5.5.0.6

* agent: fix potential race when good client can be removed instead of bad for the same transport session

* tryAgentError

* case

.github/workflows/build.yml

@@ -11,7 +11,7 @@ on:
 
 jobs:
   build:
-    name: build-${{ matrix.os }}
+    name: build-${{ matrix.os }}-${{ matrix.ghc }}
     runs-on: ${{ matrix.os }}
     strategy:
       fail-fast: false
@@ -19,17 +19,22 @@ jobs:
         include:
           - os: ubuntu-20.04
             platform_name: 20_04-x86-64
+            ghc: "8.10.7"
+          - os: ubuntu-20.04
+            platform_name: 20_04-x86-64
+            ghc: "9.6.3"
           - os: ubuntu-22.04
             platform_name: 22_04-x86-64
+            ghc: "9.6.3"
     steps:
       - name: Clone project
-        uses: actions/checkout@v2
+        uses: actions/checkout@v3
 
       - name: Setup Haskell
-        uses: haskell/actions/setup@v1
+        uses: haskell-actions/setup@v2
         with:
-          ghc-version: "8.10.7"
-          cabal-version: "latest"
+          ghc-version: ${{ matrix.ghc }}
+          cabal-version: "3.10.1.0"
 
       - name: Cache dependencies
         uses: actions/cache@v2
@@ -70,7 +75,7 @@ jobs:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
 
       - name: Create release
-        if: startsWith(github.ref, 'refs/tags/v') && matrix.os == 'ubuntu-20.04'
+        if: startsWith(github.ref, 'refs/tags/v') && matrix.os == 'ubuntu-20.04' && matrix.ghc == '9.6.3'
         uses: softprops/action-gh-release@v1
         with:
           body: |

.hlint.yaml

@@ -0,0 +1 @@
+- ignore: {name: "Use underscore"}

CHANGELOG.md

@@ -1,3 +1,115 @@
+# 5.5.0
+
+Code:
+- compatible with GHC 8.10.7 to support compilation for armv7a.
+- migrate to `crypton` from deprecated `cryptonite` (the seed for DRG is now sha512-hashed).
+- use ChaChaDRG for all random IDs, keys and nonces, only using hashed entropy as seed.
+- more efficient transaction batching in SMP protocol client and server.
+
+Agent:
+- stabilize message reception and delivery, migrate message delivery to database queue.
+- additional event MSGNTF confirming that message received via notification is processed.
+- efficient processing of messages sent to multiple recipients with batched database transactions.
+- new worker abstraction for all queued tasks resilient to race conditions and some database errors.
+- many fixed race conditions.
+- background mode for iOS NSE.
+- additional error reporting to client on critical errors (to be show as alert in the clients).
+- functional api to get worker statistics.
+
+SMP/XFTP servers:
+- fix socket and memory leak on servers with high load (inactive clients without subscriptions are disconnected after set time of inactivity).
+- control port improvements.
+- fix statistics for stored queues, messages and files.
+- make writing to store log atomic (fixes a rare bug in XFTP server).
+
+# 5.4.0
+
+Migrate to GHC 9.6.3
+
+Agent:
+- database improvements:
+  - track slow queries.
+  - better performance.
+  - "busy" error handling.
+  - create parent folder when needed.
+  - support closing and re-opening database.
+- SMP agent improvements
+  - streaming for batched SMP commands.
+  - fix asynchronous JOINing connection.
+  - handle repeating JOINs without failure.
+  - api to get subscribed connections.
+  - return simplex:/ links as invitations.
+  - fix memory leak.
+- XFTP improvements:
+  - suspend when agent is suspended.
+  - support locally encrypted files.
+  - fixes - create empty file, prevent permanent error treated as temporary.
+  - upgrade HTTP2 library (fixes error handling and flow control).
+- Remote control protocol (XRCP)
+
+SMP server:
+- control port commands for GHC threads introspection.
+- allow creating new queues without subscriptions (required for iOS).
+
+XFTP server:
+- allow 64kb file chunks.
+
+NTF server:
+- faster startup.
+
+# 5.3.0
+
+Agent:
+- improve performance, track slow database queries.
+- support delivery receipts.
+
+SMP server:
+- control port
+
+# 5.2.0 (NTF server 1.5.0)
+
+Agent:
+- treat agent INACTIVE error as temporary - fixes failed message delivery in some race conditions.
+- restore connection confirmations after client restart - fixes failed connections.
+- ratchet resynchronization protocol and API.
+- increase connection version to mutually supported by both peers on each received message.
+
+Client:
+- make timeout for batched functions dependent on the number of batches - fixes expiry on large batches.
+
+Servers:
+- add timeout in case of sending TCP traffic and in case of partial delivery of requested blocks to avoid resource leaks.
+
+# 5.1.2, 5.1.3 (NTF server 1.4.1, 1.4.2)
+
+Agent:
+- ACK message on decryption error (fixes stuck message delivery bug)
+- more robust connection switching logic, API to abort switching the address
+
+Notification server:
+- batch subscriptions to SMP servers
+
+# 5.1.1 (NTF server 1.4.0)
+
+Agent:
+- store and check hashes of previous encrypted messages to differentiate between duplicates and decryption errors
+
+Server:
+- larger processing queues
+- expire messages when restoring them
+
+# 5.1.0
+
+XFTP client:
+- check encrypted file exists when uploading
+- remove user ID from deletion API
+
+Agent:
+- vacuum database on migrations
+
+SMP server:
+- configure message expiration time in INI file
+
 # 5.0.0
 
 SimpleX File Transfer Protocol (XFTP):
@@ -28,7 +140,7 @@ SMP agent:
 - batch connection deletion
 - improve asynchronous connection deletion – it may now be completed after the client is restarted as well.
 - improve subscription logic to retry if initial attempt fails.
-- end SMP client connection after a number of failed PINGs (defalt is 3).
+- end SMP client connection after a number of failed PINGs (default is 3).
 
 # 4.3.0
 
@@ -61,7 +173,7 @@ SMP agent:
 
 Notification server (v1.3.0):
 
-- check token status when sending verification notificaiton.
+- check token status when sending verification notification.
 
 # 4.1.0
 
@@ -80,15 +192,15 @@ SMP agent:
 SMP server:
 
 - Basic authentication. The server address can now include an optional password that is required to create messaging queues, so the contacts who message you will not be able to receive messages via your server, unless you share with them the address with the password. It is recommended to enable basic authentication on all private servers by adding `create_password` parameter into AUTH section of server INI file (the previously deployed servers do not have this section, you need to add it).
-- Disable creating new queues completely with `new_queues: off` parameter in AUTH section of INI file - it can be used to simplify migrating the exising connections to another server.
+- Disable creating new queues completely with `new_queues: off` parameter in AUTH section of INI file - it can be used to simplify migrating the existing connections to another server.
 - Updated server CLI with changed defaults:
-  - interactive server initialization, use -y flag to initalize the server non-interactively.
+  - interactive server initialization, use -y flag to initialize the server non-interactively.
   - store log is now enabled by default, so messaging queues and messages are restored when the server is restarted (to restore undelivered messages the server needs to be stopped with SIGINT signal).
   - a random password is now generated by default during the server initialization.
 
 SMP agent:
 
-- API to test SMP servers. It connects to the server, creates and deletes a messaging queue. The new SimpleX Chat client uses this API to allow you to test that you have the correct server address, with the valid certificate fingerprint and pasword, before enabling the new server.
+- API to test SMP servers. It connects to the server, creates and deletes a messaging queue. The new SimpleX Chat client uses this API to allow you to test that you have the correct server address, with the valid certificate fingerprint and password, before enabling the new server.
 
 # 3.4.0
 
@@ -134,9 +246,9 @@ SMP server and agent:
 
 SMP server:
 
-- restore undeliverd messages when the server is restarted.
+- restore undelivered messages when the server is restarted.
 - SMP protocol v3 to support push notification:
-  - updated SEND and MSG to add message flags (for notification flag that contros whether the notification is sent and for any future extensions) and to move message meta-data sent to the recipient into the encrypted envelope.
+  - updated SEND and MSG to add message flags (for notification flag that confirm whether the notification is sent and for any future extensions) and to move message meta-data sent to the recipient into the encrypted envelope.
   - update NKEY and NID to add e2e encryption keys (for the notification meta-data encryption between SMP server and the client), and update NMSG to include this meta-data.
   - update ACK command to include message ID (to avoid acknowledging unprocessed message).
   - add NDEL commands to remove notification subscription credentials from SMP queue.
@@ -147,7 +259,7 @@ SMP agent:
 - new protocol for duplex connection handshake reducing traffic and connection time.
 - support for SMP notifications server and managing device token.
 - remove redundant FQDN validation from TLS handshake to prepare for access via Tor.
-- support for fully stopping agent and for termporary suspending agent operations.
+- support for fully stopping agent and for temporary suspending agent operations.
 - improve management of duplicate message delivery.
 
 SMP notifications server v1.0:

Dockerfile

@@ -5,11 +5,11 @@ FROM ubuntu:${TAG} AS build
 ### Build stage
 
 # Install curl and git and simplexmq dependencies
-RUN apt-get update && apt-get install -y curl git build-essential libgmp3-dev zlib1g-dev llvm-12 llvm-12-dev libnuma-dev
+RUN apt-get update && apt-get install -y curl git build-essential libgmp3-dev zlib1g-dev llvm-12 llvm-12-dev libnuma-dev libssl-dev
 
 # Specify bootstrap Haskell versions
-ENV BOOTSTRAP_HASKELL_GHC_VERSION=8.10.7
-ENV BOOTSTRAP_HASKELL_CABAL_VERSION=3.6.2.0
+ENV BOOTSTRAP_HASKELL_GHC_VERSION=9.6.3
+ENV BOOTSTRAP_HASKELL_CABAL_VERSION=3.10.1.0
 
 # Install ghcup
 RUN curl --proto '=https' --tlsv1.2 -sSf https://get-ghcup.haskell.org | BOOTSTRAP_HASKELL_NONINTERACTIVE=1 sh
@@ -19,7 +19,7 @@ ENV PATH="/root/.cabal/bin:/root/.ghcup/bin:$PATH"
 
 # Set both as default
 RUN ghcup set ghc "${BOOTSTRAP_HASKELL_GHC_VERSION}" && \
-    ghcup set cabal
+    ghcup set cabal "${BOOTSTRAP_HASKELL_CABAL_VERSION}"
 
 COPY . /project
 WORKDIR /project

README.md

@@ -150,7 +150,7 @@ You can install and setup servers automatically using our script:
 
 ```sh
 curl --proto '=https' --tlsv1.2 -sSf https://raw.githubusercontent.com/simplex-chat/simplexmq/stable/install.sh -o simplex-server-install.sh \
-&& if echo '1268d605e90bca1a8c7ef476038a8bd9aa9b1a28f79ea0a1485669ccf8fc23cd simplex-server-install.sh' | sha256sum -c; then chmod +x ./simplex-server-install.sh && ./simplex-server-install.sh; rm /simplex-server-install.sh; else echo "SHA-256 checksum is incorrect!" && rm ./simplex-server-install.sh; fi
+&& if echo 'b8cf2be103f21f9461d9a500bcd3db06ab7d01d68871b07f4bd245195cbead1d simplex-server-install.sh' | sha256sum -c; then chmod +x ./simplex-server-install.sh && ./simplex-server-install.sh; rm ./simplex-server-install.sh; else echo "SHA-256 checksum is incorrect!" && rm ./simplex-server-install.sh; fi
 ```
 
 ### Build from source

apps/smp-agent/Main.hs

@@ -5,6 +5,7 @@
 module Main where
 
 import Control.Logger.Simple
+import Data.ByteArray (ScrubbedBytes)
 import qualified Data.List.NonEmpty as L
 import qualified Data.Map.Strict as M
 import Simplex.Messaging.Agent.Env.SQLite
@@ -19,7 +20,7 @@ cfg = defaultAgentConfig
 agentDbFile :: String
 agentDbFile = "smp-agent.db"
 
-agentDbKey :: String
+agentDbKey :: ScrubbedBytes
 agentDbKey = ""
 
 servers :: InitialAgentServers
@@ -34,9 +35,10 @@ servers =
 logCfg :: LogConfig
 logCfg = LogConfig {lc_file = Nothing, lc_stderr = True}
 
+-- Warning: this SMP agent server is experimental - it does not work correctly with multiple connected TCP clients in some cases.
 main :: IO ()
 main = do
   putStrLn $ "SMP agent listening on port " ++ tcpPort (cfg :: AgentConfig)
   setLogLevel LogInfo -- LogError
-  Right st <- createAgentStore agentDbFile agentDbKey MCConsole
+  Right st <- createAgentStore agentDbFile agentDbKey False MCConsole
   withGlobalLogging logCfg $ runSMPAgent (transport @TLS) cfg servers st

apps/smp-server/Main.hs

@@ -1,13 +1,16 @@
+{-# LANGUAGE LambdaCase #-}
+
 module Main where
 
 import Control.Logger.Simple
 import Simplex.Messaging.Server.Main
+import System.Environment
 
-cfgPath :: FilePath
-cfgPath = "/etc/opt/simplex"
+defaultCfgPath :: FilePath
+defaultCfgPath = "/etc/opt/simplex"
 
-logPath :: FilePath
-logPath = "/var/opt/simplex"
+defaultLogPath :: FilePath
+defaultLogPath = "/var/opt/simplex"
 
 logCfg :: LogConfig
 logCfg = LogConfig {lc_file = Nothing, lc_stderr = True}
@@ -15,4 +18,9 @@ logCfg = LogConfig {lc_file = Nothing, lc_stderr = True}
 main :: IO ()
 main = do
   setLogLevel LogDebug
+  cfgPath <- getEnvPath "SMP_SERVER_CFG_PATH" defaultCfgPath
+  logPath <- getEnvPath "SMP_SERVER_LOG_PATH" defaultLogPath
   withGlobalLogging logCfg $ smpServerCLI cfgPath logPath
+    
+getEnvPath :: String -> FilePath -> IO FilePath
+getEnvPath name def = maybe def (\case "" -> def; f -> f) <$> lookupEnv name

cabal.project

@@ -2,36 +2,29 @@ packages: .
 -- packages: . ../direct-sqlcipher ../sqlcipher-simple
 -- packages: . ../hs-socks
 -- packages: . ../http2
+-- packages: . ../network-transport
 
-package *
-  coverage: True
-  library-coverage: True
+index-state: 2023-12-12T00:00:00Z
 
-package attoparsec
-  coverage: False
-  library-coverage: False
+package cryptostore
+    flags: +use_crypton
 
 source-repository-package
     type: git
     location: https://github.com/simplex-chat/aeson.git
-    tag: 3eb66f9a68f103b5f1489382aad89f5712a64db7
+    tag: aab7b5a14d6c5ea64c64dcaee418de1bb00dcc2b
 
 source-repository-package
     type: git
     location: https://github.com/simplex-chat/hs-socks.git
     tag: a30cc7a79a08d8108316094f8f2f82a0c5e1ac51
 
-source-repository-package
-    type: git
-    location: https://github.com/kazu-yamamoto/http2.git
-    tag: b5a1b7200cf5bc7044af34ba325284271f6dff25
-
 source-repository-package
     type: git
     location: https://github.com/simplex-chat/direct-sqlcipher.git
-    tag: 34309410eb2069b029b8fc1872deb1e0db123294
+    tag: f814ee68b16a9447fbb467ccc8f29bdd3546bfd9
 
 source-repository-package
     type: git
     location: https://github.com/simplex-chat/sqlcipher-simple.git
-    tag: 5e154a2aeccc33ead6c243ec07195ab673137221
+    tag: a46bd361a19376c5211f1058908fc0ae6bf42446

cbits/README.md

@@ -0,0 +1,3 @@
+# Streamlined NTRU Prime: sntrup761
+
+The implementation of sntrup761 is the _exact_ copy from this [Internet draft](https://www.ietf.org/archive/id/draft-josefsson-ntruprime-streamlined-00.html).

cbits/sha512.c

@@ -0,0 +1,9 @@
+#include <openssl/sha.h>
+#include "sha512.h"
+
+void crypto_hash_sha512 (unsigned char *out,
+                         const unsigned char *in,
+                         unsigned long long inlen)
+{
+  SHA512(in, inlen, out);
+}

cbits/sha512.h

@@ -0,0 +1,3 @@
+void crypto_hash_sha512 (unsigned char *out,
+                         const unsigned char *in,
+                         unsigned long long inlen);

cbits/sntrup761.h

@@ -0,0 +1,33 @@
+/*
+ * Derived from public domain source, written by (in alphabetical order):
+ * - Daniel J. Bernstein
+ * - Chitchanok Chuengsatiansup
+ * - Tanja Lange
+ * - Christine van Vredendaal
+ */
+
+#ifndef SNTRUP761_H
+#define SNTRUP761_H
+
+#include <string.h>
+#include <stdint.h>
+
+#define SNTRUP761_SECRETKEY_SIZE 1763
+#define SNTRUP761_PUBLICKEY_SIZE 1158
+#define SNTRUP761_CIPHERTEXT_SIZE 1039
+#define SNTRUP761_SIZE 32
+
+typedef void sntrup761_random_func (void *ctx, size_t length, uint8_t *dst);
+
+void
+sntrup761_keypair (uint8_t *pk, uint8_t *sk,
+                   void *random_ctx, sntrup761_random_func *random);
+
+void
+sntrup761_enc (uint8_t *c, uint8_t *k, const uint8_t *pk,
+               void *random_ctx, sntrup761_random_func *random);
+
+void
+sntrup761_dec (uint8_t *k, const uint8_t *c, const uint8_t *sk);
+
+#endif /* SNTRUP761_H */

fourmolu.yaml

@@ -0,0 +1,30 @@
+indentation: 2
+column-limit: none
+function-arrows: trailing
+comma-style: trailing
+import-export-style: trailing
+indent-wheres: true
+record-brace-space: true
+newlines-between-decls: 1
+haddock-style: single-line
+haddock-style-module: null
+let-style: inline
+in-style: right-align
+single-constraint-parens: never
+unicode: never
+respectful: true
+fixities:
+  - infixr 9 .
+  - infixr 8 .:, .:., .=
+  - infixr 6 <>
+  - infixr 5 ++
+  - infixl 4 <$>, <$, $>,  <$$>, <$?>
+  - infixl 4 <*>, <*, *>, <**>
+  - infix 4 ==, /=
+  - infixr 3 &&
+  - infixl 3 <|>
+  - infixr 2 ||
+  - infixl 1 >>, >>=
+  - infixr 1 =<<, >=>, <=<
+  - infixr 0 $, $!
+reexports: []

install.sh

@@ -11,6 +11,7 @@ scripts_systemd_smp="$scripts/smp-server.service"
 scripts_systemd_xftp="$scripts/xftp-server.service"
 scripts_update="$scripts/simplex-servers-update"
 scripts_uninstall="$scripts/simplex-servers-uninstall"
+scripts_stopscript="$scripts/simplex-servers-stopscript"
 
 # Default installation paths
 path_bin="/usr/local/bin"
@@ -18,6 +19,7 @@ path_bin_smp="$path_bin/smp-server"
 path_bin_xftp="$path_bin/xftp-server"
 path_bin_update="$path_bin/simplex-servers-update"
 path_bin_uninstall="$path_bin/simplex-servers-uninstall"
+path_bin_stopscript="$path_bin/simplex-servers-stopscript"
 
 path_conf_etc="/etc/opt"
 path_conf_var="/var/opt"
@@ -60,8 +62,8 @@ ${GRN}3.${NC} Setup user for each server:
 ${GRN}4.${NC} Create systemd services:
     - smp: ${YLW}${path_systemd_smp}${NC}
     - xftp: ${YLW}${path_systemd_xftp}${NC}
-${GRN}5.${NC} Install update and uninstallation script:
-    - all: ${YLW}${path_bin_update}${NC}, ${YLW}${path_bin_uninstall}${NC}
+${GRN}5.${NC} Install stopscript (systemd), update and uninstallation script:
+    - all: ${YLW}${path_bin_update}${NC}, ${YLW}${path_bin_uninstall}${NC}, ${YLW}${path_bin_stopscript}${NC}
 
 Press ${GRN}ENTER${NC} to continue or ${RED}Ctrl+C${NC} to cancel installation"
 
@@ -87,9 +89,9 @@ setup_users() {
 setup_dirs() {
  # Unquoted varibles, so field splitting can occur
  mkdir -p $path_conf_smp
- chown "$user_smp":"$user_smp" "$path_conf_smp"
+ chown "$user_smp":"$user_smp" $path_conf_smp
  mkdir -p $path_conf_xftp
- chown "$user_xftp":"$user_xftp" "$path_conf_xftp"
+ chown "$user_xftp":"$user_xftp" $path_conf_xftp
 }
 
 setup_systemd() {
@@ -100,6 +102,7 @@ setup_systemd() {
 setup_scripts() {
  curl --proto '=https' --tlsv1.2 -sSf -L "$scripts_update" -o "$path_bin_update" && chmod +x "$path_bin_update"
  curl --proto '=https' --tlsv1.2 -sSf -L "$scripts_uninstall" -o "$path_bin_uninstall" && chmod +x "$path_bin_uninstall"
+ curl --proto '=https' --tlsv1.2 -sSf -L "$scripts_stopscript" -o "$path_bin_stopscript" && chmod +x "$path_bin_stopscript"
  }
 
 checks() {
@@ -131,7 +134,7 @@ main() {
  setup_systemd
  printf "${GRN} Done!${NC}\n"
 
- printf "Installing update and uninstallation script..."
+ printf "Installing stopscript, update and uninstallation script..."
  setup_scripts
  printf "${GRN} Done!${NC}\n"
 

package.yaml

@@ -1,5 +1,5 @@
 name: simplexmq
-version: 5.1.1
+version: 5.5.0.6
 synopsis: SimpleXMQ message broker
 description: |
   This package includes <./docs/Simplex-Messaging-Server.html server>,
@@ -20,9 +20,11 @@ category: Chat, Network, Web, System, Cryptography
 extra-source-files:
   - README.md
   - CHANGELOG.md
+  - cbits/sha512.h
+  - cbits/sntrup761.h
 
 dependencies:
-  - aeson == 2.0.*
+  - aeson == 2.2.*
   - ansi-terminal >= 0.10 && < 0.12
   - asn1-encoding == 0.9.*
   - asn1-types == 0.3.*
@@ -30,49 +32,46 @@ dependencies:
   - attoparsec == 0.14.*
   - base >= 4.14 && < 5
   - base64-bytestring >= 1.0 && < 1.3
-  - bytestring == 0.10.*
   - case-insensitive == 1.2.*
   - composition == 1.0.*
   - constraints >= 0.12 && < 0.14
   - containers == 0.6.*
-  - cryptonite >= 0.27 && < 0.30
-  - cryptostore == 0.2.*
+  - crypton == 0.34.*
+  - crypton-x509 == 1.7.*
+  - crypton-x509-store == 1.6.*
+  - crypton-x509-validation == 1.6.*
+  - cryptostore == 0.3.*
   - data-default == 0.7.*
   - direct-sqlcipher == 2.3.*
   - directory == 1.3.*
   - filepath == 1.4.*
+  - hourglass == 0.2.*
   - http-types == 0.12.*
-  - http2 == 4.1.*
-  - generic-random >= 1.3 && < 1.5
+  - http2 >= 4.2.2 && < 4.3
   - ini == 0.4.1
   - iproute == 1.7.*
   - iso8601-time == 0.1.*
-  - memory == 0.15.*
-  - mtl == 2.2.*
+  - memory == 0.18.*
+  - mtl >= 2.3.1 && < 3.0
   - network >= 3.1.2.7 && < 3.2
-  - network-transport == 0.5.4
+  - network-info >= 0.2 && < 0.3
+  - network-transport == 0.5.6
+  - network-udp >= 0.0 && < 0.1
   - optparse-applicative >= 0.15 && < 0.17
-  - QuickCheck == 2.14.*
   - process == 1.6.*
   - random >= 1.1 && < 1.3
   - simple-logger == 0.1.*
   - socks == 0.6.*
   - sqlcipher-simple == 0.4.*
   - stm == 2.5.*
-  - template-haskell == 2.16.*
   - temporary == 1.3.*
-  - text == 1.2.*
-  - time == 1.9.*
-  - time-compat == 1.9.*
+  - time == 1.12.*
   - time-manager == 0.0.*
-  - tls >= 1.6.0 && < 1.7
-  - transformers == 0.5.*
+  - tls >= 1.7.0 && < 1.8
+  - transformers == 0.6.*
   - unliftio == 0.2.*
   - unliftio-core == 0.2.*
   - websockets == 0.12.*
-  - x509 == 1.7.*
-  - x509-store == 1.6.*
-  - x509-validation == 1.6.*
   - yaml == 0.11.*
 
 flags:
@@ -80,14 +79,33 @@ flags:
     description: Enable swift JSON format
     manual: True
     default: False
+  use_crypton:
+    description: Use crypton etc. in cryptostore
+    manual: True
+    default: True
 
 when:
   - condition: flag(swift)
     cpp-options:
       - -DswiftJSON
+  - condition: impl(ghc >= 9.6.2)
+    dependencies:
+      - bytestring == 0.11.*
+      - template-haskell == 2.20.*
+      - text >= 2.0.1 && < 2.2
+  - condition: impl(ghc < 9.6.2)
+    dependencies:
+      - bytestring == 0.10.*
+      - template-haskell == 2.16.*
+      - text >= 1.2.3.0 && < 1.3
 
 library:
   source-dirs: src
+  c-sources:
+    - cbits/sha512.c
+    - cbits/sntrup761.c
+  include-dirs: cbits
+  extra-libraries: crypto
 
 executables:
   smp-server:
@@ -97,6 +115,7 @@ executables:
       - simplexmq
     ghc-options:
       - -threaded
+      - -rtsopts
 
   ntf-server:
     source-dirs: apps/ntf-server
@@ -105,6 +124,7 @@ executables:
       - simplexmq
     ghc-options:
       - -threaded
+      - -rtsopts
 
   xftp-server:
     source-dirs: apps/xftp-server
@@ -113,6 +133,7 @@ executables:
       - simplexmq
     ghc-options:
       - -threaded
+      - -rtsopts
 
   smp-agent:
     source-dirs: apps/smp-agent
@@ -121,6 +142,7 @@ executables:
       - simplexmq
     ghc-options:
       - -threaded
+      - -rtsopts
 
   xftp:
     source-dirs: apps/xftp
@@ -129,6 +151,7 @@ executables:
       - simplexmq
     ghc-options:
       - -threaded
+      - -rtsopts
 
 tests:
   simplexmq-test:
@@ -137,15 +160,14 @@ tests:
     dependencies:
       - simplexmq
       - deepseq == 1.4.*
-      - hspec == 2.7.*
-      - hspec-core == 2.7.*
+      - generic-random == 1.5.*
+      - hspec == 2.11.*
+      - hspec-core == 2.11.*
       - HUnit == 1.6.*
       - QuickCheck == 2.14.*
       - silently == 1.2.*
       - main-tester == 0.2.*
       - timeit == 2.0.*
-    ghc-options:
-      - -fhpc
 
 ghc-options:
   # - -haddock

protocol/simplex-messaging.md

@@ -347,7 +347,7 @@ Simplex messaging server implementations MUST NOT create, store or send to any o
 
 - Snapshots of the database they use to store queues and messages (instead simplex messaging clients must manage redundancy by using more than one simplex messaging server). In-memory persistence is recommended.
 
-- Any other information that may compromise privacy or [forward secrecy][4] of communication between clients using simplex messaging servers.
+- Any other information that may compromise privacy or [forward secrecy][4] of communication between clients using simplex messaging servers (the servers cannot compromise forward secrecy of any application layer protocol, such as double ratchet).
 
 ## Message delivery notifications
 
@@ -364,9 +364,9 @@ The clients can optionally instruct a dedicated push notification server to subs
 
 [`SEND` command](#send-message) includes the notification flag to instruct SMP server whether to send the notification - this flag is forwarded to the recepient inside encrypted envelope, together with the timestamp and the message body, so even if TLS is compromised this flag cannot be used for traffic correlation.
 
-## SMP Transmission andtransport block structure
+## SMP Transmission and transport block structure
 
-Each transport block (SMP transmission) has a fixed size of 16384 bytes for traffic uniformity.
+Each transport block has a fixed size of 16384 bytes for traffic uniformity.
 
 From SMP version 4 each block can contain multiple transmissions, version 3 blocks have 1 transmission.
 Some parts of SMP transmission are padded to a fixed size; this padding is uniformly added as a word16 encoded in network byte order - see `paddedString` syntax.
@@ -387,17 +387,17 @@ Each transmission/block for SMP v3 between the client and the server must have t
 
 ```abnf
 paddedTransmission = <padded(transmission), 16384>
-transmission = [signature] SP signed
-signed = sessionIdentifier SP [corrId] SP [queueId] SP smpCommand
+transmission = signature signed
+signed = sessionIdentifier corrId queueId smpCommand
 ; corrId is required in client commands and server responses,
 ; it is empty in server notifications.
-corrId = 1*32(%x21-7F) ; any characters other than control/whitespace
-queueId = encoded ; max 32 bytes when decoded (24 bytes is used),
+corrId = length *OCTET
+queueId = length *OCTET
 ; empty queue ID is used with "create" command and in some server responses
-signature = encoded
+signature = length *OCTET
 ; empty signature can be used with "send" before the queue is secured with secure command
 ; signature is always empty with "ping" and "serverMsg"
-encoded = <base64 encoded binary>
+length = 1*1 OCTET
 ```
 
 `base64` encoding should be used with padding, as defined in section 4 of [RFC 4648][9]

protocol/xftp-uri-request.txt

@@ -0,0 +1,16 @@
+Scheme name: xftp
+
+Status: Provisional
+
+Applications/protocols that use this scheme name:
+This scheme is used for URIs of XFTP (SimpleX File Transfer Protocol) servers,
+a client-server protocol for asynchronous file transfer via relays,
+preserving file meta-data (including size and name) and content privacy.
+
+Contact: Evgeny Poberezkin <ep@simplex.chat>
+
+Change controller: Evgeny Poberezkin <ep@simplex.chat>
+
+References:
+The syntax for server URIs in the provisional specification for SimpleX File Transfer Protocol:
+https://github.com/simplex-chat/simplexmq/blob/master/rfcs/2022-12-26-simplex-file-transfer.md#server-address-syntax

protocol/xrcp-uri-request.txt

@@ -0,0 +1,15 @@
+Scheme name: xrcp
+
+Status: Provisional
+
+Applications/protocols that use this scheme name:
+This scheme is used for URIs of controller sessions via SimpleX Remote Control protocol (XRCP),
+a protocol for remote access and management of hosts via insecure network.
+
+Contact: Evgeny Poberezkin <ep@simplex.chat>
+
+Change controller: Evgeny Poberezkin <ep@simplex.chat>
+
+References:
+The syntax for server URIs in the provisional specification for SimpleX File Transfer Protocol:
+https://github.com/simplex-chat/simplexmq/blob/master/rfcs/2023-10-25-remote-control.md#session-invitation

rfcs/2022-12-26-simplex-file-transfer.md

@@ -149,6 +149,19 @@ parts:
 
 This file description is sent to all recipients via normal messages, split to 15780 byte chunks if needed.
 
+### Server address syntax
+
+The server address is a URI with the following format:
+
+```abnf
+xftpServerURI = %s"xftp://" xftpServer
+xftpServer = serverIdentity "@" srvHost [":" port]
+srvHost = <hostname> ; RFC1123, RFC5891
+port = 1*DIGIT
+serverIdentity = base64url
+base64url = <base64url encoded binary> ; RFC4648, section 5
+```
+
 ### Receiving file
 
 Having received the description, the recipient will:

rfcs/2023-05-02-resync-ratchets.md

@@ -0,0 +1,58 @@
+# Re-sync encryption ratchets, queue rotation, message delivery receipts
+
+This is very unfocussed doc outlining several problems that seem somewhat related, and some possible solution approaches.
+
+## Problems
+
+When old database backup is used, the sending client encrypts the messages in a way that the receiving client cannot decrypt them, because it does not store the old keys, and does not try to decrypt ratchet headers using the old keys. When receiving the messages, the client that uses the old database is unable to decrypt them, either the header or the message itself (if more than 500 messages are skipped, or if root ratchet step happened).
+
+The same symptoms happen during the queue rotation, unfortunately, as we do not keep the history of received messages once they are acknowledged, so the client reports these error to the users.
+
+While these are two different problems, to the users they manifest in the same way, so it is difficult to differentiate.
+
+Another ratchet problem that should be solved as well is not deleting unused skipped keys after some time or some number of events, possibly it should be considered at the same time.
+
+While it is tempting to consider delivery receipts as an isolated problem, it is somewhat coupled, as the requests to re-sync ratchets and to re-deliver skipped messages are effectively delivery receipts.
+
+A naive solution to delivery receipts when each message is responded to with a confirmation has two downsides:
+
+- it doubles the traffic.
+- in the absense of delays, it also leaks some information about the network latency, that can be used to track user's location.
+
+Both problems can be mitigated by sending delivery receipts with a random delay, to confirm all messages sent, not each one. But this would only reduce the traffic when the sender sends many messages. Alternatively, we could think how to reduce the block size for SMP commands. The challenge is profile pictures and image previews, so to reduce traffic we need to start sending them as XFTP files, potentially with a smaller 64kb chunks (that would also increase preview quality).
+
+## Solution for queue rotation
+
+We need to:
+- store and show the rotation status in the UI to prevent attempts to rotate the queue multiple times.
+- process multiple message delivery events while there is a redundancy, to avoid errors in terminal.
+- differentiate between decryption errors during queue rotation - that would require keeping the messages for some time, at least keeping their IDs and hashes together with some additional status. In this case when duplicate or old message arrives, we can check whether this message was already processed, and do not show error if it was.
+
+## Solution to re-sync ratchets
+
+When we establish that the receiving client is indeed unable to decrypt messages, we should:
+- notify the user - it is already happening, but it happens in case of rotation as well.
+- optionally, notify another client that message was not decrypted - this effectively becomes a delivery receipt. That could be a separate command triggered by the user in the UI.
+- clients can get into this state at the same time, how do we handle concurrent requests? Possibly, we could design these messages so this is irrelevant whether the message is sent in response or not, and calculate the new ratchet keys once the clients have both the sent and received messages. After that both clients can send "ratchet reset" message:
+  - EKEY - message containing a new set of keys to initialize the ratchets
+  - EREADY - message confirming that the ratchet is ready, already e2e encrypted
+
+The downside of this approach, is that unlike the initial handshake, where the second ratchet key is sent in e2e encrypted response, this message will be sent only encrypted with queue key. We could change this protocol by adding EREPLY message, and the clients would know to ingnore EKEY with bigger hash of the keys, so that the client who sent keys with smaller hash would be processing response from another client (that is if EKEY is received after EKEY is sent).
+
+```
+A                B
+
+EKEY 1 ->        <- EKEY 2 (e2e encrypted with per-queue key only)
+EKEY 2 <-        -> EKEY 1
+1 < 2, wait      2 > 1, reply EREPLY 3 (e2e encrypted with ratchets)
+ratchet: 1, 3   ratchet: 3, 1
+```
+
+Once ratchets are re-synched the clients could additionally request delivery of skipped messages. As ratchets being out of sync is not the only problem when messages can be lost, it may be managed independently from the ratchet re-sync. This problem seems tied with delivery receipts too, as requesting to re-deliver some messages automatically confirms that some other messages were delivered.
+
+Alternatively, the request for ratchet sync can already contain the ID of the first failed message - that would allow the agent:
+- stop sending messages until ratchets are in sync.
+- re-deliver failed messages first.
+- only then deliver any pending messages.
+
+This approach would probably result in a better UX (no out of order delivery), but requires tracking pending messages that are not attempted to deliver and to also insert old messages in the delivery queue with the old IDs.

rfcs/2023-05-03-delivery-receipts.md

@@ -0,0 +1,170 @@
+# Delivery receipts
+
+## Problems
+
+User experience - users need to know that the messages are delivered to the recipient, as this confirms that the system is functioning.
+
+The downside of communicating message delivery as it confirms that the recipient was online, and, unless there is a delay in confirming, can be used to track the location via the variation in network latency. So delivery receipts should be delayed with a randomized interval and should be opt in or opt out.
+
+Another problem of message receipts is that they increase network traffic and server load. This could be avoided if delivery receipts are communicated as part of normal message delivery flow.
+
+Some other existing and planned features implicitely confirm message delivery and, possibly, should depend on message delivery being enabled:
+- agent message to resume delivery when quota was exceeded (implemented, [rfc](./2022-12-27-queue-quota.md))
+- agent message to re-deliver skipped messages or to re-negotiate double ratchet.
+
+## Solution
+
+There are three layers where delivery receipts can be implemented:
+- chat protocol. Pro: logic of when to deliver it is decoupled from the message flow, Con: extra traffic, can only work in duplex connections.
+- agent client protocol. Pro: can be automated and combined with the protocol to re-deliver skipped messages. Con: extra traffic.
+- SMP protocol. Pro: minimal extra traffic, Con: complicates server design as it would require pushing receipts when there is no next message.
+
+The last approach seems the most promising for avoiding additional traffic:
+- modify client ACK command to include whether delivery receipt should be provided to sender, and, possibly, any e2e encrypted data that should be included in the receipt (e.g., that the receiving client already saw this message in case we use "feedback" variant of roumor-mongering protocol for groups).
+- server would manage delaying of the receipts, by randomizing the time after which the receipt will be available to the sender, and by combining the receipts when possible.
+- modify response to SEND command to include any available delivery receipts.
+- add a separate delivery receipt that will be pushed to the sender in the connection where the message was received by the server.
+
+## SMP protocol changes
+
+```haskell
+data Command (p :: Party) where
+  -- ...
+  ACK :: MsgId -> Maybe ByteString -> Command Recipient
+  -- the presense of ByteString in ACK indicates that the delivery needs to be confirmed.
+  -- the protocol does not define the format of this confirmation, it is application specific, and can be -- an empty string.
+  -- And open question is how to e2e encrypt information in this string - this probably can be handled on Agent client protocol level, and could be the same ratchet key that was used to encrypt and decrypt the message. The downside of this approach is that this key currently is not stored, and storing it requires additional logic to clear these keys if unused after some time.
+  -- TODO consider what could be a better approach.
+  SENT :: MsgId -> [(MsgId, UTCTime, ByteString)] -> Command Sender
+  -- or
+  -- SENT :: MsgId -> Command Sender
+  -- in case we just batch
+  -- this response will be sent to SEND command and will include a sender's message ID generated by the server (currently it does not exist), and posibly an empty list of delivery receipts with the same message IDs as in responses to SEND, timestamps when these receipts became available, and e2e encrypted ByteString passed in ACK command.
+  -- The ID in this response should be different from the ID used in MSG, to keep the promise of not having shared identifiers in sent/received traffic even inside TLS tunnel.
+  -- Keeping the quality of shared ciphertext also requires adding additional encryption layer between the server and the sender, this can be achieved in one of two ways:
+  -- 1) passing a separate DH key in each SEND command, and server including additional DH key in each SENT response, with computed DH secret per message later used to encrypt and decrypt the delivery receipt payloads. This is probably a bad idea as it would increase a cryptographic load on both the server and the client.
+  -- 2) agree a key per queue, in the same way it is done for the recipient. Possibly, it requires additional DH key in confirmation message that the recipient then uses to secure the queue, and passing this key in KEY (secure queue) command. The response to this secure command would the include server's DH key returned to the recipient that would be passed to the sender in HELLO message. Even though recipient could observe both public DH keys, they won't know the computed shared secret. Recipient that controls the server could perform MITM attack on this key exchange, but it doesn't give any benefit over what recipient can do when they have access to the server - the threat model remains the same. The downside of this approach is that it also requires additional changes in client protocol level (confirmation message format and HELLO message).
+  -- 3) also agree on a key per queue, but via separate commands between the sender and the server, once the sender was notified that the queue is secured. This approach is probably better, and the server would simply delay the delivery of delivery receipts until the shared secret is agreed.
+  SKEY :: C.PublicKeyX25519 -> Command Sender
+  SBKEY :: C.PublicKeyX25519 -> BrokerMsg
+  -- these are the command and response to agree secret to encrypt delivery receipt payloads for option 3
+  SSUB :: Command Sender
+  -- subscribe to receive delivery receipts for a given queue - will be sent when the conversation is opened (unless there is an active subscription already), not all queues at once, and won't be re-subscribed on losing the server connection (TBC).
+  RCVD :: MsgId -> UTCTime -> ByteString -> Command Recipient
+  -- delivery receipt. UTCTime is the time when it became available, not the time when ACK was sent by the recipient, to avoid leaking location via network latency.
+```
+
+Possibly, there is no need to include delivery receipts into SENT response and instead just use batching of responses that is already supported. As server responses are not signed, there is no per-response overhead that is substantial, and a lot of receipts that are available can be packed into one block (depending on the size of payload that has to be fixed not to leak metadata).
+
+This all seems rather complicated for SMP protocol, and the approach of doing it on a higher level seems more attractive than initially. Possibly we should reconsider, and reduce traffic by reducing block sizes... Reducing block sizes unfortunately requires supporting variable block sizes, and would leak some metadata during the transition period.
+
+## Another approach
+
+Above represents substantial complexity, and at least doubles server code complexity for the feature that is definitely not doubling the value of server software. Moving to variable block size is simpler, but also has a lot of complexity, reduces metadata privacy (at least for the duration of migration period), reduces image preview quality, and requires postponing this feature for multiple releases, until all clients migrate.
+
+Given that the main traffic is generated by the groups, and direct messages do not create a lot of traffic, a much simpler and better solution is to simply send delivery receipts as the message, in direct conversations only, either as chat protocol message or as agent client protocol message (either on the message or on the envelope layer).
+
+### Comparison of these two approaches:
+
+**Chat protocol message**
+
+Pros:
+- simpler, more contained change - SMP layer is not aware of this feature
+- easier to extend protocol with additional application specific payload, e.g. references to group DAG
+Cons:
+- ?
+
+```json
+  // ...
+  "x.msg.delivered": {
+    "properties": {
+      "msgId": {"ref": "base64url"},
+      "params": {
+        "properties": {
+          "msgId": {"ref": "base64url"},
+        },
+        "optionalProperties": {
+          "data": {} // possibly the initial protocol does not need it, with JSON can be added later
+        }
+      }
+    }
+  },
+  // ...
+```
+
+**Agent client protocol envelope**
+
+Pros:
+- possibility of using it in a wider range of the applications
+- possibility to include received message hash to increase communication integrity - the sending client would be then notified, and it can be exposed in the UI, that the received message is not the same as sent.
+
+Cons:
+- additional implementation complexity - requires additional events to communicate between chat and agent.
+
+```haskell
+data AMessage =
+  -- ...
+  | A_RCVD AgentMsgId MsgHash ByteString -- references to the received message
+  -- ...
+```
+
+The weirdness of the above design is that it refers to the data present in the header of another message, the alternative would be to have a separate envelope for delivery receipt:
+
+```haskell
+data AgentMessage =
+  -- ...
+  AgentMessageRcvd APrivHeader AgentMsgId MsgHash ByteString -- references to the received message
+  -- ...
+```
+
+But probably the first one is a bit better, TBC.
+
+In any case there should be an additional event to notify chat client:
+
+```haskell
+data ACommand (p :: AParty) (e :: AEntity) where
+  -- ...
+  RCVD AgentMsgId MsgMeta ByteString -> ACommand Agent AEConn
+  -- ...
+```
+
+On the balance of things, implementing on the level of Agent Client protocol seems better, as the additional complexity is marginal, but it allows for wider range of applications, and also allows for additional delivery integrity validation. The format for payload still requires chat protocol message encoding once we want to add it, but initially it could be just an empty string.
+
+## Implementation plan
+
+Currently, we delete sent messages once delivered to the server. It would be helpful if we could keep the records in snd_messages table, and then use them to process delivery receipts, although it may be insufficient (we could add fields). Probably it is not possible to keep them as there is a foreign key constraint with `on delete cascade`.
+
+The new table will be used to track sent message hashes to correlate their IDs with delivery receipts (that will also be stored in messages/rcv_messages). Receipts need to be processed in chat in the same way as normal messages, so they would be sent to chat with MsgMeta, and the chat will need to ack them once processed.
+
+Agent ackMessage function will be also used to automatically schedule sending delivery receipts if they are enabled for connection, only for normal messages - no sending receipts to receipts.
+
+There can be receipt re-delivery to the chat in the same cases as when normal message can be re-delivered (in case of AGENT A_DUPLICATE when it was not ack'd by the user).
+
+## Other considerations
+
+### How clients decide whether to send delivery receipts
+
+Two options are possible - local settings per conversation or chat preferences framework, that allows to have mutual on/off. The latter seems preferable as without knowing whether the other party is sending receipts, it is not possible to uderstand what the absense of the receipt means - network malfunction or receipts disabled.
+
+Groups are a special case, as while some groups may enable sending the delivery receipts, the group members should be able to disable it locally. This should probably be done via a separate conversation setting in the same way as enabling notifications or favourites. In this case the receipt would only be sent to the group if it is enabled in the group and not disabled by the member. The toggle may be located in the same page as chat preferences, but it should be a separate setting. We might want though to communicate somehow whether a given member sends delivery receipts so that other members know whether to expect them or not.
+
+### How this functionality is released
+
+5.2:
+- support for sending and receiving delivery receipts preference, both in direct messages and in groups (for forward compatibility).
+- support for sending and receiving delivery receipts in direct chats only, but disable sending them
+
+5.3:
+- show receipt preferences in the UI
+- enable sending receipts in direct chats where they are enabled
+
+A separate question is how to enable this functionality for the existing contacts. Possible options are:
+
+1. Enable (as per default) for all contacts, show notification to the user when they open the app for update that delivery notifications are now sent by default to all contacts. Pro: no extra logic to implement. Con: may be perceived as a privacy violation, as to some contacts the delivery receipts will be sent before the user had chance to disable them.
+2. Ask the user when the new version first runs whether they want to use delivery receipts and offer these options:
+  1) keep enabled for all profiles (and for all contacts)
+  2) enable for all profiles in ~12 or in ~24 hours giving users the chance to review all contacts / profiles and disable some of them. The problem here is also in possibility of the correlation in case they all start sending receipts at the same time. Possibly the option could be to set a random time in 12-15 or 24-30 hours range to avoid the possibility of such correlation.
+  3) disable for all profiles – it will require sending profile updates to all contacts, so the delivery receipts should be kept disabled and profile updates should be sent after random intervals, one by one (not scheduled all at once, as the time may pass while the app is off).
+3. Offer an option to enable globally later - we could keep it as a one-off option, available only to existing users, and visible on the top level of the Settings - once enabled, the option will disappear and it won't be possible to disable again. The downside here is that the new contacts would be receiving the profile with enabled notifications but they still won't be delivered...
+4. Another option is to have all new contacts decided based on a global user default (that can be set in the settings and in the dialog on first start), but for the existing contacts keep in unset state that is not interpreted as either on or off, but interpreted as unknown until the user makes a choice... That might be an optimal solution for the users but it would probably require changing the preferences framework or some ad-hoc hacks. That still keeps the question open how to avoid correlation between profiles.
+5. That might be the case for version agreement too - the availability of the option per contact will depend on the version. It doesn't answer the question what to do with global defaults though...

rfcs/2023-05-24-smp-delivery-proxy.md

@@ -0,0 +1,34 @@
+# SMP and XFTP delivery relays
+
+## Problem
+
+SimpleX network relies on SMP relays to hold sent messages while the recipient is offline. This design provides many advantages over pure p2p designs [1]. The threat model for relays is covered in [2].
+
+The problem with this design is that the relays are chosen by the recipients, and they as a result can be used to record transport addresses of the senders.
+
+Similar problem, but reversed, exists with XFTP design - XFTP relays are chosen by the senders and also can be modified to record transport addresses of the recipients.
+
+The current solution to this problem is to use onion routing to access relays, and many users do it already. The downside is that it requires additional configuration and can be complex to some users.
+
+## Solution approach
+
+Add a second type of relays to the network design.
+
+For SMP protocol, these relays would accept messages from the senders and deliver them to the recipients. If the recipient were to modify the relay used to receive the messages to also records senders' IP addresses, then they would only succeed in recording the IP address of the relay, not the sender.
+
+Similar relay can be introduced for XFTP protocol.
+
+The additional advantage of using the relays is the reduced number of network connections that need to be used, and reduced traffic.
+
+Compared with Tor this approach has:
+- lower latency
+- no centralized components or the registry of the servers
+- still no visibility of the whole network to the relays, as sending relays do not need to randomly choose the next relay, but would forward the messages directly to the destination relay.
+
+The downside of this approach is that relays will have more visibility of the network than in the current design.
+
+## Implementation considerations
+
+1. Should this new relays be included in the existing SMP and XFTP protocols and servers, or should they be designed as separate protocols and servers?
+
+2. Should the client aim to choose the same host for delivery relay as the destination server (to reduce the traffic), if possible, or should it aim to choose a different host (to reduce metadata available to single host)? The answer to this would affect the answer to the first question.

rfcs/2023-06-08-resync-ratchets.md

@@ -0,0 +1,366 @@
+# Re-sync encryption ratchets
+
+## Problem
+
+See https://github.com/simplex-chat/simplexmq/pull/743/files for problem and high-level solution.
+
+## Implementation
+
+### Diagnosing ratchet de-synchronization
+
+Message decryption happens in `agentClientMsg`, in `agentRatchetDecrypt`, which can return decryption result or error. Decryption error can be differentiated in `agentClientMsg` result pattern match, in Left cases, where we already differentiate duplicate error (`AGENT A_DUPLICATE`).
+
+Question: Which decryption errors can be diagnosed as ratchet de-synchronization?
+
+Possibly any `AGENT A_CRYPTO` error. Definitely on `RATCHET_HEADER`, TBC other. See `cryptoError :: C.CryptoError -> AgentErrorType` for conversion from decryption errors to `A_CRYPTO` or other agent errors. We're only interested in crypto errors, as other are either other client implementation errors, internal errors, or already processed duplicate error.
+
+Proposed classification of crypto errors, based on `AgentCryptoError`:
+
+`DECRYPT_AES` -> re-sync allowed (recommended/required?)
+
+`DECRYPT_CB` -> re-sync allowed (recommended/required?)
+
+`RATCHET_HEADER` -> **re-sync required**
+
+`RATCHET_EARLIER` -> re-sync allowed
+
+`RATCHET_SKIPPED` -> **re-sync required**
+
+Ratchet re-synchronization could be started automatically on diagnosing de-synchronization, based on these errors. As a potentially dangerous feature (e.g., implementation error could lead to infinite re-sync loop causing large traffic consumption), initially it will be available via agent functional api for client to call. Ratchet de-synchronization will instead produce an event prompting client to re-synchronize.
+
+Diagnosing possible ratchet de-synchronization also will be recorded as connection state - `ratchet_desync_state` field in `connections` table. Client should be prohibited to start ratchet re-synchronization unless `ratchet_desync_state` is set.
+
+Event should not be repeated for following received messages that can't be decrypted - based on `ratchet_desync_state`. If a received message can be decrypted, `ratchet_desync_state` should be set to NULL and a new event sent, indicating ratchet has healed.
+
+New event - `RDESYNC :: RatchetDesyncState -> ConnectionStats -> ACommand Agent AEConn`
+
+```haskell
+data RatchetDesyncState
+  = RDResyncAllowed
+  | RDResyncRequired
+  | RDHealed
+```
+
+New field should be added to `ConnectionStats` - `ratchetDesyncState :: Maybe RatchetDesyncState`, based on `ratchet_desync_state`.
+
+> On `RDESYNC` events chat should create chat item, prompting ratchet re-synchronization or notifying it has healed.
+> If connection has diagnosed ratchet de-sync, chat item should have a button to start ratchet re-sync.
+> We'd have to get `ConnectionStats` on chat level for this instead of chat info.
+> This wouldn't work for groups. One option is to add `ConnectionStats` to `GroupMember` type and update on events.
+> Same could be done for `Contact` then.
+
+To consider - allow to start ratchet re-synchronization at any time regardless of this field as an experimental feature. In chat it could be behind "Developer tools" + additional "Experimental" toggle. Agent api would have `force :: Bool` as parameter, allowing to bypass `ratchet_desync_state`. Should `ratchet_resync_state` (see below) still be honored in this case?
+
+### Re-synchronization process
+
+\*\*\*\*\*
+
+Basic idea is the following:
+
+Both agents send new ratchet keys and compute a new shared secret. Agent that starts re-synchronization should record this fact in the connection state. Agent that receives a new key should respond with a key of its own, unless it has recorded that it itself started re-synchronization in the connection state.
+
+It can happen that both agents start re-synchronizing simultaneously. In this case they both would record it in the connection state and would not respond with a new message - instead they would use each other's already sent keys.
+
+Agent has both keys if:
+
+- It initiates with the first key, and then receives the second key;
+- It receives the first key and then generates its own in response.
+
+After agent has both keys, it initiates new ratchet depending on keys ordering. The agent that sent the lower key should use `initRcvRatchet` function, the agent that sent the greater key should use `initSndRatchet` (or vice versa - but they should deterministically choose different sides).
+
+\*\*\*\*\*
+
+State whether the ratchet re-synchronization is in progress should be tracked in database via `connections` table new `ratchet_resync_state` field.
+
+New functional api:
+
+```haskell
+resyncConnectionRatchet :: AgentErrorMonad m => AgentClient -> ConnId -> m ConnectionStats
+```
+
+or if we want to allow re-synchronizing ratchet at any time even if de-synchronization wasn't diagnosed:
+
+```haskell
+resyncConnectionRatchet :: AgentErrorMonad m => AgentClient -> ConnId -> Bool -> m ConnectionStats
+resyncConnectionRatchet c connId force = ...
+```
+
+Possibly client command?
+
+``` haskell
+data ACommand (p :: AParty) (e :: AEntity) where
+  ...
+  RESYNC_RATCHET :: Bool -> ACommand Client AEConn
+```
+
+New event - `RRESYNC :: RatchetResyncState -> ConnectionStats -> ACommand Agent AEConn`
+
+```haskell
+data RatchetResyncState
+  = RRStarted
+  | RRAgreedSnd
+  | RRAgreedRcv
+  | RRComplete
+```
+
+New `ConnectionStats` field - `ratchetResyncState :: Maybe RatchetResyncState`.
+
+When called, it should:
+
+- Generate new keys.
+- Update database connection state.
+  - Set `ratchet_desync_state` to NULL.
+  - Set `ratchet_resync_state` to `RRStarted`.
+  - Delete old ratchet from `ratchets` (is it safe?), create new ratchet.
+- Send `AgentRatchetKey` message.
+- Return updated `ConnectionStats` to client.
+
+> On `RRESYNC` events chat should create chat item, and reset connection verification.
+> Parameterized `RRESYNC` allows to distinguish: start and end of re-synchronization for initiating party; chat item direction - `RRESYNC RRStarted` is snd, `RRESYNC RRAgreedSnd/Rcv` and `RRESYNC RRComplete` are rcv (`RRESYNC RRAgreedSnd/Rcv` chat item could be omitted).
+
+AgentRatchetKey is a new message on the level of AgentMsgEnvelope - encrypted with queue level e2e encryption, but not with connection level e2e encryption (since ratchet de-synchronized).
+
+```haskell
+data AgentMsgEnvelope
+  = ...
+  | AgentRatchetKey
+      { agentVersion :: Version,
+        e2eEncryption :: E2ERatchetParams 'C.X448,
+        info :: ByteString -- for extension
+      }
+```
+
+On receiving `AgentRatchetKey`, if the receiving client hasn't started the ratchet re-synchronization itself (check `ratchet_resync_state`), it should:
+
+- Generate new keys and compute new shared secret, initializing ratchet based on keys comparison.
+- Update database connection state.
+  - Set `ratchet_resync_state` to `RRAgreedSnd/Rcv` (depending on whether ratchet was initialized as sending or receiving).
+  - Delete old ratchet from `ratchets`, create new ratchet.
+- Reply with its own `AgentRatchetKey`.
+- Notify client with `RRESYNC RRAgreedSnd/Rcv`.
+- If ratchet was initialized as sending, send `EREADY` message, notifying other agent ratchet is re-synced.
+
+New agent message:
+
+```haskell
+data AMessage
+  = ...
+  | -- ratchet re-synchronization is complete, with last decrypted sender message id
+    EREADY PrevExternalSndId
+```
+
+On receiving `AgentRatchetKey`, if the receiving client started re-sync:
+
+- Compute new shared secret, initializing ratchet based on keys comparison.
+- Update database connection state.
+  - Set `ratchet_resync_state` to `RRAgreedSnd/Rcv` (depending on whether ratchet was initialized as sending or receiving).
+  - Update ratchet.
+- Notify client with `RRESYNC RRAgreedSnd/Rcv`.
+- If ratchet was initialized as sending, send `EREADY` message.
+
+After agent receives `EREADY` (or any other message that successfully decrypts):
+
+- Reset `ratchet_resync_state` to NULL.
+- Notify client with `RRESYNC RRComplete`.
+- If ratchet was initialized as receiving, send reply `EREADY` message.
+
+### State transitions
+
+For initiating party:
+
+```
+    +------------+
+    | Ratchet ok |
+    +------------+
+          |
+          | message received, decryption error
+   * ---->|-----------------------------------+
+          |                                   |
+          V          new (clarifying)         V
+  +-----------------+     error      +------------------+
+  | Re-sync allowed |--------------->| Re-sync required |
+  +-----------------+                +------------------+
+          |                                   |
+          |-----------------------------------|
+          |                                   | alternative - message received,
+          | re-sync started by client         | successfully decrypted
+          V                                   V
+  +-----------------+                   +------------+
+  | Re-sync started |                   | Ratchet ok |
+  +-----------------+                   +------------+
+          |
+          | other party replied with new ratchet key
+          V
+  +----------------+
+  | Re-sync agreed |----> * message received, decryption error
+  |   snd / rcv    |        (should remember agreed state for reply EREADY?)
+  +----------------+
+          |
+          | message received, successfully decrypted
+          | (can be, but not necessarily, EREADY)
+          V
+    +------------+
+    | Ratchet ok |
+    +------------+
+```
+
+For replying party:
+
+```
+    +------------+
+    | Ratchet ok |
+    +------------+
+          |
+          | other party sent new ratchet key
+          V
+  +----------------+
+  | Re-sync agreed |
+  |   snd / rcv    |
+  +----------------+
+          |
+          | message received, successfully decrypted
+          | (can be, but not necessarily, EREADY)
+          V
+    +------------+
+    | Ratchet ok |
+    +------------+
+```
+
+### Ratchet state model
+
+#### 2 state variables
+
+Above we considered model with separate de-sync and re-sync state.
+
+| Desync \ Resync      | Nothing | RRStarted | RRAgreedSnd | RRAgreedRcv |
+| ---                  |  :---:  |   :---:   |    :---:    |    :---:    |
+| **Nothing**          | 1       | 3         | 4           | 4           |
+| **RDResyncAllowed**  | 2       |           | 5           | 5           |
+| **RDResyncRequired** | 2       |           | 5           | 5           |
+
+1: Ratchet is ok.
+
+2: Re-sync diagnosed, not in progress.
+
+3: Rs-sync started, diagnosing de-sync is prohibited.
+
+4: Re-sync agreed.
+
+5: Re-sync agreed, new de-sync is diagnosed.
+
+Combination 5 is possible in case de-sync was diagnosed before message that could be decrypted is received, for example if `EREADY` failed to deliver and no other decryptable message followed. We shouldn't prohibit diagnosing de-sync in this case, because agent may never exit "Agreed" state (if new decryptable message is never received). We also shouldn't overwrite/forget state of re-sync, even if we diagnose new possible de-sync, because if the decryptable `EREADY` is received and ratchet is in `RRAgreedRcv` state, it should respond with reply `EREADY`.
+
+Some combinations should be impossible:
+
+  - `RDResyncAllowed` with `RRStarted`.
+
+  - `RDResyncRequired` with `RRStarted`.
+
+`RDHealed` is equivalent to `Nothing` and only used for `RDESYNC` event, `Maybe RatchetDesyncState` can be replaced with `RatchetDesyncState`, with single new constructor `RDNoDesync` replacing `Nothing` and `RDHealed`.
+
+`RRComplete` is equivalent to `Nothing` and only used for `RRESYNC` event, `Maybe RatchetResyncState` can be replaced with `RatchetResyncState`, with single new constructor `RDNoResync` replacing `Nothing` and `RRComplete`.
+
+#### Single state variable
+
+Another option is two have a single state variable describing ratchet.
+
+```haskell
+data RatchetState
+  = RSOk
+  | RSResyncAllowed
+  | RSResyncRequired
+  | RSResyncStarted
+  | RRResyncAgreedSnd
+  | RRResyncAgreedRcv
+
+-- When `resyncConnectionRatchet` is not prohibited. Can override with `force`.
+-- Currently we check:`(isJust ratchetDesyncState || force) && ratchetResyncState /= Just RRStarted`.
+resyncConnectionRatchetAllowed :: RatchetState -> Bool
+resyncConnectionRatchetAllowed = \case
+  RSOk -> False
+  RSResyncAllowed -> True
+  RSResyncRequired -> True
+  RSResyncStarted -> False -- `force` shouldn't override
+  RRResyncAgreedSnd -> False
+  RRResyncAgreedRcv -> False
+
+-- When we register and notify about ratchet de-synchronization.
+-- Currently we check: `(isNothing ratchetDesyncState && ratchetResyncState /= Just RRStarted)`.
+-- We should also allow to update from Allowed to Required.
+shouldNotifyRDESYNC :: RatchetState -> Bool
+shouldNotifyRDESYNC = \case
+  RSOk -> True
+  RSResyncAllowed -> False -- only if new error implies Required
+  RSResyncRequired -> False
+  RSResyncStarted -> False
+  RRResyncAgreedSnd -> True
+  RRResyncAgreedRcv -> True
+
+-- When we prohibit connection switch, for `checkRatchetDesync`.
+-- Currently we check: `(ratchetDesyncState == Just RDResyncRequired || ratchetResyncState == Just RRStarted)`
+-- Also use in `runSmpQueueMsgDelivery` to pause delivery?
+ratchetDesynced :: RatchetState -> Bool
+ratchetDesynced = \case
+  RSOk -> False
+  RSResyncAllowed -> False
+  RSResyncRequired -> True
+  RSResyncStarted -> True
+  RRResyncAgreedSnd -> False
+  RRResyncAgreedRcv -> False
+```
+
+Having a single state variable limits differentiation described for combination 5 in matrix. It also limits possible differentiations in client between events when ratchet is healed on its own, and when ratchet re-sync is completed after agents negotiation. Overall, since matrix is not very sparse and allows for more fine-grained decision-making, having separate state variables for de-sync and re-sync seems preferred.
+
+#### Single state variable simplified (final version)
+
+```haskell
+data RatchetSyncState
+  = RSOk
+  | RSAllowed
+  | RSRequired
+  | RSStarted
+  | RSAgreed
+
+-- event
+RSYNC :: RatchetSyncState -> ConnectionStats -> ACommand Agent AEConn`
+
+-- ConnectionStats field
+ratchetSyncState :: RatchetSyncState
+```
+
+Updated design decisions:
+
+1. Single constructor for "Agreed" state. Differentiating `RRResyncAgreedSnd` and `RRResyncAgreedRcv` allowed for easier processing of `EREADY` by helping to determine whether reply `EREADY` has to be sent. However, it duplicated information already present in ratchet's state, and can be instead worked around by remembering and analyzing ratchet state pre decryption.
+
+2. Prohibit transition from "Agreed" state to "Desync" states. This would make possible edge-cases that leave ratchet in de-synchronized state without ability to progress (e.g. failed delivery of `AgentRatchetKey`), but would simplify state machine by removing dedicated "Desync" variable. Besides, there's still a recovery way with a `force` option.
+
+3. Treat "Agreed" as unfinished state - prohibit new messages to be enqueued, etc. Reception of any decryptable message transitions ratchet to "Ok" state.
+
+Possible improvements:
+
+- Repeatedly triggering re-synchronization while in "Started"/"Agreed" state re-sends same keys and EREADY.
+- Cooldown period, during which repeat re-synchronization is prohibited.
+
+### Skipped messages
+
+Options:
+
+1. Ignore skipped messages.
+2. Stop sending new messages while connection re-synchronizes (can use `ratchet_resync_state`).
+
+- Initiator shouldn't send new messages until receives `AgentRatchetKey` from second party.
+- Second party knows new shared secret immediately after processing first `AgentRatchetKey`, so it's not necessary to limit?
+
+3. 2 + Re-send skipped messages first.
+
+- Add `last_external_snd_msg_id` to `AgentRatchetKey`? + see link above
+
+4. Re-send only messages after the latest ratchet step. \*
+
+It may be okay to ignore skipped messages, or at most implement option 2, as ratchet de-synchronization is usually caused by misuse (human error) - the most common cause of ratchet de-sync seems to be sending and receiving messages after running agent with old database backup. In this case user has already seen most skipped messages, and it can be expected to not have them after switching to an old backup. So in this case the only "really skipped" messages are those that were sent during the latest ratchet step and failed to decrypt, triggering ratchet re-sync (\* another option is to only re-send those).
+
+Besides, depending on time of backup there may be an arbitrary large number of skipped messages, which may consume a lot of traffic and may halt delivery of up-to-date messages for some time.
+
+It may be better to have request for repeat delivery as a separate feature, that can be requested in necessary contexts - for example for group stability.
+
+Can servers delivery failure lead to de-sync? If message is lost on server and never delivered, ratchet wouldn't advance, so there's no room for de-sync? If yes, re-evaluate.

rfcs/2023-09-12-second-relays.md

@@ -0,0 +1,240 @@
+# Protecting IP addresses of the users from their contacts
+
+## Problem
+
+SMP protocol relays are chosen and can be controlled by the message recipients. It means that the recipients can find out IP addresses of message senders by modifying SMP relay code (or by using proxies and timing correlation), unless the senders use VPN or some overlay network. Tor is an audequate solution in most cases to mitigate it, but it requires additional technical knowledge to install and configure (even installing Orbot on Android is seen as "complex" by many users), and reduces usability because of higher latency.
+
+The lack of in-built IP address protection is the main concern of many users, particularly given that most people do not realise that it is lacking by default - without transport protection SimpleX is not perceived as a "whole product".
+
+Similarly, XFTP protocol relays are chosen by senders, and they can be used to detect file recipients' IP addresses.
+
+## Possible solutions
+
+1. Embed Tor in the client.
+
+Pros:
+- no changes in the protocols/servers
+- acceptable threat model for most users
+- removes complexity of installing and configuring Tor
+- probably, the most attractive option for Tor users
+
+Cons:
+- higher latency and error rate
+- higher resource usage
+- requires us updating Tor client regularly
+- restriction on Tor usage in some networks, so it would require supporting bridges in the app UI
+- legislative restrictions on Tor usage, so it may require supporting multiple app versions, and won't solve the problem where Tor is not embedded
+
+2. Thin clients with hosted accounts.
+
+Host some part of the current client on the server and have another part running on the devices.
+
+Pros:
+- substantially reduces the traffic for mobile clients
+- probably the most attractive option for mass market users who expect to have an account on the server and be able to access it from multiple devices and via web.
+
+Cons:
+- a lot of design and development
+- makes accounts visible to the server
+- the new protocol design to protect connections graph and message content from the hosting server.
+- quite different threat model
+
+Overall, this is not a viable or even appropriate option for the current stage.
+
+3. SMP / XFTP proxy.
+
+Introduce SMP and XFTP protocol extenstions to allow message senders and file recipients to delegate the tasks of sending messages and receiving files to the proxies, so that peer-chosen relays can only observe IP addresses of the proxies and not of the users.
+
+Pros:
+- no dependency on and lower latency than via Tor
+- reduces client traffic, both due to retries handled by these proxies and due to the smaller number of connections to the peer-chosen relays. The flip side is that additional commands (and traffic) are needed to create the sessions with relays.
+- higher transport privacy: protects IP addresses from peers and makes transport correlation harder.
+- additional function: retrying pending messages, while the queue is not secured yet, without extra traffic for the clients.
+- improves current threat model by preventing session-based correlation of the traffic by the destination relay (assuming no information sharing with the proxies), as there will be one session between proxy and destination relay, mixing traffic from multiple users.
+
+Cons:
+- design and development cost
+- can undermine delivery stability
+
+This seems an attractive option, both from technical (reasonable complexity) and positioning (moving SimpleX closer to mix networks, while still avoiding server list visibility) points of view, and it is in the middle between embedding Tor, which would make the product more niche and has downsides and development costs too, and thin clients, which is 10x+ more work, and also creates a different threat model, that is not very attractive for the current stage and users.
+
+Below considers this design.
+
+## SMP/XFTP proxy design
+
+### Design requirements
+
+1. To avoid increasing the complexity of self-hosting, we should not create additional server types, and existing SMP and XFTP servers should be extended to provide proxy functions.
+
+2. SMP proxy should not be able to observe queue addresses and their count on the destination relays. This requirement is not needed for XFTP proxies, as each file chunk is downloaded only once, so there is no need to hide its address.
+
+3. There must be no identifiers and cyphertext in common in outgoing and incoming traffic inside TLS (the current designs have this quality).
+
+4. Traffic between the client and destination relays must be e2e encrypted, with MITM-by-proxy mitigated, relying on the relay identity (certificate fingerprint), ideally without any additional fingerprint in relay address.
+
+5. SMP proxy should implement retry logic and hold messages while they are delivered. They also should return relay replies to the client. To avoid any additional traffic the client should just add "sent to proxy" status and only show "sent" once proxy returns the response from the destination relay - there should be no additional response from the proxy confirming acceptance to delivery.
+
+This would also reduce the difference in how the traffic looks to the observer - sending via proxy may look similar to sending to the usual server (which can be further supported by friendly destination relays that could add latency for direct requests and reply quickly when response came via proxy and be undermined by hostile relays that would introduce some latency pattern to help traffic correlation. The latter problem can be mitigated by having a fixed response latency from proxy that may be "come back later for destination response").
+
+6. Sending messages to groups have to be batched in the client to avoid multiple requests for destination relay sessions - such requests can be batched to proxy, even though it leaks _some_ metadata - which destination relays are used by a given sender's IP address, it also reduces the overhead from proxies – it could be an option based on the privacy slider.
+
+6. SMP proxy may also increase utility and privacy of the platform:
+
+- holding messages and retrying them for the new connections while the receiving queue is not secured yet.
+- add delays in message delivery to make traffic correlation harder (same can be done in SMP relays).
+
+### Implementation considerations
+
+1. Block size. Possibly, there is not enough spare capacity in the current 16kb block to fit additional headers, any necessary metadata and encryption authentication tags, in which case we cannot send SMP and SMP-proxy traffic in the same transport connection (in case the same server plays both roles). In which case we will need to negotiate role during connection handshake and define a different (sub-)protocol for SMP-proxy.
+
+2. To be decided if we see it as extension of SMP protocol or as another protocol, irrespective of whether it's provided by the same or another server. Given different roles and block size it may be simpler to see it as a separate protocol, and which protocol is provided in the connection is determined during handshake.
+
+3. We probably should aim to avoid changing agent/client logic and see it instead as transport concern that can be dynamically decided at a point of sending a message, based on the current configuration.
+
+4. Configuration should probably allow to choose between not using proxies (particularly, during testing, when it would be the default), using proxies only for unknown relays, and using proxies for all relays (extra traffic, but more complex transport correlation - although randomizing this choice can be more beneficial to the transport privacy). The clients can aim to use proxy from another provider, to reduce the risks of sharing the information.
+
+### SMP-proxy protocol
+
+The flow of the messages will be:
+
+1. Client requests proxy to create session with the relay by sending `server` command with the SMP relay address and optional proxy basic AUTH (below). It should be possible to batch multiple session requests into one block, to reduce traffic.
+
+2. Proxy connects to SMP relay, negotiating a shared secret in the handshake that will be used to encrypt all sender blocks inside TLS (proxy-relay encryption). SMP relay also returns in handshake its temporary DH key to agree e2e encryption with the client (sender-relay encryption, to hide metadata sent to the destination relay from proxy).
+
+3. Proxy replies with `server_id` command including relay session ID to identify it in further requests, relay DH key for e2e encryption with the client - this key is signed with the TLS online private key associated with the certificate (its fingerprint is included in the relay address), and the TLS session ID between proxy and relay (this session ID must be used in transmissions, to mitigate replay attacks as before).
+
+A possible attack here is that proxy can use this TLS session to replay commands received from the client. Possibly, it could be mitigated with a bloom filter per proxy/SMP relay connection that would reject the repeated DH keys (that need to be used for replay), and also with DH key expiration (this mitigation should allow some acceptable rate of false positives from the bloom filter).
+
+With 32 bits per key there will be ~1/1,000,000 false positives (see https://en.wikipedia.org/wiki/Bloom_filter), and the filter would use ~32mb per each proxy connection if we reset relay key every 1000000 messages or more frequently. Given that the only commands accepted via relay would be SEND, replaying it would be interpreted by the receiving client as duplicate message delivery, so a small number of replayed messages won't cause any problems. But without mitigation it could be used to flood the receiving SMP relay queues with repeated messages, effectively causing DoS on these queues, even in case when SMP relays require basic auth to create queues.
+
+Given that the client chooses proxy it has some trust to, maybe this replay attack risk can be accepted.
+
+It is important that the same public key from destination relay is returned to all clients, so proxy does not need to repeat this request to know relays while the key did not expire, as using different keys for different clients would allow destination relays to correlate requests to the clients. A proxy that colludes with the destination relay can pass different public keys to the same client, but it is not changing threat model as colluding proxy can share information as well. It is also important that the client uses a new random key for each command, as using the same key would allow the destination relay to identify these commands as comming from the same user, and using a different key for each queue while would protect privacy of the user from the destination relay, would make it visible to the proxy how many different queues the client has on destination relay.
+
+*Unrelated cosideration for SMP protocol privacy improvement*: instead of signing commands to the destination relay, the sender could have a ratchet per queue agreed with the destination relay that would simply use authenticated encryption with per-message symmetric key to encrypt the message on the way to relay, and this encryption would be used as a proof of sender.
+
+4. Now the client sends `forward` to proxy, which it then forwards to SMP relay, applying additional encryption layer.
+
+5. SMP relay sends `response` to proxy applying additional encryption layer, which it then forwards to the client removing the additional encryption layer.
+
+Effectively it works as a simplified two-hop onion routing with the first relay (proxy) chosen by the sending client and the second relay chosen by the recipient, not only protecting senders' IP addresses from the recipients' relays, but also preventing recipients relays from correlating senders' traffic to different queues, as TLS session is owned by the proxy now and it mixes the traffic from multiple senders. To correlate traffic to users, proxy and relay would have to combine their information. SMP relays are still able to correlate traffic to receiving users via transport session.
+
+Sequence diagram for sending the message via SMP proxy:
+
+```
+-------------               -------------                     -------------              -------------
+|  sending  |               |    SMP    |                     |    SMP    |              | receiving |
+|  client   |               |   proxy   |                     |   relay   |              |  client   |
+-------------               -------------                     -------------              -------------
+     |       `server`             |                                 |                          |
+     | -------------------------> |   create TLS session, get keys  |                          |
+     |                            | ------------------------------> |                          |
+     |        `server_id`         |       (if doesn't exist)        |                          |
+     | <------------------------- |                                 |                          |
+     |                            |                                 |                          |
+     | TLS(F:s2r(SEND(e2e(msg)))) |                                 |                          |
+     | -------------------------> | TLS(F:p2r(s2r(SEND(e2e(msg))))) |                          |
+     |                            | ------------------------------> |                          |
+     |                            |                                 |                          |
+     |                            |     TLS(R:p2r(s2r(OK/ERR)))     |                          |
+     |     TLS(R:s2r(OK/ERR))     | <------------------------------ |                          |
+     | <------------------------- |                                 | TLS(MSG(r2c(e2e(msg))))  |
+     |                            |                                 | -----------------------> |
+     |                            |                                 |                          |
+     |                            |                                 |        TLS(ACK)          |
+     |                            |                                 | <----------------------- |
+     |                            |                                 |                          |
+     |                            |                                 |                          |
+
+```
+
+Below diagram shows the encrypttion layers for `forward` and `response` commands:
+
+- s2r (added) - encryption between client and SMP relay, with relay key returned in server_id command, with MITM by proxy mitigated by verifying the certificate fingerprint included in the relay address.
+- e2e (exists now) - end-to-end encryption per SMP queue, with double ratchet e2e encryption inside it.
+- p2r (added) - additional encryption between proxy and SMP relay with key agreed in the handshake, to mitigate traffic correlation inside TLS. This key could also be signed by the same certificate, if we don't want to rely on TLS security.
+- r2c (exists now) additional encryption between SMP relay and client to prevent traffic correlation inside TLS.
+
+```
+-----------------             -----------------  -- TLS --  -----------------             -----------------
+|               |  -- TLS --  |               |  -- p2r --  |               |  -- TLS --  |               |
+|               |  -- s2r --  |               |  -- s2r --  |               |  -- r2c --  |               |
+|    sending    |  -- e2e --  |               |  -- e2e --  |               |  -- e2e --  |   receiving   |
+|    client     |     MSG     |   SMP proxy   |     MSG     |   SMP relay   |     MSG     |    client     |
+|               |  -- e2e --  |               |  -- e2e --  |               |  -- e2e --  |               |
+|               |  -- s2r --  |               |  -- s2r --  |               |  -- r2c --  |               |
+|               |  -- TLS --  |               |  -- p2r --  |               |  -- TLS --  |               |
+-----------------             -----------------  -- TLS --  -----------------             -----------------
+```
+
+When proxy connects to SMP relay it would indicate in the handshake that it will use proxy protocol and the SMP relay would expect the same `forward` commands and reply with `response`s.
+
+Below syntax aims to fit in 16kb block using spare capacity in SMP protocol.
+
+```abnf
+proxy_block = padded(proxy_transmission, 16384)
+proxy_transmission = corr_id relay_session_id proxy_command
+corr_id = length *8 OCTET
+proxy_command = server / server_id / forward / response / error
+server = "S" address [relay_basic_auth] ; creates transport session between proxy and relay
+server_id = "I" relay_session_id tls_session_id signed_relay_key ;
+    ; session_id is the TLS session ID between proxy and relay, it has to be included inside encrypted block to prevent replay attacks
+forward = %s"F" random_dh_pub_key encrypted_block ; it's important that a new key is used for each command, to prevent any correlation by proxy or by destination relay
+response = %s"R" encrypted_block; response received from the destination SMP relay
+relay_session_id = length *8 OCTET
+error = %s"E" error
+```
+
+The overhead is: 1+8 (corrId) + 1+8 (relay_session_id) + 1 (command) + 1+32 (random_dh_pub_key) + 2 (original length) + 16 (auth tag for e2e encryption) + 16 (auth tag for proxy to relay encryption) = 86 bytes. The reserve for sent messages in SMP is ~84 bytes, so it should about fit with some reduced bytes somewhere.
+
+Another possible design is to allow mixing sent messages and normal SMP commands in the same transport connection, but it can make fitting in the block a bit harder, additional overhead would be: 1 (transmission count) + 2 (transmission size) + 1 (empty signature) = 4 bytes.
+
+The above assumes that the client can only send one message to an SMP relay and then has to wait for response before sending the next message. Missing the response would cause re-delivery (further improvement is possible when proxy detects these redelieveries and not send them to relays but simply reply with the same response).
+
+### Threat model for SMP proxy and changes to threat model for SMP
+
+#### SMP proxy
+
+*can:*
+
+- learn a user's IP address, as long as Tor is not used.
+
+- learn when a user with a given IP address is online.
+
+- know how many messages are sent from a given IP address and to a given destination SMP relay.
+
+- drop all messages from a given IP address or to a given destination relay.
+
+- unless SMP relay detects repeated public DH keys of senders, replay messages to a destination relay within a single session, causing either duplicate message delivery (which will be detected and ignored by the receiving clients), or, when receiving client is not connected to SMP relay, exhausting capacity of destination queues used within the session.
+
+*cannot:*
+
+- perform queue correlation (matching multiple queues to a single user), unless it has additional information from SMP relay.
+
+- undetectably add, duplicate, or corrupt individual messages.
+
+- undetectably drop individual messages, so long as a subsequent message is delivered.
+
+- learn the contents of messages.
+
+- learn the destination queues of messages.
+
+- distinguish noise messages from regular messages except via timing regularities.
+
+- compromise the user's end-to-end encryption with another user via an active attack.
+
+- compromise the user's end-to-end encryption with destination relay via an active attack.
+
+#### SMP relay accessed via SMP proxy
+
+*still can:*
+
+- perform recipients' queue correlation (matching multiple queues to a single recipient) via either a re-used transport connection, user's IP Address, or connection timing regularities
+
+*no longer can:*
+
+- perform senders' queue correlation (matching multiple queues to a single sender) via either a re-used transport connection, user's IP Address, or connection timing regularities, unless it has additional information from SMP proxy.
+
+- learn a sender's IP address, track them through other IP addresses they use to access the same queue, and infer information (e.g. employer) based on the IP addresses, even if Tor is not used.
+
+The rest of the threat model for SMP relays remains the same as in [overview](../protocol/overview-tjr.md#simplex-messaging-protocol-server).

rfcs/2023-10-25-remote-control.md

@@ -0,0 +1,361 @@
+# SimpleX Remote Control protocol
+
+Using profiles in SimpleX Chat mobile app from desktop app with minimal risk to the security/threat model of SimpleX protocols.
+
+## Problem
+
+Synchronizing profiles that use double ratchet for e2e encryption is effectively impossible in a way that tolerates partitioning between devices without reducing security of double ratchet.
+
+We are not considering replacing (or weakening) double ratchet to allow profile synchronization, as some other messengers did (e.g., Session). We are also not considering Signal model, when profile is known to the server and adding devices results in changing security code and no visibility of conversation history, as it would be substantially different from the current model, and also effectively weakens the security, as in practives most users don't revalidate security codes when they change.
+
+## Solution
+
+The proposed solution is a new remote access/control protocol, when the application on host device (usually mobile) acts as a server, and application on another device (usually desktop) acts as a controller usually running in the same local network.
+
+Existing service discovery and remote control protocols known to us are vulnerable to spoofing, spamming and MITM attacks. This design aims to solve these problems.
+
+## Requirements
+
+- Strong, cryptographically verified identity of the controller device, with the initial connection requiring out-of-band communication of public keys (QR code or link).
+- Identity verification of the host device during session handshake.
+- Protection against malicious "controllers" trying to make host connect to them instead of valid controller on the same network.
+- Protection against replay attacks, both during discovery and during control session.
+- Additional encryption layer inside TLS, with post-quantum algorithm in key agreement.
+- Protect host device from unauthorized access in case of controller compromise.
+- Post-compromised security - that is, even if long term secrets were copied from the controller, and host device was made to connect to malicious controller device, prevent malicious controller from accessing the host device data.
+- Allow general high-level interactions common for many applications:
+  - RPC pattern for commands executed by the host application.
+  - Events sent by host to update controller UI.
+  - Uploading and downloading files between host and controller, either to be processed by the host or to be presented in the controller UI.
+
+This design will allow application-level protocol that is quite close to how SimpleX Chat UI interacts with SimpleX Chat core - there is a similar RPC + events protocol and support for files. The application-level protocol is out of scope of this specification.
+
+## Protocol phases
+
+Protocol consists of four phases:
+- controller session invitation
+- establishing session TLS connection
+- session verification and protocol negotiation
+- session operation
+
+### Session invitation
+
+The invitation to the first session between host and controller pair MUST be shared out-of-band, to establish a long term identity keys/certificates of the controller to host device.
+
+The subsequent sessions can be announced via an application-defined site-local multicast group, e.g. `224.0.0.251` (also used in mDNS/bonjour) and an application-defined port (SimpleX Chat uses 5227).
+
+The session invitation contains this data:
+- supported version range for remote control protocol
+- application-specific information, e.g. device name, application name and supported version range, settings, etc.
+- session start time in seconds since epoch
+- if multicast is used, counter of announce packets sent by controller
+- network address (ipv4 address and port) of the controller
+- CA TLS certificate fingerprint of the controller - this is part of long term identity of the controller established during the first session, and repeated in the subsequent session announcements.
+- Session Ed25519 public key used to verify the announcement and commands - this mitigates the compromise of the long term signature key, as the controller will have to sign each command with this key first.
+- Long-term Ed25519 public key used to verify the announcement and commands - this is part of the long term controller identity.
+- Session X25519 DH key and sntrup761 KEM encapsulation key to agree session encryption (both for multicast announcement and for commands and responses in TLS), as described in https://datatracker.ietf.org/doc/draft-josefsson-ntruprime-hybrid/. The new keys are used for each session, and if client key is already available (from the previous session), the computed shared secret will be used to encrypt the announcement multicast packet. The out-of-band invitation is unencrypted. These DH public key and KEM encapsulation key are always sent unencrypted. NaCL Cryptobox is used for encryption.
+
+Host device decrypts (except the first session) and validates the invitation:
+- Session signature is valid.
+- Timestamp is within some window from the current time.
+- Long-term key signature is valid.
+- Long-term CA and signature key are the same as in the first session.
+- Some version in the offered range is supported.
+
+OOB session invitation is a URI with this syntax:
+
+```abnf
+sessionAddressUri = "xrcp://" encodedCAFingerprint "@" host ":" port "#/?" qsParams
+encodedCAFingerprint = base64url
+qsParams = param *("&" param)
+param = versionRangeParam / appInfoParam / sessionTsParam /
+        sessPubKeyParam / idPubKeyParam / kemEncKeyParam / dhPubKeyParam /
+        sessSignatureParam / idSignatureParam
+versionRangeParam = "v=" (versionParam / (versionParam "-" versionParam))
+versionParam = 1*DIGIT
+appInfoParam = "app=" escapedJSON ; optional
+sessionTsParam = "ts=" 1*DIGIT
+sessPubKeyParam = "skey=" base64url ; required
+idPubKeyParam = "idkey=" base64url ; required
+dhPubKeyParam = "dh=" base64url ; required
+sessSignatureParam = "ssig=" base64url ; required, signs the URI with this and idSignatureParam param removed
+idSignatureParam = "idsig=" base64url ; required, signs the URI with this param removed
+base64url = <base64url encoded binary> ; RFC4648, section 5
+```
+
+Multicast session announcement is a binary encoded packet with this syntax:
+
+```abnf
+sessionAddressPacket = dhPubKey nonce encrypted(unpaddedSize sessionAddress packetPad)
+dhPubKey = length x509encoded ; same as announced
+nonce = length *OCTET
+sessionAddress = largeLength sessionAddressUri ; as above
+length = 1*1 OCTET ; for binary data up to 255 bytes
+largeLength = 2*2 OCTET ; for binary data up to 65535 bytes
+packetPad = <pad packet size to 1450 bytes> ; possibly, we may need to move KEM agreement one step later,
+; with encapsulation key in HELLO block and KEM ciphertext in reply to HELLO.
+```
+
+### Establishing session TLS connection
+
+Host connects to controller via TCP session and validates CA credentials during TLS handshake. Controller acts as a TCP server in this connection, to avoid host device listening on a port, which might create an attack vector. During TLS handshake the controller's TCP server MUST present a self-signed two-certificate chain where the fingerprint of the first certificate MUST be the same as in the announcement.
+
+Host device presents its own client certificate chain with CA representing a long-term identity of the host device.
+
+### Session verification and protocol negotiation
+
+Once TLS session is established, both the host and controller device present a "session security code" to the user who must match them (e.g., visually or via QR code scan) and confirm on the host device. The session security code must be a digest of tlsunique channel binding. As it is computed as a digest of the TLS handshake for both the controller and the host, it will validate that the same TLS certificates are used on both sides, and that the same TLS session is established, mitigating the possibility of MITM attack in the connection.
+
+Once the session is confirmed by the user, the host device sends "hello" block to the controller. ALPN TLS extension is not used to obtain tlsunique prior to sending any packets.
+
+Block size should be 16384 bytes.
+
+Host HELLO must contain:
+- new session DH key - used to compute new shared secret with the controller keys from the announcement.
+- encrypted part of hello block (JSON object), containing:
+  - chosen protocol version.
+  - host CA TLS certificate fingerprint - part of host long term identity - must match the one presented in TLS handshake and the previous sessions, otherwise the connection is terminated.
+  - KEM encapsulation key - used to compute new shared secret for the session.
+  - additional application specific parameters, e.g host device name, application version, host settings or JSON encoding format.
+
+Hello block syntax:
+
+```abnf
+hostHello = unpaddedSize %s"HELLO " dhPubKey nonce encrypted(unpaddedSize hostHelloJSON helloPad) pad
+unpaddedSize = largeLength
+dhPubKey = length x509encoded
+pad = <pad block size to 16384 bytes>
+helloPad = <pad hello size to 12888 bytes>
+largeLength = 2*2 OCTET
+```
+
+Controller decrypts (including the first session) and validates the received hello block:
+- Chosen versions are supported (must be within offered ranges).
+- CA fingerprint matches the one presented in TLS handshake and the previous sessions - in subsequent sessions TLS connection should be rejected if the fingerprint is different.
+
+JTD schema for the encrypted part of host HELLO block `hostHelloJSON`: 
+
+```json
+{
+  "definitions": {
+    "version": {
+      "type": "string",
+      "metadata": {
+        "format": "[0-9]+"
+      }
+    },
+    "base64url": {
+      "type": "string",
+      "metadata": {
+        "format": "base64url"
+      }
+    }
+  },
+  "properties": {
+    "v": {"ref": "version"},
+    "ca": {"ref": "base64url"},
+    "kem": {"ref": "base64url"}
+  },
+  "optionalProperties": {
+    "app": {"properties": {}, "additionalProperties": true}
+  },
+  "additionalProperties": true
+}
+```
+
+Controller should reply with with `hello` or `error` response:
+
+```abnf
+ctrlHello = unpaddedSize %s"HELLO " kemCyphertext nonce encrypted(unpaddedSize ctrlHelloJSON helloPad) pad
+; ctrlHelloJSON is encrypted with the hybrid secret,
+; including both previously agreed DH secret and KEM secret from kemCyphertext
+unpaddedSize = largeLength
+kemCyphertext = largeLength *OCTET
+pad = <pad block size to 16384 bytes>
+helloPad = <pad hello size to 12888 bytes>
+largeLength = 2*2 OCTET
+
+ctrlError = unpaddedSize %s"ERROR " nonce encrypted(unpaddedSize ctrlErrorJSON helloPad) pad
+; ctrlErrorJSON is encrypted using previously agreed DH secret.
+```
+
+JTD schema for the encrypted part of controller HELLO block `ctrlHelloJSON`:
+
+```json
+{
+  "properties": {},
+  "additionalProperties": true
+}
+```
+
+JTD schema for the encrypted part of controller ERROR block `ctrlErrorJSON`:
+
+```json
+{
+  "properties": {
+    "message": {"type": "string"}
+  },
+  "additionalProperties": true
+}
+```
+
+Once controller replies HELLO to the valid host HELLO block, it should stop accepting new TCP connections.
+
+
+### Сontroller/host session operation
+
+The protocol for communication during the session is out of scope of this protocol.
+
+SimpleX Chat will use HTTP2 encoding, where host device acts as a server and controller acts as a client (these roles are reversed compared with TLS connection).
+
+Payloads in the protocol must be encrypted using NaCL cryptobox using the hybrid shared secret agreed during session establishment.
+
+Commands of the controller must be signed after the encryption using the controller's session and long term Ed25519 keys.
+
+tlsunique channel binding from TLS session MUST be included in commands (included in the signed body).
+
+The syntax for encrypted command and response body encoding:
+
+```
+commandBody = encBody sessSignature idSignature (attachment / noAttachment)
+responseBody = encBody attachment; should match counter in the command
+encBody = nonce encLength32 encrypted(tlsunique counter body)
+attachment = %x01 nonce encLength32 encrypted(attachment)
+noAttachment = %x00
+tlsunique = length 1*OCTET
+counter = 8*8 OCTET ; int64
+encLength32 = 4*4 OCTET ; uint32, includes authTag
+```
+
+If the command or response includes attachment, it's hash must be included in command/response and validated.
+
+## Key agreement for announcement packet and for session
+
+Initial announcement is shared out-of-band (URI with xrcp scheme), and it is not encrypted.
+
+This announcement contains only DH keys, as KEM key is too large to include in QR code, which are used to agree encryption key for host HELLO block. The host HELLO block will containt DH key in plaintext part and KEM encapsulation (public) key in encrypted part, that will be used to determine the shared secret (using SHA256 over concatenated DH shared secret and KEM encapsulated secret) both for controller HELLO response (that contains KEM cyphertext in plaintext part) and subsequent session commands and responses.
+
+During the next session the announcement is sent via encrypted multicast block. The shared key for this announcement and for host HELLO block is determined using the KEM shared secred from the previous session and DH shared secret computed using the host DH key from the previous session and the new controller DH key from the announcement.
+
+For the session, the shared secred is computed again using the KEM shared secret encapsulated by the controller using the new KEM key from the HOST hello block and DH shared secret computed using the host DH key from HELLO block and the new controller DH key from the announcement.
+
+To describe it in pseudocode:
+
+```
+// session 1
+hostHelloSecret(1) = dhSecret(1)
+sessionSecret(1) = sha256(dhSecret(1) || kemSecret(1)) // to encrypt session 1 data, incl. controller hello
+dhSecret(1) = dh(hostHelloDhKey(1), controllerInvitationDhKey(1))
+kemCiphertext(1) = enc(kemSecret(1), kemEncKey(1))
+// kemEncKey is included in host HELLO, kemCiphertext - in controller HELLO
+kemSecret(1) = dec(kemCiphertext(1), kemDecKey(1))
+
+// multicast announcement for session n
+announcementSecret(n) = sha256(dhSecret(n'))
+dhSecret(n') = dh(hostHelloDhKey(n - 1), controllerDhKey(n))
+
+// session n
+hostHelloSecret(n) = dhSecret(n)
+sessionSecret(n) = sha256(dhSecret(n) || kemSecret(n)) // to encrypt session n data, incl. controller hello
+dhSecret(n) = dh(hostHelloDhKey(n), controllerDhKey(n))
+// controllerDhKey(n) is either from invitation or from multicast announcement
+kemCiphertext(n) = enc(kemSecret(n), kemEncKey(n))
+kemSecret(n) = dec(kemCiphertext(n), kemDecKey(n))
+```
+
+If controller fails to store the new host DH key after receiving HELLO block, the encryption will become out of sync and the host won't be able to decrypt the next announcement. To mitigate it, the host should keep the last session DH key and also previous session DH key to try to decrypt the next announcement computing shared secret using both keys (first the new one, and in case it fails - the previous).
+
+To decrypt multicast announcement, the host should try to decrypt it using the keys of all known (paired) remote controllers.
+
+## Other options
+
+The proposed design has these pros/cons:
+
+Pros:
+- mobile host that has sensitive data doesn't act as TLS server.
+- multicast is optional - all sessions can happen via QR code only.
+
+Cons:
+- reversing of client/server roles between TLS and HTTP2.
+- in the first session mobile host TLS client credentials are verified after TLS connection is accepted.
+- cannot be used with host that runs in VM.
+
+The alternative design will use mobile host device as TLS server. The session negotiation process:
+
+- desktop shares its initial credentials via QR code, only for the first session
+- mobile sends encrypted multicast with session address, TLS CA fingerprint, DH key in clear text
+- desktop connects to mobile
+- session tlsunique presented to users on both devices - either user would have to confirm session on both devices or the mobile would have to send an additional "ready" block.
+
+Pros:
+- no reversing server role between TLS and HTTP2
+- TLS credentials are exchanged before TLS, so invalid credentials can be rejected during the handshake of the first session.
+- if some other way to pass data from host to controller is added, then it can be used with host running in VM.
+
+Cons:
+- multicast is mandatory, as there is no efficient way to communicate from mobile to desktop.
+- still needs hello or confirmation on both devices
+- mobile is now acting as TLS server creating additional attack vector
+
+In both proposed and alternative design mobile host has chat data, acts as HTTP2 server, commands are signed with desktop key presented out-of–band, and both commands and responses are encrypted inside TLS session.
+
+Other considered options:
+- SSH - more work integrating it.
+- use SMP connection to negotiate TLS session - it seems wrong to require Internet connection to negotiate a local network connection between desktop and mobile.
+- other multicast service discovery protocols - quite insecure.
+
+## Threat model
+
+#### A passive network adversary able to monitor the site-local traffic:
+
+*can:*
+- observe session times, duration and volume of the transmitted data between host and controller.
+
+*cannot:*
+- observe the content of the transmitted data.
+- substitute the transmitted commands or responses.
+- replay transmitted commands or events from the hosts.
+
+#### An active network adversary able to intercept and substitute the site-local traffic:
+
+*can:*
+- prevent host and controller devices from establishing the session
+
+*cannot:*
+- same as passive adversary, provided that user visually verified session code out-of-band.
+
+#### An active adversary with the access to the network:
+
+*can:*
+- spam controller device.
+
+*cannot:*
+- compromise host or controller devices.
+
+#### An active adversary with the access to the network who also observed OOB announcement:
+
+*can:*
+- connect to controller instead of the host.
+- present incorrect data to the controller.
+
+*cannot:*
+- connect to the host or make host connect to itself.
+
+#### Compromised controller device:
+
+*can:*
+- observe the content of the transmitted data.
+- access any data of the controlled host application, within the capabilities of the provided API.
+
+*cannot:*
+- access other data on the host device.
+- compromise host device.
+
+#### Compromised host device:
+
+*can:*
+- present incorrect data to the controller.
+- incorrectly interpret controller commands.
+
+*cannot:*
+- access controller data, even related to this host device.

scripts/docker/entrypoint-xftp-server

@@ -19,7 +19,9 @@ if [ ! -f "${confd}/file-server.ini" ]; then
   # Set quota
   case "${QUOTA}" in
     '') printf 'Please specify $QUOTA environment variable.\n'; exit 1 ;;
-    *) set -- "$@" --quota "${QUOTA}" ;;
+    *GB) QUOTA="$(printf ${QUOTA} | tr '[:upper:]' '[:lower:]')"; set -- "$@" --quota "${QUOTA}" ;;
+    *gb) set -- "$@" --quota "${QUOTA}" ;;
+    *) printf 'Wrong format. Format should be: 1gb, 10gb, 100gb.\n'; exit 1 ;;
   esac
 
   # Init the certificates and configs

scripts/main/simplex-servers-stopscript

@@ -0,0 +1,30 @@
+#!/usr/bin/env sh
+set -eu
+
+path_conf_var="/var/opt"
+path_conf_smp="$path_conf_var/simplex"
+path_conf_xftp="$path_conf_var/simplex-xftp"
+path_conf_storelog_smp="$path_conf_smp/smp-server-store.log"
+path_conf_storelog_xftp="$path_conf_xftp/file-server-store.log"
+date="$(date -u '+%Y-%m-%dT%H:%M:%S')"
+
+backup_smp() {
+ if [ -e "$path_conf_storelog_smp" ]; then
+  cp "$path_conf_storelog_smp" "${path_conf_storelog_smp}.${date:-date-failed}"
+ fi
+}
+
+backup_xftp() {
+ if [ -e "$path_conf_storelog_xftp" ]; then
+  cp "$path_conf_storelog_xftp" "${path_conf_storelog_xftp}.${date:-date-failed}"
+ fi
+}
+
+if [ "$1" = 'smp-server' ]; then
+ backup_smp
+elif [ "$1" = 'xftp-server' ]; then
+ backup_xftp
+else
+ backup_smp
+ backup_xftp
+fi

scripts/main/simplex-servers-uninstall

@@ -13,6 +13,11 @@ fi
 printf "${RED}This action will permanently remove all configs, directories, binaries from Installation Script. Please backup any relevant configs if they are needed.${NC}\n\nPress ${GRN}ENTER${NC} to continue or ${RED}Ctrl+C${NC} to cancel installation"
 read ans
 
-rm -rf /var/opt/simplex /etc/opt/simplex /var/opt/simplex-xftp /etc/opt/simplex-xftp /srv/xftp /etc/systemd/system/smp-server.service /etc/systemd/system/xftp-server.service /usr/local/bin/smp-server /usr/local/bin/xftp-server /usr/local/bin/simplex-servers-update /usr/local/bin/simplex-servers-uninstall && userdel smp && userdel xftp
+systemctl stop smp-server
+systemctl stop xftp-server
+
+rm -rf /var/opt/simplex /etc/opt/simplex /var/opt/simplex-xftp /etc/opt/simplex-xftp /srv/xftp /etc/systemd/system/smp-server.service /etc/systemd/system/xftp-server.service /usr/local/bin/smp-server /usr/local/bin/xftp-server /usr/local/bin/simplex-servers-update /usr/local/bin/simplex-servers-uninstall
+
+userdel smp && userdel xftp
 
 printf "Uninstallation is complete! Thanks for trying out SimpleX!\n"

scripts/main/simplex-servers-update

@@ -11,6 +11,7 @@ scripts_systemd_smp="$scripts/smp-server.service"
 scripts_systemd_xftp="$scripts/xftp-server.service"
 scripts_update="$scripts/simplex-servers-update"
 scripts_uninstall="$scripts/simplex-servers-uninstall"
+scripts_stopscript="$scripts/simplex-servers-stopscript"
 
 # Default installation paths
 path_bin="/usr/local/bin"
@@ -18,6 +19,7 @@ path_bin_smp="$path_bin/smp-server"
 path_bin_xftp="$path_bin/xftp-server"
 path_bin_update="$path_bin/simplex-servers-update"
 path_bin_uninstall="$path_bin/simplex-servers-uninstall"
+path_bin_stopscript="$path_bin/simplex-servers-stopscript"
 
 path_systemd="/etc/systemd/system"
 path_systemd_smp="$path_systemd/smp-server.service"
@@ -27,6 +29,7 @@ path_systemd_xftp="$path_systemd/xftp-server.service"
 path_tmp_bin="$(mktemp -d)"
 path_tmp_bin_update="$path_tmp_bin/simplex-servers-update"
 path_tmp_bin_uninstall="$path_tmp_bin/simplex-servers-uninstall"
+path_tmp_bin_stopscript="$path_tmp_bin/simplex-servers-stopscript"
 path_tmp_systemd_smp="$path_tmp_bin/smp-server.service"
 path_tmp_systemd_xftp="$path_tmp_bin/xftp-server.service"
 
@@ -43,8 +46,9 @@ remote_version="$(curl --proto '=https' --tlsv1.2 -sSf -L https://api.github.com
 update_scripts() {
   curl --proto '=https' --tlsv1.2 -sSf -L "$scripts_update" -o "$path_tmp_bin_update" && chmod +x "$path_tmp_bin_update"
   curl --proto '=https' --tlsv1.2 -sSf -L "$scripts_uninstall" -o "$path_tmp_bin_uninstall" && chmod +x "$path_tmp_bin_uninstall"
+  curl --proto '=https' --tlsv1.2 -sSf -L "$scripts_stopscript" -o "$path_tmp_bin_stopscript" && chmod +x "$path_tmp_bin_stopscript"
 
-  if diff -q "$path_bin_uninstall" "$path_tmp_bin_uninstall" > /dev/null; then
+  if diff -q "$path_bin_uninstall" "$path_tmp_bin_uninstall" > /dev/null 2>&1; then
     printf -- "- ${YLW}Uninstall script is up-to-date${NC}.\n"
     rm "$path_tmp_bin_uninstall"
   else
@@ -52,7 +56,15 @@ update_scripts() {
     mv "$path_tmp_bin_uninstall" "$path_bin_uninstall"
     printf "${GRN}Done!${NC}\n"
   fi
-  if diff -q "$path_bin_update" "$path_tmp_bin_update" > /dev/null; then
+  if diff -q "$path_bin_stopscript" "$path_tmp_bin_stopscript" > /dev/null 2>&1; then
+    printf -- "- ${YLW}Stopscript script is up-to-date${NC}.\n"
+    rm "$path_tmp_bin_stopscript"
+  else
+    printf -- "- Updating stopscript script..."
+    mv "$path_tmp_bin_stopscript" "$path_bin_stopscript"
+    printf "${GRN}Done!${NC}\n"
+  fi
+  if diff -q "$path_bin_update" "$path_tmp_bin_update" > /dev/null 2>&1; then
     printf -- "- ${YLW}Update script is up-to-date${NC}.\n"
     rm "$path_tmp_bin_update"
   else
@@ -68,7 +80,7 @@ update_systemd() {
   curl --proto '=https' --tlsv1.2 -sSf -L "$scripts_systemd_smp" -o "$path_tmp_systemd_smp"
   curl --proto '=https' --tlsv1.2 -sSf -L "$scripts_systemd_xftp" -o "$path_tmp_systemd_xftp"
 
-  if diff -q "$path_systemd_smp" "$path_tmp_systemd_smp" > /dev/null; then
+  if diff -q "$path_systemd_smp" "$path_tmp_systemd_smp" > /dev/null 2>&1; then
     printf -- "- ${YLW}smp-server service is up-to-date${NC}.\n"
     rm "$path_tmp_systemd_smp"
   else
@@ -77,7 +89,7 @@ update_systemd() {
     systemctl daemon-reload
     printf "${GRN}Done!${NC}\n"
   fi
-  if diff -q "$path_systemd_xftp" "$path_tmp_systemd_xftp" > /dev/null; then
+  if diff -q "$path_systemd_xftp" "$path_tmp_systemd_xftp" > /dev/null 2>&1; then
     printf -- "- ${YLW}xftp-server service is up-to-date${NC}.\n"
     rm "$path_tmp_systemd_xftp"
   else
@@ -95,16 +107,16 @@ update_bins() {
       systemctl stop smp-server
       printf "${GRN}Done!${NC}\n"
 
-      print -- "- Updating smp-server bin to %s..." "$remote_version"
-      curl --proto '=https' --tlsv1.2 -sSf -L "$bin_smp" -o "$bin_path_smp" && chmod +x "$bin_path_smp"
+      printf -- "- Updating smp-server bin to %s..." "$remote_version"
+      curl --proto '=https' --tlsv1.2 -sSf -L "$bin_smp" -o "$path_bin_smp" && chmod +x "$path_bin_smp"
       printf "${GRN}Done!${NC}\n"
       
       printf -- "- Starting smp-server service..."
-      systemctl stop smp-server
+      systemctl start smp-server
       printf "${GRN}Done!${NC}\n"
     else
-      print -- "- Updating smp-server bin..."
-      curl --proto '=https' --tlsv1.2 -sSf -L "$bin_smp" -o "$bin_path_smp" && chmod +x "$bin_path_smp"
+      printf -- "- Updating smp-server bin..."
+      curl --proto '=https' --tlsv1.2 -sSf -L "$bin_smp" -o "$path_bin_smp" && chmod +x "$path_bin_smp"
       printf "${GRN}Done!${NC}\n"
     fi
 
@@ -113,16 +125,16 @@ update_bins() {
       systemctl stop xftp-server
       printf "${GRN}Done!${NC}\n"
 
-      print -- "- Updating xftp-server bin to %s..." "$remote_version"
-      curl --proto '=https' --tlsv1.2 -sSf -L "$bin_xftp" -o "$bin_path_xftp" && chmod +x "$bin_path_xftp"
+      printf -- "- Updating xftp-server bin to %s..." "$remote_version"
+      curl --proto '=https' --tlsv1.2 -sSf -L "$bin_xftp" -o "$path_bin_xftp" && chmod +x "$path_bin_xftp"
       printf "${GRN}Done!${NC}\n"
       
       printf -- "- Starting xftp-server service..."
-      systemctl stop xftp-server
+      systemctl start xftp-server
       printf "${GRN}Done!${NC}\n"
     else
-      print -- "- Updating xftp-server bin..."
-      curl --proto '=https' --tlsv1.2 -sSf -L "$bin_smp" -o "$bin_path_xftp" && chmod +x "$bin_path_xftp"
+      printf -- "- Updating xftp-server bin..."
+      curl --proto '=https' --tlsv1.2 -sSf -L "$bin_smp" -o "$path_bin_xftp" && chmod +x "$path_bin_xftp"
       printf "${GRN}Done!${NC}\n"
     fi
   else
@@ -155,6 +167,8 @@ main() {
   
   printf "Updating simplex server binaries...\n"
   update_bins
+
+  rm -rf "$path_tmp_bin"
 }
 
 main "$@"

scripts/main/smp-server.service

@@ -6,7 +6,7 @@ User=smp
 Group=smp
 Type=simple
 ExecStart=/usr/local/bin/smp-server start +RTS -N -RTS 
-ExecStopPost=/usr/bin/env sh -c '[ -e "/var/opt/simplex/smp-server-store.log" ] && cp "/var/opt/simplex/smp-server-store.log" "/var/opt/simplex/smp-server-store.log.$(date +%FT%T)'
+ExecStopPost=/usr/local/bin/simplex-servers-stopscript smp-server
 LimitNOFILE=65535
 KillSignal=SIGINT
 TimeoutStopSec=infinity

scripts/main/xftp-server.service

@@ -6,7 +6,7 @@ User=xftp
 Group=xftp
 Type=simple
 ExecStart=/usr/local/bin/xftp-server start +RTS -N -RTS
-ExecStopPost=/usr/bin/env sh -c '[ -e "/var/opt/simplex-xftp/file-server-store.log" ] && cp "/var/opt/simplex-xftp/file-server-store.log" "/var/opt/simplex-xftp/file-server-store.log.$(date +%FT%T)'
+ExecStopPost=/usr/local/bin/simplex-servers-stopscript xftp-server
 LimitNOFILE=65535
 KillSignal=SIGINT
 TimeoutStopSec=infinity

simplexmq.cabal

@@ -1,11 +1,11 @@
 cabal-version: 1.12
 
--- This file has been generated from package.yaml by hpack version 0.35.1.
+-- This file has been generated from package.yaml by hpack version 0.35.0.
 --
 -- see: https://github.com/sol/hpack
 
 name:           simplexmq
-version:        5.1.1
+version:        5.5.0.6
 synopsis:       SimpleXMQ message broker
 description:    This package includes <./docs/Simplex-Messaging-Server.html server>,
                 <./docs/Simplex-Messaging-Client.html client> and
@@ -26,16 +26,23 @@ build-type:     Simple
 extra-source-files:
     README.md
     CHANGELOG.md
+    cbits/sha512.h
+    cbits/sntrup761.h
 
 flag swift
   description: Enable swift JSON format
   manual: True
   default: False
 
+flag use_crypton
+  description: Use crypton etc. in cryptostore
+  manual: True
+  default: True
+
 library
   exposed-modules:
-      Simplex.FileTransfer
       Simplex.FileTransfer.Agent
+      Simplex.FileTransfer.Chunks
       Simplex.FileTransfer.Client
       Simplex.FileTransfer.Client.Agent
       Simplex.FileTransfer.Client.Main
@@ -44,6 +51,7 @@ library
       Simplex.FileTransfer.Description
       Simplex.FileTransfer.Protocol
       Simplex.FileTransfer.Server
+      Simplex.FileTransfer.Server.Control
       Simplex.FileTransfer.Server.Env
       Simplex.FileTransfer.Server.Main
       Simplex.FileTransfer.Server.Stats
@@ -63,6 +71,8 @@ library
       Simplex.Messaging.Agent.Server
       Simplex.Messaging.Agent.Store
       Simplex.Messaging.Agent.Store.SQLite
+      Simplex.Messaging.Agent.Store.SQLite.Common
+      Simplex.Messaging.Agent.Store.SQLite.DB
       Simplex.Messaging.Agent.Store.SQLite.Migrations
       Simplex.Messaging.Agent.Store.SQLite.Migrations.M20220101_initial
       Simplex.Messaging.Agent.Store.SQLite.Migrations.M20220301_snd_queue_keys
@@ -82,13 +92,28 @@ library
       Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230401_snd_files
       Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230510_files_pending_replicas_indexes
       Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230516_encrypted_rcv_message_hashes
+      Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230531_switch_status
+      Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230615_ratchet_sync
+      Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230701_delivery_receipts
+      Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230720_delete_expired_messages
+      Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230722_indexes
+      Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230814_indexes
+      Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230829_crypto_files
+      Simplex.Messaging.Agent.Store.SQLite.Migrations.M20231222_command_created_at
+      Simplex.Messaging.Agent.Store.SQLite.Migrations.M20231225_failed_work_items
       Simplex.Messaging.Agent.TAsyncs
       Simplex.Messaging.Agent.TRcvQueues
       Simplex.Messaging.Client
       Simplex.Messaging.Client.Agent
       Simplex.Messaging.Crypto
+      Simplex.Messaging.Crypto.File
       Simplex.Messaging.Crypto.Lazy
       Simplex.Messaging.Crypto.Ratchet
+      Simplex.Messaging.Crypto.SNTRUP761
+      Simplex.Messaging.Crypto.SNTRUP761.Bindings
+      Simplex.Messaging.Crypto.SNTRUP761.Bindings.Defines
+      Simplex.Messaging.Crypto.SNTRUP761.Bindings.FFI
+      Simplex.Messaging.Crypto.SNTRUP761.Bindings.RNG
       Simplex.Messaging.Encoding
       Simplex.Messaging.Encoding.String
       Simplex.Messaging.Notifications.Client
@@ -97,6 +122,7 @@ library
       Simplex.Messaging.Notifications.Server.Env
       Simplex.Messaging.Notifications.Server.Main
       Simplex.Messaging.Notifications.Server.Push.APNS
+      Simplex.Messaging.Notifications.Server.Push.APNS.Internal
       Simplex.Messaging.Notifications.Server.Stats
       Simplex.Messaging.Notifications.Server.Store
       Simplex.Messaging.Notifications.Server.StoreLog
@@ -106,6 +132,7 @@ library
       Simplex.Messaging.Protocol
       Simplex.Messaging.Server
       Simplex.Messaging.Server.CLI
+      Simplex.Messaging.Server.Control
       Simplex.Messaging.Server.Env.STM
       Simplex.Messaging.Server.Expiration
       Simplex.Messaging.Server.Main
@@ -119,22 +146,35 @@ library
       Simplex.Messaging.Transport
       Simplex.Messaging.Transport.Buffer
       Simplex.Messaging.Transport.Client
+      Simplex.Messaging.Transport.Credentials
       Simplex.Messaging.Transport.HTTP2
       Simplex.Messaging.Transport.HTTP2.Client
+      Simplex.Messaging.Transport.HTTP2.File
       Simplex.Messaging.Transport.HTTP2.Server
       Simplex.Messaging.Transport.KeepAlive
       Simplex.Messaging.Transport.Server
       Simplex.Messaging.Transport.WebSockets
       Simplex.Messaging.Util
       Simplex.Messaging.Version
+      Simplex.RemoteControl.Client
+      Simplex.RemoteControl.Discovery
+      Simplex.RemoteControl.Discovery.Multicast
+      Simplex.RemoteControl.Invitation
+      Simplex.RemoteControl.Types
   other-modules:
       Paths_simplexmq
   hs-source-dirs:
       src
   ghc-options: -Wall -Wcompat -Werror=incomplete-patterns -Wredundant-constraints -Wincomplete-record-updates -Wincomplete-uni-patterns -Wunused-type-patterns
+  include-dirs:
+      cbits
+  c-sources:
+      cbits/sha512.c
+      cbits/sntrup761.c
+  extra-libraries:
+      crypto
   build-depends:
-      QuickCheck ==2.14.*
-    , aeson ==2.0.*
+      aeson ==2.2.*
     , ansi-terminal >=0.10 && <0.12
     , asn1-encoding ==0.9.*
     , asn1-types ==0.3.*
@@ -142,27 +182,31 @@ library
     , attoparsec ==0.14.*
     , base >=4.14 && <5
     , base64-bytestring >=1.0 && <1.3
-    , bytestring ==0.10.*
     , case-insensitive ==1.2.*
     , composition ==1.0.*
     , constraints >=0.12 && <0.14
     , containers ==0.6.*
-    , cryptonite >=0.27 && <0.30
-    , cryptostore ==0.2.*
+    , crypton ==0.34.*
+    , crypton-x509 ==1.7.*
+    , crypton-x509-store ==1.6.*
+    , crypton-x509-validation ==1.6.*
+    , cryptostore ==0.3.*
     , data-default ==0.7.*
     , direct-sqlcipher ==2.3.*
     , directory ==1.3.*
     , filepath ==1.4.*
-    , generic-random >=1.3 && <1.5
+    , hourglass ==0.2.*
     , http-types ==0.12.*
-    , http2 ==4.1.*
+    , http2 >=4.2.2 && <4.3
     , ini ==0.4.1
     , iproute ==1.7.*
     , iso8601-time ==0.1.*
-    , memory ==0.15.*
-    , mtl ==2.2.*
+    , memory ==0.18.*
+    , mtl >=2.3.1 && <3.0
     , network >=3.1.2.7 && <3.2
-    , network-transport ==0.5.4
+    , network-info ==0.2.*
+    , network-transport ==0.5.6
+    , network-udp ==0.0.*
     , optparse-applicative >=0.15 && <0.17
     , process ==1.6.*
     , random >=1.1 && <1.3
@@ -170,24 +214,28 @@ library
     , socks ==0.6.*
     , sqlcipher-simple ==0.4.*
     , stm ==2.5.*
-    , template-haskell ==2.16.*
     , temporary ==1.3.*
-    , text ==1.2.*
-    , time ==1.9.*
-    , time-compat ==1.9.*
+    , time ==1.12.*
     , time-manager ==0.0.*
-    , tls >=1.6.0 && <1.7
-    , transformers ==0.5.*
+    , tls >=1.7.0 && <1.8
+    , transformers ==0.6.*
     , unliftio ==0.2.*
     , unliftio-core ==0.2.*
     , websockets ==0.12.*
-    , x509 ==1.7.*
-    , x509-store ==1.6.*
-    , x509-validation ==1.6.*
     , yaml ==0.11.*
   default-language: Haskell2010
   if flag(swift)
     cpp-options: -DswiftJSON
+  if impl(ghc >= 9.6.2)
+    build-depends:
+        bytestring ==0.11.*
+      , template-haskell ==2.20.*
+      , text >=2.0.1 && <2.2
+  if impl(ghc < 9.6.2)
+    build-depends:
+        bytestring ==0.10.*
+      , template-haskell ==2.16.*
+      , text >=1.2.3.0 && <1.3
 
 executable ntf-server
   main-is: Main.hs
@@ -195,10 +243,9 @@ executable ntf-server
       Paths_simplexmq
   hs-source-dirs:
       apps/ntf-server
-  ghc-options: -Wall -Wcompat -Werror=incomplete-patterns -Wredundant-constraints -Wincomplete-record-updates -Wincomplete-uni-patterns -Wunused-type-patterns -threaded
+  ghc-options: -Wall -Wcompat -Werror=incomplete-patterns -Wredundant-constraints -Wincomplete-record-updates -Wincomplete-uni-patterns -Wunused-type-patterns -threaded -rtsopts
   build-depends:
-      QuickCheck ==2.14.*
-    , aeson ==2.0.*
+      aeson ==2.2.*
     , ansi-terminal >=0.10 && <0.12
     , asn1-encoding ==0.9.*
     , asn1-types ==0.3.*
@@ -206,27 +253,31 @@ executable ntf-server
     , attoparsec ==0.14.*
     , base >=4.14 && <5
     , base64-bytestring >=1.0 && <1.3
-    , bytestring ==0.10.*
     , case-insensitive ==1.2.*
     , composition ==1.0.*
     , constraints >=0.12 && <0.14
     , containers ==0.6.*
-    , cryptonite >=0.27 && <0.30
-    , cryptostore ==0.2.*
+    , crypton ==0.34.*
+    , crypton-x509 ==1.7.*
+    , crypton-x509-store ==1.6.*
+    , crypton-x509-validation ==1.6.*
+    , cryptostore ==0.3.*
     , data-default ==0.7.*
     , direct-sqlcipher ==2.3.*
     , directory ==1.3.*
     , filepath ==1.4.*
-    , generic-random >=1.3 && <1.5
+    , hourglass ==0.2.*
     , http-types ==0.12.*
-    , http2 ==4.1.*
+    , http2 >=4.2.2 && <4.3
     , ini ==0.4.1
     , iproute ==1.7.*
     , iso8601-time ==0.1.*
-    , memory ==0.15.*
-    , mtl ==2.2.*
+    , memory ==0.18.*
+    , mtl >=2.3.1 && <3.0
     , network >=3.1.2.7 && <3.2
-    , network-transport ==0.5.4
+    , network-info ==0.2.*
+    , network-transport ==0.5.6
+    , network-udp ==0.0.*
     , optparse-applicative >=0.15 && <0.17
     , process ==1.6.*
     , random >=1.1 && <1.3
@@ -235,24 +286,28 @@ executable ntf-server
     , socks ==0.6.*
     , sqlcipher-simple ==0.4.*
     , stm ==2.5.*
-    , template-haskell ==2.16.*
     , temporary ==1.3.*
-    , text ==1.2.*
-    , time ==1.9.*
-    , time-compat ==1.9.*
+    , time ==1.12.*
     , time-manager ==0.0.*
-    , tls >=1.6.0 && <1.7
-    , transformers ==0.5.*
+    , tls >=1.7.0 && <1.8
+    , transformers ==0.6.*
     , unliftio ==0.2.*
     , unliftio-core ==0.2.*
     , websockets ==0.12.*
-    , x509 ==1.7.*
-    , x509-store ==1.6.*
-    , x509-validation ==1.6.*
     , yaml ==0.11.*
   default-language: Haskell2010
   if flag(swift)
     cpp-options: -DswiftJSON
+  if impl(ghc >= 9.6.2)
+    build-depends:
+        bytestring ==0.11.*
+      , template-haskell ==2.20.*
+      , text >=2.0.1 && <2.2
+  if impl(ghc < 9.6.2)
+    build-depends:
+        bytestring ==0.10.*
+      , template-haskell ==2.16.*
+      , text >=1.2.3.0 && <1.3
 
 executable smp-agent
   main-is: Main.hs
@@ -260,10 +315,9 @@ executable smp-agent
       Paths_simplexmq
   hs-source-dirs:
       apps/smp-agent
-  ghc-options: -Wall -Wcompat -Werror=incomplete-patterns -Wredundant-constraints -Wincomplete-record-updates -Wincomplete-uni-patterns -Wunused-type-patterns -threaded
+  ghc-options: -Wall -Wcompat -Werror=incomplete-patterns -Wredundant-constraints -Wincomplete-record-updates -Wincomplete-uni-patterns -Wunused-type-patterns -threaded -rtsopts
   build-depends:
-      QuickCheck ==2.14.*
-    , aeson ==2.0.*
+      aeson ==2.2.*
     , ansi-terminal >=0.10 && <0.12
     , asn1-encoding ==0.9.*
     , asn1-types ==0.3.*
@@ -271,27 +325,31 @@ executable smp-agent
     , attoparsec ==0.14.*
     , base >=4.14 && <5
     , base64-bytestring >=1.0 && <1.3
-    , bytestring ==0.10.*
     , case-insensitive ==1.2.*
     , composition ==1.0.*
     , constraints >=0.12 && <0.14
     , containers ==0.6.*
-    , cryptonite >=0.27 && <0.30
-    , cryptostore ==0.2.*
+    , crypton ==0.34.*
+    , crypton-x509 ==1.7.*
+    , crypton-x509-store ==1.6.*
+    , crypton-x509-validation ==1.6.*
+    , cryptostore ==0.3.*
     , data-default ==0.7.*
     , direct-sqlcipher ==2.3.*
     , directory ==1.3.*
     , filepath ==1.4.*
-    , generic-random >=1.3 && <1.5
+    , hourglass ==0.2.*
     , http-types ==0.12.*
-    , http2 ==4.1.*
+    , http2 >=4.2.2 && <4.3
     , ini ==0.4.1
     , iproute ==1.7.*
     , iso8601-time ==0.1.*
-    , memory ==0.15.*
-    , mtl ==2.2.*
+    , memory ==0.18.*
+    , mtl >=2.3.1 && <3.0
     , network >=3.1.2.7 && <3.2
-    , network-transport ==0.5.4
+    , network-info ==0.2.*
+    , network-transport ==0.5.6
+    , network-udp ==0.0.*
     , optparse-applicative >=0.15 && <0.17
     , process ==1.6.*
     , random >=1.1 && <1.3
@@ -300,24 +358,28 @@ executable smp-agent
     , socks ==0.6.*
     , sqlcipher-simple ==0.4.*
     , stm ==2.5.*
-    , template-haskell ==2.16.*
     , temporary ==1.3.*
-    , text ==1.2.*
-    , time ==1.9.*
-    , time-compat ==1.9.*
+    , time ==1.12.*
     , time-manager ==0.0.*
-    , tls >=1.6.0 && <1.7
-    , transformers ==0.5.*
+    , tls >=1.7.0 && <1.8
+    , transformers ==0.6.*
     , unliftio ==0.2.*
     , unliftio-core ==0.2.*
     , websockets ==0.12.*
-    , x509 ==1.7.*
-    , x509-store ==1.6.*
-    , x509-validation ==1.6.*
     , yaml ==0.11.*
   default-language: Haskell2010
   if flag(swift)
     cpp-options: -DswiftJSON
+  if impl(ghc >= 9.6.2)
+    build-depends:
+        bytestring ==0.11.*
+      , template-haskell ==2.20.*
+      , text >=2.0.1 && <2.2
+  if impl(ghc < 9.6.2)
+    build-depends:
+        bytestring ==0.10.*
+      , template-haskell ==2.16.*
+      , text >=1.2.3.0 && <1.3
 
 executable smp-server
   main-is: Main.hs
@@ -325,10 +387,9 @@ executable smp-server
       Paths_simplexmq
   hs-source-dirs:
       apps/smp-server
-  ghc-options: -Wall -Wcompat -Werror=incomplete-patterns -Wredundant-constraints -Wincomplete-record-updates -Wincomplete-uni-patterns -Wunused-type-patterns -threaded
+  ghc-options: -Wall -Wcompat -Werror=incomplete-patterns -Wredundant-constraints -Wincomplete-record-updates -Wincomplete-uni-patterns -Wunused-type-patterns -threaded -rtsopts
   build-depends:
-      QuickCheck ==2.14.*
-    , aeson ==2.0.*
+      aeson ==2.2.*
     , ansi-terminal >=0.10 && <0.12
     , asn1-encoding ==0.9.*
     , asn1-types ==0.3.*
@@ -336,27 +397,31 @@ executable smp-server
     , attoparsec ==0.14.*
     , base >=4.14 && <5
     , base64-bytestring >=1.0 && <1.3
-    , bytestring ==0.10.*
     , case-insensitive ==1.2.*
     , composition ==1.0.*
     , constraints >=0.12 && <0.14
     , containers ==0.6.*
-    , cryptonite >=0.27 && <0.30
-    , cryptostore ==0.2.*
+    , crypton ==0.34.*
+    , crypton-x509 ==1.7.*
+    , crypton-x509-store ==1.6.*
+    , crypton-x509-validation ==1.6.*
+    , cryptostore ==0.3.*
     , data-default ==0.7.*
     , direct-sqlcipher ==2.3.*
     , directory ==1.3.*
     , filepath ==1.4.*
-    , generic-random >=1.3 && <1.5
+    , hourglass ==0.2.*
     , http-types ==0.12.*
-    , http2 ==4.1.*
+    , http2 >=4.2.2 && <4.3
     , ini ==0.4.1
     , iproute ==1.7.*
     , iso8601-time ==0.1.*
-    , memory ==0.15.*
-    , mtl ==2.2.*
+    , memory ==0.18.*
+    , mtl >=2.3.1 && <3.0
     , network >=3.1.2.7 && <3.2
-    , network-transport ==0.5.4
+    , network-info ==0.2.*
+    , network-transport ==0.5.6
+    , network-udp ==0.0.*
     , optparse-applicative >=0.15 && <0.17
     , process ==1.6.*
     , random >=1.1 && <1.3
@@ -365,24 +430,28 @@ executable smp-server
     , socks ==0.6.*
     , sqlcipher-simple ==0.4.*
     , stm ==2.5.*
-    , template-haskell ==2.16.*
     , temporary ==1.3.*
-    , text ==1.2.*
-    , time ==1.9.*
-    , time-compat ==1.9.*
+    , time ==1.12.*
     , time-manager ==0.0.*
-    , tls >=1.6.0 && <1.7
-    , transformers ==0.5.*
+    , tls >=1.7.0 && <1.8
+    , transformers ==0.6.*
     , unliftio ==0.2.*
     , unliftio-core ==0.2.*
     , websockets ==0.12.*
-    , x509 ==1.7.*
-    , x509-store ==1.6.*
-    , x509-validation ==1.6.*
     , yaml ==0.11.*
   default-language: Haskell2010
   if flag(swift)
     cpp-options: -DswiftJSON
+  if impl(ghc >= 9.6.2)
+    build-depends:
+        bytestring ==0.11.*
+      , template-haskell ==2.20.*
+      , text >=2.0.1 && <2.2
+  if impl(ghc < 9.6.2)
+    build-depends:
+        bytestring ==0.10.*
+      , template-haskell ==2.16.*
+      , text >=1.2.3.0 && <1.3
 
 executable xftp
   main-is: Main.hs
@@ -390,10 +459,9 @@ executable xftp
       Paths_simplexmq
   hs-source-dirs:
       apps/xftp
-  ghc-options: -Wall -Wcompat -Werror=incomplete-patterns -Wredundant-constraints -Wincomplete-record-updates -Wincomplete-uni-patterns -Wunused-type-patterns -threaded
+  ghc-options: -Wall -Wcompat -Werror=incomplete-patterns -Wredundant-constraints -Wincomplete-record-updates -Wincomplete-uni-patterns -Wunused-type-patterns -threaded -rtsopts
   build-depends:
-      QuickCheck ==2.14.*
-    , aeson ==2.0.*
+      aeson ==2.2.*
     , ansi-terminal >=0.10 && <0.12
     , asn1-encoding ==0.9.*
     , asn1-types ==0.3.*
@@ -401,27 +469,31 @@ executable xftp
     , attoparsec ==0.14.*
     , base >=4.14 && <5
     , base64-bytestring >=1.0 && <1.3
-    , bytestring ==0.10.*
     , case-insensitive ==1.2.*
     , composition ==1.0.*
     , constraints >=0.12 && <0.14
     , containers ==0.6.*
-    , cryptonite >=0.27 && <0.30
-    , cryptostore ==0.2.*
+    , crypton ==0.34.*
+    , crypton-x509 ==1.7.*
+    , crypton-x509-store ==1.6.*
+    , crypton-x509-validation ==1.6.*
+    , cryptostore ==0.3.*
     , data-default ==0.7.*
     , direct-sqlcipher ==2.3.*
     , directory ==1.3.*
     , filepath ==1.4.*
-    , generic-random >=1.3 && <1.5
+    , hourglass ==0.2.*
     , http-types ==0.12.*
-    , http2 ==4.1.*
+    , http2 >=4.2.2 && <4.3
     , ini ==0.4.1
     , iproute ==1.7.*
     , iso8601-time ==0.1.*
-    , memory ==0.15.*
-    , mtl ==2.2.*
+    , memory ==0.18.*
+    , mtl >=2.3.1 && <3.0
     , network >=3.1.2.7 && <3.2
-    , network-transport ==0.5.4
+    , network-info ==0.2.*
+    , network-transport ==0.5.6
+    , network-udp ==0.0.*
     , optparse-applicative >=0.15 && <0.17
     , process ==1.6.*
     , random >=1.1 && <1.3
@@ -430,24 +502,28 @@ executable xftp
     , socks ==0.6.*
     , sqlcipher-simple ==0.4.*
     , stm ==2.5.*
-    , template-haskell ==2.16.*
     , temporary ==1.3.*
-    , text ==1.2.*
-    , time ==1.9.*
-    , time-compat ==1.9.*
+    , time ==1.12.*
     , time-manager ==0.0.*
-    , tls >=1.6.0 && <1.7
-    , transformers ==0.5.*
+    , tls >=1.7.0 && <1.8
+    , transformers ==0.6.*
     , unliftio ==0.2.*
     , unliftio-core ==0.2.*
     , websockets ==0.12.*
-    , x509 ==1.7.*
-    , x509-store ==1.6.*
-    , x509-validation ==1.6.*
     , yaml ==0.11.*
   default-language: Haskell2010
   if flag(swift)
     cpp-options: -DswiftJSON
+  if impl(ghc >= 9.6.2)
+    build-depends:
+        bytestring ==0.11.*
+      , template-haskell ==2.20.*
+      , text >=2.0.1 && <2.2
+  if impl(ghc < 9.6.2)
+    build-depends:
+        bytestring ==0.10.*
+      , template-haskell ==2.16.*
+      , text >=1.2.3.0 && <1.3
 
 executable xftp-server
   main-is: Main.hs
@@ -455,10 +531,9 @@ executable xftp-server
       Paths_simplexmq
   hs-source-dirs:
       apps/xftp-server
-  ghc-options: -Wall -Wcompat -Werror=incomplete-patterns -Wredundant-constraints -Wincomplete-record-updates -Wincomplete-uni-patterns -Wunused-type-patterns -threaded
+  ghc-options: -Wall -Wcompat -Werror=incomplete-patterns -Wredundant-constraints -Wincomplete-record-updates -Wincomplete-uni-patterns -Wunused-type-patterns -threaded -rtsopts
   build-depends:
-      QuickCheck ==2.14.*
-    , aeson ==2.0.*
+      aeson ==2.2.*
     , ansi-terminal >=0.10 && <0.12
     , asn1-encoding ==0.9.*
     , asn1-types ==0.3.*
@@ -466,27 +541,31 @@ executable xftp-server
     , attoparsec ==0.14.*
     , base >=4.14 && <5
     , base64-bytestring >=1.0 && <1.3
-    , bytestring ==0.10.*
     , case-insensitive ==1.2.*
     , composition ==1.0.*
     , constraints >=0.12 && <0.14
     , containers ==0.6.*
-    , cryptonite >=0.27 && <0.30
-    , cryptostore ==0.2.*
+    , crypton ==0.34.*
+    , crypton-x509 ==1.7.*
+    , crypton-x509-store ==1.6.*
+    , crypton-x509-validation ==1.6.*
+    , cryptostore ==0.3.*
     , data-default ==0.7.*
     , direct-sqlcipher ==2.3.*
     , directory ==1.3.*
     , filepath ==1.4.*
-    , generic-random >=1.3 && <1.5
+    , hourglass ==0.2.*
     , http-types ==0.12.*
-    , http2 ==4.1.*
+    , http2 >=4.2.2 && <4.3
     , ini ==0.4.1
     , iproute ==1.7.*
     , iso8601-time ==0.1.*
-    , memory ==0.15.*
-    , mtl ==2.2.*
+    , memory ==0.18.*
+    , mtl >=2.3.1 && <3.0
     , network >=3.1.2.7 && <3.2
-    , network-transport ==0.5.4
+    , network-info ==0.2.*
+    , network-transport ==0.5.6
+    , network-udp ==0.0.*
     , optparse-applicative >=0.15 && <0.17
     , process ==1.6.*
     , random >=1.1 && <1.3
@@ -495,24 +574,28 @@ executable xftp-server
     , socks ==0.6.*
     , sqlcipher-simple ==0.4.*
     , stm ==2.5.*
-    , template-haskell ==2.16.*
     , temporary ==1.3.*
-    , text ==1.2.*
-    , time ==1.9.*
-    , time-compat ==1.9.*
+    , time ==1.12.*
     , time-manager ==0.0.*
-    , tls >=1.6.0 && <1.7
-    , transformers ==0.5.*
+    , tls >=1.7.0 && <1.8
+    , transformers ==0.6.*
     , unliftio ==0.2.*
     , unliftio-core ==0.2.*
     , websockets ==0.12.*
-    , x509 ==1.7.*
-    , x509-store ==1.6.*
-    , x509-validation ==1.6.*
     , yaml ==0.11.*
   default-language: Haskell2010
   if flag(swift)
     cpp-options: -DswiftJSON
+  if impl(ghc >= 9.6.2)
+    build-depends:
+        bytestring ==0.11.*
+      , template-haskell ==2.20.*
+      , text >=2.0.1 && <2.2
+  if impl(ghc < 9.6.2)
+    build-depends:
+        bytestring ==0.10.*
+      , template-haskell ==2.16.*
+      , text >=1.2.3.0 && <1.3
 
 test-suite simplexmq-test
   type: exitcode-stdio-1.0
@@ -527,14 +610,18 @@ test-suite simplexmq-test
       AgentTests.SchemaDump
       AgentTests.SQLiteTests
       CLITests
+      CoreTests.BatchingTests
+      CoreTests.CryptoFileTests
       CoreTests.CryptoTests
       CoreTests.EncodingTests
       CoreTests.ProtocolErrorTests
       CoreTests.RetryIntervalTests
+      CoreTests.UtilTests
       CoreTests.VersionRangeTests
       FileDescriptionTests
       NtfClient
       NtfServerTests
+      RemoteControl
       ServerTests
       SMPAgentClient
       SMPClient
@@ -545,11 +632,11 @@ test-suite simplexmq-test
       Paths_simplexmq
   hs-source-dirs:
       tests
-  ghc-options: -Wall -Wcompat -Werror=incomplete-patterns -Wredundant-constraints -Wincomplete-record-updates -Wincomplete-uni-patterns -Wunused-type-patterns -fhpc
+  ghc-options: -Wall -Wcompat -Werror=incomplete-patterns -Wredundant-constraints -Wincomplete-record-updates -Wincomplete-uni-patterns -Wunused-type-patterns
   build-depends:
       HUnit ==1.6.*
     , QuickCheck ==2.14.*
-    , aeson ==2.0.*
+    , aeson ==2.2.*
     , ansi-terminal >=0.10 && <0.12
     , asn1-encoding ==0.9.*
     , asn1-types ==0.3.*
@@ -557,31 +644,36 @@ test-suite simplexmq-test
     , attoparsec ==0.14.*
     , base >=4.14 && <5
     , base64-bytestring >=1.0 && <1.3
-    , bytestring ==0.10.*
     , case-insensitive ==1.2.*
     , composition ==1.0.*
     , constraints >=0.12 && <0.14
     , containers ==0.6.*
-    , cryptonite >=0.27 && <0.30
-    , cryptostore ==0.2.*
+    , crypton ==0.34.*
+    , crypton-x509 ==1.7.*
+    , crypton-x509-store ==1.6.*
+    , crypton-x509-validation ==1.6.*
+    , cryptostore ==0.3.*
     , data-default ==0.7.*
     , deepseq ==1.4.*
     , direct-sqlcipher ==2.3.*
     , directory ==1.3.*
     , filepath ==1.4.*
-    , generic-random >=1.3 && <1.5
-    , hspec ==2.7.*
-    , hspec-core ==2.7.*
+    , generic-random ==1.5.*
+    , hourglass ==0.2.*
+    , hspec ==2.11.*
+    , hspec-core ==2.11.*
     , http-types ==0.12.*
-    , http2 ==4.1.*
+    , http2 >=4.2.2 && <4.3
     , ini ==0.4.1
     , iproute ==1.7.*
     , iso8601-time ==0.1.*
     , main-tester ==0.2.*
-    , memory ==0.15.*
-    , mtl ==2.2.*
+    , memory ==0.18.*
+    , mtl >=2.3.1 && <3.0
     , network >=3.1.2.7 && <3.2
-    , network-transport ==0.5.4
+    , network-info ==0.2.*
+    , network-transport ==0.5.6
+    , network-udp ==0.0.*
     , optparse-applicative >=0.15 && <0.17
     , process ==1.6.*
     , random >=1.1 && <1.3
@@ -591,22 +683,26 @@ test-suite simplexmq-test
     , socks ==0.6.*
     , sqlcipher-simple ==0.4.*
     , stm ==2.5.*
-    , template-haskell ==2.16.*
     , temporary ==1.3.*
-    , text ==1.2.*
-    , time ==1.9.*
-    , time-compat ==1.9.*
+    , time ==1.12.*
     , time-manager ==0.0.*
     , timeit ==2.0.*
-    , tls >=1.6.0 && <1.7
-    , transformers ==0.5.*
+    , tls >=1.7.0 && <1.8
+    , transformers ==0.6.*
     , unliftio ==0.2.*
     , unliftio-core ==0.2.*
     , websockets ==0.12.*
-    , x509 ==1.7.*
-    , x509-store ==1.6.*
-    , x509-validation ==1.6.*
     , yaml ==0.11.*
   default-language: Haskell2010
   if flag(swift)
     cpp-options: -DswiftJSON
+  if impl(ghc >= 9.6.2)
+    build-depends:
+        bytestring ==0.11.*
+      , template-haskell ==2.20.*
+      , text >=2.0.1 && <2.2
+  if impl(ghc < 9.6.2)
+    build-depends:
+        bytestring ==0.10.*
+      , template-haskell ==2.16.*
+      , text >=1.2.3.0 && <1.3

src/Simplex/FileTransfer.hs

@@ -1,14 +0,0 @@
-module Simplex.FileTransfer where
-
--- TODO
--- Protocol
--- Store (in memory storage)
--- StoreLog (append only log)
--- FileDescription
--- Server
--- Client
--- Server/Main (server CLI)
--- Client/Main (client CLI)
---
--- Transport for HTTP2 ?
--- streaming Crypto

src/Simplex/FileTransfer/Agent.hs

@@ -7,39 +7,32 @@
 {-# LANGUAGE OverloadedStrings #-}
 {-# LANGUAGE RankNTypes #-}
 {-# LANGUAGE ScopedTypeVariables #-}
-{-# LANGUAGE TupleSections #-}
 {-# LANGUAGE TypeApplications #-}
+{-# OPTIONS_GHC -fno-warn-ambiguous-fields #-}
 
 module Simplex.FileTransfer.Agent
-  ( startWorkers,
+  ( startXFTPWorkers,
     closeXFTPAgent,
     toFSFilePath,
     -- Receiving files
-    receiveFile,
-    deleteRcvFile,
+    xftpReceiveFile',
+    xftpDeleteRcvFile',
     -- Sending files
-    sendFileExperimental,
-    sendFile,
+    xftpSendFile',
     deleteSndFileInternal,
     deleteSndFileRemote,
   )
 where
 
-import Control.Concurrent.STM (stateTVar)
 import Control.Logger.Simple (logError)
 import Control.Monad
 import Control.Monad.Except
 import Control.Monad.Reader
-import Crypto.Random (ChaChaDRG, randomBytesGenerate)
-import Data.Bifunctor (first)
-import Data.ByteString (ByteString)
-import qualified Data.ByteString.Base64.URL as U
 import qualified Data.ByteString.Char8 as B
 import qualified Data.ByteString.Lazy.Char8 as LB
 import Data.Composition ((.:))
 import Data.Int (Int64)
-import Data.List (foldl', isSuffixOf, partition, sortOn)
-import Data.List.NonEmpty (nonEmpty)
+import Data.List (foldl', sortOn)
 import qualified Data.List.NonEmpty as L
 import Data.Map (Map)
 import qualified Data.Map.Strict as M
@@ -50,7 +43,7 @@ import Simplex.FileTransfer.Client (XFTPChunkSpec (..))
 import Simplex.FileTransfer.Client.Main
 import Simplex.FileTransfer.Crypto
 import Simplex.FileTransfer.Description
-import Simplex.FileTransfer.Protocol (FileParty (..), FilePartyI, SFileParty (..))
+import Simplex.FileTransfer.Protocol (FileParty (..), SFileParty (..))
 import Simplex.FileTransfer.Transport (XFTPRcvChunkSpec (..))
 import Simplex.FileTransfer.Types
 import Simplex.FileTransfer.Util (removePath, uniqueCombine)
@@ -60,58 +53,52 @@ import Simplex.Messaging.Agent.Protocol
 import Simplex.Messaging.Agent.RetryInterval
 import Simplex.Messaging.Agent.Store.SQLite
 import qualified Simplex.Messaging.Crypto as C
+import Simplex.Messaging.Crypto.File (CryptoFile (..), CryptoFileArgs)
+import qualified Simplex.Messaging.Crypto.File as CF
 import qualified Simplex.Messaging.Crypto.Lazy as LC
 import Simplex.Messaging.Encoding
-import Simplex.Messaging.Encoding.String
-import Simplex.Messaging.Protocol (EntityId, XFTPServer, XFTPServerWithAuth)
-import Simplex.Messaging.TMap (TMap)
-import qualified Simplex.Messaging.TMap as TM
-import Simplex.Messaging.Util (liftError, liftIOEither, tshow, unlessM, whenM)
+import Simplex.Messaging.Protocol (EntityId, XFTPServer)
+import Simplex.Messaging.Util (liftError, tshow, unlessM, whenM)
 import System.FilePath (takeFileName, (</>))
 import UnliftIO
-import UnliftIO.Concurrent
 import UnliftIO.Directory
-import qualified UnliftIO.Exception as E
 
-startWorkers :: AgentMonad m => AgentClient -> Maybe FilePath -> m ()
-startWorkers c workDir = do
+startXFTPWorkers :: AgentMonad m => AgentClient -> Maybe FilePath -> m ()
+startXFTPWorkers c workDir = do
   wd <- asks $ xftpWorkDir . xftpAgent
   atomically $ writeTVar wd workDir
-  startRcvFiles
-  startSndFiles
-  startDelFiles
+  cfg <- asks config
+  startRcvFiles cfg
+  startSndFiles cfg
+  startDelFiles cfg
   where
-    startRcvFiles = do
-      rcvFilesTTL <- asks $ rcvFilesTTL . config
+    startRcvFiles AgentConfig {rcvFilesTTL} = do
       pendingRcvServers <- withStore' c (`getPendingRcvFilesServers` rcvFilesTTL)
-      forM_ pendingRcvServers $ \s -> addXFTPRcvWorker c (Just s)
+      forM_ pendingRcvServers $ \s -> resumeXFTPRcvWork c (Just s)
       -- start local worker for files pending decryption,
       -- no need to make an extra query for the check
       -- as the worker will check the store anyway
-      addXFTPRcvWorker c Nothing
-    startSndFiles = do
-      sndFilesTTL <- asks $ sndFilesTTL . config
+      resumeXFTPRcvWork c Nothing
+    startSndFiles AgentConfig {sndFilesTTL} = do
       -- start worker for files pending encryption/creation
-      addXFTPSndWorker c Nothing
+      resumeXFTPSndWork c Nothing
       pendingSndServers <- withStore' c (`getPendingSndFilesServers` sndFilesTTL)
-      forM_ pendingSndServers $ \s -> addXFTPSndWorker c (Just s)
-    startDelFiles = do
-      rcvFilesTTL <- asks $ rcvFilesTTL . config
+      forM_ pendingSndServers $ \s -> resumeXFTPSndWork c (Just s)
+    startDelFiles AgentConfig {rcvFilesTTL} = do
       pendingDelServers <- withStore' c (`getPendingDelFilesServers` rcvFilesTTL)
-      forM_ pendingDelServers $ addXFTPDelWorker c
+      forM_ pendingDelServers $ resumeXFTPDelWork c
 
 closeXFTPAgent :: MonadUnliftIO m => XFTPAgent -> m ()
-closeXFTPAgent XFTPAgent {xftpRcvWorkers, xftpSndWorkers} = do
-  stopWorkers xftpRcvWorkers
-  stopWorkers xftpSndWorkers
+closeXFTPAgent a = do
+  stopWorkers $ xftpRcvWorkers a
+  stopWorkers $ xftpSndWorkers a
+  stopWorkers $ xftpDelWorkers a
   where
-    stopWorkers wsSel = do
-      ws <- atomically $ stateTVar wsSel (,M.empty)
-      mapM_ (uninterruptibleCancel . snd) ws
+    stopWorkers workers = atomically (swapTVar workers M.empty) >>= mapM_ (liftIO . cancelWorker)
 
-receiveFile :: AgentMonad m => AgentClient -> UserId -> ValidFileDescription 'FRecipient -> m RcvFileId
-receiveFile c userId (ValidFileDescription fd@FileDescription {chunks}) = do
-  g <- asks idsDrg
+xftpReceiveFile' :: AgentMonad m => AgentClient -> UserId -> ValidFileDescription 'FRecipient -> Maybe CryptoFileArgs -> m RcvFileId
+xftpReceiveFile' c userId (ValidFileDescription fd@FileDescription {chunks}) cfArgs = do
+  g <- asks random
   prefixPath <- getPrefixPath "rcv.xftp"
   createDirectory prefixPath
   let relPrefixPath = takeFileName prefixPath
@@ -119,13 +106,14 @@ receiveFile c userId (ValidFileDescription fd@FileDescription {chunks}) = do
       relSavePath = relPrefixPath </> "xftp.decrypted"
   createDirectory =<< toFSFilePath relTmpPath
   createEmptyFile =<< toFSFilePath relSavePath
-  fId <- withStore c $ \db -> createRcvFile db g userId fd relPrefixPath relTmpPath relSavePath
+  let saveFile = CryptoFile relSavePath cfArgs
+  fId <- withStore c $ \db -> createRcvFile db g userId fd relPrefixPath relTmpPath saveFile
   forM_ chunks downloadChunk
   pure fId
   where
     downloadChunk :: AgentMonad m => FileChunk -> m ()
     downloadChunk FileChunk {replicas = (FileChunkReplica {server} : _)} = do
-      addXFTPRcvWorker c (Just server)
+      void $ getXFTPRcvWorker True c (Just server)
     downloadChunk _ = throwError $ INTERNAL "no replicas"
 
 getPrefixPath :: AgentMonad m => String -> m FilePath
@@ -139,59 +127,37 @@ toFSFilePath :: AgentMonad m => FilePath -> m FilePath
 toFSFilePath f = (</> f) <$> getXFTPWorkPath
 
 createEmptyFile :: AgentMonad m => FilePath -> m ()
-createEmptyFile fPath = do
-  h <- openFile fPath AppendMode
-  liftIO $ B.hPut h "" >> hFlush h
+createEmptyFile fPath = liftIO $ B.writeFile fPath ""
 
-addXFTPRcvWorker :: AgentMonad m => AgentClient -> Maybe XFTPServer -> m ()
-addXFTPRcvWorker c = addWorker c xftpRcvWorkers runXFTPRcvWorker runXFTPRcvLocalWorker
+resumeXFTPRcvWork :: AgentMonad' m => AgentClient -> Maybe XFTPServer -> m ()
+resumeXFTPRcvWork = void .: getXFTPRcvWorker False
 
-addWorker ::
-  AgentMonad m =>
-  AgentClient ->
-  (XFTPAgent -> TMap (Maybe XFTPServer) (TMVar (), Async ())) ->
-  (AgentClient -> XFTPServer -> TMVar () -> m ()) ->
-  (AgentClient -> TMVar () -> m ()) ->
-  Maybe XFTPServer ->
-  m ()
-addWorker c wsSel runWorker runWorkerNoSrv srv_ = do
-  ws <- asks $ wsSel . xftpAgent
-  atomically (TM.lookup srv_ ws) >>= \case
-    Nothing -> do
-      doWork <- newTMVarIO ()
-      let runWorker' = case srv_ of
-            Just srv -> runWorker c srv doWork
-            Nothing -> runWorkerNoSrv c doWork
-      worker <- async $ runWorker' `E.finally` atomically (TM.delete srv_ ws)
-      atomically $ TM.insert srv_ (doWork, worker) ws
-    Just (doWork, _) ->
-      void . atomically $ tryPutTMVar doWork ()
+getXFTPRcvWorker :: AgentMonad' m => Bool -> AgentClient -> Maybe XFTPServer -> m Worker
+getXFTPRcvWorker hasWork c server = do
+  ws <- asks $ xftpRcvWorkers . xftpAgent
+  getAgentWorker "xftp_rcv" hasWork c server ws $
+    maybe (runXFTPRcvLocalWorker c) (runXFTPRcvWorker c) server
 
-runXFTPRcvWorker :: forall m. AgentMonad m => AgentClient -> XFTPServer -> TMVar () -> m ()
-runXFTPRcvWorker c srv doWork = do
+runXFTPRcvWorker :: forall m. AgentMonad m => AgentClient -> XFTPServer -> Worker -> m ()
+runXFTPRcvWorker c srv Worker {doWork} = do
+  cfg <- asks config
   forever $ do
-    void . atomically $ readTMVar doWork
+    waitForWork doWork
     atomically $ assertAgentForeground c
-    runXFTPOperation
+    runXFTPOperation cfg
   where
-    noWorkToDo = void . atomically $ tryTakeTMVar doWork
-    runXFTPOperation :: m ()
-    runXFTPOperation = do
-      rcvFilesTTL <- asks (rcvFilesTTL . config)
-      nextChunk <- withStore' c $ \db -> getNextRcvChunkToDownload db srv rcvFilesTTL
-      case nextChunk of
-        Nothing -> noWorkToDo
-        Just RcvFileChunk {rcvFileId, rcvFileEntityId, fileTmpPath, replicas = []} -> rcvWorkerInternalError c rcvFileId rcvFileEntityId (Just fileTmpPath) "chunk has no replicas"
-        Just fc@RcvFileChunk {userId, rcvFileId, rcvFileEntityId, digest, fileTmpPath, replicas = replica@RcvFileChunkReplica {rcvChunkReplicaId, server, delay} : _} -> do
-          ri <- asks $ reconnectInterval . config
+    runXFTPOperation :: AgentConfig -> m ()
+    runXFTPOperation AgentConfig {rcvFilesTTL, reconnectInterval = ri, xftpNotifyErrsOnRetry = notifyOnRetry, xftpConsecutiveRetries} =
+      withWork c doWork (\db -> getNextRcvChunkToDownload db srv rcvFilesTTL) $ \case
+        RcvFileChunk {rcvFileId, rcvFileEntityId, fileTmpPath, replicas = []} -> rcvWorkerInternalError c rcvFileId rcvFileEntityId (Just fileTmpPath) "chunk has no replicas"
+        fc@RcvFileChunk {userId, rcvFileId, rcvFileEntityId, digest, fileTmpPath, replicas = replica@RcvFileChunkReplica {rcvChunkReplicaId, server, delay} : _} -> do
           let ri' = maybe ri (\d -> ri {initialInterval = d, increaseAfter = 0}) delay
-          withRetryInterval ri' $ \delay' loop ->
+          withRetryIntervalLimit xftpConsecutiveRetries ri' $ \delay' loop ->
             downloadFileChunk fc replica
-              `catchError` \e -> retryOnError "XFTP rcv worker" (retryLoop loop e delay') (retryDone e) e
+              `catchAgentError` \e -> retryOnError "XFTP rcv worker" (retryLoop loop e delay') (retryDone e) e
           where
             retryLoop loop e replicaDelay = do
-              flip catchError (\_ -> pure ()) $ do
-                notifyOnRetry <- asks (xftpNotifyErrsOnRetry . config)
+              flip catchAgentError (\_ -> pure ()) $ do
                 when notifyOnRetry $ notify c rcvFileEntityId $ RFERR e
                 closeXFTPServerClient c userId server digest
                 withStore' c $ \db -> updateRcvChunkReplicaDelay db rcvChunkReplicaId replicaDelay
@@ -205,6 +171,7 @@ runXFTPRcvWorker c srv doWork = do
       let chunkSpec = XFTPRcvChunkSpec chunkPath (unFileSize chunkSize) (unFileDigest digest)
           relChunkPath = fileTmpPath </> takeFileName chunkPath
       agentXFTPDownloadChunk c userId digest replica chunkSpec
+      atomically $ waitUntilForeground c
       (complete, progress) <- withStore c $ \db -> runExceptT $ do
         liftIO $ updateRcvFileChunkReceived db (rcvChunkReplicaId replica) rcvChunkId relChunkPath
         RcvFile {size = FileSize total, chunks} <- ExceptT $ getRcvFile db rcvFileId
@@ -213,7 +180,8 @@ runXFTPRcvWorker c srv doWork = do
         liftIO . when complete $ updateRcvFileStatus db rcvFileId RFSReceived
         pure (complete, RFPROG rcvd total)
       notify c rcvFileEntityId progress
-      when complete $ addXFTPRcvWorker c Nothing
+      when complete . void $
+        getXFTPRcvWorker True c Nothing
       where
         receivedSize :: [RcvFileChunk] -> Int64
         receivedSize = foldl' (\sz ch -> sz + receivedChunkSize ch) 0
@@ -222,6 +190,12 @@ runXFTPRcvWorker c srv doWork = do
           | otherwise = 0
         chunkReceived RcvFileChunk {replicas} = any received replicas
 
+-- The first call of action has n == 0, maxN is max number of retries
+withRetryIntervalLimit :: forall m. MonadIO m => Int -> RetryInterval -> (Int64 -> m () -> m ()) -> m ()
+withRetryIntervalLimit maxN ri action =
+  withRetryIntervalCount ri $ \n delay loop ->
+    when (n < maxN) $ action delay loop
+
 retryOnError :: AgentMonad m => Text -> m a -> m a -> AgentErrorType -> m a
 retryOnError name loop done e = do
   logError $ name <> " error: " <> tshow e
@@ -235,33 +209,33 @@ rcvWorkerInternalError c rcvFileId rcvFileEntityId tmpPath internalErrStr = do
   withStore' c $ \db -> updateRcvFileError db rcvFileId internalErrStr
   notify c rcvFileEntityId $ RFERR $ INTERNAL internalErrStr
 
-runXFTPRcvLocalWorker :: forall m. AgentMonad m => AgentClient -> TMVar () -> m ()
-runXFTPRcvLocalWorker c doWork = do
+runXFTPRcvLocalWorker :: forall m. AgentMonad m => AgentClient -> Worker -> m ()
+runXFTPRcvLocalWorker c Worker {doWork} = do
+  cfg <- asks config
   forever $ do
-    void . atomically $ readTMVar doWork
+    waitForWork doWork
     atomically $ assertAgentForeground c
-    runXFTPOperation
+    runXFTPOperation cfg
   where
-    runXFTPOperation :: m ()
-    runXFTPOperation = do
-      rcvFilesTTL <- asks (rcvFilesTTL . config)
-      nextFile <- withStore' c (`getNextRcvFileToDecrypt` rcvFilesTTL)
-      case nextFile of
-        Nothing -> noWorkToDo
-        Just f@RcvFile {rcvFileId, rcvFileEntityId, tmpPath} ->
-          decryptFile f `catchError` (rcvWorkerInternalError c rcvFileId rcvFileEntityId tmpPath . show)
-    noWorkToDo = void . atomically $ tryTakeTMVar doWork
+    runXFTPOperation :: AgentConfig -> m ()
+    runXFTPOperation AgentConfig {rcvFilesTTL} =
+      withWork c doWork (`getNextRcvFileToDecrypt` rcvFilesTTL) $
+        \f@RcvFile {rcvFileId, rcvFileEntityId, tmpPath} ->
+          decryptFile f `catchAgentError` (rcvWorkerInternalError c rcvFileId rcvFileEntityId tmpPath . show)
     decryptFile :: RcvFile -> m ()
-    decryptFile RcvFile {rcvFileId, rcvFileEntityId, key, nonce, tmpPath, savePath, status, chunks} = do
+    decryptFile RcvFile {rcvFileId, rcvFileEntityId, key, nonce, tmpPath, saveFile, status, chunks} = do
+      let CryptoFile savePath cfArgs = saveFile
       fsSavePath <- toFSFilePath savePath
       when (status == RFSDecrypting) $
         whenM (doesFileExist fsSavePath) (removeFile fsSavePath >> createEmptyFile fsSavePath)
       withStore' c $ \db -> updateRcvFileStatus db rcvFileId RFSDecrypting
       chunkPaths <- getChunkPaths chunks
       encSize <- liftIO $ foldM (\s path -> (s +) . fromIntegral <$> getFileSize path) 0 chunkPaths
-      void $ liftError (INTERNAL . show) $ decryptChunks encSize chunkPaths key nonce $ \_ -> pure fsSavePath
+      let destFile = CryptoFile fsSavePath cfArgs
+      void $ liftError (INTERNAL . show) $ decryptChunks encSize chunkPaths key nonce $ \_ -> pure destFile
       notify c rcvFileEntityId $ RFDONE fsSavePath
       forM_ tmpPath (removePath <=< toFSFilePath)
+      atomically $ waitUntilForeground c
       withStore' c (`updateRcvFileComplete` rcvFileId)
       where
         getChunkPaths :: [RcvFileChunk] -> m [FilePath]
@@ -273,8 +247,8 @@ runXFTPRcvLocalWorker c doWork = do
         getChunkPaths (RcvFileChunk {chunkTmpPath = Nothing} : _cs) =
           throwError $ INTERNAL "no chunk path"
 
-deleteRcvFile :: AgentMonad m => AgentClient -> RcvFileId -> m ()
-deleteRcvFile c rcvFileEntityId = do
+xftpDeleteRcvFile' :: AgentMonad m => AgentClient -> RcvFileId -> m ()
+xftpDeleteRcvFile' c rcvFileEntityId = do
   RcvFile {rcvFileId, prefixPath, status} <- withStore c $ \db -> getRcvFileByEntityId db rcvFileEntityId
   if status == RFSComplete || status == RFSError
     then do
@@ -282,129 +256,79 @@ deleteRcvFile c rcvFileEntityId = do
       withStore' c (`deleteRcvFile'` rcvFileId)
     else withStore' c (`updateRcvFileDeleted` rcvFileId)
 
-sendFileExperimental :: forall m. AgentMonad m => AgentClient -> UserId -> FilePath -> Int -> m SndFileId
-sendFileExperimental c@AgentClient {xftpServers} userId filePath numRecipients = do
-  g <- asks idsDrg
-  sndFileId <- liftIO $ randomId g 12
-  xftpSrvs <- atomically $ TM.lookup userId xftpServers
-  void $ forkIO $ sendCLI sndFileId $ maybe [] L.toList xftpSrvs
-  pure sndFileId
-  where
-    randomId :: TVar ChaChaDRG -> Int -> IO ByteString
-    randomId gVar n = U.encode <$> (atomically . stateTVar gVar $ randomBytesGenerate n)
-    sendCLI :: SndFileId -> [XFTPServerWithAuth] -> m ()
-    sendCLI sndFileId xftpSrvs = do
-      let fileName = takeFileName filePath
-      workPath <- getXFTPWorkPath
-      outputDir <- uniqueCombine workPath $ fileName <> ".descr"
-      createDirectory outputDir
-      let tempPath = workPath </> "snd"
-      createDirectoryIfMissing False tempPath
-      runSend fileName outputDir tempPath `catchError` \e -> do
-        cleanup outputDir tempPath
-        notify c sndFileId $ SFERR e
-      where
-        runSend :: String -> FilePath -> FilePath -> m ()
-        runSend fileName outputDir tempPath = do
-          let sendOptions =
-                SendOptions
-                  { filePath,
-                    outputDir = Just outputDir,
-                    numRecipients,
-                    xftpServers = xftpSrvs,
-                    retryCount = 3,
-                    tempPath = Just tempPath,
-                    verbose = False
-                  }
-          liftCLI $ cliSendFileOpts sendOptions False $ notify c sndFileId .: SFPROG
-          (sndDescr, rcvDescrs) <- readDescrs outputDir fileName
-          cleanup outputDir tempPath
-          notify c sndFileId $ SFDONE sndDescr rcvDescrs
-        cleanup :: FilePath -> FilePath -> m ()
-        cleanup outputDir tempPath = do
-          removePath tempPath
-          removePath outputDir
-    liftCLI :: ExceptT CLIError IO () -> m ()
-    liftCLI = either (throwError . INTERNAL . show) pure <=< liftIO . runExceptT
-    readDescrs :: FilePath -> FilePath -> m (ValidFileDescription 'FSender, [ValidFileDescription 'FRecipient])
-    readDescrs outDir fileName = do
-      let descrDir = outDir </> (fileName <> ".xftp")
-      files <- listDirectory descrDir
-      let (sdFiles, rdFiles) = partition ("snd.xftp.private" `isSuffixOf`) files
-          sdFile = maybe "" (\l -> descrDir </> L.head l) (nonEmpty sdFiles)
-          rdFiles' = map (descrDir </>) rdFiles
-      (,) <$> readDescr sdFile <*> mapM readDescr rdFiles'
-    readDescr :: FilePartyI p => FilePath -> m (ValidFileDescription p)
-    readDescr f = liftIOEither $ first INTERNAL . strDecode <$> B.readFile f
-
 notify :: forall m e. (MonadUnliftIO m, AEntityI e) => AgentClient -> EntityId -> ACommand 'Agent e -> m ()
 notify c entId cmd = atomically $ writeTBQueue (subQ c) ("", entId, APC (sAEntity @e) cmd)
 
-sendFile :: AgentMonad m => AgentClient -> UserId -> FilePath -> Int -> m SndFileId
-sendFile c userId filePath numRecipients = do
-  g <- asks idsDrg
+xftpSendFile' :: AgentMonad m => AgentClient -> UserId -> CryptoFile -> Int -> m SndFileId
+xftpSendFile' c userId file numRecipients = do
+  g <- asks random
   prefixPath <- getPrefixPath "snd.xftp"
   createDirectory prefixPath
   let relPrefixPath = takeFileName prefixPath
-  key <- liftIO C.randomSbKey
-  nonce <- liftIO C.randomCbNonce
+  key <- atomically $ C.randomSbKey g
+  nonce <- atomically $ C.randomCbNonce g
   -- saving absolute filePath will not allow to restore file encryption after app update, but it's a short window
-  fId <- withStore c $ \db -> createSndFile db g userId numRecipients filePath relPrefixPath key nonce
-  addXFTPSndWorker c Nothing
+  fId <- withStore c $ \db -> createSndFile db g userId file numRecipients relPrefixPath key nonce
+  void $ getXFTPSndWorker True c Nothing
   pure fId
 
-addXFTPSndWorker :: AgentMonad m => AgentClient -> Maybe XFTPServer -> m ()
-addXFTPSndWorker c = addWorker c xftpSndWorkers runXFTPSndWorker runXFTPSndPrepareWorker
+resumeXFTPSndWork :: AgentMonad' m => AgentClient -> Maybe XFTPServer -> m ()
+resumeXFTPSndWork = void .: getXFTPSndWorker False
 
-runXFTPSndPrepareWorker :: forall m. AgentMonad m => AgentClient -> TMVar () -> m ()
-runXFTPSndPrepareWorker c doWork = do
+getXFTPSndWorker :: AgentMonad' m => Bool -> AgentClient -> Maybe XFTPServer -> m Worker
+getXFTPSndWorker hasWork c server = do
+  ws <- asks $ xftpSndWorkers . xftpAgent
+  getAgentWorker "xftp_snd" hasWork c server ws $
+    maybe (runXFTPSndPrepareWorker c) (runXFTPSndWorker c) server
+
+runXFTPSndPrepareWorker :: forall m. AgentMonad m => AgentClient -> Worker -> m ()
+runXFTPSndPrepareWorker c Worker {doWork} = do
+  cfg <- asks config
   forever $ do
-    void . atomically $ readTMVar doWork
+    waitForWork doWork
     atomically $ assertAgentForeground c
-    runXFTPOperation
+    runXFTPOperation cfg
   where
-    runXFTPOperation :: m ()
-    runXFTPOperation = do
-      sndFilesTTL <- asks (sndFilesTTL . config)
-      nextFile <- withStore' c (`getNextSndFileToPrepare` sndFilesTTL)
-      case nextFile of
-        Nothing -> noWorkToDo
-        Just f@SndFile {sndFileId, sndFileEntityId, prefixPath} ->
-          prepareFile f `catchError` (sndWorkerInternalError c sndFileId sndFileEntityId prefixPath . show)
-    noWorkToDo = void . atomically $ tryTakeTMVar doWork
-    prepareFile :: SndFile -> m ()
-    prepareFile SndFile {prefixPath = Nothing} =
+    runXFTPOperation :: AgentConfig -> m ()
+    runXFTPOperation cfg@AgentConfig {sndFilesTTL} =
+      withWork c doWork (`getNextSndFileToPrepare` sndFilesTTL) $
+        \f@SndFile {sndFileId, sndFileEntityId, prefixPath} ->
+          prepareFile cfg f `catchAgentError` (sndWorkerInternalError c sndFileId sndFileEntityId prefixPath . show)
+    prepareFile :: AgentConfig -> SndFile -> m ()
+    prepareFile _ SndFile {prefixPath = Nothing} =
       throwError $ INTERNAL "no prefix path"
-    prepareFile sndFile@SndFile {sndFileId, userId, prefixPath = Just ppath, status} = do
+    prepareFile cfg sndFile@SndFile {sndFileId, userId, prefixPath = Just ppath, status} = do
       SndFile {numRecipients, chunks} <-
         if status /= SFSEncrypted -- status is SFSNew or SFSEncrypting
           then do
             fsEncPath <- toFSFilePath $ sndFileEncPath ppath
-            when (status == SFSEncrypting) $
-              whenM (doesFileExist fsEncPath) $ removeFile fsEncPath
+            when (status == SFSEncrypting) . whenM (doesFileExist fsEncPath) $
+              removeFile fsEncPath
             withStore' c $ \db -> updateSndFileStatus db sndFileId SFSEncrypting
             (digest, chunkSpecsDigests) <- encryptFileForUpload sndFile fsEncPath
             withStore c $ \db -> do
               updateSndFileEncrypted db sndFileId digest chunkSpecsDigests
               getSndFile db sndFileId
           else pure sndFile
-      maxRecipients <- asks (xftpMaxRecipientsPerRequest . config)
       let numRecipients' = min numRecipients maxRecipients
       -- concurrently?
+      -- separate worker to create chunks? record retries and delay on snd_file_chunks?
       forM_ (filter (not . chunkCreated) chunks) $ createChunk numRecipients'
       withStore' c $ \db -> updateSndFileStatus db sndFileId SFSUploading
       where
+        AgentConfig {xftpMaxRecipientsPerRequest = maxRecipients, messageRetryInterval = ri} = cfg
         encryptFileForUpload :: SndFile -> FilePath -> m (FileDigest, [(XFTPChunkSpec, FileDigest)])
-        encryptFileForUpload SndFile {key, nonce, filePath} fsEncPath = do
-          let fileName = takeFileName filePath
-          fileSize <- fromInteger <$> getFileSize filePath
+        encryptFileForUpload SndFile {key, nonce, srcFile} fsEncPath = do
+          let CryptoFile {filePath} = srcFile
+              fileName = takeFileName filePath
+          fileSize <- liftIO $ fromInteger <$> CF.getFileContentsSize srcFile
           when (fileSize > maxFileSize) $ throwError $ INTERNAL "max file size exceeded"
           let fileHdr = smpEncode FileHeader {fileName, fileExtra = Nothing}
               fileSize' = fromIntegral (B.length fileHdr) + fileSize
               chunkSizes = prepareChunkSizes $ fileSize' + fileSizeLen + authTagSize
               chunkSizes' = map fromIntegral chunkSizes
               encSize = sum chunkSizes'
-          void $ liftError (INTERNAL . show) $ encryptFile filePath fileHdr key nonce fileSize' encSize fsEncPath
+          void $ liftError (INTERNAL . show) $ encryptFile srcFile fileHdr key nonce fileSize' encSize fsEncPath
           digest <- liftIO $ LC.sha512Hash <$> LB.readFile fsEncPath
           let chunkSpecs = prepareChunkSpecs fsEncPath chunkSizes
           chunkDigests <- map FileDigest <$> mapM (liftIO . getChunkDigest) chunkSpecs
@@ -417,14 +341,13 @@ runXFTPSndPrepareWorker c doWork = do
           atomically $ assertAgentForeground c
           (replica, ProtoServerWithAuth srv _) <- tryCreate
           withStore' c $ \db -> createSndFileReplica db ch replica
-          addXFTPSndWorker c $ Just srv
+          void $ getXFTPSndWorker True c (Just srv)
           where
             tryCreate = do
-              ri <- asks $ messageRetryInterval . config
               usedSrvs <- newTVarIO ([] :: [XFTPServer])
               withRetryInterval (riFast ri) $ \_ loop ->
                 createWithNextSrv usedSrvs
-                  `catchError` \e -> retryOnError "XFTP prepare worker" (retryLoop loop) (throwError e) e
+                  `catchAgentError` \e -> retryOnError "XFTP prepare worker" (retryLoop loop) (throwError e) e
               where
                 retryLoop loop = atomically (assertAgentForeground c) >> loop
             createWithNextSrv usedSrvs = do
@@ -440,45 +363,41 @@ sndWorkerInternalError c sndFileId sndFileEntityId prefixPath internalErrStr = d
   withStore' c $ \db -> updateSndFileError db sndFileId internalErrStr
   notify c sndFileEntityId $ SFERR $ INTERNAL internalErrStr
 
-runXFTPSndWorker :: forall m. AgentMonad m => AgentClient -> XFTPServer -> TMVar () -> m ()
-runXFTPSndWorker c srv doWork = do
+runXFTPSndWorker :: forall m. AgentMonad m => AgentClient -> XFTPServer -> Worker -> m ()
+runXFTPSndWorker c srv Worker {doWork} = do
+  cfg <- asks config
   forever $ do
-    void . atomically $ readTMVar doWork
+    waitForWork doWork
     atomically $ assertAgentForeground c
-    runXFTPOperation
+    runXFTPOperation cfg
   where
-    noWorkToDo = void . atomically $ tryTakeTMVar doWork
-    runXFTPOperation :: m ()
-    runXFTPOperation = do
-      sndFilesTTL <- asks (sndFilesTTL . config)
-      nextChunk <- withStore' c $ \db -> getNextSndChunkToUpload db srv sndFilesTTL
-      case nextChunk of
-        Nothing -> noWorkToDo
-        Just SndFileChunk {sndFileId, sndFileEntityId, filePrefixPath, replicas = []} -> sndWorkerInternalError c sndFileId sndFileEntityId (Just filePrefixPath) "chunk has no replicas"
-        Just fc@SndFileChunk {userId, sndFileId, sndFileEntityId, filePrefixPath, digest, replicas = replica@SndFileChunkReplica {sndChunkReplicaId, server, delay} : _} -> do
-          ri <- asks $ reconnectInterval . config
+    runXFTPOperation :: AgentConfig -> m ()
+    runXFTPOperation cfg@AgentConfig {sndFilesTTL, reconnectInterval = ri, xftpNotifyErrsOnRetry = notifyOnRetry, xftpConsecutiveRetries} = do
+      withWork c doWork (\db -> getNextSndChunkToUpload db srv sndFilesTTL) $ \case
+        SndFileChunk {sndFileId, sndFileEntityId, filePrefixPath, replicas = []} -> sndWorkerInternalError c sndFileId sndFileEntityId (Just filePrefixPath) "chunk has no replicas"
+        fc@SndFileChunk {userId, sndFileId, sndFileEntityId, filePrefixPath, digest, replicas = replica@SndFileChunkReplica {sndChunkReplicaId, server, delay} : _} -> do
           let ri' = maybe ri (\d -> ri {initialInterval = d, increaseAfter = 0}) delay
-          withRetryInterval ri' $ \delay' loop ->
-            uploadFileChunk fc replica
-              `catchError` \e -> retryOnError "XFTP snd worker" (retryLoop loop e delay') (retryDone e) e
+          withRetryIntervalLimit xftpConsecutiveRetries ri' $ \delay' loop ->
+            uploadFileChunk cfg fc replica
+              `catchAgentError` \e -> retryOnError "XFTP snd worker" (retryLoop loop e delay') (retryDone e) e
           where
             retryLoop loop e replicaDelay = do
-              flip catchError (\_ -> pure ()) $ do
-                notifyOnRetry <- asks (xftpNotifyErrsOnRetry . config)
+              flip catchAgentError (\_ -> pure ()) $ do
                 when notifyOnRetry $ notify c sndFileEntityId $ SFERR e
                 closeXFTPServerClient c userId server digest
                 withStore' c $ \db -> updateSndChunkReplicaDelay db sndChunkReplicaId replicaDelay
               atomically $ assertAgentForeground c
               loop
             retryDone e = sndWorkerInternalError c sndFileId sndFileEntityId (Just filePrefixPath) (show e)
-    uploadFileChunk :: SndFileChunk -> SndFileChunkReplica -> m ()
-    uploadFileChunk sndFileChunk@SndFileChunk {sndFileId, userId, chunkSpec = chunkSpec@XFTPChunkSpec {filePath}, digest = chunkDigest} replica = do
+    uploadFileChunk :: AgentConfig -> SndFileChunk -> SndFileChunkReplica -> m ()
+    uploadFileChunk AgentConfig {xftpMaxRecipientsPerRequest = maxRecipients} sndFileChunk@SndFileChunk {sndFileId, userId, chunkSpec = chunkSpec@XFTPChunkSpec {filePath}, digest = chunkDigest} replica = do
       replica'@SndFileChunkReplica {sndChunkReplicaId} <- addRecipients sndFileChunk replica
       fsFilePath <- toFSFilePath filePath
       unlessM (doesFileExist fsFilePath) $ throwError $ INTERNAL "encrypted file doesn't exist on upload"
       let chunkSpec' = chunkSpec {filePath = fsFilePath} :: XFTPChunkSpec
       atomically $ assertAgentForeground c
       agentXFTPUploadChunk c userId chunkDigest replica' chunkSpec'
+      atomically $ waitUntilForeground c
       sf@SndFile {sndFileEntityId, prefixPath, chunks} <- withStore c $ \db -> do
         updateSndChunkReplicaStatus db sndChunkReplicaId SFRSUploaded
         getSndFile db sndFileId
@@ -497,11 +416,10 @@ runXFTPSndWorker c srv doWork = do
           | length rcvIdsKeys > numRecipients = throwError $ INTERNAL "too many recipients"
           | length rcvIdsKeys == numRecipients = pure cr
           | otherwise = do
-            maxRecipients <- asks $ xftpMaxRecipientsPerRequest . config
-            let numRecipients' = min (numRecipients - length rcvIdsKeys) maxRecipients
-            rcvIdsKeys' <- agentXFTPAddRecipients c userId chunkDigest cr numRecipients'
-            cr' <- withStore' c $ \db -> addSndChunkReplicaRecipients db cr $ L.toList rcvIdsKeys'
-            addRecipients ch cr'
+              let numRecipients' = min (numRecipients - length rcvIdsKeys) maxRecipients
+              rcvIdsKeys' <- agentXFTPAddRecipients c userId chunkDigest cr numRecipients'
+              cr' <- withStore' c $ \db -> addSndChunkReplicaRecipients db cr $ L.toList rcvIdsKeys'
+              addRecipients ch cr'
         sndFileToDescrs :: SndFile -> m (ValidFileDescription 'FSender, [ValidFileDescription 'FRecipient])
         sndFileToDescrs SndFile {digest = Nothing} = throwError $ INTERNAL "snd file has no digest"
         sndFileToDescrs SndFile {chunks = []} = throwError $ INTERNAL "snd file has no chunks"
@@ -542,7 +460,7 @@ runXFTPSndWorker c srv doWork = do
             rcvChunks :: [[FileChunk]]
             rcvChunks = map (sortChunks . M.elems) $ M.elems $ foldl' addRcvChunk M.empty rcvReplicas
             sortChunks :: [FileChunk] -> [FileChunk]
-            sortChunks = map reverseReplicas . sortOn (chunkNo :: FileChunk -> Int)
+            sortChunks = map reverseReplicas . sortOn (\FileChunk {chunkNo} -> chunkNo)
             reverseReplicas ch@FileChunk {replicas} = (ch :: FileChunk) {replicas = reverse replicas}
             addRcvChunk :: Map Int (Map Int FileChunk) -> SentRecipientReplica -> Map Int (Map Int FileChunk)
             addRcvChunk m SentRecipientReplica {chunkNo, server, rcvNo, replicaId, replicaKey, digest, chunkSize} =
@@ -579,61 +497,53 @@ deleteSndFileInternal c sndFileEntityId = do
 
 deleteSndFileRemote :: forall m. AgentMonad m => AgentClient -> UserId -> SndFileId -> ValidFileDescription 'FSender -> m ()
 deleteSndFileRemote c userId sndFileEntityId (ValidFileDescription FileDescription {chunks}) = do
-  deleteSndFileInternal c sndFileEntityId `catchError` (notify c sndFileEntityId . SFERR)
-  forM_ chunks $ \ch -> deleteFileChunk ch `catchError` (notify c sndFileEntityId . SFERR)
+  deleteSndFileInternal c sndFileEntityId `catchAgentError` (notify c sndFileEntityId . SFERR)
+  forM_ chunks $ \ch -> deleteFileChunk ch `catchAgentError` (notify c sndFileEntityId . SFERR)
   where
     deleteFileChunk :: FileChunk -> m ()
     deleteFileChunk FileChunk {digest, replicas = replica@FileChunkReplica {server} : _} = do
       withStore' c $ \db -> createDeletedSndChunkReplica db userId replica digest
-      addXFTPDelWorker c server
+      void $ getXFTPDelWorker True c server
     deleteFileChunk _ = pure ()
 
-addXFTPDelWorker :: AgentMonad m => AgentClient -> XFTPServer -> m ()
-addXFTPDelWorker c srv = do
-  ws <- asks $ xftpDelWorkers . xftpAgent
-  atomically (TM.lookup srv ws) >>= \case
-    Nothing -> do
-      doWork <- newTMVarIO ()
-      worker <- async $ runXFTPDelWorker c srv doWork `E.finally` atomically (TM.delete srv ws)
-      atomically $ TM.insert srv (doWork, worker) ws
-    Just (doWork, _) ->
-      void . atomically $ tryPutTMVar doWork ()
+resumeXFTPDelWork :: AgentMonad' m => AgentClient -> XFTPServer -> m ()
+resumeXFTPDelWork = void .: getXFTPDelWorker False
 
-runXFTPDelWorker :: forall m. AgentMonad m => AgentClient -> XFTPServer -> TMVar () -> m ()
-runXFTPDelWorker c srv doWork = do
+getXFTPDelWorker :: AgentMonad' m => Bool -> AgentClient -> XFTPServer -> m Worker
+getXFTPDelWorker hasWork c server = do
+  ws <- asks $ xftpDelWorkers . xftpAgent
+  getAgentWorker "xftp_del" hasWork c server ws $ runXFTPDelWorker c server
+
+runXFTPDelWorker :: forall m. AgentMonad m => AgentClient -> XFTPServer -> Worker -> m ()
+runXFTPDelWorker c srv Worker {doWork} = do
+  cfg <- asks config
   forever $ do
-    void . atomically $ readTMVar doWork
+    waitForWork doWork
     atomically $ assertAgentForeground c
-    runXFTPOperation
+    runXFTPOperation cfg
   where
-    noWorkToDo = void . atomically $ tryTakeTMVar doWork
-    runXFTPOperation :: m ()
-    runXFTPOperation = do
+    runXFTPOperation :: AgentConfig -> m ()
+    runXFTPOperation AgentConfig {rcvFilesTTL, reconnectInterval = ri, xftpNotifyErrsOnRetry = notifyOnRetry, xftpConsecutiveRetries} = do
       -- no point in deleting files older than rcv ttl, as they will be expired on server
-      rcvFilesTTL <- asks (rcvFilesTTL . config)
-      nextReplica <- withStore' c $ \db -> getNextDeletedSndChunkReplica db srv rcvFilesTTL
-      case nextReplica of
-        Nothing -> noWorkToDo
-        Just replica@DeletedSndChunkReplica {deletedSndChunkReplicaId, userId, server, chunkDigest, delay} -> do
-          ri <- asks $ reconnectInterval . config
+      withWork c doWork (\db -> getNextDeletedSndChunkReplica db srv rcvFilesTTL) processDeletedReplica
+      where
+        processDeletedReplica replica@DeletedSndChunkReplica {deletedSndChunkReplicaId, userId, server, chunkDigest, delay} = do
           let ri' = maybe ri (\d -> ri {initialInterval = d, increaseAfter = 0}) delay
-          withRetryInterval ri' $ \delay' loop ->
-            deleteChunkReplica replica
-              `catchError` \e -> retryOnError "XFTP del worker" (retryLoop loop e delay') (retryDone e) e
+          withRetryIntervalLimit xftpConsecutiveRetries ri' $ \delay' loop ->
+            deleteChunkReplica
+              `catchAgentError` \e -> retryOnError "XFTP del worker" (retryLoop loop e delay') (retryDone e) e
           where
             retryLoop loop e replicaDelay = do
-              flip catchError (\_ -> pure ()) $ do
-                notifyOnRetry <- asks (xftpNotifyErrsOnRetry . config)
+              flip catchAgentError (\_ -> pure ()) $ do
                 when notifyOnRetry $ notify c "" $ SFERR e
                 closeXFTPServerClient c userId server chunkDigest
                 withStore' c $ \db -> updateDeletedSndChunkReplicaDelay db deletedSndChunkReplicaId replicaDelay
               atomically $ assertAgentForeground c
               loop
-            retryDone e = delWorkerInternalError c deletedSndChunkReplicaId e
-    deleteChunkReplica :: DeletedSndChunkReplica -> m ()
-    deleteChunkReplica replica@DeletedSndChunkReplica {userId, deletedSndChunkReplicaId} = do
-      agentXFTPDeleteChunk c userId replica
-      withStore' c $ \db -> deleteDeletedSndChunkReplica db deletedSndChunkReplicaId
+            retryDone = delWorkerInternalError c deletedSndChunkReplicaId
+            deleteChunkReplica = do
+              agentXFTPDeleteChunk c userId replica
+              withStore' c $ \db -> deleteDeletedSndChunkReplica db deletedSndChunkReplicaId
 
 delWorkerInternalError :: AgentMonad m => AgentClient -> Int64 -> AgentErrorType -> m ()
 delWorkerInternalError c deletedSndChunkReplicaId e = do

src/Simplex/FileTransfer/Chunks.hs

@@ -0,0 +1,35 @@
+module Simplex.FileTransfer.Chunks where
+
+import Data.Word (Word32)
+
+serverChunkSizes :: [Word32]
+serverChunkSizes = [chunkSize0, chunkSize1, chunkSize2, chunkSize3]
+{-# INLINE serverChunkSizes #-}
+
+chunkSize0 :: Word32
+chunkSize0 = kb 64
+{-# INLINE chunkSize0 #-}
+
+chunkSize1 :: Word32
+chunkSize1 = kb 256
+{-# INLINE chunkSize1 #-}
+
+chunkSize2 :: Word32
+chunkSize2 = mb 1
+{-# INLINE chunkSize2 #-}
+
+chunkSize3 :: Word32
+chunkSize3 = mb 4
+{-# INLINE chunkSize3 #-}
+
+kb :: Integral a => a -> a
+kb n = 1024 * n
+{-# INLINE kb #-}
+
+mb :: Integral a => a -> a
+mb n = 1024 * kb n
+{-# INLINE mb #-}
+
+gb :: Integral a => a -> a
+gb n = 1024 * mb n
+{-# INLINE gb #-}

src/Simplex/FileTransfer/Client.hs

@@ -9,7 +9,9 @@
 
 module Simplex.FileTransfer.Client where
 
+import Control.Monad
 import Control.Monad.Except
+import Crypto.Random (ChaChaDRG)
 import Data.Bifunctor (first)
 import Data.ByteString.Builder (Builder, byteString)
 import Data.ByteString.Char8 (ByteString)
@@ -47,6 +49,7 @@ import Simplex.Messaging.Transport (supportedParameters)
 import Simplex.Messaging.Transport.Client (TransportClientConfig, TransportHost)
 import Simplex.Messaging.Transport.HTTP2
 import Simplex.Messaging.Transport.HTTP2.Client
+import Simplex.Messaging.Transport.HTTP2.File
 import Simplex.Messaging.Util (bshow, liftEitherError, whenM)
 import UnliftIO
 import UnliftIO.Directory
@@ -108,7 +111,7 @@ xftpClientServer = B.unpack . strEncode . snd3 . transportSession
     snd3 (_, s, _) = s
 
 xftpTransportHost :: XFTPClient -> TransportHost
-xftpTransportHost = (host :: HClient -> TransportHost) . client_ . http2Client
+xftpTransportHost XFTPClient {http2Client = HTTP2Client {client_ = HClient {host}}} = host
 
 xftpSessionTs :: XFTPClient -> UTCTime
 xftpSessionTs = sessionTs . http2Client
@@ -129,11 +132,15 @@ xftpClientError = \case
   HCIOError e -> PCEIOError e
 
 sendXFTPCommand :: forall p. FilePartyI p => XFTPClient -> C.APrivateSignKey -> XFTPFileId -> FileCommand p -> Maybe XFTPChunkSpec -> ExceptT XFTPClientError IO (FileResponse, HTTP2Body)
-sendXFTPCommand XFTPClient {config, http2Client = http2@HTTP2Client {sessionId}} pKey fId cmd chunkSpec_ = do
+sendXFTPCommand c@XFTPClient {http2Client = HTTP2Client {sessionId}} pKey fId cmd chunkSpec_ = do
   t <-
     liftEither . first PCETransportError $
       xftpEncodeTransmission sessionId (Just pKey) ("", fId, FileCmd (sFileParty @p) cmd)
-  let req = H.requestStreaming N.methodPost "/" [] $ streamBody t
+  sendXFTPTransmission c t chunkSpec_
+
+sendXFTPTransmission :: XFTPClient -> ByteString -> Maybe XFTPChunkSpec -> ExceptT XFTPClientError IO (FileResponse, HTTP2Body)
+sendXFTPTransmission XFTPClient {config, http2Client = http2@HTTP2Client {sessionId}} t chunkSpec_ = do
+  let req = H.requestStreaming N.methodPost "/" [] streamBody
       reqTimeout = (\XFTPChunkSpec {chunkSize} -> chunkTimeout config chunkSize) <$> chunkSpec_
   HTTP2Response {respBody = body@HTTP2Body {bodyHead}} <- liftEitherError xftpClientError $ sendRequest http2 req reqTimeout
   when (B.length bodyHead /= xftpBlockSize) $ throwError $ PCEResponseError BLOCK
@@ -145,13 +152,13 @@ sendXFTPCommand XFTPClient {config, http2Client = http2@HTTP2Client {sessionId}}
       _ -> pure (r, body)
     Left e -> throwError $ PCEResponseError e
   where
-    streamBody :: ByteString -> (Builder -> IO ()) -> IO () -> IO ()
-    streamBody t send done = do
+    streamBody :: (Builder -> IO ()) -> IO () -> IO ()
+    streamBody send done = do
       send $ byteString t
       forM_ chunkSpec_ $ \XFTPChunkSpec {filePath, chunkOffset, chunkSize} ->
         withFile filePath ReadMode $ \h -> do
           hSeek h AbsoluteSeek $ fromIntegral chunkOffset
-          sendFile h send $ fromIntegral chunkSize
+          hSendFile h send $ fromIntegral chunkSize
       done
 
 createXFTPChunk ::
@@ -176,9 +183,9 @@ uploadXFTPChunk :: XFTPClient -> C.APrivateSignKey -> XFTPFileId -> XFTPChunkSpe
 uploadXFTPChunk c spKey fId chunkSpec =
   sendXFTPCommand c spKey fId FPUT (Just chunkSpec) >>= okResponse
 
-downloadXFTPChunk :: XFTPClient -> C.APrivateSignKey -> XFTPFileId -> XFTPRcvChunkSpec -> ExceptT XFTPClientError IO ()
-downloadXFTPChunk c@XFTPClient {config} rpKey fId chunkSpec@XFTPRcvChunkSpec {filePath, chunkSize} = do
-  (rDhKey, rpDhKey) <- liftIO C.generateKeyPair'
+downloadXFTPChunk :: TVar ChaChaDRG -> XFTPClient -> C.APrivateSignKey -> XFTPFileId -> XFTPRcvChunkSpec -> ExceptT XFTPClientError IO ()
+downloadXFTPChunk g c@XFTPClient {config} rpKey fId chunkSpec@XFTPRcvChunkSpec {filePath, chunkSize} = do
+  (rDhKey, rpDhKey) <- atomically $ C.generateKeyPair g
   sendXFTPCommand c rpKey fId (FGET rDhKey) Nothing >>= \case
     (FRFile sDhKey cbNonce, HTTP2Body {bodyHead = _bg, bodySize = _bs, bodyPart}) -> case bodyPart of
       -- TODO atm bodySize is set to 0, so chunkSize will be incorrect - validate once set
@@ -204,6 +211,16 @@ deleteXFTPChunk c spKey sId = sendXFTPCommand c spKey sId FDEL Nothing >>= okRes
 ackXFTPChunk :: XFTPClient -> C.APrivateSignKey -> RecipientId -> ExceptT XFTPClientError IO ()
 ackXFTPChunk c rpKey rId = sendXFTPCommand c rpKey rId FACK Nothing >>= okResponse
 
+pingXFTP :: XFTPClient -> ExceptT XFTPClientError IO ()
+pingXFTP c@XFTPClient {http2Client = HTTP2Client {sessionId}} = do
+  t <-
+    liftEither . first PCETransportError $
+      xftpEncodeTransmission sessionId Nothing ("", "", FileCmd SFRecipient PING)
+  (r, _) <- sendXFTPTransmission c t Nothing
+  case r of
+    FRPong -> pure ()
+    _ -> throwError $ PCEUnexpectedResponse $ bshow r
+
 okResponse :: (FileResponse, HTTP2Body) -> ExceptT XFTPClientError IO ()
 okResponse = \case
   (FROk, body) -> noFile body ()

src/Simplex/FileTransfer/Client/Agent.hs

@@ -8,6 +8,7 @@
 module Simplex.FileTransfer.Client.Agent where
 
 import Control.Logger.Simple (logInfo)
+import Control.Monad
 import Control.Monad.Except
 import Data.Bifunctor (first)
 import qualified Data.ByteString.Char8 as B
@@ -21,7 +22,7 @@ import Simplex.Messaging.Encoding.String
 import Simplex.Messaging.Protocol (ProtocolServer (..), XFTPServer)
 import Simplex.Messaging.TMap (TMap)
 import qualified Simplex.Messaging.TMap as TM
-import Simplex.Messaging.Util (catchAll_, tryError)
+import Simplex.Messaging.Util (catchAll_)
 import UnliftIO
 
 type XFTPClientVar = TMVar (Either XFTPClientAgentError XFTPClient)

src/Simplex/FileTransfer/Client/Main.hs

@@ -8,18 +8,15 @@
 {-# LANGUAGE OverloadedStrings #-}
 {-# LANGUAGE ScopedTypeVariables #-}
 {-# LANGUAGE TypeApplications #-}
+{-# OPTIONS_GHC -fno-warn-ambiguous-fields #-}
 
 module Simplex.FileTransfer.Client.Main
   ( SendOptions (..),
     CLIError (..),
     xftpClientCLI,
     cliSendFile,
-    cliSendFileOpts,
     prepareChunkSizes,
     prepareChunkSpecs,
-    chunkSize1,
-    chunkSize2,
-    chunkSize3,
     maxFileSize,
     fileSizeLen,
     getChunkDigest,
@@ -27,20 +24,18 @@ module Simplex.FileTransfer.Client.Main
   )
 where
 
-import Control.Concurrent.STM (stateTVar)
 import Control.Logger.Simple
 import Control.Monad
 import Control.Monad.Except
-import Crypto.Random (getRandomBytes)
+import Crypto.Random (ChaChaDRG)
 import qualified Data.Attoparsec.ByteString.Char8 as A
 import Data.Bifunctor (first)
 import Data.ByteString.Char8 (ByteString)
 import qualified Data.ByteString.Char8 as B
 import qualified Data.ByteString.Lazy.Char8 as LB
 import Data.Char (toLower)
-import Data.Function (on)
 import Data.Int (Int64)
-import Data.List (foldl', groupBy, sortOn)
+import Data.List (foldl', sortOn)
 import Data.List.NonEmpty (NonEmpty (..), nonEmpty)
 import qualified Data.List.NonEmpty as L
 import Data.Map (Map)
@@ -50,6 +45,7 @@ import qualified Data.Text as T
 import Data.Word (Word32)
 import GHC.Records (HasField (getField))
 import Options.Applicative
+import Simplex.FileTransfer.Chunks
 import Simplex.FileTransfer.Client
 import Simplex.FileTransfer.Client.Agent
 import Simplex.FileTransfer.Client.Presets
@@ -60,13 +56,15 @@ import Simplex.FileTransfer.Transport (XFTPRcvChunkSpec (..))
 import Simplex.FileTransfer.Types
 import Simplex.FileTransfer.Util (uniqueCombine)
 import qualified Simplex.Messaging.Crypto as C
+import Simplex.Messaging.Crypto.File (CryptoFile (..), FTCryptoError (..))
+import qualified Simplex.Messaging.Crypto.File as CF
 import qualified Simplex.Messaging.Crypto.Lazy as LC
 import Simplex.Messaging.Encoding
 import Simplex.Messaging.Encoding.String (StrEncoding (..))
 import Simplex.Messaging.Parsers (parseAll)
 import Simplex.Messaging.Protocol (ProtoServerWithAuth (..), SenderId, SndPrivateSignKey, XFTPServer, XFTPServerWithAuth)
 import Simplex.Messaging.Server.CLI (getCliCommand')
-import Simplex.Messaging.Util (ifM, tshow, whenM)
+import Simplex.Messaging.Util (groupAllOn, ifM, tshow, whenM)
 import System.Exit (exitFailure)
 import System.FilePath (splitFileName, (</>))
 import System.IO.Temp (getCanonicalTemporaryDirectory)
@@ -75,16 +73,7 @@ import UnliftIO
 import UnliftIO.Directory
 
 xftpClientVersion :: String
-xftpClientVersion = "0.1.0"
-
-chunkSize1 :: Word32
-chunkSize1 = kb 256
-
-chunkSize2 :: Word32
-chunkSize2 = mb 1
-
-chunkSize3 :: Word32
-chunkSize3 = mb 4
+xftpClientVersion = "1.0.1"
 
 maxFileSize :: Int64
 maxFileSize = gb 1
@@ -103,6 +92,7 @@ cliCryptoError = \case
   FTCECryptoError e -> CLIError $ "Error decrypting file: " <> show e
   FTCEInvalidHeader e -> CLIError $ "Invalid file header: " <> e
   FTCEInvalidAuthTag -> CLIError "Error decrypting file: incorrect auth tag"
+  FTCEInvalidFileSize -> CLIError "Error decrypting file: incorrect file size"
   FTCEFileIOError e -> CLIError $ "File IO error: " <> show e
 
 data CliCommand
@@ -273,10 +263,11 @@ cliSendFileOpts :: SendOptions -> Bool -> (Int64 -> Int64 -> IO ()) -> ExceptT C
 cliSendFileOpts SendOptions {filePath, outputDir, numRecipients, xftpServers, retryCount, tempPath, verbose} printInfo notifyProgress = do
   let (_, fileName) = splitFileName filePath
   liftIO $ when printInfo $ printNoNewLine "Encrypting file..."
-  (encPath, fdRcv, fdSnd, chunkSpecs, encSize) <- encryptFileForUpload fileName
+  g <- liftIO C.newRandom
+  (encPath, fdRcv, fdSnd, chunkSpecs, encSize) <- encryptFileForUpload g fileName
   liftIO $ when printInfo $ printNoNewLine "Uploading file..."
   uploadedChunks <- newTVarIO []
-  sentChunks <- uploadFile chunkSpecs uploadedChunks encSize
+  sentChunks <- uploadFile g chunkSpecs uploadedChunks encSize
   whenM (doesFileExist encPath) $ removeFile encPath
   -- TODO if only small chunks, use different default size
   liftIO $ do
@@ -289,34 +280,35 @@ cliSendFileOpts SendOptions {filePath, outputDir, numRecipients, xftpServers, re
       putStrLn "Pass file descriptions to the recipient(s):"
     forM_ fdRcvPaths putStrLn
   where
-    encryptFileForUpload :: String -> ExceptT CLIError IO (FilePath, FileDescription 'FRecipient, FileDescription 'FSender, [XFTPChunkSpec], Int64)
-    encryptFileForUpload fileName = do
+    encryptFileForUpload :: TVar ChaChaDRG -> String -> ExceptT CLIError IO (FilePath, FileDescription 'FRecipient, FileDescription 'FSender, [XFTPChunkSpec], Int64)
+    encryptFileForUpload g fileName = do
       fileSize <- fromInteger <$> getFileSize filePath
       when (fileSize > maxFileSize) $ throwError $ CLIError $ "Files bigger than " <> maxFileSizeStr <> " are not supported"
       encPath <- getEncPath tempPath "xftp"
-      key <- liftIO C.randomSbKey
-      nonce <- liftIO C.randomCbNonce
+      key <- atomically $ C.randomSbKey g
+      nonce <- atomically $ C.randomCbNonce g
       let fileHdr = smpEncode FileHeader {fileName, fileExtra = Nothing}
           fileSize' = fromIntegral (B.length fileHdr) + fileSize
           chunkSizes = prepareChunkSizes $ fileSize' + fileSizeLen + authTagSize
           defChunkSize = head chunkSizes
           chunkSizes' = map fromIntegral chunkSizes
           encSize = sum chunkSizes'
-      withExceptT (CLIError . show) $ encryptFile filePath fileHdr key nonce fileSize' encSize encPath
+          srcFile = CF.plain filePath
+      withExceptT (CLIError . show) $ encryptFile srcFile fileHdr key nonce fileSize' encSize encPath
       digest <- liftIO $ LC.sha512Hash <$> LB.readFile encPath
       let chunkSpecs = prepareChunkSpecs encPath chunkSizes
           fdRcv = FileDescription {party = SFRecipient, size = FileSize encSize, digest = FileDigest digest, key, nonce, chunkSize = FileSize defChunkSize, chunks = []}
           fdSnd = FileDescription {party = SFSender, size = FileSize encSize, digest = FileDigest digest, key, nonce, chunkSize = FileSize defChunkSize, chunks = []}
       logInfo $ "encrypted file to " <> tshow encPath
       pure (encPath, fdRcv, fdSnd, chunkSpecs, encSize)
-    uploadFile :: [XFTPChunkSpec] -> TVar [Int64] -> Int64 -> ExceptT CLIError IO [SentFileChunk]
-    uploadFile chunks uploadedChunks encSize = do
+    uploadFile :: TVar ChaChaDRG -> [XFTPChunkSpec] -> TVar [Int64] -> Int64 -> ExceptT CLIError IO [SentFileChunk]
+    uploadFile g chunks uploadedChunks encSize = do
       a <- atomically $ newXFTPAgent defaultXFTPClientAgentConfig
       gen <- newTVarIO =<< liftIO newStdGen
       let xftpSrvs = fromMaybe defaultXFTPServers (nonEmpty xftpServers)
       srvs <- liftIO $ replicateM (length chunks) $ getXFTPServer gen xftpSrvs
       let thd3 (_, _, x) = x
-          chunks' = groupBy ((==) `on` thd3) $ sortOn thd3 $ zip3 [1 ..] chunks srvs
+          chunks' = groupAllOn thd3 $ zip3 [1 ..] chunks srvs
       -- TODO shuffle/unshuffle chunks
       -- the reason we don't do pooled downloads here within one server is that http2 library doesn't handle cleint concurrency, even though
       -- upload doesn't allow other requests within the same client until complete (but download does allow).
@@ -326,8 +318,8 @@ cliSendFileOpts SendOptions {filePath, outputDir, numRecipients, xftpServers, re
         uploadFileChunk :: XFTPClientAgent -> (Int, XFTPChunkSpec, XFTPServerWithAuth) -> ExceptT CLIError IO (Int, SentFileChunk)
         uploadFileChunk a (chunkNo, chunkSpec@XFTPChunkSpec {chunkSize}, ProtoServerWithAuth xftpServer auth) = do
           logInfo $ "uploading chunk " <> tshow chunkNo <> " to " <> showServer xftpServer <> "..."
-          (sndKey, spKey) <- liftIO $ C.generateSignatureKeyPair C.SEd25519
-          rKeys <- liftIO $ L.fromList <$> replicateM numRecipients (C.generateSignatureKeyPair C.SEd25519)
+          (sndKey, spKey) <- atomically $ C.generateSignatureKeyPair C.SEd25519 g
+          rKeys <- atomically $ L.fromList <$> replicateM numRecipients (C.generateSignatureKeyPair C.SEd25519 g)
           digest <- liftIO $ getChunkDigest chunkSpec
           let ch = FileInfo {sndKey, size = fromIntegral chunkSize, digest}
           c <- withRetry retryCount $ getXFTPServerClient a xftpServer
@@ -368,7 +360,7 @@ cliSendFileOpts SendOptions {filePath, outputDir, numRecipients, xftpServers, re
         rcvChunks :: [[FileChunk]]
         rcvChunks = map (sortChunks . M.elems) $ M.elems $ foldl' addRcvChunk M.empty rcvReplicas
         sortChunks :: [FileChunk] -> [FileChunk]
-        sortChunks = map reverseReplicas . sortOn (chunkNo :: FileChunk -> Int)
+        sortChunks = map reverseReplicas . sortOn (\FileChunk {chunkNo} -> chunkNo)
         reverseReplicas ch@FileChunk {replicas} = (ch :: FileChunk) {replicas = reverse replicas}
         addRcvChunk :: Map Int (Map Int FileChunk) -> SentRecipientReplica -> Map Int (Map Int FileChunk)
         addRcvChunk m SentRecipientReplica {chunkNo, server, rcvNo, replicaId, replicaKey, digest, chunkSize} =
@@ -418,36 +410,39 @@ getChunkDigest XFTPChunkSpec {filePath = chunkPath, chunkOffset, chunkSize} =
 
 cliReceiveFile :: ReceiveOptions -> ExceptT CLIError IO ()
 cliReceiveFile ReceiveOptions {fileDescription, filePath, retryCount, tempPath, verbose, yes} =
-  getFileDescription' fileDescription >>= receiveFile
+  getFileDescription' fileDescription >>= receive
   where
-    receiveFile :: ValidFileDescription 'FRecipient -> ExceptT CLIError IO ()
-    receiveFile (ValidFileDescription FileDescription {size, digest, key, nonce, chunks}) = do
+    receive :: ValidFileDescription 'FRecipient -> ExceptT CLIError IO ()
+    receive (ValidFileDescription FileDescription {size, digest, key, nonce, chunks}) = do
       encPath <- getEncPath tempPath "xftp"
       createDirectory encPath
       a <- atomically $ newXFTPAgent defaultXFTPClientAgentConfig
       liftIO $ printNoNewLine "Downloading file..."
       downloadedChunks <- newTVarIO []
-      let srv FileChunk {replicas} = server (head replicas :: FileChunkReplica)
-          srvChunks = groupBy ((==) `on` srv) $ sortOn srv chunks
-      chunkPaths <- map snd . sortOn fst . concat <$> pooledForConcurrentlyN 16 srvChunks (mapM $ downloadFileChunk a encPath size downloadedChunks)
+      let srv FileChunk {replicas} = case replicas of
+            [] -> error "empty FileChunk.replicas"
+            FileChunkReplica {server} : _ -> server
+          srvChunks = groupAllOn srv chunks
+      g <- liftIO C.newRandom
+      chunkPaths <- map snd . sortOn fst . concat <$> pooledForConcurrentlyN 16 srvChunks (mapM $ downloadFileChunk g a encPath size downloadedChunks)
       encDigest <- liftIO $ LC.sha512Hash <$> readChunks chunkPaths
       when (encDigest /= unFileDigest digest) $ throwError $ CLIError "File digest mismatch"
       encSize <- liftIO $ foldM (\s path -> (s +) . fromIntegral <$> getFileSize path) 0 chunkPaths
       when (FileSize encSize /= size) $ throwError $ CLIError "File size mismatch"
       liftIO $ printNoNewLine "Decrypting file..."
-      path <- withExceptT cliCryptoError $ decryptChunks encSize chunkPaths key nonce getFilePath
+      CryptoFile path _ <- withExceptT cliCryptoError $ decryptChunks encSize chunkPaths key nonce $ fmap CF.plain . getFilePath
       forM_ chunks $ acknowledgeFileChunk a
       whenM (doesPathExist encPath) $ removeDirectoryRecursive encPath
       liftIO $ do
         printNoNewLine $ "File downloaded: " <> path
         removeFD yes fileDescription
-    downloadFileChunk :: XFTPClientAgent -> FilePath -> FileSize Int64 -> TVar [Int64] -> FileChunk -> ExceptT CLIError IO (Int, FilePath)
-    downloadFileChunk a encPath (FileSize encSize) downloadedChunks FileChunk {chunkNo, chunkSize, digest, replicas = replica : _} = do
+    downloadFileChunk :: TVar ChaChaDRG -> XFTPClientAgent -> FilePath -> FileSize Int64 -> TVar [Int64] -> FileChunk -> ExceptT CLIError IO (Int, FilePath)
+    downloadFileChunk g a encPath (FileSize encSize) downloadedChunks FileChunk {chunkNo, chunkSize, digest, replicas = replica : _} = do
       let FileChunkReplica {server, replicaId, replicaKey} = replica
       logInfo $ "downloading chunk " <> tshow chunkNo <> " from " <> showServer server <> "..."
       chunkPath <- uniqueCombine encPath $ show chunkNo
       let chunkSpec = XFTPRcvChunkSpec chunkPath (unFileSize chunkSize) (unFileDigest digest)
-      withReconnect a server retryCount $ \c -> downloadXFTPChunk c replicaKey (unChunkReplicaId replicaId) chunkSpec
+      withReconnect a server retryCount $ \c -> downloadXFTPChunk g c replicaKey (unChunkReplicaId replicaId) chunkSpec
       logInfo $ "downloaded chunk " <> tshow chunkNo <> " to " <> T.pack chunkPath
       downloaded <- atomically . stateTVar downloadedChunks $ \cs ->
         let cs' = fromIntegral (unFileSize chunkSize) : cs in (sum cs', cs')
@@ -455,7 +450,7 @@ cliReceiveFile ReceiveOptions {fileDescription, filePath, retryCount, tempPath,
         printProgress "Downloaded" downloaded encSize
         when verbose $ putStrLn ""
       pure (chunkNo, chunkPath)
-    downloadFileChunk _ _ _ _ _ = throwError $ CLIError "chunk has no replicas"
+    downloadFileChunk _ _ _ _ _ _ = throwError $ CLIError "chunk has no replicas"
     getFilePath :: String -> ExceptT String IO FilePath
     getFilePath name =
       case filePath of
@@ -505,10 +500,9 @@ cliFileDescrInfo InfoOptions {fileDescription} = do
         printParty
         putStrLn $ "File download size: " <> strEnc size
         putStrLn "File server(s):"
-        forM_ replicas $ \srvReplicas -> do
-          let srv = replicaServer $ head srvReplicas
-              chSizes = map (\FileServerReplica {chunkSize = chSize_} -> unFileSize $ fromMaybe chunkSize chSize_) srvReplicas
-          putStrLn $ strEnc srv <> ": " <> strEnc (FileSize $ sum chSizes)
+        forM_ replicas $ \srvReplicas@(FileServerReplica {server} :| _) -> do
+          let chSizes = fmap (\FileServerReplica {chunkSize = chSize_} -> unFileSize $ fromMaybe chunkSize chSize_) srvReplicas
+          putStrLn $ strEnc server <> ": " <> strEnc (FileSize $ sum chSizes)
       where
         printParty :: IO ()
         printParty = case party of
@@ -530,7 +524,11 @@ getFileDescription' path =
 prepareChunkSizes :: Int64 -> [Word32]
 prepareChunkSizes size' = prepareSizes size'
   where
-    (smallSize, bigSize) = if size' > size34 chunkSize3 then (chunkSize2, chunkSize3) else (chunkSize1, chunkSize2)
+    (smallSize, bigSize)
+      | size' > size34 chunkSize3 = (chunkSize2, chunkSize3)
+      | otherwise = (chunkSize1, chunkSize2)
+    --  | size' > size34 chunkSize2 = (chunkSize1, chunkSize2)
+    --  | otherwise = (chunkSize0, chunkSize1)
     size34 sz = (fromIntegral sz * 3) `div` 4
     prepareSizes 0 = []
     prepareSizes size
@@ -573,11 +571,11 @@ withRetry retryCount = withRetry' retryCount . withExceptT (CLIError . show)
 removeFD :: Bool -> FilePath -> IO ()
 removeFD yes fd
   | yes = do
-    removeFile fd
-    putStrLn $ "\nFile description " <> fd <> " is deleted."
+      removeFile fd
+      putStrLn $ "\nFile description " <> fd <> " is deleted."
   | otherwise = do
-    y <- liftIO . getConfirmation $ "\nFile description " <> fd <> " can't be used again. Delete it"
-    when y $ removeFile fd
+      y <- liftIO . getConfirmation $ "\nFile description " <> fd <> " can't be used again. Delete it"
+      when y $ removeFile fd
 
 getConfirmation :: String -> IO Bool
 getConfirmation prompt = do
@@ -596,7 +594,8 @@ cliRandomFile RandomFileOptions {filePath, fileSize = FileSize size} = do
   putStrLn $ "File created: " <> filePath
   where
     saveRandomFile h sz = do
-      bytes <- getRandomBytes $ fromIntegral $ min mb' sz
+      g <- C.newRandom
+      bytes <- atomically $ C.randomBytes (fromIntegral $ min mb' sz) g
       B.hPut h bytes
       when (sz > mb') $ saveRandomFile h (sz - mb')
     mb' = mb 1

src/Simplex/FileTransfer/Crypto.hs

@@ -6,6 +6,7 @@
 
 module Simplex.FileTransfer.Crypto where
 
+import Control.Monad
 import Control.Monad.Except
 import qualified Data.Attoparsec.ByteString.Char8 as A
 import Data.Bifunctor (first)
@@ -16,6 +17,8 @@ import qualified Data.ByteString.Lazy.Char8 as LB
 import Data.Int (Int64)
 import Simplex.FileTransfer.Types (FileHeader (..), authTagSize)
 import qualified Simplex.Messaging.Crypto as C
+import Simplex.Messaging.Crypto.File (CryptoFile (..), FTCryptoError (..))
+import qualified Simplex.Messaging.Crypto.File as CF
 import Simplex.Messaging.Crypto.Lazy (LazyByteString)
 import qualified Simplex.Messaging.Crypto.Lazy as LC
 import Simplex.Messaging.Encoding
@@ -23,54 +26,55 @@ import Simplex.Messaging.Util (liftEitherWith)
 import UnliftIO
 import UnliftIO.Directory (removeFile)
 
-encryptFile :: FilePath -> ByteString -> C.SbKey -> C.CbNonce -> Int64 -> Int64 -> FilePath -> ExceptT FTCryptoError IO ()
-encryptFile filePath fileHdr key nonce fileSize' encSize encFile = do
+encryptFile :: CryptoFile -> ByteString -> C.SbKey -> C.CbNonce -> Int64 -> Int64 -> FilePath -> ExceptT FTCryptoError IO ()
+encryptFile srcFile fileHdr key nonce fileSize' encSize encFile = do
   sb <- liftEitherWith FTCECryptoError $ LC.sbInit key nonce
-  withFile filePath ReadMode $ \r -> withFile encFile WriteMode $ \w -> do
+  CF.withFile srcFile ReadMode $ \r -> withFile encFile WriteMode $ \w -> do
     let lenStr = smpEncode fileSize'
         (hdr, !sb') = LC.sbEncryptChunk sb $ lenStr <> fileHdr
         padLen = encSize - authTagSize - fileSize' - 8
     liftIO $ B.hPut w hdr
     sb2 <- encryptChunks r w (sb', fileSize' - fromIntegral (B.length fileHdr))
+    CF.hGetTag r
     sb3 <- encryptPad w (sb2, padLen)
     let tag = BA.convert $ LC.sbAuth sb3
     liftIO $ B.hPut w tag
   where
-    encryptChunks r = encryptChunks_ $ liftIO . B.hGet r . fromIntegral
+    encryptChunks r = encryptChunks_ $ liftIO . CF.hGet r . fromIntegral
     encryptPad = encryptChunks_ $ \sz -> pure $ B.replicate (fromIntegral sz) '#'
     encryptChunks_ :: (Int64 -> IO ByteString) -> Handle -> (LC.SbState, Int64) -> ExceptT FTCryptoError IO LC.SbState
     encryptChunks_ get w (!sb, !len)
       | len == 0 = pure sb
       | otherwise = do
-        let chSize = min len 65536
-        ch <- liftIO $ get chSize
-        when (B.length ch /= fromIntegral chSize) $ throwError $ FTCEFileIOError "encrypting file: unexpected EOF"
-        let (ch', sb') = LC.sbEncryptChunk sb ch
-        liftIO $ B.hPut w ch'
-        encryptChunks_ get w (sb', len - chSize)
+          let chSize = min len 65536
+          ch <- liftIO $ get chSize
+          when (B.length ch /= fromIntegral chSize) $ throwError $ FTCEFileIOError "encrypting file: unexpected EOF"
+          let (ch', sb') = LC.sbEncryptChunk sb ch
+          liftIO $ B.hPut w ch'
+          encryptChunks_ get w (sb', len - chSize)
 
-decryptChunks :: Int64 -> [FilePath] -> C.SbKey -> C.CbNonce -> (String -> ExceptT String IO String) -> ExceptT FTCryptoError IO FilePath
+decryptChunks :: Int64 -> [FilePath] -> C.SbKey -> C.CbNonce -> (String -> ExceptT String IO CryptoFile) -> ExceptT FTCryptoError IO CryptoFile
 decryptChunks _ [] _ _ _ = throwError $ FTCEInvalidHeader "empty"
-decryptChunks encSize (chPath : chPaths) key nonce getFilePath = case reverse chPaths of
+decryptChunks encSize (chPath : chPaths) key nonce getDestFile = case reverse chPaths of
   [] -> do
     (!authOk, !f) <- liftEither . first FTCECryptoError . LC.sbDecryptTailTag key nonce (encSize - authTagSize) =<< liftIO (LB.readFile chPath)
     unless authOk $ throwError FTCEInvalidAuthTag
     (FileHeader {fileName}, !f') <- parseFileHeader f
-    path <- withExceptT FTCEFileIOError $ getFilePath fileName
-    liftIO $ LB.writeFile path f'
-    pure path
+    destFile <- withExceptT FTCEFileIOError $ getDestFile fileName
+    CF.writeFile destFile f'
+    pure destFile
   lastPath : chPaths' -> do
     (state, expectedLen, ch) <- decryptFirstChunk
     (FileHeader {fileName}, ch') <- parseFileHeader ch
-    path <- withExceptT FTCEFileIOError $ getFilePath fileName
-    authOk <- liftIO . withFile path WriteMode $ \h -> do
-      liftIO $ LB.hPut h ch'
+    destFile@(CryptoFile path _) <- withExceptT FTCEFileIOError $ getDestFile fileName
+    authOk <- CF.withFile destFile WriteMode $ \h -> liftIO $ do
+      CF.hPut h ch'
       state' <- foldM (decryptChunk h) state $ reverse chPaths'
       decryptLastChunk h state' expectedLen
     unless authOk $ do
       removeFile path
       throwError FTCEInvalidAuthTag
-    pure path
+    pure destFile
     where
       decryptFirstChunk = do
         sb <- liftEitherWith FTCECryptoError $ LC.sbInit key nonce
@@ -83,7 +87,7 @@ decryptChunks encSize (chPath : chPaths) key nonce getFilePath = case reverse ch
         ch <- LB.readFile chPth
         let len' = len + LB.length ch
             (ch', sb') = LC.sbDecryptChunkLazy sb ch
-        LB.hPut h ch'
+        CF.hPut h ch'
         pure (sb', len')
       decryptLastChunk h (!sb, !len) expectedLen = do
         ch <- LB.readFile lastPath
@@ -93,7 +97,8 @@ decryptChunks encSize (chPath : chPaths) key nonce getFilePath = case reverse ch
             len' = len + LB.length ch2
             ch3 = LB.take (LB.length ch2 - len' + expectedLen) ch2
             tag :: ByteString = BA.convert (LC.sbAuth sb')
-        LB.hPut h ch3
+        CF.hPut h ch3
+        CF.hPutTag h
         pure $ B.length tag'' == 16 && BA.constEq tag'' tag
   where
     parseFileHeader :: LazyByteString -> ExceptT FTCryptoError IO (FileHeader, LazyByteString)
@@ -106,10 +111,3 @@ decryptChunks encSize (chPath : chPaths) key nonce getFilePath = case reverse ch
 
 readChunks :: [FilePath] -> IO LB.ByteString
 readChunks = foldM (\s path -> (s <>) <$> LB.readFile path) ""
-
-data FTCryptoError
-  = FTCECryptoError C.CryptoError
-  | FTCEInvalidHeader String
-  | FTCEInvalidAuthTag
-  | FTCEFileIOError String
-  deriving (Show, Eq, Exception)

src/Simplex/FileTransfer/Description.hs

@@ -1,6 +1,5 @@
 {-# LANGUAGE DataKinds #-}
 {-# LANGUAGE DeriveAnyClass #-}
-{-# LANGUAGE DeriveGeneric #-}
 {-# LANGUAGE DuplicateRecordFields #-}
 {-# LANGUAGE GADTs #-}
 {-# LANGUAGE KindSignatures #-}
@@ -8,6 +7,8 @@
 {-# LANGUAGE OverloadedStrings #-}
 {-# LANGUAGE PatternSynonyms #-}
 {-# LANGUAGE ScopedTypeVariables #-}
+{-# LANGUAGE TemplateHaskell #-}
+{-# OPTIONS_GHC -fno-warn-ambiguous-fields #-}
 
 module Simplex.FileTransfer.Description
   ( FileDescription (..),
@@ -25,7 +26,6 @@ module Simplex.FileTransfer.Description
     YAMLServerReplicas (..), -- for tests
     validateFileDescription,
     groupReplicasByServer,
-    replicaServer,
     fdSeparator,
     kb,
     mb,
@@ -36,15 +36,16 @@ where
 import Control.Applicative (optional)
 import Control.Monad ((<=<))
 import Data.Aeson (FromJSON, ToJSON)
-import qualified Data.Aeson as J
+import qualified Data.Aeson.TH as J
 import Data.Attoparsec.ByteString.Char8 (Parser)
 import qualified Data.Attoparsec.ByteString.Char8 as A
 import Data.Bifunctor (first)
 import Data.ByteString.Char8 (ByteString)
 import qualified Data.ByteString.Char8 as B
-import Data.Function (on)
 import Data.Int (Int64)
-import Data.List (foldl', groupBy, sortOn)
+import Data.List (foldl', sortOn)
+import Data.List.NonEmpty (NonEmpty (..))
+import qualified Data.List.NonEmpty as L
 import Data.Map (Map)
 import qualified Data.Map as M
 import Data.Maybe (fromMaybe)
@@ -53,11 +54,11 @@ import Data.Word (Word32)
 import qualified Data.Yaml as Y
 import Database.SQLite.Simple.FromField (FromField (..))
 import Database.SQLite.Simple.ToField (ToField (..))
-import GHC.Generics (Generic)
+import Simplex.FileTransfer.Chunks
 import Simplex.FileTransfer.Protocol
 import qualified Simplex.Messaging.Crypto as C
 import Simplex.Messaging.Encoding.String
-import Simplex.Messaging.Parsers (parseAll)
+import Simplex.Messaging.Parsers (defaultJSON, parseAll)
 import Simplex.Messaging.Protocol (XFTPServer)
 import Simplex.Messaging.Util (bshow, (<$?>))
 
@@ -148,21 +149,13 @@ data YAMLFileDescription = YAMLFileDescription
     chunkSize :: String,
     replicas :: [YAMLServerReplicas]
   }
-  deriving (Eq, Show, Generic, FromJSON)
-
-instance ToJSON YAMLFileDescription where
-  toJSON = J.genericToJSON J.defaultOptions
-  toEncoding = J.genericToEncoding J.defaultOptions
+  deriving (Eq, Show)
 
 data YAMLServerReplicas = YAMLServerReplicas
   { server :: XFTPServer,
     chunks :: [String]
   }
-  deriving (Eq, Show, Generic, FromJSON)
-
-instance ToJSON YAMLServerReplicas where
-  toJSON = J.genericToJSON J.defaultOptions
-  toEncoding = J.genericToEncoding J.defaultOptions
+  deriving (Eq, Show)
 
 data FileServerReplica = FileServerReplica
   { chunkNo :: Int,
@@ -174,6 +167,13 @@ data FileServerReplica = FileServerReplica
   }
   deriving (Show)
 
+newtype FileSize a = FileSize {unFileSize :: a}
+  deriving (Eq, Show)
+
+$(J.deriveJSON defaultJSON ''YAMLServerReplicas)
+
+$(J.deriveJSON defaultJSON ''YAMLFileDescription)
+
 instance FilePartyI p => StrEncoding (ValidFileDescription p) where
   strEncode (ValidFD fd) = strEncode fd
   strDecode s = strDecode s >>= (\(AVFD fd) -> checkParty fd)
@@ -200,7 +200,7 @@ validateFileDescription fd@FileDescription {size, chunks}
   | chunksSize chunks /= unFileSize size = Left "chunks total size is different than file size"
   | otherwise = Right $ ValidFD fd
   where
-    chunkNos = map (chunkNo :: FileChunk -> Int) chunks
+    chunkNos = map (\FileChunk {chunkNo} -> chunkNo) chunks
     chunksSize = fromIntegral . foldl' (\s FileChunk {chunkSize} -> s + unFileSize chunkSize) 0
 
 encodeFileDescription :: FileDescription p -> YAMLFileDescription
@@ -215,9 +215,6 @@ encodeFileDescription FileDescription {party, size, digest, key, nonce, chunkSiz
       replicas = encodeFileReplicas chunkSize chunks
     }
 
-newtype FileSize a = FileSize {unFileSize :: a}
-  deriving (Eq, Show)
-
 instance (Integral a, Show a) => StrEncoding (FileSize a) where
   strEncode (FileSize b)
     | b' /= 0 = bshow b
@@ -237,44 +234,27 @@ instance (Integral a, Show a) => StrEncoding (FileSize a) where
           A.decimal
         ]
 
-kb :: Integral a => a -> a
-kb n = 1024 * n
-{-# INLINE kb #-}
-
-mb :: Integral a => a -> a
-mb n = 1024 * kb n
-{-# INLINE mb #-}
-
-gb :: Integral a => a -> a
-gb n = 1024 * mb n
-{-# INLINE gb #-}
-
 instance (Integral a, Show a) => IsString (FileSize a) where
   fromString = either error id . strDecode . B.pack
 
-instance (FromField a) => FromField (FileSize a) where fromField f = FileSize <$> fromField f
+instance FromField a => FromField (FileSize a) where fromField f = FileSize <$> fromField f
 
-instance (ToField a) => ToField (FileSize a) where toField (FileSize s) = toField s
+instance ToField a => ToField (FileSize a) where toField (FileSize s) = toField s
 
-groupReplicasByServer :: FileSize Word32 -> [FileChunk] -> [[FileServerReplica]]
+groupReplicasByServer :: FileSize Word32 -> [FileChunk] -> [NonEmpty FileServerReplica]
 groupReplicasByServer defChunkSize =
-  groupBy ((==) `on` replicaServer)
-    . sortOn replicaServer
-    . unfoldChunksToReplicas defChunkSize
+  L.groupAllWith (\FileServerReplica {server} -> server) . unfoldChunksToReplicas defChunkSize
 
 encodeFileReplicas :: FileSize Word32 -> [FileChunk] -> [YAMLServerReplicas]
 encodeFileReplicas defChunkSize =
   map encodeServerReplicas . groupReplicasByServer defChunkSize
   where
-    encodeServerReplicas fs =
+    encodeServerReplicas fs@(FileServerReplica {server} :| _) =
       YAMLServerReplicas
-        { server = replicaServer $ head fs, -- groupBy guarantees that fs is not empty
-          chunks = map (B.unpack . encodeServerReplica) fs
+        { server,
+          chunks = map (B.unpack . encodeServerReplica) $ L.toList fs
         }
 
-replicaServer :: FileServerReplica -> XFTPServer
-replicaServer = server
-
 encodeServerReplica :: FileServerReplica -> ByteString
 encodeServerReplica FileServerReplica {chunkNo, replicaId, replicaKey, digest, chunkSize} =
   bshow chunkNo
@@ -326,7 +306,7 @@ foldReplicasToChunks :: FileSize Word32 -> [FileServerReplica] -> Either String
 foldReplicasToChunks defChunkSize fs = do
   sd <- foldSizesDigests fs
   -- TODO validate (check that chunks match) or in separate function
-  sortOn (chunkNo :: FileChunk -> Int) . map reverseReplicas . M.elems <$> foldChunks sd fs
+  sortOn (\FileChunk {chunkNo} -> chunkNo) . map reverseReplicas . M.elems <$> foldChunks sd fs
   where
     foldSizesDigests :: [FileServerReplica] -> Either String (Map Int (FileSize Word32), Map Int FileDigest)
     foldSizesDigests = foldl' addSizeDigest $ Right (M.empty, M.empty)

src/Simplex/FileTransfer/Protocol.hs

@@ -1,5 +1,4 @@
 {-# LANGUAGE DataKinds #-}
-{-# LANGUAGE DeriveGeneric #-}
 {-# LANGUAGE GADTs #-}
 {-# LANGUAGE LambdaCase #-}
 {-# LANGUAGE MultiParamTypeClasses #-}
@@ -7,6 +6,7 @@
 {-# LANGUAGE OverloadedStrings #-}
 {-# LANGUAGE ScopedTypeVariables #-}
 {-# LANGUAGE StandaloneDeriving #-}
+{-# LANGUAGE TemplateHaskell #-}
 {-# LANGUAGE TypeApplications #-}
 {-# LANGUAGE TypeFamilies #-}
 {-# OPTIONS_GHC -fno-warn-unticked-promoted-constructors #-}
@@ -14,8 +14,7 @@
 module Simplex.FileTransfer.Protocol where
 
 import Control.Applicative ((<|>))
-import Data.Aeson (FromJSON, ToJSON)
-import qualified Data.Aeson as J
+import qualified Data.Aeson.TH as J
 import qualified Data.Attoparsec.ByteString.Char8 as A
 import Data.Bifunctor (first)
 import Data.ByteString.Char8 (ByteString)
@@ -25,8 +24,6 @@ import Data.List.NonEmpty (NonEmpty (..))
 import Data.Maybe (isNothing)
 import Data.Type.Equality
 import Data.Word (Word32)
-import GHC.Generics (Generic)
-import Generic.Random (genericArbitraryU)
 import qualified Simplex.Messaging.Crypto as C
 import Simplex.Messaging.Encoding
 import Simplex.Messaging.Encoding.String
@@ -51,15 +48,12 @@ import Simplex.Messaging.Protocol
     encodeTransmission,
     messageTagP,
     tDecodeParseValidate,
-    tEncode,
-    tEncodeBatch,
+    tEncodeBatch1,
     tParse,
-    _smpP,
   )
 import Simplex.Messaging.Transport (SessionId, TransportError (..))
 import Simplex.Messaging.Util (bshow, (<$?>))
 import Simplex.Messaging.Version
-import Test.QuickCheck (Arbitrary (..))
 
 currentXFTPVersion :: Version
 currentXFTPVersion = 1
@@ -69,14 +63,7 @@ xftpBlockSize = 16384
 
 -- | File protocol clients
 data FileParty = FRecipient | FSender
-  deriving (Eq, Show, Generic)
-
-instance FromJSON FileParty where
-  parseJSON = J.genericParseJSON . enumJSON $ dropPrefix "F"
-
-instance ToJSON FileParty where
-  toJSON = J.genericToJSON . enumJSON $ dropPrefix "F"
-  toEncoding = J.genericToEncoding . enumJSON $ dropPrefix "F"
+  deriving (Eq, Show)
 
 data SFileParty :: FileParty -> Type where
   SFRecipient :: SFileParty FRecipient
@@ -355,11 +342,7 @@ data XFTPErrorType
     INTERNAL
   | -- | used internally, never returned by the server (to be removed)
     DUPLICATE_ -- not part of SMP protocol, used internally
-  deriving (Eq, Generic, Read, Show)
-
-instance ToJSON XFTPErrorType where
-  toJSON = J.genericToJSON $ sumTypeJSON id
-  toEncoding = J.genericToEncoding $ sumTypeJSON id
+  deriving (Eq, Read, Show)
 
 instance StrEncoding XFTPErrorType where
   strEncode = \case
@@ -367,8 +350,6 @@ instance StrEncoding XFTPErrorType where
     e -> bshow e
   strP = "CMD " *> (CMD <$> parseRead1) <|> parseRead1
 
-instance Arbitrary XFTPErrorType where arbitrary = genericArbitraryU
-
 instance Encoding XFTPErrorType where
   smpEncode = \case
     BLOCK -> "BLOCK"
@@ -421,10 +402,8 @@ xftpEncodeTransmission sessionId pKey (corrId, fId, msg) = do
     signTransmission t = ((`C.sign` t) <$> pKey, t)
 
 -- this function uses batch syntax but puts only one transmission in the batch
-xftpEncodeBatch1 :: (Maybe C.ASignature, ByteString) -> Either TransportError ByteString
-xftpEncodeBatch1 (sig, t) =
-  let t' = tEncodeBatch 1 . smpEncode . Large $ tEncode (sig, t)
-   in first (const TELargeMsg) $ C.pad t' xftpBlockSize
+xftpEncodeBatch1 :: SentRawTransmission -> Either TransportError ByteString
+xftpEncodeBatch1 t = first (const TELargeMsg) $ C.pad (tEncodeBatch1 t) xftpBlockSize
 
 xftpDecodeTransmission :: ProtocolEncoding e c => SessionId -> ByteString -> Either XFTPErrorType (SignedTransmission e c)
 xftpDecodeTransmission sessionId t = do
@@ -432,3 +411,7 @@ xftpDecodeTransmission sessionId t = do
   case tParse True t' of
     t'' :| [] -> Right $ tDecodeParseValidate sessionId currentXFTPVersion t''
     _ -> Left BLOCK
+
+$(J.deriveJSON (enumJSON $ dropPrefix "F") ''FileParty)
+
+$(J.deriveJSON (sumTypeJSON id) ''XFTPErrorType)

src/Simplex/FileTransfer/Server.hs

@@ -1,3 +1,4 @@
+{-# LANGUAGE BangPatterns #-}
 {-# LANGUAGE DataKinds #-}
 {-# LANGUAGE DuplicateRecordFields #-}
 {-# LANGUAGE FlexibleContexts #-}
@@ -12,10 +13,9 @@
 module Simplex.FileTransfer.Server where
 
 import Control.Logger.Simple
+import Control.Monad
 import Control.Monad.Except
-import Control.Monad.IO.Unlift (MonadUnliftIO)
 import Control.Monad.Reader
-import Crypto.Random (getRandomBytes)
 import Data.Bifunctor (first)
 import qualified Data.ByteString.Base64.URL as B64
 import Data.ByteString.Builder (byteString)
@@ -32,9 +32,13 @@ import Data.Time.Clock (UTCTime (..), diffTimeToPicoseconds, getCurrentTime)
 import Data.Time.Clock.System (SystemTime (..), getSystemTime)
 import Data.Time.Format.ISO8601 (iso8601Show)
 import Data.Word (Word32)
+import GHC.IO.Handle (hSetNewlineMode)
+import GHC.Stats (getRTSStats)
 import qualified Network.HTTP.Types as N
 import qualified Network.HTTP2.Server as H
+import Network.Socket
 import Simplex.FileTransfer.Protocol
+import Simplex.FileTransfer.Server.Control
 import Simplex.FileTransfer.Server.Env
 import Simplex.FileTransfer.Server.Stats
 import Simplex.FileTransfer.Server.Store
@@ -47,17 +51,18 @@ import Simplex.Messaging.Protocol (CorrId, RcvPublicDhKey, RcvPublicVerifyKey, R
 import Simplex.Messaging.Server (dummyVerifyCmd, verifyCmdSignature)
 import Simplex.Messaging.Server.Expiration
 import Simplex.Messaging.Server.Stats
+import Simplex.Messaging.Transport.Buffer (trimCR)
 import Simplex.Messaging.Transport.HTTP2
 import Simplex.Messaging.Transport.HTTP2.Server
+import Simplex.Messaging.Transport.Server (runTCPServer)
 import Simplex.Messaging.Util
 import System.Exit (exitFailure)
 import System.FilePath ((</>))
-import System.IO (BufferMode (..), hPutStrLn, hSetBuffering)
-import UnliftIO (IOMode (..), withFile)
+import System.IO (hPrint, hPutStrLn, universalNewlineMode)
+import UnliftIO
 import UnliftIO.Concurrent (threadDelay)
 import UnliftIO.Directory (doesFileExist, removeFile, renameFile)
-import UnliftIO.Exception
-import UnliftIO.STM
+import qualified UnliftIO.Exception as E
 
 type M a = ReaderT XFTPEnv IO a
 
@@ -70,16 +75,16 @@ runXFTPServerBlocking :: TMVar Bool -> XFTPServerConfig -> IO ()
 runXFTPServerBlocking started cfg = newXFTPServerEnv cfg >>= runReaderT (xftpServer cfg started)
 
 xftpServer :: XFTPServerConfig -> TMVar Bool -> M ()
-xftpServer cfg@XFTPServerConfig {xftpPort, logTLSErrors} started = do
+xftpServer cfg@XFTPServerConfig {xftpPort, transportConfig, inactiveClientExpiration} started = do
   restoreServerStats
-  raceAny_ (runServer : expireFilesThread_ cfg <> serverStatsThread_ cfg) `finally` stopServer
+  raceAny_ (runServer : expireFilesThread_ cfg <> serverStatsThread_ cfg <> controlPortThread_ cfg) `finally` stopServer
   where
     runServer :: M ()
     runServer = do
       serverParams <- asks tlsServerParams
       env <- ask
       liftIO $
-        runHTTP2Server started xftpPort defaultHTTP2BufferSize serverParams logTLSErrors $ \sessionId r sendResponse -> do
+        runHTTP2Server started xftpPort defaultHTTP2BufferSize serverParams transportConfig inactiveClientExpiration $ \sessionId r sendResponse -> do
           reqBody <- getHTTP2Body r xftpBlockSize
           processRequest HTTP2Request {sessionId, request = r, reqBody, sendResponse} `runReaderT` env
 
@@ -103,16 +108,19 @@ xftpServer cfg@XFTPServerConfig {xftpPort, logTLSErrors} started = do
         forM_ sIds $ \sId -> do
           threadDelay 100000
           atomically (expiredFilePath st sId old)
-            >>= mapM_ (remove $ delete st sId)
+            >>= mapM_ (maybeRemove $ delete st sId)
       where
+        maybeRemove del = maybe del (remove del)
         remove del filePath =
           ifM
             (doesFileExist filePath)
-            (removeFile filePath >> del `catch` \(e :: SomeException) -> logError $ "failed to remove expired file " <> tshow filePath <> ": " <> tshow e)
+            ((removeFile filePath >> del) `catch` \(e :: SomeException) -> logError $ "failed to remove expired file " <> tshow filePath <> ": " <> tshow e)
             del
         delete st sId = do
           withFileLog (`logDeleteFile` sId)
           void $ atomically $ deleteFile st sId
+          FileServerStats {filesExpired} <- asks serverStats
+          atomically $ modifyTVar' filesExpired (+ 1)
 
     serverStatsThread_ :: XFTPServerConfig -> [M ()]
     serverStatsThread_ XFTPServerConfig {logStatsInterval = Just interval, logStatsStartTime, serverStatsLogFile} =
@@ -124,7 +132,7 @@ xftpServer cfg@XFTPServerConfig {xftpPort, logTLSErrors} started = do
       initialDelay <- (startAt -) . fromIntegral . (`div` 1000000_000000) . diffTimeToPicoseconds . utctDayTime <$> liftIO getCurrentTime
       liftIO $ putStrLn $ "server stats log enabled: " <> statsFilePath
       liftIO $ threadDelay' $ 1_000_000 * (initialDelay + if initialDelay < 0 then 86_400 else 0)
-      FileServerStats {fromTime, filesCreated, fileRecipients, filesUploaded, filesDeleted, filesDownloaded, fileDownloads, fileDownloadAcks, filesCount, filesSize} <- asks serverStats
+      FileServerStats {fromTime, filesCreated, fileRecipients, filesUploaded, filesExpired, filesDeleted, filesDownloaded, fileDownloads, fileDownloadAcks, filesCount, filesSize} <- asks serverStats
       let interval = 1_000_000 * logInterval
       forever $ do
         withFile statsFilePath AppendMode $ \h -> liftIO $ do
@@ -134,12 +142,13 @@ xftpServer cfg@XFTPServerConfig {xftpPort, logTLSErrors} started = do
           filesCreated' <- atomically $ swapTVar filesCreated 0
           fileRecipients' <- atomically $ swapTVar fileRecipients 0
           filesUploaded' <- atomically $ swapTVar filesUploaded 0
+          filesExpired' <- atomically $ swapTVar filesExpired 0
           filesDeleted' <- atomically $ swapTVar filesDeleted 0
           files <- atomically $ periodStatCounts filesDownloaded ts
           fileDownloads' <- atomically $ swapTVar fileDownloads 0
           fileDownloadAcks' <- atomically $ swapTVar fileDownloadAcks 0
-          filesCount' <- atomically $ swapTVar filesCount 0
-          filesSize' <- atomically $ swapTVar filesSize 0
+          filesCount' <- readTVarIO filesCount
+          filesSize' <- readTVarIO filesSize
           hPutStrLn h $
             intercalate
               ","
@@ -154,10 +163,52 @@ xftpServer cfg@XFTPServerConfig {xftpPort, logTLSErrors} started = do
                 show fileDownloads',
                 show fileDownloadAcks',
                 show filesCount',
-                show filesSize'
+                show filesSize',
+                show filesExpired'
               ]
         liftIO $ threadDelay' interval
 
+    controlPortThread_ :: XFTPServerConfig -> [M ()]
+    controlPortThread_ XFTPServerConfig {controlPort = Just port} = [runCPServer port]
+    controlPortThread_ _ = []
+
+    runCPServer :: ServiceName -> M ()
+    runCPServer port = do
+      cpStarted <- newEmptyTMVarIO
+      u <- askUnliftIO
+      liftIO $ do
+        labelMyThread "control port server"
+        runTCPServer cpStarted port $ runCPClient u
+      where
+        runCPClient :: UnliftIO (ReaderT XFTPEnv IO) -> Socket -> IO ()
+        runCPClient u sock = do
+          labelMyThread "control port client"
+          h <- socketToHandle sock ReadWriteMode
+          hSetBuffering h LineBuffering
+          hSetNewlineMode h universalNewlineMode
+          hPutStrLn h "XFTP server control port\n'help' for supported commands"
+          cpLoop h
+          where
+            cpLoop h = do
+              s <- B.hGetLine h
+              case strDecode $ trimCR s of
+                Right CPQuit -> hClose h
+                Right cmd -> processCP h cmd >> cpLoop h
+                Left err -> hPutStrLn h ("error: " <> err) >> cpLoop h
+            processCP h = \case
+              CPStatsRTS -> E.tryAny getRTSStats >>= either (hPrint h) (hPrint h)
+              CPDelete fileId -> unliftIO u $ do
+                fs <- asks store
+                r <- runExceptT $ do
+                  let asSender = ExceptT . atomically $ getFile fs SFSender fileId
+                  let asRecipient = ExceptT . atomically $ getFile fs SFRecipient fileId
+                  (fr, _) <- asSender `catchError` const asRecipient
+                  ExceptT $ deleteServerFile_ fr
+                liftIO . hPutStrLn h $ either (\e -> "error: " <> show e) (\() -> "ok") r
+              CPHelp -> hPutStrLn h "commands: stats-rts, delete, help, quit"
+              CPQuit -> pure ()
+              CPSkip -> pure ()
+
 data ServerFile = ServerFile
   { filePath :: FilePath,
     fileSize :: Word32,
@@ -168,17 +219,17 @@ processRequest :: HTTP2Request -> M ()
 processRequest HTTP2Request {sessionId, reqBody = body@HTTP2Body {bodyHead}, sendResponse}
   | B.length bodyHead /= xftpBlockSize = sendXFTPResponse ("", "", FRErr BLOCK) Nothing
   | otherwise = do
-    case xftpDecodeTransmission sessionId bodyHead of
-      Right (sig_, signed, (corrId, fId, cmdOrErr)) -> do
-        case cmdOrErr of
-          Right cmd -> do
-            verifyXFTPTransmission sig_ signed fId cmd >>= \case
-              VRVerified req -> uncurry send =<< processXFTPRequest body req
-              VRFailed -> send (FRErr AUTH) Nothing
-          Left e -> send (FRErr e) Nothing
-        where
-          send resp = sendXFTPResponse (corrId, fId, resp)
-      Left e -> sendXFTPResponse ("", "", FRErr e) Nothing
+      case xftpDecodeTransmission sessionId bodyHead of
+        Right (sig_, signed, (corrId, fId, cmdOrErr)) -> do
+          case cmdOrErr of
+            Right cmd -> do
+              verifyXFTPTransmission sig_ signed fId cmd >>= \case
+                VRVerified req -> uncurry send =<< processXFTPRequest body req
+                VRFailed -> send (FRErr AUTH) Nothing
+            Left e -> send (FRErr e) Nothing
+          where
+            send resp = sendXFTPResponse (corrId, fId, resp)
+        Left e -> sendXFTPResponse ("", "", FRErr e) Nothing
   where
     sendXFTPResponse :: (CorrId, XFTPFileId, FileResponse) -> Maybe ServerFile -> M ()
     sendXFTPResponse (corrId, fId, resp) serverFile_ = do
@@ -317,9 +368,10 @@ processXFTPRequest HTTP2Body {bodyPart} = \case
     sendServerFile FileRec {senderId, filePath, fileInfo = FileInfo {size}} rDhKey = do
       readTVarIO filePath >>= \case
         Just path -> do
-          (sDhKey, spDhKey) <- liftIO C.generateKeyPair'
+          g <- asks random
+          (sDhKey, spDhKey) <- atomically $ C.generateKeyPair g
           let dhSecret = C.dh' rDhKey spDhKey
-          cbNonce <- liftIO C.randomCbNonce
+          cbNonce <- atomically $ C.randomCbNonce g
           case LC.cbInit dhSecret cbNonce of
             Right sbState -> do
               stats <- asks serverStats
@@ -330,21 +382,7 @@ processXFTPRequest HTTP2Body {bodyPart} = \case
         _ -> pure (FRErr NO_FILE, Nothing)
 
     deleteServerFile :: FileRec -> M FileResponse
-    deleteServerFile FileRec {senderId, fileInfo, filePath} = do
-      withFileLog (`logDeleteFile` senderId)
-      r <- runExceptT $ do
-        path <- readTVarIO filePath
-        stats <- asks serverStats
-        ExceptT $ first (\(_ :: SomeException) -> FILE_IO) <$> try (forM_ path $ \p -> whenM (doesFileExist p) (removeFile p >> deletedStats stats))
-        st <- asks store
-        void $ atomically $ deleteFile st senderId
-        atomically $ modifyTVar' (filesDeleted stats) (+ 1)
-        pure FROk
-      either (pure . FRErr) pure r
-      where
-        deletedStats stats = do
-          atomically $ modifyTVar' (filesCount stats) (subtract 1)
-          atomically $ modifyTVar' (filesSize stats) (subtract $ fromIntegral $ size fileInfo)
+    deleteServerFile fr = either FRErr (\() -> FROk) <$> deleteServerFile_ fr
 
     logFileError :: SomeException -> IO ()
     logFileError e = logError $ "Error deleting file: " <> tshow e
@@ -358,13 +396,28 @@ processXFTPRequest HTTP2Body {bodyPart} = \case
       atomically $ modifyTVar' (fileDownloadAcks stats) (+ 1)
       pure FROk
 
+deleteServerFile_ :: FileRec -> M (Either XFTPErrorType ())
+deleteServerFile_ FileRec {senderId, fileInfo, filePath} = do
+  withFileLog (`logDeleteFile` senderId)
+  runExceptT $ do
+    path <- readTVarIO filePath
+    stats <- asks serverStats
+    ExceptT $ first (\(_ :: SomeException) -> FILE_IO) <$> try (forM_ path $ \p -> whenM (doesFileExist p) (removeFile p >> deletedStats stats))
+    st <- asks store
+    void $ atomically $ deleteFile st senderId
+    atomically $ modifyTVar' (filesDeleted stats) (+ 1)
+  where
+    deletedStats stats = do
+      atomically $ modifyTVar' (filesCount stats) (subtract 1)
+      atomically $ modifyTVar' (filesSize stats) (subtract $ fromIntegral $ size fileInfo)
+
 randomId :: (MonadUnliftIO m, MonadReader XFTPEnv m) => Int -> m ByteString
-randomId n = do
-  gVar <- asks idsDrg
-  atomically (C.pseudoRandomBytes n gVar)
+randomId n = atomically . C.randomBytes n =<< asks random
 
 getFileId :: M XFTPFileId
-getFileId = liftIO . getRandomBytes =<< asks (fileIdSize . config)
+getFileId = do
+  size <- asks (fileIdSize . config)
+  atomically . C.randomBytes size =<< asks random
 
 withFileLog :: (MonadIO m, MonadReader XFTPEnv m) => (StoreLog 'WriteMode -> IO a) -> m ()
 withFileLog action = liftIO . mapM_ action =<< asks storeLog
@@ -390,14 +443,17 @@ restoreServerStats = asks (serverStatsBackupFile . config) >>= mapM_ restoreStat
     restoreStats f = whenM (doesFileExist f) $ do
       logInfo $ "restoring server stats from file " <> T.pack f
       liftIO (strDecode <$> B.readFile f) >>= \case
-        Right d -> do
+        Right d@FileServerStatsData {_filesCount = statsFilesCount, _filesSize = statsFilesSize} -> do
           s <- asks serverStats
-          fs <- readTVarIO . files =<< asks store
-          let _filesCount = length $ M.keys fs
-              _filesSize = M.foldl' (\n -> (n +) . fromIntegral . size . fileInfo) 0 fs
+          FileStore {files, usedStorage} <- asks store
+          _filesCount <- M.size <$> readTVarIO files
+          _filesSize <- readTVarIO usedStorage
           atomically $ setFileServerStats s d {_filesCount, _filesSize}
           renameFile f $ f <> ".bak"
           logInfo "server stats restored"
+          when (statsFilesCount /= _filesCount) $ logWarn $ "Files count differs: stats: " <> tshow statsFilesCount <> ", store: " <> tshow _filesCount
+          when (statsFilesSize /= _filesSize) $ logWarn $ "Files size differs: stats: " <> tshow statsFilesSize <> ", store: " <> tshow _filesSize
+          logInfo $ "Restored " <> tshow (_filesSize `div` 1048576) <> " MBs in " <> tshow _filesCount <> " files"
         Left e -> do
           logInfo $ "error restoring server stats: " <> T.pack e
           liftIO exitFailure

src/Simplex/FileTransfer/Server/Control.hs

@@ -0,0 +1,31 @@
+{-# LANGUAGE LambdaCase #-}
+{-# LANGUAGE OverloadedStrings #-}
+
+module Simplex.FileTransfer.Server.Control where
+
+import qualified Data.Attoparsec.ByteString.Char8 as A
+import Data.ByteString (ByteString)
+import Simplex.Messaging.Encoding.String
+
+data ControlProtocol
+  = CPStatsRTS
+  | CPDelete ByteString
+  | CPHelp
+  | CPQuit
+  | CPSkip
+
+instance StrEncoding ControlProtocol where
+  strEncode = \case
+    CPStatsRTS -> "stats-rts"
+    CPDelete bs -> "delete " <> strEncode bs
+    CPHelp -> "help"
+    CPQuit -> "quit"
+    CPSkip -> ""
+  strP =
+    A.takeTill (== ' ') >>= \case
+      "stats-rts" -> pure CPStatsRTS
+      "delete" -> CPDelete <$> (A.space *> strP)
+      "help" -> pure CPHelp
+      "quit" -> pure CPQuit
+      "" -> pure CPSkip
+      _ -> fail "bad ControlProtocol command"

src/Simplex/FileTransfer/Server/Env.hs

@@ -26,13 +26,14 @@ import Simplex.FileTransfer.Server.StoreLog
 import qualified Simplex.Messaging.Crypto as C
 import Simplex.Messaging.Protocol (BasicAuth, RcvPublicVerifyKey)
 import Simplex.Messaging.Server.Expiration
-import Simplex.Messaging.Transport.Server (loadFingerprint, loadTLSServerParams)
+import Simplex.Messaging.Transport.Server (TransportServerConfig, loadFingerprint, loadTLSServerParams)
 import Simplex.Messaging.Util (tshow)
 import System.IO (IOMode (..))
 import UnliftIO.STM
 
 data XFTPServerConfig = XFTPServerConfig
   { xftpPort :: ServiceName,
+    controlPort :: Maybe ServiceName,
     fileIdSize :: Int,
     storeLogFile :: Maybe FilePath,
     filesPath :: FilePath,
@@ -46,6 +47,8 @@ data XFTPServerConfig = XFTPServerConfig
     newFileBasicAuth :: Maybe BasicAuth,
     -- | time after which the files can be removed and check interval, seconds
     fileExpiration :: Maybe ExpirationConfig,
+    -- | time after which inactive clients can be disconnected and check interval, seconds
+    inactiveClientExpiration :: Maybe ExpirationConfig,
     -- CA certificate private key is not needed for initialization
     caCertificateFile :: FilePath,
     privateKeyFile :: FilePath,
@@ -55,14 +58,21 @@ data XFTPServerConfig = XFTPServerConfig
     logStatsStartTime :: Int64,
     serverStatsLogFile :: FilePath,
     serverStatsBackupFile :: Maybe FilePath,
-    logTLSErrors :: Bool
+    transportConfig :: TransportServerConfig
   }
 
+defaultInactiveClientExpiration :: ExpirationConfig
+defaultInactiveClientExpiration =
+  ExpirationConfig
+    { ttl = 43200, -- seconds, 12 hours
+      checkInterval = 3600 -- seconds, 1 hours
+    }
+
 data XFTPEnv = XFTPEnv
   { config :: XFTPServerConfig,
     store :: FileStore,
     storeLog :: Maybe (StoreLog 'WriteMode),
-    idsDrg :: TVar ChaChaDRG,
+    random :: TVar ChaChaDRG,
     serverIdentity :: C.KeyHash,
     tlsServerParams :: T.ServerParams,
     serverStats :: FileServerStats
@@ -80,7 +90,7 @@ defaultFileExpiration =
 
 newXFTPServerEnv :: (MonadUnliftIO m, MonadRandom m) => XFTPServerConfig -> m XFTPEnv
 newXFTPServerEnv config@XFTPServerConfig {storeLogFile, fileSizeQuota, caCertificateFile, certificateFile, privateKeyFile} = do
-  idsDrg <- drgNew >>= newTVarIO
+  random <- liftIO C.newRandom
   store <- atomically newFileStore
   storeLog <- liftIO $ mapM (`readWriteFileStore` store) storeLogFile
   used <- readTVarIO (usedStorage store)
@@ -90,7 +100,7 @@ newXFTPServerEnv config@XFTPServerConfig {storeLogFile, fileSizeQuota, caCertifi
   tlsServerParams <- liftIO $ loadTLSServerParams caCertificateFile certificateFile privateKeyFile
   Fingerprint fp <- liftIO $ loadFingerprint caCertificateFile
   serverStats <- atomically . newFileServerStats =<< liftIO getCurrentTime
-  pure XFTPEnv {config, store, storeLog, idsDrg, tlsServerParams, serverIdentity = C.KeyHash fp, serverStats}
+  pure XFTPEnv {config, store, storeLog, random, tlsServerParams, serverIdentity = C.KeyHash fp, serverStats}
 
 data XFTPRequest
   = XFTPReqNew FileInfo (NonEmpty RcvPublicVerifyKey) (Maybe BasicAuth)

src/Simplex/FileTransfer/Server/Main.hs

@@ -16,22 +16,24 @@ import Data.Maybe (fromMaybe)
 import qualified Data.Text as T
 import Network.Socket (HostName)
 import Options.Applicative
-import Simplex.FileTransfer.Description (FileSize (..), kb, mb)
+import Simplex.FileTransfer.Chunks
+import Simplex.FileTransfer.Description (FileSize (..))
 import Simplex.FileTransfer.Server (runXFTPServer)
-import Simplex.FileTransfer.Server.Env (XFTPServerConfig (..), defaultFileExpiration, defFileExpirationHours)
+import Simplex.FileTransfer.Server.Env (XFTPServerConfig (..), defFileExpirationHours, defaultFileExpiration, defaultInactiveClientExpiration)
 import qualified Simplex.Messaging.Crypto as C
 import Simplex.Messaging.Encoding.String
 import Simplex.Messaging.Protocol (ProtoServerWithAuth (..), pattern XFTPServer)
 import Simplex.Messaging.Server.CLI
 import Simplex.Messaging.Server.Expiration
 import Simplex.Messaging.Transport.Client (TransportHost (..))
+import Simplex.Messaging.Transport.Server (TransportServerConfig (..), defaultTransportServerConfig)
 import System.Directory (createDirectoryIfMissing, doesFileExist)
 import System.FilePath (combine)
 import System.IO (BufferMode (..), hSetBuffering, stderr, stdout)
 import Text.Read (readMaybe)
 
 xftpServerVersion :: String
-xftpServerVersion = "1.0.0"
+xftpServerVersion = "1.2.0.6"
 
 xftpServerCLI :: FilePath -> FilePath -> IO ()
 xftpServerCLI cfgPath logPath = do
@@ -98,10 +100,17 @@ xftpServerCLI cfgPath logPath = do
             <> ("host: " <> host <> "\n")
             <> ("port: " <> defaultServerPort <> "\n")
             <> "log_tls_errors: off\n\
+               \# control_port: 5226\n\
                \\n\
                \[FILES]\n"
             <> ("path: " <> filesPath <> "\n")
             <> ("storage_quota: " <> B.unpack (strEncode fileSizeQuota) <> "\n")
+            <> "\n\
+               \[INACTIVE_CLIENTS]\n\
+               \# TTL and interval to check inactive clients\n\
+               \disconnect: off\n"
+            <> ("# ttl: " <> show (ttl defaultInactiveClientExpiration) <> "\n")
+            <> ("# check_interval: " <> show (checkInterval defaultInactiveClientExpiration) <> "\n")
     runServer ini = do
       hSetBuffering stdout LineBuffering
       hSetBuffering stderr LineBuffering
@@ -116,13 +125,16 @@ xftpServerCLI cfgPath logPath = do
         enableStoreLog = settingIsOn "STORE_LOG" "enable" ini
         logStats = settingIsOn "STORE_LOG" "log_stats" ini
         c = combine cfgPath . ($ defaultX509Config)
-        printXFTPConfig XFTPServerConfig {allowNewFiles, newFileBasicAuth, xftpPort, storeLogFile, fileExpiration} = do
+        printXFTPConfig XFTPServerConfig {allowNewFiles, newFileBasicAuth, xftpPort, storeLogFile, fileExpiration, inactiveClientExpiration} = do
           putStrLn $ case storeLogFile of
             Just f -> "Store log: " <> f
             _ -> "Store log disabled."
           putStrLn $ case fileExpiration of
             Just ExpirationConfig {ttl} -> "expiring files after " <> showTTL ttl
             _ -> "not expiring files"
+          putStrLn $ case inactiveClientExpiration of
+            Just ExpirationConfig {ttl, checkInterval} -> "expiring clients inactive for " <> show ttl <> " seconds every " <> show checkInterval <> " seconds"
+            _ -> "not expiring inactive clients"
           putStrLn $
             "Uploading new files "
               <> if allowNewFiles
@@ -133,17 +145,25 @@ xftpServerCLI cfgPath logPath = do
         serverConfig =
           XFTPServerConfig
             { xftpPort = T.unpack $ strictIni "TRANSPORT" "port" ini,
+              controlPort = either (const Nothing) (Just . T.unpack) $ lookupValue "TRANSPORT" "control_port" ini,
               fileIdSize = 16,
               storeLogFile = enableStoreLog $> storeLogFilePath,
               filesPath = T.unpack $ strictIni "FILES" "path" ini,
               fileSizeQuota = either error unFileSize <$> strDecodeIni "FILES" "storage_quota" ini,
-              allowedChunkSizes = [kb 256, mb 1, mb 4],
+              allowedChunkSizes = serverChunkSizes,
               allowNewFiles = fromMaybe True $ iniOnOff "AUTH" "new_files" ini,
               newFileBasicAuth = either error id <$> strDecodeIni "AUTH" "create_password" ini,
               fileExpiration =
-                Just defaultFileExpiration
-                  { ttl = 3600 * readIniDefault defFileExpirationHours "STORE_LOG" "expire_files_hours" ini
-                  },
+                Just
+                  defaultFileExpiration
+                    { ttl = 3600 * readIniDefault defFileExpirationHours "STORE_LOG" "expire_files_hours" ini
+                    },
+              inactiveClientExpiration =
+                settingIsOn "INACTIVE_CLIENTS" "disconnect" ini
+                  $> ExpirationConfig
+                    { ttl = readStrictIni "INACTIVE_CLIENTS" "ttl" ini,
+                      checkInterval = readStrictIni "INACTIVE_CLIENTS" "check_interval" ini
+                    },
               caCertificateFile = c caCrtFile,
               privateKeyFile = c serverKeyFile,
               certificateFile = c serverCrtFile,
@@ -151,7 +171,10 @@ xftpServerCLI cfgPath logPath = do
               logStatsStartTime = 0, -- seconds from 00:00 UTC
               serverStatsLogFile = combine logPath "file-server-stats.daily.log",
               serverStatsBackupFile = logStats $> combine logPath "file-server-stats.log",
-              logTLSErrors = fromMaybe False $ iniOnOff "TRANSPORT" "log_tls_errors" ini
+              transportConfig =
+                defaultTransportServerConfig
+                  { logTLSErrors = fromMaybe False $ iniOnOff "TRANSPORT" "log_tls_errors" ini
+                  }
             }
 
 data CliCommand

src/Simplex/FileTransfer/Server/Stats.hs

@@ -4,6 +4,7 @@
 
 module Simplex.FileTransfer.Server.Stats where
 
+import Control.Applicative ((<|>))
 import qualified Data.Attoparsec.ByteString.Char8 as A
 import qualified Data.ByteString.Char8 as B
 import Data.Int (Int64)
@@ -18,6 +19,7 @@ data FileServerStats = FileServerStats
     filesCreated :: TVar Int,
     fileRecipients :: TVar Int,
     filesUploaded :: TVar Int,
+    filesExpired :: TVar Int,
     filesDeleted :: TVar Int,
     filesDownloaded :: PeriodStats SenderId,
     fileDownloads :: TVar Int,
@@ -31,6 +33,7 @@ data FileServerStatsData = FileServerStatsData
     _filesCreated :: Int,
     _fileRecipients :: Int,
     _filesUploaded :: Int,
+    _filesExpired :: Int,
     _filesDeleted :: Int,
     _filesDownloaded :: PeriodStatsData SenderId,
     _fileDownloads :: Int,
@@ -46,13 +49,14 @@ newFileServerStats ts = do
   filesCreated <- newTVar 0
   fileRecipients <- newTVar 0
   filesUploaded <- newTVar 0
+  filesExpired <- newTVar 0
   filesDeleted <- newTVar 0
   filesDownloaded <- newPeriodStats
   fileDownloads <- newTVar 0
   fileDownloadAcks <- newTVar 0
   filesCount <- newTVar 0
   filesSize <- newTVar 0
-  pure FileServerStats {fromTime, filesCreated, fileRecipients, filesUploaded, filesDeleted, filesDownloaded, fileDownloads, fileDownloadAcks, filesCount, filesSize}
+  pure FileServerStats {fromTime, filesCreated, fileRecipients, filesUploaded, filesExpired, filesDeleted, filesDownloaded, fileDownloads, fileDownloadAcks, filesCount, filesSize}
 
 getFileServerStatsData :: FileServerStats -> STM FileServerStatsData
 getFileServerStatsData s = do
@@ -60,13 +64,14 @@ getFileServerStatsData s = do
   _filesCreated <- readTVar $ filesCreated s
   _fileRecipients <- readTVar $ fileRecipients s
   _filesUploaded <- readTVar $ filesUploaded s
+  _filesExpired <- readTVar $ filesExpired s
   _filesDeleted <- readTVar $ filesDeleted s
   _filesDownloaded <- getPeriodStatsData $ filesDownloaded s
   _fileDownloads <- readTVar $ fileDownloads s
   _fileDownloadAcks <- readTVar $ fileDownloadAcks s
   _filesCount <- readTVar $ filesCount s
   _filesSize <- readTVar $ filesSize s
-  pure FileServerStatsData {_fromTime, _filesCreated, _fileRecipients, _filesUploaded, _filesDeleted, _filesDownloaded, _fileDownloads, _fileDownloadAcks, _filesCount, _filesSize}
+  pure FileServerStatsData {_fromTime, _filesCreated, _fileRecipients, _filesUploaded, _filesExpired, _filesDeleted, _filesDownloaded, _fileDownloads, _fileDownloadAcks, _filesCount, _filesSize}
 
 setFileServerStats :: FileServerStats -> FileServerStatsData -> STM ()
 setFileServerStats s d = do
@@ -74,6 +79,7 @@ setFileServerStats s d = do
   writeTVar (filesCreated s) $! _filesCreated d
   writeTVar (fileRecipients s) $! _fileRecipients d
   writeTVar (filesUploaded s) $! _filesUploaded d
+  writeTVar (filesExpired s) $! _filesExpired d
   writeTVar (filesDeleted s) $! _filesDeleted d
   setPeriodStats (filesDownloaded s) $! _filesDownloaded d
   writeTVar (fileDownloads s) $! _fileDownloads d
@@ -82,13 +88,16 @@ setFileServerStats s d = do
   writeTVar (filesSize s) $! _filesSize d
 
 instance StrEncoding FileServerStatsData where
-  strEncode FileServerStatsData {_fromTime, _filesCreated, _fileRecipients, _filesUploaded, _filesDeleted, _filesDownloaded, _fileDownloads, _fileDownloadAcks} =
+  strEncode FileServerStatsData {_fromTime, _filesCreated, _fileRecipients, _filesUploaded, _filesExpired, _filesDeleted, _filesDownloaded, _fileDownloads, _fileDownloadAcks, _filesCount, _filesSize} =
     B.unlines
       [ "fromTime=" <> strEncode _fromTime,
         "filesCreated=" <> strEncode _filesCreated,
         "fileRecipients=" <> strEncode _fileRecipients,
         "filesUploaded=" <> strEncode _filesUploaded,
+        "filesExpired=" <> strEncode _filesExpired,
         "filesDeleted=" <> strEncode _filesDeleted,
+        "filesCount=" <> strEncode _filesCount,
+        "filesSize=" <> strEncode _filesSize,
         "filesDownloaded:",
         strEncode _filesDownloaded,
         "fileDownloads=" <> strEncode _fileDownloads,
@@ -99,8 +108,11 @@ instance StrEncoding FileServerStatsData where
     _filesCreated <- "filesCreated=" *> strP <* A.endOfLine
     _fileRecipients <- "fileRecipients=" *> strP <* A.endOfLine
     _filesUploaded <- "filesUploaded=" *> strP <* A.endOfLine
+    _filesExpired <- "filesExpired=" *> strP <* A.endOfLine <|> pure 0
     _filesDeleted <- "filesDeleted=" *> strP <* A.endOfLine
+    _filesCount <- "filesCount=" *> strP <* A.endOfLine <|> pure 0
+    _filesSize <- "filesSize=" *> strP <* A.endOfLine <|> pure 0
     _filesDownloaded <- "filesDownloaded:" *> A.endOfLine *> strP <* A.endOfLine
     _fileDownloads <- "fileDownloads=" *> strP <* A.endOfLine
     _fileDownloadAcks <- "fileDownloadAcks=" *> strP <* A.endOfLine
-    pure FileServerStatsData {_fromTime, _filesCreated, _fileRecipients, _filesUploaded, _filesDeleted, _filesDownloaded, _fileDownloads, _fileDownloadAcks, _filesCount = 0, _filesSize = 0}
+    pure FileServerStatsData {_fromTime, _filesCreated, _fileRecipients, _filesUploaded, _filesExpired, _filesDeleted, _filesDownloaded, _fileDownloads, _fileDownloadAcks, _filesCount, _filesSize}

src/Simplex/FileTransfer/Server/Store.hs

@@ -121,12 +121,12 @@ getFile st party fId = case party of
       Just (sId, rKey) -> withFile st sId $ pure . Right . (,rKey)
       _ -> pure $ Left AUTH
 
-expiredFilePath :: FileStore -> XFTPFileId -> Int64 -> STM (Maybe FilePath)
+expiredFilePath :: FileStore -> XFTPFileId -> Int64 -> STM (Maybe (Maybe FilePath))
 expiredFilePath FileStore {files} sId old =
   TM.lookup sId files
     $>>= \FileRec {filePath, createdAt} ->
       if systemSeconds createdAt < old
-        then readTVar filePath
+        then Just <$> readTVar filePath
         else pure Nothing
 
 ackFile :: FileStore -> RecipientId -> STM (Either XFTPErrorType ())

src/Simplex/FileTransfer/Transport.hs

@@ -1,32 +1,34 @@
 {-# LANGUAGE DuplicateRecordFields #-}
-{-# LANGUAGE LambdaCase #-}
 {-# LANGUAGE MultiWayIf #-}
 {-# LANGUAGE NamedFieldPuns #-}
-{-# LANGUAGE OverloadedStrings #-}
 {-# LANGUAGE ScopedTypeVariables #-}
 
 module Simplex.FileTransfer.Transport
   ( supportedFileServerVRange,
     XFTPRcvChunkSpec (..),
-    sendFile,
+    ReceiveFileError (..),
     receiveFile,
     sendEncFile,
     receiveEncFile,
+    receiveSbFile,
   )
 where
 
 import qualified Control.Exception as E
+import Control.Monad
 import Control.Monad.Except
+import Control.Monad.IO.Class
+import Data.Bifunctor (first)
 import qualified Data.ByteArray as BA
 import Data.ByteString.Builder (Builder, byteString)
 import Data.ByteString.Char8 (ByteString)
 import qualified Data.ByteString.Char8 as B
 import qualified Data.ByteString.Lazy.Char8 as LB
 import Data.Word (Word32)
-import GHC.IO.Handle.Internals (ioe_EOF)
 import Simplex.FileTransfer.Protocol (XFTPErrorType (..))
 import qualified Simplex.Messaging.Crypto as C
 import qualified Simplex.Messaging.Crypto.Lazy as LC
+import Simplex.Messaging.Transport.HTTP2.File
 import Simplex.Messaging.Version
 import System.IO (Handle, IOMode (..), withFile)
 
@@ -40,18 +42,6 @@ data XFTPRcvChunkSpec = XFTPRcvChunkSpec
 supportedFileServerVRange :: VersionRange
 supportedFileServerVRange = mkVersionRange 1 1
 
-fileBlockSize :: Int
-fileBlockSize = 16 * 1024
-
-sendFile :: Handle -> (Builder -> IO ()) -> Word32 -> IO ()
-sendFile h send = go
-  where
-    go 0 = pure ()
-    go sz =
-      getFileChunk h sz >>= \ch -> do
-        send $ byteString ch
-        go $ sz - fromIntegral (B.length ch)
-
 sendEncFile :: Handle -> (Builder -> IO ()) -> LC.SbState -> Word32 -> IO ()
 sendEncFile h send = go
   where
@@ -64,46 +54,42 @@ sendEncFile h send = go
         send (byteString encCh) `E.catch` \(e :: E.SomeException) -> print e >> E.throwIO e
         go sbState' $ sz - fromIntegral (B.length ch)
 
-getFileChunk :: Handle -> Word32 -> IO ByteString
-getFileChunk h sz =
-  B.hGet h fileBlockSize >>= \case
-    "" -> ioe_EOF
-    ch -> pure $ B.take (fromIntegral sz) ch -- sz >= xftpBlockSize
-
 receiveFile :: (Int -> IO ByteString) -> XFTPRcvChunkSpec -> ExceptT XFTPErrorType IO ()
 receiveFile getBody = receiveFile_ receive
   where
-    receive h sz = do
-      ch <- getBody fileBlockSize
-      let chSize = fromIntegral $ B.length ch
-      if
-          | chSize > sz -> pure $ Left SIZE
-          | chSize > 0 -> B.hPut h ch >> receive h (sz - chSize)
-          | sz == 0 -> pure $ Right ()
-          | otherwise -> pure $ Left SIZE
+    receive h sz = hReceiveFile getBody h sz >>= \sz' -> pure $ if sz' == 0 then Right () else Left SIZE
 
 receiveEncFile :: (Int -> IO ByteString) -> LC.SbState -> XFTPRcvChunkSpec -> ExceptT XFTPErrorType IO ()
 receiveEncFile getBody = receiveFile_ . receive
   where
-    receive sbState h sz = do
+    receive sbState h sz = first err <$> receiveSbFile getBody h sbState sz
+    err RFESize = SIZE
+    err RFECrypto = CRYPTO
+
+data ReceiveFileError = RFESize | RFECrypto
+
+receiveSbFile :: (Int -> IO ByteString) -> Handle -> LC.SbState -> Word32 -> IO (Either ReceiveFileError ())
+receiveSbFile getBody h = receive
+  where
+    receive sbState sz = do
       ch <- getBody fileBlockSize
       let chSize = fromIntegral $ B.length ch
       if
-          | chSize > sz + authSz -> pure $ Left SIZE
-          | chSize > 0 -> do
+        | chSize > sz + authSz -> pure $ Left RFESize
+        | chSize > 0 -> do
             let (ch', rest) = B.splitAt (fromIntegral sz) ch
                 (decCh, sbState') = LC.sbDecryptChunk sbState ch'
                 sz' = sz - fromIntegral (B.length ch')
             B.hPut h decCh
             if sz' > 0
-              then receive sbState' h sz'
+              then receive sbState' sz'
               else do
                 let tag' = B.take C.authTagSize rest
                     tagSz = B.length tag'
                     tag = LC.sbAuth sbState'
                 tag'' <- if tagSz == C.authTagSize then pure tag' else (tag' <>) <$> getBody (C.authTagSize - tagSz)
-                pure $ if BA.constEq tag'' tag then Right () else Left CRYPTO
-          | otherwise -> pure $ Left SIZE
+                pure $ if BA.constEq tag'' tag then Right () else Left RFECrypto
+        | otherwise -> pure $ Left RFESize
     authSz = fromIntegral C.authTagSize
 
 receiveFile_ :: (Handle -> Word32 -> IO (Either XFTPErrorType ())) -> XFTPRcvChunkSpec -> ExceptT XFTPErrorType IO ()

src/Simplex/FileTransfer/Types.hs

@@ -13,6 +13,7 @@ import Simplex.FileTransfer.Client (XFTPChunkSpec (..))
 import Simplex.FileTransfer.Description
 import Simplex.Messaging.Agent.Protocol (RcvFileId, SndFileId)
 import qualified Simplex.Messaging.Crypto as C
+import Simplex.Messaging.Crypto.File (CryptoFile (..))
 import Simplex.Messaging.Encoding
 import Simplex.Messaging.Encoding.String
 import Simplex.Messaging.Parsers (fromTextField_)
@@ -49,7 +50,7 @@ data RcvFile = RcvFile
     chunks :: [RcvFileChunk],
     prefixPath :: FilePath,
     tmpPath :: Maybe FilePath,
-    savePath :: FilePath,
+    saveFile :: CryptoFile,
     status :: RcvFileStatus,
     deleted :: Bool
   }
@@ -120,7 +121,7 @@ data SndFile = SndFile
     key :: C.SbKey,
     nonce :: C.CbNonce,
     chunks :: [SndFileChunk],
-    filePath :: FilePath,
+    srcFile :: CryptoFile,
     prefixPath :: Maybe FilePath,
     status :: SndFileStatus,
     deleted :: Bool

src/Simplex/Messaging/Agent/Env/SQLite.hs

@@ -6,6 +6,7 @@
 {-# LANGUAGE NamedFieldPuns #-}
 {-# LANGUAGE NumericUnderscores #-}
 {-# LANGUAGE RankNTypes #-}
+{-# LANGUAGE ScopedTypeVariables #-}
 {-# LANGUAGE TypeApplications #-}
 {-# OPTIONS_GHC -fno-warn-unticked-promoted-constructors #-}
 
@@ -17,23 +18,33 @@ module Simplex.Messaging.Agent.Env.SQLite
     NetworkConfig (..),
     defaultAgentConfig,
     defaultReconnectInterval,
+    tryAgentError,
+    tryAgentError',
+    catchAgentError,
+    agentFinally,
     Env (..),
     newSMPAgentEnv,
     createAgentStore,
     NtfSupervisor (..),
     NtfSupervisorCommand (..),
     XFTPAgent (..),
+    Worker (..),
+    RestartCount (..),
+    updateRestartCount,
   )
 where
 
+import Control.Monad
 import Control.Monad.Except
 import Control.Monad.IO.Unlift
 import Control.Monad.Reader
 import Crypto.Random
+import Data.ByteArray (ScrubbedBytes)
 import Data.Int (Int64)
 import Data.List.NonEmpty (NonEmpty)
 import Data.Map (Map)
 import Data.Time.Clock (NominalDiffTime, nominalDay)
+import Data.Time.Clock.System (SystemTime (..))
 import Data.Word (Word16)
 import Network.Socket
 import Numeric.Natural
@@ -52,9 +63,10 @@ import Simplex.Messaging.TMap (TMap)
 import qualified Simplex.Messaging.TMap as TM
 import Simplex.Messaging.Transport (TLS, Transport (..))
 import Simplex.Messaging.Transport.Client (defaultSMPPort)
+import Simplex.Messaging.Util (allFinally, catchAllErrors, tryAllErrors)
 import Simplex.Messaging.Version
 import System.Random (StdGen, newStdGen)
-import UnliftIO (Async)
+import UnliftIO (Async, SomeException)
 import UnliftIO.STM
 
 type AgentMonad' m = (MonadUnliftIO m, MonadReader Env m)
@@ -82,10 +94,14 @@ data AgentConfig = AgentConfig
     helloTimeout :: NominalDiffTime,
     initialCleanupDelay :: Int64,
     cleanupInterval :: Int64,
-    rcvMsgHashesTTL :: NominalDiffTime,
+    cleanupStepInterval :: Int,
+    maxWorkerRestartsPerMin :: Int,
+    maxSubscriptionTimeouts :: Int,
+    storedMsgDataTTL :: NominalDiffTime,
     rcvFilesTTL :: NominalDiffTime,
     sndFilesTTL :: NominalDiffTime,
     xftpNotifyErrsOnRetry :: Bool,
+    xftpConsecutiveRetries :: Int,
     xftpMaxRecipientsPerRequest :: Int,
     deleteErrorCount :: Int,
     ntfCron :: Word16,
@@ -98,8 +114,7 @@ data AgentConfig = AgentConfig
     certificateFile :: FilePath,
     e2eEncryptVRange :: VersionRange,
     smpAgentVRange :: VersionRange,
-    smpClientVRange :: VersionRange,
-    initialClientId :: Int
+    smpClientVRange :: VersionRange
   }
 
 defaultReconnectInterval :: RetryInterval
@@ -146,17 +161,23 @@ defaultAgentConfig =
       helloTimeout = 2 * nominalDay,
       initialCleanupDelay = 30 * 1000000, -- 30 seconds
       cleanupInterval = 30 * 60 * 1000000, -- 30 minutes
-      rcvMsgHashesTTL = 30 * nominalDay,
+      cleanupStepInterval = 200000, -- 200ms
+      maxWorkerRestartsPerMin = 5,
+      -- 3 consecutive subscription timeouts will result in alert to the user
+      -- this is a fallback, as the timeout set to 3x of expected timeout, to avoid potential locking.
+      maxSubscriptionTimeouts = 3,
+      storedMsgDataTTL = 21 * nominalDay,
       rcvFilesTTL = 2 * nominalDay,
       sndFilesTTL = nominalDay,
       xftpNotifyErrsOnRetry = True,
+      xftpConsecutiveRetries = 3,
       xftpMaxRecipientsPerRequest = 200,
       deleteErrorCount = 10,
       ntfCron = 20, -- minutes
       ntfWorkerDelay = 100000, -- microseconds
       ntfSMPWorkerDelay = 500000, -- microseconds
       ntfSubCheckInterval = nominalDay,
-      ntfMaxMessages = 4,
+      ntfMaxMessages = 3,
       -- CA certificate private key is not needed for initialization
       -- ! we do not generate these
       caCertificateFile = "/etc/opt/simplex-agent/ca.crt",
@@ -164,37 +185,36 @@ defaultAgentConfig =
       certificateFile = "/etc/opt/simplex-agent/agent.crt",
       e2eEncryptVRange = supportedE2EEncryptVRange,
       smpAgentVRange = supportedSMPAgentVRange,
-      smpClientVRange = supportedSMPClientVRange,
-      initialClientId = 0
+      smpClientVRange = supportedSMPClientVRange
     }
 
 data Env = Env
   { config :: AgentConfig,
     store :: SQLiteStore,
-    idsDrg :: TVar ChaChaDRG,
-    clientCounter :: TVar Int,
+    random :: TVar ChaChaDRG,
     randomServer :: TVar StdGen,
     ntfSupervisor :: NtfSupervisor,
-    xftpAgent :: XFTPAgent
+    xftpAgent :: XFTPAgent,
+    multicastSubscribers :: TMVar Int
   }
 
 newSMPAgentEnv :: AgentConfig -> SQLiteStore -> IO Env
-newSMPAgentEnv config@AgentConfig {initialClientId} store = do
-  idsDrg <- newTVarIO =<< liftIO drgNew
-  clientCounter <- newTVarIO initialClientId
+newSMPAgentEnv config store = do
+  random <- C.newRandom
   randomServer <- newTVarIO =<< liftIO newStdGen
   ntfSupervisor <- atomically . newNtfSubSupervisor $ tbqSize config
   xftpAgent <- atomically newXFTPAgent
-  pure Env {config, store, idsDrg, clientCounter, randomServer, ntfSupervisor, xftpAgent}
+  multicastSubscribers <- newTMVarIO 0
+  pure Env {config, store, random, randomServer, ntfSupervisor, xftpAgent, multicastSubscribers}
 
-createAgentStore :: FilePath -> String -> MigrationConfirmation -> IO (Either MigrationError SQLiteStore)
-createAgentStore dbFilePath dbKey = createSQLiteStore dbFilePath dbKey Migrations.app
+createAgentStore :: FilePath -> ScrubbedBytes -> Bool -> MigrationConfirmation -> IO (Either MigrationError SQLiteStore)
+createAgentStore dbFilePath dbKey keepKey = createSQLiteStore dbFilePath dbKey keepKey Migrations.app
 
 data NtfSupervisor = NtfSupervisor
   { ntfTkn :: TVar (Maybe NtfToken),
     ntfSubQ :: TBQueue (ConnId, NtfSupervisorCommand),
-    ntfWorkers :: TMap NtfServer (TMVar (), Async ()),
-    ntfSMPWorkers :: TMap SMPServer (TMVar (), Async ())
+    ntfWorkers :: TMap NtfServer Worker,
+    ntfSMPWorkers :: TMap SMPServer Worker
   }
 
 data NtfSupervisorCommand = NSCCreate | NSCDelete | NSCSmpDelete | NSCNtfWorker NtfServer | NSCNtfSMPWorker SMPServer
@@ -211,9 +231,9 @@ newNtfSubSupervisor qSize = do
 data XFTPAgent = XFTPAgent
   { -- if set, XFTP file paths will be considered as relative to this directory
     xftpWorkDir :: TVar (Maybe FilePath),
-    xftpRcvWorkers :: TMap (Maybe XFTPServer) (TMVar (), Async ()),
-    xftpSndWorkers :: TMap (Maybe XFTPServer) (TMVar (), Async ()),
-    xftpDelWorkers :: TMap XFTPServer (TMVar (), Async ())
+    xftpRcvWorkers :: TMap (Maybe XFTPServer) Worker,
+    xftpSndWorkers :: TMap (Maybe XFTPServer) Worker,
+    xftpDelWorkers :: TMap XFTPServer Worker
   }
 
 newXFTPAgent :: STM XFTPAgent
@@ -223,3 +243,41 @@ newXFTPAgent = do
   xftpSndWorkers <- TM.empty
   xftpDelWorkers <- TM.empty
   pure XFTPAgent {xftpWorkDir, xftpRcvWorkers, xftpSndWorkers, xftpDelWorkers}
+
+tryAgentError :: AgentMonad m => m a -> m (Either AgentErrorType a)
+tryAgentError = tryAllErrors mkInternal
+{-# INLINE tryAgentError #-}
+
+-- unlike runExceptT, this ensures we catch IO exceptions as well
+tryAgentError' :: AgentMonad' m => ExceptT AgentErrorType m a -> m (Either AgentErrorType a)
+tryAgentError' = fmap join . runExceptT . tryAgentError
+{-# INLINE tryAgentError' #-}
+
+catchAgentError :: AgentMonad m => m a -> (AgentErrorType -> m a) -> m a
+catchAgentError = catchAllErrors mkInternal
+{-# INLINE catchAgentError #-}
+
+agentFinally :: AgentMonad m => m a -> m b -> m a
+agentFinally = allFinally mkInternal
+{-# INLINE agentFinally #-}
+
+mkInternal :: SomeException -> AgentErrorType
+mkInternal = INTERNAL . show
+{-# INLINE mkInternal #-}
+
+data Worker = Worker
+  { workerId :: Int,
+    doWork :: TMVar (),
+    action :: TMVar (Maybe (Async ())),
+    restarts :: TVar RestartCount
+  }
+
+data RestartCount = RestartCount
+  { restartMinute :: Int64,
+    restartCount :: Int
+  }
+
+updateRestartCount :: SystemTime -> RestartCount -> RestartCount
+updateRestartCount t (RestartCount minute count) = do
+  let min' = systemSeconds t `div` 60
+   in RestartCount min' $ if minute == min' then count + 1 else 1

src/Simplex/Messaging/Agent/Lock.hs

@@ -1,10 +1,18 @@
 {-# LANGUAGE NamedFieldPuns #-}
 
-module Simplex.Messaging.Agent.Lock where
+module Simplex.Messaging.Agent.Lock
+  ( Lock,
+    createLock,
+    withLock,
+    withGetLock,
+    withGetLocks,
+  )
+where
 
 import Control.Monad (void)
 import Control.Monad.IO.Unlift
 import Data.Functor (($>))
+import UnliftIO.Async (forConcurrently)
 import qualified UnliftIO.Exception as E
 import UnliftIO.STM
 
@@ -20,9 +28,22 @@ withLock lock name =
     (atomically $ putTMVar lock name)
     (void . atomically $ takeTMVar lock)
 
-withGetLock :: MonadUnliftIO m => STM Lock -> String -> m a -> m a
-withGetLock getLock name a =
+withGetLock :: MonadUnliftIO m => (k -> STM Lock) -> k -> String -> m a -> m a
+withGetLock getLock key name a =
   E.bracket
-    (atomically $ getLock >>= \l -> putTMVar l name $> l)
+    (atomically $ getPutLock getLock key name)
     (atomically . takeTMVar)
     (const a)
+
+withGetLocks :: MonadUnliftIO m => (k -> STM Lock) -> [k] -> String -> m a -> m a
+withGetLocks getLock keys name = E.bracket holdLocks releaseLocks . const
+  where
+    holdLocks = forConcurrently keys $ \key -> atomically $ getPutLock getLock key name
+    -- only this withGetLocks would be holding the locks,
+    -- so it's safe to combine all lock releases into one transaction
+    releaseLocks = atomically . mapM_ takeTMVar
+
+-- getLock and putTMVar can be in one transaction on the assumption that getLock doesn't write in case the lock already exists,
+-- and in case it is created and added to some shared resource (we use TMap) it also helps avoid contention for the newly created lock.
+getPutLock :: (k -> STM Lock) -> k -> String -> STM Lock
+getPutLock getLock key name = getLock key >>= \l -> putTMVar l name $> l

src/Simplex/Messaging/Agent/NtfSubSupervisor.hs

@@ -5,6 +5,7 @@
 {-# LANGUAGE OverloadedStrings #-}
 {-# LANGUAGE ScopedTypeVariables #-}
 {-# LANGUAGE TupleSections #-}
+{-# OPTIONS_GHC -fno-warn-ambiguous-fields #-}
 
 module Simplex.Messaging.Agent.NtfSubSupervisor
   ( runNtfSupervisor,
@@ -17,7 +18,6 @@ module Simplex.Messaging.Agent.NtfSubSupervisor
   )
 where
 
-import Control.Concurrent.STM (stateTVar)
 import Control.Logger.Simple (logError, logInfo)
 import Control.Monad
 import Control.Monad.Except
@@ -37,9 +37,7 @@ import Simplex.Messaging.Client.Agent ()
 import qualified Simplex.Messaging.Crypto as C
 import Simplex.Messaging.Notifications.Protocol (NtfSubStatus (..), NtfTknStatus (..), SMPQueueNtf (..))
 import Simplex.Messaging.Notifications.Types
-import Simplex.Messaging.Protocol (NtfServer, ProtocolServer, SMPServer, sameSrvAddr)
-import Simplex.Messaging.TMap (TMap)
-import qualified Simplex.Messaging.TMap as TM
+import Simplex.Messaging.Protocol (NtfServer, SMPServer, sameSrvAddr)
 import Simplex.Messaging.Util (diffToMicroseconds, threadDelay', tshow, unlessM)
 import System.Random (randomR)
 import UnliftIO
@@ -51,13 +49,13 @@ runNtfSupervisor c = do
   ns <- asks ntfSupervisor
   forever $ do
     cmd@(connId, _) <- atomically . readTBQueue $ ntfSubQ ns
-    handleError connId . agentOperationBracket c AONtfNetwork waitUntilActive $
+    handleErr connId . agentOperationBracket c AONtfNetwork waitUntilActive $
       runExceptT (processNtfSub c cmd) >>= \case
         Left e -> notifyErr connId e
         Right _ -> return ()
   where
-    handleError :: ConnId -> m () -> m ()
-    handleError connId = E.handle $ \(e :: E.SomeException) -> do
+    handleErr :: ConnId -> m () -> m ()
+    handleErr connId = E.handle $ \(e :: E.SomeException) -> do
       logError $ "runNtfSupervisor error " <> tshow e
       notifyErr connId e
     notifyErr connId e = notifyInternalError c connId $ "runNtfSupervisor error " <> show e
@@ -79,11 +77,11 @@ processNtfSub c (connId, cmd) = do
               Just ClientNtfCreds {notifierId} -> do
                 let newSub = newNtfSubscription connId smpServer (Just notifierId) ntfServer NASKey
                 withStore c $ \db -> createNtfSubscription db newSub $ NtfSubNTFAction NSACreate
-                addNtfNTFWorker ntfServer
+                void $ getNtfNTFWorker True c ntfServer
               Nothing -> do
                 let newSub = newNtfSubscription connId smpServer Nothing ntfServer NASNew
                 withStore c $ \db -> createNtfSubscription db newSub $ NtfSubSMPAction NSASmpKey
-                addNtfSMPWorker smpServer
+                void $ getNtfSMPWorker True c smpServer
         (Just (sub@NtfSubscription {ntfSubStatus, ntfServer = subNtfServer, smpServer = smpServer', ntfQueueId}, action_)) -> do
           case (clientNtfCreds, ntfQueueId) of
             (Just ClientNtfCreds {notifierId}, Just ntfQueueId')
@@ -99,84 +97,74 @@ processNtfSub c (connId, cmd) = do
               Just (action, _)
                 -- subscription was marked for deletion / is being deleted
                 | isDeleteNtfSubAction action -> do
-                  if ntfSubStatus == NASNew || ntfSubStatus == NASOff || ntfSubStatus == NASDeleted
-                    then resetSubscription
-                    else withNtfServer c $ \ntfServer -> do
-                      withStore' c $ \db -> supervisorUpdateNtfSub db sub {ntfServer} (NtfSubNTFAction NSACreate)
-                      addNtfNTFWorker ntfServer
+                    if ntfSubStatus == NASNew || ntfSubStatus == NASOff || ntfSubStatus == NASDeleted
+                      then resetSubscription
+                      else withNtfServer c $ \ntfServer -> do
+                        withStore' c $ \db -> supervisorUpdateNtfSub db sub {ntfServer} (NtfSubNTFAction NSACreate)
+                        void $ getNtfNTFWorker True c ntfServer
                 | otherwise -> case action of
-                  NtfSubNTFAction _ -> addNtfNTFWorker subNtfServer
-                  NtfSubSMPAction _ -> addNtfSMPWorker smpServer
+                    NtfSubNTFAction _ -> void $ getNtfNTFWorker True c subNtfServer
+                    NtfSubSMPAction _ -> void $ getNtfSMPWorker True c smpServer
             rotate :: m ()
             rotate = do
               withStore' c $ \db -> supervisorUpdateNtfSub db sub (NtfSubNTFAction NSARotate)
-              addNtfNTFWorker subNtfServer
+              void $ getNtfNTFWorker True c subNtfServer
             resetSubscription :: m ()
             resetSubscription =
               withNtfServer c $ \ntfServer -> do
                 let sub' = sub {ntfQueueId = Nothing, ntfServer, ntfSubId = Nothing, ntfSubStatus = NASNew}
                 withStore' c $ \db -> supervisorUpdateNtfSub db sub' (NtfSubSMPAction NSASmpKey)
-                addNtfSMPWorker smpServer
+                void $ getNtfSMPWorker True c smpServer
     NSCDelete -> do
       sub_ <- withStore' c $ \db -> do
         supervisorUpdateNtfAction db connId (NtfSubNTFAction NSADelete)
         getNtfSubscription db connId
       logInfo $ "processNtfSub, NSCDelete - sub_ = " <> tshow sub_
       case sub_ of
-        (Just (NtfSubscription {ntfServer}, _)) -> addNtfNTFWorker ntfServer
+        (Just (NtfSubscription {ntfServer}, _)) -> void $ getNtfNTFWorker True c ntfServer
         _ -> pure () -- err "NSCDelete - no subscription"
     NSCSmpDelete -> do
       withStore' c (`getPrimaryRcvQueue` connId) >>= \case
         Right rq@RcvQueue {server = smpServer} -> do
           logInfo $ "processNtfSub, NSCSmpDelete - rq = " <> tshow rq
           withStore' c $ \db -> supervisorUpdateNtfAction db connId (NtfSubSMPAction NSASmpDelete)
-          addNtfSMPWorker smpServer
+          void $ getNtfSMPWorker True c smpServer
         _ -> notifyInternalError c connId "NSCSmpDelete - no rcv queue"
-    NSCNtfWorker ntfServer -> addNtfNTFWorker ntfServer
-    NSCNtfSMPWorker smpServer -> addNtfSMPWorker smpServer
-  where
-    addNtfNTFWorker = addWorker ntfWorkers runNtfWorker
-    addNtfSMPWorker = addWorker ntfSMPWorkers runNtfSMPWorker
-    addWorker ::
-      (NtfSupervisor -> TMap (ProtocolServer s) (TMVar (), Async ())) ->
-      (AgentClient -> ProtocolServer s -> TMVar () -> m ()) ->
-      ProtocolServer s ->
-      m ()
-    addWorker wsSel runWorker srv = do
-      ws <- asks $ wsSel . ntfSupervisor
-      atomically (TM.lookup srv ws) >>= \case
-        Nothing -> do
-          doWork <- newTMVarIO ()
-          worker <- async $ runWorker c srv doWork `E.finally` atomically (TM.delete srv ws)
-          atomically $ TM.insert srv (doWork, worker) ws
-        Just (doWork, _) ->
-          void . atomically $ tryPutTMVar doWork ()
+    NSCNtfWorker ntfServer -> void $ getNtfNTFWorker True c ntfServer
+    NSCNtfSMPWorker smpServer -> void $ getNtfSMPWorker True c smpServer
+
+getNtfNTFWorker :: AgentMonad' m => Bool -> AgentClient -> NtfServer -> m Worker
+getNtfNTFWorker hasWork c server = do
+  ws <- asks $ ntfWorkers . ntfSupervisor
+  getAgentWorker "ntf_ntf" hasWork c server ws $ runNtfWorker c server
+
+getNtfSMPWorker :: AgentMonad' m => Bool -> AgentClient -> SMPServer -> m Worker
+getNtfSMPWorker hasWork c server = do
+  ws <- asks $ ntfSMPWorkers . ntfSupervisor
+  getAgentWorker "ntf_smp" hasWork c server ws $ runNtfSMPWorker c server
 
 withNtfServer :: AgentMonad' m => AgentClient -> (NtfServer -> m ()) -> m ()
 withNtfServer c action = getNtfServer c >>= mapM_ action
 
-runNtfWorker :: forall m. AgentMonad m => AgentClient -> NtfServer -> TMVar () -> m ()
-runNtfWorker c srv doWork = do
+runNtfWorker :: forall m. AgentMonad m => AgentClient -> NtfServer -> Worker -> m ()
+runNtfWorker c srv Worker {doWork} = do
   delay <- asks $ ntfWorkerDelay . config
   forever $ do
-    void . atomically $ readTMVar doWork
+    waitForWork doWork
     agentOperationBracket c AONtfNetwork throwWhenInactive runNtfOperation
     threadDelay delay
   where
     runNtfOperation :: m ()
-    runNtfOperation = do
-      nextSub_ <- withStore' c (`getNextNtfSubNTFAction` srv)
-      logInfo $ "runNtfWorker, nextSub_ " <> tshow nextSub_
-      case nextSub_ of
-        Nothing -> noWorkToDo
-        Just a@(NtfSubscription {connId}, _, _) -> do
+    runNtfOperation =
+      withWork c doWork (`getNextNtfSubNTFAction` srv) $
+        \nextSub@(NtfSubscription {connId}, _, _) -> do
+          logInfo $ "runNtfWorker, nextSub " <> tshow nextSub
           ri <- asks $ reconnectInterval . config
           withRetryInterval ri $ \_ loop ->
-            processAction a
-              `catchError` retryOnError c "NtfWorker" loop (workerInternalError c connId . show)
-    noWorkToDo = void . atomically $ tryTakeTMVar doWork
-    processAction :: (NtfSubscription, NtfSubNTFAction, NtfActionTs) -> m ()
-    processAction (sub@NtfSubscription {connId, smpServer, ntfSubId}, action, actionTs) = do
+            processSub nextSub
+              `catchAgentError` retryOnError c "NtfWorker" loop (workerInternalError c connId . show)
+    processSub :: (NtfSubscription, NtfSubNTFAction, NtfActionTs) -> m ()
+    processSub (sub@NtfSubscription {connId, smpServer, ntfSubId}, action, actionTs) = do
       ts <- liftIO getCurrentTime
       unlessM (rescheduleAction doWork ts actionTs) $
         case action of
@@ -213,7 +201,7 @@ runNtfWorker c srv doWork = do
           NSADelete -> case ntfSubId of
             Just nSubId ->
               (getNtfToken >>= mapM_ (agentNtfDeleteSubscription c nSubId))
-                `E.finally` continueDeletion
+                `agentFinally` continueDeletion
             _ -> continueDeletion
             where
               continueDeletion = do
@@ -224,7 +212,7 @@ runNtfWorker c srv doWork = do
           NSARotate -> case ntfSubId of
             Just nSubId ->
               (getNtfToken >>= mapM_ (agentNtfDeleteSubscription c nSubId))
-                `E.finally` deleteCreate
+                `agentFinally` deleteCreate
             _ -> deleteCreate
             where
               deleteCreate = do
@@ -240,27 +228,24 @@ runNtfWorker c srv doWork = do
           withStore' c $ \db ->
             updateNtfSubscription db sub {ntfSubStatus = toStatus} toAction actionTs'
 
-runNtfSMPWorker :: forall m. AgentMonad m => AgentClient -> SMPServer -> TMVar () -> m ()
-runNtfSMPWorker c srv doWork = do
+runNtfSMPWorker :: forall m. AgentMonad m => AgentClient -> SMPServer -> Worker -> m ()
+runNtfSMPWorker c srv Worker {doWork} = do
   delay <- asks $ ntfSMPWorkerDelay . config
   forever $ do
-    void . atomically $ readTMVar doWork
+    waitForWork doWork
     agentOperationBracket c AONtfNetwork throwWhenInactive runNtfSMPOperation
     threadDelay delay
   where
-    runNtfSMPOperation = do
-      nextSub_ <- withStore' c (`getNextNtfSubSMPAction` srv)
-      logInfo $ "runNtfSMPWorker, nextSub_ " <> tshow nextSub_
-      case nextSub_ of
-        Nothing -> noWorkToDo
-        Just a@(NtfSubscription {connId}, _, _) -> do
+    runNtfSMPOperation =
+      withWork c doWork (`getNextNtfSubSMPAction` srv) $
+        \nextSub@(NtfSubscription {connId}, _, _) -> do
+          logInfo $ "runNtfSMPWorker, nextSub " <> tshow nextSub
           ri <- asks $ reconnectInterval . config
           withRetryInterval ri $ \_ loop ->
-            processAction a
-              `catchError` retryOnError c "NtfSMPWorker" loop (workerInternalError c connId . show)
-    noWorkToDo = void . atomically $ tryTakeTMVar doWork
-    processAction :: (NtfSubscription, NtfSubSMPAction, NtfActionTs) -> m ()
-    processAction (sub@NtfSubscription {connId, ntfServer}, smpAction, actionTs) = do
+            processSub nextSub
+              `catchAgentError` retryOnError c "NtfSMPWorker" loop (workerInternalError c connId . show)
+    processSub :: (NtfSubscription, NtfSubSMPAction, NtfActionTs) -> m ()
+    processSub (sub@NtfSubscription {connId, ntfServer}, smpAction, actionTs) = do
       ts <- liftIO getCurrentTime
       unlessM (rescheduleAction doWork ts actionTs) $
         case smpAction of
@@ -269,8 +254,9 @@ runNtfSMPWorker c srv doWork = do
               Just NtfToken {ntfTknStatus = NTActive, ntfMode = NMInstant} -> do
                 rq <- withStore c (`getPrimaryRcvQueue` connId)
                 C.SignAlg a <- asks (cmdSignAlg . config)
-                (ntfPublicKey, ntfPrivateKey) <- liftIO $ C.generateSignatureKeyPair a
-                (rcvNtfPubDhKey, rcvNtfPrivDhKey) <- liftIO C.generateKeyPair'
+                g <- asks random
+                (ntfPublicKey, ntfPrivateKey) <- atomically $ C.generateSignatureKeyPair a g
+                (rcvNtfPubDhKey, rcvNtfPrivDhKey) <- atomically $ C.generateKeyPair g
                 (notifierId, rcvNtfSrvPubDhKey) <- enableQueueNotifications c rq ntfPublicKey rcvNtfPubDhKey
                 let rcvNtfDhSecret = C.dh' rcvNtfSrvPubDhKey rcvNtfPrivDhKey
                 withStore' c $ \db -> do
@@ -290,11 +276,11 @@ rescheduleAction :: AgentMonad' m => TMVar () -> UTCTime -> UTCTime -> m Bool
 rescheduleAction doWork ts actionTs
   | actionTs <= ts = pure False
   | otherwise = do
-    void . atomically $ tryTakeTMVar doWork
-    void . forkIO $ do
-      liftIO $ threadDelay' $ diffToMicroseconds $ diffUTCTime actionTs ts
-      void . atomically $ tryPutTMVar doWork ()
-    pure True
+      void . atomically $ tryTakeTMVar doWork
+      void . forkIO $ do
+        liftIO $ threadDelay' $ diffToMicroseconds $ diffUTCTime actionTs ts
+        atomically $ hasWorkToDo' doWork
+      pure True
 
 retryOnError :: AgentMonad' m => AgentClient -> Text -> m () -> (AgentErrorType -> m ()) -> AgentErrorType -> m ()
 retryOnError c name loop done e = do
@@ -342,13 +328,10 @@ instantNotifications = \case
 
 closeNtfSupervisor :: MonadUnliftIO m => NtfSupervisor -> m ()
 closeNtfSupervisor ns = do
-  cancelNtfWorkers_ $ ntfWorkers ns
-  cancelNtfWorkers_ $ ntfSMPWorkers ns
-
-cancelNtfWorkers_ :: MonadUnliftIO m => TMap (ProtocolServer s) (TMVar (), Async ()) -> m ()
-cancelNtfWorkers_ wsVar = do
-  ws <- atomically $ stateTVar wsVar (,M.empty)
-  mapM_ (uninterruptibleCancel . snd) ws
+  stopWorkers $ ntfWorkers ns
+  stopWorkers $ ntfSMPWorkers ns
+  where
+    stopWorkers workers = atomically (swapTVar workers M.empty) >>= mapM_ (liftIO . cancelWorker)
 
 getNtfServer :: AgentMonad' m => AgentClient -> m (Maybe NtfServer)
 getNtfServer c = do

src/Simplex/Messaging/Agent/RetryInterval.hs

@@ -1,6 +1,5 @@
 {-# LANGUAGE LambdaCase #-}
 {-# LANGUAGE NamedFieldPuns #-}
-{-# LANGUAGE OverloadedStrings #-}
 {-# LANGUAGE ScopedTypeVariables #-}
 
 module Simplex.Messaging.Agent.RetryInterval
@@ -9,6 +8,7 @@ module Simplex.Messaging.Agent.RetryInterval
     RetryIntervalMode (..),
     RI2State (..),
     withRetryInterval,
+    withRetryIntervalCount,
     withRetryLock2,
     updateRetryInterval2,
   )
@@ -49,15 +49,18 @@ data RetryIntervalMode = RISlow | RIFast
   deriving (Eq, Show)
 
 withRetryInterval :: forall m a. MonadIO m => RetryInterval -> (Int64 -> m a -> m a) -> m a
-withRetryInterval ri action = callAction 0 $ initialInterval ri
+withRetryInterval ri = withRetryIntervalCount ri . const
+
+withRetryIntervalCount :: forall m a. MonadIO m => RetryInterval -> (Int -> Int64 -> m a -> m a) -> m a
+withRetryIntervalCount ri action = callAction 0 0 $ initialInterval ri
   where
-    callAction :: Int64 -> Int64 -> m a
-    callAction elapsed delay = action delay loop
+    callAction :: Int -> Int64 -> Int64 -> m a
+    callAction n elapsed delay = action n delay loop
       where
         loop = do
           liftIO $ threadDelay' delay
           let elapsed' = elapsed + delay
-          callAction elapsed' $ nextDelay elapsed' delay ri
+          callAction (n + 1) elapsed' $ nextDelay elapsed' delay ri
 
 -- This function allows action to toggle between slow and fast retry intervals.
 withRetryLock2 :: forall m. MonadIO m => RetryInterval2 -> TMVar () -> (RI2State -> (RetryIntervalMode -> m ()) -> m ()) -> m ()

src/Simplex/Messaging/Agent/Server.hs

@@ -11,7 +11,7 @@ module Simplex.Messaging.Agent.Server
 where
 
 import Control.Logger.Simple (logInfo)
-import Control.Monad.Except
+import Control.Monad
 import Control.Monad.IO.Unlift (MonadUnliftIO)
 import Control.Monad.Reader
 import Crypto.Random (MonadRandom)
@@ -23,7 +23,7 @@ import Simplex.Messaging.Agent.Env.SQLite
 import Simplex.Messaging.Agent.Protocol
 import Simplex.Messaging.Agent.Store.SQLite (SQLiteStore)
 import Simplex.Messaging.Transport (ATransport (..), TProxy, Transport (..), simplexMQVersion)
-import Simplex.Messaging.Transport.Server (loadTLSServerParams, runTransportServer)
+import Simplex.Messaging.Transport.Server (defaultTransportServerConfig, loadTLSServerParams, runTransportServer)
 import Simplex.Messaging.Util (bshow)
 import UnliftIO.Async (race_)
 import qualified UnliftIO.Exception as E
@@ -48,9 +48,11 @@ runSMPAgentBlocking (ATransport t) cfg@AgentConfig {tcpPort, caCertificateFile,
     smpAgent _ = do
       -- tlsServerParams is not in Env to avoid breaking functional API w/t key and certificate generation
       tlsServerParams <- liftIO $ loadTLSServerParams caCertificateFile certificateFile privateKeyFile
-      runTransportServer started tcpPort tlsServerParams True $ \(h :: c) -> do
+      clientId <- newTVarIO 0
+      runTransportServer started tcpPort tlsServerParams defaultTransportServerConfig $ \(h :: c) -> do
         liftIO . putLn h $ "Welcome to SMP agent v" <> B.pack simplexMQVersion
-        c <- getAgentClient initServers
+        cId <- atomically $ stateTVar clientId $ \i -> (i + 1, i + 1)
+        c <- getAgentClient cId initServers
         logConnection c True
         race_ (connectClient h c) (runAgentClient c)
           `E.finally` disconnectAgentClient c

src/Simplex/Messaging/Agent/Store.hs

@@ -1,6 +1,9 @@
+{-# LANGUAGE ConstraintKinds #-}
 {-# LANGUAGE DataKinds #-}
 {-# LANGUAGE DeriveAnyClass #-}
 {-# LANGUAGE DuplicateRecordFields #-}
+{-# LANGUAGE FlexibleContexts #-}
+{-# LANGUAGE FlexibleInstances #-}
 {-# LANGUAGE GADTs #-}
 {-# LANGUAGE KindSignatures #-}
 {-# LANGUAGE LambdaCase #-}
@@ -21,6 +24,7 @@ import Data.Kind (Type)
 import Data.List (find)
 import Data.List.NonEmpty (NonEmpty)
 import qualified Data.List.NonEmpty as L
+import Data.Maybe (isJust)
 import Data.Time (UTCTime)
 import Data.Type.Equality
 import Simplex.Messaging.Agent.Protocol
@@ -35,7 +39,6 @@ import Simplex.Messaging.Protocol
     NotifierId,
     NtfPrivateSignKey,
     NtfPublicVerifyKey,
-    QueueId,
     RcvDhSecret,
     RcvNtfDhSecret,
     RcvPrivateSignKey,
@@ -47,8 +50,26 @@ import Simplex.Messaging.Version
 
 -- * Queue types
 
+data QueueStored = QSStored | QSNew
+
+data SQueueStored (q :: QueueStored) where
+  SQSStored :: SQueueStored 'QSStored
+  SQSNew :: SQueueStored 'QSNew
+
+data DBQueueId (q :: QueueStored) where
+  DBQueueId :: Int64 -> DBQueueId 'QSStored
+  DBNewQueue :: DBQueueId 'QSNew
+
+deriving instance Eq (DBQueueId q)
+
+deriving instance Show (DBQueueId q)
+
+type RcvQueue = StoredRcvQueue 'QSStored
+
+type NewRcvQueue = StoredRcvQueue 'QSNew
+
 -- | A receive queue. SMP queue through which the agent receives messages from a sender.
-data RcvQueue = RcvQueue
+data StoredRcvQueue (q :: QueueStored) = RcvQueue
   { userId :: UserId,
     connId :: ConnId,
     server :: SMPServer,
@@ -66,12 +87,13 @@ data RcvQueue = RcvQueue
     sndId :: SMP.SenderId,
     -- | queue status
     status :: QueueStatus,
-    -- | database queue ID (within connection), can be Nothing for old queues
-    dbQueueId :: Int64,
+    -- | database queue ID (within connection)
+    dbQueueId :: DBQueueId q,
     -- | True for a primary or a next primary queue of the connection (next if dbReplaceQueueId is set)
     primary :: Bool,
     -- | database queue ID to replace, Nothing if this queue is not replacing another, `Just Nothing` is used for replacing old queues
     dbReplaceQueueId :: Maybe Int64,
+    rcvSwchStatus :: Maybe RcvSwitchStatus,
     -- | SMP client version
     smpClientVersion :: Version,
     -- | credentials used in context of notifications
@@ -80,6 +102,22 @@ data RcvQueue = RcvQueue
   }
   deriving (Eq, Show)
 
+rcvQueueInfo :: RcvQueue -> RcvQueueInfo
+rcvQueueInfo rq@RcvQueue {server, rcvSwchStatus} =
+  RcvQueueInfo {rcvServer = server, rcvSwitchStatus = rcvSwchStatus, canAbortSwitch = canAbortRcvSwitch rq}
+
+canAbortRcvSwitch :: RcvQueue -> Bool
+canAbortRcvSwitch = maybe False canAbort . rcvSwchStatus
+  where
+    canAbort = \case
+      RSSwitchStarted -> True
+      RSSendingQADD -> True
+      -- if switch is in RSSendingQUSE, a race condition with sender deleting the original queue is possible
+      RSSendingQUSE -> False
+      -- if switch is in RSReceivedMessage status, aborting switch (deleting new queue)
+      -- will break the connection because the sender would have original queue deleted
+      RSReceivedMessage -> False
+
 data ClientNtfCreds = ClientNtfCreds
   { -- | key pair to be used by the notification server to sign transmissions
     ntfPublicKey :: NtfPublicVerifyKey,
@@ -91,8 +129,12 @@ data ClientNtfCreds = ClientNtfCreds
   }
   deriving (Eq, Show)
 
+type SndQueue = StoredSndQueue 'QSStored
+
+type NewSndQueue = StoredSndQueue 'QSNew
+
 -- | A send queue. SMP queue through which the agent sends messages to a recipient.
-data SndQueue = SndQueue
+data StoredSndQueue (q :: QueueStored) = SndQueue
   { userId :: UserId,
     connId :: ConnId,
     server :: SMPServer,
@@ -107,32 +149,33 @@ data SndQueue = SndQueue
     e2eDhSecret :: C.DhSecretX25519,
     -- | queue status
     status :: QueueStatus,
-    -- | database queue ID (within connection), can be Nothing for old queues
-    dbQueueId :: Int64,
+    -- | database queue ID (within connection)
+    dbQueueId :: DBQueueId q,
     -- | True for a primary or a next primary queue of the connection (next if dbReplaceQueueId is set)
     primary :: Bool,
     -- | ID of the queue this one is replacing
     dbReplaceQueueId :: Maybe Int64,
+    sndSwchStatus :: Maybe SndSwitchStatus,
     -- | SMP client version
     smpClientVersion :: Version
   }
   deriving (Eq, Show)
 
+sndQueueInfo :: SndQueue -> SndQueueInfo
+sndQueueInfo SndQueue {server, sndSwchStatus} =
+  SndQueueInfo {sndServer = server, sndSwitchStatus = sndSwchStatus}
+
 instance SMPQueue RcvQueue where
   qServer RcvQueue {server} = server
   {-# INLINE qServer #-}
-  qAddress RcvQueue {server, rcvId} = (server, rcvId)
-  {-# INLINE qAddress #-}
-  sameQueue addr q = sameQAddress addr (qAddress q)
-  {-# INLINE sameQueue #-}
+  queueId RcvQueue {rcvId} = rcvId
+  {-# INLINE queueId #-}
 
 instance SMPQueue SndQueue where
   qServer SndQueue {server} = server
   {-# INLINE qServer #-}
-  qAddress SndQueue {server, sndId} = (server, sndId)
-  {-# INLINE qAddress #-}
-  sameQueue addr q = sameQAddress addr (qAddress q)
-  {-# INLINE sameQueue #-}
+  queueId SndQueue {sndId} = sndId
+  {-# INLINE queueId #-}
 
 findQ :: SMPQueue q => (SMPServer, SMP.QueueId) -> NonEmpty q -> Maybe q
 findQ = find . sameQueue
@@ -155,26 +198,39 @@ findRQ :: (SMPServer, SMP.SenderId) -> NonEmpty RcvQueue -> Maybe RcvQueue
 findRQ sAddr = find $ sameQAddress sAddr . sndAddress
 {-# INLINE findRQ #-}
 
+switchingRQ :: NonEmpty RcvQueue -> Maybe RcvQueue
+switchingRQ = find $ isJust . rcvSwchStatus
+{-# INLINE switchingRQ #-}
+
+updatedQs :: SMPQueueRec q => q -> NonEmpty q -> NonEmpty q
+updatedQs q = L.map $ \q' -> if dbQId q == dbQId q' then q else q'
+{-# INLINE updatedQs #-}
+
 class SMPQueue q => SMPQueueRec q where
   qUserId :: q -> UserId
   qConnId :: q -> ConnId
-  queueId :: q -> QueueId
+  dbQId :: q -> Int64
+  dbReplaceQId :: q -> Maybe Int64
 
 instance SMPQueueRec RcvQueue where
-  qUserId = userId
+  qUserId RcvQueue {userId} = userId
   {-# INLINE qUserId #-}
-  qConnId = connId
+  qConnId RcvQueue {connId} = connId
   {-# INLINE qConnId #-}
-  queueId = rcvId
-  {-# INLINE queueId #-}
+  dbQId RcvQueue {dbQueueId = DBQueueId qId} = qId
+  {-# INLINE dbQId #-}
+  dbReplaceQId RcvQueue {dbReplaceQueueId} = dbReplaceQueueId
+  {-# INLINE dbReplaceQId #-}
 
 instance SMPQueueRec SndQueue where
-  qUserId = userId
+  qUserId SndQueue {userId} = userId
   {-# INLINE qUserId #-}
-  qConnId = connId
+  qConnId SndQueue {connId} = connId
   {-# INLINE qConnId #-}
-  queueId = sndId
-  {-# INLINE queueId #-}
+  dbQId SndQueue {dbQueueId = DBQueueId qId} = qId
+  {-# INLINE dbQId #-}
+  dbReplaceQId SndQueue {dbReplaceQueueId} = dbReplaceQueueId
+  {-# INLINE dbReplaceQId #-}
 
 -- * Connection types
 
@@ -202,14 +258,22 @@ deriving instance Eq (Connection d)
 
 deriving instance Show (Connection d)
 
-connData :: Connection d -> ConnData
-connData = \case
+toConnData :: Connection d -> ConnData
+toConnData = \case
   NewConnection cData -> cData
   RcvConnection cData _ -> cData
   SndConnection cData _ -> cData
   DuplexConnection cData _ _ -> cData
   ContactConnection cData _ -> cData
 
+updateConnection :: ConnData -> Connection d -> Connection d
+updateConnection cData = \case
+  NewConnection _ -> NewConnection cData
+  RcvConnection _ rq -> RcvConnection cData rq
+  SndConnection _ sq -> SndConnection cData sq
+  DuplexConnection _ rqs sqs -> DuplexConnection cData rqs sqs
+  ContactConnection _ rq -> ContactConnection cData rq
+
 data SConnType :: ConnType -> Type where
   SCNew :: SConnType CNew
   SCRcv :: SConnType CRcv
@@ -252,12 +316,31 @@ data ConnData = ConnData
     connAgentVersion :: Version,
     enableNtfs :: Bool,
     duplexHandshake :: Maybe Bool, -- added in agent protocol v2
-    deleted :: Bool
+    lastExternalSndId :: PrevExternalSndId,
+    deleted :: Bool,
+    ratchetSyncState :: RatchetSyncState
   }
   deriving (Eq, Show)
 
+-- this function should be mirrored in the clients
+ratchetSyncAllowed :: ConnData -> Bool
+ratchetSyncAllowed cData@ConnData {ratchetSyncState} =
+  ratchetSyncSupported' cData && (ratchetSyncState `elem` ([RSAllowed, RSRequired] :: [RatchetSyncState]))
+
+ratchetSyncSupported' :: ConnData -> Bool
+ratchetSyncSupported' ConnData {connAgentVersion} = connAgentVersion >= 3
+
+messageRcptsSupported :: ConnData -> Bool
+messageRcptsSupported ConnData {connAgentVersion} = connAgentVersion >= 4
+
+-- this function should be mirrored in the clients
+ratchetSyncSendProhibited :: ConnData -> Bool
+ratchetSyncSendProhibited ConnData {ratchetSyncState} =
+  ratchetSyncState `elem` ([RSRequired, RSStarted, RSAgreed] :: [RatchetSyncState])
+
 data PendingCommand = PendingCommand
-  { corrId :: ACorrId,
+  { cmdId :: AsyncCmdId,
+    corrId :: ACorrId,
     userId :: UserId,
     connId :: ConnId,
     command :: AgentCommand
@@ -308,6 +391,7 @@ data InternalCommand
   | ICAllowSecure SMP.RecipientId SMP.SndPublicVerifyKey
   | ICDuplexSecure SMP.RecipientId SMP.SndPublicVerifyKey
   | ICDeleteConn
+  | ICDeleteRcvQueue SMP.RecipientId
   | ICQSecure SMP.RecipientId SMP.SndPublicVerifyKey
   | ICQDelete SMP.RecipientId
 
@@ -317,6 +401,7 @@ data InternalCommandTag
   | ICAllowSecure_
   | ICDuplexSecure_
   | ICDeleteConn_
+  | ICDeleteRcvQueue_
   | ICQSecure_
   | ICQDelete_
   deriving (Show)
@@ -328,6 +413,7 @@ instance StrEncoding InternalCommand where
     ICAllowSecure rId sndKey -> strEncode (ICAllowSecure_, rId, sndKey)
     ICDuplexSecure rId sndKey -> strEncode (ICDuplexSecure_, rId, sndKey)
     ICDeleteConn -> strEncode ICDeleteConn_
+    ICDeleteRcvQueue rId -> strEncode (ICDeleteRcvQueue_, rId)
     ICQSecure rId senderKey -> strEncode (ICQSecure_, rId, senderKey)
     ICQDelete rId -> strEncode (ICQDelete_, rId)
   strP =
@@ -337,6 +423,7 @@ instance StrEncoding InternalCommand where
       ICAllowSecure_ -> ICAllowSecure <$> _strP <*> _strP
       ICDuplexSecure_ -> ICDuplexSecure <$> _strP <*> _strP
       ICDeleteConn_ -> pure ICDeleteConn
+      ICDeleteRcvQueue_ -> ICDeleteRcvQueue <$> _strP
       ICQSecure_ -> ICQSecure <$> _strP <*> _strP
       ICQDelete_ -> ICQDelete <$> _strP
 
@@ -347,6 +434,7 @@ instance StrEncoding InternalCommandTag where
     ICAllowSecure_ -> "ALLOW_SECURE"
     ICDuplexSecure_ -> "DUPLEX_SECURE"
     ICDeleteConn_ -> "DELETE_CONN"
+    ICDeleteRcvQueue_ -> "DELETE_RCV_QUEUE"
     ICQSecure_ -> "QSECURE"
     ICQDelete_ -> "QDELETE"
   strP =
@@ -356,6 +444,7 @@ instance StrEncoding InternalCommandTag where
       "ALLOW_SECURE" -> pure ICAllowSecure_
       "DUPLEX_SECURE" -> pure ICDuplexSecure_
       "DELETE_CONN" -> pure ICDeleteConn_
+      "DELETE_RCV_QUEUE" -> pure ICDeleteRcvQueue_
       "QSECURE" -> pure ICQSecure_
       "QDELETE" -> pure ICQDelete_
       _ -> fail "bad InternalCommandTag"
@@ -372,6 +461,7 @@ internalCmdTag = \case
   ICAllowSecure {} -> ICAllowSecure_
   ICDuplexSecure {} -> ICDuplexSecure_
   ICDeleteConn -> ICDeleteConn_
+  ICDeleteRcvQueue {} -> ICDeleteRcvQueue_
   ICQSecure {} -> ICQSecure_
   ICQDelete _ -> ICQDelete_
 
@@ -435,7 +525,10 @@ data RcvMsgData = RcvMsgData
 data RcvMsg = RcvMsg
   { internalId :: InternalId,
     msgMeta :: MsgMeta,
+    msgType :: AgentMessageType,
     msgBody :: MsgBody,
+    internalHash :: MsgHash,
+    msgReceipt :: Maybe MsgReceipt, -- if this message is a delivery receipt
     userAck :: Bool
   }
 
@@ -450,6 +543,14 @@ data SndMsgData = SndMsgData
     prevMsgHash :: MsgHash
   }
 
+data SndMsg = SndMsg
+  { internalId :: InternalId,
+    internalSndId :: InternalSndId,
+    msgType :: AgentMessageType,
+    internalHash :: MsgHash,
+    msgReceipt :: Maybe MsgReceipt
+  }
+
 data PendingMsgData = PendingMsgData
   { msgId :: InternalId,
     msgType :: AgentMessageType,
@@ -515,6 +616,8 @@ data StoreError
     SEServerNotFound
   | -- | Connection already used.
     SEConnDuplicate
+  | -- | Confirmed snd queue already exists.
+    SESndQueueExists
   | -- | Wrong connection type, e.g. "send" connection when "receive" or "duplex" is expected, or vice versa.
     -- 'upgradeRcvConnToDuplex' and 'upgradeSndConnToDuplex' do not allow duplex connections - they would also return this error.
     SEBadConnType ConnType
@@ -542,4 +645,6 @@ data StoreError
     SEFileNotFound
   | -- | XFTP Deleted snd chunk replica not found.
     SEDeletedSndChunkReplicaNotFound
+  | -- | Error when reading work item that suspends worker - do not use!
+    SEWorkItemError ByteString
   deriving (Eq, Show, Exception)

src/Simplex/Messaging/Agent/Store/SQLite/Common.hs

@@ -0,0 +1,78 @@
+{-# LANGUAGE DuplicateRecordFields #-}
+{-# LANGUAGE NamedFieldPuns #-}
+{-# LANGUAGE ScopedTypeVariables #-}
+
+module Simplex.Messaging.Agent.Store.SQLite.Common
+  ( SQLiteStore (..),
+    withConnection,
+    withConnection',
+    withTransaction,
+    withTransaction',
+    withTransactionCtx,
+    dbBusyLoop,
+    storeKey,
+  )
+where
+
+import Control.Concurrent (threadDelay)
+import Data.ByteArray (ScrubbedBytes)
+import qualified Data.ByteArray as BA
+import Data.Time.Clock (diffUTCTime, getCurrentTime)
+import Database.SQLite.Simple (SQLError)
+import qualified Database.SQLite.Simple as SQL
+import qualified Simplex.Messaging.Agent.Store.SQLite.DB as DB
+import Simplex.Messaging.Util (diffToMilliseconds)
+import UnliftIO.Exception (bracket)
+import qualified UnliftIO.Exception as E
+import UnliftIO.STM
+
+storeKey :: ScrubbedBytes -> Bool -> Maybe ScrubbedBytes
+storeKey key keepKey = if keepKey || BA.null key then Just key else Nothing
+
+data SQLiteStore = SQLiteStore
+  { dbFilePath :: FilePath,
+    dbKey :: TVar (Maybe ScrubbedBytes),
+    dbConnection :: TMVar DB.Connection,
+    dbClosed :: TVar Bool,
+    dbNew :: Bool
+  }
+
+withConnection :: SQLiteStore -> (DB.Connection -> IO a) -> IO a
+withConnection SQLiteStore {dbConnection} =
+  bracket
+    (atomically $ takeTMVar dbConnection)
+    (atomically . putTMVar dbConnection)
+
+withConnection' :: SQLiteStore -> (SQL.Connection -> IO a) -> IO a
+withConnection' st action = withConnection st $ action . DB.conn
+
+withTransaction :: SQLiteStore -> (DB.Connection -> IO a) -> IO a
+withTransaction = withTransactionCtx Nothing
+
+withTransaction' :: SQLiteStore -> (SQL.Connection -> IO a) -> IO a
+withTransaction' st action = withTransaction st $ action . DB.conn
+
+withTransactionCtx :: Maybe String -> SQLiteStore -> (DB.Connection -> IO a) -> IO a
+withTransactionCtx ctx_ st action = withConnection st $ dbBusyLoop . transactionWithCtx
+  where
+    transactionWithCtx db@DB.Connection {conn} = case ctx_ of
+      Nothing -> SQL.withImmediateTransaction conn $ action db
+      Just ctx -> do
+        t1 <- getCurrentTime
+        r <- SQL.withImmediateTransaction conn $ action db
+        t2 <- getCurrentTime
+        putStrLn $ "withTransactionCtx start :: " <> show t1 <> " :: " <> ctx
+        putStrLn $ "withTransactionCtx end   :: " <> show t2 <> " :: " <> ctx <> " :: duration=" <> show (diffToMilliseconds $ diffUTCTime t2 t1)
+        pure r
+
+dbBusyLoop :: forall a. IO a -> IO a
+dbBusyLoop action = loop 500 3000000
+  where
+    loop :: Int -> Int -> IO a
+    loop t tLim =
+      action `E.catch` \(e :: SQLError) ->
+        if tLim > t && SQL.sqlError e == SQL.ErrorBusy
+          then do
+            threadDelay t
+            loop (t * 9 `div` 8) (tLim - t)
+          else E.throwIO e

src/Simplex/Messaging/Agent/Store/SQLite/DB.hs

@@ -0,0 +1,101 @@
+{-# LANGUAGE DeriveAnyClass #-}
+{-# LANGUAGE NamedFieldPuns #-}
+{-# LANGUAGE StrictData #-}
+{-# LANGUAGE TemplateHaskell #-}
+
+module Simplex.Messaging.Agent.Store.SQLite.DB
+  ( Connection (..),
+    SlowQueryStats (..),
+    open,
+    close,
+    execute,
+    execute_,
+    executeNamed,
+    executeMany,
+    query,
+    query_,
+    queryNamed,
+  )
+where
+
+import Control.Concurrent.STM
+import Control.Monad (when)
+import qualified Data.Aeson.TH as J
+import Data.Int (Int64)
+import Data.Time (diffUTCTime, getCurrentTime)
+import Database.SQLite.Simple (FromRow, NamedParam, Query, ToRow)
+import qualified Database.SQLite.Simple as SQL
+import Simplex.Messaging.Parsers (defaultJSON)
+import Simplex.Messaging.TMap (TMap)
+import qualified Simplex.Messaging.TMap as TM
+import Simplex.Messaging.Util (diffToMilliseconds)
+
+data Connection = Connection
+  { conn :: SQL.Connection,
+    slow :: TMap Query SlowQueryStats
+  }
+
+data SlowQueryStats = SlowQueryStats
+  { count :: Int64,
+    timeMax :: Int64,
+    timeAvg :: Int64
+  }
+  deriving (Show)
+
+timeIt :: TMap Query SlowQueryStats -> Query -> IO a -> IO a
+timeIt slow sql a = do
+  t <- getCurrentTime
+  r <- a
+  t' <- getCurrentTime
+  let diff = diffToMilliseconds $ diffUTCTime t' t
+  atomically $ when (diff > 5) $ TM.alter (updateQueryStats diff) sql slow
+  pure r
+  where
+    updateQueryStats :: Int64 -> Maybe SlowQueryStats -> Maybe SlowQueryStats
+    updateQueryStats diff Nothing = Just $ SlowQueryStats 1 diff diff
+    updateQueryStats diff (Just SlowQueryStats {count, timeMax, timeAvg}) =
+      Just $
+        SlowQueryStats
+          { count = count + 1,
+            timeMax = max timeMax diff,
+            timeAvg = (timeAvg * count + diff) `div` (count + 1)
+          }
+
+open :: String -> IO Connection
+open f = do
+  conn <- SQL.open f
+  slow <- atomically $ TM.empty
+  pure Connection {conn, slow}
+
+close :: Connection -> IO ()
+close = SQL.close . conn
+
+execute :: ToRow q => Connection -> Query -> q -> IO ()
+execute Connection {conn, slow} sql = timeIt slow sql . SQL.execute conn sql
+{-# INLINE execute #-}
+
+execute_ :: Connection -> Query -> IO ()
+execute_ Connection {conn, slow} sql = timeIt slow sql $ SQL.execute_ conn sql
+{-# INLINE execute_ #-}
+
+executeNamed :: Connection -> Query -> [NamedParam] -> IO ()
+executeNamed Connection {conn, slow} sql = timeIt slow sql . SQL.executeNamed conn sql
+{-# INLINE executeNamed #-}
+
+executeMany :: ToRow q => Connection -> Query -> [q] -> IO ()
+executeMany Connection {conn, slow} sql = timeIt slow sql . SQL.executeMany conn sql
+{-# INLINE executeMany #-}
+
+query :: (ToRow q, FromRow r) => Connection -> Query -> q -> IO [r]
+query Connection {conn, slow} sql = timeIt slow sql . SQL.query conn sql
+{-# INLINE query #-}
+
+query_ :: FromRow r => Connection -> Query -> IO [r]
+query_ Connection {conn, slow} sql = timeIt slow sql $ SQL.query_ conn sql
+{-# INLINE query_ #-}
+
+queryNamed :: FromRow r => Connection -> Query -> [NamedParam] -> IO [r]
+queryNamed Connection {conn, slow} sql = timeIt slow sql . SQL.queryNamed conn sql
+{-# INLINE queryNamed #-}
+
+$(J.deriveJSON defaultJSON ''SlowQueryStats)

src/Simplex/Messaging/Agent/Store/SQLite/Migrations.hs

@@ -1,4 +1,3 @@
-{-# LANGUAGE DeriveGeneric #-}
 {-# LANGUAGE LambdaCase #-}
 {-# LANGUAGE NamedFieldPuns #-}
 {-# LANGUAGE OverloadedLists #-}
@@ -27,8 +26,7 @@ module Simplex.Messaging.Agent.Store.SQLite.Migrations
 where
 
 import Control.Monad (forM_, when)
-import Data.Aeson (ToJSON)
-import qualified Data.Aeson as J
+import qualified Data.Aeson.TH as J
 import Data.List (intercalate, sortOn)
 import Data.List.NonEmpty (NonEmpty)
 import qualified Data.Map as M
@@ -40,8 +38,8 @@ import Database.SQLite.Simple (Connection, Only (..), Query (..))
 import qualified Database.SQLite.Simple as DB
 import Database.SQLite.Simple.QQ (sql)
 import qualified Database.SQLite3 as SQLite3
-import GHC.Generics (Generic)
 import Simplex.Messaging.Agent.Protocol (extraSMPServerHosts)
+import Simplex.Messaging.Agent.Store.SQLite.Common
 import Simplex.Messaging.Agent.Store.SQLite.Migrations.M20220101_initial
 import Simplex.Messaging.Agent.Store.SQLite.Migrations.M20220301_snd_queue_keys
 import Simplex.Messaging.Agent.Store.SQLite.Migrations.M20220322_notifications
@@ -60,6 +58,15 @@ import Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230320_retry_state
 import Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230401_snd_files
 import Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230510_files_pending_replicas_indexes
 import Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230516_encrypted_rcv_message_hashes
+import Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230531_switch_status
+import Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230615_ratchet_sync
+import Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230701_delivery_receipts
+import Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230720_delete_expired_messages
+import Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230722_indexes
+import Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230814_indexes
+import Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230829_crypto_files
+import Simplex.Messaging.Agent.Store.SQLite.Migrations.M20231222_command_created_at
+import Simplex.Messaging.Agent.Store.SQLite.Migrations.M20231225_failed_work_items
 import Simplex.Messaging.Encoding.String
 import Simplex.Messaging.Parsers (dropPrefix, sumTypeJSON)
 import Simplex.Messaging.Transport.Client (TransportHost)
@@ -86,7 +93,16 @@ schemaMigrations =
     ("m20230320_retry_state", m20230320_retry_state, Just down_m20230320_retry_state),
     ("m20230401_snd_files", m20230401_snd_files, Just down_m20230401_snd_files),
     ("m20230510_files_pending_replicas_indexes", m20230510_files_pending_replicas_indexes, Just down_m20230510_files_pending_replicas_indexes),
-    ("m20230516_encrypted_rcv_message_hashes", m20230516_encrypted_rcv_message_hashes, Just down_m20230516_encrypted_rcv_message_hashes)
+    ("m20230516_encrypted_rcv_message_hashes", m20230516_encrypted_rcv_message_hashes, Just down_m20230516_encrypted_rcv_message_hashes),
+    ("m20230531_switch_status", m20230531_switch_status, Just down_m20230531_switch_status),
+    ("m20230615_ratchet_sync", m20230615_ratchet_sync, Just down_m20230615_ratchet_sync),
+    ("m20230701_delivery_receipts", m20230701_delivery_receipts, Just down_m20230701_delivery_receipts),
+    ("m20230720_delete_expired_messages", m20230720_delete_expired_messages, Just down_m20230720_delete_expired_messages),
+    ("m20230722_indexes", m20230722_indexes, Just down_m20230722_indexes),
+    ("m20230814_indexes", m20230814_indexes, Just down_m20230814_indexes),
+    ("m20230829_crypto_files", m20230829_crypto_files, Just down_m20230829_crypto_files),
+    ("m20231222_command_created_at", m20231222_command_created_at, Just down_m20231222_command_created_at),
+    ("m20231225_failed_work_items", m20231225_failed_work_items, Just down_m20231225_failed_work_items)
   ]
 
 -- | The list of migrations in ascending order by date
@@ -95,44 +111,42 @@ app = sortOn name $ map migration schemaMigrations
   where
     migration (name, up, down) = Migration {name, up = fromQuery up, down = fromQuery <$> down}
 
-get :: Connection -> [Migration] -> IO (Either MTRError MigrationsToRun)
-get db migrations = migrationsToRun migrations <$> getCurrent db
+get :: SQLiteStore -> [Migration] -> IO (Either MTRError MigrationsToRun)
+get st migrations = migrationsToRun migrations <$> withTransaction' st getCurrent
 
 getCurrent :: Connection -> IO [Migration]
 getCurrent db = map toMigration <$> DB.query_ db "SELECT name, down FROM migrations ORDER BY name ASC;"
   where
     toMigration (name, down) = Migration {name, up = "", down}
 
-run :: Connection -> MigrationsToRun -> IO ()
-run db = \case
+run :: SQLiteStore -> MigrationsToRun -> IO ()
+run st = \case
   MTRUp [] -> pure ()
-  MTRUp ms -> mapM_ runUp ms >> execSQL "VACUUM;"
+  MTRUp ms -> mapM_ runUp ms >> withConnection' st (`execSQL` "VACUUM;")
   MTRDown ms -> mapM_ runDown $ reverse ms
   MTRNone -> pure ()
   where
-    runUp Migration {name, up, down} = do
-      when (name == "m20220811_onion_hosts") updateServers
-      DB.withImmediateTransaction db $ insert >> execSQL up
+    runUp Migration {name, up, down} = withTransaction' st $ \db -> do
+      when (name == "m20220811_onion_hosts") $ updateServers db
+      insert db >> execSQL db up
       where
-        insert = DB.execute db "INSERT INTO migrations (name, down, ts) VALUES (?,?,?)" . (name,down,) =<< getCurrentTime
-        updateServers = forM_ (M.assocs extraSMPServerHosts) $ \(h, h') ->
-          DB.withImmediateTransaction db $
-            let hs = decodeLatin1 . strEncode $ ([h, h'] :: NonEmpty TransportHost)
-             in DB.execute db "UPDATE servers SET host = ? WHERE host = ?" (hs, decodeLatin1 $ strEncode h)
-    runDown DownMigration {downName, downQuery} = do
-      DB.withImmediateTransaction db $ do
-        execSQL downQuery
-        DB.execute db "DELETE FROM migrations WHERE name = ?" (Only downName)
-    execSQL = SQLite3.exec $ DB.connectionHandle db
+        insert db = DB.execute db "INSERT INTO migrations (name, down, ts) VALUES (?,?,?)" . (name,down,) =<< getCurrentTime
+        updateServers db = forM_ (M.assocs extraSMPServerHosts) $ \(h, h') ->
+          let hs = decodeLatin1 . strEncode $ ([h, h'] :: NonEmpty TransportHost)
+           in DB.execute db "UPDATE servers SET host = ? WHERE host = ?" (hs, decodeLatin1 $ strEncode h)
+    runDown DownMigration {downName, downQuery} = withTransaction' st $ \db -> do
+      execSQL db downQuery
+      DB.execute db "DELETE FROM migrations WHERE name = ?" (Only downName)
+    execSQL db = SQLite3.exec $ DB.connectionHandle db
 
-initialize :: Connection -> IO ()
-initialize db = do
+initialize :: SQLiteStore -> IO ()
+initialize st = withTransaction' st $ \db -> do
   cs :: [Text] <- map fromOnly <$> DB.query_ db "SELECT name FROM pragma_table_info('migrations')"
   case cs of
-    [] -> createMigrations
+    [] -> createMigrations db
     _ -> when ("down" `notElem` cs) $ DB.execute_ db "ALTER TABLE migrations ADD COLUMN down TEXT"
   where
-    createMigrations =
+    createMigrations db =
       DB.execute_
         db
         [sql|
@@ -156,11 +170,7 @@ data MigrationsToRun = MTRUp [Migration] | MTRDown [DownMigration] | MTRNone
 data MTRError
   = MTRENoDown {dbMigrations :: [String]}
   | MTREDifferent {appMigration :: String, dbMigration :: String}
-  deriving (Eq, Show, Generic)
-
-instance ToJSON MTRError where
-  toJSON = J.genericToJSON . sumTypeJSON $ dropPrefix "MTRE"
-  toEncoding = J.genericToEncoding . sumTypeJSON $ dropPrefix "MTRE"
+  deriving (Eq, Show)
 
 mtrErrorDescription :: MTRError -> String
 mtrErrorDescription = \case
@@ -179,3 +189,5 @@ migrationsToRun [] dbMs
 migrationsToRun (a : as) (d : ds)
   | name a == name d = migrationsToRun as ds
   | otherwise = Left $ MTREDifferent (name a) (name d)
+
+$(J.deriveJSON (sumTypeJSON $ dropPrefix "MTRE") ''MTRError)

src/Simplex/Messaging/Agent/Store/SQLite/Migrations/M20230531_switch_status.hs

@@ -0,0 +1,22 @@
+{-# LANGUAGE QuasiQuotes #-}
+
+module Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230531_switch_status where
+
+import Database.SQLite.Simple (Query)
+import Database.SQLite.Simple.QQ (sql)
+
+m20230531_switch_status :: Query
+m20230531_switch_status =
+  [sql|
+ALTER TABLE rcv_queues ADD COLUMN switch_status TEXT;
+ALTER TABLE rcv_queues ADD COLUMN deleted INTEGER NOT NULL DEFAULT 0;
+ALTER TABLE snd_queues ADD COLUMN switch_status TEXT;
+|]
+
+down_m20230531_switch_status :: Query
+down_m20230531_switch_status =
+  [sql|
+ALTER TABLE snd_queues DROP COLUMN switch_status;
+ALTER TABLE rcv_queues DROP COLUMN deleted;
+ALTER TABLE rcv_queues DROP COLUMN switch_status;
+|]

src/Simplex/Messaging/Agent/Store/SQLite/Migrations/M20230615_ratchet_sync.hs

@@ -0,0 +1,41 @@
+{-# LANGUAGE QuasiQuotes #-}
+
+module Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230615_ratchet_sync where
+
+import Database.SQLite.Simple (Query)
+import Database.SQLite.Simple.QQ (sql)
+
+-- Ratchet public keys are saved when ratchet re-synchronization is started - upon receiving other party's public keys,
+-- keys are compared to determine ratchet initialization ordering for both parties.
+-- This solves a possible race when both parties start ratchet re-synchronization at the same time.
+m20230615_ratchet_sync :: Query
+m20230615_ratchet_sync =
+  [sql|
+ALTER TABLE connections ADD COLUMN ratchet_sync_state TEXT NOT NULL DEFAULT 'ok';
+
+ALTER TABLE ratchets ADD COLUMN x3dh_pub_key_1 BLOB;
+ALTER TABLE ratchets ADD COLUMN x3dh_pub_key_2 BLOB;
+
+CREATE TABLE processed_ratchet_key_hashes(
+  processed_ratchet_key_hash_id INTEGER PRIMARY KEY,
+  conn_id BLOB NOT NULL REFERENCES connections ON DELETE CASCADE,
+  hash BLOB NOT NULL,
+  created_at TEXT NOT NULL DEFAULT (datetime('now')),
+  updated_at TEXT NOT NULL DEFAULT (datetime('now'))
+);
+
+CREATE INDEX idx_processed_ratchet_key_hashes_hash ON processed_ratchet_key_hashes(conn_id, hash);
+|]
+
+down_m20230615_ratchet_sync :: Query
+down_m20230615_ratchet_sync =
+  [sql|
+DROP INDEX idx_processed_ratchet_key_hashes_hash;
+
+DROP TABLE processed_ratchet_key_hashes;
+
+ALTER TABLE ratchets DROP COLUMN x3dh_pub_key_2;
+ALTER TABLE ratchets DROP COLUMN x3dh_pub_key_1;
+
+ALTER TABLE connections DROP COLUMN ratchet_sync_state;
+|]

src/Simplex/Messaging/Agent/Store/SQLite/Migrations/M20230701_delivery_receipts.hs

@@ -0,0 +1,24 @@
+{-# LANGUAGE QuasiQuotes #-}
+
+module Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230701_delivery_receipts where
+
+import Database.SQLite.Simple (Query)
+import Database.SQLite.Simple.QQ (sql)
+
+m20230701_delivery_receipts :: Query
+m20230701_delivery_receipts =
+  [sql|
+ALTER TABLE snd_messages ADD COLUMN rcpt_internal_id INTEGER;
+ALTER TABLE snd_messages ADD COLUMN rcpt_status TEXT;
+
+CREATE INDEX idx_snd_messages_rcpt_internal_id ON snd_messages(conn_id, rcpt_internal_id);
+|]
+
+down_m20230701_delivery_receipts :: Query
+down_m20230701_delivery_receipts =
+  [sql|
+DROP INDEX idx_snd_messages_rcpt_internal_id;
+
+ALTER TABLE snd_messages DROP COLUMN rcpt_internal_id;
+ALTER TABLE snd_messages DROP COLUMN rcpt_status;
+|]

src/Simplex/Messaging/Agent/Store/SQLite/Migrations/M20230720_delete_expired_messages.hs

@@ -0,0 +1,20 @@
+{-# LANGUAGE QuasiQuotes #-}
+
+module Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230720_delete_expired_messages where
+
+import Database.SQLite.Simple (Query)
+import Database.SQLite.Simple.QQ (sql)
+
+m20230720_delete_expired_messages :: Query
+m20230720_delete_expired_messages =
+  [sql|
+CREATE INDEX idx_messages_internal_snd_id_ts ON messages(internal_snd_id, internal_ts);
+
+DELETE FROM messages WHERE internal_snd_id IS NOT NULL AND internal_ts < datetime('now', '-3 days');
+|]
+
+down_m20230720_delete_expired_messages :: Query
+down_m20230720_delete_expired_messages =
+  [sql|
+DROP INDEX idx_messages_internal_snd_id_ts;
+|]

src/Simplex/Messaging/Agent/Store/SQLite/Migrations/M20230722_indexes.hs

@@ -0,0 +1,20 @@
+{-# LANGUAGE QuasiQuotes #-}
+
+module Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230722_indexes where
+
+import Database.SQLite.Simple (Query)
+import Database.SQLite.Simple.QQ (sql)
+
+m20230722_indexes :: Query
+m20230722_indexes =
+  [sql|
+CREATE INDEX idx_processed_ratchet_key_hashes_created_at ON processed_ratchet_key_hashes(created_at);
+CREATE INDEX idx_encrypted_rcv_message_hashes_created_at ON encrypted_rcv_message_hashes(created_at);
+|]
+
+down_m20230722_indexes :: Query
+down_m20230722_indexes =
+  [sql|
+DROP INDEX idx_encrypted_rcv_message_hashes_created_at;
+DROP INDEX idx_processed_ratchet_key_hashes_created_at;
+|]

src/Simplex/Messaging/Agent/Store/SQLite/Migrations/M20230814_indexes.hs

@@ -0,0 +1,22 @@
+{-# LANGUAGE QuasiQuotes #-}
+
+module Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230814_indexes where
+
+import Database.SQLite.Simple (Query)
+import Database.SQLite.Simple.QQ (sql)
+
+m20230814_indexes :: Query
+m20230814_indexes =
+  [sql|
+DROP INDEX idx_messages_internal_snd_id_ts;
+
+CREATE INDEX idx_messages_internal_ts ON messages(internal_ts);
+|]
+
+down_m20230814_indexes :: Query
+down_m20230814_indexes =
+  [sql|
+DROP INDEX idx_messages_internal_ts;
+
+CREATE INDEX idx_messages_internal_snd_id_ts ON messages(internal_snd_id, internal_ts);
+|]

src/Simplex/Messaging/Agent/Store/SQLite/Migrations/M20230829_crypto_files.hs

@@ -0,0 +1,24 @@
+{-# LANGUAGE QuasiQuotes #-}
+
+module Simplex.Messaging.Agent.Store.SQLite.Migrations.M20230829_crypto_files where
+
+import Database.SQLite.Simple (Query)
+import Database.SQLite.Simple.QQ (sql)
+
+m20230829_crypto_files :: Query
+m20230829_crypto_files =
+  [sql|
+ALTER TABLE rcv_files ADD COLUMN save_file_key BLOB;
+ALTER TABLE rcv_files ADD COLUMN save_file_nonce BLOB;
+ALTER TABLE snd_files ADD COLUMN src_file_key BLOB;
+ALTER TABLE snd_files ADD COLUMN src_file_nonce BLOB;
+|]
+
+down_m20230829_crypto_files :: Query
+down_m20230829_crypto_files =
+  [sql|
+ALTER TABLE rcv_files DROP COLUMN save_file_key;
+ALTER TABLE rcv_files DROP COLUMN save_file_nonce;
+ALTER TABLE snd_files DROP COLUMN src_file_key;
+ALTER TABLE snd_files DROP COLUMN src_file_nonce;
+|]

src/Simplex/Messaging/Agent/Store/SQLite/Migrations/M20231222_command_created_at.hs

@@ -0,0 +1,20 @@
+{-# LANGUAGE QuasiQuotes #-}
+
+module Simplex.Messaging.Agent.Store.SQLite.Migrations.M20231222_command_created_at where
+
+import Database.SQLite.Simple (Query)
+import Database.SQLite.Simple.QQ (sql)
+
+m20231222_command_created_at :: Query
+m20231222_command_created_at =
+  [sql|
+ALTER TABLE commands ADD COLUMN created_at TEXT NOT NULL DEFAULT('1970-01-01 00:00:00');
+CREATE INDEX idx_commands_server_commands ON commands(host, port, created_at, command_id);
+|]
+
+down_m20231222_command_created_at :: Query
+down_m20231222_command_created_at =
+  [sql|
+DROP INDEX idx_commands_server_commands;
+ALTER TABLE commands DROP COLUMN created_at;
+|]

src/Simplex/Messaging/Agent/Store/SQLite/Migrations/M20231225_failed_work_items.hs

@@ -0,0 +1,38 @@
+{-# LANGUAGE QuasiQuotes #-}
+
+module Simplex.Messaging.Agent.Store.SQLite.Migrations.M20231225_failed_work_items where
+
+import Database.SQLite.Simple (Query)
+import Database.SQLite.Simple.QQ (sql)
+
+m20231225_failed_work_items :: Query
+m20231225_failed_work_items =
+    [sql|
+ALTER TABLE snd_message_deliveries ADD COLUMN failed INTEGER DEFAULT 0;
+ALTER TABLE commands ADD COLUMN failed INTEGER DEFAULT 0;
+ALTER TABLE ntf_subscriptions ADD COLUMN ntf_failed INTEGER DEFAULT 0;
+ALTER TABLE ntf_subscriptions ADD COLUMN smp_failed INTEGER DEFAULT 0;
+ALTER TABLE rcv_files ADD COLUMN failed INTEGER DEFAULT 0;
+ALTER TABLE snd_files ADD COLUMN failed INTEGER DEFAULT 0;
+ALTER TABLE deleted_snd_chunk_replicas ADD COLUMN failed INTEGER DEFAULT 0;
+
+CREATE INDEX idx_rcv_files_status_created_at ON rcv_files(status, created_at);
+CREATE INDEX idx_snd_files_status_created_at ON snd_files(status, created_at);
+CREATE INDEX idx_snd_files_snd_file_entity_id ON snd_files(snd_file_entity_id);
+|]
+
+down_m20231225_failed_work_items :: Query
+down_m20231225_failed_work_items =
+    [sql|
+DROP INDEX idx_rcv_files_status_created_at;
+DROP INDEX idx_snd_files_status_created_at;
+DROP INDEX idx_snd_files_snd_file_entity_id;
+
+ALTER TABLE snd_message_deliveries DROP COLUMN failed;
+ALTER TABLE commands DROP COLUMN failed;
+ALTER TABLE ntf_subscriptions DROP COLUMN ntf_failed;
+ALTER TABLE ntf_subscriptions DROP COLUMN smp_failed;
+ALTER TABLE rcv_files DROP COLUMN failed;
+ALTER TABLE snd_files DROP COLUMN failed;
+ALTER TABLE deleted_snd_chunk_replicas DROP COLUMN failed;
+|]

src/Simplex/Messaging/Agent/Store/SQLite/Migrations/agent_schema.sql

@@ -25,7 +25,8 @@ CREATE TABLE connections(
   enable_ntfs INTEGER,
   deleted INTEGER DEFAULT 0 CHECK(deleted NOT NULL),
   user_id INTEGER CHECK(user_id NOT NULL)
-  REFERENCES users ON DELETE CASCADE
+  REFERENCES users ON DELETE CASCADE,
+  ratchet_sync_state TEXT NOT NULL DEFAULT 'ok'
 ) WITHOUT ROWID;
 CREATE TABLE rcv_queues(
   host TEXT NOT NULL,
@@ -50,6 +51,8 @@ CREATE TABLE rcv_queues(
   replace_rcv_queue_id INTEGER NULL,
   delete_errors INTEGER DEFAULT 0 CHECK(delete_errors NOT NULL),
   server_key_hash BLOB,
+  switch_status TEXT,
+  deleted INTEGER NOT NULL DEFAULT 0,
   PRIMARY KEY(host, port, rcv_id),
   FOREIGN KEY(host, port) REFERENCES servers
   ON DELETE RESTRICT ON UPDATE CASCADE,
@@ -71,6 +74,7 @@ CREATE TABLE snd_queues(
   snd_primary INTEGER CHECK(snd_primary NOT NULL),
   replace_snd_queue_id INTEGER NULL,
   server_key_hash BLOB,
+  switch_status TEXT,
   PRIMARY KEY(host, port, snd_id),
   FOREIGN KEY(host, port) REFERENCES servers
   ON DELETE RESTRICT ON UPDATE CASCADE
@@ -115,6 +119,8 @@ CREATE TABLE snd_messages(
   previous_msg_hash BLOB NOT NULL DEFAULT x'',
   retry_int_slow INTEGER,
   retry_int_fast INTEGER,
+  rcpt_internal_id INTEGER,
+  rcpt_status TEXT,
   PRIMARY KEY(conn_id, internal_snd_id),
   FOREIGN KEY(conn_id, internal_id) REFERENCES messages
   ON DELETE CASCADE
@@ -151,6 +157,9 @@ CREATE TABLE ratchets(
   -- ratchet is initially empty on the receiving side(the side offering the connection)
   ratchet_state BLOB,
   e2e_version INTEGER NOT NULL DEFAULT 1
+  ,
+  x3dh_pub_key_1 BLOB,
+  x3dh_pub_key_2 BLOB
 ) WITHOUT ROWID;
 CREATE TABLE skipped_messages(
   skipped_message_id INTEGER PRIMARY KEY,
@@ -204,6 +213,8 @@ CREATE TABLE ntf_subscriptions(
   created_at TEXT NOT NULL DEFAULT(datetime('now')),
   updated_at TEXT NOT NULL DEFAULT(datetime('now')),
   smp_server_key_hash BLOB,
+  ntf_failed INTEGER DEFAULT 0,
+  smp_failed INTEGER DEFAULT 0,
   PRIMARY KEY(conn_id),
   FOREIGN KEY(smp_host, smp_port) REFERENCES servers(host, port)
   ON DELETE SET NULL ON UPDATE CASCADE,
@@ -220,6 +231,8 @@ CREATE TABLE commands(
   command BLOB NOT NULL,
   agent_version INTEGER NOT NULL DEFAULT 1,
   server_key_hash BLOB,
+  created_at TEXT NOT NULL DEFAULT('1970-01-01 00:00:00'),
+  failed INTEGER DEFAULT 0,
   FOREIGN KEY(host, port) REFERENCES servers
   ON DELETE RESTRICT ON UPDATE CASCADE
 );
@@ -228,6 +241,7 @@ CREATE TABLE snd_message_deliveries(
   conn_id BLOB NOT NULL REFERENCES connections ON DELETE CASCADE,
   snd_queue_id INTEGER NOT NULL,
   internal_id INTEGER NOT NULL,
+  failed INTEGER DEFAULT 0,
   FOREIGN KEY(conn_id, internal_id) REFERENCES messages ON DELETE CASCADE DEFERRABLE INITIALLY DEFERRED
 );
 CREATE TABLE sqlite_sequence(name,seq);
@@ -262,6 +276,9 @@ CREATE TABLE rcv_files(
   error TEXT,
   created_at TEXT NOT NULL DEFAULT(datetime('now')),
   updated_at TEXT NOT NULL DEFAULT(datetime('now')),
+  save_file_key BLOB,
+  save_file_nonce BLOB,
+  failed INTEGER DEFAULT 0,
   UNIQUE(rcv_file_entity_id)
 );
 CREATE TABLE rcv_file_chunks(
@@ -302,6 +319,10 @@ CREATE TABLE snd_files(
   error TEXT,
   created_at TEXT NOT NULL DEFAULT(datetime('now')),
   updated_at TEXT NOT NULL DEFAULT(datetime('now'))
+  ,
+  src_file_key BLOB,
+  src_file_nonce BLOB,
+  failed INTEGER DEFAULT 0
 );
 CREATE TABLE snd_file_chunks(
   snd_file_chunk_id INTEGER PRIMARY KEY,
@@ -345,6 +366,8 @@ CREATE TABLE deleted_snd_chunk_replicas(
   retries INTEGER NOT NULL DEFAULT 0,
   created_at TEXT NOT NULL DEFAULT(datetime('now')),
   updated_at TEXT NOT NULL DEFAULT(datetime('now'))
+  ,
+  failed INTEGER DEFAULT 0
 );
 CREATE TABLE encrypted_rcv_message_hashes(
   encrypted_rcv_message_hash_id INTEGER PRIMARY KEY,
@@ -353,6 +376,13 @@ CREATE TABLE encrypted_rcv_message_hashes(
   created_at TEXT NOT NULL DEFAULT(datetime('now')),
   updated_at TEXT NOT NULL DEFAULT(datetime('now'))
 );
+CREATE TABLE processed_ratchet_key_hashes(
+  processed_ratchet_key_hash_id INTEGER PRIMARY KEY,
+  conn_id BLOB NOT NULL REFERENCES connections ON DELETE CASCADE,
+  hash BLOB NOT NULL,
+  created_at TEXT NOT NULL DEFAULT(datetime('now')),
+  updated_at TEXT NOT NULL DEFAULT(datetime('now'))
+);
 CREATE UNIQUE INDEX idx_rcv_queues_ntf ON rcv_queues(host, port, ntf_id);
 CREATE UNIQUE INDEX idx_rcv_queue_id ON rcv_queues(conn_id, rcv_queue_id);
 CREATE UNIQUE INDEX idx_snd_queue_id ON snd_queues(conn_id, snd_queue_id);
@@ -443,3 +473,27 @@ CREATE INDEX idx_encrypted_rcv_message_hashes_hash ON encrypted_rcv_message_hash
   conn_id,
   hash
 );
+CREATE INDEX idx_processed_ratchet_key_hashes_hash ON processed_ratchet_key_hashes(
+  conn_id,
+  hash
+);
+CREATE INDEX idx_snd_messages_rcpt_internal_id ON snd_messages(
+  conn_id,
+  rcpt_internal_id
+);
+CREATE INDEX idx_processed_ratchet_key_hashes_created_at ON processed_ratchet_key_hashes(
+  created_at
+);
+CREATE INDEX idx_encrypted_rcv_message_hashes_created_at ON encrypted_rcv_message_hashes(
+  created_at
+);
+CREATE INDEX idx_messages_internal_ts ON messages(internal_ts);
+CREATE INDEX idx_commands_server_commands ON commands(
+  host,
+  port,
+  created_at,
+  command_id
+);
+CREATE INDEX idx_rcv_files_status_created_at ON rcv_files(status, created_at);
+CREATE INDEX idx_snd_files_status_created_at ON snd_files(status, created_at);
+CREATE INDEX idx_snd_files_snd_file_entity_id ON snd_files(snd_file_entity_id);

src/Simplex/Messaging/Agent/TAsyncs.hs

@@ -1,6 +1,5 @@
 module Simplex.Messaging.Agent.TAsyncs where
 
-import Control.Concurrent.STM (stateTVar)
 import Control.Monad.IO.Unlift (MonadUnliftIO)
 import Simplex.Messaging.TMap (TMap)
 import qualified Simplex.Messaging.TMap as TM

src/Simplex/Messaging/Agent/TRcvQueues.hs

@@ -1,5 +1,5 @@
 module Simplex.Messaging.Agent.TRcvQueues
-  ( TRcvQueues,
+  ( TRcvQueues (getRcvQueues),
     empty,
     clear,
     deleteConn,
@@ -9,6 +9,7 @@ module Simplex.Messaging.Agent.TRcvQueues
     deleteQueue,
     getSessQueues,
     getDelSessQueues,
+    qKey,
   )
 where
 
@@ -17,12 +18,12 @@ import qualified Data.Map.Strict as M
 import Data.Set (Set)
 import qualified Data.Set as S
 import Simplex.Messaging.Agent.Protocol (ConnId, UserId)
-import Simplex.Messaging.Agent.Store (RcvQueue (..))
+import Simplex.Messaging.Agent.Store (RcvQueue, StoredRcvQueue (..))
 import Simplex.Messaging.Protocol (RecipientId, SMPServer)
 import Simplex.Messaging.TMap (TMap)
 import qualified Simplex.Messaging.TMap as TM
 
-newtype TRcvQueues = TRcvQueues (TMap (UserId, SMPServer, RecipientId) RcvQueue)
+newtype TRcvQueues = TRcvQueues {getRcvQueues :: TMap (UserId, SMPServer, RecipientId) RcvQueue}
 
 empty :: STM TRcvQueues
 empty = TRcvQueues <$> TM.empty

src/Simplex/Messaging/Client.hs

@@ -1,6 +1,5 @@
 {-# LANGUAGE DataKinds #-}
 {-# LANGUAGE DeriveAnyClass #-}
-{-# LANGUAGE DeriveGeneric #-}
 {-# LANGUAGE DuplicateRecordFields #-}
 {-# LANGUAGE FlexibleContexts #-}
 {-# LANGUAGE GADTs #-}
@@ -10,6 +9,8 @@
 {-# LANGUAGE OverloadedLists #-}
 {-# LANGUAGE OverloadedStrings #-}
 {-# LANGUAGE ScopedTypeVariables #-}
+{-# LANGUAGE TemplateHaskell #-}
+{-# LANGUAGE TupleSections #-}
 {-# LANGUAGE TypeApplications #-}
 
 -- |
@@ -39,8 +40,10 @@ module Simplex.Messaging.Client
     createSMPQueue,
     subscribeSMPQueue,
     subscribeSMPQueues,
+    streamSubscribeSMPQueues,
     getSMPMessage,
     subscribeSMPQueueNotifications,
+    subscribeSMPQueuesNtfs,
     secureSMPQueue,
     enableSMPQueueNotifications,
     disableSMPQueueNotifications,
@@ -67,6 +70,11 @@ module Simplex.Messaging.Client
     temporaryClientError,
     ServerTransmission,
     ClientCommand,
+
+    -- * For testing
+    PCTransmission,
+    mkTransmission,
+    clientStub,
   )
 where
 
@@ -76,25 +84,22 @@ import Control.Exception
 import Control.Monad
 import Control.Monad.IO.Class (liftIO)
 import Control.Monad.Trans.Except
-import Data.Aeson (FromJSON (..), ToJSON (..))
-import qualified Data.Aeson as J
+import qualified Data.Aeson.TH as J
 import Data.ByteString.Char8 (ByteString)
 import qualified Data.ByteString.Char8 as B
-import Data.Either (rights)
 import Data.Functor (($>))
 import Data.Int (Int64)
 import Data.List (find)
-import Data.List.NonEmpty (NonEmpty)
+import Data.List.NonEmpty (NonEmpty (..))
 import qualified Data.List.NonEmpty as L
 import Data.Maybe (fromMaybe)
 import Data.Time.Clock (UTCTime, getCurrentTime)
-import GHC.Generics (Generic)
 import Network.Socket (ServiceName)
 import Numeric.Natural
 import qualified Simplex.Messaging.Crypto as C
 import Simplex.Messaging.Encoding.String
-import Simplex.Messaging.Parsers (dropPrefix, enumJSON)
-import Simplex.Messaging.Protocol as SMP
+import Simplex.Messaging.Parsers (defaultJSON, dropPrefix, enumJSON)
+import Simplex.Messaging.Protocol
 import Simplex.Messaging.TMap (TMap)
 import qualified Simplex.Messaging.TMap as TM
 import Simplex.Messaging.Transport
@@ -113,6 +118,9 @@ data ProtocolClient err msg = ProtocolClient
     sessionId :: SessionId,
     sessionTs :: UTCTime,
     thVersion :: Version,
+    timeoutPerBlock :: Int,
+    blockSize :: Int,
+    batch :: Bool,
     client_ :: PClient err msg
   }
 
@@ -121,21 +129,54 @@ data PClient err msg = PClient
     transportSession :: TransportSession msg,
     transportHost :: TransportHost,
     tcpTimeout :: Int,
+    batchDelay :: Maybe Int,
     pingErrorCount :: TVar Int,
     clientCorrId :: TVar Natural,
     sentCommands :: TMap CorrId (Request err msg),
-    sndQ :: TBQueue (NonEmpty SentRawTransmission),
+    sndQ :: TBQueue ByteString,
     rcvQ :: TBQueue (NonEmpty (SignedTransmission err msg)),
     msgQ :: Maybe (TBQueue (ServerTransmission msg))
   }
 
-type SMPClient = ProtocolClient ErrorType SMP.BrokerMsg
+clientStub :: ByteString -> STM (ProtocolClient err msg)
+clientStub sessionId = do
+  connected <- newTVar False
+  clientCorrId <- newTVar 0
+  sentCommands <- TM.empty
+  sndQ <- newTBQueue 100
+  rcvQ <- newTBQueue 100
+  return
+    ProtocolClient
+      { action = Nothing,
+        sessionId,
+        sessionTs = undefined,
+        thVersion = 5,
+        timeoutPerBlock = undefined,
+        blockSize = smpBlockSize,
+        batch = undefined,
+        client_ =
+          PClient
+            { connected,
+              transportSession = undefined,
+              transportHost = undefined,
+              tcpTimeout = undefined,
+              batchDelay = Nothing,
+              pingErrorCount = undefined,
+              clientCorrId,
+              sentCommands,
+              sndQ,
+              rcvQ,
+              msgQ = Nothing
+            }
+      }
+
+type SMPClient = ProtocolClient ErrorType BrokerMsg
 
 -- | Type for client command data
-type ClientCommand msg = (Maybe C.APrivateSignKey, QueueId, ProtoCommand msg)
+type ClientCommand msg = (Maybe C.APrivateSignKey, EntityId, ProtoCommand msg)
 
 -- | Type synonym for transmission from some SPM server queue.
-type ServerTransmission msg = (TransportSession msg, Version, SessionId, QueueId, msg)
+type ServerTransmission msg = (TransportSession msg, Version, SessionId, EntityId, msg)
 
 data HostMode
   = -- | prefer (or require) onion hosts when connecting via SOCKS proxy
@@ -144,14 +185,7 @@ data HostMode
     HMOnion
   | -- | prefer (or require) public hosts
     HMPublic
-  deriving (Eq, Show, Generic)
-
-instance FromJSON HostMode where
-  parseJSON = J.genericParseJSON . enumJSON $ dropPrefix "HM"
-
-instance ToJSON HostMode where
-  toJSON = J.genericToJSON . enumJSON $ dropPrefix "HM"
-  toEncoding = J.genericToEncoding . enumJSON $ dropPrefix "HM"
+  deriving (Eq, Show)
 
 -- | network configuration for the client
 data NetworkConfig = NetworkConfig
@@ -167,6 +201,8 @@ data NetworkConfig = NetworkConfig
     tcpConnectTimeout :: Int,
     -- | timeout of protocol commands (microseconds)
     tcpTimeout :: Int,
+    -- | additional timeout per kilobyte (1024 bytes) to be sent
+    tcpTimeoutPerKb :: Int,
     -- | TCP keep-alive options, Nothing to skip enabling keep-alive
     tcpKeepAlive :: Maybe KeepAliveOpts,
     -- | period for SMP ping commands (microseconds, 0 to disable)
@@ -175,21 +211,10 @@ data NetworkConfig = NetworkConfig
     smpPingCount :: Int,
     logTLSErrors :: Bool
   }
-  deriving (Eq, Show, Generic, FromJSON)
-
-instance ToJSON NetworkConfig where
-  toJSON = J.genericToJSON J.defaultOptions {J.omitNothingFields = True}
-  toEncoding = J.genericToEncoding J.defaultOptions {J.omitNothingFields = True}
+  deriving (Eq, Show)
 
 data TransportSessionMode = TSMUser | TSMEntity
-  deriving (Eq, Show, Generic)
-
-instance ToJSON TransportSessionMode where
-  toJSON = J.genericToJSON . enumJSON $ dropPrefix "TSM"
-  toEncoding = J.genericToEncoding . enumJSON $ dropPrefix "TSM"
-
-instance FromJSON TransportSessionMode where
-  parseJSON = J.genericParseJSON . enumJSON $ dropPrefix "TSM"
+  deriving (Eq, Show)
 
 defaultNetworkConfig :: NetworkConfig
 defaultNetworkConfig =
@@ -198,8 +223,9 @@ defaultNetworkConfig =
       hostMode = HMOnionViaSocks,
       requiredHostMode = False,
       sessionMode = TSMUser,
-      tcpConnectTimeout = 7_500_000,
-      tcpTimeout = 5_000_000,
+      tcpConnectTimeout = 20_000_000,
+      tcpTimeout = 15_000_000,
+      tcpTimeoutPerKb = 45_000, -- 45ms, should be less than 130ms to avoid Int overflow on 32 bit systems
       tcpKeepAlive = Just defaultKeepAliveOpts,
       smpPingInterval = 600_000_000, -- 10min
       smpPingCount = 3,
@@ -208,7 +234,7 @@ defaultNetworkConfig =
 
 transportClientConfig :: NetworkConfig -> TransportClientConfig
 transportClientConfig NetworkConfig {socksProxy, tcpKeepAlive, logTLSErrors} =
-  TransportClientConfig {socksProxy, tcpKeepAlive, logTLSErrors}
+  TransportClientConfig {socksProxy, tcpKeepAlive, logTLSErrors, clientCredentials = Nothing}
 
 -- | protocol client configuration.
 data ProtocolClientConfig = ProtocolClientConfig
@@ -218,8 +244,10 @@ data ProtocolClientConfig = ProtocolClientConfig
     defaultTransport :: (ServiceName, ATransport),
     -- | network configuration
     networkConfig :: NetworkConfig,
-    -- | SMP client-server protocol version range
-    smpServerVRange :: VersionRange
+    -- | client-server protocol version range
+    serverVRange :: VersionRange,
+    -- | delay between sending batches of commands (microseconds)
+    batchDelay :: Maybe Int
   }
 
 -- | Default protocol client configuration.
@@ -229,15 +257,19 @@ defaultClientConfig =
     { qSize = 64,
       defaultTransport = ("443", transport @TLS),
       networkConfig = defaultNetworkConfig,
-      smpServerVRange = supportedSMPServerVRange
+      serverVRange = supportedSMPServerVRange,
+      batchDelay = Nothing
     }
 
 data Request err msg = Request
-  { queueId :: QueueId,
-    responseVar :: TMVar (Response err msg)
+  { entityId :: EntityId,
+    responseVar :: TMVar (Either (ProtocolClientError err) msg)
   }
 
-type Response err msg = Either (ProtocolClientError err) msg
+data Response err msg = Response
+  { entityId :: EntityId,
+    response :: Either (ProtocolClientError err) msg
+  }
 
 chooseTransportHost :: NetworkConfig -> NonEmpty TransportHost -> Either (ProtocolClientError err) TransportHost
 chooseTransportHost NetworkConfig {socksProxy, hostMode, requiredHostMode} hosts =
@@ -275,14 +307,14 @@ type TransportSession msg = (UserId, ProtoServer msg, Maybe EntityId)
 -- A single queue can be used for multiple 'SMPClient' instances,
 -- as 'SMPServerTransmission' includes server information.
 getProtocolClient :: forall err msg. Protocol err msg => TransportSession msg -> ProtocolClientConfig -> Maybe (TBQueue (ServerTransmission msg)) -> (ProtocolClient err msg -> IO ()) -> IO (Either (ProtocolClientError err) (ProtocolClient err msg))
-getProtocolClient transportSession@(_, srv, _) cfg@ProtocolClientConfig {qSize, networkConfig, smpServerVRange} msgQ disconnected = do
+getProtocolClient transportSession@(_, srv, _) cfg@ProtocolClientConfig {qSize, networkConfig, serverVRange, batchDelay} msgQ disconnected = do
   case chooseTransportHost networkConfig (host srv) of
     Right useHost ->
       (atomically (mkProtocolClient useHost) >>= runClient useTransport useHost)
         `catch` \(e :: IOException) -> pure . Left $ PCEIOError e
     Left e -> pure $ Left e
   where
-    NetworkConfig {tcpConnectTimeout, tcpTimeout, smpPingInterval} = networkConfig
+    NetworkConfig {tcpConnectTimeout, tcpTimeout, tcpTimeoutPerKb, smpPingInterval} = networkConfig
     mkProtocolClient :: TransportHost -> STM (PClient err msg)
     mkProtocolClient transportHost = do
       connected <- newTVar False
@@ -297,6 +329,7 @@ getProtocolClient transportSession@(_, srv, _) cfg@ProtocolClientConfig {qSize,
             transportSession,
             transportHost,
             tcpTimeout,
+            batchDelay,
             pingErrorCount,
             clientCorrId,
             sentCommands,
@@ -328,11 +361,12 @@ getProtocolClient transportSession@(_, srv, _) cfg@ProtocolClientConfig {qSize,
 
     client :: forall c. Transport c => TProxy c -> PClient err msg -> TMVar (Either (ProtocolClientError err) (ProtocolClient err msg)) -> c -> IO ()
     client _ c cVar h =
-      runExceptT (protocolClientHandshake @err @msg h (keyHash srv) smpServerVRange) >>= \case
+      runExceptT (protocolClientHandshake @err @msg h (keyHash srv) serverVRange) >>= \case
         Left e -> atomically . putTMVar cVar . Left $ PCETransportError e
-        Right th@THandle {sessionId, thVersion} -> do
+        Right th@THandle {sessionId, thVersion, blockSize, batch} -> do
           sessionTs <- getCurrentTime
-          let c' = ProtocolClient {action = Nothing, client_ = c, sessionId, thVersion, sessionTs}
+          let timeoutPerBlock = (blockSize * tcpTimeoutPerKb) `div` 1024
+              c' = ProtocolClient {action = Nothing, client_ = c, sessionId, thVersion, sessionTs, timeoutPerBlock, blockSize, batch}
           atomically $ do
             writeTVar (connected c) True
             putTMVar cVar $ Right c'
@@ -340,7 +374,7 @@ getProtocolClient transportSession@(_, srv, _) cfg@ProtocolClientConfig {qSize,
             `finally` disconnected c'
 
     send :: Transport c => ProtocolClient err msg -> THandle c -> IO ()
-    send ProtocolClient {client_ = PClient {sndQ}} h = forever $ atomically (readTBQueue sndQ) >>= tPut h
+    send ProtocolClient {client_ = PClient {sndQ}} h = forever $ atomically (readTBQueue sndQ) >>= tPutLog h
 
     receive :: Transport c => ProtocolClient err msg -> THandle c -> IO ()
     receive ProtocolClient {client_ = PClient {rcvQ}} h = forever $ tGet h >>= atomically . writeTBQueue rcvQ
@@ -360,26 +394,27 @@ getProtocolClient transportSession@(_, srv, _) cfg@ProtocolClientConfig {qSize,
     process c = forever $ atomically (readTBQueue $ rcvQ $ client_ c) >>= mapM_ (processMsg c)
 
     processMsg :: ProtocolClient err msg -> SignedTransmission err msg -> IO ()
-    processMsg c@ProtocolClient {client_ = PClient {sentCommands}} (_, _, (corrId, qId, respOrErr)) =
+    processMsg c@ProtocolClient {client_ = PClient {sentCommands}} (_, _, (corrId, entId, respOrErr)) =
       if B.null $ bs corrId
         then sendMsg respOrErr
         else do
           atomically (TM.lookup corrId sentCommands) >>= \case
             Nothing -> sendMsg respOrErr
-            Just Request {queueId, responseVar} -> atomically $ do
+            Just Request {entityId, responseVar} -> atomically $ do
               TM.delete corrId sentCommands
-              putTMVar responseVar $
-                if queueId == qId
-                  then case respOrErr of
-                    Left e -> Left $ PCEResponseError e
-                    Right r -> case protocolError r of
-                      Just e -> Left $ PCEProtocolError e
-                      _ -> Right r
-                  else Left . PCEUnexpectedResponse $ bshow respOrErr
+              putTMVar responseVar $ response entityId
       where
+        response entityId
+          | entityId == entId =
+              case respOrErr of
+                Left e -> Left $ PCEResponseError e
+                Right r -> case protocolError r of
+                  Just e -> Left $ PCEProtocolError e
+                  _ -> Right r
+          | otherwise = Left . PCEUnexpectedResponse $ bshow respOrErr
         sendMsg :: Either err msg -> IO ()
         sendMsg = \case
-          Right msg -> atomically $ mapM_ (`writeTBQueue` serverTransmission c qId msg) msgQ
+          Right msg -> atomically $ mapM_ (`writeTBQueue` serverTransmission c entId msg) msgQ
           Left e -> putStrLn $ "SMP client error: " <> show e
 
 proxyUsername :: TransportSession msg -> ByteString
@@ -437,9 +472,10 @@ createSMPQueue ::
   RcvPublicVerifyKey ->
   RcvPublicDhKey ->
   Maybe BasicAuth ->
+  SubscriptionMode ->
   ExceptT SMPClientError IO QueueIdsKeys
-createSMPQueue c rpKey rKey dhKey auth =
-  sendSMPCommand c (Just rpKey) "" (NEW rKey dhKey auth) >>= \case
+createSMPQueue c rpKey rKey dhKey auth subMode =
+  sendSMPCommand c (Just rpKey) "" (NEW rKey dhKey auth subMode) >>= \case
     IDS qik -> pure qik
     r -> throwE . PCEUnexpectedResponse $ bshow r
 
@@ -455,14 +491,22 @@ subscribeSMPQueue c rpKey rId =
 
 -- | Subscribe to multiple SMP queues batching commands if supported.
 subscribeSMPQueues :: SMPClient -> NonEmpty (RcvPrivateSignKey, RecipientId) -> IO (NonEmpty (Either SMPClientError ()))
-subscribeSMPQueues c qs = sendProtocolCommands c cs >>= mapM response . L.zip qs
+subscribeSMPQueues c qs = sendProtocolCommands c cs >>= mapM (processSUBResponse c)
   where
     cs = L.map (\(rpKey, rId) -> (Just rpKey, rId, Cmd SRecipient SUB)) qs
-    response ((_, rId), r) = case r of
-      Right OK -> pure $ Right ()
-      Right cmd@MSG {} -> writeSMPMessage c rId cmd $> Right ()
-      Right r' -> pure . Left . PCEUnexpectedResponse $ bshow r'
-      Left e -> pure $ Left e
+
+streamSubscribeSMPQueues :: SMPClient -> NonEmpty (RcvPrivateSignKey, RecipientId) -> ([(RecipientId, Either SMPClientError ())] -> IO ()) -> IO ()
+streamSubscribeSMPQueues c qs cb = streamProtocolCommands c cs $ mapM process >=> cb
+  where
+    cs = L.map (\(rpKey, rId) -> (Just rpKey, rId, Cmd SRecipient SUB)) qs
+    process r@(Response rId _) = (rId,) <$> processSUBResponse c r
+
+processSUBResponse :: SMPClient -> Response ErrorType BrokerMsg -> IO (Either SMPClientError ())
+processSUBResponse c (Response rId r) = case r of
+  Right OK -> pure $ Right ()
+  Right cmd@MSG {} -> writeSMPMessage c rId cmd $> Right ()
+  Right r' -> pure . Left . PCEUnexpectedResponse $ bshow r'
+  Left e -> pure $ Left e
 
 writeSMPMessage :: SMPClient -> RecipientId -> BrokerMsg -> IO ()
 writeSMPMessage c rId msg = atomically $ mapM_ (`writeTBQueue` serverTransmission c rId msg) (msgQ $ client_ c)
@@ -487,6 +531,10 @@ getSMPMessage c rpKey rId =
 subscribeSMPQueueNotifications :: SMPClient -> NtfPrivateSignKey -> NotifierId -> ExceptT SMPClientError IO ()
 subscribeSMPQueueNotifications = okSMPCommand NSUB
 
+-- | Subscribe to multiple SMP queues notifications batching commands if supported.
+subscribeSMPQueuesNtfs :: SMPClient -> NonEmpty (NtfPrivateSignKey, NotifierId) -> IO (NonEmpty (Either SMPClientError ()))
+subscribeSMPQueuesNtfs = okSMPCommands NSUB
+
 -- | Secure the SMP queue by adding a sender public key.
 --
 -- https://github.com/simplex-chat/simplexmq/blob/master/protocol/simplex-messaging.md#secure-queue-command
@@ -504,12 +552,12 @@ enableSMPQueueNotifications c rpKey rId notifierKey rcvNtfPublicDhKey =
 
 -- | Enable notifications for the multiple queues for push notifications server.
 enableSMPQueuesNtfs :: SMPClient -> NonEmpty (RcvPrivateSignKey, RecipientId, NtfPublicVerifyKey, RcvNtfPublicDhKey) -> IO (NonEmpty (Either SMPClientError (NotifierId, RcvNtfPublicDhKey)))
-enableSMPQueuesNtfs c qs = L.map response <$> sendProtocolCommands c cs
+enableSMPQueuesNtfs c qs = L.map process <$> sendProtocolCommands c cs
   where
     cs = L.map (\(rpKey, rId, notifierKey, rcvNtfPublicDhKey) -> (Just rpKey, rId, Cmd SRecipient $ NKEY notifierKey rcvNtfPublicDhKey)) qs
-    response = \case
+    process (Response _ r) = case r of
       Right (NID nId rcvNtfSrvPublicDhKey) -> Right (nId, rcvNtfSrvPublicDhKey)
-      Right r -> Left . PCEUnexpectedResponse $ bshow r
+      Right r' -> Left . PCEUnexpectedResponse $ bshow r'
       Left e -> Left e
 
 -- | Disable notifications for the queue for push notifications server.
@@ -565,49 +613,88 @@ okSMPCommand cmd c pKey qId =
     r -> throwE . PCEUnexpectedResponse $ bshow r
 
 okSMPCommands :: PartyI p => Command p -> SMPClient -> NonEmpty (C.APrivateSignKey, QueueId) -> IO (NonEmpty (Either SMPClientError ()))
-okSMPCommands cmd c qs = L.map response <$> sendProtocolCommands c cs
+okSMPCommands cmd c qs = L.map process <$> sendProtocolCommands c cs
   where
     aCmd = Cmd sParty cmd
     cs = L.map (\(pKey, qId) -> (Just pKey, qId, aCmd)) qs
-    response = \case
+    process (Response _ r) = case r of
       Right OK -> Right ()
-      Right r -> Left . PCEUnexpectedResponse $ bshow r
+      Right r' -> Left . PCEUnexpectedResponse $ bshow r'
       Left e -> Left e
 
 -- | Send SMP command
 sendSMPCommand :: PartyI p => SMPClient -> Maybe C.APrivateSignKey -> QueueId -> Command p -> ExceptT SMPClientError IO BrokerMsg
 sendSMPCommand c pKey qId cmd = sendProtocolCommand c pKey qId (Cmd sParty cmd)
 
+type PCTransmission err msg = (SentRawTransmission, Request err msg)
+
 -- | Send multiple commands with batching and collect responses
-sendProtocolCommands :: forall err msg. ProtocolEncoding err (ProtoCommand msg) => ProtocolClient err msg -> NonEmpty (ClientCommand msg) -> IO (NonEmpty (Either (ProtocolClientError err) msg))
-sendProtocolCommands c@ProtocolClient {client_ = PClient {sndQ}} cs = do
-  ts <- mapM (runExceptT . mkTransmission c) cs
-  mapM_ (atomically . writeTBQueue sndQ . L.map fst) . L.nonEmpty . rights $ L.toList ts
-  forConcurrently ts $ \case
-    Right (_, r) -> withTimeout c $ atomically $ takeTMVar r
-    Left e -> pure $ Left e
+sendProtocolCommands :: forall err msg. ProtocolEncoding err (ProtoCommand msg) => ProtocolClient err msg -> NonEmpty (ClientCommand msg) -> IO (NonEmpty (Response err msg))
+sendProtocolCommands c@ProtocolClient {batch, blockSize} cs = do
+  bs <- batchTransmissions' batch blockSize <$> mapM (mkTransmission c) cs
+  validate . concat =<< mapM (sendBatch c) bs
+  where
+    validate :: [Response err msg] -> IO (NonEmpty (Response err msg))
+    validate rs
+      | diff == 0 = pure $ L.fromList rs
+      | diff > 0 = do
+          putStrLn "send error: fewer responses than expected"
+          pure $ L.fromList $ rs <> replicate diff (Response "" $ Left $ PCETransportError TEBadBlock)
+      | otherwise = do
+          putStrLn "send error: more responses than expected"
+          pure $ L.fromList $ take (L.length cs) rs
+      where
+        diff = L.length cs - length rs
+
+streamProtocolCommands :: forall err msg. ProtocolEncoding err (ProtoCommand msg) => ProtocolClient err msg -> NonEmpty (ClientCommand msg) -> ([Response err msg] -> IO ()) -> IO ()
+streamProtocolCommands c@ProtocolClient {batch, blockSize} cs cb = do
+  bs <- batchTransmissions' batch blockSize <$> mapM (mkTransmission c) cs
+  mapM_ (cb <=< sendBatch c) bs
+
+sendBatch :: ProtocolClient err msg -> TransportBatch (Request err msg) -> IO [Response err msg]
+sendBatch c@ProtocolClient {client_ = PClient {sndQ}} b = do
+  case b of
+    TBLargeTransmission Request {entityId} -> do
+      putStrLn "send error: large message"
+      pure [Response entityId $ Left $ PCETransportError TELargeMsg]
+    TBTransmissions s n rs
+      | n > 0 -> do
+          atomically $ writeTBQueue sndQ s
+          mapConcurrently (getResponse c) rs
+      | otherwise -> pure []
+    TBTransmission s r -> do
+      atomically $ writeTBQueue sndQ s
+      (: []) <$> getResponse c r
 
 -- | Send Protocol command
-sendProtocolCommand :: forall err msg. ProtocolEncoding err (ProtoCommand msg) => ProtocolClient err msg -> Maybe C.APrivateSignKey -> QueueId -> ProtoCommand msg -> ExceptT (ProtocolClientError err) IO msg
-sendProtocolCommand c@ProtocolClient {client_ = PClient {sndQ}} pKey qId cmd = do
-  (t, r) <- mkTransmission c (pKey, qId, cmd)
-  ExceptT $ sendRecv t r
+sendProtocolCommand :: forall err msg. ProtocolEncoding err (ProtoCommand msg) => ProtocolClient err msg -> Maybe C.APrivateSignKey -> EntityId -> ProtoCommand msg -> ExceptT (ProtocolClientError err) IO msg
+sendProtocolCommand c@ProtocolClient {client_ = PClient {sndQ}, batch, blockSize} pKey entId cmd =
+  ExceptT $ uncurry sendRecv =<< mkTransmission c (pKey, entId, cmd)
   where
     -- two separate "atomically" needed to avoid blocking
-    sendRecv :: SentRawTransmission -> TMVar (Response err msg) -> IO (Response err msg)
-    sendRecv t r = atomically (writeTBQueue sndQ [t]) >> withTimeout c (atomically $ takeTMVar r)
+    sendRecv :: SentRawTransmission -> Request err msg -> IO (Either (ProtocolClientError err) msg)
+    sendRecv t r
+      | B.length s > blockSize - 2 = pure $ Left $ PCETransportError TELargeMsg
+      | otherwise = atomically (writeTBQueue sndQ s) >> response <$> getResponse c r
+      where
+        s
+          | batch = tEncodeBatch1 t
+          | otherwise = tEncode t
 
-withTimeout :: ProtocolClient err msg -> IO (Either (ProtocolClientError err) msg) -> IO (Either (ProtocolClientError err) msg)
-withTimeout ProtocolClient {client_ = PClient {tcpTimeout, pingErrorCount}} a =
-  timeout tcpTimeout a >>= \case
-    Just r -> atomically (writeTVar pingErrorCount 0) >> pure r
-    _ -> pure $ Left PCEResponseTimeout
+-- TODO switch to timeout or TimeManager that supports Int64
+getResponse :: ProtocolClient err msg -> Request err msg -> IO (Response err msg)
+getResponse ProtocolClient {client_ = PClient {tcpTimeout, pingErrorCount}} Request {entityId, responseVar} = do
+  response <-
+    timeout tcpTimeout (atomically (takeTMVar responseVar)) >>= \case
+      Just r -> atomically (writeTVar pingErrorCount 0) $> r
+      Nothing -> pure $ Left PCEResponseTimeout
+  pure Response {entityId, response}
 
-mkTransmission :: forall err msg. ProtocolEncoding err (ProtoCommand msg) => ProtocolClient err msg -> ClientCommand msg -> ExceptT (ProtocolClientError err) IO (SentRawTransmission, TMVar (Response err msg))
-mkTransmission ProtocolClient {sessionId, thVersion, client_ = PClient {clientCorrId, sentCommands}} (pKey, qId, cmd) = do
-  corrId <- liftIO $ atomically getNextCorrId
-  let t = signTransmission $ encodeTransmission thVersion sessionId (corrId, qId, cmd)
-  r <- liftIO . atomically $ mkRequest corrId
+mkTransmission :: forall err msg. ProtocolEncoding err (ProtoCommand msg) => ProtocolClient err msg -> ClientCommand msg -> IO (PCTransmission err msg)
+mkTransmission ProtocolClient {sessionId, thVersion, client_ = PClient {clientCorrId, sentCommands}} (pKey, entId, cmd) = do
+  corrId <- atomically getNextCorrId
+  let t = signTransmission $ encodeTransmission thVersion sessionId (corrId, entId, cmd)
+  r <- atomically $ mkRequest corrId
   pure (t, r)
   where
     getNextCorrId :: STM CorrId
@@ -616,8 +703,14 @@ mkTransmission ProtocolClient {sessionId, thVersion, client_ = PClient {clientCo
       pure . CorrId $ bshow i
     signTransmission :: ByteString -> SentRawTransmission
     signTransmission t = ((`C.sign` t) <$> pKey, t)
-    mkRequest :: CorrId -> STM (TMVar (Response err msg))
+    mkRequest :: CorrId -> STM (Request err msg)
     mkRequest corrId = do
-      r <- newEmptyTMVar
-      TM.insert corrId (Request qId r) sentCommands
+      r <- Request entId <$> newEmptyTMVar
+      TM.insert corrId r sentCommands
       pure r
+
+$(J.deriveJSON (enumJSON $ dropPrefix "HM") ''HostMode)
+
+$(J.deriveJSON (enumJSON $ dropPrefix "TSM") ''TransportSessionMode)
+
+$(J.deriveJSON defaultJSON ''NetworkConfig)

src/Simplex/Messaging/Client/Agent.hs

@@ -5,6 +5,7 @@
 {-# LANGUAGE OverloadedStrings #-}
 {-# LANGUAGE RankNTypes #-}
 {-# LANGUAGE ScopedTypeVariables #-}
+{-# LANGUAGE TupleSections #-}
 {-# LANGUAGE TypeApplications #-}
 {-# OPTIONS_GHC -fno-warn-orphans #-}
 
@@ -13,27 +14,35 @@ module Simplex.Messaging.Client.Agent where
 import Control.Concurrent (forkIO)
 import Control.Concurrent.Async (Async, uninterruptibleCancel)
 import Control.Logger.Simple
+import Control.Monad
 import Control.Monad.Except
 import Control.Monad.IO.Unlift
 import Control.Monad.Trans.Except
+import Data.Bifunctor (bimap, first)
 import Data.ByteString.Char8 (ByteString)
 import qualified Data.ByteString.Char8 as B
+import Data.Either (partitionEithers)
+import Data.List (partition)
+import Data.List.NonEmpty (NonEmpty)
+import qualified Data.List.NonEmpty as L
 import Data.Map.Strict (Map)
 import qualified Data.Map.Strict as M
+import Data.Maybe (listToMaybe)
 import Data.Set (Set)
 import Data.Text.Encoding
+import Data.Tuple (swap)
 import Numeric.Natural
 import Simplex.Messaging.Agent.RetryInterval
 import Simplex.Messaging.Client
 import qualified Simplex.Messaging.Crypto as C
 import Simplex.Messaging.Encoding.String
-import Simplex.Messaging.Protocol (BrokerMsg, ProtocolServer (..), QueueId, SMPServer)
+import Simplex.Messaging.Protocol (BrokerMsg, NotifierId, NtfPrivateSignKey, ProtocolServer (..), QueueId, RcvPrivateSignKey, RecipientId, SMPServer)
 import Simplex.Messaging.TMap (TMap)
 import qualified Simplex.Messaging.TMap as TM
 import Simplex.Messaging.Transport
-import Simplex.Messaging.Util (catchAll_, tryE, unlessM, ($>>=))
+import Simplex.Messaging.Util (catchAll_, toChunks, ($>>=))
 import System.Timeout (timeout)
-import UnliftIO (async, forConcurrently_)
+import UnliftIO (async)
 import UnliftIO.Exception (Exception)
 import qualified UnliftIO.Exception as E
 import UnliftIO.STM
@@ -44,8 +53,8 @@ data SMPClientAgentEvent
   = CAConnected SMPServer
   | CADisconnected SMPServer (Set SMPSub)
   | CAReconnected SMPServer
-  | CAResubscribed SMPServer SMPSub
-  | CASubError SMPServer SMPSub SMPClientError
+  | CAResubscribed SMPServer (NonEmpty SMPSub)
+  | CASubError SMPServer (NonEmpty (SMPSub, SMPClientError))
 
 data SMPSubParty = SPRecipient | SPNotifier
   deriving (Eq, Ord, Show)
@@ -58,7 +67,8 @@ data SMPClientAgentConfig = SMPClientAgentConfig
   { smpCfg :: ProtocolClientConfig,
     reconnectInterval :: RetryInterval,
     msgQSize :: Natural,
-    agentQSize :: Natural
+    agentQSize :: Natural,
+    agentSubsBatchSize :: Int
   }
 
 defaultSMPClientAgentConfig :: SMPClientAgentConfig
@@ -71,8 +81,9 @@ defaultSMPClientAgentConfig =
             increaseAfter = 10 * second,
             maxInterval = 10 * second
           },
-      msgQSize = 64,
-      agentQSize = 64
+      msgQSize = 256,
+      agentQSize = 256,
+      agentSubsBatchSize = 900
     }
   where
     second = 1000000
@@ -184,9 +195,9 @@ getSMPServerClient' ca@SMPClientAgent {agentCfg, smpClients, msgQ} srv =
             _ -> TM.insert srv sVar ps
 
     serverDown :: Map SMPSub C.APrivateSignKey -> IO ()
-    serverDown ss = unless (M.null ss) . void . runExceptT $ do
+    serverDown ss = unless (M.null ss) $ do
       notify . CADisconnected srv $ M.keysSet ss
-      reconnectServer
+      void $ runExceptT reconnectServer
 
     reconnectServer :: ExceptT SMPClientError IO ()
     reconnectServer = do
@@ -201,27 +212,37 @@ getSMPServerClient' ca@SMPClientAgent {agentCfg, smpClients, msgQ} srv =
     reconnectClient :: ExceptT SMPClientError IO ()
     reconnectClient = do
       withSMP ca srv $ \smp -> do
-        notify $ CAReconnected srv
-        cs <- atomically $ mapM readTVar =<< TM.lookup srv (pendingSrvSubs ca)
-        forConcurrently_ (maybe [] M.assocs cs) $ \sub@(s, _) ->
-          unlessM (atomically $ hasSub (srvSubs ca) srv s) $
-            subscribe_ smp sub `catchE` handleError s
+        liftIO $ notify $ CAReconnected srv
+        cs_ <- atomically $ mapM readTVar =<< TM.lookup srv (pendingSrvSubs ca)
+        forM_ cs_ $ \cs -> do
+          subs' <- filterM (fmap not . atomically . hasSub (srvSubs ca) srv . fst) $ M.assocs cs
+          let (nSubs, rSubs) = partition (isNotifier . fst . fst) subs'
+          subscribe_ smp SPNotifier nSubs
+          subscribe_ smp SPRecipient rSubs
       where
-        subscribe_ :: SMPClient -> (SMPSub, C.APrivateSignKey) -> ExceptT SMPClientError IO ()
-        subscribe_ smp sub@(s, _) = do
-          smpSubscribe smp sub
-          atomically $ addSubscription ca srv sub
-          notify $ CAResubscribed srv s
+        isNotifier = \case
+          SPNotifier -> True
+          SPRecipient -> False
 
-        handleError :: SMPSub -> SMPClientError -> ExceptT SMPClientError IO ()
-        handleError s = \case
-          e@PCEResponseTimeout -> throwE e
-          e@PCENetworkError -> throwE e
-          e -> do
-            notify $ CASubError srv s e
-            atomically $ removePendingSubscription ca srv s
+        subscribe_ :: SMPClient -> SMPSubParty -> [(SMPSub, C.APrivateSignKey)] -> ExceptT SMPClientError IO ()
+        subscribe_ smp party = mapM_ subscribeBatch . toChunks (agentSubsBatchSize agentCfg)
+          where
+            subscribeBatch subs' = do
+              let subs'' :: (NonEmpty (QueueId, C.APrivateSignKey)) = L.map (first snd) subs'
+              rs <- liftIO $ smpSubscribeQueues party ca smp srv subs''
+              let rs' :: (NonEmpty ((SMPSub, C.APrivateSignKey), Either SMPClientError ())) =
+                    L.zipWith (first . const) subs' rs
+                  rs'' :: [Either (SMPSub, SMPClientError) (SMPSub, C.APrivateSignKey)] =
+                    map (\(sub, r) -> bimap (fst sub,) (const sub) r) $ L.toList rs'
+                  (errs, oks) = partitionEithers rs''
+                  (tempErrs, finalErrs) = partition (temporaryClientError . snd) errs
+              mapM_ (atomically . addSubscription ca srv) oks
+              mapM_ (liftIO . notify . CAResubscribed srv) $ L.nonEmpty $ map fst oks
+              mapM_ (atomically . removePendingSubscription ca srv . fst) finalErrs
+              mapM_ (liftIO . notify . CASubError srv) $ L.nonEmpty finalErrs
+              mapM_ (throwE . snd) $ listToMaybe tempErrs
 
-    notify :: SMPClientAgentEvent -> ExceptT SMPClientError IO ()
+    notify :: SMPClientAgentEvent -> IO ()
     notify evt = atomically $ writeTBQueue (agentQ ca) evt
 
 closeSMPClientAgent :: MonadUnliftIO m => SMPClientAgent -> m ()
@@ -252,17 +273,47 @@ withSMP ca srv action = (getSMPServerClient' ca srv >>= action) `catchE` logSMPE
 subscribeQueue :: SMPClientAgent -> SMPServer -> (SMPSub, C.APrivateSignKey) -> ExceptT SMPClientError IO ()
 subscribeQueue ca srv sub = do
   atomically $ addPendingSubscription ca srv sub
-  withSMP ca srv $ \smp -> subscribe_ smp `catchE` handleError
+  withSMP ca srv $ \smp -> subscribe_ smp `catchE` handleErr
   where
     subscribe_ smp = do
       smpSubscribe smp sub
       atomically $ addSubscription ca srv sub
 
-    handleError e = do
+    handleErr e = do
       atomically . when (e /= PCENetworkError && e /= PCEResponseTimeout) $
-        removePendingSubscription ca srv $ fst sub
+        removePendingSubscription ca srv (fst sub)
       throwE e
 
+subscribeQueuesSMP :: SMPClientAgent -> SMPServer -> NonEmpty (RecipientId, RcvPrivateSignKey) -> IO (NonEmpty (RecipientId, Either SMPClientError ()))
+subscribeQueuesSMP = subscribeQueues_ SPRecipient
+
+subscribeQueuesNtfs :: SMPClientAgent -> SMPServer -> NonEmpty (NotifierId, NtfPrivateSignKey) -> IO (NonEmpty (NotifierId, Either SMPClientError ()))
+subscribeQueuesNtfs = subscribeQueues_ SPNotifier
+
+subscribeQueues_ :: SMPSubParty -> SMPClientAgent -> SMPServer -> NonEmpty (QueueId, C.APrivateSignKey) -> IO (NonEmpty (QueueId, Either SMPClientError ()))
+subscribeQueues_ party ca srv subs = do
+  atomically $ forM_ subs $ addPendingSubscription ca srv . first (party,)
+  runExceptT (getSMPServerClient' ca srv) >>= \case
+    Left e -> pure $ L.map ((,Left e) . fst) subs
+    Right smp -> smpSubscribeQueues party ca smp srv subs
+
+smpSubscribeQueues :: SMPSubParty -> SMPClientAgent -> SMPClient -> SMPServer -> NonEmpty (QueueId, C.APrivateSignKey) -> IO (NonEmpty (QueueId, Either SMPClientError ()))
+smpSubscribeQueues party ca smp srv subs = do
+  rs <- L.zip subs <$> subscribe smp (L.map swap subs)
+  atomically $ forM rs $ \(sub, r) ->
+    (fst sub,) <$> case r of
+      Right () -> do
+        addSubscription ca srv $ first (party,) sub
+        pure $ Right ()
+      Left e -> do
+        when (e /= PCENetworkError && e /= PCEResponseTimeout) $
+          removePendingSubscription ca srv (party, fst sub)
+        pure $ Left e
+  where
+    subscribe = case party of
+      SPRecipient -> subscribeSMPQueues
+      SPNotifier -> subscribeSMPQueuesNtfs
+
 showServer :: SMPServer -> ByteString
 showServer ProtocolServer {host, port} =
   strEncode host <> B.pack (if null port then "" else ':' : port)

src/Simplex/Messaging/Crypto.hs

@@ -3,6 +3,7 @@
 {-# LANGUAGE DataKinds #-}
 {-# LANGUAGE DeriveAnyClass #-}
 {-# LANGUAGE DuplicateRecordFields #-}
+{-# LANGUAGE FlexibleContexts #-}
 {-# LANGUAGE FlexibleInstances #-}
 {-# LANGUAGE GADTs #-}
 {-# LANGUAGE LambdaCase #-}
@@ -40,6 +41,8 @@ module Simplex.Messaging.Crypto
     DhAlgorithm,
     PrivateKey (..),
     PublicKey (..),
+    PrivateKeyEd25519,
+    PublicKeyEd25519,
     PrivateKeyX25519,
     PublicKeyX25519,
     PrivateKeyX448,
@@ -58,12 +61,15 @@ module Simplex.Messaging.Crypto
     DhSecretX25519,
     ADhSecret (..),
     KeyHash (..),
+    newRandom,
+    generateAKeyPair,
     generateKeyPair,
-    generateKeyPair',
     generateSignatureKeyPair,
     generateDhKeyPair,
     privateToX509,
     publicKey,
+    signatureKeyPair,
+    publicToX509,
 
     -- * key encoding/decoding
     encodePubKey,
@@ -80,6 +86,7 @@ module Simplex.Messaging.Crypto
     SignatureAlgorithm,
     AlgorithmI (..),
     sign,
+    sign',
     verify,
     verify',
     validSignatureSize,
@@ -99,7 +106,6 @@ module Simplex.Messaging.Crypto
     decryptAESNoPad,
     authTagSize,
     randomAesKey,
-    randomIV,
     randomGCMIV,
     ivSize,
     gcmIVSize,
@@ -111,9 +117,10 @@ module Simplex.Messaging.Crypto
     cbEncrypt,
     cbEncryptMaxLenBS,
     cbDecrypt,
+    sbDecrypt_,
+    sbEncrypt_,
     cbNonce,
     randomCbNonce,
-    pseudoRandomCbNonce,
 
     -- * NaCl crypto_secretbox
     SbKey (unSbKey),
@@ -125,7 +132,7 @@ module Simplex.Messaging.Crypto
     randomSbKey,
 
     -- * pseudo-random bytes
-    pseudoRandomBytes,
+    randomBytes,
 
     -- * digests
     sha256Hash,
@@ -135,6 +142,16 @@ module Simplex.Messaging.Crypto
     pad,
     unPad,
 
+    -- * X509 Certificates
+    SignedCertificate,
+    Certificate,
+    signCertificate,
+    signX509,
+    certificateFingerprint,
+    signedFingerprint,
+    SignatureAlgorithmX509 (..),
+    SignedObject (..),
+
     -- * Cryptography error type
     CryptoError (..),
 
@@ -149,19 +166,20 @@ where
 
 import Control.Concurrent.STM
 import Control.Exception (Exception)
+import Control.Monad
 import Control.Monad.Except
 import Control.Monad.Trans.Except
 import Crypto.Cipher.AES (AES256)
 import qualified Crypto.Cipher.Types as AES
 import qualified Crypto.Cipher.XSalsa as XSalsa
 import qualified Crypto.Error as CE
-import Crypto.Hash (Digest, SHA256, SHA512, hash)
+import Crypto.Hash (Digest, SHA256 (..), SHA512, hash)
 import qualified Crypto.MAC.Poly1305 as Poly1305
 import qualified Crypto.PubKey.Curve25519 as X25519
 import qualified Crypto.PubKey.Curve448 as X448
 import qualified Crypto.PubKey.Ed25519 as Ed25519
 import qualified Crypto.PubKey.Ed448 as Ed448
-import Crypto.Random (ChaChaDRG, getRandomBytes, randomBytesGenerate)
+import Crypto.Random (ChaChaDRG, MonadPseudoRandom, drgNew, randomBytesGenerate, withDRG)
 import Data.ASN1.BinaryEncoding
 import Data.ASN1.Encoding
 import Data.ASN1.Types
@@ -182,6 +200,7 @@ import Data.Type.Equality
 import Data.Typeable (Proxy (Proxy), Typeable)
 import Data.Word (Word32)
 import Data.X509
+import Data.X509.Validation (Fingerprint (..), getFingerprint)
 import Database.SQLite.Simple.FromField (FromField (..))
 import Database.SQLite.Simple.ToField (ToField (..))
 import GHC.TypeLits (ErrorMessage (..), KnownNat, Nat, TypeError, natVal, type (+))
@@ -262,6 +281,8 @@ instance Eq APublicKey where
 
 deriving instance Show APublicKey
 
+type PublicKeyEd25519 = PublicKey Ed25519
+
 type PublicKeyX25519 = PublicKey X25519
 
 type PublicKeyX448 = PublicKey X448
@@ -277,6 +298,10 @@ deriving instance Eq (PrivateKey a)
 
 deriving instance Show (PrivateKey a)
 
+-- Do not enable, to avoid leaking key data
+-- instance StrEncoding (PrivateKey Ed25519) where
+
+-- Used in notification store log
 instance StrEncoding (PrivateKey X25519) where
   strEncode = strEncode . encodePrivKey
   {-# INLINE strEncode #-}
@@ -295,6 +320,8 @@ instance Eq APrivateKey where
 
 deriving instance Show APrivateKey
 
+type PrivateKeyEd25519 = PrivateKey Ed25519
+
 type PrivateKeyX25519 = PrivateKey X25519
 
 type PrivateKeyX448 = PrivateKey X448
@@ -537,6 +564,10 @@ publicKey = \case
   PrivateKeyX25519 _ k -> PublicKeyX25519 k
   PrivateKeyX448 _ k -> PublicKeyX448 k
 
+-- | Expand signature private key to a key pair.
+signatureKeyPair :: APrivateSignKey -> ASignatureKeyPair
+signatureKeyPair ak@(APrivateSignKey a k) = (APublicVerifyKey a (publicKey k), ak)
+
 encodePrivKey :: CryptoPrivateKey pk => pk -> ByteString
 encodePrivKey = toPrivKey $ encodeASNObj . privateToX509
 
@@ -563,17 +594,23 @@ type ASignatureKeyPair = KeyPairType APrivateSignKey
 
 type ADhKeyPair = KeyPairType APrivateDhKey
 
-generateKeyPair :: AlgorithmI a => SAlgorithm a -> IO AKeyPair
-generateKeyPair a = bimap (APublicKey a) (APrivateKey a) <$> generateKeyPair'
+newRandom :: IO (TVar ChaChaDRG)
+newRandom = newTVarIO =<< drgNew
 
-generateSignatureKeyPair :: (AlgorithmI a, SignatureAlgorithm a) => SAlgorithm a -> IO ASignatureKeyPair
-generateSignatureKeyPair a = bimap (APublicVerifyKey a) (APrivateSignKey a) <$> generateKeyPair'
+generateAKeyPair :: AlgorithmI a => SAlgorithm a -> TVar ChaChaDRG -> STM AKeyPair
+generateAKeyPair a g = bimap (APublicKey a) (APrivateKey a) <$> generateKeyPair g
 
-generateDhKeyPair :: (AlgorithmI a, DhAlgorithm a) => SAlgorithm a -> IO ADhKeyPair
-generateDhKeyPair a = bimap (APublicDhKey a) (APrivateDhKey a) <$> generateKeyPair'
+generateSignatureKeyPair :: (AlgorithmI a, SignatureAlgorithm a) => SAlgorithm a -> TVar ChaChaDRG -> STM ASignatureKeyPair
+generateSignatureKeyPair a g = bimap (APublicVerifyKey a) (APrivateSignKey a) <$> generateKeyPair g
 
-generateKeyPair' :: forall a. AlgorithmI a => IO (KeyPair a)
-generateKeyPair' = case sAlgorithm @a of
+generateDhKeyPair :: (AlgorithmI a, DhAlgorithm a) => SAlgorithm a -> TVar ChaChaDRG -> STM ADhKeyPair
+generateDhKeyPair a g = bimap (APublicDhKey a) (APrivateDhKey a) <$> generateKeyPair g
+
+generateKeyPair :: forall a. AlgorithmI a => TVar ChaChaDRG -> STM (KeyPair a)
+generateKeyPair g = stateTVar g (`withDRG` generateKeyPair_)
+
+generateKeyPair_ :: forall a. AlgorithmI a => MonadPseudoRandom ChaChaDRG (KeyPair a)
+generateKeyPair_ = case sAlgorithm @a of
   SEd25519 ->
     Ed25519.generateSecretKey >>= \pk ->
       let k = Ed25519.toPublic pk
@@ -648,13 +685,17 @@ class CryptoSignature s where
   signatureBytes :: s -> ByteString
   decodeSignature :: ByteString -> Either String s
 
+instance CryptoSignature (Signature s) => StrEncoding (Signature s) where
+  strEncode = serializeSignature
+  strDecode = decodeSignature
+
 instance CryptoSignature ASignature where
   signatureBytes (ASignature _ sig) = signatureBytes sig
   decodeSignature s
     | B.length s == Ed25519.signatureSize =
-      ASignature SEd25519 . SignatureEd25519 <$> ed Ed25519.signature s
+        ASignature SEd25519 . SignatureEd25519 <$> ed Ed25519.signature s
     | B.length s == Ed448.signatureSize =
-      ASignature SEd448 . SignatureEd448 <$> ed Ed448.signature s
+        ASignature SEd448 . SignatureEd448 <$> ed Ed448.signature s
     | otherwise = Left "bad signature size"
     where
       ed alg = first show . CE.eitherCryptoError . alg
@@ -680,6 +721,9 @@ instance SignatureSize (Signature a) where
     SignatureEd25519 _ -> Ed25519.signatureSize
     SignatureEd448 _ -> Ed448.signatureSize
 
+instance SignatureSize ASignature where
+  signatureSize (ASignature _ s) = signatureSize s
+
 instance SignatureSize APrivateSignKey where
   signatureSize (APrivateSignKey _ k) = signatureSize k
 
@@ -800,6 +844,13 @@ instance StrEncoding KeyHash where
   strEncode = strEncode . unKeyHash
   strP = KeyHash <$> strP
 
+instance ToJSON KeyHash where
+  toEncoding = strToJEncoding
+  toJSON = strToJSON
+
+instance FromJSON KeyHash where
+  parseJSON = strParseJSON "KeyHash"
+
 instance IsString KeyHash where
   fromString = parseString $ parseAll strP
 
@@ -928,15 +979,11 @@ initAEADGCM (Key aesKey) (GCMIV ivBytes) = cryptoFailable $ do
   AES.aeadInit AES.AEAD_GCM cipher ivBytes
 
 -- | Random AES256 key.
-randomAesKey :: IO Key
-randomAesKey = Key <$> getRandomBytes aesKeySize
+randomAesKey :: TVar ChaChaDRG -> STM Key
+randomAesKey = fmap Key . randomBytes aesKeySize
 
--- | Random IV bytes for AES256 encryption.
-randomIV :: IO IV
-randomIV = IV <$> getRandomBytes (ivSize @AES256)
-
-randomGCMIV :: IO GCMIV
-randomGCMIV = GCMIV <$> getRandomBytes gcmIVSize
+randomGCMIV :: TVar ChaChaDRG -> STM GCMIV
+randomGCMIV = fmap GCMIV . randomBytes gcmIVSize
 
 ivSize :: forall c. AES.BlockCipher c => Int
 ivSize = AES.blockSize (undefined :: c)
@@ -963,6 +1010,53 @@ sign' (PrivateKeyEd448 pk k) msg = SignatureEd448 $ Ed448.sign pk k msg
 sign :: APrivateSignKey -> ByteString -> ASignature
 sign (APrivateSignKey a k) = ASignature a . sign' k
 
+signCertificate :: APrivateSignKey -> Certificate -> SignedCertificate
+signCertificate = signX509
+
+signX509 :: (ASN1Object o, Eq o, Show o) => APrivateSignKey -> o -> SignedExact o
+signX509 key = fst . objectToSignedExact f
+  where
+    f bytes =
+      ( signatureBytes $ sign key bytes,
+        signatureAlgorithmX509 key,
+        ()
+      )
+
+certificateFingerprint :: SignedCertificate -> KeyHash
+certificateFingerprint = signedFingerprint
+
+signedFingerprint :: (ASN1Object o, Eq o, Show o) => SignedExact o -> KeyHash
+signedFingerprint o = KeyHash fp
+  where
+    Fingerprint fp = getFingerprint o HashSHA256
+
+class SignatureAlgorithmX509 a where
+  signatureAlgorithmX509 :: a -> SignatureALG
+
+instance SignatureAlgorithm a => SignatureAlgorithmX509 (SAlgorithm a) where
+  signatureAlgorithmX509 = \case
+    SEd25519 -> SignatureALG_IntrinsicHash PubKeyALG_Ed25519
+    SEd448 -> SignatureALG_IntrinsicHash PubKeyALG_Ed448
+
+instance SignatureAlgorithmX509 APrivateSignKey where
+  signatureAlgorithmX509 (APrivateSignKey a _) = signatureAlgorithmX509 a
+
+instance SignatureAlgorithmX509 APublicVerifyKey where
+  signatureAlgorithmX509 (APublicVerifyKey a _) = signatureAlgorithmX509 a
+
+-- | An instance for 'ASignatureKeyPair' / ('PublicKeyType' pk, pk), without touching its type family.
+instance SignatureAlgorithmX509 pk => SignatureAlgorithmX509 (a, pk) where
+  signatureAlgorithmX509 = signatureAlgorithmX509 . snd
+
+-- | A wrapper to marshall signed ASN1 objects, like certificates.
+newtype SignedObject a = SignedObject (SignedExact a)
+
+instance (Typeable a, Eq a, Show a, ASN1Object a) => FromField (SignedObject a) where
+  fromField = fmap SignedObject . blobFieldDecoder decodeSignedObject
+
+instance (Eq a, Show a, ASN1Object a) => ToField (SignedObject a) where
+  toField (SignedObject s) = toField $ encodeSignedObject s
+
 -- | Signature verification.
 --
 -- Used by SMP servers to authorize SMP commands and by SMP agents to verify messages.
@@ -1050,14 +1144,11 @@ cbNonce s
   where
     len = B.length s
 
-randomCbNonce :: IO CbNonce
-randomCbNonce = CryptoBoxNonce <$> getRandomBytes 24
+randomCbNonce :: TVar ChaChaDRG -> STM CbNonce
+randomCbNonce = fmap CryptoBoxNonce . randomBytes 24
 
-pseudoRandomCbNonce :: TVar ChaChaDRG -> STM CbNonce
-pseudoRandomCbNonce gVar = CryptoBoxNonce <$> pseudoRandomBytes 24 gVar
-
-pseudoRandomBytes :: Int -> TVar ChaChaDRG -> STM ByteString
-pseudoRandomBytes n gVar = stateTVar gVar $ randomBytesGenerate n
+randomBytes :: Int -> TVar ChaChaDRG -> STM ByteString
+randomBytes n gVar = stateTVar gVar $ randomBytesGenerate n
 
 instance Encoding CbNonce where
   smpEncode = unCbNonce
@@ -1094,8 +1185,8 @@ sbKey s
 unsafeSbKey :: ByteString -> SbKey
 unsafeSbKey s = either error id $ sbKey s
 
-randomSbKey :: IO SbKey
-randomSbKey = SecretBoxKey <$> getRandomBytes 32
+randomSbKey :: TVar ChaChaDRG -> STM SbKey
+randomSbKey gVar = SecretBoxKey <$> randomBytes 32 gVar
 
 xSalsa20 :: ByteArrayAccess key => key -> ByteString -> ByteString -> (ByteString, ByteString)
 xSalsa20 secret nonce msg = (rs, msg')

src/Simplex/Messaging/Crypto/File.hs

@@ -0,0 +1,123 @@
+{-# LANGUAGE DeriveAnyClass #-}
+{-# LANGUAGE ScopedTypeVariables #-}
+{-# LANGUAGE TemplateHaskell #-}
+
+module Simplex.Messaging.Crypto.File
+  ( CryptoFile (..),
+    CryptoFileArgs (..),
+    CryptoFileHandle (..),
+    FTCryptoError (..),
+    Simplex.Messaging.Crypto.File.readFile,
+    Simplex.Messaging.Crypto.File.writeFile,
+    withFile,
+    hPut,
+    hPutTag,
+    hGet,
+    hGetTag,
+    plain,
+    randomArgs,
+    getFileContentsSize,
+  )
+where
+
+import Control.Exception
+import Control.Monad
+import Control.Monad.Except
+import Crypto.Random (ChaChaDRG)
+import qualified Data.Aeson.TH as J
+import qualified Data.ByteArray as BA
+import Data.ByteString.Char8 (ByteString)
+import qualified Data.ByteString.Char8 as B
+import qualified Data.ByteString.Lazy as LB
+import Data.List.NonEmpty (NonEmpty (..))
+import Data.Maybe (isJust)
+import Simplex.Messaging.Client.Agent ()
+import qualified Simplex.Messaging.Crypto as C
+import Simplex.Messaging.Crypto.Lazy (LazyByteString)
+import qualified Simplex.Messaging.Crypto.Lazy as LC
+import Simplex.Messaging.Parsers (defaultJSON)
+import Simplex.Messaging.Util (liftEitherWith)
+import System.Directory (getFileSize)
+import UnliftIO (Handle, IOMode (..), liftIO)
+import qualified UnliftIO as IO
+import UnliftIO.STM
+
+-- Possibly encrypted local file
+data CryptoFile = CryptoFile {filePath :: FilePath, cryptoArgs :: Maybe CryptoFileArgs}
+  deriving (Eq, Show)
+
+data CryptoFileArgs = CFArgs {fileKey :: C.SbKey, fileNonce :: C.CbNonce}
+  deriving (Eq, Show)
+
+data CryptoFileHandle = CFHandle Handle (Maybe (TVar LC.SbState))
+
+readFile :: CryptoFile -> ExceptT FTCryptoError IO LazyByteString
+readFile (CryptoFile path cfArgs) = do
+  s <- liftIO $ LB.readFile path
+  case cfArgs of
+    Just (CFArgs (C.SbKey key) (C.CbNonce nonce)) -> do
+      let len = LB.length s - fromIntegral C.authTagSize
+      when (len < 0) $ throwError FTCEInvalidFileSize
+      let (s', tag') = LB.splitAt len s
+      (tag :| cs) <- liftEitherWith FTCECryptoError $ LC.secretBox LC.sbDecryptChunk key nonce s'
+      unless (BA.constEq (LB.toStrict tag') tag) $ throwError FTCEInvalidAuthTag
+      pure $ LB.fromChunks cs
+    Nothing -> pure s
+
+writeFile :: CryptoFile -> LazyByteString -> ExceptT FTCryptoError IO ()
+writeFile (CryptoFile path cfArgs) s = do
+  s' <- case cfArgs of
+    Just (CFArgs (C.SbKey key) (C.CbNonce nonce)) ->
+      liftEitherWith FTCECryptoError $ LB.fromChunks <$> LC.secretBoxTailTag LC.sbEncryptChunk key nonce s
+    Nothing -> pure s
+  liftIO $ LB.writeFile path s'
+
+withFile :: CryptoFile -> IOMode -> (CryptoFileHandle -> ExceptT FTCryptoError IO a) -> ExceptT FTCryptoError IO a
+withFile (CryptoFile path cfArgs) mode action = do
+  sb <- forM cfArgs $ \(CFArgs key nonce) ->
+    liftEitherWith FTCECryptoError (LC.sbInit key nonce) >>= newTVarIO
+  IO.withFile path mode $ \h -> action $ CFHandle h sb
+
+hPut :: CryptoFileHandle -> LazyByteString -> IO ()
+hPut (CFHandle h sb_) s = LB.hPut h =<< maybe (pure s) encrypt sb_
+  where
+    encrypt sb = atomically $ stateTVar sb (`LC.sbEncryptChunkLazy` s)
+
+hPutTag :: CryptoFileHandle -> IO ()
+hPutTag (CFHandle h sb_) = forM_ sb_ $ B.hPut h . BA.convert . LC.sbAuth <=< readTVarIO
+
+hGet :: CryptoFileHandle -> Int -> IO ByteString
+hGet (CFHandle h sb_) n = B.hGet h n >>= maybe pure decrypt sb_
+  where
+    decrypt sb s = atomically $ stateTVar sb (`LC.sbDecryptChunk` s)
+
+-- | Read and validate the auth tag.
+-- This function should be called after reading the whole file, it assumes you know the file size and read only the needed bytes.
+hGetTag :: CryptoFileHandle -> ExceptT FTCryptoError IO ()
+hGetTag (CFHandle h sb_) = forM_ sb_ $ \sb -> do
+  tag <- liftIO $ B.hGet h C.authTagSize
+  tag' <- LC.sbAuth <$> readTVarIO sb
+  unless (BA.constEq tag tag') $ throwError FTCEInvalidAuthTag
+
+data FTCryptoError
+  = FTCECryptoError C.CryptoError
+  | FTCEInvalidHeader String
+  | FTCEInvalidFileSize
+  | FTCEInvalidAuthTag
+  | FTCEFileIOError String
+  deriving (Show, Eq, Exception)
+
+plain :: FilePath -> CryptoFile
+plain = (`CryptoFile` Nothing)
+
+randomArgs :: TVar ChaChaDRG -> STM CryptoFileArgs
+randomArgs g = CFArgs <$> C.randomSbKey g <*> C.randomCbNonce g
+
+getFileContentsSize :: CryptoFile -> IO Integer
+getFileContentsSize (CryptoFile path cfArgs) = do
+  size <- getFileSize path
+  pure $ if isJust cfArgs then size - fromIntegral C.authTagSize else size
+
+$(J.deriveJSON defaultJSON ''CryptoFileArgs)
+
+$(J.deriveJSON defaultJSON ''CryptoFile)

src/Simplex/Messaging/Crypto/Lazy.hs

@@ -1,7 +1,6 @@
 {-# LANGUAGE BangPatterns #-}
 {-# LANGUAGE GADTs #-}
 {-# LANGUAGE LambdaCase #-}
-{-# LANGUAGE OverloadedStrings #-}
 {-# LANGUAGE PatternSynonyms #-}
 {-# LANGUAGE ScopedTypeVariables #-}
 {-# LANGUAGE TupleSections #-}
@@ -15,11 +14,16 @@ module Simplex.Messaging.Crypto.Lazy
     sbEncrypt,
     sbDecrypt,
     sbEncryptTailTag,
+    kcbEncryptTailTag,
     sbDecryptTailTag,
+    kcbDecryptTailTag,
     fastReplicate,
+    secretBox,
+    secretBoxTailTag,
     SbState,
     cbInit,
     sbInit,
+    kcbInit,
     sbEncryptChunk,
     sbDecryptChunk,
     sbEncryptChunkLazy,
@@ -37,7 +41,6 @@ import qualified Crypto.MAC.Poly1305 as Poly1305
 import Data.Bifunctor (first)
 import Data.ByteArray (ByteArrayAccess)
 import qualified Data.ByteArray as BA
-import qualified Data.ByteString as S
 import Data.ByteString.Char8 (ByteString)
 import qualified Data.ByteString.Char8 as B
 import qualified Data.ByteString.Lazy.Char8 as LB
@@ -47,6 +50,7 @@ import Data.Int (Int64)
 import Data.List.NonEmpty (NonEmpty (..))
 import Foreign (sizeOf)
 import Simplex.Messaging.Crypto (CbNonce, CryptoError (..), DhSecret (..), DhSecretX25519, SbKey, pattern CbNonce, pattern SbKey)
+import Simplex.Messaging.Crypto.SNTRUP761 (KEMHybridSecret (..))
 import Simplex.Messaging.Encoding
 
 type LazyByteString = LB.ByteString
@@ -88,10 +92,10 @@ unPad = fmap snd . splitLen
 splitLen :: LazyByteString -> Either CryptoError (Int64, LazyByteString)
 splitLen padded
   | LB.length lenStr == 8 = case smpDecode $ LB.toStrict lenStr of
-    Right len
-      | len < 0 -> Left CryptoInvalidMsgError
-      | otherwise -> Right (len, LB.take len rest)
-    Left _ -> Left CryptoInvalidMsgError
+      Right len
+        | len < 0 -> Left CryptoInvalidMsgError
+        | otherwise -> Right (len, LB.take len rest)
+      Left _ -> Left CryptoInvalidMsgError
   | otherwise = Left CryptoInvalidMsgError
   where
     (lenStr, rest) = LB.splitAt 8 padded
@@ -110,10 +114,10 @@ sbDecrypt :: SbKey -> CbNonce -> LazyByteString -> Either CryptoError LazyByteSt
 sbDecrypt (SbKey key) (CbNonce nonce) packet
   | LB.length tag' < 16 = Left CBDecryptError
   | otherwise = case secretBox sbDecryptChunk key nonce c of
-    Right (tag :| cs)
-      | BA.constEq (LB.toStrict tag') tag -> unPad $ LB.fromChunks cs
-      | otherwise -> Left CBDecryptError
-    Left e -> Left e
+      Right (tag :| cs)
+        | BA.constEq (LB.toStrict tag') tag -> unPad $ LB.fromChunks cs
+        | otherwise -> Left CBDecryptError
+      Left e -> Left e
   where
     (tag', c) = LB.splitAt 16 packet
 
@@ -126,13 +130,33 @@ secretBox sbProcess secret nonce msg = run <$> sbInit_ secret nonce
 
 -- | NaCl @secret_box@ lazy encrypt with a symmetric 256-bit key and 192-bit nonce with appended auth tag (more efficient with large files).
 sbEncryptTailTag :: SbKey -> CbNonce -> LazyByteString -> Int64 -> Int64 -> Either CryptoError LazyByteString
-sbEncryptTailTag (SbKey key) (CbNonce nonce) msg len paddedLen =
+sbEncryptTailTag (SbKey key) = sbEncryptTailTag_ key
+{-# INLINE sbEncryptTailTag #-}
+
+-- | NaCl @crypto_box@ lazy encrypt with with a shared hybrid KEM+DH 256-bit secret and 192-bit nonce with appended auth tag (more efficient with large strings/files).
+kcbEncryptTailTag :: KEMHybridSecret -> CbNonce -> LazyByteString -> Int64 -> Int64 -> Either CryptoError LazyByteString
+kcbEncryptTailTag (KEMHybridSecret key) = sbEncryptTailTag_ key
+{-# INLINE kcbEncryptTailTag #-}
+
+sbEncryptTailTag_ :: ByteArrayAccess key => key -> CbNonce -> LazyByteString -> Int64 -> Int64 -> Either CryptoError LazyByteString
+sbEncryptTailTag_ key (CbNonce nonce) msg len paddedLen =
   LB.fromChunks <$> (secretBoxTailTag sbEncryptChunk key nonce =<< pad msg len paddedLen)
 
 -- | NaCl @secret_box@ decrypt with a symmetric 256-bit key and 192-bit nonce with appended auth tag (more efficient with large files).
 -- paddedLen should NOT include the tag length, it should be the same number that is passed to sbEncrypt / sbEncryptTailTag.
 sbDecryptTailTag :: SbKey -> CbNonce -> Int64 -> LazyByteString -> Either CryptoError (Bool, LazyByteString)
-sbDecryptTailTag (SbKey key) (CbNonce nonce) paddedLen packet =
+sbDecryptTailTag (SbKey key) = sbDecryptTailTag_ key
+{-# INLINE sbDecryptTailTag #-}
+
+-- | NaCl @crypto_box@ lazy decrypt with a shared hybrid KEM+DH 256-bit secret and 192-bit nonce with appended auth tag (more efficient with large strings/files).
+-- paddedLen should NOT include the tag length, it should be the same number that is passed to sbEncrypt / sbEncryptTailTag.
+kcbDecryptTailTag :: KEMHybridSecret -> CbNonce -> Int64 -> LazyByteString -> Either CryptoError (Bool, LazyByteString)
+kcbDecryptTailTag (KEMHybridSecret key) = sbDecryptTailTag_ key
+{-# INLINE kcbDecryptTailTag #-}
+
+-- paddedLen should NOT include the tag length, it should be the same number that is passed to sbEncrypt / sbEncryptTailTag.
+sbDecryptTailTag_ :: ByteArrayAccess key => key -> CbNonce -> Int64 -> LazyByteString -> Either CryptoError (Bool, LazyByteString)
+sbDecryptTailTag_ key (CbNonce nonce) paddedLen packet =
   case secretBox sbDecryptChunk key nonce c of
     Right (tag :| cs) ->
       let valid = LB.length tag' == 16 && BA.constEq (LB.toStrict tag') tag
@@ -150,7 +174,7 @@ secretBoxTailTag sbProcess secret nonce msg = run <$> sbInit_ secret nonce
 
 -- passes lazy bytestring via initialized secret box returning the reversed list of chunks
 secretBoxLazy_ :: (SbState -> ByteString -> (ByteString, SbState)) -> SbState -> LazyByteString -> ([ByteString], SbState)
-secretBoxLazy_ sbProcess state = foldlChunks update ([], state)
+secretBoxLazy_ sbProcess state = LB.foldlChunks update ([], state)
   where
     update (cs, st) chunk = let (!c, !st') = sbProcess st chunk in (c : cs, st')
 
@@ -164,6 +188,10 @@ sbInit :: SbKey -> CbNonce -> Either CryptoError SbState
 sbInit (SbKey secret) (CbNonce nonce) = sbInit_ secret nonce
 {-# INLINE sbInit #-}
 
+kcbInit :: KEMHybridSecret -> CbNonce -> Either CryptoError SbState
+kcbInit (KEMHybridSecret k) (CbNonce nonce) = sbInit_ k nonce
+{-# INLINE kcbInit #-}
+
 sbInit_ :: ByteArrayAccess key => key -> ByteString -> Either CryptoError SbState
 sbInit_ secret nonce = (state2,) <$> cryptoPassed (Poly1305.initialize rs)
   where
@@ -202,10 +230,3 @@ cryptoPassed :: CE.CryptoFailable b -> Either CryptoError b
 cryptoPassed = \case
   CE.CryptoPassed a -> Right a
   CE.CryptoFailed e -> Left $ CryptoPoly1305Error e
-
-foldlChunks :: (a -> S.ByteString -> a) -> a -> LazyByteString -> a
-foldlChunks f = go
-  where
-    go !a LB.Empty = a
-    go !a (LB.Chunk c cs) = go (f a c) cs
-{-# INLINE foldlChunks #-}

src/Simplex/Messaging/Crypto/Ratchet.hs

@@ -1,12 +1,12 @@
 {-# LANGUAGE DataKinds #-}
 {-# LANGUAGE DeriveAnyClass #-}
-{-# LANGUAGE DeriveGeneric #-}
 {-# LANGUAGE DuplicateRecordFields #-}
 {-# LANGUAGE GADTs #-}
 {-# LANGUAGE LambdaCase #-}
 {-# LANGUAGE NamedFieldPuns #-}
 {-# LANGUAGE OverloadedStrings #-}
 {-# LANGUAGE ScopedTypeVariables #-}
+{-# LANGUAGE TemplateHaskell #-}
 {-# LANGUAGE TupleSections #-}
 {-# LANGUAGE TypeApplications #-}
 {-# LANGUAGE TypeFamilies #-}
@@ -19,8 +19,10 @@ import Control.Monad.Trans.Except
 import Crypto.Cipher.AES (AES256)
 import Crypto.Hash (SHA512)
 import qualified Crypto.KDF.HKDF as H
-import Data.Aeson (FromJSON, ToJSON)
+import Crypto.Random (ChaChaDRG)
+import Data.Aeson (FromJSON (..), ToJSON (..))
 import qualified Data.Aeson as J
+import qualified Data.Aeson.TH as JQ
 import Data.ByteString.Char8 (ByteString)
 import qualified Data.ByteString.Char8 as B
 import qualified Data.ByteString.Lazy as LB
@@ -32,14 +34,13 @@ import Data.Typeable (Typeable)
 import Data.Word (Word32)
 import Database.SQLite.Simple.FromField (FromField (..))
 import Database.SQLite.Simple.ToField (ToField (..))
-import GHC.Generics
 import Simplex.Messaging.Agent.QueryString
 import Simplex.Messaging.Crypto
 import Simplex.Messaging.Encoding
 import Simplex.Messaging.Encoding.String
-import Simplex.Messaging.Parsers (blobFieldDecoder, parseE, parseE')
-import Simplex.Messaging.Util (tryE)
+import Simplex.Messaging.Parsers (blobFieldDecoder, defaultJSON, parseE, parseE')
 import Simplex.Messaging.Version
+import UnliftIO.STM
 
 currentE2EEncryptVersion :: Version
 currentE2EEncryptVersion = 2
@@ -81,10 +82,10 @@ instance AlgorithmI a => StrEncoding (E2ERatchetParamsUri a) where
       [key1, key2] -> pure $ E2ERatchetParamsUri vs key1 key2
       _ -> fail "bad e2e params"
 
-generateE2EParams :: (AlgorithmI a, DhAlgorithm a) => Version -> IO (PrivateKey a, PrivateKey a, E2ERatchetParams a)
-generateE2EParams v = do
-  (k1, pk1) <- generateKeyPair'
-  (k2, pk2) <- generateKeyPair'
+generateE2EParams :: (AlgorithmI a, DhAlgorithm a) => TVar ChaChaDRG -> Version -> STM (PrivateKey a, PrivateKey a, E2ERatchetParams a)
+generateE2EParams g v = do
+  (k1, pk1) <- generateKeyPair g
+  (k2, pk2) <- generateKeyPair g
   pure (pk1, pk2, E2ERatchetParams v k1 k2)
 
 data RatchetInitParams = RatchetInitParams
@@ -112,11 +113,11 @@ x3dh v (sk1, rk1) dh1 dh2 dh3 =
     (hk, nhk, sk)
       -- for backwards compatibility with clients using agent version before 3.4.0
       | v == 1 =
-        let (hk', rest) = B.splitAt 32 dhs
-         in uncurry (hk',,) $ B.splitAt 32 rest
+          let (hk', rest) = B.splitAt 32 dhs
+           in uncurry (hk',,) $ B.splitAt 32 rest
       | otherwise =
-        let salt = B.replicate 64 '\0'
-         in hkdf3 salt dhs "SimpleXX3DH"
+          let salt = B.replicate 64 '\0'
+           in hkdf3 salt dhs "SimpleXX3DH"
 
 type RatchetX448 = Ratchet 'X448
 
@@ -135,29 +136,20 @@ data Ratchet a = Ratchet
     rcNHKs :: HeaderKey,
     rcNHKr :: HeaderKey
   }
-  deriving (Eq, Show, Generic, FromJSON)
-
-instance AlgorithmI a => ToJSON (Ratchet a) where
-  toEncoding = J.genericToEncoding J.defaultOptions
+  deriving (Eq, Show)
 
 data SndRatchet a = SndRatchet
   { rcDHRr :: PublicKey a,
     rcCKs :: RatchetKey,
     rcHKs :: HeaderKey
   }
-  deriving (Eq, Show, Generic, FromJSON)
-
-instance AlgorithmI a => ToJSON (SndRatchet a) where
-  toEncoding = J.genericToEncoding J.defaultOptions
+  deriving (Eq, Show)
 
 data RcvRatchet = RcvRatchet
   { rcCKr :: RatchetKey,
     rcHKr :: HeaderKey
   }
-  deriving (Eq, Show, Generic, FromJSON)
-
-instance ToJSON RcvRatchet where
-  toEncoding = J.genericToEncoding J.defaultOptions
+  deriving (Eq, Show)
 
 type SkippedMsgKeys = Map HeaderKey SkippedHdrMsgKeys
 
@@ -204,10 +196,6 @@ instance ToJSON RatchetKey where
 instance FromJSON RatchetKey where
   parseJSON = fmap RatchetKey . strParseJSON "Key"
 
-instance AlgorithmI a => ToField (Ratchet a) where toField = toField . LB.toStrict . J.encode
-
-instance (AlgorithmI a, Typeable a) => FromField (Ratchet a) where fromField = blobFieldDecoder J.eitherDecodeStrict'
-
 instance ToField MessageKey where toField = toField . smpEncode
 
 instance FromField MessageKey where fromField = blobFieldDecoder smpDecode
@@ -358,11 +346,12 @@ maxSkip = 512
 rcDecrypt ::
   forall a.
   (AlgorithmI a, DhAlgorithm a) =>
+  TVar ChaChaDRG ->
   Ratchet a ->
   SkippedMsgKeys ->
   ByteString ->
   ExceptT CryptoError IO (DecryptResult a)
-rcDecrypt rc@Ratchet {rcRcv, rcAD = Str rcAD} rcMKSkipped msg' = do
+rcDecrypt g rc@Ratchet {rcRcv, rcAD = Str rcAD} rcMKSkipped msg' = do
   encMsg@EncRatchetMessage {emHeader} <- parseE CryptoHeaderError smpP msg'
   encHdr <- parseE CryptoHeaderError smpP emHeader
   -- plaintext = TrySkippedMessageKeysHE(state, enc_header, cipher-text, AD)
@@ -402,7 +391,7 @@ rcDecrypt rc@Ratchet {rcRcv, rcAD = Str rcAD} rcMKSkipped msg' = do
             Left e -> throwE e
             Right (rc'@Ratchet {rcDHRs, rcRK, rcNHKs, rcNHKr}, hmks) -> do
               -- DHRatchetHE(state, header)
-              (_, rcDHRs') <- liftIO $ generateKeyPair' @a
+              (_, rcDHRs') <- atomically $ generateKeyPair @a g
               -- state.RK, state.CKr, state.NHKr = KDF_RK_HE(state.RK, DH(state.DHRs, state.DHRr))
               let (rcRK', rcCKr', rcNHKr') = rootKdf rcRK msgDHRs rcDHRs
                   -- state.RK, state.CKs, state.NHKs = KDF_RK_HE(state.RK, DH(state.DHRs, state.DHRr))
@@ -428,9 +417,9 @@ rcDecrypt rc@Ratchet {rcRcv, rcAD = Str rcAD} rcMKSkipped msg' = do
       | rcNr + maxSkip < untilN = Left $ CERatchetTooManySkipped (untilN + 1 - rcNr)
       | rcNr == untilN = Right (r, M.empty)
       | otherwise =
-        let (rcCKr', rcNr', mks) = advanceRcvRatchet (untilN - rcNr) rcCKr rcNr M.empty
-            r' = r {rcRcv = Just rr {rcCKr = rcCKr'}, rcNr = rcNr'}
-         in Right (r', M.singleton rcHKr mks)
+          let (rcCKr', rcNr', mks) = advanceRcvRatchet (untilN - rcNr) rcCKr rcNr M.empty
+              r' = r {rcRcv = Just rr {rcCKr = rcCKr'}, rcNr = rcNr'}
+           in Right (r', M.singleton rcHKr mks)
     advanceRcvRatchet :: Word32 -> RatchetKey -> Word32 -> SkippedHdrMsgKeys -> (RatchetKey, Word32, SkippedHdrMsgKeys)
     advanceRcvRatchet 0 ck msgNs mks = (ck, msgNs, mks)
     advanceRcvRatchet n ck msgNs mks =
@@ -493,3 +482,23 @@ hkdf3 salt ikm info = (s1, s2, s3)
     out = H.expand prk info 96
     (s1, rest) = B.splitAt 32 out
     (s2, s3) = B.splitAt 32 rest
+
+$(JQ.deriveJSON defaultJSON ''RcvRatchet)
+
+instance AlgorithmI a => ToJSON (SndRatchet a) where
+  toEncoding = $(JQ.mkToEncoding defaultJSON ''SndRatchet)
+  toJSON = $(JQ.mkToJSON defaultJSON ''SndRatchet)
+
+instance AlgorithmI a => FromJSON (SndRatchet a) where
+  parseJSON = $(JQ.mkParseJSON defaultJSON ''SndRatchet)
+
+instance AlgorithmI a => ToJSON (Ratchet a) where
+  toEncoding = $(JQ.mkToEncoding defaultJSON ''Ratchet)
+  toJSON = $(JQ.mkToJSON defaultJSON ''Ratchet)
+
+instance AlgorithmI a => FromJSON (Ratchet a) where
+  parseJSON = $(JQ.mkParseJSON defaultJSON ''Ratchet)
+
+instance AlgorithmI a => ToField (Ratchet a) where toField = toField . LB.toStrict . J.encode
+
+instance (AlgorithmI a, Typeable a) => FromField (Ratchet a) where fromField = blobFieldDecoder J.eitherDecodeStrict'

src/Simplex/Messaging/Crypto/SNTRUP761.hs

@@ -0,0 +1,31 @@
+{-# LANGUAGE DataKinds #-}
+{-# LANGUAGE GADTs #-}
+{-# LANGUAGE LambdaCase #-}
+
+module Simplex.Messaging.Crypto.SNTRUP761 where
+
+import Crypto.Hash (Digest, SHA256, hash)
+import Data.ByteArray (ScrubbedBytes)
+import qualified Data.ByteArray as BA
+import Data.ByteString (ByteString)
+import Simplex.Messaging.Crypto
+import qualified Simplex.Messaging.Crypto as C
+import Simplex.Messaging.Crypto.SNTRUP761.Bindings
+
+-- Hybrid shared secret for crypto_box is defined as SHA256(DHSecret || KEMSharedKey),
+-- similar to https://datatracker.ietf.org/doc/draft-josefsson-ntruprime-hybrid/
+
+newtype KEMHybridSecret = KEMHybridSecret ScrubbedBytes
+
+-- | NaCl @crypto_box@ decrypt with a shared hybrid DH + KEM secret and 192-bit nonce.
+kcbDecrypt :: KEMHybridSecret -> CbNonce -> ByteString -> Either CryptoError ByteString
+kcbDecrypt (KEMHybridSecret k) = sbDecrypt_ k
+
+-- | NaCl @crypto_box@ encrypt with a shared hybrid DH + KEM secret and 192-bit nonce.
+kcbEncrypt :: KEMHybridSecret -> CbNonce -> ByteString -> Int -> Either CryptoError ByteString
+kcbEncrypt (KEMHybridSecret k) = sbEncrypt_ k
+
+kemHybridSecret :: PublicKeyX25519 -> PrivateKeyX25519 -> KEMSharedKey -> KEMHybridSecret
+kemHybridSecret k pk (KEMSharedKey kem) =
+  let DhSecretX25519 dh = C.dh' k pk
+   in KEMHybridSecret $ BA.convert (hash $ BA.convert dh <> kem :: Digest SHA256)

src/Simplex/Messaging/Crypto/SNTRUP761/Bindings.hs

@@ -0,0 +1,82 @@
+{-# LANGUAGE TypeApplications #-}
+
+module Simplex.Messaging.Crypto.SNTRUP761.Bindings where
+
+import Control.Concurrent.STM
+import Crypto.Random (ChaChaDRG)
+import Data.Aeson (FromJSON (..), ToJSON (..))
+import Data.Bifunctor (bimap)
+import Data.ByteArray (ScrubbedBytes)
+import qualified Data.ByteArray as BA
+import Data.ByteString (ByteString)
+import Database.SQLite.Simple.FromField
+import Database.SQLite.Simple.ToField
+import Foreign (nullPtr)
+import Simplex.Messaging.Crypto.SNTRUP761.Bindings.Defines
+import Simplex.Messaging.Crypto.SNTRUP761.Bindings.FFI
+import Simplex.Messaging.Crypto.SNTRUP761.Bindings.RNG (withDRG)
+import Simplex.Messaging.Encoding
+import Simplex.Messaging.Encoding.String
+
+newtype KEMPublicKey = KEMPublicKey ByteString
+  deriving (Show)
+
+newtype KEMSecretKey = KEMSecretKey ScrubbedBytes
+  deriving (Show)
+
+newtype KEMCiphertext = KEMCiphertext ByteString
+  deriving (Show)
+
+newtype KEMSharedKey = KEMSharedKey ScrubbedBytes
+  deriving (Show)
+
+type KEMKeyPair = (KEMPublicKey, KEMSecretKey)
+
+sntrup761Keypair :: TVar ChaChaDRG -> IO KEMKeyPair
+sntrup761Keypair drg =
+  bimap KEMPublicKey KEMSecretKey
+    <$> BA.allocRet
+      c_SNTRUP761_SECRETKEY_SIZE
+      ( \skPtr ->
+          BA.alloc c_SNTRUP761_PUBLICKEY_SIZE $ \pkPtr ->
+            withDRG drg $ c_sntrup761_keypair pkPtr skPtr nullPtr
+      )
+
+sntrup761Enc :: TVar ChaChaDRG -> KEMPublicKey -> IO (KEMCiphertext, KEMSharedKey)
+sntrup761Enc drg (KEMPublicKey pk) =
+  BA.withByteArray pk $ \pkPtr ->
+    bimap KEMCiphertext KEMSharedKey
+      <$> BA.allocRet
+        c_SNTRUP761_SIZE
+        ( \kPtr ->
+            BA.alloc c_SNTRUP761_CIPHERTEXT_SIZE $ \cPtr ->
+              withDRG drg $ c_sntrup761_enc cPtr kPtr pkPtr nullPtr
+        )
+
+sntrup761Dec :: KEMCiphertext -> KEMSecretKey -> IO KEMSharedKey
+sntrup761Dec (KEMCiphertext c) (KEMSecretKey sk) =
+  BA.withByteArray sk $ \skPtr ->
+    BA.withByteArray c $ \cPtr ->
+      KEMSharedKey
+        <$> BA.alloc c_SNTRUP761_SIZE (\kPtr -> c_sntrup761_dec kPtr cPtr skPtr)
+
+instance StrEncoding KEMPublicKey where
+  strEncode (KEMPublicKey pk) = strEncode (BA.convert pk :: ByteString)
+  strP = KEMPublicKey . BA.convert <$> strP @ByteString
+
+instance Encoding KEMCiphertext where
+  smpEncode (KEMCiphertext c) = smpEncode . Large $ BA.convert c
+  smpP = KEMCiphertext . BA.convert . unLarge <$> smpP
+
+instance ToJSON KEMPublicKey where
+  toJSON = strToJSON
+  toEncoding = strToJEncoding
+
+instance FromJSON KEMPublicKey where
+  parseJSON = strParseJSON "KEMPublicKey"
+
+instance ToField KEMSharedKey where
+  toField (KEMSharedKey k) = toField (BA.convert k :: ByteString)
+
+instance FromField KEMSharedKey where
+  fromField f = KEMSharedKey . BA.convert @ByteString <$> fromField f

src/Simplex/Messaging/Crypto/SNTRUP761/Bindings/Defines.hsc

@@ -0,0 +1,15 @@
+module Simplex.Messaging.Crypto.SNTRUP761.Bindings.Defines where
+
+#include "sntrup761.h"
+
+c_SNTRUP761_SECRETKEY_SIZE :: Int
+c_SNTRUP761_SECRETKEY_SIZE = #{const SNTRUP761_SECRETKEY_SIZE}
+
+c_SNTRUP761_PUBLICKEY_SIZE :: Int
+c_SNTRUP761_PUBLICKEY_SIZE = #{const SNTRUP761_PUBLICKEY_SIZE}
+
+c_SNTRUP761_CIPHERTEXT_SIZE :: Int
+c_SNTRUP761_CIPHERTEXT_SIZE = #{const SNTRUP761_CIPHERTEXT_SIZE}
+
+c_SNTRUP761_SIZE :: Int
+c_SNTRUP761_SIZE = #{const SNTRUP761_SIZE}

src/Simplex/Messaging/Crypto/SNTRUP761/Bindings/FFI.hs

@@ -0,0 +1,23 @@
+{-# LANGUAGE ForeignFunctionInterface #-}
+
+module Simplex.Messaging.Crypto.SNTRUP761.Bindings.FFI
+  ( c_sntrup761_keypair,
+    c_sntrup761_enc,
+    c_sntrup761_dec,
+  ) where
+
+import Foreign
+import Foreign.C
+import Simplex.Messaging.Crypto.SNTRUP761.Bindings.RNG (RNGContext, RNGFunc)
+
+-- void sntrup761_keypair (uint8_t *pk, uint8_t *sk, void *random_ctx, sntrup761_random_func *random);
+foreign import ccall "sntrup761_keypair"
+  c_sntrup761_keypair :: Ptr Word8 -> Ptr Word8 -> Ptr RNGContext -> FunPtr RNGFunc -> IO ()
+
+-- void sntrup761_enc (uint8_t *c, uint8_t *k, const uint8_t *pk, void *random_ctx, sntrup761_random_func *random);
+foreign import ccall "sntrup761_enc"
+  c_sntrup761_enc :: Ptr Word8 -> Ptr Word8 -> Ptr Word8 -> Ptr RNGContext -> FunPtr RNGFunc -> IO ()
+
+-- void sntrup761_dec (uint8_t *k, const uint8_t *c, const uint8_t *sk);
+foreign import ccall "sntrup761_dec"
+  c_sntrup761_dec :: Ptr Word8 -> Ptr Word8 -> Ptr Word8 -> IO ()

src/Simplex/Messaging/Crypto/SNTRUP761/Bindings/RNG.hs

@@ -0,0 +1,30 @@
+module Simplex.Messaging.Crypto.SNTRUP761.Bindings.RNG
+  ( withDRG,
+    RNGContext,
+    RNGFunc,
+  ) where
+
+import Control.Concurrent.STM
+import Control.Exception (bracket)
+import Crypto.Random (ChaChaDRG)
+import Data.ByteArray (ByteArrayAccess (copyByteArrayToPtr))
+import Foreign
+import Foreign.C
+import qualified Simplex.Messaging.Crypto as C
+
+withDRG :: TVar ChaChaDRG -> (FunPtr RNGFunc -> IO a) -> IO a
+withDRG drg = bracket (createRNGFunc drg) freeHaskellFunPtr
+
+createRNGFunc :: TVar ChaChaDRG -> IO (FunPtr RNGFunc)
+createRNGFunc drg =
+  mkRNGFunc $ \_ctx sz buf -> do
+    bs <- atomically $ C.randomBytes (fromIntegral sz) drg
+    copyByteArrayToPtr bs buf
+
+type RNGContext = ()
+
+-- typedef void random_func (void *ctx, size_t length, uint8_t *dst);
+type RNGFunc = Ptr RNGContext -> CSize -> Ptr Word8 -> IO ()
+
+foreign import ccall "wrapper"
+  mkRNGFunc :: RNGFunc -> IO (FunPtr RNGFunc)

src/Simplex/Messaging/Encoding.hs

@@ -11,6 +11,7 @@ module Simplex.Messaging.Encoding
   ( Encoding (..),
     Tail (..),
     Large (..),
+    _smpP,
     smpEncodeList,
     smpListP,
     lenEncode,
@@ -109,7 +110,7 @@ lenP = fromIntegral . c2w <$> A.anyChar
 {-# INLINE lenP #-}
 
 instance Encoding a => Encoding (Maybe a) where
-  smpEncode s = maybe "0" (("1" <>) . smpEncode) s
+  smpEncode = maybe "0" (("1" <>) . smpEncode)
   {-# INLINE smpEncode #-}
   smpP =
     smpP >>= \case
@@ -143,6 +144,9 @@ instance Encoding SystemTime where
   smpP = MkSystemTime <$> smpP <*> pure 0
   {-# INLINE smpP #-}
 
+_smpP :: Encoding a => Parser a
+_smpP = A.space *> smpP
+
 -- lists encode/parse as a sequence of items prefixed with list length (as 1 byte)
 smpEncodeList :: Encoding a => [a] -> ByteString
 smpEncodeList xs = B.cons (lenEncode $ length xs) . B.concat $ map smpEncode xs
@@ -170,37 +174,37 @@ instance (Encoding a, Encoding b) => Encoding (a, b) where
   {-# INLINE smpP #-}
 
 instance (Encoding a, Encoding b, Encoding c) => Encoding (a, b, c) where
-  smpEncode (a, b, c) = smpEncode a <> smpEncode b <> smpEncode c
+  smpEncode (a, b, c) = B.concat [smpEncode a, smpEncode b, smpEncode c]
   {-# INLINE smpEncode #-}
   smpP = (,,) <$> smpP <*> smpP <*> smpP
   {-# INLINE smpP #-}
 
 instance (Encoding a, Encoding b, Encoding c, Encoding d) => Encoding (a, b, c, d) where
-  smpEncode (a, b, c, d) = smpEncode a <> smpEncode b <> smpEncode c <> smpEncode d
+  smpEncode (a, b, c, d) = B.concat [smpEncode a, smpEncode b, smpEncode c, smpEncode d]
   {-# INLINE smpEncode #-}
   smpP = (,,,) <$> smpP <*> smpP <*> smpP <*> smpP
   {-# INLINE smpP #-}
 
 instance (Encoding a, Encoding b, Encoding c, Encoding d, Encoding e) => Encoding (a, b, c, d, e) where
-  smpEncode (a, b, c, d, e) = smpEncode a <> smpEncode b <> smpEncode c <> smpEncode d <> smpEncode e
+  smpEncode (a, b, c, d, e) = B.concat [smpEncode a, smpEncode b, smpEncode c, smpEncode d, smpEncode e]
   {-# INLINE smpEncode #-}
   smpP = (,,,,) <$> smpP <*> smpP <*> smpP <*> smpP <*> smpP
   {-# INLINE smpP #-}
 
 instance (Encoding a, Encoding b, Encoding c, Encoding d, Encoding e, Encoding f) => Encoding (a, b, c, d, e, f) where
-  smpEncode (a, b, c, d, e, f) = smpEncode a <> smpEncode b <> smpEncode c <> smpEncode d <> smpEncode e <> smpEncode f
+  smpEncode (a, b, c, d, e, f) = B.concat [smpEncode a, smpEncode b, smpEncode c, smpEncode d, smpEncode e, smpEncode f]
   {-# INLINE smpEncode #-}
   smpP = (,,,,,) <$> smpP <*> smpP <*> smpP <*> smpP <*> smpP <*> smpP
   {-# INLINE smpP #-}
 
 instance (Encoding a, Encoding b, Encoding c, Encoding d, Encoding e, Encoding f, Encoding g) => Encoding (a, b, c, d, e, f, g) where
-  smpEncode (a, b, c, d, e, f, g) = smpEncode a <> smpEncode b <> smpEncode c <> smpEncode d <> smpEncode e <> smpEncode f <> smpEncode g
+  smpEncode (a, b, c, d, e, f, g) = B.concat [smpEncode a, smpEncode b, smpEncode c, smpEncode d, smpEncode e, smpEncode f, smpEncode g]
   {-# INLINE smpEncode #-}
   smpP = (,,,,,,) <$> smpP <*> smpP <*> smpP <*> smpP <*> smpP <*> smpP <*> smpP
   {-# INLINE smpP #-}
 
 instance (Encoding a, Encoding b, Encoding c, Encoding d, Encoding e, Encoding f, Encoding g, Encoding h) => Encoding (a, b, c, d, e, f, g, h) where
-  smpEncode (a, b, c, d, e, f, g, h) = smpEncode a <> smpEncode b <> smpEncode c <> smpEncode d <> smpEncode e <> smpEncode f <> smpEncode g <> smpEncode h
+  smpEncode (a, b, c, d, e, f, g, h) = B.concat [smpEncode a, smpEncode b, smpEncode c, smpEncode d, smpEncode e, smpEncode f, smpEncode g, smpEncode h]
   {-# INLINE smpEncode #-}
   smpP = (,,,,,,,) <$> smpP <*> smpP <*> smpP <*> smpP <*> smpP <*> smpP <*> smpP <*> smpP
   {-# INLINE smpP #-}

src/Simplex/Messaging/Notifications/Protocol.hs

@@ -11,7 +11,7 @@
 
 module Simplex.Messaging.Notifications.Protocol where
 
-import Data.Aeson (FromJSON (..), ToJSON (..), (.=))
+import Data.Aeson (FromJSON (..), ToJSON (..), (.:), (.=))
 import qualified Data.Aeson as J
 import qualified Data.Aeson.Encoding as JE
 import qualified Data.Attoparsec.ByteString.Char8 as A
@@ -408,6 +408,12 @@ instance ToJSON DeviceToken where
   toEncoding (DeviceToken pp t) = J.pairs $ "pushProvider" .= decodeLatin1 (strEncode pp) <> "token" .= decodeLatin1 t
   toJSON (DeviceToken pp t) = J.object ["pushProvider" .= decodeLatin1 (strEncode pp), "token" .= decodeLatin1 t]
 
+instance FromJSON DeviceToken where
+  parseJSON = J.withObject "DeviceToken" $ \o -> do
+    pp <- strDecode . encodeUtf8 <$?> o .: "pushProvider"
+    t <- encodeUtf8 <$> o .: "token"
+    pure $ DeviceToken pp t
+
 type NtfEntityId = ByteString
 
 type NtfSubscriptionId = NtfEntityId
@@ -429,7 +435,7 @@ data NtfSubStatus
     NSAuth
   | -- | SMP error other than AUTH
     NSErr ByteString
-  deriving (Eq, Show)
+  deriving (Eq, Ord, Show)
 
 ntfShouldSubscribe :: NtfSubStatus -> Bool
 ntfShouldSubscribe = \case
@@ -510,6 +516,9 @@ instance ToJSON NtfTknStatus where
   toEncoding = JE.text . decodeLatin1 . smpEncode
   toJSON = J.String . decodeLatin1 . smpEncode
 
+instance FromJSON NtfTknStatus where
+  parseJSON = J.withText "NtfTknStatus" $ either fail pure . smpDecode . encodeUtf8
+
 checkEntity :: forall t e e'. (NtfEntityI e, NtfEntityI e') => t e' -> Either String (t e)
 checkEntity c = case testEquality (sNtfEntity @e) (sNtfEntity @e') of
   Just Refl -> Right c

src/Simplex/Messaging/Notifications/Server.hs

@@ -12,16 +12,20 @@
 
 module Simplex.Messaging.Notifications.Server where
 
-import Control.Concurrent.STM (stateTVar)
 import Control.Logger.Simple
+import Control.Monad
 import Control.Monad.Except
 import Control.Monad.Reader
 import Data.ByteString.Char8 (ByteString)
 import qualified Data.ByteString.Char8 as B
 import Data.Functor (($>))
 import Data.Int (Int64)
-import Data.List (intercalate)
+import Data.List (intercalate, sort)
+import Data.List.NonEmpty (NonEmpty (..))
+import qualified Data.List.NonEmpty as L
 import Data.Map.Strict (Map)
+import qualified Data.Map.Strict as M
+import Data.Maybe (catMaybes)
 import qualified Data.Text as T
 import Data.Text.Encoding (decodeLatin1)
 import Data.Time.Clock (UTCTime (..), diffTimeToPicoseconds, getCurrentTime)
@@ -51,7 +55,7 @@ import System.Exit (exitFailure)
 import System.IO (BufferMode (..), hPutStrLn, hSetBuffering)
 import System.Mem.Weak (deRefWeak)
 import UnliftIO (IOMode (..), async, uninterruptibleCancel, withFile)
-import UnliftIO.Concurrent (forkIO, killThread, mkWeakThreadId, threadDelay)
+import UnliftIO.Concurrent (forkIO, killThread, mkWeakThreadId)
 import UnliftIO.Directory (doesFileExist, renameFile)
 import UnliftIO.Exception
 import UnliftIO.STM
@@ -67,7 +71,7 @@ runNtfServerBlocking started cfg = runReaderT (ntfServer cfg started) =<< newNtf
 type M a = ReaderT NtfEnv IO a
 
 ntfServer :: NtfServerConfig -> TMVar Bool -> M ()
-ntfServer cfg@NtfServerConfig {transports, logTLSErrors} started = do
+ntfServer cfg@NtfServerConfig {transports, transportConfig = tCfg} started = do
   restoreServerStats
   s <- asks subscriber
   ps <- asks pushServer
@@ -78,7 +82,7 @@ ntfServer cfg@NtfServerConfig {transports, logTLSErrors} started = do
     runServer :: (ServiceName, ATransport) -> M ()
     runServer (tcpPort, ATransport t) = do
       serverParams <- asks tlsServerParams
-      runTransportServer started tcpPort serverParams logTLSErrors (runClient t)
+      runTransportServer started tcpPort serverParams tCfg (runClient t)
 
     runClient :: Transport c => TProxy c -> c -> M ()
     runClient _ h = do
@@ -141,24 +145,27 @@ ntfServer cfg@NtfServerConfig {transports, logTLSErrors} started = do
 
 resubscribe :: NtfSubscriber -> Map NtfSubscriptionId NtfSubData -> M ()
 resubscribe NtfSubscriber {newSubQ} subs = do
-  d <- asks $ resubscribeDelay . config
-  forM_ subs $ \sub@NtfSubData {} ->
-    whenM (ntfShouldSubscribe <$> readTVarIO (subStatus sub)) $ do
-      atomically $ writeTBQueue newSubQ $ NtfSub sub
-      threadDelay d
-  liftIO $ logInfo "SMP connections resubscribed"
+  subs' <- atomically $ filterM (fmap ntfShouldSubscribe . readTVar . subStatus) $ M.elems subs
+  atomically . writeTBQueue newSubQ $ map NtfSub subs'
+  liftIO $ logInfo $ "SMP resubscriptions queued (" <> tshow (length subs') <> " subscriptions)"
 
 ntfSubscriber :: NtfSubscriber -> M ()
 ntfSubscriber NtfSubscriber {smpSubscribers, newSubQ, smpAgent = ca@SMPClientAgent {msgQ, agentQ}} = do
   raceAny_ [subscribe, receiveSMP, receiveAgent]
   where
     subscribe :: M ()
-    subscribe =
-      forever $
-        atomically (readTBQueue newSubQ) >>= \case
-          sub@(NtfSub NtfSubData {smpQueue = SMPQueueNtf {smpServer}}) -> do
-            SMPSubscriber {newSubQ = subscriberSubQ} <- getSMPSubscriber smpServer
-            atomically $ writeTQueue subscriberSubQ sub
+    subscribe = forever $ do
+      subs <- atomically (readTBQueue newSubQ)
+      let ss = L.groupAllWith server subs
+      batchSize <- asks $ subsBatchSize . config
+      forM_ ss $ \serverSubs -> do
+        let srv = server $ L.head serverSubs
+            batches = toChunks batchSize $ L.toList serverSubs
+        SMPSubscriber {newSubQ = subscriberSubQ} <- getSMPSubscriber srv
+        mapM_ (atomically . writeTQueue subscriberSubQ) batches
+
+    server :: NtfEntityRec 'Subscription -> SMPServer
+    server (NtfSub sub) = ntfSubServer sub
 
     getSMPSubscriber :: SMPServer -> M SMPSubscriber
     getSMPSubscriber smpServer =
@@ -173,21 +180,36 @@ ntfSubscriber NtfSubscriber {smpSubscribers, newSubQ, smpAgent = ca@SMPClientAge
 
     runSMPSubscriber :: SMPSubscriber -> M ()
     runSMPSubscriber SMPSubscriber {newSubQ = subscriberSubQ} =
-      forever $
-        atomically (peekTQueue subscriberSubQ)
-          >>= \(NtfSub NtfSubData {smpQueue, notifierKey}) -> do
-            updateSubStatus smpQueue NSPending
-            let SMPQueueNtf {smpServer, notifierId} = smpQueue
-            liftIO (runExceptT $ subscribeQueue ca smpServer ((SPNotifier, notifierId), notifierKey)) >>= \case
-              Right _ -> do
-                updateSubStatus smpQueue NSActive
-                void . atomically $ readTQueue subscriberSubQ
-              Left err -> do
-                handleSubError smpQueue err
-                case err of
-                  PCEResponseTimeout -> pure ()
-                  PCENetworkError -> pure ()
-                  _ -> void . atomically $ readTQueue subscriberSubQ
+      forever $ do
+        subs <- atomically (peekTQueue subscriberSubQ)
+        let subs' = L.map (\(NtfSub sub) -> sub) subs
+            srv = server $ L.head subs
+        logSubStatus srv "subscribing" $ length subs
+        mapM_ (\NtfSubData {smpQueue} -> updateSubStatus smpQueue NSPending) subs'
+        rs <- liftIO $ subscribeQueues srv subs'
+        (subs'', oks, errs) <- foldM process ([], 0, []) rs
+        atomically $ do
+          void $ readTQueue subscriberSubQ
+          mapM_ (writeTQueue subscriberSubQ . L.map NtfSub) $ L.nonEmpty subs''
+        logSubStatus srv "retrying" $ length subs''
+        logSubStatus srv "subscribed" oks
+        logSubErrors srv errs
+      where
+        process :: ([NtfSubData], Int, [NtfSubStatus]) -> (NtfSubData, Either SMPClientError ()) -> M ([NtfSubData], Int, [NtfSubStatus])
+        process (subs, oks, errs) (sub@NtfSubData {smpQueue}, r) = case r of
+          Right _ -> updateSubStatus smpQueue NSActive $> (subs, oks + 1, errs)
+          Left e -> update <$> handleSubError smpQueue e
+          where
+            update = \case
+              Just err -> (subs, oks, err : errs) -- permanent error, log and don't retry subscription
+              Nothing -> (sub : subs, oks, errs) -- temporary error, retry subscription
+
+    -- \| Subscribe to queues. The list of results can have a different order.
+    subscribeQueues :: SMPServer -> NonEmpty NtfSubData -> IO (NonEmpty (NtfSubData, Either SMPClientError ()))
+    subscribeQueues srv subs =
+      L.zipWith (\s r -> (s, snd r)) subs <$> subscribeQueuesNtfs ca srv (L.map sub subs)
+      where
+        sub NtfSubData {smpQueue = SMPQueueNtf {notifierId}, notifierKey} = (notifierId, notifierKey)
 
     receiveSMP :: M ()
     receiveSMP = forever $ do
@@ -212,46 +234,52 @@ ntfSubscriber NtfSubscriber {smpSubscribers, newSubQ, smpAgent = ca@SMPClientAge
         atomically (readTBQueue agentQ) >>= \case
           CAConnected _ -> pure ()
           CADisconnected srv subs -> do
-            logInfo $ "SMP server disconnected " <> showServer' srv <> " (" <> tshow (length subs) <> ") subscriptions"
+            logSubStatus srv "disconnected" $ length subs
             forM_ subs $ \(_, ntfId) -> do
               let smpQueue = SMPQueueNtf srv ntfId
               updateSubStatus smpQueue NSInactive
           CAReconnected srv ->
             logInfo $ "SMP server reconnected " <> showServer' srv
-          CAResubscribed srv sub -> do
-            let ntfId = snd sub
-                smpQueue = SMPQueueNtf srv ntfId
-            updateSubStatus smpQueue NSActive
-          CASubError srv (_, ntfId) err -> do
-            logError $ "SMP subscription error on server " <> showServer' srv <> ": " <> tshow err
-            handleSubError (SMPQueueNtf srv ntfId) err
-      where
-        showServer' = decodeLatin1 . strEncode . host
+          CAResubscribed srv subs -> do
+            forM_ subs $ \(_, ntfId) -> updateSubStatus (SMPQueueNtf srv ntfId) NSActive
+            logSubStatus srv "resubscribed" $ length subs
+          CASubError srv errs ->
+            forM errs (\((_, ntfId), err) -> handleSubError (SMPQueueNtf srv ntfId) err)
+              >>= logSubErrors srv . catMaybes . L.toList
 
-    handleSubError :: SMPQueueNtf -> SMPClientError -> M ()
+    logSubStatus srv event n =
+      when (n > 0) . logInfo $
+        "SMP server " <> event <> " " <> showServer' srv <> " (" <> tshow n <> " subscriptions)"
+
+    logSubErrors :: SMPServer -> [NtfSubStatus] -> M ()
+    logSubErrors srv errs = forM_ (L.group $ sort errs) $ \errs' -> do
+      logError $ "SMP subscription errors on server " <> showServer' srv <> ": " <> tshow (L.head errs') <> " (" <> tshow (length errs') <> " errors)"
+
+    showServer' = decodeLatin1 . strEncode . host
+
+    handleSubError :: SMPQueueNtf -> SMPClientError -> M (Maybe NtfSubStatus)
     handleSubError smpQueue = \case
-      PCEProtocolError AUTH -> updateSubStatus smpQueue NSAuth
+      PCEProtocolError AUTH -> updateSubStatus smpQueue NSAuth $> Just NSAuth
       PCEProtocolError e -> updateErr "SMP error " e
-      PCEIOError e -> updateErr "IOError " e
       PCEResponseError e -> updateErr "ResponseError " e
       PCEUnexpectedResponse r -> updateErr "UnexpectedResponse " r
       PCETransportError e -> updateErr "TransportError " e
       PCECryptoError e -> updateErr "CryptoError " e
-      PCEIncompatibleHost -> updateSubStatus smpQueue $ NSErr "IncompatibleHost"
-      PCEResponseTimeout -> pure ()
-      PCENetworkError -> pure ()
+      PCEIncompatibleHost -> let e = NSErr "IncompatibleHost" in updateSubStatus smpQueue e $> Just e
+      PCEResponseTimeout -> pure Nothing
+      PCENetworkError -> pure Nothing
+      PCEIOError _ -> pure Nothing
       where
-        updateErr :: Show e => ByteString -> e -> M ()
-        updateErr errType e = updateSubStatus smpQueue . NSErr $ errType <> bshow e
+        updateErr :: Show e => ByteString -> e -> M (Maybe NtfSubStatus)
+        updateErr errType e = updateSubStatus smpQueue (NSErr $ errType <> bshow e) $> Just (NSErr errType)
 
     updateSubStatus smpQueue status = do
       st <- asks store
-      atomically (findNtfSubscription st smpQueue)
-        >>= mapM_
-          ( \NtfSubData {ntfSubId, subStatus} -> do
-              atomically $ writeTVar subStatus status
-              withNtfLog $ \sl -> logSubscriptionStatus sl ntfSubId status
-          )
+      atomically (findNtfSubscription st smpQueue) >>= mapM_ update
+      where
+        update NtfSubData {ntfSubId, subStatus} = do
+          old <- atomically $ stateTVar subStatus (,status)
+          when (old /= status) $ withNtfLog $ \sl -> logSubscriptionStatus sl ntfSubId status
 
 ntfPush :: NtfPushServer -> M ()
 ntfPush s@NtfPushServer {pushQ} = forever $ do
@@ -261,11 +289,14 @@ ntfPush s@NtfPushServer {pushQ} = forever $ do
   case ntf of
     PNVerification _
       | status /= NTInvalid && status /= NTExpired ->
-        deliverNotification pp tkn ntf >>= \case
-          Right _ -> do
-            status_ <- atomically $ stateTVar tknStatus $ \status' -> if status' == NTActive then (Nothing, NTActive) else (Just NTConfirmed, NTConfirmed)
-            forM_ status_ $ \status' -> withNtfLog $ \sl -> logTokenStatus sl ntfTknId status'
-          _ -> pure ()
+          deliverNotification pp tkn ntf >>= \case
+            Right _ -> do
+              status_ <- atomically $ stateTVar tknStatus $ \case
+                NTActive -> (Nothing, NTActive)
+                NTConfirmed -> (Nothing, NTConfirmed)
+                _ -> (Just NTConfirmed, NTConfirmed)
+              forM_ status_ $ \status' -> withNtfLog $ \sl -> logTokenStatus sl ntfTknId status'
+            _ -> pure ()
       | otherwise -> logError "bad notification token status"
     PNCheckMessages -> checkActiveTkn status $ do
       void $ deliverNotification pp tkn ntf
@@ -280,7 +311,7 @@ ntfPush s@NtfPushServer {pushQ} = forever $ do
       | status == NTActive = action
       | otherwise = liftIO $ logError "bad notification token status"
     deliverNotification :: PushProvider -> NtfTknData -> PushNotification -> M (Either PushProviderError ())
-    deliverNotification pp tkn@NtfTknData {ntfTknId, tknStatus} ntf = do
+    deliverNotification pp tkn ntf = do
       deliver <- liftIO $ getPushClient s pp
       liftIO (runExceptT $ deliver tkn ntf) >>= \case
         Right _ -> pure $ Right ()
@@ -289,19 +320,20 @@ ntfPush s@NtfPushServer {pushQ} = forever $ do
           PPRetryLater -> retryDeliver
           PPCryptoError _ -> err e
           PPResponseError _ _ -> err e
-          PPTokenInvalid -> updateTknStatus NTInvalid >> err e
+          PPTokenInvalid -> updateTknStatus tkn NTInvalid >> err e
           PPPermanentError -> err e
       where
         retryDeliver :: M (Either PushProviderError ())
         retryDeliver = do
           deliver <- liftIO $ newPushClient s pp
           liftIO (runExceptT $ deliver tkn ntf) >>= either err (pure . Right)
-        updateTknStatus :: NtfTknStatus -> M ()
-        updateTknStatus status = do
-          atomically $ writeTVar tknStatus status
-          withNtfLog $ \sl -> logTokenStatus sl ntfTknId status
         err e = logError (T.pack $ "Push provider error (" <> show pp <> "): " <> show e) $> Left e
 
+updateTknStatus :: NtfTknData -> NtfTknStatus -> M ()
+updateTknStatus NtfTknData {ntfTknId, tknStatus} status = do
+  old <- atomically $ stateTVar tknStatus (,status)
+  when (old /= status) $ withNtfLog $ \sl -> logTokenStatus sl ntfTknId status
+
 runNtfClientTransport :: Transport c => THandle c -> M ()
 runNtfClientTransport th@THandle {sessionId} = do
   qSize <- asks $ clientQSize . config
@@ -313,16 +345,17 @@ runNtfClientTransport th@THandle {sessionId} = do
   raceAny_ ([liftIO $ send th c, client c s ps, receive th c] <> disconnectThread_ c expCfg)
     `finally` liftIO (clientDisconnected c)
   where
-    disconnectThread_ c expCfg = maybe [] ((: []) . liftIO . disconnectTransport th c activeAt) expCfg
+    disconnectThread_ c (Just expCfg) = [liftIO $ disconnectTransport th (rcvActiveAt c) (sndActiveAt c) expCfg (pure True)]
+    disconnectThread_ _ _ = []
 
 clientDisconnected :: NtfServerClient -> IO ()
 clientDisconnected NtfServerClient {connected} = atomically $ writeTVar connected False
 
 receive :: Transport c => THandle c -> NtfServerClient -> M ()
-receive th NtfServerClient {rcvQ, sndQ, activeAt} = forever $ do
+receive th NtfServerClient {rcvQ, sndQ, rcvActiveAt} = forever $ do
   ts <- liftIO $ tGet th
   forM_ ts $ \t@(_, _, (corrId, entId, cmdOrError)) -> do
-    atomically . writeTVar activeAt =<< liftIO getSystemTime
+    atomically . writeTVar rcvActiveAt =<< liftIO getSystemTime
     logDebug "received transmission"
     case cmdOrError of
       Left e -> write sndQ (corrId, entId, NRErr e)
@@ -334,10 +367,10 @@ receive th NtfServerClient {rcvQ, sndQ, activeAt} = forever $ do
     write q t = atomically $ writeTBQueue q t
 
 send :: Transport c => THandle c -> NtfServerClient -> IO ()
-send h@THandle {thVersion = v} NtfServerClient {sndQ, sessionId, activeAt} = forever $ do
+send h@THandle {thVersion = v} NtfServerClient {sndQ, sessionId, sndActiveAt} = forever $ do
   t <- atomically $ readTBQueue sndQ
   void . liftIO $ tPut h [(Nothing, encodeTransmission v sessionId t)]
-  atomically . writeTVar activeAt =<< liftIO getSystemTime
+  atomically . writeTVar sndActiveAt =<< liftIO getSystemTime
 
 -- instance Show a => Show (TVar a) where
 --   show x = unsafePerformIO $ show <$> readTVarIO x
@@ -407,7 +440,7 @@ client NtfServerClient {rcvQ, sndQ} NtfSubscriber {newSubQ, smpAgent = ca} NtfPu
       NtfReqNew corrId (ANE SToken newTkn@(NewNtfTkn _ _ dhPubKey)) -> do
         logDebug "TNEW - new token"
         st <- asks store
-        ks@(srvDhPubKey, srvDhPrivKey) <- liftIO C.generateKeyPair'
+        ks@(srvDhPubKey, srvDhPrivKey) <- atomically . C.generateKeyPair =<< asks random
         let dhSecret = C.dh' dhPubKey srvDhPrivKey
         tknId <- getId
         regCode <- getRegCode
@@ -431,17 +464,16 @@ client NtfServerClient {rcvQ, sndQ} NtfSubscriber {newSubQ, smpAgent = ca} NtfPu
               else pure $ NRErr AUTH
           TVFY code -- this allows repeated verification for cases when client connection dropped before server response
             | (status == NTRegistered || status == NTConfirmed || status == NTActive) && tknRegCode == code -> do
-              logDebug "TVFY - token verified"
-              st <- asks store
-              atomically $ writeTVar tknStatus NTActive
-              tIds <- atomically $ removeInactiveTokenRegistrations st tkn
-              forM_ tIds cancelInvervalNotifications
-              withNtfLog $ \s -> logTokenStatus s tknId NTActive
-              incNtfStat tknVerified
-              pure NROk
+                logDebug "TVFY - token verified"
+                st <- asks store
+                updateTknStatus tkn NTActive
+                tIds <- atomically $ removeInactiveTokenRegistrations st tkn
+                forM_ tIds cancelInvervalNotifications
+                incNtfStat tknVerified
+                pure NROk
             | otherwise -> do
-              logDebug "TVFY - incorrect code or token status"
-              pure $ NRErr AUTH
+                logDebug "TVFY - incorrect code or token status"
+                pure $ NRErr AUTH
           TCHK -> do
             logDebug "TCHK"
             pure $ NRTkn status
@@ -478,16 +510,16 @@ client NtfServerClient {rcvQ, sndQ} NtfSubscriber {newSubQ, smpAgent = ca} NtfPu
           TCRN int
             | int < 20 -> pure $ NRErr QUOTA
             | otherwise -> do
-              logDebug "TCRN"
-              atomically $ writeTVar tknCronInterval int
-              atomically (TM.lookup tknId intervalNotifiers) >>= \case
-                Nothing -> runIntervalNotifier int
-                Just IntervalNotifier {interval, action} ->
-                  unless (interval == int) $ do
-                    uninterruptibleCancel action
-                    runIntervalNotifier int
-              withNtfLog $ \s -> logTokenCron s tknId int
-              pure NROk
+                logDebug "TCRN"
+                atomically $ writeTVar tknCronInterval int
+                atomically (TM.lookup tknId intervalNotifiers) >>= \case
+                  Nothing -> runIntervalNotifier int
+                  Just IntervalNotifier {interval, action} ->
+                    unless (interval == int) $ do
+                      uninterruptibleCancel action
+                      runIntervalNotifier int
+                withNtfLog $ \s -> logTokenCron s tknId int
+                pure NROk
             where
               runIntervalNotifier interval = do
                 action <- async . intervalNotifier $ fromIntegral interval * 1000000 * 60
@@ -504,7 +536,7 @@ client NtfServerClient {rcvQ, sndQ} NtfSubscriber {newSubQ, smpAgent = ca} NtfPu
         sub <- atomically $ mkNtfSubData subId newSub
         resp <-
           atomically (addNtfSubscription st subId sub) >>= \case
-            Just _ -> atomically (writeTBQueue newSubQ $ NtfSub sub) $> NRSubId subId
+            Just _ -> atomically (writeTBQueue newSubQ [NtfSub sub]) $> NRSubId subId
             _ -> pure $ NRErr AUTH
         withNtfLog (`logCreateSubscription` sub)
         incNtfStat subCreated
@@ -533,13 +565,11 @@ client NtfServerClient {rcvQ, sndQ} NtfSubscriber {newSubQ, smpAgent = ca} NtfPu
           PING -> pure NRPong
       NtfReqPing corrId entId -> pure (corrId, entId, NRPong)
     getId :: M NtfEntityId
-    getId = getRandomBytes =<< asks (subIdBytes . config)
+    getId = randomBytes =<< asks (subIdBytes . config)
     getRegCode :: M NtfRegCode
-    getRegCode = NtfRegCode <$> (getRandomBytes =<< asks (regCodeBytes . config))
-    getRandomBytes :: Int -> M ByteString
-    getRandomBytes n = do
-      gVar <- asks idsDrg
-      atomically (C.pseudoRandomBytes n gVar)
+    getRegCode = NtfRegCode <$> (randomBytes =<< asks (regCodeBytes . config))
+    randomBytes :: Int -> M ByteString
+    randomBytes n = atomically . C.randomBytes n =<< asks random
     cancelInvervalNotifications :: NtfTokenId -> M ()
     cancelInvervalNotifications tknId =
       atomically (TM.lookupDelete tknId intervalNotifiers)

src/Simplex/Messaging/Notifications/Server/Env.hs

@@ -3,15 +3,18 @@
 {-# LANGUAGE GADTs #-}
 {-# LANGUAGE KindSignatures #-}
 {-# LANGUAGE NamedFieldPuns #-}
+{-# LANGUAGE OverloadedStrings #-}
 
 module Simplex.Messaging.Notifications.Server.Env where
 
 import Control.Concurrent (ThreadId)
 import Control.Concurrent.Async (Async)
+import Control.Logger.Simple
 import Control.Monad.IO.Unlift
 import Crypto.Random
 import Data.ByteString.Char8 (ByteString)
 import Data.Int (Int64)
+import Data.List.NonEmpty (NonEmpty)
 import Data.Time.Clock (getCurrentTime)
 import Data.Time.Clock.System (SystemTime)
 import Data.Word (Word16)
@@ -31,7 +34,7 @@ import Simplex.Messaging.Server.Expiration
 import Simplex.Messaging.TMap (TMap)
 import qualified Simplex.Messaging.TMap as TM
 import Simplex.Messaging.Transport (ATransport)
-import Simplex.Messaging.Transport.Server (loadFingerprint, loadTLSServerParams)
+import Simplex.Messaging.Transport.Server (TransportServerConfig, loadFingerprint, loadTLSServerParams)
 import System.IO (IOMode (..))
 import System.Mem.Weak (Weak)
 import UnliftIO.STM
@@ -45,9 +48,9 @@ data NtfServerConfig = NtfServerConfig
     pushQSize :: Natural,
     smpAgentCfg :: SMPClientAgentConfig,
     apnsConfig :: APNSPushClientConfig,
+    subsBatchSize :: Int,
     inactiveClientExpiration :: Maybe ExpirationConfig,
     storeLogFile :: Maybe FilePath,
-    resubscribeDelay :: Int, -- microseconds
     -- CA certificate private key is not needed for initialization
     caCertificateFile :: FilePath,
     privateKeyFile :: FilePath,
@@ -57,14 +60,14 @@ data NtfServerConfig = NtfServerConfig
     logStatsStartTime :: Int64,
     serverStatsLogFile :: FilePath,
     serverStatsBackupFile :: Maybe FilePath,
-    logTLSErrors :: Bool
+    transportConfig :: TransportServerConfig
   }
 
 defaultInactiveClientExpiration :: ExpirationConfig
 defaultInactiveClientExpiration =
   ExpirationConfig
-    { ttl = 7200, -- 2 hours
-      checkInterval = 3600 -- seconds, 1 hour
+    { ttl = 43200, -- seconds, 12 hours
+      checkInterval = 3600 -- seconds, 1 hours
     }
 
 data NtfEnv = NtfEnv
@@ -73,7 +76,7 @@ data NtfEnv = NtfEnv
     pushServer :: NtfPushServer,
     store :: NtfStore,
     storeLog :: Maybe (StoreLog 'WriteMode),
-    idsDrg :: TVar ChaChaDRG,
+    random :: TVar ChaChaDRG,
     tlsServerParams :: T.ServerParams,
     serverIdentity :: C.KeyHash,
     serverStats :: NtfServerStats
@@ -81,19 +84,21 @@ data NtfEnv = NtfEnv
 
 newNtfServerEnv :: (MonadUnliftIO m, MonadRandom m) => NtfServerConfig -> m NtfEnv
 newNtfServerEnv config@NtfServerConfig {subQSize, pushQSize, smpAgentCfg, apnsConfig, storeLogFile, caCertificateFile, certificateFile, privateKeyFile} = do
-  idsDrg <- newTVarIO =<< drgNew
+  random <- liftIO C.newRandom
   store <- atomically newNtfStore
+  logInfo "restoring subscriptions..."
   storeLog <- liftIO $ mapM (`readWriteNtfStore` store) storeLogFile
+  logInfo "restored subscriptions"
   subscriber <- atomically $ newNtfSubscriber subQSize smpAgentCfg
   pushServer <- atomically $ newNtfPushServer pushQSize apnsConfig
   tlsServerParams <- liftIO $ loadTLSServerParams caCertificateFile certificateFile privateKeyFile
   Fingerprint fp <- liftIO $ loadFingerprint caCertificateFile
   serverStats <- atomically . newNtfServerStats =<< liftIO getCurrentTime
-  pure NtfEnv {config, subscriber, pushServer, store, storeLog, idsDrg, tlsServerParams, serverIdentity = C.KeyHash fp, serverStats}
+  pure NtfEnv {config, subscriber, pushServer, store, storeLog, random, tlsServerParams, serverIdentity = C.KeyHash fp, serverStats}
 
 data NtfSubscriber = NtfSubscriber
   { smpSubscribers :: TMap SMPServer SMPSubscriber,
-    newSubQ :: TBQueue (NtfEntityRec 'Subscription),
+    newSubQ :: TBQueue [NtfEntityRec 'Subscription],
     smpAgent :: SMPClientAgent
   }
 
@@ -105,7 +110,7 @@ newNtfSubscriber qSize smpAgentCfg = do
   pure NtfSubscriber {smpSubscribers, newSubQ, smpAgent}
 
 data SMPSubscriber = SMPSubscriber
-  { newSubQ :: TQueue (NtfEntityRec 'Subscription),
+  { newSubQ :: TQueue (NonEmpty (NtfEntityRec 'Subscription)),
     subThreadId :: TVar (Maybe (Weak ThreadId))
   }
 
@@ -155,7 +160,8 @@ data NtfServerClient = NtfServerClient
     sndQ :: TBQueue (Transmission NtfResponse),
     sessionId :: ByteString,
     connected :: TVar Bool,
-    activeAt :: TVar SystemTime
+    rcvActiveAt :: TVar SystemTime,
+    sndActiveAt :: TVar SystemTime
   }
 
 newNtfServerClient :: Natural -> ByteString -> SystemTime -> STM NtfServerClient
@@ -163,5 +169,6 @@ newNtfServerClient qSize sessionId ts = do
   rcvQ <- newTBQueue qSize
   sndQ <- newTBQueue qSize
   connected <- newTVar True
-  activeAt <- newTVar ts
-  return NtfServerClient {rcvQ, sndQ, sessionId, connected, activeAt}
+  rcvActiveAt <- newTVar ts
+  sndActiveAt <- newTVar ts
+  return NtfServerClient {rcvQ, sndQ, sessionId, connected, rcvActiveAt, sndActiveAt}

src/Simplex/Messaging/Notifications/Server/Main.hs

@@ -7,28 +7,33 @@
 
 module Simplex.Messaging.Notifications.Server.Main where
 
-import Data.Either (fromRight)
 import Data.Functor (($>))
 import Data.Ini (lookupValue, readIniFile)
 import Data.Maybe (fromMaybe)
 import qualified Data.Text as T
 import Network.Socket (HostName)
 import Options.Applicative
-import Simplex.Messaging.Client.Agent (defaultSMPClientAgentConfig)
+import Simplex.Messaging.Client (ProtocolClientConfig (..))
+import Simplex.Messaging.Client.Agent (SMPClientAgentConfig (..), defaultSMPClientAgentConfig)
 import qualified Simplex.Messaging.Crypto as C
 import Simplex.Messaging.Notifications.Server (runNtfServer)
-import Simplex.Messaging.Notifications.Server.Env (NtfServerConfig (..))
+import Simplex.Messaging.Notifications.Server.Env (NtfServerConfig (..), defaultInactiveClientExpiration)
 import Simplex.Messaging.Notifications.Server.Push.APNS (defaultAPNSPushClientConfig)
 import Simplex.Messaging.Protocol (ProtoServerWithAuth (..), pattern NtfServer)
 import Simplex.Messaging.Server.CLI
+import Simplex.Messaging.Server.Expiration
 import Simplex.Messaging.Transport.Client (TransportHost (..))
+import Simplex.Messaging.Transport.Server (TransportServerConfig (..), defaultTransportServerConfig)
 import System.Directory (createDirectoryIfMissing, doesFileExist)
 import System.FilePath (combine)
 import System.IO (BufferMode (..), hSetBuffering, stderr, stdout)
 import Text.Read (readMaybe)
 
 ntfServerVersion :: String
-ntfServerVersion = "1.4.0"
+ntfServerVersion = "1.7.0.6"
+
+defaultSMPBatchDelay :: Int
+defaultSMPBatchDelay = 10000
 
 ntfServerCLI :: FilePath -> FilePath -> IO ()
 ntfServerCLI cfgPath logPath =
@@ -80,12 +85,19 @@ ntfServerCLI cfgPath logPath =
             <> ("host: " <> host <> "\n")
             <> ("port: " <> defaultServerPort <> "\n")
             <> "log_tls_errors: off\n\
-               \websockets: off\n"
+               \# delay between command batches sent to SMP relays (microseconds), 0 to disable\n"
+            <> ("smp_batch_delay: " <> show defaultSMPBatchDelay <> "\n")
+            <> "websockets: off\n\n\
+               \[INACTIVE_CLIENTS]\n\
+               \# TTL and interval to check inactive clients\n\
+               \disconnect: off\n"
+            <> ("# ttl: " <> show (ttl defaultInactiveClientExpiration) <> "\n")
+            <> ("# check_interval: " <> show (checkInterval defaultInactiveClientExpiration) <> "\n")
     runServer ini = do
       hSetBuffering stdout LineBuffering
       hSetBuffering stderr LineBuffering
       fp <- checkSavedFingerprint cfgPath defaultX509Config
-      let host = fromRight "<hostnames>" $ T.unpack <$> lookupValue "TRANSPORT" "host" ini
+      let host = either (const "<hostnames>") T.unpack $ lookupValue "TRANSPORT" "host" ini
           port = T.unpack $ strictIni "TRANSPORT" "port" ini
           cfg@NtfServerConfig {transports, storeLogFile} = serverConfig
           srv = ProtoServerWithAuth (NtfServer [THDomainName host] (if port == "443" then "" else port) (C.KeyHash fp)) Nothing
@@ -96,19 +108,26 @@ ntfServerCLI cfgPath logPath =
         enableStoreLog = settingIsOn "STORE_LOG" "enable" ini
         logStats = settingIsOn "STORE_LOG" "log_stats" ini
         c = combine cfgPath . ($ defaultX509Config)
+        smpBatchDelay = readIniDefault defaultSMPBatchDelay "TRANSPORT" "smp_batch_delay" ini
+        batchDelay = if smpBatchDelay <= 0 then Nothing else Just smpBatchDelay
         serverConfig =
           NtfServerConfig
             { transports = iniTransports ini,
               subIdBytes = 24,
               regCodeBytes = 32,
-              clientQSize = 16,
-              subQSize = 64,
-              pushQSize = 128,
-              smpAgentCfg = defaultSMPClientAgentConfig,
+              clientQSize = 64,
+              subQSize = 512,
+              pushQSize = 1048,
+              smpAgentCfg = defaultSMPClientAgentConfig {smpCfg = (smpCfg defaultSMPClientAgentConfig) {batchDelay}},
               apnsConfig = defaultAPNSPushClientConfig,
-              inactiveClientExpiration = Nothing,
+              subsBatchSize = 900,
+              inactiveClientExpiration =
+                settingIsOn "INACTIVE_CLIENTS" "disconnect" ini
+                  $> ExpirationConfig
+                    { ttl = readStrictIni "INACTIVE_CLIENTS" "ttl" ini,
+                      checkInterval = readStrictIni "INACTIVE_CLIENTS" "check_interval" ini
+                    },
               storeLogFile = enableStoreLog $> storeLogFilePath,
-              resubscribeDelay = 50000, -- 50ms
               caCertificateFile = c caCrtFile,
               privateKeyFile = c serverKeyFile,
               certificateFile = c serverCrtFile,
@@ -116,7 +135,10 @@ ntfServerCLI cfgPath logPath =
               logStatsStartTime = 0, -- seconds from 00:00 UTC
               serverStatsLogFile = combine logPath "ntf-server-stats.daily.log",
               serverStatsBackupFile = logStats $> combine logPath "ntf-server-stats.log",
-              logTLSErrors = fromMaybe False $ iniOnOff "TRANSPORT" "log_tls_errors" ini
+              transportConfig =
+                defaultTransportServerConfig
+                  { logTLSErrors = fromMaybe False $ iniOnOff "TRANSPORT" "log_tls_errors" ini
+                  }
             }
 
 data CliCommand

src/Simplex/Messaging/Notifications/Server/Push/APNS.hs

@@ -1,9 +1,9 @@
 {-# LANGUAGE DeriveAnyClass #-}
-{-# LANGUAGE DeriveGeneric #-}
 {-# LANGUAGE DuplicateRecordFields #-}
 {-# LANGUAGE LambdaCase #-}
 {-# LANGUAGE NamedFieldPuns #-}
 {-# LANGUAGE OverloadedStrings #-}
+{-# LANGUAGE TemplateHaskell #-}
 {-# OPTIONS_GHC -Wno-unrecognised-pragmas #-}
 
 {-# HLINT ignore "Use newtype instead of data" #-}
@@ -12,24 +12,26 @@ module Simplex.Messaging.Notifications.Server.Push.APNS where
 
 import Control.Exception (Exception)
 import Control.Logger.Simple
+import Control.Monad
 import Control.Monad.Except
+import Control.Monad.IO.Class
 import Crypto.Hash.Algorithms (SHA256 (..))
 import qualified Crypto.PubKey.ECC.ECDSA as EC
 import qualified Crypto.PubKey.ECC.Types as ECT
-import Crypto.Random (ChaChaDRG, drgNew)
+import Crypto.Random (ChaChaDRG)
 import qualified Crypto.Store.PKCS8 as PK
 import Data.ASN1.BinaryEncoding (DER (..))
 import Data.ASN1.Encoding
 import Data.ASN1.Types
-import Data.Aeson (FromJSON, ToJSON, (.=))
+import Data.Aeson (ToJSON, (.=))
 import qualified Data.Aeson as J
 import qualified Data.Aeson.Encoding as JE
+import qualified Data.Aeson.TH as JQ
 import Data.Bifunctor (first)
 import qualified Data.ByteString.Base64.URL as U
 import Data.ByteString.Builder (lazyByteString)
 import Data.ByteString.Char8 (ByteString)
 import qualified Data.ByteString.Lazy.Char8 as LB
-import qualified Data.CaseInsensitive as CI
 import Data.Int (Int64)
 import Data.Map.Strict (Map)
 import Data.Maybe (isNothing)
@@ -38,8 +40,7 @@ import qualified Data.Text as T
 import Data.Time.Clock.System
 import qualified Data.X509 as X
 import qualified Data.X509.CertificateStore as XS
-import GHC.Generics (Generic)
-import Network.HTTP.Types (HeaderName, Status)
+import Network.HTTP.Types (Status)
 import qualified Network.HTTP.Types as N
 import Network.HTTP2.Client (Request)
 import qualified Network.HTTP2.Client as H
@@ -47,7 +48,9 @@ import Network.Socket (HostName, ServiceName)
 import qualified Simplex.Messaging.Crypto as C
 import Simplex.Messaging.Encoding.String
 import Simplex.Messaging.Notifications.Protocol
+import Simplex.Messaging.Notifications.Server.Push.APNS.Internal
 import Simplex.Messaging.Notifications.Server.Store (NtfTknData (..))
+import Simplex.Messaging.Parsers (defaultJSON)
 import Simplex.Messaging.Protocol (EncNMsgMeta)
 import Simplex.Messaging.Transport.HTTP2 (HTTP2Body (..))
 import Simplex.Messaging.Transport.HTTP2.Client
@@ -59,17 +62,13 @@ data JWTHeader = JWTHeader
   { alg :: Text, -- key algorithm, ES256 for APNS
     kid :: Text -- key ID
   }
-  deriving (Show, Generic)
-
-instance ToJSON JWTHeader where toEncoding = J.genericToEncoding J.defaultOptions
+  deriving (Show)
 
 data JWTClaims = JWTClaims
   { iss :: Text, -- issuer, team ID for APNS
     iat :: Int64 -- issue time, seconds from epoch
   }
-  deriving (Show, Generic)
-
-instance ToJSON JWTClaims where toEncoding = J.genericToEncoding J.defaultOptions
+  deriving (Show)
 
 data JWTToken = JWTToken JWTHeader JWTClaims
   deriving (Show)
@@ -81,6 +80,10 @@ mkJWTToken hdr iss = do
 
 type SignedJWTToken = ByteString
 
+$(JQ.deriveToJSON defaultJSON ''JWTHeader)
+
+$(JQ.deriveToJSON defaultJSON ''JWTClaims)
+
 signedJWTToken :: EC.PrivateKey -> JWTToken -> IO SignedJWTToken
 signedJWTToken pk (JWTToken hdr claims) = do
   let hc = jwtEncode hdr <> "." <> jwtEncode claims
@@ -120,24 +123,13 @@ instance StrEncoding PNMessageData where
     pure PNMessageData {smpQueue, ntfTs, nmsgNonce, encNMsgMeta}
 
 data APNSNotification = APNSNotification {aps :: APNSNotificationBody, notificationData :: Maybe J.Value}
-  deriving (Show, Generic)
-
-instance ToJSON APNSNotification where
-  toJSON = J.genericToJSON J.defaultOptions {J.omitNothingFields = True}
-  toEncoding = J.genericToEncoding J.defaultOptions {J.omitNothingFields = True}
+  deriving (Show)
 
 data APNSNotificationBody
   = APNSBackground {contentAvailable :: Int}
   | APNSMutableContent {mutableContent :: Int, alert :: APNSAlertBody, category :: Maybe Text}
   | APNSAlert {alert :: APNSAlertBody, badge :: Maybe Int, sound :: Maybe Text, category :: Maybe Text}
-  deriving (Show, Generic)
-
-apnsJSONOptions :: J.Options
-apnsJSONOptions = J.defaultOptions {J.omitNothingFields = True, J.sumEncoding = J.UntaggedValue, J.fieldLabelModifier = J.camelTo2 '-'}
-
-instance ToJSON APNSNotificationBody where
-  toJSON = J.genericToJSON apnsJSONOptions
-  toEncoding = J.genericToEncoding apnsJSONOptions
+  deriving (Show)
 
 type APNSNotificationData = Map Text Text
 
@@ -240,7 +232,7 @@ createAPNSPushClient apnsHost apnsCfg@APNSPushClientConfig {authKeyFileEnv, auth
   authKeyId <- T.pack <$> getEnv authKeyIdEnv
   let jwtHeader = JWTHeader {alg = authKeyAlg, kid = authKeyId}
   jwtToken <- newTVarIO =<< mkApnsJWTToken appTeamId jwtHeader privateKey
-  nonceDrg <- drgNew >>= newTVarIO
+  nonceDrg <- C.newRandom
   pure APNSPushClient {https2Client, privateKey, jwtHeader, jwtToken, nonceDrg, apnsHost, apnsCfg}
 
 getApnsJWTToken :: APNSPushClient -> IO SignedJWTToken
@@ -303,6 +295,10 @@ apnsNotification NtfTknData {tknDhSecret} nonce paddedLen = \case
 
 -- apnAlert alert = APNSAlert {alert, badge = Nothing, sound = Nothing, category = Nothing}
 
+$(JQ.deriveToJSON apnsJSONOptions ''APNSNotificationBody)
+
+$(JQ.deriveToJSON defaultJSON ''APNSNotification)
+
 apnsRequest :: APNSPushClient -> ByteString -> APNSNotification -> IO Request
 apnsRequest c tkn ntf@APNSNotification {aps} = do
   signedJWT <- getApnsJWTToken c
@@ -335,12 +331,13 @@ type PushProviderClient = NtfTknData -> PushNotification -> ExceptT PushProvider
 
 -- this is not a newtype on purpose to have a correct JSON encoding as a record
 data APNSErrorResponse = APNSErrorResponse {reason :: Text}
-  deriving (Generic, FromJSON)
+
+$(JQ.deriveFromJSON defaultJSON ''APNSErrorResponse)
 
 apnsPushProviderClient :: APNSPushClient -> PushProviderClient
 apnsPushProviderClient c@APNSPushClient {nonceDrg, apnsCfg} tkn@NtfTknData {token = DeviceToken _ tknStr} pn = do
   http2 <- liftHTTPS2 $ getApnsHTTP2Client c
-  nonce <- atomically $ C.pseudoRandomCbNonce nonceDrg
+  nonce <- atomically $ C.randomCbNonce nonceDrg
   apnsNtf <- liftEither $ first PPCryptoError $ apnsNotification tkn nonce (paddedNtfLength apnsCfg) pn
   req <- liftIO $ apnsRequest c tknStr apnsNtf
   -- TODO when HTTP2 client is thread-safe, we can use sendRequestDirect
@@ -354,15 +351,15 @@ apnsPushProviderClient c@APNSPushClient {nonceDrg, apnsCfg} tkn@NtfTknData {toke
     result status reason'
       | status == Just N.ok200 = pure ()
       | status == Just N.badRequest400 =
-        case reason' of
-          "BadDeviceToken" -> throwError PPTokenInvalid
-          "DeviceTokenNotForTopic" -> throwError PPTokenInvalid
-          "TopicDisallowed" -> throwError PPPermanentError
-          _ -> err status reason'
+          case reason' of
+            "BadDeviceToken" -> throwError PPTokenInvalid
+            "DeviceTokenNotForTopic" -> throwError PPTokenInvalid
+            "TopicDisallowed" -> throwError PPPermanentError
+            _ -> err status reason'
       | status == Just N.forbidden403 = case reason' of
-        "ExpiredProviderToken" -> throwError PPPermanentError -- there should be no point retrying it as the token was refreshed
-        "InvalidProviderToken" -> throwError PPPermanentError
-        _ -> err status reason'
+          "ExpiredProviderToken" -> throwError PPPermanentError -- there should be no point retrying it as the token was refreshed
+          "InvalidProviderToken" -> throwError PPPermanentError
+          _ -> err status reason'
       | status == Just N.gone410 = throwError PPTokenInvalid
       | status == Just N.serviceUnavailable503 = liftIO (disconnectApnsHTTP2Client c) >> throwError PPRetryLater
       -- Just tooManyRequests429 -> TooManyRequests - too many requests for the same token
@@ -370,12 +367,3 @@ apnsPushProviderClient c@APNSPushClient {nonceDrg, apnsCfg} tkn@NtfTknData {toke
     err :: Maybe Status -> Text -> ExceptT PushProviderError IO ()
     err s r = throwError $ PPResponseError s r
     liftHTTPS2 a = ExceptT $ first PPConnection <$> a
-
-hApnsTopic :: HeaderName
-hApnsTopic = CI.mk "apns-topic"
-
-hApnsPushType :: HeaderName
-hApnsPushType = CI.mk "apns-push-type"
-
-hApnsPriority :: HeaderName
-hApnsPriority = CI.mk "apns-priority"

src/Simplex/Messaging/Notifications/Server/Push/APNS/Internal.hs

@@ -0,0 +1,20 @@
+{-# LANGUAGE OverloadedStrings #-}
+
+module Simplex.Messaging.Notifications.Server.Push.APNS.Internal where
+
+import qualified Data.Aeson as J
+import qualified Data.CaseInsensitive as CI
+import Network.HTTP.Types (HeaderName)
+import Simplex.Messaging.Parsers (defaultJSON)
+
+hApnsTopic :: HeaderName
+hApnsTopic = CI.mk "apns-topic"
+
+hApnsPushType :: HeaderName
+hApnsPushType = CI.mk "apns-push-type"
+
+hApnsPriority :: HeaderName
+hApnsPriority = CI.mk "apns-priority"
+
+apnsJSONOptions :: J.Options
+apnsJSONOptions = defaultJSON {J.sumEncoding = J.UntaggedValue, J.fieldLabelModifier = J.camelTo2 '-'}

src/Simplex/Messaging/Notifications/Server/Stats.hs

@@ -55,8 +55,8 @@ newNtfServerStats ts = do
   pure NtfServerStats {fromTime, tknCreated, tknVerified, tknDeleted, subCreated, subDeleted, ntfReceived, ntfDelivered, activeTokens, activeSubs}
 
 getNtfServerStatsData :: NtfServerStats -> STM NtfServerStatsData
-getNtfServerStatsData s = do
-  _fromTime <- readTVar $ fromTime (s :: NtfServerStats)
+getNtfServerStatsData s@NtfServerStats {fromTime} = do
+  _fromTime <- readTVar fromTime
   _tknCreated <- readTVar $ tknCreated s
   _tknVerified <- readTVar $ tknVerified s
   _tknDeleted <- readTVar $ tknDeleted s
@@ -69,8 +69,8 @@ getNtfServerStatsData s = do
   pure NtfServerStatsData {_fromTime, _tknCreated, _tknVerified, _tknDeleted, _subCreated, _subDeleted, _ntfReceived, _ntfDelivered, _activeTokens, _activeSubs}
 
 setNtfServerStats :: NtfServerStats -> NtfServerStatsData -> STM ()
-setNtfServerStats s d = do
-  writeTVar (fromTime (s :: NtfServerStats)) $! _fromTime (d :: NtfServerStatsData)
+setNtfServerStats s@NtfServerStats {fromTime} d@NtfServerStatsData {_fromTime} = do
+  writeTVar fromTime $! _fromTime
   writeTVar (tknCreated s) $! _tknCreated d
   writeTVar (tknVerified s) $! _tknVerified d
   writeTVar (tknDeleted s) $! _tknDeleted d

src/Simplex/Messaging/Notifications/Server/Store.hs

@@ -4,7 +4,6 @@
 {-# LANGUAGE KindSignatures #-}
 {-# LANGUAGE LambdaCase #-}
 {-# LANGUAGE NamedFieldPuns #-}
-{-# LANGUAGE OverloadedStrings #-}
 {-# LANGUAGE ScopedTypeVariables #-}
 
 module Simplex.Messaging.Notifications.Server.Store where
@@ -20,7 +19,7 @@ import qualified Data.Set as S
 import Data.Word (Word16)
 import qualified Simplex.Messaging.Crypto as C
 import Simplex.Messaging.Notifications.Protocol
-import Simplex.Messaging.Protocol (NtfPrivateSignKey)
+import Simplex.Messaging.Protocol (NtfPrivateSignKey, SMPServer)
 import Simplex.Messaging.TMap (TMap)
 import qualified Simplex.Messaging.TMap as TM
 import Simplex.Messaging.Util (whenM, ($>>=))
@@ -68,6 +67,9 @@ data NtfSubData = NtfSubData
     subStatus :: TVar NtfSubStatus
   }
 
+ntfSubServer :: NtfSubData -> SMPServer
+ntfSubServer NtfSubData {smpQueue = SMPQueueNtf {smpServer}} = smpServer
+
 data NtfEntityRec (e :: NtfEntity) where
   NtfTkn :: NtfTknData -> NtfEntityRec 'Token
   NtfSub :: NtfSubData -> NtfEntityRec 'Subscription

src/Simplex/Messaging/Notifications/Server/StoreLog.hs

@@ -4,6 +4,8 @@
 {-# LANGUAGE LambdaCase #-}
 {-# LANGUAGE NamedFieldPuns #-}
 {-# LANGUAGE OverloadedStrings #-}
+{-# LANGUAGE StrictData #-}
+{-# OPTIONS_GHC -fno-warn-ambiguous-fields #-}
 
 module Simplex.Messaging.Notifications.Server.StoreLog
   ( StoreLog,
@@ -22,7 +24,7 @@ module Simplex.Messaging.Notifications.Server.StoreLog
 where
 
 import Control.Concurrent.STM
-import Control.Monad (void)
+import Control.Monad
 import qualified Data.Attoparsec.ByteString.Char8 as A
 import qualified Data.ByteString.Char8 as B
 import Data.Word (Word16)
@@ -151,6 +153,7 @@ instance StrEncoding NtfSubRec where
 
 logNtfStoreRecord :: StoreLog 'WriteMode -> NtfStoreLogRecord -> IO ()
 logNtfStoreRecord = writeStoreLogRecord
+{-# INLINE logNtfStoreRecord #-}
 
 logCreateToken :: StoreLog 'WriteMode -> NtfTknData -> IO ()
 logCreateToken s tkn = logNtfStoreRecord s . CreateToken =<< atomically (mkTknRec tkn)
@@ -194,9 +197,11 @@ readNtfStore f st = mapM_ addNtfLogRecord . B.lines =<< B.readFile f
         CreateToken r@NtfTknRec {ntfTknId} -> do
           tkn <- mkTknData r
           addNtfToken st ntfTknId tkn
-        TokenStatus tknId status ->
-          getNtfToken st tknId
-            >>= mapM_ (\NtfTknData {tknStatus} -> writeTVar tknStatus status)
+        TokenStatus tknId status -> do
+          tkn_ <- getNtfToken st tknId
+          forM_ tkn_ $ \tkn@NtfTknData {tknStatus} -> do
+            writeTVar tknStatus status
+            when (status == NTActive) $ void $ removeInactiveTokenRegistrations st tkn
         UpdateToken tknId token' tknRegCode ->
           getNtfToken st tknId
             >>= mapM_
@@ -221,7 +226,5 @@ readNtfStore f st = mapM_ addNtfLogRecord . B.lines =<< B.readFile f
 
 writeNtfStore :: StoreLog 'WriteMode -> NtfStore -> IO ()
 writeNtfStore s NtfStore {tokens, subscriptions} = do
-  atomically (readTVar tokens >>= mapM mkTknRec)
-    >>= mapM_ (writeStoreLogRecord s . CreateToken)
-  atomically (readTVar subscriptions >>= mapM mkSubRec)
-    >>= mapM_ (writeStoreLogRecord s . CreateSubscription)
+  mapM_ (logCreateToken s) =<< readTVarIO tokens
+  mapM_ (logCreateSubscription s) =<< readTVarIO subscriptions

src/Simplex/Messaging/Notifications/Transport.hs

@@ -50,9 +50,9 @@ ntfServerHandshake c kh ntfVRange = do
   getHandshake th >>= \case
     NtfClientHandshake {ntfVersion, keyHash}
       | keyHash /= kh ->
-        throwError $ TEHandshake IDENTITY
-      | ntfVersion `isCompatible` ntfVRange -> do
-        pure (th :: THandle c) {thVersion = ntfVersion}
+          throwError $ TEHandshake IDENTITY
+      | ntfVersion `isCompatible` ntfVRange ->
+          pure (th :: THandle c) {thVersion = ntfVersion}
       | otherwise -> throwError $ TEHandshake VERSION
 
 -- | Notifcations server client transport handshake.

src/Simplex/Messaging/Parsers.hs

@@ -1,6 +1,7 @@
 {-# LANGUAGE CPP #-}
 {-# LANGUAGE LambdaCase #-}
 {-# LANGUAGE OverloadedStrings #-}
+{-# LANGUAGE PatternSynonyms #-}
 
 module Simplex.Messaging.Parsers where
 
@@ -13,6 +14,7 @@ import Data.ByteString.Base64
 import Data.ByteString.Char8 (ByteString)
 import qualified Data.ByteString.Char8 as B
 import Data.Char (isAlphaNum, toLower)
+import Data.String
 import Data.Text (Text)
 import qualified Data.Text as T
 import Data.Time.Clock (UTCTime)
@@ -83,7 +85,7 @@ blobFieldDecoder dec = \case
       Left e -> returnError ConversionFailed f ("couldn't parse field: " ++ e)
   f -> returnError ConversionFailed f "expecting SQLBlob column type"
 
-fromTextField_ :: (Typeable a) => (Text -> Maybe a) -> Field -> Ok a
+fromTextField_ :: Typeable a => (Text -> Maybe a) -> Field -> Ok a
 fromTextField_ fromText = \case
   f@(Field (SQLText t) _) ->
     case fromText t of
@@ -107,17 +109,21 @@ enumJSON tagModifier =
       J.allNullaryToStringTag = True
     }
 
+-- used in platform-specific encoding, includes tag for single-field encoding of sum types to allow conversion to tagged objects
 sumTypeJSON :: (String -> String) -> J.Options
 #if defined(darwin_HOST_OS) && defined(swiftJSON)
-sumTypeJSON = singleFieldJSON
+sumTypeJSON = singleFieldJSON_ $ Just SingleFieldJSONTag
 #else
 sumTypeJSON = taggedObjectJSON
 #endif
 
+pattern SingleFieldJSONTag :: (Eq a, IsString a) => a
+pattern SingleFieldJSONTag = "_owsf"
+
 taggedObjectJSON :: (String -> String) -> J.Options
 taggedObjectJSON tagModifier =
   J.defaultOptions
-    { J.sumEncoding = J.TaggedObject "type" "data",
+    { J.sumEncoding = J.TaggedObject TaggedObjectJSONTag TaggedObjectJSONData,
       J.tagSingleConstructors = True,
       J.constructorTagModifier = tagModifier,
       J.allNullaryToStringTag = False,
@@ -125,13 +131,26 @@ taggedObjectJSON tagModifier =
       J.omitNothingFields = True
     }
 
+pattern TaggedObjectJSONTag :: (Eq a, IsString a) => a
+pattern TaggedObjectJSONTag = "type"
+
+pattern TaggedObjectJSONData :: (Eq a, IsString a) => a
+pattern TaggedObjectJSONData = "data"
+
+-- used in platform independent encoding, doesn't include tag for single-field encoding of sum types
 singleFieldJSON :: (String -> String) -> J.Options
-singleFieldJSON tagModifier =
+singleFieldJSON = singleFieldJSON_ Nothing
+
+singleFieldJSON_ :: Maybe String -> (String -> String) -> J.Options
+singleFieldJSON_ objectTag tagModifier =
   J.defaultOptions
-    { J.sumEncoding = J.ObjectWithSingleField,
+    { J.sumEncoding = J.ObjectWithSingleField objectTag,
       J.tagSingleConstructors = True,
       J.constructorTagModifier = tagModifier,
       J.allNullaryToStringTag = False,
       J.nullaryToObject = True,
       J.omitNothingFields = True
     }
+
+defaultJSON :: J.Options
+defaultJSON = J.defaultOptions {J.omitNothingFields = True}

src/Simplex/Messaging/Protocol.hs

@@ -1,6 +1,6 @@
 {-# LANGUAGE AllowAmbiguousTypes #-}
 {-# LANGUAGE DataKinds #-}
-{-# LANGUAGE DeriveGeneric #-}
+{-# LANGUAGE DeriveAnyClass #-}
 {-# LANGUAGE DuplicateRecordFields #-}
 {-# LANGUAGE FlexibleContexts #-}
 {-# LANGUAGE FlexibleInstances #-}
@@ -15,6 +15,8 @@
 {-# LANGUAGE RankNTypes #-}
 {-# LANGUAGE ScopedTypeVariables #-}
 {-# LANGUAGE StandaloneDeriving #-}
+{-# LANGUAGE StrictData #-}
+{-# LANGUAGE TemplateHaskell #-}
 {-# LANGUAGE TupleSections #-}
 {-# LANGUAGE TypeApplications #-}
 {-# LANGUAGE TypeFamilyDependencies #-}
@@ -46,6 +48,7 @@ module Simplex.Messaging.Protocol
     -- * SMP protocol types
     ProtocolEncoding (..),
     Command (..),
+    SubscriptionMode (..),
     Party (..),
     Cmd (..),
     BrokerMsg (..),
@@ -115,6 +118,8 @@ module Simplex.Messaging.Protocol
     userProtocol,
     rcvMessageMeta,
     noMsgFlags,
+    messageId,
+    messageTs,
 
     -- * Parse and serialize
     ProtocolMsgTag (..),
@@ -132,12 +137,16 @@ module Simplex.Messaging.Protocol
     noAuthSrv,
 
     -- * TCP transport functions
+    TransportBatch (..),
     tPut,
+    tPutLog,
     tGet,
     tParse,
     tDecodeParseValidate,
     tEncode,
-    tEncodeBatch,
+    tEncodeBatch1,
+    batchTransmissions,
+    batchTransmissions',
 
     -- * exports for tests
     CommandTag (..),
@@ -148,8 +157,8 @@ where
 import Control.Applicative (optional, (<|>))
 import Control.Monad.Except
 import Data.Aeson (FromJSON (..), ToJSON (..))
-import qualified Data.Aeson as J
-import Data.Attoparsec.ByteString.Char8 (Parser)
+import qualified Data.Aeson.TH as J
+import Data.Attoparsec.ByteString.Char8 (Parser, (<?>))
 import qualified Data.Attoparsec.ByteString.Char8 as A
 import Data.ByteString.Char8 (ByteString)
 import qualified Data.ByteString.Char8 as B
@@ -163,9 +172,7 @@ import Data.Maybe (isJust, isNothing)
 import Data.String
 import Data.Time.Clock.System (SystemTime (..))
 import Data.Type.Equality
-import GHC.Generics (Generic)
 import GHC.TypeLits (ErrorMessage (..), TypeError, type (+))
-import Generic.Random (genericArbitraryU)
 import Network.Socket (HostName, ServiceName)
 import qualified Simplex.Messaging.Crypto as C
 import Simplex.Messaging.Encoding
@@ -175,7 +182,6 @@ import Simplex.Messaging.Transport
 import Simplex.Messaging.Transport.Client (TransportHost, TransportHosts (..))
 import Simplex.Messaging.Util (bshow, eitherToMaybe, (<$?>))
 import Simplex.Messaging.Version
-import Test.QuickCheck (Arbitrary (..))
 
 currentSMPClientVersion :: Version
 currentSMPClientVersion = 2
@@ -271,7 +277,7 @@ type EntityId = ByteString
 -- | Parameterized type for SMP protocol commands from all clients.
 data Command (p :: Party) where
   -- SMP recipient commands
-  NEW :: RcvPublicVerifyKey -> RcvPublicDhKey -> Maybe BasicAuth -> Command Recipient
+  NEW :: RcvPublicVerifyKey -> RcvPublicDhKey -> Maybe BasicAuth -> SubscriptionMode -> Command Recipient
   SUB :: Command Recipient
   KEY :: SndPublicVerifyKey -> Command Recipient
   NKEY :: NtfPublicVerifyKey -> RcvNtfPublicDhKey -> Command Recipient
@@ -294,6 +300,28 @@ deriving instance Show (Command p)
 
 deriving instance Eq (Command p)
 
+data SubscriptionMode = SMSubscribe | SMOnlyCreate
+  deriving (Eq, Show)
+
+instance StrEncoding SubscriptionMode where
+  strEncode = \case
+    SMSubscribe -> "subscribe"
+    SMOnlyCreate -> "only-create"
+  strP =
+    (A.string "subscribe" $> SMSubscribe)
+      <|> (A.string "only-create" $> SMOnlyCreate)
+        <?> "SubscriptionMode"
+
+instance Encoding SubscriptionMode where
+  smpEncode = \case
+    SMSubscribe -> "S"
+    SMOnlyCreate -> "C"
+  smpP =
+    A.anyChar >>= \case
+      'S' -> pure SMSubscribe
+      'C' -> pure SMOnlyCreate
+      _ -> fail "bad SubscriptionMode"
+
 data BrokerMsg where
   -- SMP broker messages (responses, client messages, notifications)
   IDS :: QueueIdsKeys -> BrokerMsg
@@ -330,6 +358,16 @@ data Message
         msgTs :: SystemTime
       }
 
+messageId :: Message -> MsgId
+messageId = \case
+  Message {msgId} -> msgId
+  MessageQuota {msgId} -> msgId
+
+messageTs :: Message -> SystemTime
+messageTs = \case
+  Message {msgTs} -> msgTs
+  MessageQuota {msgTs} -> msgTs
+
 instance StrEncoding RcvMessage where
   strEncode RcvMessage {msgId, msgTs, msgFlags, msgBody = EncRcvMsgBody body} =
     B.unwords
@@ -422,7 +460,14 @@ data SMPMsgMeta = SMPMsgMeta
     msgTs :: SystemTime,
     msgFlags :: MsgFlags
   }
-  deriving (Show)
+  deriving (Eq, Show)
+
+instance StrEncoding SMPMsgMeta where
+  strEncode SMPMsgMeta {msgId, msgTs, msgFlags} =
+    strEncode (msgId, msgTs, msgFlags)
+  strP = do
+    (msgId, msgTs, msgFlags) <- strP
+    pure SMPMsgMeta {msgId, msgTs, msgFlags}
 
 rcvMessageMeta :: MsgId -> ClientRcvMsgBody -> SMPMsgMeta
 rcvMessageMeta msgId = \case
@@ -445,9 +490,7 @@ instance Encoding NMsgMeta where
 
 -- it must be data for correct JSON encoding
 data MsgFlags = MsgFlags {notification :: Bool}
-  deriving (Eq, Show, Generic)
-
-instance ToJSON MsgFlags where toEncoding = J.genericToEncoding J.defaultOptions
+  deriving (Eq, Show)
 
 -- this encoding should not become bigger than 7 bytes (currently it is 1 byte)
 instance Encoding MsgFlags where
@@ -711,10 +754,16 @@ instance ProtocolTypeI p => ToJSON (SProtocolType p) where
   toEncoding = strToJEncoding
   toJSON = strToJSON
 
+instance ProtocolTypeI p => FromJSON (SProtocolType p) where
+  parseJSON = strParseJSON "SProtocolType"
+
 instance ToJSON AProtocolType where
   toEncoding = strToJEncoding
   toJSON = strToJSON
 
+instance FromJSON AProtocolType where
+  parseJSON = strParseJSON "AProtocolType"
+
 checkProtocolType :: forall t p p'. (ProtocolTypeI p, ProtocolTypeI p') => t p' -> Either String (t p)
 checkProtocolType p = case testEquality (protocolTypeI @p) (protocolTypeI @p') of
   Just Refl -> Right p
@@ -891,6 +940,9 @@ instance ToJSON CorrId where
   toJSON = strToJSON
   toEncoding = strToJEncoding
 
+instance FromJSON CorrId where
+  parseJSON = strParseJSON "CorrId"
+
 -- | Queue IDs and keys
 data QueueIdsKeys = QIK
   { rcvId :: RecipientId,
@@ -961,11 +1013,7 @@ data ErrorType
     INTERNAL
   | -- | used internally, never returned by the server (to be removed)
     DUPLICATE_ -- not part of SMP protocol, used internally
-  deriving (Eq, Generic, Read, Show)
-
-instance ToJSON ErrorType where
-  toJSON = J.genericToJSON $ sumTypeJSON id
-  toEncoding = J.genericToEncoding $ sumTypeJSON id
+  deriving (Eq, Read, Show)
 
 instance StrEncoding ErrorType where
   strEncode = \case
@@ -987,15 +1035,7 @@ data CommandError
     HAS_AUTH
   | -- | transmission has no required entity ID (e.g. SMP queue)
     NO_ENTITY
-  deriving (Eq, Generic, Read, Show)
-
-instance ToJSON CommandError where
-  toJSON = J.genericToJSON $ sumTypeJSON id
-  toEncoding = J.genericToEncoding $ sumTypeJSON id
-
-instance Arbitrary ErrorType where arbitrary = genericArbitraryU
-
-instance Arbitrary CommandError where arbitrary = genericArbitraryU
+  deriving (Eq, Read, Show)
 
 -- | SMP transmission parser.
 transmissionP :: Parser RawTransmission
@@ -1039,13 +1079,13 @@ class ProtocolMsgTag (Tag msg) => ProtocolEncoding err msg | msg -> err where
 instance PartyI p => ProtocolEncoding ErrorType (Command p) where
   type Tag (Command p) = CommandTag p
   encodeProtocol v = \case
-    NEW rKey dhKey auth_ -> case auth_ of
-      Just auth
-        | v >= 5 -> new <> e ('A', auth)
-        | otherwise -> new
-      _ -> new
+    NEW rKey dhKey auth_ subMode
+      | v >= 6 -> new <> auth <> e subMode
+      | v == 5 -> new <> auth
+      | otherwise -> new
       where
         new = e (NEW_, ' ', rKey, dhKey)
+        auth = maybe "" (e . ('A',)) auth_
     SUB -> e SUB_
     KEY k -> e (KEY_, ' ', k)
     NKEY k dhKey -> e (NKEY_, ' ', k, dhKey)
@@ -1097,10 +1137,12 @@ instance ProtocolEncoding ErrorType Cmd where
     CT SRecipient tag ->
       Cmd SRecipient <$> case tag of
         NEW_
-          | v >= 5 -> new <*> optional (A.char 'A' *> smpP)
-          | otherwise -> new <*> pure Nothing
+          | v >= 6 -> new <*> auth <*> smpP
+          | v == 5 -> new <*> auth <*> pure SMSubscribe
+          | otherwise -> new <*> pure Nothing <*> pure SMSubscribe
           where
             new = NEW <$> _smpP <*> smpP
+            auth = optional (A.char 'A' *> smpP)
         SUB_ -> pure SUB
         KEY_ -> KEY <$> _smpP
         NKEY_ -> NKEY <$> _smpP <*> smpP
@@ -1180,9 +1222,6 @@ instance ProtocolEncoding ErrorType BrokerMsg where
       | B.null queueId -> Left $ CMD NO_ENTITY
       | otherwise -> Right cmd
 
-_smpP :: Encoding a => Parser a
-_smpP = A.space *> smpP
-
 -- | Parse SMP protocol commands and broker messages
 parseProtocol :: forall err msg. ProtocolEncoding err msg => Version -> ByteString -> Either err msg
 parseProtocol v s =
@@ -1245,50 +1284,78 @@ instance Encoding CommandError where
 
 -- | Send signed SMP transmission to TCP transport.
 tPut :: Transport c => THandle c -> NonEmpty SentRawTransmission -> IO [Either TransportError ()]
-tPut th trs
-  | batch th = tPutBatch [] $ L.map tEncode trs
-  | otherwise = forM (L.toList trs) $ tPutLog . tEncode
+tPut th = fmap concat . mapM tPutBatch . batchTransmissions (batch th) (blockSize th)
   where
-    tPutBatch :: [Either TransportError ()] -> NonEmpty ByteString -> IO [Either TransportError ()]
-    tPutBatch rs ts = do
-      let (n, s, ts_) = encodeBatch 0 "" ts
-      r <- if n == 0 then largeMsg else replicate n <$> tPutLog (tEncodeBatch n s)
-      let rs' = rs <> r
-      case ts_ of
-        Just ts' -> tPutBatch rs' ts'
-        _ -> pure rs'
-    largeMsg = putStrLn "tPut error: large message" >> pure [Left TELargeMsg]
-    tPutLog s = do
-      r <- tPutBlock th s
-      case r of
-        Left e -> putStrLn ("tPut error: " <> show e)
-        _ -> pure ()
-      pure r
-    encodeBatch :: Int -> ByteString -> NonEmpty ByteString -> (Int, ByteString, Maybe (NonEmpty ByteString))
-    encodeBatch n s ts@(t :| ts_)
-      | n == 255 = (n, s, Just ts)
-      | otherwise =
-        let s' = s <> smpEncode (Large t)
-            n' = n + 1
-         in if B.length s' > blockSize th - 1 -- one byte is reserved for the number of messages in the batch
-              then (n,s,) $ if n == 0 then L.nonEmpty ts_ else Just ts
-              else case L.nonEmpty ts_ of
-                Just ts' -> encodeBatch n' s' ts'
-                _ -> (n', s', Nothing)
+    tPutBatch :: TransportBatch () -> IO [Either TransportError ()]
+    tPutBatch = \case
+      TBLargeTransmission _ -> [Left TELargeMsg] <$ putStrLn "tPut error: large message"
+      TBTransmissions s n _ -> replicate n <$> (tPutLog th s)
+      TBTransmission s _ -> (: []) <$> tPutLog th s
 
-tEncode :: (Maybe C.ASignature, ByteString) -> ByteString
+tPutLog :: Transport c => THandle c -> ByteString -> IO (Either TransportError ())
+tPutLog th s = do
+  r <- tPutBlock th s
+  case r of
+    Left e -> putStrLn ("tPut error: " <> show e)
+    _ -> pure ()
+  pure r
+
+-- ByteString in TBTransmissions includes byte with transmissions count
+data TransportBatch r = TBTransmissions ByteString Int [r] | TBTransmission ByteString r | TBLargeTransmission r
+
+batchTransmissions :: Bool -> Int -> NonEmpty SentRawTransmission -> [TransportBatch ()]
+batchTransmissions batch bSize = batchTransmissions' batch bSize . L.map (,())
+
+-- | encodes and batches transmissions into blocks,
+batchTransmissions' :: forall r. Bool -> Int -> NonEmpty (SentRawTransmission, r) -> [TransportBatch r]
+batchTransmissions' batch bSize
+  | batch = addBatch . foldr addTransmission ([], 0, 0, [], [])
+  | otherwise = map mkBatch1 . L.toList
+  where
+    mkBatch1 :: (SentRawTransmission, r) -> TransportBatch r
+    mkBatch1 (t, r)
+      -- 2 bytes are reserved for pad size
+      | B.length s <= bSize - 2 = TBTransmission s r
+      | otherwise = TBLargeTransmission r
+      where
+        s = tEncode t
+    -- 3 = 2 bytes reserved for pad size + 1 for transmission count
+    bSize' = bSize - 3
+    addTransmission :: (SentRawTransmission, r) -> ([TransportBatch r], Int, Int, [ByteString], [r]) -> ([TransportBatch r], Int, Int, [ByteString], [r])
+    addTransmission (t, r) acc@(bs, len, n, ss, rs)
+      | len' <= bSize' && n < 255 = (bs, len', 1 + n, s : ss, r : rs)
+      | sLen <= bSize' = (addBatch acc, sLen, 1, [s], [r])
+      | otherwise = (TBLargeTransmission r : addBatch acc, 0, 0, [], [])
+      where
+        s = tEncodeForBatch t
+        sLen = B.length s
+        len' = len + sLen
+    addBatch :: ([TransportBatch r], Int, Int, [ByteString], [r]) -> [TransportBatch r]
+    addBatch (bs, _len, n, ss, rs) = if n == 0 then bs else TBTransmissions b n rs : bs
+      where
+        b = B.concat $ B.singleton (lenEncode n) : ss
+
+tEncode :: SentRawTransmission -> ByteString
 tEncode (sig, t) = smpEncode (C.signatureBytes sig) <> t
+{-# INLINE tEncode #-}
 
-tEncodeBatch :: Int -> ByteString -> ByteString
-tEncodeBatch n s = lenEncode n `B.cons` s
+tEncodeForBatch :: SentRawTransmission -> ByteString
+tEncodeForBatch = smpEncode . Large . tEncode
+{-# INLINE tEncodeForBatch #-}
+
+tEncodeBatch1 :: SentRawTransmission -> ByteString
+tEncodeBatch1 t = lenEncode 1 `B.cons` tEncodeForBatch t
+{-# INLINE tEncodeBatch1 #-}
 
 encodeTransmission :: ProtocolEncoding e c => Version -> ByteString -> Transmission c -> ByteString
 encodeTransmission v sessionId (CorrId corrId, queueId, command) =
   smpEncode (sessionId, corrId, queueId) <> encodeProtocol v command
+{-# INLINE encodeTransmission #-}
 
 -- | Receive and parse transmission from the TCP transport (ignoring any trailing padding).
 tGetParse :: Transport c => THandle c -> IO (NonEmpty (Either TransportError RawTransmission))
 tGetParse th = eitherList (tParse $ batch th) <$> tGetBlock th
+{-# INLINE tGetParse #-}
 
 tParse :: Bool -> ByteString -> NonEmpty (Either TransportError RawTransmission)
 tParse batch s
@@ -1309,8 +1376,8 @@ tDecodeParseValidate :: forall err cmd. ProtocolEncoding err cmd => SessionId ->
 tDecodeParseValidate sessionId v = \case
   Right RawTransmission {signature, signed, sessId, corrId, entityId, command}
     | sessId == sessionId ->
-      let decodedTransmission = (,corrId,entityId,command) <$> C.decodeSignature signature
-       in either (const $ tError corrId) (tParseValidate signed) decodedTransmission
+        let decodedTransmission = (,corrId,entityId,command) <$> C.decodeSignature signature
+         in either (const $ tError corrId) (tParseValidate signed) decodedTransmission
     | otherwise -> (Nothing, "", (CorrId corrId, "", Left $ fromProtocolError @err @cmd PESession))
   Left _ -> tError ""
   where
@@ -1321,3 +1388,9 @@ tDecodeParseValidate sessionId v = \case
     tParseValidate signed t@(sig, corrId, entityId, command) =
       let cmd = parseProtocol @err @cmd v command >>= checkCredentials t
        in (sig, signed, (CorrId corrId, entityId, cmd))
+
+$(J.deriveJSON defaultJSON ''MsgFlags)
+
+$(J.deriveJSON (sumTypeJSON id) ''CommandError)
+
+$(J.deriveJSON (sumTypeJSON id) ''ErrorType)

src/Simplex/Messaging/Server.hs

@@ -1,3 +1,5 @@
+{-# LANGUAGE BangPatterns #-}
+{-# LANGUAGE CPP #-}
 {-# LANGUAGE DataKinds #-}
 {-# LANGUAGE DuplicateRecordFields #-}
 {-# LANGUAGE FlexibleContexts #-}
@@ -31,6 +33,7 @@ module Simplex.Messaging.Server
     disconnectTransport,
     verifyCmdSignature,
     dummyVerifyCmd,
+    randomId,
   )
 where
 
@@ -57,33 +60,44 @@ import Data.Time.Clock (UTCTime (..), diffTimeToPicoseconds, getCurrentTime)
 import Data.Time.Clock.System (SystemTime (..), getSystemTime)
 import Data.Time.Format.ISO8601 (iso8601Show)
 import Data.Type.Equality
+import GHC.Stats (getRTSStats)
 import GHC.TypeLits (KnownNat)
-import Network.Socket (ServiceName)
+import Network.Socket (ServiceName, Socket, socketToHandle)
+import Simplex.Messaging.Agent.Lock
 import qualified Simplex.Messaging.Crypto as C
 import Simplex.Messaging.Encoding (Encoding (smpEncode))
 import Simplex.Messaging.Encoding.String
 import Simplex.Messaging.Protocol
-import Simplex.Messaging.Server.Env.STM
+import Simplex.Messaging.Server.Control
+import Simplex.Messaging.Server.Env.STM as Env
 import Simplex.Messaging.Server.Expiration
 import Simplex.Messaging.Server.MsgStore
 import Simplex.Messaging.Server.MsgStore.STM
 import Simplex.Messaging.Server.QueueStore
-import Simplex.Messaging.Server.QueueStore.STM
+import Simplex.Messaging.Server.QueueStore.STM as QS
 import Simplex.Messaging.Server.Stats
 import Simplex.Messaging.Server.StoreLog
 import Simplex.Messaging.TMap (TMap)
 import qualified Simplex.Messaging.TMap as TM
 import Simplex.Messaging.Transport
+import Simplex.Messaging.Transport.Buffer (trimCR)
 import Simplex.Messaging.Transport.Server
 import Simplex.Messaging.Util
 import System.Exit (exitFailure)
-import System.IO (hPutStrLn)
+import System.IO (hPrint, hPutStrLn, hSetNewlineMode, universalNewlineMode)
 import System.Mem.Weak (deRefWeak)
+import UnliftIO (timeout)
 import UnliftIO.Concurrent
 import UnliftIO.Directory (doesFileExist, renameFile)
 import UnliftIO.Exception
 import UnliftIO.IO
 import UnliftIO.STM
+#if MIN_VERSION_base(4,18,0)
+import Data.List (sort)
+import Data.Maybe (fromMaybe)
+import GHC.Conc (listThreads, threadStatus)
+import GHC.Conc.Sync (threadLabel)
+#endif
 
 -- | Runs an SMP server using passed configuration.
 --
@@ -103,39 +117,46 @@ runSMPServerBlocking started cfg = newEnv cfg >>= runReaderT (smpServer started
 type M a = ReaderT Env IO a
 
 smpServer :: TMVar Bool -> ServerConfig -> M ()
-smpServer started cfg@ServerConfig {transports, logTLSErrors} = do
+smpServer started cfg@ServerConfig {transports, transportConfig = tCfg} = do
   s <- asks server
-  restoreServerMessages
-  restoreServerStats
+  expired <- restoreServerMessages
+  restoreServerStats expired
   raceAny_
-    ( serverThread s subscribedQ subscribers subscriptions cancelSub :
-      serverThread s ntfSubscribedQ notifiers ntfSubscriptions (\_ -> pure ()) :
-      map runServer transports <> expireMessagesThread_ cfg <> serverStatsThread_ cfg
+    ( serverThread s "server subscribedQ" subscribedQ subscribers subscriptions cancelSub
+        : serverThread s "server ntfSubscribedQ" ntfSubscribedQ Env.notifiers ntfSubscriptions (\_ -> pure ())
+        : map runServer transports <> expireMessagesThread_ cfg <> serverStatsThread_ cfg <> controlPortThread_ cfg
     )
-    `finally` (withLog closeStoreLog >> saveServerMessages >> saveServerStats)
+    `finally` withLock (savingLock s) "final" (saveServer False)
   where
     runServer :: (ServiceName, ATransport) -> M ()
     runServer (tcpPort, ATransport t) = do
       serverParams <- asks tlsServerParams
-      runTransportServer started tcpPort serverParams logTLSErrors (runClient t)
+      ss <- asks sockets
+      runTransportServerState ss started tcpPort serverParams tCfg (runClient t)
+
+    saveServer :: Bool -> M ()
+    saveServer keepMsgs = withLog closeStoreLog >> saveServerMessages keepMsgs >> saveServerStats
 
     serverThread ::
       forall s.
       Server ->
-      (Server -> TBQueue (QueueId, Client)) ->
+      String ->
+      (Server -> TQueue (QueueId, Client)) ->
       (Server -> TMap QueueId Client) ->
       (Client -> TMap QueueId s) ->
       (s -> IO ()) ->
       M ()
-    serverThread s subQ subs clientSubs unsub = forever $ do
-      atomically updateSubscribers
-        $>>= endPreviousSubscriptions
-        >>= liftIO . mapM_ unsub
+    serverThread s label subQ subs clientSubs unsub = do
+      labelMyThread label
+      forever $
+        atomically updateSubscribers
+          $>>= endPreviousSubscriptions
+          >>= liftIO . mapM_ unsub
       where
         updateSubscribers :: STM (Maybe (QueueId, Client))
         updateSubscribers = do
-          (qId, clnt) <- readTBQueue $ subQ s
-          let clientToBeNotified = \c' ->
+          (qId, clnt) <- readTQueue $ subQ s
+          let clientToBeNotified c' =
                 if sameClientSession clnt c'
                   then pure Nothing
                   else do
@@ -144,8 +165,9 @@ smpServer started cfg@ServerConfig {transports, logTLSErrors} = do
           TM.lookupInsert qId clnt (subs s) $>>= clientToBeNotified
         endPreviousSubscriptions :: (QueueId, Client) -> M (Maybe s)
         endPreviousSubscriptions (qId, c) = do
-          void . forkIO . atomically $
-            writeTBQueue (sndQ c) [(CorrId "", qId, END)]
+          void . forkIO $ do
+            labelMyThread $ label <> ".endPreviousSubscriptions"
+            atomically $ writeTBQueue (sndQ c) [(CorrId "", qId, END)]
           atomically $ TM.lookupDelete qId (clientSubs c)
 
     expireMessagesThread_ :: ServerConfig -> [M ()]
@@ -157,13 +179,16 @@ smpServer started cfg@ServerConfig {transports, logTLSErrors} = do
       ms <- asks msgStore
       quota <- asks $ msgQueueQuota . config
       let interval = checkInterval expCfg * 1000000
+      stats <- asks serverStats
+      labelMyThread "expireMessages"
       forever $ do
         liftIO $ threadDelay' interval
         old <- liftIO $ expireBeforeEpoch expCfg
         rIds <- M.keysSet <$> readTVarIO ms
-        forM_ rIds $ \rId ->
-          atomically (getMsgQueue ms rId quota)
-            >>= atomically . (`deleteExpiredMsgs` old)
+        forM_ rIds $ \rId -> do
+          q <- atomically (getMsgQueue ms rId quota)
+          deleted <- atomically $ deleteExpiredMsgs q old
+          atomically $ modifyTVar' (msgExpired stats) (+ deleted)
 
     serverStatsThread_ :: ServerConfig -> [M ()]
     serverStatsThread_ ServerConfig {logStatsInterval = Just interval, logStatsStartTime, serverStatsLogFile} =
@@ -172,10 +197,11 @@ smpServer started cfg@ServerConfig {transports, logTLSErrors} = do
 
     logServerStats :: Int64 -> Int64 -> FilePath -> M ()
     logServerStats startAt logInterval statsFilePath = do
+      labelMyThread "logServerStats"
       initialDelay <- (startAt -) . fromIntegral . (`div` 1000000_000000) . diffTimeToPicoseconds . utctDayTime <$> liftIO getCurrentTime
       liftIO $ putStrLn $ "server stats log enabled: " <> statsFilePath
       liftIO $ threadDelay' $ 1000000 * (initialDelay + if initialDelay < 0 then 86400 else 0)
-      ServerStats {fromTime, qCreated, qSecured, qDeleted, msgSent, msgRecv, activeQueues, msgSentNtf, msgRecvNtf, activeQueuesNtf, qCount, msgCount} <- asks serverStats
+      ServerStats {fromTime, qCreated, qSecured, qDeleted, msgSent, msgRecv, msgExpired, activeQueues, msgSentNtf, msgRecvNtf, activeQueuesNtf, qCount, msgCount} <- asks serverStats
       let interval = 1000000 * logInterval
       withFile statsFilePath AppendMode $ \h -> liftIO $ do
         hSetBuffering h LineBuffering
@@ -187,6 +213,7 @@ smpServer started cfg@ServerConfig {transports, logTLSErrors} = do
           qDeleted' <- atomically $ swapTVar qDeleted 0
           msgSent' <- atomically $ swapTVar msgSent 0
           msgRecv' <- atomically $ swapTVar msgRecv 0
+          msgExpired' <- atomically $ swapTVar msgExpired 0
           ps <- atomically $ periodStatCounts activeQueues ts
           msgSentNtf' <- atomically $ swapTVar msgSentNtf 0
           msgRecvNtf' <- atomically $ swapTVar msgRecvNtf 0
@@ -211,39 +238,166 @@ smpServer started cfg@ServerConfig {transports, logTLSErrors} = do
                 weekCount psNtf,
                 monthCount psNtf,
                 show qCount',
-                show msgCount'
+                show msgCount',
+                show msgExpired'
               ]
           threadDelay' interval
 
     runClient :: Transport c => TProxy c -> c -> M ()
-    runClient _ h = do
+    runClient tp h = do
       kh <- asks serverIdentity
-      smpVRange <- asks $ smpServerVRange . config
-      liftIO (runExceptT $ smpServerHandshake h kh smpVRange) >>= \case
-        Right th -> runClientTransport th
-        Left _ -> pure ()
+      ServerConfig {smpServerVRange, smpHandshakeTimeout} <- asks config
+      labelMyThread $ "smp handshake for " <> transportName tp
+      liftIO (timeout smpHandshakeTimeout . runExceptT $ smpServerHandshake h kh smpServerVRange) >>= \case
+        Just (Right th) -> runClientTransport th
+        _ -> pure ()
+
+    controlPortThread_ :: ServerConfig -> [M ()]
+    controlPortThread_ ServerConfig {controlPort = Just port} = [runCPServer port]
+    controlPortThread_ _ = []
+
+    runCPServer :: ServiceName -> M ()
+    runCPServer port = do
+      srv <- asks server
+      cpStarted <- newEmptyTMVarIO
+      u <- askUnliftIO
+      liftIO $ do
+        labelMyThread "control port server"
+        runTCPServer cpStarted port $ runCPClient u srv
+      where
+        runCPClient :: UnliftIO (ReaderT Env IO) -> Server -> Socket -> IO ()
+        runCPClient u srv sock = do
+          labelMyThread "control port client"
+          h <- socketToHandle sock ReadWriteMode
+          hSetBuffering h LineBuffering
+          hSetNewlineMode h universalNewlineMode
+          hPutStrLn h "SMP server control port\n'help' for supported commands"
+          cpLoop h
+          where
+            cpLoop h = do
+              s <- B.hGetLine h
+              case strDecode $ trimCR s of
+                Right CPQuit -> hClose h
+                Right cmd -> processCP h cmd >> cpLoop h
+                Left err -> hPutStrLn h ("error: " <> err) >> cpLoop h
+            processCP h = \case
+              CPSuspend -> hPutStrLn h "suspend not implemented"
+              CPResume -> hPutStrLn h "resume not implemented"
+              CPClients -> do
+                active <- unliftIO u (asks clients) >>= readTVarIO
+                hPutStrLn h $ "clientId,sessionId,connected,createdAt,rcvActiveAt,sndActiveAt,age,subscriptions"
+                forM_ (M.toList active) $ \(cid, Client {sessionId, connected, createdAt, rcvActiveAt, sndActiveAt, subscriptions}) -> do
+                  connected' <- bshow <$> readTVarIO connected
+                  rcvActiveAt' <- strEncode <$> readTVarIO rcvActiveAt
+                  sndActiveAt' <- strEncode <$> readTVarIO sndActiveAt
+                  now <- liftIO getSystemTime
+                  let age = systemSeconds now - systemSeconds createdAt
+                  subscriptions' <- bshow . M.size <$> readTVarIO subscriptions
+                  hPutStrLn h . B.unpack $ B.intercalate "," [bshow cid, encode sessionId, connected', strEncode createdAt, rcvActiveAt', sndActiveAt', bshow age, subscriptions']
+              CPStats -> do
+                ServerStats {fromTime, qCreated, qSecured, qDeleted, msgSent, msgRecv, msgSentNtf, msgRecvNtf, qCount, msgCount} <- unliftIO u $ asks serverStats
+                putStat "fromTime" fromTime
+                putStat "qCreated" qCreated
+                putStat "qSecured" qSecured
+                putStat "qDeleted" qDeleted
+                putStat "msgSent" msgSent
+                putStat "msgRecv" msgRecv
+                putStat "msgSentNtf" msgSentNtf
+                putStat "msgRecvNtf" msgRecvNtf
+                putStat "qCount" qCount
+                putStat "msgCount" msgCount
+                where
+                  putStat :: Show a => String -> TVar a -> IO ()
+                  putStat label var = readTVarIO var >>= \v -> hPutStrLn h $ label <> ": " <> show v
+              CPStatsRTS -> getRTSStats >>= hPrint h
+              CPThreads -> do
+#if MIN_VERSION_base(4,18,0)
+                threads <- liftIO listThreads
+                hPutStrLn h $ "Threads: " <> show (length threads)
+                forM_ (sort threads) $ \tid -> do
+                  label <- threadLabel tid
+                  status <- threadStatus tid
+                  hPutStrLn h $ show tid <> " (" <> show status <> ") " <> fromMaybe "" label
+#else
+                hPutStrLn h "Not available on GHC 8.10"
+#endif
+              CPSockets -> do
+                (accepted', closed', active') <- unliftIO u $ asks sockets
+                (accepted, closed, active) <- atomically $ (,,) <$> readTVar accepted' <*> readTVar closed' <*> readTVar active'
+                hPutStrLn h "Sockets: "
+                hPutStrLn h $ "accepted: " <> show accepted
+                hPutStrLn h $ "closed: " <> show closed
+                hPutStrLn h $ "active: " <> show (M.size active)
+                hPutStrLn h $ "leaked: " <> show (accepted - closed - M.size active)
+              CPSocketThreads -> do
+#if MIN_VERSION_base(4,18,0)
+                (_, _, active') <- unliftIO u $ asks sockets
+                active <- readTVarIO active'
+                forM_ (M.toList active) $ \(sid, tid') ->
+                  deRefWeak tid' >>= \case
+                    Nothing -> hPutStrLn h $ intercalate "," [show sid, "", "gone", ""]
+                    Just tid -> do
+                      label <- threadLabel tid
+                      status <- threadStatus tid
+                      hPutStrLn h $ intercalate "," [show sid, show tid, show status, fromMaybe "" label]
+#else
+                hPutStrLn h "Not available on GHC 8.10"
+#endif
+              CPDelete queueId' -> unliftIO u $ do
+                st <- asks queueStore
+                ms <- asks msgStore
+                stats <- asks serverStats
+                queueId <- atomically (getQueue st SSender queueId') >>= \case
+                  Left _ -> pure queueId' -- fallback to using as recipientId directly
+                  Right QueueRec {recipientId} -> pure recipientId
+                r <- atomically $
+                  deleteQueue st queueId $>>= \() ->
+                    Right <$> delMsgQueueSize ms queueId
+                case r of
+                  Left e -> liftIO . hPutStrLn h $ "error: " <> show e
+                  Right numDeleted -> do
+                    withLog (`logDeleteQueue` queueId)
+                    atomically $ modifyTVar' (qDeleted stats) (+ 1)
+                    atomically $ modifyTVar' (qCount stats) (subtract 1)
+                    liftIO . hPutStrLn h $ "ok, " <> show numDeleted <> " messages deleted"
+              CPSave -> withLock (savingLock srv) "control" $ do
+                hPutStrLn h "saving server state..."
+                unliftIO u $ saveServer True
+                hPutStrLn h "server state saved!"
+              CPHelp -> hPutStrLn h "commands: stats, stats-rts, clients, sockets, socket-threads, threads, delete, save, help, quit"
+              CPQuit -> pure ()
+              CPSkip -> pure ()
 
 runClientTransport :: Transport c => THandle c -> M ()
 runClientTransport th@THandle {thVersion, sessionId} = do
   q <- asks $ tbqSize . config
   ts <- liftIO getSystemTime
-  c <- atomically $ newClient q thVersion sessionId ts
+  active <- asks clients
+  nextClientId <- asks clientSeq
+  c <- atomically $ do
+    new@Client {clientId} <- newClient nextClientId q thVersion sessionId ts
+    TM.insert clientId new active
+    pure new
   s <- asks server
   expCfg <- asks $ inactiveClientExpiration . config
+  labelMyThread . B.unpack $ "client $" <> encode sessionId
   raceAny_ ([liftIO $ send th c, client c s, receive th c] <> disconnectThread_ c expCfg)
     `finally` clientDisconnected c
   where
-    disconnectThread_ c (Just expCfg) = [liftIO $ disconnectTransport th c activeAt expCfg]
+    disconnectThread_ c (Just expCfg) = [liftIO $ disconnectTransport th (rcvActiveAt c) (sndActiveAt c) expCfg (noSubscriptions c)]
     disconnectThread_ _ _ = []
+    noSubscriptions c = atomically $ (&&) <$> TM.null (subscriptions c) <*> TM.null (ntfSubscriptions c)
 
 clientDisconnected :: Client -> M ()
-clientDisconnected c@Client {subscriptions, connected} = do
+clientDisconnected c@Client {clientId, subscriptions, connected, sessionId} = do
+  labelMyThread . B.unpack $ "client $" <> encode sessionId <> " disc"
   atomically $ writeTVar connected False
   subs <- readTVarIO subscriptions
   liftIO $ mapM_ cancelSub subs
   atomically $ writeTVar subscriptions M.empty
   cs <- asks $ subscribers . server
   atomically . mapM_ (\rId -> TM.update deleteCurrentClient rId cs) $ M.keys subs
+  asks clients >>= atomically . TM.delete clientId
   where
     deleteCurrentClient :: Client -> Maybe Client
     deleteCurrentClient c'
@@ -260,12 +414,14 @@ cancelSub sub =
     _ -> return ()
 
 receive :: Transport c => THandle c -> Client -> M ()
-receive th Client {rcvQ, sndQ, activeAt} = forever $ do
-  ts <- L.toList <$> liftIO (tGet th)
-  atomically . writeTVar activeAt =<< liftIO getSystemTime
-  as <- partitionEithers <$> mapM cmdAction ts
-  write sndQ $ fst as
-  write rcvQ $ snd as
+receive th Client {rcvQ, sndQ, rcvActiveAt, sessionId} = do
+  labelMyThread . B.unpack $ "client $" <> encode sessionId <> " receive"
+  forever $ do
+    ts <- L.toList <$> liftIO (tGet th)
+    atomically . writeTVar rcvActiveAt =<< liftIO getSystemTime
+    as <- partitionEithers <$> mapM cmdAction ts
+    write sndQ $ fst as
+    write rcvQ $ snd as
   where
     cmdAction :: SignedTransmission ErrorType Cmd -> M (Either (Transmission BrokerMsg) (Maybe QueueRec, Transmission Cmd))
     cmdAction (sig, signed, (corrId, queueId, cmdOrError)) =
@@ -279,31 +435,38 @@ receive th Client {rcvQ, sndQ, activeAt} = forever $ do
     write q = mapM_ (atomically . writeTBQueue q) . L.nonEmpty
 
 send :: Transport c => THandle c -> Client -> IO ()
-send h@THandle {thVersion = v} Client {sndQ, sessionId, activeAt} = forever $ do
-  ts <- atomically $ L.sortWith tOrder <$> readTBQueue sndQ
-  void . liftIO . tPut h $ L.map ((Nothing,) . encodeTransmission v sessionId) ts
-  atomically . writeTVar activeAt =<< liftIO getSystemTime
+send h@THandle {thVersion = v} Client {sndQ, sessionId, sndActiveAt} = do
+  labelMyThread . B.unpack $ "client $" <> encode sessionId <> " send"
+  forever $ do
+    ts <- atomically $ L.sortWith tOrder <$> readTBQueue sndQ
+    void . liftIO . tPut h $ L.map ((Nothing,) . encodeTransmission v sessionId) ts
+    atomically . writeTVar sndActiveAt =<< liftIO getSystemTime
   where
     tOrder :: Transmission BrokerMsg -> Int
     tOrder (_, _, cmd) = case cmd of
       MSG {} -> 0
+      NMSG {} -> 0
       _ -> 1
 
-disconnectTransport :: Transport c => THandle c -> client -> (client -> TVar SystemTime) -> ExpirationConfig -> IO ()
-disconnectTransport THandle {connection} c activeAt expCfg = do
-  let interval = checkInterval expCfg * 1000000
-  forever . liftIO $ do
-    threadDelay' interval
-    old <- expireBeforeEpoch expCfg
-    ts <- readTVarIO $ activeAt c
-    when (systemSeconds ts < old) $ closeConnection connection
+disconnectTransport :: Transport c => THandle c -> TVar SystemTime -> TVar SystemTime -> ExpirationConfig -> IO Bool -> IO ()
+disconnectTransport THandle {connection, sessionId} rcvActiveAt sndActiveAt expCfg noSubscriptions = do
+  labelMyThread . B.unpack $ "client $" <> encode sessionId <> " disconnectTransport"
+  loop
+  where
+    loop = do
+      threadDelay' $ checkInterval expCfg * 1000000
+      ifM noSubscriptions checkExpired loop
+    checkExpired = do
+      old <- expireBeforeEpoch expCfg
+      ts <- max <$> readTVarIO rcvActiveAt <*> readTVarIO sndActiveAt
+      if systemSeconds ts < old then closeConnection connection else loop
 
 data VerificationResult = VRVerified (Maybe QueueRec) | VRFailed
 
 verifyTransmission :: Maybe C.ASignature -> ByteString -> QueueId -> Cmd -> M VerificationResult
 verifyTransmission sig_ signed queueId cmd =
   case cmd of
-    Cmd SRecipient (NEW k _ _) -> pure $ Nothing `verified` verifyCmdSignature sig_ signed k
+    Cmd SRecipient (NEW k _ _ _) -> pure $ Nothing `verified` verifyCmdSignature sig_ signed k
     Cmd SRecipient _ -> verifyCmd SRecipient $ verifyCmdSignature sig_ signed . recipientKey
     Cmd SSender SEND {} -> verifyCmd SSender $ verifyMaybe . senderKey
     Cmd SSender PING -> pure $ VRVerified Nothing
@@ -346,7 +509,8 @@ dummyKeyEd448 :: C.PublicKey 'C.Ed448
 dummyKeyEd448 = "MEMwBQYDK2VxAzoA6ibQc9XpkSLtwrf7PLvp81qW/etiumckVFImCMRdftcG/XopbOSaq9qyLhrgJWKOLyNrQPNVvpMA"
 
 client :: forall m. (MonadUnliftIO m, MonadReader Env m) => Client -> Server -> m ()
-client clnt@Client {thVersion, subscriptions, ntfSubscriptions, rcvQ, sndQ} Server {subscribedQ, ntfSubscribedQ, notifiers} =
+client clnt@Client {thVersion, subscriptions, ntfSubscriptions, rcvQ, sndQ, sessionId} Server {subscribedQ, ntfSubscribedQ, notifiers} = do
+  labelMyThread . B.unpack $ "client $" <> encode sessionId <> " commands"
   forever $
     atomically (readTBQueue rcvQ)
       >>= mapM processCommand
@@ -363,10 +527,10 @@ client clnt@Client {thVersion, subscriptions, ntfSubscriptions, rcvQ, sndQ} Serv
         Cmd SNotifier NSUB -> subscribeNotifications
         Cmd SRecipient command ->
           case command of
-            NEW rKey dhKey auth ->
+            NEW rKey dhKey auth subMode ->
               ifM
                 allowNew
-                (createQueue st rKey dhKey)
+                (createQueue st rKey dhKey subMode)
                 (pure (corrId, queueId, ERR AUTH))
               where
                 allowNew = do
@@ -381,9 +545,9 @@ client clnt@Client {thVersion, subscriptions, ntfSubscriptions, rcvQ, sndQ} Serv
             OFF -> suspendQueue_ st
             DEL -> delQueueAndMsgs st
       where
-        createQueue :: QueueStore -> RcvPublicVerifyKey -> RcvPublicDhKey -> m (Transmission BrokerMsg)
-        createQueue st recipientKey dhKey = time "NEW" $ do
-          (rcvPublicDhKey, privDhKey) <- liftIO C.generateKeyPair'
+        createQueue :: QueueStore -> RcvPublicVerifyKey -> RcvPublicDhKey -> SubscriptionMode -> m (Transmission BrokerMsg)
+        createQueue st recipientKey dhKey subMode = time "NEW" $ do
+          (rcvPublicDhKey, privDhKey) <- atomically . C.generateKeyPair =<< asks random
           let rcvDhSecret = C.dh' dhKey privDhKey
               qik (rcvId, sndId) = QIK {rcvId, sndId, rcvPublicDhKey}
               qRec (recipientId, senderId) =
@@ -413,7 +577,10 @@ client clnt@Client {thVersion, subscriptions, ntfSubscriptions, rcvQ, sndQ} Serv
                   stats <- asks serverStats
                   atomically $ modifyTVar' (qCreated stats) (+ 1)
                   atomically $ modifyTVar' (qCount stats) (+ 1)
-                  subscribeQueue qr rId $> IDS (qik ids)
+                  case subMode of
+                    SMOnlyCreate -> pure ()
+                    SMSubscribe -> void $ subscribeQueue qr rId
+                  pure $ IDS (qik ids)
 
             logCreateById :: StoreLog 'WriteMode -> RecipientId -> IO ()
             logCreateById s rId =
@@ -435,7 +602,7 @@ client clnt@Client {thVersion, subscriptions, ntfSubscriptions, rcvQ, sndQ} Serv
 
         addQueueNotifier_ :: QueueStore -> NtfPublicVerifyKey -> RcvNtfPublicDhKey -> m (Transmission BrokerMsg)
         addQueueNotifier_ st notifierKey dhKey = time "NKEY" $ do
-          (rcvPublicDhKey, privDhKey) <- liftIO C.generateKeyPair'
+          (rcvPublicDhKey, privDhKey) <- atomically . C.generateKeyPair =<< asks random
           let rcvNtfDhSecret = C.dh' dhKey privDhKey
           (corrId,queueId,) <$> addNotifierRetry 3 rcvPublicDhKey rcvNtfDhSecret
           where
@@ -476,7 +643,7 @@ client clnt@Client {thVersion, subscriptions, ntfSubscriptions, rcvQ, sndQ} Serv
           where
             newSub :: m (TVar Sub)
             newSub = time "SUB newSub" . atomically $ do
-              writeTBQueue subscribedQ (rId, clnt)
+              writeTQueue subscribedQ (rId, clnt)
               sub <- newTVar =<< newSubscription NoSub
               TM.insert rId sub subscriptions
               pure sub
@@ -521,7 +688,7 @@ client clnt@Client {thVersion, subscriptions, ntfSubscriptions, rcvQ, sndQ} Serv
         subscribeNotifications :: m (Transmission BrokerMsg)
         subscribeNotifications = time "NSUB" . atomically $ do
           unlessM (TM.member queueId ntfSubscriptions) $ do
-            writeTBQueue ntfSubscribedQ (queueId, clnt)
+            writeTQueue ntfSubscribedQ (queueId, clnt)
             TM.insert queueId () ntfSubscriptions
           pure ok
 
@@ -567,27 +734,27 @@ client clnt@Client {thVersion, subscriptions, ntfSubscriptions, rcvQ, sndQ} Serv
         sendMessage qr msgFlags msgBody
           | B.length msgBody > maxMessageLength = pure $ err LARGE_MSG
           | otherwise = case status qr of
-            QueueOff -> return $ err AUTH
-            QueueActive ->
-              case C.maxLenBS msgBody of
-                Left _ -> pure $ err LARGE_MSG
-                Right body -> do
-                  msg_ <- time "SEND" $ do
-                    q <- getStoreMsgQueue "SEND" $ recipientId qr
-                    expireMessages q
-                    atomically . writeMsg q =<< mkMessage body
-                  case msg_ of
-                    Nothing -> pure $ err QUOTA
-                    Just msg -> time "SEND ok" $ do
-                      stats <- asks serverStats
-                      when (notification msgFlags) $ do
-                        atomically . trySendNotification msg =<< asks idsDrg
-                        atomically $ modifyTVar' (msgSentNtf stats) (+ 1)
-                        atomically $ updatePeriodStats (activeQueuesNtf stats) (recipientId qr)
-                      atomically $ modifyTVar' (msgSent stats) (+ 1)
-                      atomically $ modifyTVar' (msgCount stats) (subtract 1)
-                      atomically $ updatePeriodStats (activeQueues stats) (recipientId qr)
-                      pure ok
+              QueueOff -> return $ err AUTH
+              QueueActive ->
+                case C.maxLenBS msgBody of
+                  Left _ -> pure $ err LARGE_MSG
+                  Right body -> do
+                    msg_ <- time "SEND" $ do
+                      q <- getStoreMsgQueue "SEND" $ recipientId qr
+                      expireMessages q
+                      atomically . writeMsg q =<< mkMessage body
+                    case msg_ of
+                      Nothing -> pure $ err QUOTA
+                      Just msg -> time "SEND ok" $ do
+                        stats <- asks serverStats
+                        when (notification msgFlags) $ do
+                          atomically . trySendNotification msg =<< asks random
+                          atomically $ modifyTVar' (msgSentNtf stats) (+ 1)
+                          atomically $ updatePeriodStats (activeQueuesNtf stats) (recipientId qr)
+                        atomically $ modifyTVar' (msgSent stats) (+ 1)
+                        atomically $ modifyTVar' (msgCount stats) (subtract 1)
+                        atomically $ updatePeriodStats (activeQueues stats) (recipientId qr)
+                        pure ok
           where
             mkMessage :: C.MaxLenBS MaxMessageLen -> m Message
             mkMessage body = do
@@ -599,7 +766,9 @@ client clnt@Client {thVersion, subscriptions, ntfSubscriptions, rcvQ, sndQ} Serv
             expireMessages q = do
               msgExp <- asks $ messageExpiration . config
               old <- liftIO $ mapM expireBeforeEpoch msgExp
-              atomically $ mapM_ (deleteExpiredMsgs q) old
+              stats <- asks serverStats
+              deleted <- atomically $ sum <$> mapM (deleteExpiredMsgs q) old
+              atomically $ modifyTVar' (msgExpired stats) (+ deleted)
 
             trySendNotification :: Message -> TVar ChaChaDRG -> STM ()
             trySendNotification msg ntfNonceDrg =
@@ -616,7 +785,7 @@ client clnt@Client {thVersion, subscriptions, ntfSubscriptions, rcvQ, sndQ} Serv
 
             mkMessageNotification :: ByteString -> SystemTime -> RcvNtfDhSecret -> TVar ChaChaDRG -> STM (C.CbNonce, EncNMsgMeta)
             mkMessageNotification msgId msgTs rcvNtfDhSecret ntfNonceDrg = do
-              cbNonce <- C.pseudoRandomCbNonce ntfNonceDrg
+              cbNonce <- C.randomCbNonce ntfNonceDrg
               let msgMeta = NMsgMeta {msgId, msgTs}
                   encNMsgMeta = C.cbEncrypt rcvNtfDhSecret cbNonce (smpEncode msgMeta) 128
               pure . (cbNonce,) $ fromRight "" encNMsgMeta
@@ -641,6 +810,7 @@ client clnt@Client {thVersion, subscriptions, ntfSubscriptions, rcvQ, sndQ} Serv
                 s -> s
               where
                 subscriber = do
+                  labelMyThread $ B.unpack ("client $" <> encode sessionId) <> " subscriber/" <> T.unpack name
                   msg <- atomically $ peekMsg q
                   time "subscriber" . atomically $ do
                     let encMsg = encryptMsg qr msg
@@ -664,11 +834,11 @@ client clnt@Client {thVersion, subscriptions, ntfSubscriptions, rcvQ, sndQ} Serv
             encrypt msgFlags body =
               let encBody = EncRcvMsgBody $ C.cbEncryptMaxLenBS (rcvDhSecret qr) (C.cbNonce msgId') body
                in RcvMessage msgId' msgTs' msgFlags encBody
-            msgId' = msgId (msg :: Message)
-            msgTs' = msgTs (msg :: Message)
+            msgId' = messageId msg
+            msgTs' = messageTs msg
 
         setDelivered :: Sub -> Message -> STM Bool
-        setDelivered s msg = tryPutTMVar (delivered s) $ msgId (msg :: Message)
+        setDelivered s msg = tryPutTMVar (delivered s) (messageId msg)
 
         getStoreMsgQueue :: T.Text -> RecipientId -> m MsgQueue
         getStoreMsgQueue name rId = time (name <> " getMsgQueue") $ do
@@ -715,12 +885,10 @@ timed name qId a = do
     sec = 1000_000000
 
 randomId :: (MonadUnliftIO m, MonadReader Env m) => Int -> m ByteString
-randomId n = do
-  gVar <- asks idsDrg
-  atomically (C.pseudoRandomBytes n gVar)
+randomId n = atomically . C.randomBytes n =<< asks random
 
-saveServerMessages :: (MonadUnliftIO m, MonadReader Env m) => m ()
-saveServerMessages = asks (storeMsgsFile . config) >>= mapM_ saveMessages
+saveServerMessages :: (MonadUnliftIO m, MonadReader Env m) => Bool -> m ()
+saveServerMessages keepMsgs = asks (storeMsgsFile . config) >>= mapM_ saveMessages
   where
     saveMessages f = do
       logInfo $ "saving messages to file " <> T.pack f
@@ -729,28 +897,32 @@ saveServerMessages = asks (storeMsgsFile . config) >>= mapM_ saveMessages
         readTVarIO ms >>= mapM_ (saveQueueMsgs ms h) . M.keys
       logInfo "messages saved"
       where
+        getMessages = if keepMsgs then snapshotMsgQueue else flushMsgQueue
         saveQueueMsgs ms h rId =
-          atomically (flushMsgQueue ms rId)
+          atomically (getMessages ms rId)
             >>= mapM_ (B.hPutStrLn h . strEncode . MLRv3 rId)
 
-restoreServerMessages :: forall m. (MonadUnliftIO m, MonadReader Env m) => m ()
-restoreServerMessages = asks (storeMsgsFile . config) >>= mapM_ restoreMessages
+restoreServerMessages :: forall m. (MonadUnliftIO m, MonadReader Env m) => m Int
+restoreServerMessages = asks (storeMsgsFile . config) >>= \case
+  Just f -> ifM (doesFileExist f) (restoreMessages f) (pure 0)
+  Nothing -> pure 0
   where
-    restoreMessages f = whenM (doesFileExist f) $ do
+    restoreMessages f = do
       logInfo $ "restoring messages from file " <> T.pack f
       st <- asks queueStore
       ms <- asks msgStore
       quota <- asks $ msgQueueQuota . config
       old_ <- asks (messageExpiration . config) $>>= (liftIO . fmap Just . expireBeforeEpoch)
-      runExceptT (liftIO (B.readFile f) >>= mapM_ (restoreMsg st ms quota old_) . B.lines) >>= \case
+      runExceptT (liftIO (B.readFile f) >>= foldM (\expired -> restoreMsg expired st ms quota old_) 0 . B.lines) >>= \case
         Left e -> do
           logError . T.pack $ "error restoring messages: " <> e
           liftIO exitFailure
-        _ -> do
+        Right expired -> do
           renameFile f $ f <> ".bak"
           logInfo "messages restored"
+          pure expired
       where
-        restoreMsg st ms quota old_ s = do
+        restoreMsg !expired st ms quota old_ s = do
           r <- liftEither . first (msgErr "parsing") $ strDecode s
           case r of
             MLRv3 rId msg -> addToMsgQueue rId msg
@@ -760,14 +932,15 @@ restoreServerMessages = asks (storeMsgsFile . config) >>= mapM_ restoreMessages
               addToMsgQueue rId msg'
           where
             addToMsgQueue rId msg = do
-              logFull <- atomically $ do
+              (isExpired, logFull) <- atomically $ do
                 q <- getMsgQueue ms rId quota
                 case msg of
                   Message {msgTs}
-                    | maybe True (systemSeconds msgTs >=) old_ -> isNothing <$> writeMsg q msg
-                    | otherwise -> pure False
-                  MessageQuota {} -> writeMsg q msg $> False
-              when logFull . logError . decodeLatin1 $ "message queue " <> strEncode rId <> " is full, message not restored: " <> strEncode (msgId (msg :: Message))
+                    | maybe True (systemSeconds msgTs >=) old_ -> (False,) . isNothing <$> writeMsg q msg
+                    | otherwise -> pure (True, False)
+                  MessageQuota {} -> writeMsg q msg $> (False, False)
+              when logFull . logError . decodeLatin1 $ "message queue " <> strEncode rId <> " is full, message not restored: " <> strEncode (messageId msg)
+              pure $ if isExpired then expired + 1 else expired
             updateMsgV1toV3 QueueRec {rcvDhSecret} RcvMessage {msgId, msgTs, msgFlags, msgBody = EncRcvMsgBody body} = do
               let nonce = C.cbNonce msgId
               msgBody <- liftEither . first (msgErr "v1 message decryption") $ C.maxLenBS =<< C.cbDecrypt rcvDhSecret nonce body
@@ -785,19 +958,21 @@ saveServerStats =
       B.writeFile f $ strEncode stats
       logInfo "server stats saved"
 
-restoreServerStats :: (MonadUnliftIO m, MonadReader Env m) => m ()
-restoreServerStats = asks (serverStatsBackupFile . config) >>= mapM_ restoreStats
+restoreServerStats :: (MonadUnliftIO m, MonadReader Env m) => Int -> m ()
+restoreServerStats expiredWhileRestoring = asks (serverStatsBackupFile . config) >>= mapM_ restoreStats
   where
     restoreStats f = whenM (doesFileExist f) $ do
       logInfo $ "restoring server stats from file " <> T.pack f
       liftIO (strDecode <$> B.readFile f) >>= \case
-        Right d -> do
+        Right d@ServerStatsData {_qCount = statsQCount} -> do
           s <- asks serverStats
-          _qCount <- fmap (length . M.keys) . readTVarIO . queues =<< asks queueStore
-          _msgCount <- foldM (\n q -> (n +) <$> readTVarIO (size q)) 0 =<< readTVarIO =<< asks msgStore
-          atomically $ setServerStats s d {_qCount, _msgCount}
+          _qCount <- fmap M.size . readTVarIO . queues =<< asks queueStore
+          _msgCount <- foldM (\(!n) q -> (n +) <$> readTVarIO (size q)) 0 =<< readTVarIO =<< asks msgStore
+          atomically $ setServerStats s d {_qCount, _msgCount, _msgExpired = _msgExpired d + expiredWhileRestoring}
           renameFile f $ f <> ".bak"
           logInfo "server stats restored"
+          when (_qCount /= statsQCount) $ logWarn $ "Queue count differs: stats: " <> tshow statsQCount <> ", store: " <> tshow _qCount
+          logInfo $ "Restored " <> tshow _msgCount <> " messages in " <> tshow _qCount <> " queues"
         Left e -> do
           logInfo $ "error restoring server stats: " <> T.pack e
           liftIO exitFailure