bench tweetnacl bindings

benchmarks/Bench.hs

@@ -0,0 +1,28 @@
+{- Benchmark harness
+
+Run with: cabal bench -O2 simplexmq-bench
+
+List cases: cabal bench -O2 simplexmq-bench --benchmark-options "-l"
+Pick one or group: cabal bench -O2 simplexmq-bench --benchmark-options "-p TRcvQueues.getDelSessQueues"
+-}
+
+module Main where
+
+import Bench.Base64
+import Bench.BsConcat
+import Bench.Compression
+import Bench.Crypto
+import Bench.SNTRUP761
+import Bench.TRcvQueues
+import Test.Tasty.Bench
+
+main :: IO ()
+main =
+  defaultMain
+    [ bgroup "TRcvQueues" benchTRcvQueues,
+      bgroup "SNTRUP761" benchSNTRUP761,
+      bgroup "Crypto" benchCrypto,
+      bgroup "Compression" benchCompression,
+      bgroup "BsConcat" benchBsConcat,
+      bgroup "Base64" benchBase64
+    ]

benchmarks/Bench/Base64.hs

@@ -0,0 +1,71 @@
+{-# LANGUAGE DataKinds #-}
+{-# LANGUAGE OverloadedStrings #-}
+{-# LANGUAGE PackageImports #-}
+{-# LANGUAGE TypeApplications #-}
+
+module Bench.Base64 where
+
+import qualified Data.Attoparsec.ByteString.Char8 as A
+import Data.ByteString (ByteString)
+import qualified Data.ByteString.Char8 as B
+import Data.Char (isAlphaNum)
+import Test.Tasty.Bench
+import qualified "base64" Data.Base64.Types as New
+import qualified "base64" Data.ByteString.Base64 as New
+import qualified "base64" Data.ByteString.Base64.URL as NewUrl
+import qualified "base64-bytestring" Data.ByteString.Base64 as Old
+import qualified "base64-bytestring" Data.ByteString.Base64.URL as OldUrl
+
+benchBase64 :: [Benchmark]
+benchBase64 =
+  [ bgroup
+      "encode"
+      [ bench "e-old" $ nf Old.encode decoded,
+        bcompare "e-old" . bench "e-new" $ nf New.encodeBase64' decoded
+      ],
+    bgroup
+      "decode"
+      [ bench "d-old" $ nf Old.decode encoded,
+        bcompare "d-old" . bench "d-new" $ nf New.decodeBase64Untyped encoded,
+        bcompare "d-old" . bench "d-typed" $ nf (New.decodeBase64 . New.assertBase64 @New.StdPadded) encoded
+      ],
+    bgroup
+      "encode url"
+      [ bench "eu-old" $ nf OldUrl.encode decoded,
+        bcompare "eu-old" . bench "eu-new" $ nf NewUrl.encodeBase64' decoded
+      ],
+    bgroup
+      "decode url"
+      [ bench "du-old" $ nf OldUrl.decode encodedUrl,
+        bcompare "du-old" . bench "du-new" $ nf NewUrl.decodeBase64Untyped encodedUrl,
+        bcompare "du-old" . bench "du-typed" $ nf (NewUrl.decodeBase64 . New.assertBase64 @New.UrlPadded) encodedUrl
+      ],
+    bgroup
+      "parsing"
+      [ bench "predicates" $ nf parsePredicates encoded,
+        bcompare "predicates" . bench "alphabet" $ nf parseAlphabet encoded
+      ]
+  ]
+
+parsePredicates :: ByteString -> Either String ByteString
+parsePredicates = A.parseOnly $ do
+  str <- A.takeWhile1 (\c -> isAlphaNum c || c == '+' || c == '/')
+  pad <- A.takeWhile (== '=')
+  either fail pure $ Old.decode (str <> pad)
+
+parseAlphabet :: ByteString -> Either String ByteString
+parseAlphabet = A.parseOnly $ do
+  str <- A.takeWhile1 (`B.elem` base64Alphabet)
+  pad <- A.takeWhile (== '=')
+  either fail pure $ Old.decode (str <> pad)
+  where
+    base64Alphabet = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/"
+
+encoded :: ByteString
+encoded = "e8JK+8V3fq6kOLqco/SaKlpNaQ7i1gfOrXoqekEl42u4mF8Bgu14T5j0189CGcUhJHw2RwCMvON+qbvQ9ecJAA=="
+
+encodedUrl :: ByteString
+encodedUrl = "e8JK-8V3fq6kOLqco_SaKlpNaQ7i1gfOrXoqekEl42u4mF8Bgu14T5j0189CGcUhJHw2RwCMvON-qbvQ9ecJAA=="
+
+decoded :: ByteString
+decoded = "{\194J\251\197w~\174\164\&8\186\156\163\244\154*ZMi\SO\226\214\a\206\173z*zA%\227k\184\152_\SOH\130\237xO\152\244\215\207B\EM\197!$|6G\NUL\140\188\227~\169\187\208\245\231\t\NUL"

benchmarks/Bench/BsConcat.hs

@@ -0,0 +1,23 @@
+{-# LANGUAGE OverloadedStrings #-}
+
+module Bench.BsConcat where
+
+import Data.ByteString (ByteString)
+import qualified Data.ByteString.Char8 as B
+import Test.Tasty.Bench
+
+benchBsConcat :: [Benchmark]
+benchBsConcat =
+  [ bgroup "3 elements"
+      [ bench "(3-tuple baseline)" $ nf (\(a, s, b) -> a `seq` s `seq` b `seq` "" :: ByteString) ("aaa" :: ByteString, " " :: ByteString, "bbb" :: ByteString),
+        bench "a <> s <> b" $ nf (\(a, s, b) -> a <> s <> b :: ByteString) ("aaa", " ", "bbb"),
+        bench "concat [a, s, b]" $ nf (\(a, s, b) -> B.concat [a, s, b] :: ByteString) ("aaa", " ", "bbb"),
+        bench "unwords [a, b]" $ nf (\(a, b) -> B.unwords [a, b] :: ByteString) ("aaa", "bbb")
+      ],
+    bgroup "5 elements"
+      [ bench "a <> s <> b <> s <> c" $ nf (\(a, s1, b, s2, c) -> a <> s1 <> b <> s2 <> c :: ByteString) ("aaa", " ", "bbb", " ", "ccc"),
+        bench "(a <> s <> b) <> (s <> c)" $ nf (\(a, s1, b, s2, c) -> (a <> s1 <> b) <> (s2 <> c) :: ByteString) ("aaa", " ", "bbb", " ", "ccc"),
+        bench "concat [a, s, b, s c]" $ nf (\(a, s1, b, s2, c) -> B.concat [a, s1, b, s2, c] :: ByteString) ("aaa", " ", "bbb", " ", "ccc"),
+        bench "unwords [a, b, c]" $ nf (\(a, b, c) -> B.unwords [a, b, c] :: ByteString) ("aaa", "bbb", "ccc")
+      ]
+  ]

benchmarks/Bench/Compression.hs

@@ -0,0 +1,30 @@
+{-# LANGUAGE OverloadedStrings #-}
+
+module Bench.Compression where
+
+import qualified Codec.Compression.Zstd as Z
+import Data.Aeson
+import qualified Data.ByteString.Char8 as B
+import qualified Data.ByteString.Lazy as LB
+import Simplex.Messaging.Compression
+import Test.Tasty.Bench
+
+benchCompression :: [Benchmark]
+benchCompression =
+  [ bgroup
+      "stateless"
+      [ bench "1" $ nf (Z.compress 1) testJson,
+        bench "3" $ nf (Z.compress 3) testJson,
+        bench "5" $ nf (Z.compress 5) testJson,
+        bench "9" $ nf (Z.compress 9) testJson,
+        bench "15" $ nf (Z.compress 19) testJson
+      ]
+  ]
+
+shortJson :: B.ByteString
+shortJson = B.take maxLengthPassthrough testJson
+
+testJson :: B.ByteString
+testJson = LB.toStrict . encode $ object ["some stuff" .= [obj, obj, obj, obj]]
+  where
+    obj = object ["test" .= [True, False, True], "arr" .= [0 :: Int .. 50], "loooooooooong key" .= String "is loooooooooooooooooooooooong-ish"]

benchmarks/Bench/Crypto.hs

@@ -0,0 +1,75 @@
+{-# LANGUAGE DataKinds #-}
+{-# LANGUAGE GADTs #-}
+{-# LANGUAGE ScopedTypeVariables #-}
+{-# OPTIONS_GHC -Wno-unrecognised-pragmas #-}
+
+{-# HLINT ignore "Use camelCase" #-}
+
+module Bench.Crypto where
+
+import qualified Simplex.Messaging.Crypto as C
+import Simplex.Messaging.Crypto.NaCl.Bindings
+import Test.Tasty.Bench
+
+import Control.Concurrent.STM (TVar)
+import Control.Monad.STM (atomically)
+import qualified Crypto.Cipher.XSalsa as XSalsa
+import qualified Crypto.MAC.Poly1305 as Poly1305
+import Crypto.Random (ChaChaDRG)
+import qualified Data.ByteArray as BA
+import Data.ByteString (ByteString)
+import qualified Data.ByteString as B
+-- import Test.Tasty (withResource)
+
+benchCrypto :: [Benchmark]
+benchCrypto =
+  [ bgroup
+      "cryptoBox"
+      [ env randomChunk $ bench "crypton" . nf cryptoBoxRef,
+        env randomChunk $ bcompare "crypton" . bench "tweetnacl" . nf cryptoBoxNaCl
+      ]
+  ]
+
+randomChunk :: IO (ByteString, ByteString, ByteString)
+randomChunk = do
+  g <- C.newRandom
+  (aPub :: C.PublicKeyX25519, _aPriv :: C.PrivateKeyX25519) <- atomically $ C.generateKeyPair g
+  (_bPub :: C.PublicKeyX25519, bPriv :: C.PrivateKeyX25519) <- atomically $ C.generateKeyPair g
+  let C.DhSecretX25519 sk = C.dh' aPub bPriv
+  -- let baShared = C.dh' bPub aPriv
+  C.CbNonce n <- atomically $ C.randomCbNonce g
+  msg <- atomically $ C.randomBytes 12345 g
+  pure (BA.convert sk, n, msg)
+
+cryptoBoxRef :: (ByteString, ByteString, ByteString) -> ByteString
+cryptoBoxRef (k, nonce', msg) = ref_cryptoBox k nonce' msg
+
+cryptoBoxNaCl :: (ByteString, ByteString, ByteString) -> ByteString
+cryptoBoxNaCl (k, nonce', msg) = C.cryptoBox k nonce' msg
+
+ref_cryptoBox :: BA.ByteArrayAccess key => key -> ByteString -> ByteString -> ByteString
+ref_cryptoBox secret nonce s = BA.convert tag <> c
+  where
+    (rs, c) = ref_xSalsa20 secret nonce s
+    tag = Poly1305.auth rs c
+
+-- | NaCl @crypto_box@ decrypt with a shared DH secret and 192-bit nonce (without unpadding).
+ref_sbDecryptNoPad_ :: BA.ByteArrayAccess key => key -> C.CbNonce -> ByteString -> Either C.CryptoError ByteString
+ref_sbDecryptNoPad_ secret (C.CbNonce nonce) packet
+  | B.length packet < 16 = Left C.CBDecryptError
+  | BA.constEq tag' tag = Right msg
+  | otherwise = Left C.CBDecryptError
+  where
+    (tag', c) = B.splitAt 16 packet
+    (rs, msg) = ref_xSalsa20 secret nonce c
+    tag = Poly1305.auth rs c
+
+ref_xSalsa20 :: BA.ByteArrayAccess key => key -> ByteString -> ByteString -> (ByteString, ByteString)
+ref_xSalsa20 secret nonce msg = (rs, msg')
+  where
+    zero = B.replicate 16 $ toEnum 0
+    (iv0, iv1) = B.splitAt 8 nonce
+    state0 = XSalsa.initialize 20 secret (zero `B.append` iv0)
+    state1 = XSalsa.derive state0 iv1
+    (rs, state2) = XSalsa.generate state1 32
+    (msg', _) = XSalsa.combine state2 msg

benchmarks/Bench/SNTRUP761.hs

@@ -0,0 +1,15 @@
+module Bench.SNTRUP761 where
+
+import qualified Simplex.Messaging.Crypto as C
+import Simplex.Messaging.Crypto.SNTRUP761.Bindings
+import Test.Tasty.Bench
+
+import Test.Tasty (withResource)
+
+benchSNTRUP761 :: [Benchmark]
+benchSNTRUP761 =
+  [ bgroup
+      "sntrup761Keypair"
+      [ withResource C.newRandom (\_ -> pure ()) $ bench "current" . whnfAppIO (>>= sntrup761Keypair)
+      ]
+  ]

benchmarks/Bench/TRcvQueues.hs

@@ -0,0 +1,137 @@
+{-# LANGUAGE NamedFieldPuns #-}
+{-# LANGUAGE OverloadedStrings #-}
+{-# LANGUAGE TypeApplications #-}
+
+module Bench.TRcvQueues where
+
+import Control.Monad (replicateM, unless)
+import Crypto.Random
+import Data.Bifunctor (bimap)
+import Data.ByteString (ByteString)
+import Data.Hashable (hash)
+import Simplex.Messaging.Agent.Protocol (ConnId, QueueStatus (..), UserId)
+import Simplex.Messaging.Agent.Store (DBQueueId (..), RcvQueue, StoredRcvQueue (..))
+import qualified Simplex.Messaging.Agent.TRcvQueues as Current
+import qualified Simplex.Messaging.Crypto as C
+import Simplex.Messaging.Protocol (ProtocolServer (..), SMPServer, SProtocolType (..), currentSMPClientVersion)
+import Simplex.Messaging.Transport.Client (TransportHost (..))
+import Test.Tasty.Bench
+import qualified Data.Map.Strict as M
+import UnliftIO
+
+-- For quick equivalence tests
+-- import GHC.IO (unsafePerformIO)
+-- import Test.Hspec
+-- import Test.Tasty.Hspec (testSpec)
+
+
+benchTRcvQueues :: [Benchmark]
+benchTRcvQueues =
+  [ bgroup
+      "addQueue"
+      [ bench "aq-current" $ nfIO prepareCurrent,
+        bcompare "aq-current" . bench "aq-batch" $ nfIO prepareCurrentBatch
+      ],
+    bgroup "getDelSessQueues" benchGDS,
+    bgroup "resubscribe" benchResubscribe
+  ]
+
+benchGDS :: [Benchmark]
+benchGDS =
+  [ env prepareCurrent $ bench "gds-current" . nfAppIO (fmap (bimap length length) . benchGDSCurrent)
+  -- unsafePerformIO $ testSpec "gds-equiv" testGDSequivalent
+  ]
+  where
+    benchGDSCurrent (tSess, qs) = atomically $ Current.getDelSessQueues tSess qs
+
+-- testGDSequivalent = it "same" $ do
+--   m@(mKey, _) <- prepareMaster
+--   c@(cKey, _) <- prepareCurrent
+--   mKey `shouldBe` cKey
+--   qsMaster <- benchGDSMaster m
+--   (qsCurrent, _connIds) <- benchGDSCurrent c
+--   length qsMaster `shouldNotBe` 0
+--   length qsMaster `shouldBe` length qsCurrent
+--   qsMaster `shouldBe` qsCurrent
+
+benchResubscribe :: [Benchmark]
+benchResubscribe =
+  [ env (prepareCurrent >>= pickActiveCurrent 1.0) $ bench "resub-current-full" . nfAppIO benchResubCurrent,
+    env (prepareCurrent >>= pickActiveCurrent 0.5) $ bench "resub-current-half" . nfAppIO benchResubCurrent,
+    env (prepareCurrent >>= pickActiveCurrent 0.0) $ bench "resub-current-none" . nfAppIO benchResubCurrent
+  ]
+  where
+    pickActiveCurrent rOk (_tsess, activeSubs) = do
+      ok <- readTVarIO $ Current.getConnections activeSubs
+      let num = fromIntegral (M.size ok) * rOk :: Float
+      let ok' = take (round num) $ M.keys ok
+      pure (ok', activeSubs)
+    benchResubCurrent (okConns, activeSubs) = do
+      cs <- readTVarIO $ Current.getConnections activeSubs
+      let conns = filter (`M.notMember` cs) okConns
+      unless (null conns) $ pure ()
+
+type TSessKey = (UserId, SMPServer, Maybe ConnId)
+
+prepareCurrent :: IO (TSessKey, Current.TRcvQueues)
+prepareCurrent = prepareWith Current.empty Current.addQueue
+
+prepareCurrentBatch :: IO (TSessKey, Current.TRcvQueues)
+prepareCurrentBatch = prepareQueues Current.empty Current.batchAddQueues
+
+prepareWith :: STM qs -> (RcvQueue -> qs -> STM ()) -> IO (TSessKey, qs)
+prepareWith initQS addQueue = prepareQueues initQS (\trqs qs -> mapM_ (`addQueue` trqs) qs)
+
+prepareQueues :: STM qs -> (qs -> [RcvQueue] -> STM ()) -> IO (TSessKey, qs)
+prepareQueues initQS addQueues = do
+  let (servers, gen1) = genServers gen0 nServers
+  let (qs, _gen2) = genQueues gen1 servers nUsers nQueues
+  atomically $ do
+    trqs <- initQS
+    addQueues trqs qs
+    pure (fmap (const Nothing) . Current.qKey $ head qs, trqs)
+  where
+    nUsers = 4
+    nServers = 10
+    nQueues = 10000
+
+genServers :: ChaChaDRG -> Int -> ([SMPServer], ChaChaDRG)
+genServers random nServers =
+  withDRG random . replicateM nServers $ do
+    host <- THOnionHost <$> getRandomBytes 32
+    keyHash <- C.KeyHash <$> getRandomBytes 64
+    pure ProtocolServer {scheme = SPSMP, host = pure host, port = "12345", keyHash}
+
+genQueues :: ChaChaDRG -> [SMPServer] -> Int -> Int -> ([RcvQueue], ChaChaDRG)
+genQueues random servers nUsers nQueues =
+  withDRG random . replicateM nQueues $ do
+    userRandom <- hash @ByteString <$> getRandomBytes 8
+    let userId = fromIntegral $ userRandom `mod` nUsers
+    connId <- getRandomBytes 10
+    serverRandom <- hash @ByteString <$> getRandomBytes 8
+    let server = servers !! (serverRandom `mod` nServers)
+    pure
+      RcvQueue
+        { userId,
+          connId,
+          server,
+          rcvId = "",
+          rcvPrivateKey = C.APrivateAuthKey C.SEd25519 "MC4CAQAwBQYDK2VwBCIEIDfEfevydXXfKajz3sRkcQ7RPvfWUPoq6pu1TYHV1DEe",
+          rcvDhSecret = "01234567890123456789012345678901",
+          e2ePrivKey = "MC4CAQAwBQYDK2VuBCIEINCzbVFaCiYHoYncxNY8tSIfn0pXcIAhLBfFc0m+gOpk",
+          e2eDhSecret = Nothing,
+          sndId = "",
+          status = New,
+          dbQueueId = DBQueueId 0,
+          primary = True,
+          dbReplaceQueueId = Nothing,
+          rcvSwchStatus = Nothing,
+          smpClientVersion = currentSMPClientVersion,
+          clientNtfCreds = Nothing,
+          deleteErrors = 0
+        }
+  where
+    nServers = length servers
+
+gen0 :: ChaChaDRG
+gen0 = drgNewSeed (seedFromInteger 100500)

cabal.project

@@ -28,3 +28,8 @@ source-repository-package
     type: git
     location: https://github.com/simplex-chat/sqlcipher-simple.git
     tag: a46bd361a19376c5211f1058908fc0ae6bf42446
+
+source-repository-package
+    type: git
+    location: https://github.com/emilypi/base64.git
+    tag: e67505b35084040c91c833bae6a9e6592863fd04

cbits/tweetnacl.c

@@ -0,0 +1,814 @@
+#include "tweetnacl.h"
+#define FOR(i,n) for (i = 0;i < n;++i)
+#define sv static void
+
+typedef unsigned char u8;
+typedef unsigned long u32;
+typedef unsigned long long u64;
+typedef long long i64;
+typedef i64 gf[16];
+// extern void randombytes(u8 *,u64);
+// fudge unused function
+#include "string.h"
+static void randombytes(u8 * dst, u64 len) {
+  memset(dst, 42, len);
+}
+
+static const u8
+  _0[16],
+  _9[32] = {9};
+static const gf
+  gf0,
+  gf1 = {1},
+  _121665 = {0xDB41,1},
+  D = {0x78a3, 0x1359, 0x4dca, 0x75eb, 0xd8ab, 0x4141, 0x0a4d, 0x0070, 0xe898, 0x7779, 0x4079, 0x8cc7, 0xfe73, 0x2b6f, 0x6cee, 0x5203},
+  D2 = {0xf159, 0x26b2, 0x9b94, 0xebd6, 0xb156, 0x8283, 0x149a, 0x00e0, 0xd130, 0xeef3, 0x80f2, 0x198e, 0xfce7, 0x56df, 0xd9dc, 0x2406},
+  X = {0xd51a, 0x8f25, 0x2d60, 0xc956, 0xa7b2, 0x9525, 0xc760, 0x692c, 0xdc5c, 0xfdd6, 0xe231, 0xc0a4, 0x53fe, 0xcd6e, 0x36d3, 0x2169},
+  Y = {0x6658, 0x6666, 0x6666, 0x6666, 0x6666, 0x6666, 0x6666, 0x6666, 0x6666, 0x6666, 0x6666, 0x6666, 0x6666, 0x6666, 0x6666, 0x6666},
+  I = {0xa0b0, 0x4a0e, 0x1b27, 0xc4ee, 0xe478, 0xad2f, 0x1806, 0x2f43, 0xd7a7, 0x3dfb, 0x0099, 0x2b4d, 0xdf0b, 0x4fc1, 0x2480, 0x2b83};
+
+static u32 L32(u32 x,int c) { return (x << c) | ((x&0xffffffff) >> (32 - c)); }
+
+static u32 ld32(const u8 *x)
+{
+  u32 u = x[3];
+  u = (u<<8)|x[2];
+  u = (u<<8)|x[1];
+  return (u<<8)|x[0];
+}
+
+static u64 dl64(const u8 *x)
+{
+  u64 i,u=0;
+  FOR(i,8) u=(u<<8)|x[i];
+  return u;
+}
+
+sv st32(u8 *x,u32 u)
+{
+  int i;
+  FOR(i,4) { x[i] = u; u >>= 8; }
+}
+
+sv ts64(u8 *x,u64 u)
+{
+  int i;
+  for (i = 7;i >= 0;--i) { x[i] = u; u >>= 8; }
+}
+
+static int vn(const u8 *x,const u8 *y,int n)
+{
+  u32 i,d = 0;
+  FOR(i,n) d |= x[i]^y[i];
+  return (1 & ((d - 1) >> 8)) - 1;
+}
+
+int crypto_verify_16(const u8 *x,const u8 *y)
+{
+  return vn(x,y,16);
+}
+
+int crypto_verify_32(const u8 *x,const u8 *y)
+{
+  return vn(x,y,32);
+}
+
+sv core(u8 *out,const u8 *in,const u8 *k,const u8 *c,int h)
+{
+  u32 w[16],x[16],y[16],t[4];
+  int i,j,m;
+
+  FOR(i,4) {
+    x[5*i] = ld32(c+4*i);
+    x[1+i] = ld32(k+4*i);
+    x[6+i] = ld32(in+4*i);
+    x[11+i] = ld32(k+16+4*i);
+  }
+
+  FOR(i,16) y[i] = x[i];
+
+  FOR(i,20) {
+    FOR(j,4) {
+      FOR(m,4) t[m] = x[(5*j+4*m)%16];
+      t[1] ^= L32(t[0]+t[3], 7);
+      t[2] ^= L32(t[1]+t[0], 9);
+      t[3] ^= L32(t[2]+t[1],13);
+      t[0] ^= L32(t[3]+t[2],18);
+      FOR(m,4) w[4*j+(j+m)%4] = t[m];
+    }
+    FOR(m,16) x[m] = w[m];
+  }
+
+  if (h) {
+    FOR(i,16) x[i] += y[i];
+    FOR(i,4) {
+      x[5*i] -= ld32(c+4*i);
+      x[6+i] -= ld32(in+4*i);
+    }
+    FOR(i,4) {
+      st32(out+4*i,x[5*i]);
+      st32(out+16+4*i,x[6+i]);
+    }
+  } else
+    FOR(i,16) st32(out + 4 * i,x[i] + y[i]);
+}
+
+int crypto_core_salsa20(u8 *out,const u8 *in,const u8 *k,const u8 *c)
+{
+  core(out,in,k,c,0);
+  return 0;
+}
+
+int crypto_core_hsalsa20(u8 *out,const u8 *in,const u8 *k,const u8 *c)
+{
+  core(out,in,k,c,1);
+  return 0;
+}
+
+static const u8 sigma[16] = "expand 32-byte k";
+
+int crypto_stream_salsa20_xor(u8 *c,const u8 *m,u64 b,const u8 *n,const u8 *k)
+{
+  u8 z[16],x[64];
+  u32 u,i;
+  if (!b) return 0;
+  FOR(i,16) z[i] = 0;
+  FOR(i,8) z[i] = n[i];
+  while (b >= 64) {
+    crypto_core_salsa20(x,z,k,sigma);
+    FOR(i,64) c[i] = (m?m[i]:0) ^ x[i];
+    u = 1;
+    for (i = 8;i < 16;++i) {
+      u += (u32) z[i];
+      z[i] = u;
+      u >>= 8;
+    }
+    b -= 64;
+    c += 64;
+    if (m) m += 64;
+  }
+  if (b) {
+    crypto_core_salsa20(x,z,k,sigma);
+    FOR(i,b) c[i] = (m?m[i]:0) ^ x[i];
+  }
+  return 0;
+}
+
+int crypto_stream_salsa20(u8 *c,u64 d,const u8 *n,const u8 *k)
+{
+  return crypto_stream_salsa20_xor(c,0,d,n,k);
+}
+
+int crypto_stream(u8 *c,u64 d,const u8 *n,const u8 *k)
+{
+  u8 s[32];
+  crypto_core_hsalsa20(s,n,k,sigma);
+  return crypto_stream_salsa20(c,d,n+16,s);
+}
+
+int crypto_stream_xor(u8 *c,const u8 *m,u64 d,const u8 *n,const u8 *k)
+{
+  u8 s[32];
+  crypto_core_hsalsa20(s,n,k,sigma);
+  return crypto_stream_salsa20_xor(c,m,d,n+16,s);
+}
+
+sv add1305(u32 *h,const u32 *c)
+{
+  u32 j,u = 0;
+  FOR(j,17) {
+    u += h[j] + c[j];
+    h[j] = u & 255;
+    u >>= 8;
+  }
+}
+
+static const u32 minusp[17] = {
+  5, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 252
+} ;
+
+int crypto_onetimeauth(u8 *out,const u8 *m,u64 n,const u8 *k)
+{
+  u32 s,i,j,u,x[17],r[17],h[17],c[17],g[17];
+
+  FOR(j,17) r[j]=h[j]=0;
+  FOR(j,16) r[j]=k[j];
+  r[3]&=15;
+  r[4]&=252;
+  r[7]&=15;
+  r[8]&=252;
+  r[11]&=15;
+  r[12]&=252;
+  r[15]&=15;
+
+  while (n > 0) {
+    FOR(j,17) c[j] = 0;
+    for (j = 0;(j < 16) && (j < n);++j) c[j] = m[j];
+    c[j] = 1;
+    m += j; n -= j;
+    add1305(h,c);
+    FOR(i,17) {
+      x[i] = 0;
+      FOR(j,17) x[i] += h[j] * ((j <= i) ? r[i - j] : 320 * r[i + 17 - j]);
+    }
+    FOR(i,17) h[i] = x[i];
+    u = 0;
+    FOR(j,16) {
+      u += h[j];
+      h[j] = u & 255;
+      u >>= 8;
+    }
+    u += h[16]; h[16] = u & 3;
+    u = 5 * (u >> 2);
+    FOR(j,16) {
+      u += h[j];
+      h[j] = u & 255;
+      u >>= 8;
+    }
+    u += h[16]; h[16] = u;
+  }
+
+  FOR(j,17) g[j] = h[j];
+  add1305(h,minusp);
+  s = -(h[16] >> 7);
+  FOR(j,17) h[j] ^= s & (g[j] ^ h[j]);
+
+  FOR(j,16) c[j] = k[j + 16];
+  c[16] = 0;
+  add1305(h,c);
+  FOR(j,16) out[j] = h[j];
+  return 0;
+}
+
+int crypto_onetimeauth_verify(const u8 *h,const u8 *m,u64 n,const u8 *k)
+{
+  u8 x[16];
+  crypto_onetimeauth(x,m,n,k);
+  return crypto_verify_16(h,x);
+}
+
+int crypto_secretbox(u8 *c,const u8 *m,u64 d,const u8 *n,const u8 *k)
+{
+  int i;
+  if (d < 32) return -1;
+  crypto_stream_xor(c,m,d,n,k);
+  crypto_onetimeauth(c + 16,c + 32,d - 32,c);
+  FOR(i,16) c[i] = 0;
+  return 0;
+}
+
+int crypto_secretbox_open(u8 *m,const u8 *c,u64 d,const u8 *n,const u8 *k)
+{
+  int i;
+  u8 x[32];
+  if (d < 32) return -1;
+  crypto_stream(x,32,n,k);
+  if (crypto_onetimeauth_verify(c + 16,c + 32,d - 32,x) != 0) return -1;
+  crypto_stream_xor(m,c,d,n,k);
+  FOR(i,32) m[i] = 0;
+  return 0;
+}
+
+sv set25519(gf r, const gf a)
+{
+  int i;
+  FOR(i,16) r[i]=a[i];
+}
+
+sv car25519(gf o)
+{
+  int i;
+  i64 c;
+  FOR(i,16) {
+    o[i]+=(1LL<<16);
+    c=o[i]>>16;
+    o[(i+1)*(i<15)]+=c-1+37*(c-1)*(i==15);
+    o[i]-=c<<16;
+  }
+}
+
+sv sel25519(gf p,gf q,int b)
+{
+  i64 t,i,c=~(b-1);
+  FOR(i,16) {
+    t= c&(p[i]^q[i]);
+    p[i]^=t;
+    q[i]^=t;
+  }
+}
+
+sv pack25519(u8 *o,const gf n)
+{
+  int i,j,b;
+  gf m,t;
+  FOR(i,16) t[i]=n[i];
+  car25519(t);
+  car25519(t);
+  car25519(t);
+  FOR(j,2) {
+    m[0]=t[0]-0xffed;
+    for(i=1;i<15;i++) {
+      m[i]=t[i]-0xffff-((m[i-1]>>16)&1);
+      m[i-1]&=0xffff;
+    }
+    m[15]=t[15]-0x7fff-((m[14]>>16)&1);
+    b=(m[15]>>16)&1;
+    m[14]&=0xffff;
+    sel25519(t,m,1-b);
+  }
+  FOR(i,16) {
+    o[2*i]=t[i]&0xff;
+    o[2*i+1]=t[i]>>8;
+  }
+}
+
+static int neq25519(const gf a, const gf b)
+{
+  u8 c[32],d[32];
+  pack25519(c,a);
+  pack25519(d,b);
+  return crypto_verify_32(c,d);
+}
+
+static u8 par25519(const gf a)
+{
+  u8 d[32];
+  pack25519(d,a);
+  return d[0]&1;
+}
+
+sv unpack25519(gf o, const u8 *n)
+{
+  int i;
+  FOR(i,16) o[i]=n[2*i]+((i64)n[2*i+1]<<8);
+  o[15]&=0x7fff;
+}
+
+sv A(gf o,const gf a,const gf b)
+{
+  int i;
+  FOR(i,16) o[i]=a[i]+b[i];
+}
+
+sv Z(gf o,const gf a,const gf b)
+{
+  int i;
+  FOR(i,16) o[i]=a[i]-b[i];
+}
+
+sv M(gf o,const gf a,const gf b)
+{
+  i64 i,j,t[31];
+  FOR(i,31) t[i]=0;
+  FOR(i,16) FOR(j,16) t[i+j]+=a[i]*b[j];
+  FOR(i,15) t[i]+=38*t[i+16];
+  FOR(i,16) o[i]=t[i];
+  car25519(o);
+  car25519(o);
+}
+
+sv S(gf o,const gf a)
+{
+  M(o,a,a);
+}
+
+sv inv25519(gf o,const gf i)
+{
+  gf c;
+  int a;
+  FOR(a,16) c[a]=i[a];
+  for(a=253;a>=0;a--) {
+    S(c,c);
+    if(a!=2&&a!=4) M(c,c,i);
+  }
+  FOR(a,16) o[a]=c[a];
+}
+
+sv pow2523(gf o,const gf i)
+{
+  gf c;
+  int a;
+  FOR(a,16) c[a]=i[a];
+  for(a=250;a>=0;a--) {
+    S(c,c);
+    if(a!=1) M(c,c,i);
+  }
+  FOR(a,16) o[a]=c[a];
+}
+
+int crypto_scalarmult(u8 *q,const u8 *n,const u8 *p)
+{
+  u8 z[32];
+  i64 x[80],r,i;
+  gf a,b,c,d,e,f;
+  FOR(i,31) z[i]=n[i];
+  z[31]=(n[31]&127)|64;
+  z[0]&=248;
+  unpack25519(x,p);
+  FOR(i,16) {
+    b[i]=x[i];
+    d[i]=a[i]=c[i]=0;
+  }
+  a[0]=d[0]=1;
+  for(i=254;i>=0;--i) {
+    r=(z[i>>3]>>(i&7))&1;
+    sel25519(a,b,r);
+    sel25519(c,d,r);
+    A(e,a,c);
+    Z(a,a,c);
+    A(c,b,d);
+    Z(b,b,d);
+    S(d,e);
+    S(f,a);
+    M(a,c,a);
+    M(c,b,e);
+    A(e,a,c);
+    Z(a,a,c);
+    S(b,a);
+    Z(c,d,f);
+    M(a,c,_121665);
+    A(a,a,d);
+    M(c,c,a);
+    M(a,d,f);
+    M(d,b,x);
+    S(b,e);
+    sel25519(a,b,r);
+    sel25519(c,d,r);
+  }
+  FOR(i,16) {
+    x[i+16]=a[i];
+    x[i+32]=c[i];
+    x[i+48]=b[i];
+    x[i+64]=d[i];
+  }
+  inv25519(x+32,x+32);
+  M(x+16,x+16,x+32);
+  pack25519(q,x+16);
+  return 0;
+}
+
+int crypto_scalarmult_base(u8 *q,const u8 *n)
+{ 
+  return crypto_scalarmult(q,n,_9);
+}
+
+int crypto_box_keypair(u8 *y,u8 *x)
+{
+  randombytes(x,32);
+  return crypto_scalarmult_base(y,x);
+}
+
+int crypto_box_beforenm(u8 *k,const u8 *y,const u8 *x)
+{
+  u8 s[32];
+  crypto_scalarmult(s,x,y);
+  return crypto_core_hsalsa20(k,_0,s,sigma);
+}
+
+int crypto_box_afternm(u8 *c,const u8 *m,u64 d,const u8 *n,const u8 *k)
+{
+  return crypto_secretbox(c,m,d,n,k);
+}
+
+int crypto_box_open_afternm(u8 *m,const u8 *c,u64 d,const u8 *n,const u8 *k)
+{
+  return crypto_secretbox_open(m,c,d,n,k);
+}
+
+int crypto_box(u8 *c,const u8 *m,u64 d,const u8 *n,const u8 *y,const u8 *x)
+{
+  u8 k[32];
+  crypto_box_beforenm(k,y,x);
+  return crypto_box_afternm(c,m,d,n,k);
+}
+
+int crypto_box_open(u8 *m,const u8 *c,u64 d,const u8 *n,const u8 *y,const u8 *x)
+{
+  u8 k[32];
+  crypto_box_beforenm(k,y,x);
+  return crypto_box_open_afternm(m,c,d,n,k);
+}
+
+static u64 R(u64 x,int c) { return (x >> c) | (x << (64 - c)); }
+static u64 Ch(u64 x,u64 y,u64 z) { return (x & y) ^ (~x & z); }
+static u64 Maj(u64 x,u64 y,u64 z) { return (x & y) ^ (x & z) ^ (y & z); }
+static u64 Sigma0(u64 x) { return R(x,28) ^ R(x,34) ^ R(x,39); }
+static u64 Sigma1(u64 x) { return R(x,14) ^ R(x,18) ^ R(x,41); }
+static u64 sigma0(u64 x) { return R(x, 1) ^ R(x, 8) ^ (x >> 7); }
+static u64 sigma1(u64 x) { return R(x,19) ^ R(x,61) ^ (x >> 6); }
+
+static const u64 K[80] = 
+{
+  0x428a2f98d728ae22ULL, 0x7137449123ef65cdULL, 0xb5c0fbcfec4d3b2fULL, 0xe9b5dba58189dbbcULL,
+  0x3956c25bf348b538ULL, 0x59f111f1b605d019ULL, 0x923f82a4af194f9bULL, 0xab1c5ed5da6d8118ULL,
+  0xd807aa98a3030242ULL, 0x12835b0145706fbeULL, 0x243185be4ee4b28cULL, 0x550c7dc3d5ffb4e2ULL,
+  0x72be5d74f27b896fULL, 0x80deb1fe3b1696b1ULL, 0x9bdc06a725c71235ULL, 0xc19bf174cf692694ULL,
+  0xe49b69c19ef14ad2ULL, 0xefbe4786384f25e3ULL, 0x0fc19dc68b8cd5b5ULL, 0x240ca1cc77ac9c65ULL,
+  0x2de92c6f592b0275ULL, 0x4a7484aa6ea6e483ULL, 0x5cb0a9dcbd41fbd4ULL, 0x76f988da831153b5ULL,
+  0x983e5152ee66dfabULL, 0xa831c66d2db43210ULL, 0xb00327c898fb213fULL, 0xbf597fc7beef0ee4ULL,
+  0xc6e00bf33da88fc2ULL, 0xd5a79147930aa725ULL, 0x06ca6351e003826fULL, 0x142929670a0e6e70ULL,
+  0x27b70a8546d22ffcULL, 0x2e1b21385c26c926ULL, 0x4d2c6dfc5ac42aedULL, 0x53380d139d95b3dfULL,
+  0x650a73548baf63deULL, 0x766a0abb3c77b2a8ULL, 0x81c2c92e47edaee6ULL, 0x92722c851482353bULL,
+  0xa2bfe8a14cf10364ULL, 0xa81a664bbc423001ULL, 0xc24b8b70d0f89791ULL, 0xc76c51a30654be30ULL,
+  0xd192e819d6ef5218ULL, 0xd69906245565a910ULL, 0xf40e35855771202aULL, 0x106aa07032bbd1b8ULL,
+  0x19a4c116b8d2d0c8ULL, 0x1e376c085141ab53ULL, 0x2748774cdf8eeb99ULL, 0x34b0bcb5e19b48a8ULL,
+  0x391c0cb3c5c95a63ULL, 0x4ed8aa4ae3418acbULL, 0x5b9cca4f7763e373ULL, 0x682e6ff3d6b2b8a3ULL,
+  0x748f82ee5defb2fcULL, 0x78a5636f43172f60ULL, 0x84c87814a1f0ab72ULL, 0x8cc702081a6439ecULL,
+  0x90befffa23631e28ULL, 0xa4506cebde82bde9ULL, 0xbef9a3f7b2c67915ULL, 0xc67178f2e372532bULL,
+  0xca273eceea26619cULL, 0xd186b8c721c0c207ULL, 0xeada7dd6cde0eb1eULL, 0xf57d4f7fee6ed178ULL,
+  0x06f067aa72176fbaULL, 0x0a637dc5a2c898a6ULL, 0x113f9804bef90daeULL, 0x1b710b35131c471bULL,
+  0x28db77f523047d84ULL, 0x32caab7b40c72493ULL, 0x3c9ebe0a15c9bebcULL, 0x431d67c49c100d4cULL,
+  0x4cc5d4becb3e42b6ULL, 0x597f299cfc657e2aULL, 0x5fcb6fab3ad6faecULL, 0x6c44198c4a475817ULL
+};
+
+int crypto_hashblocks(u8 *x,const u8 *m,u64 n)
+{
+  u64 z[8],b[8],a[8],w[16],t;
+  int i,j;
+
+  FOR(i,8) z[i] = a[i] = dl64(x + 8 * i);
+
+  while (n >= 128) {
+    FOR(i,16) w[i] = dl64(m + 8 * i);
+
+    FOR(i,80) {
+      FOR(j,8) b[j] = a[j];
+      t = a[7] + Sigma1(a[4]) + Ch(a[4],a[5],a[6]) + K[i] + w[i%16];
+      b[7] = t + Sigma0(a[0]) + Maj(a[0],a[1],a[2]);
+      b[3] += t;
+      FOR(j,8) a[(j+1)%8] = b[j];
+      if (i%16 == 15)
+	FOR(j,16)
+	  w[j] += w[(j+9)%16] + sigma0(w[(j+1)%16]) + sigma1(w[(j+14)%16]);
+    }
+
+    FOR(i,8) { a[i] += z[i]; z[i] = a[i]; }
+
+    m += 128;
+    n -= 128;
+  }
+
+  FOR(i,8) ts64(x+8*i,z[i]);
+
+  return n;
+}
+
+static const u8 iv[64] = {
+  0x6a,0x09,0xe6,0x67,0xf3,0xbc,0xc9,0x08,
+  0xbb,0x67,0xae,0x85,0x84,0xca,0xa7,0x3b,
+  0x3c,0x6e,0xf3,0x72,0xfe,0x94,0xf8,0x2b,
+  0xa5,0x4f,0xf5,0x3a,0x5f,0x1d,0x36,0xf1,
+  0x51,0x0e,0x52,0x7f,0xad,0xe6,0x82,0xd1,
+  0x9b,0x05,0x68,0x8c,0x2b,0x3e,0x6c,0x1f,
+  0x1f,0x83,0xd9,0xab,0xfb,0x41,0xbd,0x6b,
+  0x5b,0xe0,0xcd,0x19,0x13,0x7e,0x21,0x79
+} ;
+
+int crypto_hash(u8 *out,const u8 *m,u64 n)
+{
+  u8 h[64],x[256];
+  u64 i,b = n;
+
+  FOR(i,64) h[i] = iv[i];
+
+  crypto_hashblocks(h,m,n);
+  m += n;
+  n &= 127;
+  m -= n;
+
+  FOR(i,256) x[i] = 0;
+  FOR(i,n) x[i] = m[i];
+  x[n] = 128;
+
+  n = 256-128*(n<112);
+  x[n-9] = b >> 61;
+  ts64(x+n-8,b<<3);
+  crypto_hashblocks(h,x,n);
+
+  FOR(i,64) out[i] = h[i];
+
+  return 0;
+}
+
+sv add(gf p[4],gf q[4])
+{
+  gf a,b,c,d,t,e,f,g,h;
+  
+  Z(a, p[1], p[0]);
+  Z(t, q[1], q[0]);
+  M(a, a, t);
+  A(b, p[0], p[1]);
+  A(t, q[0], q[1]);
+  M(b, b, t);
+  M(c, p[3], q[3]);
+  M(c, c, D2);
+  M(d, p[2], q[2]);
+  A(d, d, d);
+  Z(e, b, a);
+  Z(f, d, c);
+  A(g, d, c);
+  A(h, b, a);
+
+  M(p[0], e, f);
+  M(p[1], h, g);
+  M(p[2], g, f);
+  M(p[3], e, h);
+}
+
+sv cswap(gf p[4],gf q[4],u8 b)
+{
+  int i;
+  FOR(i,4)
+    sel25519(p[i],q[i],b);
+}
+
+sv pack(u8 *r,gf p[4])
+{
+  gf tx, ty, zi;
+  inv25519(zi, p[2]); 
+  M(tx, p[0], zi);
+  M(ty, p[1], zi);
+  pack25519(r, ty);
+  r[31] ^= par25519(tx) << 7;
+}
+
+sv scalarmult(gf p[4],gf q[4],const u8 *s)
+{
+  int i;
+  set25519(p[0],gf0);
+  set25519(p[1],gf1);
+  set25519(p[2],gf1);
+  set25519(p[3],gf0);
+  for (i = 255;i >= 0;--i) {
+    u8 b = (s[i/8]>>(i&7))&1;
+    cswap(p,q,b);
+    add(q,p);
+    add(p,p);
+    cswap(p,q,b);
+  }
+}
+
+sv scalarbase(gf p[4],const u8 *s)
+{
+  gf q[4];
+  set25519(q[0],X);
+  set25519(q[1],Y);
+  set25519(q[2],gf1);
+  M(q[3],X,Y);
+  scalarmult(p,q,s);
+}
+
+int crypto_sign_keypair(u8 *pk, u8 *sk)
+{
+  u8 d[64];
+  gf p[4];
+  int i;
+
+  randombytes(sk, 32);
+  crypto_hash(d, sk, 32);
+  d[0] &= 248;
+  d[31] &= 127;
+  d[31] |= 64;
+
+  scalarbase(p,d);
+  pack(pk,p);
+
+  FOR(i,32) sk[32 + i] = pk[i];
+  return 0;
+}
+
+static const u64 L[32] = {0xed, 0xd3, 0xf5, 0x5c, 0x1a, 0x63, 0x12, 0x58, 0xd6, 0x9c, 0xf7, 0xa2, 0xde, 0xf9, 0xde, 0x14, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x10};
+
+sv modL(u8 *r,i64 x[64])
+{
+  i64 carry,i,j;
+  for (i = 63;i >= 32;--i) {
+    carry = 0;
+    for (j = i - 32;j < i - 12;++j) {
+      x[j] += carry - 16 * x[i] * L[j - (i - 32)];
+      carry = (x[j] + 128) >> 8;
+      x[j] -= carry << 8;
+    }
+    x[j] += carry;
+    x[i] = 0;
+  }
+  carry = 0;
+  FOR(j,32) {
+    x[j] += carry - (x[31] >> 4) * L[j];
+    carry = x[j] >> 8;
+    x[j] &= 255;
+  }
+  FOR(j,32) x[j] -= carry * L[j];
+  FOR(i,32) {
+    x[i+1] += x[i] >> 8;
+    r[i] = x[i] & 255;
+  }
+}
+
+sv reduce(u8 *r)
+{
+  i64 x[64],i;
+  FOR(i,64) x[i] = (u64) r[i];
+  FOR(i,64) r[i] = 0;
+  modL(r,x);
+}
+
+int crypto_sign(u8 *sm,u64 *smlen,const u8 *m,u64 n,const u8 *sk)
+{
+  u8 d[64],h[64],r[64];
+  i64 i,j,x[64];
+  gf p[4];
+
+  crypto_hash(d, sk, 32);
+  d[0] &= 248;
+  d[31] &= 127;
+  d[31] |= 64;
+
+  *smlen = n+64;
+  FOR(i,n) sm[64 + i] = m[i];
+  FOR(i,32) sm[32 + i] = d[32 + i];
+
+  crypto_hash(r, sm+32, n+32);
+  reduce(r);
+  scalarbase(p,r);
+  pack(sm,p);
+
+  FOR(i,32) sm[i+32] = sk[i+32];
+  crypto_hash(h,sm,n + 64);
+  reduce(h);
+
+  FOR(i,64) x[i] = 0;
+  FOR(i,32) x[i] = (u64) r[i];
+  FOR(i,32) FOR(j,32) x[i+j] += h[i] * (u64) d[j];
+  modL(sm + 32,x);
+
+  return 0;
+}
+
+static int unpackneg(gf r[4],const u8 p[32])
+{
+  gf t, chk, num, den, den2, den4, den6;
+  set25519(r[2],gf1);
+  unpack25519(r[1],p);
+  S(num,r[1]);
+  M(den,num,D);
+  Z(num,num,r[2]);
+  A(den,r[2],den);
+
+  S(den2,den);
+  S(den4,den2);
+  M(den6,den4,den2);
+  M(t,den6,num);
+  M(t,t,den);
+
+  pow2523(t,t);
+  M(t,t,num);
+  M(t,t,den);
+  M(t,t,den);
+  M(r[0],t,den);
+
+  S(chk,r[0]);
+  M(chk,chk,den);
+  if (neq25519(chk, num)) M(r[0],r[0],I);
+
+  S(chk,r[0]);
+  M(chk,chk,den);
+  if (neq25519(chk, num)) return -1;
+
+  if (par25519(r[0]) == (p[31]>>7)) Z(r[0],gf0,r[0]);
+
+  M(r[3],r[0],r[1]);
+  return 0;
+}
+
+int crypto_sign_open(u8 *m,u64 *mlen,const u8 *sm,u64 n,const u8 *pk)
+{
+  int i;
+  u8 t[32],h[64];
+  gf p[4],q[4];
+
+  *mlen = -1;
+  if (n < 64) return -1;
+
+  if (unpackneg(q,pk)) return -1;
+
+  FOR(i,n) m[i] = sm[i];
+  FOR(i,32) m[i+32] = pk[i];
+  crypto_hash(h,m,n);
+  reduce(h);
+  scalarmult(p,q,h);
+
+  scalarbase(q,sm + 32);
+  add(p,q);
+  pack(t,p);
+
+  n -= 64;
+  if (crypto_verify_32(sm, t)) {
+    FOR(i,n) m[i] = 0;
+    return -1;
+  }
+
+  FOR(i,n) m[i] = sm[i + 64];
+  *mlen = n;
+  return 0;
+}

cbits/tweetnacl.h

@@ -0,0 +1,272 @@
+#ifndef TWEETNACL_H
+#define TWEETNACL_H
+#define crypto_auth_PRIMITIVE "hmacsha512256"
+#define crypto_auth crypto_auth_hmacsha512256
+#define crypto_auth_verify crypto_auth_hmacsha512256_verify
+#define crypto_auth_BYTES crypto_auth_hmacsha512256_BYTES
+#define crypto_auth_KEYBYTES crypto_auth_hmacsha512256_KEYBYTES
+#define crypto_auth_IMPLEMENTATION crypto_auth_hmacsha512256_IMPLEMENTATION
+#define crypto_auth_VERSION crypto_auth_hmacsha512256_VERSION
+#define crypto_auth_hmacsha512256_tweet_BYTES 32
+#define crypto_auth_hmacsha512256_tweet_KEYBYTES 32
+extern int crypto_auth_hmacsha512256_tweet(unsigned char *,const unsigned char *,unsigned long long,const unsigned char *);
+extern int crypto_auth_hmacsha512256_tweet_verify(const unsigned char *,const unsigned char *,unsigned long long,const unsigned char *);
+#define crypto_auth_hmacsha512256_tweet_VERSION "-"
+#define crypto_auth_hmacsha512256 crypto_auth_hmacsha512256_tweet
+#define crypto_auth_hmacsha512256_verify crypto_auth_hmacsha512256_tweet_verify
+#define crypto_auth_hmacsha512256_BYTES crypto_auth_hmacsha512256_tweet_BYTES
+#define crypto_auth_hmacsha512256_KEYBYTES crypto_auth_hmacsha512256_tweet_KEYBYTES
+#define crypto_auth_hmacsha512256_VERSION crypto_auth_hmacsha512256_tweet_VERSION
+#define crypto_auth_hmacsha512256_IMPLEMENTATION "crypto_auth/hmacsha512256/tweet"
+#define crypto_box_PRIMITIVE "curve25519xsalsa20poly1305"
+#define crypto_box crypto_box_curve25519xsalsa20poly1305
+#define crypto_box_open crypto_box_curve25519xsalsa20poly1305_open
+#define crypto_box_keypair crypto_box_curve25519xsalsa20poly1305_keypair
+#define crypto_box_beforenm crypto_box_curve25519xsalsa20poly1305_beforenm
+#define crypto_box_afternm crypto_box_curve25519xsalsa20poly1305_afternm
+#define crypto_box_open_afternm crypto_box_curve25519xsalsa20poly1305_open_afternm
+#define crypto_box_PUBLICKEYBYTES crypto_box_curve25519xsalsa20poly1305_PUBLICKEYBYTES
+#define crypto_box_SECRETKEYBYTES crypto_box_curve25519xsalsa20poly1305_SECRETKEYBYTES
+#define crypto_box_BEFORENMBYTES crypto_box_curve25519xsalsa20poly1305_BEFORENMBYTES
+#define crypto_box_NONCEBYTES crypto_box_curve25519xsalsa20poly1305_NONCEBYTES
+#define crypto_box_ZEROBYTES crypto_box_curve25519xsalsa20poly1305_ZEROBYTES
+#define crypto_box_BOXZEROBYTES crypto_box_curve25519xsalsa20poly1305_BOXZEROBYTES
+#define crypto_box_IMPLEMENTATION crypto_box_curve25519xsalsa20poly1305_IMPLEMENTATION
+#define crypto_box_VERSION crypto_box_curve25519xsalsa20poly1305_VERSION
+#define crypto_box_curve25519xsalsa20poly1305_tweet_PUBLICKEYBYTES 32
+#define crypto_box_curve25519xsalsa20poly1305_tweet_SECRETKEYBYTES 32
+#define crypto_box_curve25519xsalsa20poly1305_tweet_BEFORENMBYTES 32
+#define crypto_box_curve25519xsalsa20poly1305_tweet_NONCEBYTES 24
+#define crypto_box_curve25519xsalsa20poly1305_tweet_ZEROBYTES 32
+#define crypto_box_curve25519xsalsa20poly1305_tweet_BOXZEROBYTES 16
+extern int crypto_box_curve25519xsalsa20poly1305_tweet(unsigned char *,const unsigned char *,unsigned long long,const unsigned char *,const unsigned char *,const unsigned char *);
+extern int crypto_box_curve25519xsalsa20poly1305_tweet_open(unsigned char *,const unsigned char *,unsigned long long,const unsigned char *,const unsigned char *,const unsigned char *);
+extern int crypto_box_curve25519xsalsa20poly1305_tweet_keypair(unsigned char *,unsigned char *);
+extern int crypto_box_curve25519xsalsa20poly1305_tweet_beforenm(unsigned char *,const unsigned char *,const unsigned char *);
+extern int crypto_box_curve25519xsalsa20poly1305_tweet_afternm(unsigned char *,const unsigned char *,unsigned long long,const unsigned char *,const unsigned char *);
+extern int crypto_box_curve25519xsalsa20poly1305_tweet_open_afternm(unsigned char *,const unsigned char *,unsigned long long,const unsigned char *,const unsigned char *);
+#define crypto_box_curve25519xsalsa20poly1305_tweet_VERSION "-"
+#define crypto_box_curve25519xsalsa20poly1305 crypto_box_curve25519xsalsa20poly1305_tweet
+#define crypto_box_curve25519xsalsa20poly1305_open crypto_box_curve25519xsalsa20poly1305_tweet_open
+#define crypto_box_curve25519xsalsa20poly1305_keypair crypto_box_curve25519xsalsa20poly1305_tweet_keypair
+#define crypto_box_curve25519xsalsa20poly1305_beforenm crypto_box_curve25519xsalsa20poly1305_tweet_beforenm
+#define crypto_box_curve25519xsalsa20poly1305_afternm crypto_box_curve25519xsalsa20poly1305_tweet_afternm
+#define crypto_box_curve25519xsalsa20poly1305_open_afternm crypto_box_curve25519xsalsa20poly1305_tweet_open_afternm
+#define crypto_box_curve25519xsalsa20poly1305_PUBLICKEYBYTES crypto_box_curve25519xsalsa20poly1305_tweet_PUBLICKEYBYTES
+#define crypto_box_curve25519xsalsa20poly1305_SECRETKEYBYTES crypto_box_curve25519xsalsa20poly1305_tweet_SECRETKEYBYTES
+#define crypto_box_curve25519xsalsa20poly1305_BEFORENMBYTES crypto_box_curve25519xsalsa20poly1305_tweet_BEFORENMBYTES
+#define crypto_box_curve25519xsalsa20poly1305_NONCEBYTES crypto_box_curve25519xsalsa20poly1305_tweet_NONCEBYTES
+#define crypto_box_curve25519xsalsa20poly1305_ZEROBYTES crypto_box_curve25519xsalsa20poly1305_tweet_ZEROBYTES
+#define crypto_box_curve25519xsalsa20poly1305_BOXZEROBYTES crypto_box_curve25519xsalsa20poly1305_tweet_BOXZEROBYTES
+#define crypto_box_curve25519xsalsa20poly1305_VERSION crypto_box_curve25519xsalsa20poly1305_tweet_VERSION
+#define crypto_box_curve25519xsalsa20poly1305_IMPLEMENTATION "crypto_box/curve25519xsalsa20poly1305/tweet"
+#define crypto_core_PRIMITIVE "salsa20"
+#define crypto_core crypto_core_salsa20
+#define crypto_core_OUTPUTBYTES crypto_core_salsa20_OUTPUTBYTES
+#define crypto_core_INPUTBYTES crypto_core_salsa20_INPUTBYTES
+#define crypto_core_KEYBYTES crypto_core_salsa20_KEYBYTES
+#define crypto_core_CONSTBYTES crypto_core_salsa20_CONSTBYTES
+#define crypto_core_IMPLEMENTATION crypto_core_salsa20_IMPLEMENTATION
+#define crypto_core_VERSION crypto_core_salsa20_VERSION
+#define crypto_core_salsa20_tweet_OUTPUTBYTES 64
+#define crypto_core_salsa20_tweet_INPUTBYTES 16
+#define crypto_core_salsa20_tweet_KEYBYTES 32
+#define crypto_core_salsa20_tweet_CONSTBYTES 16
+extern int crypto_core_salsa20_tweet(unsigned char *,const unsigned char *,const unsigned char *,const unsigned char *);
+#define crypto_core_salsa20_tweet_VERSION "-"
+#define crypto_core_salsa20 crypto_core_salsa20_tweet
+#define crypto_core_salsa20_OUTPUTBYTES crypto_core_salsa20_tweet_OUTPUTBYTES
+#define crypto_core_salsa20_INPUTBYTES crypto_core_salsa20_tweet_INPUTBYTES
+#define crypto_core_salsa20_KEYBYTES crypto_core_salsa20_tweet_KEYBYTES
+#define crypto_core_salsa20_CONSTBYTES crypto_core_salsa20_tweet_CONSTBYTES
+#define crypto_core_salsa20_VERSION crypto_core_salsa20_tweet_VERSION
+#define crypto_core_salsa20_IMPLEMENTATION "crypto_core/salsa20/tweet"
+#define crypto_core_hsalsa20_tweet_OUTPUTBYTES 32
+#define crypto_core_hsalsa20_tweet_INPUTBYTES 16
+#define crypto_core_hsalsa20_tweet_KEYBYTES 32
+#define crypto_core_hsalsa20_tweet_CONSTBYTES 16
+extern int crypto_core_hsalsa20_tweet(unsigned char *,const unsigned char *,const unsigned char *,const unsigned char *);
+#define crypto_core_hsalsa20_tweet_VERSION "-"
+#define crypto_core_hsalsa20 crypto_core_hsalsa20_tweet
+#define crypto_core_hsalsa20_OUTPUTBYTES crypto_core_hsalsa20_tweet_OUTPUTBYTES
+#define crypto_core_hsalsa20_INPUTBYTES crypto_core_hsalsa20_tweet_INPUTBYTES
+#define crypto_core_hsalsa20_KEYBYTES crypto_core_hsalsa20_tweet_KEYBYTES
+#define crypto_core_hsalsa20_CONSTBYTES crypto_core_hsalsa20_tweet_CONSTBYTES
+#define crypto_core_hsalsa20_VERSION crypto_core_hsalsa20_tweet_VERSION
+#define crypto_core_hsalsa20_IMPLEMENTATION "crypto_core/hsalsa20/tweet"
+#define crypto_hashblocks_PRIMITIVE "sha512"
+#define crypto_hashblocks crypto_hashblocks_sha512
+#define crypto_hashblocks_STATEBYTES crypto_hashblocks_sha512_STATEBYTES
+#define crypto_hashblocks_BLOCKBYTES crypto_hashblocks_sha512_BLOCKBYTES
+#define crypto_hashblocks_IMPLEMENTATION crypto_hashblocks_sha512_IMPLEMENTATION
+#define crypto_hashblocks_VERSION crypto_hashblocks_sha512_VERSION
+#define crypto_hashblocks_sha512_tweet_STATEBYTES 64
+#define crypto_hashblocks_sha512_tweet_BLOCKBYTES 128
+extern int crypto_hashblocks_sha512_tweet(unsigned char *,const unsigned char *,unsigned long long);
+#define crypto_hashblocks_sha512_tweet_VERSION "-"
+#define crypto_hashblocks_sha512 crypto_hashblocks_sha512_tweet
+#define crypto_hashblocks_sha512_STATEBYTES crypto_hashblocks_sha512_tweet_STATEBYTES
+#define crypto_hashblocks_sha512_BLOCKBYTES crypto_hashblocks_sha512_tweet_BLOCKBYTES
+#define crypto_hashblocks_sha512_VERSION crypto_hashblocks_sha512_tweet_VERSION
+#define crypto_hashblocks_sha512_IMPLEMENTATION "crypto_hashblocks/sha512/tweet"
+#define crypto_hashblocks_sha256_tweet_STATEBYTES 32
+#define crypto_hashblocks_sha256_tweet_BLOCKBYTES 64
+extern int crypto_hashblocks_sha256_tweet(unsigned char *,const unsigned char *,unsigned long long);
+#define crypto_hashblocks_sha256_tweet_VERSION "-"
+#define crypto_hashblocks_sha256 crypto_hashblocks_sha256_tweet
+#define crypto_hashblocks_sha256_STATEBYTES crypto_hashblocks_sha256_tweet_STATEBYTES
+#define crypto_hashblocks_sha256_BLOCKBYTES crypto_hashblocks_sha256_tweet_BLOCKBYTES
+#define crypto_hashblocks_sha256_VERSION crypto_hashblocks_sha256_tweet_VERSION
+#define crypto_hashblocks_sha256_IMPLEMENTATION "crypto_hashblocks/sha256/tweet"
+#define crypto_hash_PRIMITIVE "sha512"
+#define crypto_hash crypto_hash_sha512
+#define crypto_hash_BYTES crypto_hash_sha512_BYTES
+#define crypto_hash_IMPLEMENTATION crypto_hash_sha512_IMPLEMENTATION
+#define crypto_hash_VERSION crypto_hash_sha512_VERSION
+#define crypto_hash_sha512_tweet_BYTES 64
+extern int crypto_hash_sha512_tweet(unsigned char *,const unsigned char *,unsigned long long);
+#define crypto_hash_sha512_tweet_VERSION "-"
+#define crypto_hash_sha512 crypto_hash_sha512_tweet
+#define crypto_hash_sha512_BYTES crypto_hash_sha512_tweet_BYTES
+#define crypto_hash_sha512_VERSION crypto_hash_sha512_tweet_VERSION
+#define crypto_hash_sha512_IMPLEMENTATION "crypto_hash/sha512/tweet"
+#define crypto_hash_sha256_tweet_BYTES 32
+extern int crypto_hash_sha256_tweet(unsigned char *,const unsigned char *,unsigned long long);
+#define crypto_hash_sha256_tweet_VERSION "-"
+#define crypto_hash_sha256 crypto_hash_sha256_tweet
+#define crypto_hash_sha256_BYTES crypto_hash_sha256_tweet_BYTES
+#define crypto_hash_sha256_VERSION crypto_hash_sha256_tweet_VERSION
+#define crypto_hash_sha256_IMPLEMENTATION "crypto_hash/sha256/tweet"
+#define crypto_onetimeauth_PRIMITIVE "poly1305"
+#define crypto_onetimeauth crypto_onetimeauth_poly1305
+#define crypto_onetimeauth_verify crypto_onetimeauth_poly1305_verify
+#define crypto_onetimeauth_BYTES crypto_onetimeauth_poly1305_BYTES
+#define crypto_onetimeauth_KEYBYTES crypto_onetimeauth_poly1305_KEYBYTES
+#define crypto_onetimeauth_IMPLEMENTATION crypto_onetimeauth_poly1305_IMPLEMENTATION
+#define crypto_onetimeauth_VERSION crypto_onetimeauth_poly1305_VERSION
+#define crypto_onetimeauth_poly1305_tweet_BYTES 16
+#define crypto_onetimeauth_poly1305_tweet_KEYBYTES 32
+extern int crypto_onetimeauth_poly1305_tweet(unsigned char *,const unsigned char *,unsigned long long,const unsigned char *);
+extern int crypto_onetimeauth_poly1305_tweet_verify(const unsigned char *,const unsigned char *,unsigned long long,const unsigned char *);
+#define crypto_onetimeauth_poly1305_tweet_VERSION "-"
+#define crypto_onetimeauth_poly1305 crypto_onetimeauth_poly1305_tweet
+#define crypto_onetimeauth_poly1305_verify crypto_onetimeauth_poly1305_tweet_verify
+#define crypto_onetimeauth_poly1305_BYTES crypto_onetimeauth_poly1305_tweet_BYTES
+#define crypto_onetimeauth_poly1305_KEYBYTES crypto_onetimeauth_poly1305_tweet_KEYBYTES
+#define crypto_onetimeauth_poly1305_VERSION crypto_onetimeauth_poly1305_tweet_VERSION
+#define crypto_onetimeauth_poly1305_IMPLEMENTATION "crypto_onetimeauth/poly1305/tweet"
+#define crypto_scalarmult_PRIMITIVE "curve25519"
+#define crypto_scalarmult crypto_scalarmult_curve25519
+#define crypto_scalarmult_base crypto_scalarmult_curve25519_base
+#define crypto_scalarmult_BYTES crypto_scalarmult_curve25519_BYTES
+#define crypto_scalarmult_SCALARBYTES crypto_scalarmult_curve25519_SCALARBYTES
+#define crypto_scalarmult_IMPLEMENTATION crypto_scalarmult_curve25519_IMPLEMENTATION
+#define crypto_scalarmult_VERSION crypto_scalarmult_curve25519_VERSION
+#define crypto_scalarmult_curve25519_tweet_BYTES 32
+#define crypto_scalarmult_curve25519_tweet_SCALARBYTES 32
+extern int crypto_scalarmult_curve25519_tweet(unsigned char *,const unsigned char *,const unsigned char *);
+extern int crypto_scalarmult_curve25519_tweet_base(unsigned char *,const unsigned char *);
+#define crypto_scalarmult_curve25519_tweet_VERSION "-"
+#define crypto_scalarmult_curve25519 crypto_scalarmult_curve25519_tweet
+#define crypto_scalarmult_curve25519_base crypto_scalarmult_curve25519_tweet_base
+#define crypto_scalarmult_curve25519_BYTES crypto_scalarmult_curve25519_tweet_BYTES
+#define crypto_scalarmult_curve25519_SCALARBYTES crypto_scalarmult_curve25519_tweet_SCALARBYTES
+#define crypto_scalarmult_curve25519_VERSION crypto_scalarmult_curve25519_tweet_VERSION
+#define crypto_scalarmult_curve25519_IMPLEMENTATION "crypto_scalarmult/curve25519/tweet"
+#define crypto_secretbox_PRIMITIVE "xsalsa20poly1305"
+#define crypto_secretbox crypto_secretbox_xsalsa20poly1305
+#define crypto_secretbox_open crypto_secretbox_xsalsa20poly1305_open
+#define crypto_secretbox_KEYBYTES crypto_secretbox_xsalsa20poly1305_KEYBYTES
+#define crypto_secretbox_NONCEBYTES crypto_secretbox_xsalsa20poly1305_NONCEBYTES
+#define crypto_secretbox_ZEROBYTES crypto_secretbox_xsalsa20poly1305_ZEROBYTES
+#define crypto_secretbox_BOXZEROBYTES crypto_secretbox_xsalsa20poly1305_BOXZEROBYTES
+#define crypto_secretbox_IMPLEMENTATION crypto_secretbox_xsalsa20poly1305_IMPLEMENTATION
+#define crypto_secretbox_VERSION crypto_secretbox_xsalsa20poly1305_VERSION
+#define crypto_secretbox_xsalsa20poly1305_tweet_KEYBYTES 32
+#define crypto_secretbox_xsalsa20poly1305_tweet_NONCEBYTES 24
+#define crypto_secretbox_xsalsa20poly1305_tweet_ZEROBYTES 32
+#define crypto_secretbox_xsalsa20poly1305_tweet_BOXZEROBYTES 16
+extern int crypto_secretbox_xsalsa20poly1305_tweet(unsigned char *,const unsigned char *,unsigned long long,const unsigned char *,const unsigned char *);
+extern int crypto_secretbox_xsalsa20poly1305_tweet_open(unsigned char *,const unsigned char *,unsigned long long,const unsigned char *,const unsigned char *);
+#define crypto_secretbox_xsalsa20poly1305_tweet_VERSION "-"
+#define crypto_secretbox_xsalsa20poly1305 crypto_secretbox_xsalsa20poly1305_tweet
+#define crypto_secretbox_xsalsa20poly1305_open crypto_secretbox_xsalsa20poly1305_tweet_open
+#define crypto_secretbox_xsalsa20poly1305_KEYBYTES crypto_secretbox_xsalsa20poly1305_tweet_KEYBYTES
+#define crypto_secretbox_xsalsa20poly1305_NONCEBYTES crypto_secretbox_xsalsa20poly1305_tweet_NONCEBYTES
+#define crypto_secretbox_xsalsa20poly1305_ZEROBYTES crypto_secretbox_xsalsa20poly1305_tweet_ZEROBYTES
+#define crypto_secretbox_xsalsa20poly1305_BOXZEROBYTES crypto_secretbox_xsalsa20poly1305_tweet_BOXZEROBYTES
+#define crypto_secretbox_xsalsa20poly1305_VERSION crypto_secretbox_xsalsa20poly1305_tweet_VERSION
+#define crypto_secretbox_xsalsa20poly1305_IMPLEMENTATION "crypto_secretbox/xsalsa20poly1305/tweet"
+#define crypto_sign_PRIMITIVE "ed25519"
+#define crypto_sign crypto_sign_ed25519
+#define crypto_sign_open crypto_sign_ed25519_open
+#define crypto_sign_keypair crypto_sign_ed25519_keypair
+#define crypto_sign_BYTES crypto_sign_ed25519_BYTES
+#define crypto_sign_PUBLICKEYBYTES crypto_sign_ed25519_PUBLICKEYBYTES
+#define crypto_sign_SECRETKEYBYTES crypto_sign_ed25519_SECRETKEYBYTES
+#define crypto_sign_IMPLEMENTATION crypto_sign_ed25519_IMPLEMENTATION
+#define crypto_sign_VERSION crypto_sign_ed25519_VERSION
+#define crypto_sign_ed25519_tweet_BYTES 64
+#define crypto_sign_ed25519_tweet_PUBLICKEYBYTES 32
+#define crypto_sign_ed25519_tweet_SECRETKEYBYTES 64
+extern int crypto_sign_ed25519_tweet(unsigned char *,unsigned long long *,const unsigned char *,unsigned long long,const unsigned char *);
+extern int crypto_sign_ed25519_tweet_open(unsigned char *,unsigned long long *,const unsigned char *,unsigned long long,const unsigned char *);
+extern int crypto_sign_ed25519_tweet_keypair(unsigned char *,unsigned char *);
+#define crypto_sign_ed25519_tweet_VERSION "-"
+#define crypto_sign_ed25519 crypto_sign_ed25519_tweet
+#define crypto_sign_ed25519_open crypto_sign_ed25519_tweet_open
+#define crypto_sign_ed25519_keypair crypto_sign_ed25519_tweet_keypair
+#define crypto_sign_ed25519_BYTES crypto_sign_ed25519_tweet_BYTES
+#define crypto_sign_ed25519_PUBLICKEYBYTES crypto_sign_ed25519_tweet_PUBLICKEYBYTES
+#define crypto_sign_ed25519_SECRETKEYBYTES crypto_sign_ed25519_tweet_SECRETKEYBYTES
+#define crypto_sign_ed25519_VERSION crypto_sign_ed25519_tweet_VERSION
+#define crypto_sign_ed25519_IMPLEMENTATION "crypto_sign/ed25519/tweet"
+#define crypto_stream_PRIMITIVE "xsalsa20"
+#define crypto_stream crypto_stream_xsalsa20
+#define crypto_stream_xor crypto_stream_xsalsa20_xor
+#define crypto_stream_KEYBYTES crypto_stream_xsalsa20_KEYBYTES
+#define crypto_stream_NONCEBYTES crypto_stream_xsalsa20_NONCEBYTES
+#define crypto_stream_IMPLEMENTATION crypto_stream_xsalsa20_IMPLEMENTATION
+#define crypto_stream_VERSION crypto_stream_xsalsa20_VERSION
+#define crypto_stream_xsalsa20_tweet_KEYBYTES 32
+#define crypto_stream_xsalsa20_tweet_NONCEBYTES 24
+extern int crypto_stream_xsalsa20_tweet(unsigned char *,unsigned long long,const unsigned char *,const unsigned char *);
+extern int crypto_stream_xsalsa20_tweet_xor(unsigned char *,const unsigned char *,unsigned long long,const unsigned char *,const unsigned char *);
+#define crypto_stream_xsalsa20_tweet_VERSION "-"
+#define crypto_stream_xsalsa20 crypto_stream_xsalsa20_tweet
+#define crypto_stream_xsalsa20_xor crypto_stream_xsalsa20_tweet_xor
+#define crypto_stream_xsalsa20_KEYBYTES crypto_stream_xsalsa20_tweet_KEYBYTES
+#define crypto_stream_xsalsa20_NONCEBYTES crypto_stream_xsalsa20_tweet_NONCEBYTES
+#define crypto_stream_xsalsa20_VERSION crypto_stream_xsalsa20_tweet_VERSION
+#define crypto_stream_xsalsa20_IMPLEMENTATION "crypto_stream/xsalsa20/tweet"
+#define crypto_stream_salsa20_tweet_KEYBYTES 32
+#define crypto_stream_salsa20_tweet_NONCEBYTES 8
+extern int crypto_stream_salsa20_tweet(unsigned char *,unsigned long long,const unsigned char *,const unsigned char *);
+extern int crypto_stream_salsa20_tweet_xor(unsigned char *,const unsigned char *,unsigned long long,const unsigned char *,const unsigned char *);
+#define crypto_stream_salsa20_tweet_VERSION "-"
+#define crypto_stream_salsa20 crypto_stream_salsa20_tweet
+#define crypto_stream_salsa20_xor crypto_stream_salsa20_tweet_xor
+#define crypto_stream_salsa20_KEYBYTES crypto_stream_salsa20_tweet_KEYBYTES
+#define crypto_stream_salsa20_NONCEBYTES crypto_stream_salsa20_tweet_NONCEBYTES
+#define crypto_stream_salsa20_VERSION crypto_stream_salsa20_tweet_VERSION
+#define crypto_stream_salsa20_IMPLEMENTATION "crypto_stream/salsa20/tweet"
+#define crypto_verify_PRIMITIVE "16"
+#define crypto_verify crypto_verify_16
+#define crypto_verify_BYTES crypto_verify_16_BYTES
+#define crypto_verify_IMPLEMENTATION crypto_verify_16_IMPLEMENTATION
+#define crypto_verify_VERSION crypto_verify_16_VERSION
+#define crypto_verify_16_tweet_BYTES 16
+extern int crypto_verify_16_tweet(const unsigned char *,const unsigned char *);
+#define crypto_verify_16_tweet_VERSION "-"
+#define crypto_verify_16 crypto_verify_16_tweet
+#define crypto_verify_16_BYTES crypto_verify_16_tweet_BYTES
+#define crypto_verify_16_VERSION crypto_verify_16_tweet_VERSION
+#define crypto_verify_16_IMPLEMENTATION "crypto_verify/16/tweet"
+#define crypto_verify_32_tweet_BYTES 32
+extern int crypto_verify_32_tweet(const unsigned char *,const unsigned char *);
+#define crypto_verify_32_tweet_VERSION "-"
+#define crypto_verify_32 crypto_verify_32_tweet
+#define crypto_verify_32_BYTES crypto_verify_32_tweet_BYTES
+#define crypto_verify_32_VERSION crypto_verify_32_tweet_VERSION
+#define crypto_verify_32_IMPLEMENTATION "crypto_verify/32/tweet"
+#endif

package.yaml

@@ -22,6 +22,7 @@ extra-source-files:
   - CHANGELOG.md
   - cbits/sha512.h
   - cbits/sntrup761.h
+  - cbits/tweetnacl.h
   - apps/smp-server/static/*.html
   - apps/smp-server/static/media/*
 
@@ -44,6 +45,7 @@ dependencies:
   - 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.*
@@ -110,6 +112,7 @@ library:
   c-sources:
     - cbits/sha512.c
     - cbits/sntrup761.c
+    - cbits/tweetnacl.c
   include-dirs: cbits
   extra-libraries: crypto
 
@@ -171,7 +174,6 @@ tests:
     main: Test.hs
     dependencies:
       - simplexmq
-      - deepseq == 1.4.*
       - generic-random == 1.5.*
       - hspec == 2.11.*
       - hspec-core == 2.11.*
@@ -186,6 +188,31 @@ tests:
       - -with-rtsopts=-A64M
       - -with-rtsopts=-N1
 
+benchmarks:
+  simplexmq-bench:
+    source-dirs: benchmarks
+    main: Bench.hs
+    dependencies:
+      - base64 >= 1.0
+      - base64-bytestring
+      - containers
+      - hashable == 1.4.*
+      - hspec
+      - simplexmq
+      - tasty
+      - tasty-bench
+      - tasty-hspec
+      - unliftio
+      - unordered-containers
+      - zstd
+    ghc-options:
+      - -fproc-alignment=64
+      - -rtsopts
+      - -threaded
+      - -with-rtsopts=-A64m
+      - -with-rtsopts=-N1
+      - -with-rtsopts=-T
+
 ghc-options:
   # - -haddock
   - -Weverything

simplexmq.cabal

@@ -28,6 +28,7 @@ extra-source-files:
     CHANGELOG.md
     cbits/sha512.h
     cbits/sntrup761.h
+    cbits/tweetnacl.h
     apps/smp-server/static/index.html
     apps/smp-server/static/link.html
     apps/smp-server/static/media/apk_icon.png
@@ -140,6 +141,7 @@ library
       Simplex.Messaging.Crypto
       Simplex.Messaging.Crypto.File
       Simplex.Messaging.Crypto.Lazy
+      Simplex.Messaging.Crypto.NaCl.Bindings
       Simplex.Messaging.Crypto.Ratchet
       Simplex.Messaging.Crypto.SNTRUP761
       Simplex.Messaging.Crypto.SNTRUP761.Bindings
@@ -210,6 +212,7 @@ library
   c-sources:
       cbits/sha512.c
       cbits/sntrup761.c
+      cbits/tweetnacl.c
   extra-libraries:
       crypto
   build-depends:
@@ -231,6 +234,7 @@ library
     , 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.*
@@ -305,6 +309,7 @@ executable ntf-server
     , 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.*
@@ -380,6 +385,7 @@ executable smp-agent
     , 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.*
@@ -458,6 +464,7 @@ executable smp-server
     , crypton-x509-validation ==1.6.*
     , cryptostore ==0.3.*
     , data-default ==0.7.*
+    , deepseq ==1.4.*
     , direct-sqlcipher ==2.3.*
     , directory ==1.3.*
     , file-embed
@@ -537,6 +544,7 @@ executable xftp
     , 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.*
@@ -612,6 +620,7 @@ executable xftp-server
     , 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.*
@@ -775,3 +784,93 @@ test-suite simplexmq-test
         bytestring ==0.10.*
       , template-haskell ==2.16.*
       , text >=1.2.3.0 && <1.3
+
+benchmark simplexmq-bench
+  type: exitcode-stdio-1.0
+  main-is: Bench.hs
+  other-modules:
+      Bench.Base64
+      Bench.BsConcat
+      Bench.Compression
+      Bench.Crypto
+      Bench.SNTRUP761
+      Bench.TRcvQueues
+      Paths_simplexmq
+  hs-source-dirs:
+      benchmarks
+  default-extensions:
+      StrictData
+  ghc-options: -Weverything -Wno-missing-exported-signatures -Wno-missing-import-lists -Wno-missed-specialisations -Wno-all-missed-specialisations -Wno-unsafe -Wno-safe -Wno-missing-local-signatures -Wno-missing-kind-signatures -Wno-missing-deriving-strategies -Wno-monomorphism-restriction -Wno-prepositive-qualified-module -Wno-unused-packages -Wno-implicit-prelude -Wno-missing-safe-haskell-mode -Wno-missing-export-lists -Wno-partial-fields -Wcompat -Werror=incomplete-record-updates -Werror=incomplete-patterns -Werror=incomplete-uni-patterns -Werror=missing-methods -Werror=tabs -Wredundant-constraints -Wincomplete-record-updates -Wunused-type-patterns -O2 -fproc-alignment=64 -rtsopts -threaded -with-rtsopts=-A64m -with-rtsopts=-N1 -with-rtsopts=-T
+  build-depends:
+      aeson ==2.2.*
+    , ansi-terminal >=0.10 && <0.12
+    , asn1-encoding ==0.9.*
+    , asn1-types ==0.3.*
+    , async ==2.2.*
+    , attoparsec ==0.14.*
+    , base >=4.14 && <5
+    , base64 >=1.0
+    , base64-bytestring
+    , case-insensitive ==1.2.*
+    , composition ==1.0.*
+    , constraints >=0.12 && <0.14
+    , containers
+    , 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.*
+    , hashable ==1.4.*
+    , hourglass ==0.2.*
+    , hspec
+    , http-types ==0.12.*
+    , http2 >=4.2.2 && <4.3
+    , ini ==0.4.1
+    , iproute ==1.7.*
+    , iso8601-time ==0.1.*
+    , memory ==0.18.*
+    , mtl >=2.3.1 && <3.0
+    , network >=3.1.2.7 && <3.2
+    , 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
+    , simple-logger ==0.1.*
+    , simplexmq
+    , socks ==0.6.*
+    , sqlcipher-simple ==0.4.*
+    , stm ==2.5.*
+    , tasty
+    , tasty-bench
+    , tasty-hspec
+    , temporary ==1.3.*
+    , time ==1.12.*
+    , time-manager ==0.0.*
+    , tls >=1.7.0 && <1.8
+    , transformers ==0.6.*
+    , unliftio
+    , unliftio-core ==0.2.*
+    , unordered-containers
+    , websockets ==0.12.*
+    , yaml ==0.11.*
+    , zstd
+  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/Messaging/Agent/TRcvQueues.hs

@@ -1,4 +1,5 @@
 {-# LANGUAGE LambdaCase #-}
+{-# LANGUAGE StrictData #-}
 
 module Simplex.Messaging.Agent.TRcvQueues
   ( TRcvQueues (getRcvQueues, getConnections),
@@ -16,6 +17,7 @@ module Simplex.Messaging.Agent.TRcvQueues
 where
 
 import Control.Concurrent.STM
+import Control.DeepSeq (NFData (..))
 import Data.Foldable (foldl')
 import Data.List.NonEmpty (NonEmpty (..), (<|))
 import qualified Data.List.NonEmpty as L
@@ -33,6 +35,8 @@ data TRcvQueues = TRcvQueues
     getConnections :: TMap ConnId (NonEmpty (UserId, SMPServer, RecipientId))
   }
 
+instance NFData TRcvQueues where rnf TRcvQueues {} = ()
+
 empty :: STM TRcvQueues
 empty = TRcvQueues <$> TM.empty <*> TM.empty
 

src/Simplex/Messaging/Crypto.hs

@@ -185,6 +185,10 @@ module Simplex.Messaging.Crypto
     maxLenBS,
     unsafeMaxLenBS,
     appendMaxLenBS,
+    hsalsa20,
+    cryptoBox,
+    secretBox,
+    secretBoxOpen,
   )
 where
 
@@ -195,10 +199,8 @@ 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, hashDigestSize)
-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
@@ -228,12 +230,15 @@ import Data.X509
 import Data.X509.Validation (Fingerprint (..), getFingerprint)
 import Database.SQLite.Simple.FromField (FromField (..))
 import Database.SQLite.Simple.ToField (ToField (..))
+import Foreign.C.ConstPtr (ConstPtr (..))
 import GHC.TypeLits (ErrorMessage (..), KnownNat, Nat, TypeError, natVal, type (+))
 import Network.Transport.Internal (decodeWord16, encodeWord16)
+import qualified Simplex.Messaging.Crypto.NaCl.Bindings as NaCl
 import Simplex.Messaging.Encoding
 import Simplex.Messaging.Encoding.String
 import Simplex.Messaging.Parsers (blobFieldDecoder, parseAll, parseString)
 import Simplex.Messaging.Util ((<$?>))
+import System.IO.Unsafe (unsafePerformIO)
 
 -- | Cryptographic algorithms.
 data Algorithm = Ed25519 | Ed448 | X25519 | X448
@@ -1210,10 +1215,7 @@ cbEncryptMaxLenBS :: KnownNat i => DhSecret X25519 -> CbNonce -> MaxLenBS i -> B
 cbEncryptMaxLenBS (DhSecretX25519 secret) (CbNonce nonce) = cryptoBox secret nonce . unMaxLenBS . padMaxLenBS
 
 cryptoBox :: ByteArrayAccess key => key -> ByteString -> ByteString -> ByteString
-cryptoBox secret nonce s = BA.convert tag <> c
-  where
-    (rs, c) = xSalsa20 secret nonce s
-    tag = Poly1305.auth rs c
+cryptoBox secret nonce msg = either error id $! hsalsa20 secret >>= \sk -> secretBox sk nonce msg -- TODO: fuse
 
 -- | NaCl @crypto_box@ decrypt with a shared DH secret and 192-bit nonce.
 cbDecrypt :: DhSecret X25519 -> CbNonce -> ByteString -> Either CryptoError ByteString
@@ -1233,14 +1235,7 @@ sbDecrypt_ secret nonce = unPad <=< sbDecryptNoPad_ secret nonce
 
 -- | NaCl @crypto_box@ decrypt with a shared DH secret and 192-bit nonce (without unpadding).
 sbDecryptNoPad_ :: ByteArrayAccess key => key -> CbNonce -> ByteString -> Either CryptoError ByteString
-sbDecryptNoPad_ secret (CbNonce nonce) packet
-  | B.length packet < 16 = Left CBDecryptError
-  | BA.constEq tag' tag = Right msg
-  | otherwise = Left CBDecryptError
-  where
-    (tag', c) = B.splitAt 16 packet
-    (rs, msg) = xSalsa20 secret nonce c
-    tag = Poly1305.auth rs c
+sbDecryptNoPad_ secret (CbNonce nonce) packet = first (const CBDecryptError) $! hsalsa20 secret >>= \sk -> secretBoxOpen sk nonce packet -- TODO: fuse
 
 -- type for authentication scheme using NaCl @crypto_box@ over the sha512 digest of the message.
 newtype CbAuthenticator = CbAuthenticator ByteString deriving (Eq, Show)
@@ -1334,16 +1329,6 @@ unsafeSbKey s = either error id $ sbKey s
 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')
-  where
-    zero = B.replicate 16 $ toEnum 0
-    (iv0, iv1) = B.splitAt 8 nonce
-    state0 = XSalsa.initialize 20 secret (zero `B.append` iv0)
-    state1 = XSalsa.derive state0 iv1
-    (rs, state2) = XSalsa.generate state1 32
-    (msg', _) = XSalsa.combine state2 msg
-
 publicToX509 :: PublicKey a -> PubKey
 publicToX509 = \case
   PublicKeyEd25519 k -> PubKeyEd25519 k
@@ -1392,3 +1377,53 @@ keyError :: (a, [ASN1]) -> Either String b
 keyError = \case
   (_, []) -> Left "unknown key algorithm"
   _ -> Left "more than one key"
+
+type NaclDhSecret = BA.ScrubbedBytes
+
+-- type NaclDhSecret = C.DhSecret 'C.X25519h -- hashed DH used by NaCl "afternm" functions.
+
+-- Run salsa20 in a hash mode to make our DH keys match 'c_crypto_box_beforenm' output.
+hsalsa20 :: ByteArrayAccess key => key -> Either String NaclDhSecret
+hsalsa20 key = unsafePerformIO $ do
+  (r, ba :: NaclDhSecret) <- BA.withByteArray c_0 $ \inpPtr ->
+    BA.withByteArray key $ \keyPtr ->
+      BA.withByteArray sigma $ \sigmaPtr ->
+        BA.allocRet 32 $ \outPtr ->
+          NaCl.c_crypto_core_hsalsa20 outPtr (ConstPtr inpPtr) (ConstPtr keyPtr) (ConstPtr sigmaPtr)
+  pure $! if r /= 0 then Left "crypto_core_hsalsa20" else Right ba
+  where
+    -- sigma[16] = "expand 32-byte k";
+    sigma :: ByteString
+    sigma = "expand 32-byte k"
+    c_0 :: ByteString
+    c_0 = B.replicate 16 '\0'
+{-# NOINLINE hsalsa20 #-}
+
+secretBox :: NaclDhSecret -> ByteString -> ByteString -> Either String ByteString  -- TODO: tag?
+secretBox sk nonce msg = unsafePerformIO $ do
+  (r, c) <-
+    BA.withByteArray msg0 $ \mPtr ->
+      BA.withByteArray nonce $ \noncePtr ->
+        BA.withByteArray sk $ \skPtr ->
+          BA.allocRet (B.length msg0) $ \cPtr ->
+            NaCl.c_crypto_secretbox cPtr (ConstPtr mPtr) (fromIntegral $ B.length msg0) (ConstPtr noncePtr) (ConstPtr skPtr)
+  pure $! if r /= 0 then Left "crypto_secretbox" else Right (B.drop NaCl.crypto_box_BOXZEROBYTES c)
+  where
+    zeroBytes = B.replicate NaCl.crypto_box_ZEROBYTES '\0'
+    msg0 = zeroBytes <> BA.convert msg
+{-# NOINLINE secretBox #-}
+
+secretBoxOpen :: NaclDhSecret -> ByteString -> ByteString -> Either String ByteString
+secretBoxOpen sk nonce ciphertext = unsafePerformIO $ do
+  (r, m) <-
+    BA.withByteArray ciphertext0 $ \cPtr ->
+      BA.withByteArray nonce $ \noncePtr ->
+        BA.withByteArray sk $ \skPtr ->
+          BA.allocRet cLen $ \mPtr ->
+            NaCl.c_crypto_secretbox_open mPtr (ConstPtr cPtr) (fromIntegral cLen) (ConstPtr noncePtr) (ConstPtr skPtr)
+  pure $! if r /= 0 then Left "crypto_secretbox_open" else Right (B.drop NaCl.crypto_box_ZEROBYTES m)
+  where
+    ciphertext0 = boxZeroBytes <> ciphertext
+    boxZeroBytes = B.replicate NaCl.crypto_box_BOXZEROBYTES '\0'
+    cLen = B.length ciphertext0
+{-# NOINLINE secretBoxOpen #-}

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

@@ -0,0 +1,93 @@
+{-# LANGUAGE CApiFFI #-}
+{-# LANGUAGE DataKinds #-}
+{-# LANGUAGE GADTs #-}
+{-# LANGUAGE ScopedTypeVariables #-}
+{-# OPTIONS_GHC -Wno-unrecognised-pragmas #-}
+
+{-# HLINT ignore "Use camelCase" #-}
+
+module Simplex.Messaging.Crypto.NaCl.Bindings where
+
+import Data.ByteArray (ScrubbedBytes)
+import Foreign
+import Foreign.C.ConstPtr
+import Foreign.C.Types
+
+crypto_box_PUBLICKEYBYTES :: Num a => a
+crypto_box_PUBLICKEYBYTES = 32
+
+crypto_box_SECRETKEYBYTES :: Num a => a
+crypto_box_SECRETKEYBYTES = 32
+
+crypto_box_BEFORENMBYTES :: Num a => a
+crypto_box_BEFORENMBYTES = 32
+
+crypto_box_NONCEBYTES :: Num a => a
+crypto_box_NONCEBYTES = 24
+
+crypto_box_ZEROBYTES :: Num a => a
+crypto_box_ZEROBYTES = 32
+
+crypto_box_BOXZEROBYTES :: Num a => a
+crypto_box_BOXZEROBYTES = 16
+
+-- int crypto_box(u8 *c,const u8 *m,u64 d,const u8 *n,const u8 *y,const u8 *x)
+-- {
+--   u8 k[32];
+--   crypto_box_beforenm(k,y,x);
+--   return crypto_box_afternm(c,m,d,n,k);
+-- }
+foreign import capi "tweetnacl.h crypto_box"
+  c_crypto_box :: Ptr Word8 -> ConstPtr Word8 -> Word64 -> ConstPtr Word8 -> ConstPtr Word8 -> ConstPtr Word8 -> IO CInt
+
+-- int crypto_box_open(u8 *m,const u8 *c,u64 d,const u8 *n,const u8 *y,const u8 *x)
+-- {
+--   u8 k[32];
+--   crypto_box_beforenm(k,y,x);
+--   return crypto_box_open_afternm(m,c,d,n,k);
+-- }
+foreign import capi "crypto_box_open"
+  c_crypto_box_open :: Ptr Word8 -> ConstPtr Word8 -> Word64 -> ConstPtr Word8 -> ConstPtr Word8 -> ConstPtr Word8 -> IO CInt
+
+-- XXX: requires randombytes extern symbol available. Not relevant as we can random it ourselves.
+-- foreign import ccall "crypto_box_keypair"
+--   c_crypto_box_keypair :: Ptr Word8 -> Ptr Word8 -> IO CInt
+
+foreign import capi "tweetnacl.h crypto_scalarmult"
+  crypto_scalarmult :: Ptr Word8 -> ConstPtr Word8 -> ConstPtr Word8 -> IO CInt
+
+-- XXX: does NOT result in the same DH key we/crypton use as it throws HSalsa20 at the result of the scalarmult op above
+-- int crypto_box_beforenm(u8 *k,const u8 *y,const u8 *x)
+-- {
+--   u8 s[32];
+--   crypto_scalarmult(s,x,y);
+--   return crypto_core_hsalsa20(k,_0,s,sigma);
+-- }
+foreign import capi "tweetnacl.h crypto_box_beforenm"
+  c_crypto_box_beforenm :: Ptr Word8 -> ConstPtr Word8 -> ConstPtr Word8 -> IO CInt
+
+foreign import capi "tweetnacl.h crypto_core_hsalsa20"
+  c_crypto_core_hsalsa20 :: Ptr Word8 -> ConstPtr Word8 -> ConstPtr Word8 -> ConstPtr Word8 -> IO CInt
+
+foreign import capi "tweetnacl.h crypto_secretbox"
+  c_crypto_secretbox :: Ptr Word8 -> ConstPtr Word8 -> Word64 -> ConstPtr Word8 -> ConstPtr Word8 -> IO CInt
+
+foreign import capi "tweetnacl.h crypto_secretbox_open"
+  c_crypto_secretbox_open :: Ptr Word8 -> ConstPtr Word8 -> Word64 -> ConstPtr Word8 -> ConstPtr Word8 -> IO CInt
+
+-- type NaclDhSecret = C.DhSecret 'C.X25519
+type NaclDhSecret = ScrubbedBytes
+
+-- int crypto_box_afternm(u8 *c,const u8 *m,u64 d,const u8 *n,const u8 *k)
+-- {
+--   return crypto_secretbox(c,m,d,n,k);
+-- }
+foreign import capi "tweetnacl.h crypto_box_afternm"
+  c_crypto_box_afternm :: Ptr Word8 -> ConstPtr Word8 -> Word64 -> ConstPtr Word8 -> ConstPtr Word8 -> IO CInt
+
+-- int crypto_box_open_afternm(u8 *m,const u8 *c,u64 d,const u8 *n,const u8 *k)
+-- {
+--   return crypto_secretbox_open(m,c,d,n,k);
+-- }
+foreign import capi "tweetnacl.h crypto_box_open_afternm"
+  c_crypto_box_open_afternm :: Ptr Word8 -> ConstPtr Word8 -> Word64 -> ConstPtr Word8 -> ConstPtr Word8 -> IO CInt

src/Simplex/Messaging/Protocol.hs

@@ -133,6 +133,7 @@ module Simplex.Messaging.Protocol
     FwdTransmission (..),
     MsgFlags (..),
     initialSMPClientVersion,
+    currentSMPClientVersion,
     userProtocol,
     rcvMessageMeta,
     noMsgFlags,
@@ -178,6 +179,7 @@ module Simplex.Messaging.Protocol
 where
 
 import Control.Applicative (optional, (<|>))
+import Control.DeepSeq (NFData (..))
 import Control.Exception (Exception)
 import Control.Monad
 import Control.Monad.Except
@@ -886,6 +888,8 @@ deriving instance Ord (SProtocolType p)
 
 deriving instance Show (SProtocolType p)
 
+instance NFData (SProtocolType p) where rnf spt = spt `seq` ()
+
 data AProtocolType = forall p. ProtocolTypeI p => AProtocolType (SProtocolType p)
 
 instance Eq AProtocolType where
@@ -970,6 +974,8 @@ data ProtocolServer p = ProtocolServer
 
 data AProtocolServer = forall p. ProtocolTypeI p => AProtocolServer (SProtocolType p) (ProtocolServer p)
 
+instance NFData (ProtocolServer p) where rnf ProtocolServer {} = ()
+
 instance ProtocolTypeI p => IsString (ProtocolServer p) where
   fromString = parseString strDecode
 

tests/CoreTests/CryptoTests.hs

@@ -1,3 +1,4 @@
+{-# LANGUAGE DataKinds #-}
 {-# LANGUAGE GADTs #-}
 {-# LANGUAGE OverloadedStrings #-}
 {-# LANGUAGE ScopedTypeVariables #-}
@@ -7,10 +8,19 @@ module CoreTests.CryptoTests (cryptoTests) where
 
 import Control.Concurrent.STM
 import Control.Monad.Except
+import qualified Crypto.Cipher.XSalsa as XSalsa
+import qualified Crypto.Error as CE
+import qualified Crypto.MAC.Poly1305 as Poly1305
+import qualified Crypto.PubKey.Curve25519 as X25519
+import Data.Bifunctor (bimap)
+import Data.ByteArray (ScrubbedBytes)
+import qualified Data.ByteArray as BA
+import Data.ByteString (ByteString)
 import qualified Data.ByteString.Char8 as B
 import qualified Data.ByteString.Lazy.Char8 as LB
 import Data.Either (isRight)
 import Data.Int (Int64)
+import Data.Memory.PtrMethods (memSet)
 import qualified Data.Text as T
 import Data.Text.Encoding (encodeUtf8)
 import qualified Data.Text.Lazy as LT
@@ -19,11 +29,14 @@ import Data.Type.Equality
 import qualified Data.X509 as X
 import qualified Data.X509.CertificateStore as XS
 import qualified Data.X509.Validation as XV
+import Foreign.C.ConstPtr (ConstPtr (..))
 import qualified SMPClient
 import qualified Simplex.Messaging.Crypto as C
 import qualified Simplex.Messaging.Crypto.Lazy as LC
+import qualified Simplex.Messaging.Crypto.NaCl.Bindings as NaCl
 import Simplex.Messaging.Crypto.SNTRUP761.Bindings
 import Simplex.Messaging.Transport.Client
+import System.IO.Unsafe (unsafePerformIO)
 import Test.Hspec
 import Test.Hspec.QuickCheck (modifyMaxSuccess)
 import Test.QuickCheck
@@ -100,6 +113,8 @@ cryptoTests = do
     it "should validate certificates" testValidateX509
   describe "sntrup761" $
     it "should enc/dec key" testSNTRUP761
+  describe "NaCl" $
+    it "cryptobox is compatible" testNaCl
 
 instance Eq C.APublicKey where
   C.APublicKey a k == C.APublicKey a' k' = case testEquality a a' of
@@ -270,3 +285,111 @@ testSNTRUP761 = do
   (c, KEMSharedKey k) <- sntrup761Enc drg pk
   KEMSharedKey k' <- sntrup761Dec c sk
   k' `shouldBe` k
+
+testNaCl :: IO ()
+testNaCl = do
+  drg <- C.newRandom
+  (aPub :: C.PublicKeyX25519, aPriv :: C.PrivateKeyX25519) <- atomically $ C.generateKeyPair drg
+  (bPub :: C.PublicKeyX25519, bPriv :: C.PrivateKeyX25519) <- atomically $ C.generateKeyPair drg
+  let abShared@(C.DhSecretX25519 abShared') = C.dh' aPub bPriv
+  let baShared@(C.DhSecretX25519 baShared') = C.dh' bPub aPriv
+  abShared `shouldBe` baShared
+
+  naclShared <- either error pure $ dhNacl aPub bPriv
+  naclShared `shouldBe` abShared
+
+  naclBeforeNm <- either error pure $ cryptoBoxBeforenm aPub bPriv
+  abSharedH <- either error pure $ C.hsalsa20 abShared'
+  naclBeforeNm `shouldBe` BA.convert abSharedH
+
+  let msg = "hello long-enough world"
+  nonce@(C.CbNonce nonce') <- atomically $ C.randomCbNonce drg
+  naclCiphertext <- either (error . mappend "cryptoBox: " . show) pure $ cryptoBoxNaCl aPub bPriv nonce msg
+  let refCiphertext = ref_cryptoBox abShared' nonce' msg
+  (B.length naclCiphertext, naclCiphertext) `shouldBe` (B.length refCiphertext, refCiphertext)
+  naclCiphertextAfternm <- either error pure $ C.secretBox abSharedH nonce' msg
+  (B.length naclCiphertext, naclCiphertext) `shouldBe` (B.length naclCiphertextAfternm, naclCiphertextAfternm)
+
+  refMsg <- either (error . show) pure $ ref_sbDecryptNoPad_ baShared' nonce naclCiphertext
+  refMsg `shouldBe` msg
+
+  naclMsg <- either error pure $ cryptoBoxOpenNaCl bPub aPriv nonce refCiphertext
+  naclMsg `shouldBe` msg
+
+  naclMsgAfternm <- either error pure $ C.secretBoxOpen abSharedH nonce' naclCiphertext
+  naclMsgAfternm `shouldBe` msg
+
+-- | A replica of C.dh' using NaCl (sans hsalsa20 step)
+dhNacl :: C.PublicKeyX25519 -> C.PrivateKeyX25519 -> Either String (C.DhSecret 'C.X25519)
+dhNacl (C.PublicKeyX25519 pub) (C.PrivateKeyX25519 priv _) = unsafePerformIO $ do
+  (r, ba :: ScrubbedBytes) <- BA.withByteArray pub $ \pubPtr ->
+    BA.withByteArray priv $ \privPtr ->
+      BA.allocRet 32 $ \sharedPtr -> do
+        memSet sharedPtr 0 32
+        NaCl.crypto_scalarmult sharedPtr (ConstPtr privPtr) (ConstPtr pubPtr)
+  pure $! if r /= 0 then Left "crypto_scalarmult" else bimap show C.DhSecretX25519 $ CE.eitherCryptoError (X25519.dhSecret ba)
+
+cryptoBoxBeforenm :: C.PublicKeyX25519 -> C.PrivateKeyX25519 -> Either String ScrubbedBytes
+cryptoBoxBeforenm (C.PublicKeyX25519 pub) (C.PrivateKeyX25519 priv _) = unsafePerformIO $ do
+  (r, ba :: ScrubbedBytes) <- BA.withByteArray pub $ \pubPtr ->
+    BA.withByteArray priv $ \privPtr ->
+      BA.allocRet 32 $ \kPtr -> do
+        memSet kPtr 0 32
+        NaCl.c_crypto_box_beforenm kPtr (ConstPtr pubPtr) (ConstPtr privPtr)
+  pure $! if r /= 0 then Left "crypto_box_beforenm" else Right ba
+
+cryptoBoxNaCl :: BA.ByteArrayAccess msg => C.PublicKeyX25519 -> C.PrivateKeyX25519 -> C.CbNonce -> msg -> Either String ByteString
+cryptoBoxNaCl (C.PublicKeyX25519 pk) (C.PrivateKeyX25519 sk _) (C.CbNonce n) msg = unsafePerformIO $ do
+  (r, c) <-
+    BA.withByteArray msg0 $ \mPtr ->
+      BA.withByteArray n $ \nPtr ->
+        BA.withByteArray pk $ \pkPtr ->
+          BA.withByteArray sk $ \skPtr ->
+            BA.allocRet (B.length msg0) $ \cPtr ->
+              NaCl.c_crypto_box cPtr (ConstPtr mPtr) (fromIntegral $ B.length msg0) (ConstPtr nPtr) (ConstPtr pkPtr) (ConstPtr skPtr)
+  pure $! if r /= 0 then Left "crypto_box" else Right (B.drop NaCl.crypto_box_BOXZEROBYTES c)
+  where
+    msg0 = zeroBytes <> BA.convert msg
+    zeroBytes = B.replicate NaCl.crypto_box_ZEROBYTES '\0'
+
+cryptoBoxOpenNaCl :: C.PublicKeyX25519 -> C.PrivateKeyX25519 -> C.CbNonce -> ByteString -> Either String ByteString
+cryptoBoxOpenNaCl (C.PublicKeyX25519 pub) (C.PrivateKeyX25519 priv _) (C.CbNonce n) ciphertext = unsafePerformIO $ do
+  (r, msg) <-
+    BA.withByteArray ciphertext0 $ \cPtr ->
+      BA.withByteArray n $ \nPtr ->
+        BA.withByteArray pub $ \pubPtr ->
+          BA.withByteArray priv $ \privPtr ->
+            BA.allocRet cLen $ \mPtr ->
+              NaCl.c_crypto_box_open mPtr (ConstPtr cPtr) (fromIntegral cLen) (ConstPtr nPtr) (ConstPtr pubPtr) (ConstPtr privPtr)
+  pure $! if r /= 0 then Left "crypto_box_open" else Right (B.drop NaCl.crypto_box_ZEROBYTES msg)
+  where
+    ciphertext0 = boxZeroBytes <> ciphertext
+    boxZeroBytes = B.replicate NaCl.crypto_box_BOXZEROBYTES '\0'
+    cLen = B.length ciphertext0
+
+ref_cryptoBox :: BA.ByteArrayAccess key => key -> ByteString -> ByteString -> ByteString
+ref_cryptoBox secret nonce s = BA.convert tag <> c
+  where
+    (rs, c) = ref_xSalsa20 secret nonce s
+    tag = Poly1305.auth rs c
+
+-- | NaCl @crypto_box@ decrypt with a shared DH secret and 192-bit nonce (without unpadding).
+ref_sbDecryptNoPad_ :: BA.ByteArrayAccess key => key -> C.CbNonce -> ByteString -> Either C.CryptoError ByteString
+ref_sbDecryptNoPad_ secret (C.CbNonce nonce) packet
+  | B.length packet < 16 = Left C.CBDecryptError
+  | BA.constEq tag' tag = Right msg
+  | otherwise = Left C.CBDecryptError
+  where
+    (tag', c) = B.splitAt 16 packet
+    (rs, msg) = ref_xSalsa20 secret nonce c
+    tag = Poly1305.auth rs c
+
+ref_xSalsa20 :: BA.ByteArrayAccess key => key -> ByteString -> ByteString -> (ByteString, ByteString)
+ref_xSalsa20 secret nonce msg = (rs, msg')
+  where
+    zero = B.replicate 16 $ toEnum 0
+    (iv0, iv1) = B.splitAt 8 nonce
+    state0 = XSalsa.initialize 20 secret (zero `B.append` iv0)
+    state1 = XSalsa.derive state0 iv1
+    (rs, state2) = XSalsa.generate state1 32
+    (msg', _) = XSalsa.combine state2 msg