A spoofed GPS signal can move a fixed node anywhere and set its clock to
any time; the clock only moves forward, so a time pushed into the future
breaks timestamps and replay protection until fixed by hand.
With GPS_GUARD, MicroNMEALocationProvider asks a GpsGuard before a fix is
stored as the last fix (satellites, HDOP, optional region box, fence
around the configured position, jump speed, N stable fixes) and before
GPS time is written to the clock (trusted fix current, not before the
build date, small steps once the clock is valid, drift-sized after the
first sync). Without a trusted fix no live position is reported.
CLI in CommonGpsPrefs: 'gps guard', 'gps guard accept', 'gps fence',
'gps ref'; CommonCLI keeps the reference on set lat/lon and gps setloc,
resets the time anchor on manual clock sets and adds 'time force'.
Ported from #3736 onto the GpsDriver/LocationProvider refactor.
Bytes arriving on the UART does not mean the baud rate is right: at the
wrong rate the parser only sees garbage and never gets a fix. Modules
such as the M5Stack GROVE GPS now default to 115200, while boards
configure 9600.
With GPS_BAUD_AUTODETECT, detect() probes common GNSS rates, starting
with the configured one, and keeps the first one that yields an NMEA
sentence with a valid checksum. The rate is stored in the driver, so
later power cycles reopen the UART with it. Without the flag nothing
changes.
Ported from #3729 onto the GpsDriver refactor.
On a RAK19007 the RAK12501's supply is 3V3_S, switched by WB_IO2, so
gps off, periodic mode and shutdown left it powered. Pass WB_IO2 to the
NMEA driver as peripheralPower under GPS_RAIL_POWER_CONTROL, set by the
new RAK_4631_repeater_gps_power env. It is opt-in because the rail feeds
every slot module, and refused with ETHERNET_ENABLED (RAK13800).
The RAK12500 (I2C) driver keeps switching its own power pin, so this
only applies on the NMEA path.
RefCountedDigitalPin now sets OUTPUT on claim and release: WB_IO2 left
as an input (as probeRak12500() leaves it after a failed probe) keeps
3V3_S in its last state, so writing the latch alone did not cut it.
Measured with a KM003C, battery and panel out: gps off 40.5 -> 12.4 mA,
shutdown 29.7 -> 1.30 mA.
'sensor list' appended 'name=value' lines into the 160-byte CLI reply with
unbounded sprintf, only checking the position before each write (< 134).
A single long value (or several medium ones) could write past the end of
the buffer.
Write each entry with snprintf against the remaining space, keep room for
the '... next:N' paging hint, and move entries that do not fit to the next
page. Also guard against NULL names/values from a sensor manager.
Use GxEPD2's busy callback to run the mesh, interface and sensor loops
(loopCore(), shared with loop()) while the panel refreshes, instead of
blocking the main loop. Covers the partial refresh in endFrame(), the
full wipe in startFrame(), and the sleep icon / blank refresh on turnOff().
Compile out code when the corresponding macro defines are missing.
Failing builds: comp_radio_usb, repeatr, room_svr,
terminal_chat
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ESP32Board.h defined _rtc_backup_time and _rtc_backup_magic as static in the
header, so every translation unit that includes it has its own copy, and the
inline members of ESP32RTCClock read and write the copy of the unit they are
compiled in. Today one unit per image uses them; a second one could restore a
copy that setCurrentTime() never wrote. The header now declares them, and
ESP32Board.cpp defines them once.
"set af nan" is accepted and stores a NaN in airtime_factor, the duty cycle
budget the roles hand to their airtime accounting. strtof() consumes the
token, so the parse guard does not fire, and no ordered comparison is true for
a NaN, so "af < 0 || af > 9" cannot reject it. "set dutycycle" writes the same
field and has the same hole.
Both commands now check the text before it is parsed: after optional white
space and sign, a number has to start with a digit or a '.'. That refuses
"nan", "inf" and their variants, and keeps every spelling that was accepted
before (signs, leading white space, ".5"). The ranges and both reply strings
are unchanged.
The check is on the text rather than on the parsed float on purpose. nRF52
builds use -Ofast, and under -ffast-math the compiler may assume a float is
never NaN or infinite. A test on the value, even one that reads its bits, can
then be folded away: clang does exactly that with a bit test at -Ofast, while
arm-none-eabi-gcc keeps it. A character test does not depend on that.
CMD_SEND_CHANNEL_TXT_MSG reads txt_type, channel_idx and a 4 byte timestamp, so
the frame needs 7 bytes before the text starts, and the branch had no length
guard. For a shorter frame "len - i" is computed in size_t, wraps, and reaches
sendGroupMessage() as a negative int. The clamp there tests
"text_len + prefix_len > MAX_TEXT_LEN", which a negative value passes, and the
following memcpy() converts the length back to a size_t: a 3 byte frame asks for
4294967292 bytes into a 197 byte stack buffer.
Two guards, because there are two mistakes. The branch now states its minimum
like most other branches of handleCmdFrame do, so a frame below 7 bytes is
answered through the existing path instead of being parsed. And
sendGroupMessage() refuses a negative length the way sendGroupData() in the same
file already does, which protects every other caller.
A frame of exactly 7 bytes keeps working and sends an empty message, as before.
Evidence. On a T-Echo running dev 428a95c6, built with
FIRMWARE_BUILD_DATE "f1-dev" and confirmed by "ver => v1.18.0 (Build: f1-dev)":
CH_TXT len=7 -> RESP_CODE_OK
GET_STATS core -> uptime=78s
CH_TXT len=3 -> NO REPLY
DEVICE_QUERY -> NO REPLY (three times over 8 s, node never came back)