The W5100S Ethernet chip has only 4 hardware sockets. On RAK4631
Ethernet gateways with syslog and NTP enabled, all 4 sockets were
permanently consumed (NTP UDP + Syslog UDP + TCP API listener + TCP
API client), leaving none for MQTT, DHCP lease renewal, or additional
TCP connections.
- NTP: Remove permanent timeClient.begin() at startup; NTPClient::update()
auto-initializes when needed. Add timeClient.end() after each query to
release the UDP socket immediately.
- Syslog: Remove socket allocation from Syslog::enable(). Open and close
the UDP socket on-demand in _sendLog() around each message send.
- MQTT: Fix socket leak in isValidConfig() where a successful test
connection was never closed (PubSubClient destructor does not call
disconnect). Add explicit pubSub->disconnect() before returning.
Made-with: Cursor
Co-authored-by: Ben Meadors <benmmeadors@gmail.com>
* Fix TransmitHistory to improve epoch handling
* Enable epoch handling in unit tests
* Improve comments and test handling for epoch persistence in TransmitHistory
* Add boot-relative timestamp handling and unit tests for TransmitHistory
* loadFromDisk should handle legacy entries and clean up old v1 files after migration
* Revert "loadFromDisk should handle legacy entries and clean up old v1 files after migration"
This reverts commit eb7e5c7acfa4ac077fe50980be752e4b42a739b8.
* Add NodeInfoModule integration for RTC quality changes and trigger immediate checks
* Update test conditions for RTC quality checks
* Fix issue 9792, decode packet for TR test
* Fix 9792: Assure packet id decoded for TR test
* Potential fix for pull request finding
Log improvement for failure to decode packet.
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
* trunk fmt
---------
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Tom Fifield <tom@tomfifield.net>
* Fix for preserving pki_encrypted and public_key when relaying UDP multicast packets to radio.
PKI DMs sent over UDP multicast had their pki_encrypted flag and public_key fields explicitly cleared before being forwarded to the LoRa radio. This caused the receiving node to treat the packet as a channel-encrypted message it couldn't decrypt, silently dropping it.
The MQTT ingress path correctly preserves these fields. The UDP multicast ingress path should behave the same way.
* Zeroize MeshPacket before decoding
Zeroize MeshPacket before decoding to prevent data leakage.
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
---------
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Ben Meadors <benmmeadors@gmail.com>
* Fix RAK4631 Ethernet gateway API connection loss after W5100S brownout
PoE power instability can brownout the W5100S while the nRF52 MCU keeps
running, causing all chip registers (MAC, IP, sockets) to revert to
defaults. The firmware had no mechanism to detect or recover from this.
Changes:
- Detect W5100S chip reset by periodically verifying MAC address register
in reconnectETH(); on mismatch, perform full hardware reset and
re-initialize Ethernet interface and services
- Add deInitApiServer() for clean API server teardown during recovery
- Add ~APIServerPort destructor to prevent memory leaks
- Switch nRF52 from EthernetServer::available() to accept() to prevent
the same connected client from being repeatedly re-reported
- Add proactive dead-connection cleanup in APIServerPort::runOnce()
- Add 15-minute TCP idle timeout to close half-open connections that
consume limited W5100S hardware sockets
Fixesmeshtastic/firmware#6970
Made-with: Cursor
* Log actual elapsed idle time instead of constant timeout value
Address Copilot review comment: log millis() - lastContactMsec to show
the real time since last client activity, rather than always logging the
TCP_IDLE_TIMEOUT_MS constant.
Made-with: Cursor
* Update src/mesh/api/ServerAPI.h
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Stop UDP multicast handler during W5100S brownout recovery
After a W5100S chip brownout, the udpHandler isRunning flag stays
true while the underlying socket is dead. Without calling stop(),
the subsequent start() no-ops and multicast is silently broken
after recovery.
Made-with: Cursor
* Address Copilot review: recovery flags and timeout constant
Move ethStartupComplete and ntp_renew reset to immediately after
service teardown, before Ethernet.begin(). Previously, if DHCP
failed the early return left ethStartupComplete=true, preventing
service re-initialization on subsequent retries.
Replace #define TCP_IDLE_TIMEOUT_MS with static constexpr uint32_t
for type safety and better C++ practice.
Made-with: Cursor
---------
Co-authored-by: Ben Meadors <benmmeadors@gmail.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Add C++17 support
* Add C++17 runtime probes and update build configurations for STM32 and Portduino
* Remove unflags
* Update C++ standard flags across the platforms
* Convert a couple of instances to structured bindings
* NRF52 platform.txt to add C++17 support
* Still need the unflags apparently
* Remove C++17 runtime probe tests from test_main.cpp
* Reconfigured doesnt need a nodiscard
* Remove nodiscard attribute from init() method in RadioInterface
* Remove mbedtls/error.h from build flags
Removed include directive for mbedtls/error.h from build flags.
* Fix IRAM overflow
* Fix IRAM overflow
* Add build flag to exclude MQTT from rak11200
* Update C++ standard from gnu17 to gnu++17
socketSendUDP() calls yield() while waiting for SEND_OK, allowing the
cooperative scheduler to run AsyncUDP::runOnce(). When runOnce() calls
parsePacket() on the same socket during an active send, the multicast
loopback packet arriving from the switch produces a garbled 8-byte RX
header (wrong source IP, wrong payload length), leading to protobuf
decode errors in UdpMulticastHandler.
Add a volatile isSending flag that writeTo() sets around endPacket()
(the only yield point). runOnce() checks this flag and skips
parsePacket() while a send is in progress. Loopback packets stay
buffered in the W5100S RX buffer and are read cleanly on the next
poll cycle.
Also propagate writeTo() return value in UdpMulticastHandler::onSend()
instead of unconditionally returning true.
Made-with: Cursor
Co-authored-by: Ben Meadors <benmmeadors@gmail.com>
* Update Heltec Tracker v2 to version KCT8103L.
* Fixed the issue of inaccurate comments.
---------
Co-authored-by: Ben Meadors <benmmeadors@gmail.com>
Adds ROUTER_LATE and CLIENT_BASE to preferred rebroadcaster check
(skip unsolicited NodeInfo) and prevents ROUTER_LATE from setting
rebroadcast mode to NONE, which would silently break relaying.
ROUTER_LATE is an infrastructure relay that should use the same power
assumptions, interval defaults, and LoRa wake behavior as ROUTER.
Without this, ROUTER_LATE uses client-class defaults and will not
wake on LoRa activity from deep sleep, making it a broken relay.
Co-authored-by: Ben Meadors <benmmeadors@gmail.com>
ROUTER_LATE already has high base intervals and should not be further
scaled by congestion. TAK_TRACKER is a tracker variant that should
skip congestion scaling like TRACKER does.
Co-authored-by: Ben Meadors <benmmeadors@gmail.com>
ROUTER_LATE now preserves node_info_broadcast_secs during factory
reset and auto-enables Store & Forward server mode, matching ROUTER
infrastructure behavior.
* refactor: update throttling factor calculation and add unit tests for scaling behavior
* refactor: adjust throttling factor calculation for improved accuracy in different configurations
* Update src/mesh/Default.h
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Update src/mesh/Default.h
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* refactor: enhance throttling factor calculation and introduce pow_of_2 utility function
* refactor: improve expected ms calculation in unit tests for Default::getConfiguredOrDefaultMsScaled
* refactor: improve scaling logic for routers and sensors in computeExpectedMs function
---------
Co-authored-by: Ben Meadors <benmmeadors@gmail.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Improve resource cleanup on connection close
* Copilot had some good feedback. Let's just make the api a unique pointer
* Update src/mesh/api/ServerAPI.cpp
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Copilot stupidly suggesting we call protected methods
* Gotta do it in the superclasses as well
* Fix moar
* Refactor MQTT unit test to ensure proper subscription handling and clear side effects
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
getByIndex allocates memory and returns dummy channel whenever
chIndex validation fails. Comment implies this may happen
when malformed packet is received. The fix changes implementation
so static dummyChannel is returned in such case.
Co-authored-by: Ben Meadors <benmmeadors@gmail.com>
* enhance tx queue priority management
In busy environments, especially for ROUTER_LATE role, tx queue
fills very quickly. Delayed packets became late but new packets
to be retransmitted won't be put into the tx queue as old ones
stay there for a very long time (even a minute or more). This change
makes meshtastic prioritize new packets over late packets from tx queue
and allows to remove late packet from back of tx queue when there
is no space for a new one.
* apply copilot recommendation for cast
---------
Co-authored-by: Ben Meadors <benmmeadors@gmail.com>
* Add transmit history for throttling that persists between reboots
* Fix RAK long press detection to prevent phantom shutdowns from floating pins
* Update test/test_transmit_history/test_main.cpp
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Test fixes and placeholder for content handler tests
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Align telemetry broadcast want_response behavior with traceroute
* Fixes
* Reduce side-effects by making the telemetry modules handle the ignorerequest
* Remove unnecessary ignoreRequest flag
* Try inheriting from MeshModule
* Add exclusion for sensor/router roles and add base telem module
* Implement 'agc' reset for SX126x chip family
There's no actual agc on SX126x chips but you can reset the analog
registers by doing a warm sleep & running calibration.
* Address PR comments & implement for LR11x0 too
* calibrate for configured frequency band
* Gate LR11X0_AGC_RESET
* Apply SX1262 register 0x8B5 patch for improved GC1109 RX sensitivity
Sets the LSB of undocumented SX1262 register 0x8B5 on Heltec V4 and
Wireless Tracker V2 boards with the GC1109 FEM. This patch was
recommended by Heltec/Semtech and tested in MeshCore PR #1398, where
it significantly reduced packet loss on the Heltec V4.
* Use higher level function
* Add .venv/ to .gitignore
* Reduce maximum concurrent HTTPS connections to save memory
* Add heap check and limit connections to prevent HTTPS connection related crashes
* Use Throttle::isWithinTimespanMs for overflow-safe heap warning throttle
* Update src/mesh/http/WebServer.cpp
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Fix double-free heap corruption and parser leak in web server handlers
- Remove double-free in handleFsBrowseStatic, handleNodes, handleScanNetworks:
JSONValue(const JSONArray&) shallow-copies pointers, so delete value already
recursively frees all elements. The explicit cleanup loops were deleting the
same pointers a second time, corrupting the ESP32 heap allocator metadata.
This is the likely root cause of #8827 (SSL setup failures after uptime).
- Fix memory leak in handleFormUpload: two early-return paths were missing
delete parser.
- Remove unused global HTTPClient httpClient and its includes.
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Hold GC1109 PA_POWER and PA_EN during deep sleep for LNA RX wake
Use rtc_gpio_hold_en to latch PA_POWER (LDO) and PA_EN (CSD) HIGH
during deep sleep so the GC1109 LNA remains powered for wake-on-packet
RX. Previously these pins used weak pull-ups which could lose state.
On deep sleep wake, skip these pins in the blanket RTC hold release and
instead release them in SX126xInterface::init() after GPIO registers are
set HIGH first, avoiding a power glitch on the GC1109.
Trade-off: ~6.5mA additional deep sleep current for significantly
improved wake-on-packet RX sensitivity (~17dB).
Reference: https://github.com/meshcore-dev/MeshCore/pull/1600
* Add LDO startup delay before GC1109 chip enable
TLV75733P LDO has ~550us startup time (datasheet tSTR). On cold boot,
wait 1ms for VBAT to stabilise before driving CSD/CPS, per GC1109
power-on sequence requirement. On deep sleep wake the LDO is held on
via RTC latch so no delay is needed.
Kittiwut Khongkaeo