* fix(fs): bound getFiles() result size and harden traversal
The recursive filesystem traversal in getFiles() could grow its result
vector without limit, contributing to heap pressure on memory-constrained
devices during client config starts. The implementation also relied on
unbounded strcpy and per-entry String() allocations for dot-suffix
filtering, and leaked the directory handle when the opened path was not
a directory.
Rework the traversal to address all of those at once:
- Add optional maxCount (default 64) and wasLimited out-param. The walk
stops once maxCount entries have been collected and wasLimited is set,
so callers can distinguish a complete listing from a truncated one.
Default arguments preserve backward compatibility for existing callers.
- Extract the recursive walk into a collectFiles() helper in an
anonymous namespace, accumulating into a single caller-owned vector
instead of building and inserting temporary vectors at every recursion
level.
- Replace strcpy with strlcpy so file_name writes are bounded by the
destination buffer.
- Replace String(name).endsWith(".") with a stack-based
pathEndsWithDot() helper to avoid a heap allocation per entry.
- Close the directory handle when the opened path is not a directory,
and close the file in both branches of recursion instead of only the
leaf branch (the previous code leaked the handle when depth was
exhausted).
- Compile-gate the std::bad_alloc reserve probe-down behind
__cpp_exceptions / __EXCEPTIONS so the fallback only compiles where
bad_alloc is actually catchable. No-exception targets (nRF52, RP2040,
STM32, ESP32-C6) skip the probe and rely on the bounded maxCount
alone, which is safe because the default cap is small enough to fit
comfortably in available RAM on every supported platform.
* fix(phone-api): skip manifest scan for node-info-only config requests
A client can request a node-info-only config response by sending
want_config_id with the SPECIAL_NONCE_ONLY_NODES nonce. The manifest
walk in handleStartConfig() ran unconditionally even though
STATE_SEND_FILEMANIFEST short-circuits past the manifest on that path,
so the SPI lock and recursive filesystem traversal were wasted work.
On memory-constrained devices this extra heap pressure was enough to
trigger allocation aborts and spontaneous reboots before the node-info
response could complete.
Gate the manifest call behind config_nonce != SPECIAL_NONCE_ONLY_NODES
so the node-info-only path no longer touches the filesystem at all, and
release any stale manifest storage left over from a previous run.
For normal full-config requests, also tighten the scan limits from
depth 10 / unbounded to depth 3 / max 64 entries via the new getFiles()
parameters. A 10-level recursive listing was never useful for a client
file picker and amplified heap pressure for no benefit. Log a warning
when the listing was truncated so operators can distinguish a bounded
result from a genuinely empty filesystem.
A few small cleanups fall out of the above:
- Replace raw spiLock->lock()/unlock() with a LockGuard so the lock is
released even if getFiles() throws on exception-enabled targets.
- Use a releaseFilesManifest() helper (swap-with-empty idiom) in
close() and the STATE_SEND_FILEMANIFEST drain path instead of the
previous clear()+shrink_to_fit() pair. Equivalent behavior, but the
swap idiom is the canonical vector-freeing pattern and centralizing
it avoids divergence between call sites.
- Use %zu for the size_t log arguments.
* fix(fs): harden bounded directory traversal
Close each child directory entry before recursive descent so bounded manifest scans do not stack file handles on constrained filesystems.
Treat null or overlong paths as truncated listings, handle reserve length failures like allocation failures, and keep manifest constants aligned with repository naming.
* Potential fix for pull request finding
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
---------
Co-authored-by: Ben Meadors <benmmeadors@gmail.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
* fix(esp32): skip RTC timer wake on user shutdown
Do not arm esp_sleep_enable_timer_wakeup when msecToWake is portMAX_DELAY (UI shutdown), matching nRF52 system_off semantics.
fix(rak_wismesh_tap_v2): Tag OCV curve and 16MB partition
Add OCV_ARRAY matching WisMesh Tag for accurate SOC. Use 16MB flash partition scheme for TAP V2 hardware.
Co-authored-by: Cursor <cursoragent@cursor.com>
* Trunkt
---------
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Ben Meadors <benmmeadors@gmail.com>
* Wipe message Store on factory reset
* Check for destination 0 in a new message, and convert to broadcast
* Make sure CHARGE_LED_state gets turned off, to avoid stuck LEDs
* Take the spiLock in MessageStore when clearing messages
* Trunk
* Add thinknode M5 voltage curve
* Fix the oops
* Use standard GPS enable pin, for smarter power control on M3
* Enable GPS pin in variant.cpp initialization
---------
Co-authored-by: Ben Meadors <benmmeadors@gmail.com>
13302 and 13300 had missing module config for slot 2, also the rf switch and voltage info was missing.
Without that the auto setting in the config.yaml will try to use the default lora-hat-rak-6421-pi-hat.yaml and things do not work correctly.
alex
