MeshCoreOTA moves firmware images over LoRa to a repeater, from a phone that already has a trusted relationship with the companion it is connected to. The design goal: an attacker who can transmit on the same radio can never make the node boot a different image, and a broken image can never stay booted. Availability attacks (jamming) are out of scope — that is physics, not policy.
Layered controls, in order:
- Arming requires an operator. A repeater-side
ota stage <size> <sha256> <ver>registers a pending arm{size, sha256, token}with a 120-second single-use TTL. Only an authenticated admin session (the same ECDH-encrypted TXT_MSG path as every other repeater command) can do this. - OP_START is bound to the arm. It carries
size:u32 sha256[32] target:u32 token:u32; the node accepts it only while an unexpired pending arm exists whose token AND{size, sha256}match — and the arm is consumed on acceptance. An overheard OP_START arms nothing but the operator-approved image. Anything else is dropped beforeesp_ota_begin()runs, so a flood of forged packets cannot wear out the inactive slot. target:u32 = sha256(pio_env)[0..3]binds the stream to the board this image was built for. Flashing a different board's image is refused at OP_START witherr=1.- SHA-256 gate at
OP_END. The staged slot is hashed; mismatch → staged slot scrapped,state=error,err=4. This is the only integrity decision in the design; the op stream itself is trustless once armed. - Commit requires an admin.
ota applyrefuses unless staging isstate=ready, then flips the boot partition and reboots. - Boot health check + auto rollback. The freshly applied image runs a
boot health check (LoRa init + one zero-hop packet TX). Failure flips the
boot partition back to the previous slot and reboots — with no app at all.
The app additionally re-logs in and checks
versionafter the reboot.
What an on-air attacker can do: observe that an update is happening
(public release metadata: board, size, SHA-256), re-deliver the genuine
release to a node whose operator armed it (zero gain), jam or corrupt the
transfer (denial of availability; recoverable with ota abort + re-stage),
and fuzz the op interpreter (bounded — every frame is CRC-checked and size-
capped; errors reach state=error, never panic).
What they cannot do: install a modified image (4), boot it without an
admin (5) and a passing SHA gate (4), target the wrong board (3), arm the
node at all (1), or learn anything secret in transit — firmware images are
public releases, and the arm token never travels in the clear inside the
authenticated CLI channel except inside ota stage's encrypted reply (the
OP_START frame itself is cleartext by design; its safety is (2), not secrecy).
Deliberately not encrypted: the LoRa op stream. Wire confidentiality of public firmware buys nothing, and E2E encryption per hop would add cost and code with no new guarantee at the arm/SHA/apply gates.
- Repeater admin password — per session only, in memory, never logged,
never persisted (no
flutter_secure_storageuse). Closing the app forgets it. A blank password is a legitimate login when the companion's key is already an admin in the node's ACL; the app sends the stock blank-login request and the node decides. Anyone with the unlocked phone during a session has the same admin session as the user — same as any logged-in repeater app. - Companion connection — BLE pairing is handled by the OS; the app never sees or stores pairing keys.
- No telemetry, no analytics, no crash reporting, no network access except
the firmware catalog/download endpoints (
api.github.com,objects.githubusercontent.comfor the fork's releases). The offline switch disables even those: the app never contacts GitHub while it is on. - No persistent copy of uploaded images beyond the local library (explicit
user action), which holds only public release images and their SHA-256.
The only other app-owned state is
ota_settings.json, which records the offline flag. - No store listing dependency — releases are GitHub Releases (signed Android APKs and unsigned desktop bundles).
This is a personal project for a self-hosted mesh fork. Report issues by opening a private security advisory on jhuebert/MeshCoreOTA (or on the fork, for firmware-side issues: jhuebert/MeshCore). Administration session hygiene is your responsibility: change the repeater admin password if a session may have leaked, exactly as you would for any remote administration.