7.1 KiB
Admin-pushed firmware update (experimental -ota images)
The *-ota and *-observer-ota images of this fork's releases can take a new
firmware from their admin -- over the USB or BLE console, or over the mesh,
as remote admin CLI. Nothing changes in the bootloader and there is no second
bank: the node stores the package in free flash, checks all of it, and only on
an explicit fw apply rewrites itself from a routine running in RAM.
The point is the delta: a package against the image the node runs is usually a few kB, so it goes through LoRa in minutes. A full image (~300 kB) goes only over USB or BLE.
Rolling release on dev, for experienced users. These -ota images track
MeshCore dev, merged continuously, so a new release lands roughly twice a day
and each delta stays small. There is no stable channel, no second bank, and no
automatic rollback: a bad image that still boots keeps running until you push the
next one, and one that will not boot needs USB/BLE DFU in hand. Run it only on
nodes you can reach, and try every release on one within reach first.
What a release carries
For every <board>_<role>-ota / -observer-ota image:
| file | what |
|---|---|
<image>-<release>.zip / .uf2 |
the image, flashed the usual way (first time) |
<image>-<release>.mcfw |
full package: USB/BLE console only |
<image>-<release>.from-<earlier>.mcfw |
delta from an earlier release's same image |
manifest.json |
image sha256, target, build epoch, packages, deltas by base |
SHA256SUMS, SHA256SUMS.sig |
checksums, signed by meshcore-ota@tth-pub |
Check a download:
ssh-keygen -Y verify -f tools/ota_allowed_signers -I meshcore-ota@tth-pub \
-n meshcore-ota -s SHA256SUMS.sig < SHA256SUMS
sha256sum -c --ignore-missing SHA256SUMS
tools/ota_pack.py fetch <image> --base <sha8> does both and picks the delta
for the node's image (--full for the full package, --release <tag> for
another than the latest).
On the node
fw status state, battery, the running image's size and sha256 prefix
fw ver target and build epoch a package for this image needs
fw begin <bytes> [sid] start a transfer
fw w <offset> <base64> one 96 B chunk (text)
fw missing what is still missing
fw verify queue the full check, poll fw status
fw apply battery gate, then rewrite and reboot
fw apply downgrade local console only: an older (or the same) image
fw abort
What the node checks before it touches its own image:
- The manifest, as soon as it is in: an Ed25519 signature by the key the
image was built with (
[ota] pubkey), the target (board + variant: an-observer-otanode takes no-otapackage), the size, for a delta the base (the sha256 of the image it runs), and the build epoch. - Over the mesh: only a delta, and only a newer build. An older or equal
one is refused right after the manifest; a full image is refused when the
transfer begins. Locally a full image is fine, and an older one takes
fw apply downgrade. fw verify: the payload hash, and a delta applied dry -- the same code the real run uses, with the flash untouched -- must produce the signed image hash.fw apply: battery at 3.8 V or more unless the node runs on USB. The bootloader settings get the new image's CRC first, so a power cut during the rewrite leaves the node in the bootloader (back with USB/BLE DFU), not in half an image.
When an update goes wrong
No second bank, no automatic rollback. The checks above stop a corrupt package, but firmware that passes them and still boots into a broken state keeps running. Two watchdogs catch that, both off by default -- turn them on for nodes you push to:
set radio.silence <min> reboot if the radio hears nothing for <min> (persisted);
catches an OTA that boots with a dead radio
set radio.txstall <min> reboot if the TX queue stalls (RAM only)
get radio.reboots watchdog reboots so far (survives power loss)
get boot why the board last reset, plus the hardfault record
A node that will not boot at all has to be reflashed over USB/BLE DFU. After an
fw apply over BLE the bond can drop (a supervision timeout while the node
reboots) -- re-pair from the same host; it is not a brick.
Tools
# a delta for the node's image, from the latest release
tools/ota_pack.py fetch Xiao_nrf52_repeater-ota --base <sha8 from fw status>
# over the mesh, through a companion (USB port or tcp://host:port), as admin
tools/ota_mesh_push.py /dev/serial/by-id/<companion> <node pubkey prefix> pkg.mcfw --apply
# over the node's own USB console
tools/ota_push.py /dev/serial/by-id/<node> pkg.mcfw --apply
Off the shared channel: --side
tools/ota_mesh_push.py <companion> <node> pkg.mcfw --apply \
--side 869.525,62.5,7,5 --duty 10 --gap 1
Login, fw status and get radio still go over the mesh; then the node gets
tempradio <side>,3, the companion retunes to the same preset, logs in again
zero-hop, and the chunks go there. The window grows to --window (15 min) once
the link is proven and is renewed while the transfer runs. The node comes back
by itself when the window runs out, on fw apply (the reboot), or when the
tool sends it back with tempradio <its preset>,1. Nothing new in the firmware:
any node with tempradio takes it.
- The companion must be in direct range of the node, and it must be a
companion firmware that retunes its own radio. An openHop virtual
companion answers
set_radiowith OK and stays put; the tool checks that and stops, and the node returns within 3 minutes. - The companion's preset is put back at the end -- unless the tool is killed
hard; then
set_radioit back yourself. - The example preset sits inside 869.4-869.65 MHz (10 % duty) between the CZ mesh (869.432/62.5) and EU Narrow (869.618/62.5). The node keeps its TX power on the side channel: pick a band where that power is allowed.
ota_mesh_push.py needs the meshcore Python library. It logs in as admin
(password from [secrets] admin_pw of platformio.local.ini, or asked for),
sends binary chunks with a pause between them, asks which are missing and
resends only those, and keeps an airtime budget (--duty, default 1 % per
hour: the 865-868 MHz band). A refused package costs the two manifest chunks,
not the whole transfer. Check get radio on both ends first: this transmits.
Your own builds
A build with -D FW_OTA=1 trusts [ota] pubkey of platformio.ini -- the
release key. To sign your own packages, set your key in the gitignored
platformio.local.ini:
[ota]
pubkey = <tools/ota_pack.py pubkey --key your.key>
and pack with tools/ota_pack.py pack|diff <env> --key your.key. Target and
build epoch come from the ELF, never from the command line. Keep every image
you flash: a delta needs the exact bytes the node runs as its base.
Cost: about 10 kB of RAM on a repeater, the applier code in RAM, a 32 kB
buffer while a full package is verified or applied. Classic images (without
-ota) have none of it.