Build and release engineering: teardown order, slot safety, release gates

- Controllers stop at K80, before forgectrl at K90: runlevel 0/6 no longer tears down the cooling engine, fire gates, and broker while a controller may still be executing a job. - The grblhal/gfcloud init scripts are real emergency levers: stop routes through the supervisor (POST /controller/stop - a bare pkill was safed and respawned seconds later), start resumes supervision, status exists, and the pkill fallback matches full executable paths instead of truncated names or bare substrings. - slotmigrate: the partition grow gets the same 2048-sector tolerance as the filesystem branch (an exact compare rewrote the MBR at S02 on every boot on disks where the grow cannot land on the last sector), verifies it made progress, and the resize2fs retry is bounded at three attempts with the counter kept on p3 itself. - Installer: archive product/platform are verified after the signature, and a validly signed OLDER release now requires an explicit yes instead of installing as a silent downgrade. All predictable /tmp paths in the installer and ffboot are mktemp now. - release.sh rejects multiple positional versions (the last one used to win silently) and a release without factory-era verification dies unless explicitly bypassed; mkfw.sh refuses to pack when the public key for the post-sign self-check is missing. - forgefirm-logrotate: size-capped rotation (boot + hourly) for the /data logs - a full /data breaks settings, update staging, and the controllers own writes. - Bench build scripts derive every path from their own location or FF_SRC_TOP/FF_BUILD_TOP and log to mktemp files.
2026-09-28 09:11:11 -07:00 · 2026-08-14 18:23:13 -04:00
parent cbabf7d111
commit 58d4d91948
15 changed files with 301 additions and 45 deletions
@@ -5,7 +5,10 @@ require forgefirm-app.inc
 inherit update-rc.d

 INITSCRIPT_NAME = "gfcloud"
-INITSCRIPT_PARAMS = "defaults 92"
+# stop 80 < forgectrl's 90: at runlevel 0/6 the controller goes down
+# BEFORE the daemon that carries the cooling engine, fire gates, and
+# broker - never the other way around.
+INITSCRIPT_PARAMS = "start 92 2 3 4 5 . stop 80 0 1 6 ."

 do_install() {
    install -Dm 0755 ${S}/forgefirm-app/gfcloud.py ${D}${sbindir}/gfcloud.py
@@ -0,0 +1,43 @@
+#!/bin/sh
+### BEGIN INIT INFO
+# Provides:          forgefirm-logrotate
+# Required-Start:    $local_fs
+# Required-Stop:
+# Default-Start:     2 3 4 5
+# Default-Stop:
+# Short-Description: rotate the /data logs at boot and hourly
+### END INIT INFO
+
+# The image runs no cron daemon, so rotation is driven here: once at
+# every boot (bounds carry-over bloat), then hourly from a small
+# background loop (bounds growth across long uptimes - a full /data
+# breaks settings, update staging, and the controllers' own writes).
+
+LOGROTATE=/usr/sbin/logrotate
+CONF=/etc/logrotate.conf
+PIDFILE=/var/run/forgefirm-logrotate.pid
+
+case "$1" in
+  start)
+    [ -x "$LOGROTATE" ] || exit 0
+    "$LOGROTATE" "$CONF" 2>/dev/null
+    ( while :; do
+        sleep 3600
+        "$LOGROTATE" "$CONF" 2>/dev/null
+      done ) &
+    echo $! > "$PIDFILE"
+    ;;
+  stop)
+    [ -r "$PIDFILE" ] && kill "$(cat "$PIDFILE")" 2>/dev/null
+    rm -f "$PIDFILE"
+    ;;
+  restart)
+    $0 stop
+    $0 start
+    ;;
+  *)
+    echo "Usage: $0 {start|stop|restart}"
+    exit 1
+    ;;
+esac
+exit 0
@@ -0,0 +1,12 @@
+# ForgeFIRM logs on the persistent /data partition. copytruncate: the
+# daemon and both controllers keep their log fds open for their whole
+# lifetime, so the file must be truncated in place, never moved out
+# from under them.
+/data/forgectrl.log /data/glowforge.log /data/gfcloud.log /data/gfhome.log {
+    size 1M
+    rotate 3
+    copytruncate
+    compress
+    missingok
+    notifempty
+}
@@ -0,0 +1,26 @@
+SUMMARY = "Log rotation for the ForgeFIRM logs on /data"
+DESCRIPTION = "Size-capped rotation for the daemon and controller logs \
+on the persistent /data partition: a full /data breaks settings, update \
+staging, and the controllers' own writes. Rotation runs at every boot \
+and hourly while the machine is up."
+LICENSE = "MIT"
+LIC_FILES_CHKSUM = "file://${COMMON_LICENSE_DIR}/MIT;md5=0835ade698e0bcf8506ecda2f7b4f302"
+
+SRC_URI = " \
+    file://forgefirm.logrotate \
+    file://forgefirm-logrotate.init \
+"
+
+S = "${WORKDIR}"
+
+inherit update-rc.d
+
+INITSCRIPT_NAME = "forgefirm-logrotate"
+INITSCRIPT_PARAMS = "defaults 40"
+
+RDEPENDS:${PN} = "logrotate"
+
+do_install() {
+    install -Dm 0644 ${WORKDIR}/forgefirm.logrotate ${D}${sysconfdir}/logrotate.d/forgefirm
+    install -Dm 0755 ${WORKDIR}/forgefirm-logrotate.init ${D}${sysconfdir}/init.d/forgefirm-logrotate
+}
@@ -20,7 +20,10 @@ S = "${WORKDIR}/git"
 inherit cmake update-rc.d

 INITSCRIPT_NAME = "grblhal"
-INITSCRIPT_PARAMS = "defaults 92"
+# stop 80 < forgectrl's 90: at runlevel 0/6 the controller goes down
+# BEFORE the daemon that carries the cooling engine, fire gates, and
+# broker - never the other way around.
+INITSCRIPT_PARAMS = "start 92 2 3 4 5 . stop 80 0 1 6 ."

 do_install:append() {
    install -d ${D}${sysconfdir}/init.d
@@ -11,21 +11,54 @@
 # The controller lifecycle belongs to the forgectrl supervisor: it
 # spawns the controller selected by controller_mode as a direct child
 # (that relationship carries the pulse-device fd and death detection).
-# This script remains only as a manual emergency stop.
+# This script is the manual emergency lever, routed THROUGH the
+# supervisor: stop asks it to stop the child and hold it down (a bare
+# pkill would be safed and respawned seconds later); start resumes
+# supervision.
+
+DAEMON_BIN=/usr/bin/grblHAL_glowforge
+TOKEN_FILE=/data/forgefirm/panel.token
+CTRL_URL="http://127.0.0.1:${FORGECTRL_PORT:-8080}"
+
+super_post() {
+  [ -r "$TOKEN_FILE" ] || return 1
+  curl -sf -m 20 -X POST \
+    -H "X-ForgeFIRM-Token: $(cat "$TOKEN_FILE")" \
+    "$CTRL_URL/controller/$1" >/dev/null 2>&1
+}

 case "$1" in
  start)
-    echo "grblHAL is managed by forgectrl (POST /mode to switch controllers)"
+    if super_post start; then
+      echo "grblHAL supervision resumed (forgectrl spawns the selected controller)"
+    else
+      echo "grblHAL is managed by forgectrl (POST /mode to switch controllers)"
+    fi
    ;;
  stop)
    echo "Stopping grblHAL"
-    pkill -x grblHAL_glowfor 2>/dev/null
+    if ! super_post stop; then
+      # Supervisor unreachable: kill the controller directly. If a
+      # supervisor is alive after all, the child's exit still lands on
+      # its safing path. Match the full executable path, not the
+      # 15-char comm truncation.
+      pkill -f "^$DAEMON_BIN" 2>/dev/null
+    fi
+    ;;
+  status)
+    if pgrep -f "^$DAEMON_BIN" >/dev/null 2>&1; then
+      echo "grblHAL is running"
+      exit 0
+    else
+      echo "grblHAL is not running"
+      exit 3
+    fi
    ;;
  restart)
    $0 stop
    ;;
  *)
-    echo "Usage: $0 {start|stop|restart}"
+    echo "Usage: $0 {start|stop|status|restart}"
    exit 1
    ;;
 esac
@@ -23,7 +23,9 @@ IMAGE_INSTALL:remove = "gfui-client"
 # to the inactive rootfs slot.
 # ffboot: boot-slot inventory and switching (also ships fw_env.config).
 # slotmigrate: boot-time reclaim of the legacy p4 layout (grows /data).
-IMAGE_INSTALL:append = " grblhal-glowforge forgectrl gfhome gfcloud v4l-utils fwup ffboot slotmigrate"
+# forgefirm-logrotate: size-capped rotation of the /data logs (boot +
+# hourly; a full /data breaks settings, updates, and controller writes).
+IMAGE_INSTALL:append = " grblhal-glowforge forgectrl gfhome gfcloud v4l-utils fwup ffboot slotmigrate forgefirm-logrotate"

 # NXP's firmware EULA covers the i.MX VPU/EPDC blobs the BSP installs, so the
 # image ships the license text with them (/usr/share/licenses/firmware-imx).
@@ -57,11 +57,20 @@ P3_SIZE=$(sfdisk -d "$DISK" 2>/dev/null | sed -n "s|^${P3} .*size=[ ]*\([0-9]*\)
 [ -n "$DISK_SECT" ] && [ -n "$P3_START" ] && [ -n "$P3_SIZE" ] \
  || { log "cannot read disk/p3 geometry"; exit 0; }

-if [ $((P3_START + P3_SIZE)) -lt "$DISK_SECT" ]; then
-  log "growing p3 to the end of the disk ($((P3_START + P3_SIZE)) -> $DISK_SECT sectors)"
+# 2048-sector tolerance (mirrors the filesystem branch): on a disk where
+# the grow cannot land exactly on the last sector, an exact comparison
+# would rewrite the MBR at S02 on EVERY boot - and a power loss inside
+# that window costs the partition table and /data.
+if [ $((P3_START + P3_SIZE)) -lt $((DISK_SECT - 2048)) ]; then
+  log "growing p3 toward the end of the disk ($((P3_START + P3_SIZE)) -> $DISK_SECT sectors)"
  echo ", +" | sfdisk --no-reread --force -N 3 "$DISK" >/dev/null 2>&1 \
    || { log "p3 grow FAILED"; exit 0; }
  partx -u --nr 3 "$DISK" 2>/dev/null
+  P3_NEW=$(sfdisk -d "$DISK" 2>/dev/null | sed -n "s|^${P3} .*size=[ ]*\([0-9]*\),.*|\1|p")
+  if [ -n "$P3_NEW" ] && [ "$P3_NEW" = "$P3_SIZE" ]; then
+    log "p3 grow made no progress ($P3_SIZE sectors); leaving the table alone"
+    exit 0
+  fi
 fi

 # --- filesystem ----------------------------------------------------------
@@ -71,8 +80,42 @@ FS_BLOCKS=$(tune2fs -l "$P3" 2>/dev/null | sed -n 's/^Block count:[ ]*//p')
 FS_BSIZE=$(tune2fs -l "$P3" 2>/dev/null | sed -n 's/^Block size:[ ]*//p')
 [ -n "$FS_BLOCKS" ] && [ -n "$FS_BSIZE" ] || { log "cannot read p3 filesystem"; exit 0; }

+# Bounded retry: a resize that keeps failing must not cost a full
+# e2fsck pass on every boot forever. Nothing else is mounted this
+# early, so the attempt counter lives on p3 itself.
+TRY_FILE=.slotmigrate-resize-tries
+read_tries () {
+  TRIES=0
+  T=$(mktemp -d) || return
+  if mount "$P3" "$T" 2>/dev/null; then
+    TRIES=$(cat "$T/$TRY_FILE" 2>/dev/null)
+    umount "$T" 2>/dev/null
+  fi
+  rmdir "$T" 2>/dev/null
+  case "$TRIES" in
+    ''|*[!0-9]*) TRIES=0 ;;
+  esac
+}
+write_tries () {
+  T=$(mktemp -d) || return
+  if mount "$P3" "$T" 2>/dev/null; then
+    if [ "$1" -gt 0 ]; then
+      echo "$1" > "$T/$TRY_FILE"
+    else
+      rm -f "$T/$TRY_FILE"
+    fi
+    umount "$T" 2>/dev/null
+  fi
+  rmdir "$T" 2>/dev/null
+}
+
 FS_SECT=$((FS_BLOCKS * (FS_BSIZE / 512)))
 if [ "$FS_SECT" -lt $((PART_SECT - 2048)) ]; then
+  read_tries
+  if [ "$TRIES" -ge 3 ]; then
+    log "resize2fs failed $TRIES times; giving up (grow /data manually with resize2fs $P3)"
+    exit 0
+  fi
  log "growing /data filesystem ($FS_SECT -> $PART_SECT sectors)"
  e2fsck -f -p "$P3" >/dev/null 2>&1
  RC=$?
@@ -80,9 +123,13 @@ if [ "$FS_SECT" -lt $((PART_SECT - 2048)) ]; then
    log "e2fsck found errors (rc=$RC), NOT resizing"
    exit 0
  fi
-  resize2fs "$P3" >/dev/null 2>&1 \
-    && log "/data grown to full size" \
-    || log "resize2fs FAILED (will retry next boot)"
+  if resize2fs "$P3" >/dev/null 2>&1; then
+    log "/data grown to full size"
+    write_tries 0
+  else
+    log "resize2fs FAILED (attempt $((TRIES + 1)) of 3)"
+    write_tries $((TRIES + 1))
+  fi
 fi

 exit 0