feat(package): survive attended sysupgrade - registry self-heal + real v3 feed

.github/workflows/openwrt-package.yml

@@ -249,15 +249,17 @@ jobs:
     needs: [package, package-luci]
     runs-on: ubuntu-24.04
     container:
-      image: alpine:3.21
+      # apk index needs apk-tools 3.x to read OpenWrt 25 .apk packages;
+      # alpine:3.21 ships 2.14 and fails every package with IO ERROR (#296).
+      image: alpine:edge
     if: github.event_name != 'workflow_dispatch' || inputs.version != ''
     permissions:
       contents: write
       pages: write
     steps:
       - name: Install dependencies
         run: |
-          apk add --no-cache git nodejs npm openssl abuild apk-tools curl github-cli
+          apk add --no-cache git nodejs npm openssl curl github-cli zstd tar bash
 
       - uses: actions/checkout@v4
 
@@ -285,21 +287,30 @@ jobs:
 
       - name: Set up signing key
         env:
-          APK_SIGN_KEY: ${{ secrets.APK_SIGN_KEY }}
+          NETGRIP_FEED_KEY: ${{ secrets.NETGRIP_FEED_KEY }}
         run: |
-          if [ -z "$APK_SIGN_KEY" ]; then
-            echo "::error::APK_SIGN_KEY secret not set. Generate a key with: abuild-keygen -n"
-            echo "::error::Then add the private key content as APK_SIGN_KEY in repo secrets."
+          if [ -z "$NETGRIP_FEED_KEY" ]; then
+            echo "::error::NETGRIP_FEED_KEY secret not set. Generate a key with:"
+            echo "::error::  usign -G -s netgrip-feed-sec -p deploy/openwrt/keys/netgrip-feed.pub"
+            echo "::error::Then add the secret key content as NETGRIP_FEED_KEY in repo secrets."
             exit 1
           fi
           mkdir -p ~/.abuild
-          echo "$APK_SIGN_KEY" > ~/.abuild/netgrip-signing-key.rsa
-          chmod 600 ~/.abuild/netgrip-signing-key.rsa
+          echo "$NETGRIP_FEED_KEY" > ~/.abuild/netgrip-feed-sec
+          chmod 600 ~/.abuild/netgrip-feed-sec
 
       - name: Sign APKs and generate index
         run: |
           chmod +x deploy/openwrt/sign-apk.sh
-          ./deploy/openwrt/sign-apk.sh feed ~/.abuild/netgrip-signing-key.rsa
+          # The feed index must be built with the OpenWrt apk-tools: the
+          # Alpine one (also 3.x) speaks a different package format and
+          # fails on every apk with "file format not supported" (#296).
+          SDK_URL_DIR="https://downloads.openwrt.org/releases/25.12.5/targets/x86/64/"
+          SDK_NAME="$(curl -fsSL "$SDK_URL_DIR" | grep -o 'openwrt-sdk-[^"]*\.tar\.zst' | head -1)"
+          curl -fsSL "${SDK_URL_DIR}${SDK_NAME}" -o sdk.tar.zst
+          tar --zstd -xf sdk.tar.zst
+          export PATH="${PWD}/${SDK_NAME%.tar.zst}/staging_dir/host/bin:$PATH"
+          ./deploy/openwrt/sign-apk.sh feed ~/.abuild/netgrip-feed-sec deploy/openwrt/keys/netgrip-feed.pub
 
       - name: Upload feed artifact
         uses: actions/upload-artifact@v4

cmd/netgrip/main.go

@@ -16,9 +16,14 @@ var version = "dev"
 
 func main() {
 	listen := flag.String("listen", "0.0.0.0", "listen address")
+	showVersion := flag.Bool("version", false, "print version and exit")
 	port := flag.Int("port", 8090, "listen port")
 	rpcdURL := flag.String("rpcd-url", auth.DefaultRPCdURL, "rpcd JSON-RPC endpoint used for login validation")
 	flag.Parse()
+	if *showVersion {
+		fmt.Println(version)
+		return
+	}
 
 	// The flag always has a value (its default), so only treat it as an
 	// explicit override when it differs from the default endpoint.

deploy/openwrt/keys/netgrip-feed.pub

@@ -0,0 +1,2 @@
+untrusted comment: netgrip apk feed
+RWT74wNSfNVjV5+i6KacA0hd9nHVN17YDGuZ0nH/Jx5rgjadiww/WyQv

deploy/openwrt/netgrip/Makefile

@@ -32,6 +32,7 @@ define Package/netgrip/install
 	$(INSTALL_BIN) ./files/etc/init.d/netgrip $(1)/etc/init.d/netgrip
 	$(INSTALL_DIR) $(1)/usr/libexec
 	$(INSTALL_BIN) ./files/usr/libexec/netgrip-restore-rules $(1)/usr/libexec/netgrip-restore-rules
+	$(INSTALL_BIN) ./files/usr/libexec/netgrip-heal-register $(1)/usr/libexec/netgrip-heal-register
 endef
 
 define Package/netgrip/postinst
@@ -75,7 +76,7 @@ if [ -z "$${IPKG_INSTROOT}" ]; then
 	# Survive sysupgrades: the apk registry does not survive, but the
 	# preserved files do, so procd starts the panel on first boot.
 	# /etc/netgrip/ keeps the netpulse embedded-agent env across upgrades.
-	for f in /usr/sbin/netgrip /etc/init.d/netgrip /etc/rc.d/S99netgrip /usr/libexec/netgrip-restore-rules /etc/netgrip/; do
+	for f in /usr/sbin/netgrip /etc/init.d/netgrip /etc/rc.d/S99netgrip /usr/libexec/netgrip-restore-rules /usr/libexec/netgrip-heal-register /etc/netgrip/; do
 		grep -qxF "$$f" /etc/sysupgrade.conf 2>/dev/null || echo "$$f" >> /etc/sysupgrade.conf
 	done
 fi

deploy/openwrt/netgrip/files/netgrip.init

@@ -25,4 +25,8 @@ start() {
 	start_service
 	# Reapply MAC ACL and storm control rules after boot
 	[ -x /usr/libexec/netgrip-restore-rules ] && /usr/libexec/netgrip-restore-rules &
+	# Restore the package registry entry after attended sysupgrades (#296):
+	# owut/ASU drops packages outside official feeds from the image, files
+	# survive via /etc/sysupgrade.conf but the apk/opkg registry does not.
+	[ -x /usr/libexec/netgrip-heal-register ] && /usr/libexec/netgrip-heal-register &
 }

deploy/openwrt/netgrip/files/usr/libexec/netgrip-heal-register

@@ -0,0 +1,114 @@
+#!/bin/sh
+# netgrip-heal-register: restore the package registry entry after an
+# attended sysupgrade (#296).
+#
+# owut/ASU only builds images with packages from official feeds, so an
+# attended upgrade drops netgrip from the image. The files survive the
+# flash via /etc/sysupgrade.conf (postinst adds them) and procd starts the
+# panel on first boot, but the apk/opkg registry entry is gone: the package
+# stops being "installed" and package managers never upgrade it again.
+#
+# This script runs in the background from the init script and reinstalls
+# the running version from the latest GitHub release:
+#   - apk (OpenWrt 25): the matching netgrip-<ver>-r1-<arch>.apk asset,
+#     installed with --allow-untrusted (custom feeds would need a
+#     PGP-signed packages.adb, which the CI cannot produce yet)
+#   - opkg (OpenWrt 24): the matching netgrip_<ver>-1_<arch>.ipk asset
+# Dev builds newer than the latest release are left alone. Failures are
+# logged and retried on the next boot. Never blocks the panel.
+
+PKG=netgrip
+BIN=/usr/sbin/netgrip
+RELEASES_API=https://api.github.com/repos/gnacho/netgrip/releases/latest
+
+log() { logger -t netgrip-heal "$*"; }
+
+lock=/var/run/netgrip-heal.pid
+if [ -f "$lock" ] && kill -0 "$(cat "$lock" 2>/dev/null)" 2>/dev/null; then
+	exit 0
+fi
+echo $$ > "$lock"
+trap 'rm -f "$lock"' EXIT INT TERM
+
+registered() {
+	if command -v apk >/dev/null 2>&1; then
+		apk info "$PKG" >/dev/null 2>&1
+	else
+		opkg list-installed "$PKG" 2>/dev/null | grep -q .
+	fi
+}
+
+if registered; then
+	exit 0
+fi
+
+ver=$("$BIN" -version 2>/dev/null)
+log "registry entry lost (running ${ver:-unknown}): restoring"
+
+# Resolve the release asset for this flavor and architecture.
+if command -v apk >/dev/null 2>&1; then
+	arch=$(apk --print-arch 2>/dev/null)
+	case "$arch" in
+		aarch64*) asset=netgrip-*-arm64.apk ;;
+		x86_64*) asset=netgrip-*-amd64.apk ;;
+		*) log "no release asset for apk arch $arch"; exit 1 ;;
+	esac
+	pkgfile=/tmp/netgrip-heal.apk
+else
+	arch=$(opkg print-architecture 2>/dev/null | awk 'tolower($1)=="arch"{print $2; exit}')
+	[ -n "$arch" ] || arch=$(opkg status 2>/dev/null | sed -n 's/^Architecture: *//p' | head -n1)
+	[ -n "$arch" ] || { log "cannot determine architecture"; exit 1; }
+	asset="netgrip*_${arch}.ipk"
+	pkgfile=/tmp/netgrip-heal.ipk
+fi
+
+# Wait for WAN, up to ~3 minutes.
+i=0
+while [ "$i" -lt 36 ]; do
+	if wget -q -O /dev/null -T 5 https://api.github.com 2>/dev/null; then
+		break
+	fi
+	sleep 5
+	i=$((i + 1))
+done
+
+json=$(wget -q -O - -T 15 "$RELEASES_API" 2>/dev/null)
+[ -n "$json" ] || { log "cannot reach GitHub releases API"; exit 1; }
+
+url=$(printf '%s' "$json" | tr ',' '\n' | grep -o '"browser_download_url": *"[^"]*"' | cut -d'"' -f4 | grep -E "/$(printf '%s' "$asset" | sed 's/\*/[^/]*/g')$" | head -n1)
+[ -n "$url" ] || { log "no release asset matching $asset"; exit 1; }
+
+# Never downgrade a dev build: only heal when the asset matches the
+# running version.
+if [ -n "$ver" ]; then
+	case "$url" in
+		*"netgrip-${ver}-"*|*"netgrip_${ver}-"*) ;;
+		*)
+			log "latest release does not match running $ver: skipping (reinstall manually)"
+			exit 0
+			;;
+	esac
+fi
+
+if ! wget -q -O "$pkgfile" -T 120 "$url"; then
+	rm -f "$pkgfile"
+	log "download failed: $url"
+	exit 1
+fi
+
+if command -v apk >/dev/null 2>&1; then
+	if apk add --allow-untrusted "$pkgfile" >/dev/null 2>&1; then
+		rm -f "$pkgfile"
+		log "restored from $url"
+		exit 0
+	fi
+else
+	if opkg install "$pkgfile" >/dev/null 2>&1; then
+		rm -f "$pkgfile"
+		log "restored from $url"
+		exit 0
+	fi
+fi
+rm -f "$pkgfile"
+log "install failed"
+exit 1

deploy/openwrt/package.sh

@@ -69,6 +69,8 @@ cp "$REPO_ROOT/deploy/openwrt/netgrip/files/netgrip.init" "$PKG_DIR/etc/init.d/n
 chmod 755 "$PKG_DIR/etc/init.d/netgrip"
 cp "$REPO_ROOT/deploy/openwrt/netgrip/files/netgrip-restore-rules" "$PKG_DIR/usr/libexec/netgrip-restore-rules"
 chmod 755 "$PKG_DIR/usr/libexec/netgrip-restore-rules"
+cp "$REPO_ROOT/deploy/openwrt/netgrip/files/usr/libexec/netgrip-heal-register" "$PKG_DIR/usr/libexec/netgrip-heal-register"
+chmod 755 "$PKG_DIR/usr/libexec/netgrip-heal-register"
 
 # CONTROL files
 cat > "$PKG_DIR/CONTROL/control" << CTRL
@@ -125,7 +127,7 @@ fi
 # Survive sysupgrades: the apk registry does not survive, but the
 # preserved files do, so procd starts the panel on first boot.
 # /etc/netgrip/ keeps the netpulse embedded-agent env across upgrades.
-for f in /usr/sbin/netgrip /etc/init.d/netgrip /etc/rc.d/S99netgrip /usr/libexec/netgrip-restore-rules /etc/netgrip/; do
+for f in /usr/sbin/netgrip /etc/init.d/netgrip /etc/rc.d/S99netgrip /usr/libexec/netgrip-restore-rules /usr/libexec/netgrip-heal-register /etc/netgrip/; do
   grep -qxF "$f" /etc/sysupgrade.conf 2>/dev/null || echo "$f" >> /etc/sysupgrade.conf
 done
 exit 0

deploy/openwrt/sign-apk.sh

@@ -1,46 +1,66 @@
 #!/bin/sh
-# sign-apk.sh - Generates and signs a signed APKINDEX for OpenWrt .apk feeds.
+# sign-apk.sh - Builds the OpenWrt 25 .apk feed index and signs it.
 #
 # Usage:
-#   ./deploy/openwrt/sign-apk.sh [output-dir] [private-key]
+#   sign-apk.sh [output-dir] [usign-secret-key] [public-key]
 #
-# Requires: apk-tools (3.x), abuild-sign
+# Requires: apk-tools 3 built for OpenWrt (the SDK host build; the Alpine
+# apk-tools speaks a different package format) and usign, both under
+# staging_dir/host/bin of an extracted OpenWrt SDK.
 #
-# For apk v3 (OpenWrt 25.12+) packages are pre-built by the SDK and do not need
-# per-package repacking/signature. The feed only needs a signed APKINDEX.tar.gz.
-# This script:
-#   1. Generates APKINDEX.tar.gz from all *.apk in output-dir
-#   2. Signs APKINDEX.tar.gz with the private key
-#   3. Extracts the matching public key into output-dir
+# Produces in output-dir:
+#   packages.adb      - repository index (v3), the file repositories.d
+#                       entries must point at
+#   packages.adb.asc  - usign detached signature of the index
+#   <public-key copy> - the matching public key, for /etc/apk/keys
+#
+# The release apks are signed by the SDK build keys, which this run does
+# not trust; the index is built with --allow-untrusted and the trust lives
+# in the usign signature of packages.adb.
+#
+# NOTE (#296): OpenWrt 25.12 routers currently only verify indexes signed
+# with the official OpenWrt build-system PGP key, so third-party adb feeds
+# are not verifiable on-device yet. The structure published here is the
+# correct feed layout; routers can still install the apks directly with
+# `apk add --allow-untrusted`, which is what netgrip-heal-register does.
 
 set -eu
 
 out_dir="${1:-feed}"
-privkey="${2:-~/.abuild/netgrip-signing-key.rsa}"
+privkey="${2:?usage: sign-apk.sh <output-dir> <usign-secret-key> <public-key>}"
+pubkey="${3:?usage: sign-apk.sh <output-dir> <usign-secret-key> <public-key>}"
 
 if [ ! -d "$out_dir" ]; then
   echo "Error: output directory not found: $out_dir" >&2
   exit 1
 fi
 
 if [ ! -f "$privkey" ]; then
-  echo "Error: Private key not found: $privkey" >&2
+  echo "Error: usign secret key not found: $privkey" >&2
+  exit 1
+fi
+
+if [ ! -f "$pubkey" ]; then
+  echo "Error: public key not found: $pubkey" >&2
   exit 1
 fi
 
+# Resolve inputs to absolute paths before changing directory.
+out_dir="$(cd "$(dirname "$out_dir")" && pwd)/$(basename "$out_dir")"
+pubkey="$(cd "$(dirname "$pubkey")" && pwd)/$(basename "$pubkey")"
+
 cd "$out_dir"
 
-echo "==> Generating APKINDEX"
-apk index -o APKINDEX.tar.gz --description "NetGrip $(date +%Y-%m-%d)" *.apk || {
-  echo "Warning: apk index failed, continuing without index" >&2
-  exit 0
-}
+echo "==> Generating packages.adb"
+# Fail closed: a feed without a signed index is a broken feed. This used
+# to warn and exit 0, which published apk files nobody could install (#296).
+apk mkndx --allow-untrusted -o packages.adb *.apk
 
-echo "==> Signing APKINDEX"
-abuild-sign -k "$privkey" APKINDEX.tar.gz
+echo "==> Signing packages.adb"
+usign -S -s "$privkey" -m packages.adb -x packages.adb.asc
 
-echo "==> Extracting public key"
-openssl rsa -in "$privkey" -pubout -out netgrip.rsa.pub 2>/dev/null
+echo "==> Publishing public key"
+cp "$pubkey" .
 
 echo "==> Done"
-echo "Signed index: $out_dir/APKINDEX.tar.gz"
+echo "Index: $out_dir/packages.adb (signed: packages.adb.asc)"