Files
unraid-podman/plugin/sbin/podman-preflight.sh
T
maggesandClaude Sonnet 5 7f6fcb9166
Build Packages / Build .txz packages (push) Successful in 9m17s
Lint / ShellCheck (push) Successful in 12s
Lint / Validate .plg XML (push) Successful in 13s
Lint / EditorConfig (push) Successful in 6s
Fix real STORAGE_PATH bug; add Start Podman + format-disk from the WebUI
Root cause of a fresh-install "cannot reach the Podman API socket" report
(a friend's Unraid box, cache pool present and mounted): unlike
Docker-for-Unraid's docker.img path, this plugin never auto-created
STORAGE_PATH itself — only podman.img inside it. A perfectly normal,
already-mounted cache pool still failed preflight/storage-create with
"does not exist", just because its own .../system/podman subdirectory
had never been created. Fixed by walking up to the nearest existing
ancestor and checking whether it's on a different device than / (real
mount vs. nothing mounted at all) — see podman-common.sh's new
podman_path_has_real_mount_ancestor(), used by both podman-preflight.sh
and podman-storage.sh.

Settings gets a "Podman Service" card (status chip + Start/Restart,
backed by new ajax/settings.php service_status/start/restart actions
that just shell out to rc.podman) so a fresh install that failed to start
can be diagnosed and retried without SSH/terminal access at all — exactly
what was missing when this was first needed live.

Also adds "Format a Disk for Podman Storage" (new ajax/disks.php) for a
single-disk system with no cache pool at all. Only ever lists disks with
literally no existing partition/filesystem/RAID-or-ZFS-membership
signature and that aren't Unraid's boot flash — found live, twice, during
development: the boot USB (FAT, labeled "UNRAID") passed the initial
mounted-only check because this host's /boot is backed by a ZFS dataset
rather than a direct partition mount, and active RAID-member cache disks
passed a data-vs-blank *warning* rather than a hard exclusion. Both are
now excluded outright, not just flagged — see disks.php's
device_or_children_labeled_unraid() and the hasData exclusion in
list_candidate_disks(). A disk formatted this way is remounted by UUID on
every boot via a new plugin/sbin/podman-mount-managed-disk.sh, called
from plugin/event/disks_mounted before rc.podman start.

Unrelated fix bundled in: scripts/lib/slackbuild-common.sh now sets
SOURCE_DATE_EPOCH (derived from the repo's last commit) before calling
makepkg, so two separate builds of the same commit produce byte-identical
.txz files — makepkg already supports this (`--clamp-mtime` when
$SOURCE_DATE_EPOCH is set, confirmed by reading a real host's
/sbin/makepkg) but nothing was setting the variable, so release.yml's
"rebuild in CI and verify it matches the committed checksums" step was
guaranteed to fail on the first package it checked alphabetically
(observed live: aardvark-dns).

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
2026-07-12 22:55:14 +00:00

159 lines
6.8 KiB
Bash
Executable File

#!/bin/bash
# =============================================================================
# plugin/sbin/podman-preflight.sh
#
# Startup validation, run by `rc.podman start` BEFORE anything is mounted or
# started. The goal is to fail fast with one clear message, instead of
# letting `podman system service` fail three steps later with a cryptic
# error. See docs/ARCHITECTURE.md section 16.1 (Startup-Robustheit).
#
# Every check below is independent and all of them run even if an earlier
# one fails, so a single invocation reports every problem at once rather
# than forcing the user through a fix-one-rerun-find-the-next loop.
#
# Exit code: 0 if every check passed, 1 if any failed (with all failures
# already logged/notified by that point).
# =============================================================================
set -u # deliberately not -e: see the "run every check" note above
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
# shellcheck source=./podman-common.sh
. "$SCRIPT_DIR/podman-common.sh"
podman_load_cfg
FAILURES=0
fail() {
podman_log_error "preflight: $1"
FAILURES=$((FAILURES + 1))
}
ok() {
podman_log "preflight: OK - $1"
}
# -----------------------------------------------------------------------------
# 1. Required binaries are actually installed.
# -----------------------------------------------------------------------------
for bin in podman conmon crun mount umount xfs_repair mkfs.xfs; do
if command -v "$bin" > /dev/null 2>&1; then
ok "$bin found"
else
fail "required binary '$bin' not found on PATH — is the plugin fully installed?"
fi
done
for helper in /usr/libexec/podman/netavark /usr/libexec/podman/aardvark-dns; do
if [ -x "$helper" ]; then
ok "$helper found"
else
fail "required helper '$helper' not found — is the plugin fully installed?"
fi
done
# -----------------------------------------------------------------------------
# 2. Storage path configured, safe, and mounted (cache pool / disk, not
# /mnt/user — see docs/ARCHITECTURE.md section 4.3).
# -----------------------------------------------------------------------------
if podman_storage_path_is_safe "$STORAGE_PATH" 2> /tmp/podman-preflight-storage-safety.log; then
ok "STORAGE_PATH ($STORAGE_PATH) is not under /mnt/user"
else
fail "$(cat /tmp/podman-preflight-storage-safety.log)"
fi
rm -f /tmp/podman-preflight-storage-safety.log
if [ -d "$STORAGE_PATH" ] && mountpoint -q "$STORAGE_PATH" 2> /dev/null; then
ok "STORAGE_PATH ($STORAGE_PATH) is a mounted filesystem"
elif [ -d "$STORAGE_PATH" ]; then
# Not every valid target is a mountpoint itself (e.g. a subdirectory of a
# cache pool root) — warn rather than fail, but only if the parent chain
# is mounted somewhere.
if findmnt -T "$STORAGE_PATH" > /dev/null 2>&1; then
ok "STORAGE_PATH ($STORAGE_PATH) resolves onto a mounted filesystem"
else
fail "STORAGE_PATH ($STORAGE_PATH) does not appear to be on a mounted filesystem"
fi
elif podman_path_has_real_mount_ancestor "$STORAGE_PATH"; then
# The leaf directory doesn't exist yet, but a real filesystem IS mounted
# somewhere along its path (e.g. the cache pool itself) — podman-storage.sh
# create will mkdir -p it. Not a failure; see that check's own comment
# for the live bug this used to cause (a normal, already-mounted cache
# pool failing preflight just because its .../system/podman subdirectory
# had never been created).
ok "STORAGE_PATH ($STORAGE_PATH) doesn't exist yet, but resolves onto a mounted filesystem — will be created"
else
fail "STORAGE_PATH ($STORAGE_PATH) does not exist — is the configured cache pool/disk present and started?"
fi
# -----------------------------------------------------------------------------
# 3. Free space check (only meaningful once STORAGE_PATH exists).
# -----------------------------------------------------------------------------
if [ -d "$STORAGE_PATH" ]; then
available_kb=$(df --output=avail -k "$STORAGE_PATH" 2> /dev/null | tail -n1 | tr -d '[:space:]')
if [ -n "${available_kb:-}" ]; then
available_gb=$((available_kb / 1024 / 1024))
if [ -f "$PODMAN_STORAGE_IMAGE" ]; then
# Image already exists — just warn if the pool itself is nearly full,
# since podman.img growth or new image pulls need headroom too.
if [ "$available_gb" -lt 2 ]; then
fail "less than 2G free on $STORAGE_PATH (${available_gb}G) — image pulls will likely fail"
else
ok "${available_gb}G free on $STORAGE_PATH"
fi
elif [ "$available_gb" -lt "$STORAGE_IMAGE_SIZE_GB" ]; then
fail "not enough free space on $STORAGE_PATH to create a ${STORAGE_IMAGE_SIZE_GB}G podman.img (only ${available_gb}G free)"
else
ok "${available_gb}G free on $STORAGE_PATH (enough for a fresh ${STORAGE_IMAGE_SIZE_GB}G podman.img)"
fi
fi
fi
# -----------------------------------------------------------------------------
# 4. Kernel supports cgroup v2 (required by crun/netavark's expectations).
# -----------------------------------------------------------------------------
if [ -f /sys/fs/cgroup/cgroup.controllers ]; then
ok "cgroup v2 unified hierarchy is active"
else
fail "cgroup v2 unified hierarchy not detected (/sys/fs/cgroup/cgroup.controllers missing) — check Unraid's syslinux cgroup boot parameters"
fi
# -----------------------------------------------------------------------------
# 5. No orphaned socket/pidfile from a previous unclean shutdown.
# -----------------------------------------------------------------------------
if [ -S "$PODMAN_SOCKET" ]; then
if [ -f "$PODMAN_SERVICE_PID_FILE" ] && kill -0 "$(cat "$PODMAN_SERVICE_PID_FILE")" 2> /dev/null; then
fail "podman system service already appears to be running (pid $(cat "$PODMAN_SERVICE_PID_FILE")) — is rc.podman already started?"
else
podman_log "preflight: removing orphaned socket $PODMAN_SOCKET from a previous unclean shutdown"
rm -f "$PODMAN_SOCKET"
ok "cleared orphaned socket"
fi
else
ok "no orphaned podman.sock"
fi
# -----------------------------------------------------------------------------
# 6. Boot-config directory exists (i.e. 'seed' has run at least once).
# -----------------------------------------------------------------------------
if [ -f "$PODMAN_CFG_FILE" ]; then
ok "$PODMAN_CFG_FILE present"
else
fail "$PODMAN_CFG_FILE missing — run 'podman-config.sh seed' (should happen automatically on install)"
fi
# -----------------------------------------------------------------------------
# Summary
# -----------------------------------------------------------------------------
if [ "$FAILURES" -gt 0 ]; then
podman_notify "Podman preflight checks failed" \
"$FAILURES check(s) failed — podman was not started. See $PODMAN_PLUGIN_LOG for details." \
"alert"
podman_log_error "preflight: $FAILURES check(s) failed, aborting start"
exit 1
fi
podman_log "preflight: all checks passed"
exit 0