Files
CarloandAllen Houchins 0594f653dd Propagate errors in macOS FMA install scripts (#50198)
<!-- Add the related story/sub-task/bug number, like Resolves #123, or
remove if NA -->
**Related issue:** Resolves #50056

## Summary

macOS FMA install scripts never checked the exit code of the install
command (`installer -pkg` / `cp -R`) — the script's last statement is
always `relaunch_application`, which exits 0 — so a failed install
exited 0 and Fleet reported it installed.

**Generated scripts.** The generator now propagates failure: both
`installer -pkg` variants end with `|| exit $?`, and the `cp -R` path
exits non-zero on a failed copy, removes the partial copy (so a failed
fresh install isn't inventoried as the new version), and restores the
app it moved aside. Regenerated `outputs/` for non-frozen generated apps
are produced by the `ingest-maintained-apps` job, so they aren't
committed here.

**Custom scripts.** 9 of the 18 custom input scripts had the same bug
and are fixed with the same pattern: Google Chrome, Zoom, Microsoft
Edge, GitHub Desktop, Webex, Cycling '74 Max, Pd, Grammarly Desktop, and
P4V. The DMG-based ones also now fail before removing/moving the
existing app when the mount or staging copy fails, so a bad download
can't leave a host with nothing. Their `outputs/*/darwin.json` are
updated in the same commit (script content + recomputed 8-char sha256
ref, versions untouched), following the precedent of #49033. Docker
Desktop (`set -euo pipefail`), 1Password/Slack/LogiTune (installer is
the last statement), and the rest already propagated errors.

**Frozen apps.** The ingest job never rewrites frozen outputs, so the 10
frozen apps with generated scripts (adobe-acrobat-pro, comet, evernote,
firealpaca, keeper-password-manager, nvidia-geforce-now, pritunl,
vnc-viewer, wins, worksheet-crafter) had the fix applied directly to
their published `darwin.json` scripts — the exact text the current
generator would emit, with pinned versions/URLs/hashes untouched. The
11th frozen app (logi-options+) uses a custom script that was already
correct and in sync.

# Checklist for submitter

- [x] Changes file added for user-visible changes in `changes/`
(`changes/50056-fma-install-scripts-ignore-errors`).
- [x] Untrusted data interpolated into shell scripts is validated
against shell metacharacters. (No new untrusted interpolation: the guard
reuses the same curated cask-derived name the adjacent lines already
interpolate.)

## Testing

- [x] Added/updated automated tests (three generator tests: pkg,
pkg-with-choices, cp-R restore — the last now pins the exact emitted
block).
- [x] All 19 updated output manifests validated: embedded scripts pass
`bash -n`, refs match `sha256(script)[:8]`, refs map stays key-sorted
like Go's encoder.
- [x] QA'd all new/changed functionality manually.


<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit

* **Bug Fixes**
* macOS Fleet-maintained app installations now fail fast when
installers, DMG extraction/mounting, or app copy steps error.
* If an upgrade fails, the system removes any partial app and restores
the previously installed version when available.
* Improved robustness during app staging/copying, including safer
handling of paths with spaces or special characters.
* **Tests**
* Added unit coverage to verify installer failure propagation and
rollback behavior.
* **Documentation**
* Clarified that the install-script error handling applies to both
generated and custom scripts, including already-published frozen apps.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->

---------

Co-authored-by: Allen Houchins <allenhouchins@mac.com>
2026-07-30 14:22:02 -05:00

139 lines
4.6 KiB
Bash

#!/bin/bash
# variables
APPDIR="/Applications/"
TMPDIR=$(dirname "$(realpath "$INSTALLER_PATH")")
# functions
quit_and_track_application() {
local bundle_id="$1"
local var_name="APP_WAS_RUNNING_$(echo "$bundle_id" | tr '.-' '__')"
local timeout_duration=10
# check if the application is running
local app_running
app_running=$(osascript -e "application id \"$bundle_id\" is running" 2>/dev/null)
if [[ "$app_running" != "true" ]]; then
eval "export $var_name=0"
return
fi
local console_user
console_user=$(stat -f "%Su" /dev/console)
if [[ -z "$console_user" || "$console_user" == "root" || "$console_user" == "loginwindow" ]]; then
echo "Not logged into a non-root GUI; skipping quitting application ID '$bundle_id'."
eval "export $var_name=0"
return
fi
# App was running, mark it for relaunch
eval "export $var_name=1"
echo "Application '$bundle_id' was running; will relaunch after installation."
echo "Quitting application '$bundle_id'..."
# try to quit the application within the timeout period
local quit_success=false
SECONDS=0
while (( SECONDS < timeout_duration )); do
if osascript -e "tell application id \"$bundle_id\" to quit" >/dev/null 2>&1; then
if ! pgrep -f "$bundle_id" >/dev/null 2>&1; then
echo "Application '$bundle_id' quit successfully."
quit_success=true
break
fi
fi
sleep 1
done
if [[ "$quit_success" = false ]]; then
echo "Application '$bundle_id' did not quit."
fi
}
relaunch_application() {
local bundle_id="$1"
local var_name="APP_WAS_RUNNING_$(echo "$bundle_id" | tr '.-' '__')"
local was_running
# Check if the app was running before installation
eval "was_running=\$$var_name"
if [[ "$was_running" != "1" ]]; then
return
fi
local console_user
console_user=$(stat -f "%Su" /dev/console)
if [[ -z "$console_user" || "$console_user" == "root" || "$console_user" == "loginwindow" ]]; then
echo "Not logged into a non-root GUI; skipping relaunching application ID '$bundle_id'."
return
fi
echo "Relaunching application '$bundle_id'..."
# Launch the app in the logged-in user's GUI session. Apps launched by root
# won't register with the user's Dock/GUI, so run 'open' as the console user.
# Use 'launchctl asuser' to bootstrap into the console user's Mach namespace
# and GUI session — 'sudo -u' alone doesn't do this, which can cause
# LSOpenURLsWithRole() failures even when 'open' exits 0.
local open_status=0
if [[ $EUID -eq 0 ]]; then
local console_uid
console_uid=$(id -u "$console_user")
/bin/launchctl asuser "$console_uid" sudo -u "$console_user" open -b "$bundle_id" >/dev/null 2>&1 || open_status=$?
else
open -b "$bundle_id" >/dev/null 2>&1 || open_status=$?
fi
if [[ $open_status -eq 0 ]]; then
echo "Application '$bundle_id' relaunched successfully."
else
echo "Failed to relaunch application '$bundle_id'."
fi
}
remove_stale_upgrade_bundles() {
# Webex's own auto-updater stages downloaded updates under
# "Cisco Spark/Webexteams_upgrades_*" (one dir per architecture) and leaves
# older, fully formed Webex.app bundles behind after applying them. osquery's
# apps table indexes those staged bundles by their (old) bundle_short_version
# and bundle_identifier ('Cisco-Systems.Spark'), so version-based patch
# policies keep reporting the host as out of date even after the app in
# /Applications has been updated. Removing them is safe: they are cached,
# already-applied updates that Webex re-downloads as needed, and Webex only
# ever stages versions at or newer than what is installed.
local home
for home in /Users/*; do
[ -d "$home" ] || continue
rm -rf "$home/Library/Application Support/Cisco Spark/Webexteams_upgrades_"*
done
}
# extract contents
MOUNT_POINT=$(mktemp -d /tmp/dmg_mount_XXXXXX)
yes | hdiutil attach -plist -nobrowse -readonly -mountpoint "$MOUNT_POINT" "$INSTALLER_PATH" || exit 1
if ! sudo cp -R "$MOUNT_POINT"/* "$TMPDIR"; then
hdiutil detach "$MOUNT_POINT" || true
exit 1
fi
hdiutil detach "$MOUNT_POINT" || true
# copy to the applications folder
quit_and_track_application 'Cisco-Systems.Spark'
if [ -d "$APPDIR/Webex.app" ]; then
sudo mv "$APPDIR/Webex.app" "$TMPDIR/Webex.app.bkp" || exit $?
fi
if ! sudo cp -R "$TMPDIR/Webex.app" "$APPDIR"; then
# remove the partial copy so a failed install isn't inventoried as the new
# version, then restore the previous version if there was one
sudo rm -rf "$APPDIR/Webex.app"
if [ -d "$TMPDIR/Webex.app.bkp" ]; then
sudo mv "$TMPDIR/Webex.app.bkp" "$APPDIR/Webex.app"
fi
exit 1
fi
remove_stale_upgrade_bundles
relaunch_application 'Cisco-Systems.Spark'