fix(agent): fix pacman update check under systemd sandbox

- store checkupdates' private DB in the agent data dir, since
  ProtectSystem=strict makes /tmp read-only
- sync the private DB every 12h and use `checkupdates -n` in between,
  forcing a sync when the DB is missing so -n never reports a false 0
- set WaitDelay so a timed-out command can't hang on child processes
This commit is contained in:
henrygd
2026-09-25 11:22:38 -04:00
parent c25408651f
commit f50fb4f8e5
3 changed files with 102 additions and 14 deletions

View File

@@ -156,7 +156,7 @@ func NewAgent(dataDir ...string) (agent *Agent, err error) {
slog.Debug("SMART", "err", err) slog.Debug("SMART", "err", err)
} }
agent.packageUpdates = newPackageUpdatesManager() agent.packageUpdates = newPackageUpdatesManager(agent.dataDir)
// initialize GPU manager // initialize GPU manager
agent.gpuManager, err = NewGPUManager() agent.gpuManager, err = NewGPUManager()

View File

@@ -7,6 +7,7 @@ import (
"log/slog" "log/slog"
"os" "os"
"os/exec" "os/exec"
"path/filepath"
"runtime" "runtime"
"slices" "slices"
"strings" "strings"
@@ -19,6 +20,9 @@ import (
const ( const (
defaultPackageUpdatesInterval = time.Hour defaultPackageUpdatesInterval = time.Hour
packageUpdatesTimeout = 5 * time.Minute packageUpdatesTimeout = 5 * time.Minute
// pacmanSyncInterval limits how often checkupdates downloads fresh sync
// databases. Checks in between reuse the last synced copy.
pacmanSyncInterval = 12 * time.Hour
) )
// packageUpdatesCheck returns [total] or [total, security] pending package updates. // packageUpdatesCheck returns [total] or [total, security] pending package updates.
@@ -37,8 +41,8 @@ type packageUpdatesManager struct {
// newPackageUpdatesManager returns nil if disabled or no supported package manager // newPackageUpdatesManager returns nil if disabled or no supported package manager
// is found. Agents running in a container are skipped because the container's // is found. Agents running in a container are skipped because the container's
// package database is not the host's. // package database is not the host's. dataDir holds pacman's private sync databases.
func newPackageUpdatesManager() *packageUpdatesManager { func newPackageUpdatesManager(dataDir string) *packageUpdatesManager {
if runtime.GOOS != "linux" || runningInContainer() { if runtime.GOOS != "linux" || runningInContainer() {
return nil return nil
} }
@@ -54,7 +58,7 @@ func newPackageUpdatesManager() *packageUpdatesManager {
slog.Warn("Invalid PACKAGE_UPDATES_INTERVAL", "value", env) slog.Warn("Invalid PACKAGE_UPDATES_INTERVAL", "value", env)
} }
} }
name, check := detectPackageManager() name, check := detectPackageManager(dataDir)
if check == nil { if check == nil {
return nil return nil
} }
@@ -97,7 +101,7 @@ func runningInContainer() bool {
return false return false
} }
func detectPackageManager() (string, packageUpdatesCheck) { func detectPackageManager(dataDir string) (string, packageUpdatesCheck) {
switch { switch {
case commandExists("apt-get"): case commandExists("apt-get"):
return "apt", checkApt return "apt", checkApt
@@ -106,7 +110,7 @@ func detectPackageManager() (string, packageUpdatesCheck) {
case commandExists("zypper"): case commandExists("zypper"):
return "zypper", checkZypper return "zypper", checkZypper
case commandExists("checkupdates"): case commandExists("checkupdates"):
return "pacman", checkPacman return "pacman", newPacmanCheck(dataDir)
case commandExists("apk"): case commandExists("apk"):
return "apk", checkApk return "apk", checkApk
} }
@@ -121,8 +125,17 @@ func commandExists(name string) bool {
// runPackageCommand runs a read-only package manager command and returns stdout. // runPackageCommand runs a read-only package manager command and returns stdout.
// okCodes lists non-zero exit codes that still mean success. // okCodes lists non-zero exit codes that still mean success.
func runPackageCommand(ctx context.Context, okCodes []int, name string, args ...string) (string, error) { func runPackageCommand(ctx context.Context, okCodes []int, name string, args ...string) (string, error) {
return runPackageCommandEnv(ctx, nil, okCodes, name, args...)
}
// runPackageCommandEnv is runPackageCommand with extra environment variables.
func runPackageCommandEnv(ctx context.Context, env []string, okCodes []int, name string, args ...string) (string, error) {
cmd := exec.CommandContext(ctx, name, args...) cmd := exec.CommandContext(ctx, name, args...)
cmd.Env = append(os.Environ(), "LC_ALL=C") cmd.Env = append(os.Environ(), "LC_ALL=C")
cmd.Env = append(cmd.Env, env...)
// checkupdates is a shell script, so a timeout kills only the script and its
// children can keep stdout open. WaitDelay stops Output from waiting on them.
cmd.WaitDelay = 10 * time.Second
out, err := cmd.Output() out, err := cmd.Output()
if exitErr, ok := errors.AsType[*exec.ExitError](err); ok && slices.Contains(okCodes, exitErr.ExitCode()) { if exitErr, ok := errors.AsType[*exec.ExitError](err); ok && slices.Contains(okCodes, exitErr.ExitCode()) {
return string(out), nil return string(out), nil
@@ -168,14 +181,44 @@ func checkZypper(ctx context.Context) ([]uint16, error) {
return []uint16{total, parseZypperTable(out)}, nil return []uint16{total, parseZypperTable(out)}, nil
} }
// checkPacman uses checkupdates (pacman-contrib), which syncs a private copy of // newPacmanCheck uses checkupdates (pacman-contrib), which syncs a private copy of
// the databases and never touches pacman's own. Exit code 2 means no updates. // the databases and never touches pacman's own. The copy lives in dataDir because
func checkPacman(ctx context.Context) ([]uint16, error) { // the systemd unit's ProtectSystem=strict makes the default /tmp location read-only.
out, err := runPackageCommand(ctx, []int{2}, "checkupdates") // It syncs every pacmanSyncInterval and uses the existing copy (-n) in between.
if err != nil { // Local upgrades show up right away since checkupdates links the live local DB.
return nil, err // Exit code 2 means no updates.
func newPacmanCheck(dataDir string) packageUpdatesCheck {
var env []string
var syncDir string
if dataDir != "" {
dbPath := filepath.Join(dataDir, "checkup-db")
env = []string{"CHECKUPDATES_DB=" + dbPath}
syncDir = filepath.Join(dbPath, "sync")
}
// checks never overlap (packageUpdatesManager.running), so no lock is needed
var lastSync time.Time
return func(ctx context.Context) ([]uint16, error) {
// -n with a missing database reports no updates rather than failing,
// so always sync first and whenever the private copy is missing
sync := lastSync.IsZero() || time.Since(lastSync) >= pacmanSyncInterval
if !sync && syncDir != "" {
if _, err := os.Stat(syncDir); err != nil {
sync = true
}
}
var args []string
if !sync {
args = append(args, "-n")
}
out, err := runPackageCommandEnv(ctx, env, []int{2}, "checkupdates", args...)
if err != nil {
return nil, err
}
if sync {
lastSync = time.Now()
}
return []uint16{parsePacmanCheckUpdates(out)}, nil
} }
return []uint16{parsePacmanCheckUpdates(out)}, nil
} }
func checkApk(ctx context.Context) ([]uint16, error) { func checkApk(ctx context.Context) ([]uint16, error) {

View File

@@ -7,6 +7,8 @@ import (
"errors" "errors"
"os" "os"
"path/filepath" "path/filepath"
"runtime"
"strings"
"testing" "testing"
"time" "time"
@@ -25,7 +27,7 @@ func readPackageUpdatesTestData(t *testing.T, name string) string {
func TestParseAptSimulate(t *testing.T) { func TestParseAptSimulate(t *testing.T) {
tests := []struct { tests := []struct {
file string file string
total, security uint16 total, security uint16
}{ }{
{"apt_debian12.txt", 44, 5}, {"apt_debian12.txt", 44, 5},
@@ -149,3 +151,46 @@ func TestPackageUpdatesManagerCaching(t *testing.T) {
// failed check clears the counts // failed check clears the counts
assert.Nil(t, pm.get(time.Now())) assert.Nil(t, pm.get(time.Now()))
} }
func TestPacmanCheckSync(t *testing.T) {
if runtime.GOOS == "windows" {
t.Skip("requires a shell script on PATH")
}
binDir := t.TempDir()
dataDir := t.TempDir()
logFile := filepath.Join(binDir, "calls.log")
// fake checkupdates logs its args and db path, and creates the sync dir when syncing
script := `#!/bin/sh
echo "args=[$*] db=$CHECKUPDATES_DB" >> ` + logFile + `
[ "$1" = "-n" ] || mkdir -p "$CHECKUPDATES_DB/sync"
echo "linux 6.1-1 -> 6.2-1"
`
require.NoError(t, os.WriteFile(filepath.Join(binDir, "checkupdates"), []byte(script), 0o755))
t.Setenv("PATH", binDir+string(os.PathListSeparator)+os.Getenv("PATH"))
check := newPacmanCheck(dataDir)
dbPath := filepath.Join(dataDir, "checkup-db")
readCalls := func() []string {
data, err := os.ReadFile(logFile)
require.NoError(t, err)
return strings.Split(strings.TrimSpace(string(data)), "\n")
}
// first check syncs
counts, err := check(context.Background())
require.NoError(t, err)
assert.Equal(t, []uint16{1}, counts)
// later checks reuse the synced copy
_, err = check(context.Background())
require.NoError(t, err)
// a missing private copy forces a sync
require.NoError(t, os.RemoveAll(dbPath))
_, err = check(context.Background())
require.NoError(t, err)
assert.Equal(t, []string{
"args=[] db=" + dbPath,
"args=[-n] db=" + dbPath,
"args=[] db=" + dbPath,
}, readCalls())
}