1 // Copyright (C) The Arvados Authors. All rights reserved.
3 // SPDX-License-Identifier: AGPL-3.0
21 lockprefix = "crunch-run-"
25 // procinfo is saved in each process's lockfile.
26 type procinfo struct {
31 // Detach acquires a lock for the given uuid, and starts the current
32 // program as a child process (with -no-detach prepended to the given
33 // arguments so the child knows not to detach again). The lock is
34 // passed along to the child process.
36 // Stdout and stderr in the child process are sent to the systemd
37 // journal using the systemd-cat program.
38 func Detach(uuid string, args []string, stdout, stderr io.Writer) int {
39 return exitcode(stderr, detach(uuid, args, stdout, stderr))
41 func detach(uuid string, args []string, stdout, stderr io.Writer) error {
42 lockfile, err := func() (*os.File, error) {
43 // We must hold the dir-level lock between
44 // opening/creating the lockfile and acquiring LOCK_EX
45 // on it, to avoid racing with the ListProcess's
46 // alive-checking and garbage collection.
47 dirlock, err := lockall()
52 lockfilename := filepath.Join(lockdir, lockprefix+uuid+locksuffix)
53 lockfile, err := os.OpenFile(lockfilename, os.O_CREATE|os.O_RDWR, 0700)
55 return nil, fmt.Errorf("open %s: %s", lockfilename, err)
57 err = syscall.Flock(int(lockfile.Fd()), syscall.LOCK_EX|syscall.LOCK_NB)
60 return nil, fmt.Errorf("lock %s: %s", lockfilename, err)
67 defer lockfile.Close()
70 cmd := exec.Command("systemd-cat", append([]string{"--identifier=crunch-run", args[0], "-no-detach"}, args[1:]...)...)
71 // Child inherits lockfile.
72 cmd.ExtraFiles = []*os.File{lockfile}
73 // Ensure child isn't interrupted even if we receive signals
74 // from parent (sshd) while sending lockfile content to
76 cmd.SysProcAttr = &syscall.SysProcAttr{Setpgid: true}
79 return fmt.Errorf("exec %s: %s", cmd.Path, err)
82 w := io.MultiWriter(stdout, lockfile)
83 return json.NewEncoder(w).Encode(procinfo{
89 // KillProcess finds the crunch-run process corresponding to the given
90 // uuid, and sends the given signal to it. It then waits up to 1
91 // second for the process to die. It returns 0 if the process is
92 // successfully killed or didn't exist in the first place.
93 func KillProcess(uuid string, signal syscall.Signal, stdout, stderr io.Writer) int {
94 return exitcode(stderr, kill(uuid, signal, stdout, stderr))
97 func kill(uuid string, signal syscall.Signal, stdout, stderr io.Writer) error {
98 path := filepath.Join(lockdir, lockprefix+uuid+locksuffix)
99 f, err := os.Open(path)
100 if os.IsNotExist(err) {
102 } else if err != nil {
103 return fmt.Errorf("open %s: %s", path, err)
108 err = json.NewDecoder(f).Decode(&pi)
110 return fmt.Errorf("decode %s: %s\n", path, err)
113 if pi.UUID != uuid || pi.PID == 0 {
114 return fmt.Errorf("%s: bogus procinfo: %+v", path, pi)
117 proc, err := os.FindProcess(pi.PID)
119 return fmt.Errorf("%s: find process %d: %s", uuid, pi.PID, err)
122 err = proc.Signal(signal)
123 for deadline := time.Now().Add(time.Second); err == nil && time.Now().Before(deadline); time.Sleep(time.Second / 100) {
124 err = proc.Signal(syscall.Signal(0))
127 return fmt.Errorf("%s: pid %d: sent signal %d (%s) but process is still alive", uuid, pi.PID, signal, signal)
129 fmt.Fprintf(stderr, "%s: pid %d: %s\n", uuid, pi.PID, err)
133 // List UUIDs of active crunch-run processes.
134 func ListProcesses(stdout, stderr io.Writer) int {
135 // filepath.Walk does not follow symlinks, so we must walk
136 // lockdir+"/." in case lockdir itself is a symlink.
137 walkdir := lockdir + "/."
138 return exitcode(stderr, filepath.Walk(walkdir, func(path string, info os.FileInfo, err error) error {
139 if info.IsDir() && path != walkdir {
140 return filepath.SkipDir
142 if name := info.Name(); !strings.HasPrefix(name, lockprefix) || !strings.HasSuffix(name, locksuffix) {
145 if info.Size() == 0 {
146 // race: process has opened/locked but hasn't yet written pid/uuid
150 f, err := os.Open(path)
156 // Ensure other processes don't acquire this lockfile
157 // after we have decided it is abandoned but before we
159 dirlock, err := lockall()
163 err = syscall.Flock(int(f.Fd()), syscall.LOCK_SH|syscall.LOCK_NB)
166 err := os.Remove(path)
169 fmt.Fprintf(stderr, "unlink %s: %s\n", f.Name(), err)
176 err = json.NewDecoder(f).Decode(&pi)
178 fmt.Fprintf(stderr, "%s: %s\n", path, err)
181 if pi.UUID == "" || pi.PID == 0 {
182 fmt.Fprintf(stderr, "%s: bogus procinfo: %+v", path, pi)
186 fmt.Fprintln(stdout, pi.UUID)
191 // If err is nil, return 0 ("success"); otherwise, print err to stderr
193 func exitcode(stderr io.Writer, err error) int {
195 fmt.Fprintln(stderr, err)
201 // Acquire a dir-level lock. Must be held while creating or deleting
202 // container-specific lockfiles, to avoid races during the intervals
203 // when those container-specific lockfiles are open but not locked.
205 // Caller releases the lock by closing the returned file.
206 func lockall() (*os.File, error) {
207 lockfile := filepath.Join(lockdir, lockprefix+"all"+locksuffix)
208 f, err := os.OpenFile(lockfile, os.O_CREATE|os.O_RDWR, 0700)
210 return nil, fmt.Errorf("open %s: %s", lockfile, err)
212 err = syscall.Flock(int(f.Fd()), syscall.LOCK_EX)
215 return nil, fmt.Errorf("lock %s: %s", lockfile, err)