1 // Copyright (C) The Arvados Authors. All rights reserved.
3 // SPDX-License-Identifier: AGPL-3.0
21 lockprefix = "crunch-run-"
23 brokenfile = "crunch-run-broken"
26 // procinfo is saved in each process's lockfile.
27 type procinfo struct {
32 // Detach acquires a lock for the given uuid, and starts the current
33 // program as a child process (with -no-detach prepended to the given
34 // arguments so the child knows not to detach again). The lock is
35 // passed along to the child process.
37 // Stdout and stderr in the child process are sent to the systemd
38 // journal using the systemd-cat program.
39 func Detach(uuid string, args []string, stdout, stderr io.Writer) int {
40 return exitcode(stderr, detach(uuid, args, stdout, stderr))
42 func detach(uuid string, args []string, stdout, stderr io.Writer) error {
43 lockfile, err := func() (*os.File, error) {
44 // We must hold the dir-level lock between
45 // opening/creating the lockfile and acquiring LOCK_EX
46 // on it, to avoid racing with the ListProcess's
47 // alive-checking and garbage collection.
48 dirlock, err := lockall()
53 lockfilename := filepath.Join(lockdir, lockprefix+uuid+locksuffix)
54 lockfile, err := os.OpenFile(lockfilename, os.O_CREATE|os.O_RDWR, 0700)
56 return nil, fmt.Errorf("open %s: %s", lockfilename, err)
58 err = syscall.Flock(int(lockfile.Fd()), syscall.LOCK_EX|syscall.LOCK_NB)
61 return nil, fmt.Errorf("lock %s: %s", lockfilename, err)
68 defer lockfile.Close()
71 cmd := exec.Command("systemd-cat", append([]string{"--identifier=crunch-run", args[0], "-no-detach"}, args[1:]...)...)
72 // Child inherits lockfile.
73 cmd.ExtraFiles = []*os.File{lockfile}
74 // Ensure child isn't interrupted even if we receive signals
75 // from parent (sshd) while sending lockfile content to
77 cmd.SysProcAttr = &syscall.SysProcAttr{Setpgid: true}
80 return fmt.Errorf("exec %s: %s", cmd.Path, err)
83 w := io.MultiWriter(stdout, lockfile)
84 return json.NewEncoder(w).Encode(procinfo{
90 // KillProcess finds the crunch-run process corresponding to the given
91 // uuid, and sends the given signal to it. It then waits up to 1
92 // second for the process to die. It returns 0 if the process is
93 // successfully killed or didn't exist in the first place.
94 func KillProcess(uuid string, signal syscall.Signal, stdout, stderr io.Writer) int {
95 return exitcode(stderr, kill(uuid, signal, stdout, stderr))
98 func kill(uuid string, signal syscall.Signal, stdout, stderr io.Writer) error {
99 path := filepath.Join(lockdir, lockprefix+uuid+locksuffix)
100 f, err := os.Open(path)
101 if os.IsNotExist(err) {
103 } else if err != nil {
104 return fmt.Errorf("open %s: %s", path, err)
109 err = json.NewDecoder(f).Decode(&pi)
111 return fmt.Errorf("decode %s: %s\n", path, err)
114 if pi.UUID != uuid || pi.PID == 0 {
115 return fmt.Errorf("%s: bogus procinfo: %+v", path, pi)
118 proc, err := os.FindProcess(pi.PID)
120 // FindProcess should have succeeded, even if the
121 // process does not exist.
122 return fmt.Errorf("%s: find process %d: %s", uuid, pi.PID, err)
125 // Send the requested signal once, then send signal 0 a few
126 // times. When proc.Signal() returns an error (process no
127 // longer exists), return success. If that doesn't happen
128 // within 1 second, return an error.
129 err = proc.Signal(signal)
130 for deadline := time.Now().Add(time.Second); err == nil && time.Now().Before(deadline); time.Sleep(time.Second / 100) {
131 err = proc.Signal(syscall.Signal(0))
134 // Reached deadline without a proc.Signal() error.
135 return fmt.Errorf("%s: pid %d: sent signal %d (%s) but process is still alive", uuid, pi.PID, signal, signal)
137 fmt.Fprintf(stderr, "%s: pid %d: %s\n", uuid, pi.PID, err)
141 // List UUIDs of active crunch-run processes.
142 func ListProcesses(stdout, stderr io.Writer) int {
143 // filepath.Walk does not follow symlinks, so we must walk
144 // lockdir+"/." in case lockdir itself is a symlink.
145 walkdir := lockdir + "/."
146 return exitcode(stderr, filepath.Walk(walkdir, func(path string, info os.FileInfo, err error) error {
147 if info.IsDir() && path != walkdir {
148 return filepath.SkipDir
150 if name := info.Name(); name == brokenfile {
151 fmt.Fprintln(stdout, "broken")
153 } else if !strings.HasPrefix(name, lockprefix) || !strings.HasSuffix(name, locksuffix) {
156 if info.Size() == 0 {
157 // race: process has opened/locked but hasn't yet written pid/uuid
161 f, err := os.Open(path)
167 // Ensure other processes don't acquire this lockfile
168 // after we have decided it is abandoned but before we
170 dirlock, err := lockall()
174 err = syscall.Flock(int(f.Fd()), syscall.LOCK_SH|syscall.LOCK_NB)
177 err := os.Remove(path)
180 fmt.Fprintf(stderr, "unlink %s: %s\n", f.Name(), err)
187 err = json.NewDecoder(f).Decode(&pi)
189 fmt.Fprintf(stderr, "%s: %s\n", path, err)
192 if pi.UUID == "" || pi.PID == 0 {
193 fmt.Fprintf(stderr, "%s: bogus procinfo: %+v", path, pi)
197 fmt.Fprintln(stdout, pi.UUID)
202 // If err is nil, return 0 ("success"); otherwise, print err to stderr
204 func exitcode(stderr io.Writer, err error) int {
206 fmt.Fprintln(stderr, err)
212 // Acquire a dir-level lock. Must be held while creating or deleting
213 // container-specific lockfiles, to avoid races during the intervals
214 // when those container-specific lockfiles are open but not locked.
216 // Caller releases the lock by closing the returned file.
217 func lockall() (*os.File, error) {
218 lockfile := filepath.Join(lockdir, lockprefix+"all"+locksuffix)
219 f, err := os.OpenFile(lockfile, os.O_CREATE|os.O_RDWR, 0700)
221 return nil, fmt.Errorf("open %s: %s", lockfile, err)
223 err = syscall.Flock(int(f.Fd()), syscall.LOCK_EX)
226 return nil, fmt.Errorf("lock %s: %s", lockfile, err)