2 * Copyright 2006-2007 Free Software Foundation, Inc.
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
18 * Reap any leftover children possibly holding file descriptors.
19 * Children are identified by the stale file descriptor or PGID / SID.
20 * Both can be missed but only the stale file descriptors are important for us.
21 * PGID / SID may be set by the children on their own.
22 * If we fine a candidate we kill it will all its process tree (grandchildren).
23 * The child process is run with `2>&1' redirection (due to forkpty(3)).
24 * 2007-07-10 Jan Kratochvil <jan.kratochvil@redhat.com>
32 #include <sys/types.h>
46 #define LENGTH(x) (sizeof (x) / sizeof (*(x)))
48 static const char *progname;
50 static volatile int signal_chld_hit = 0;
52 static void signal_chld (int signo)
57 static volatile int signal_alrm_hit = 0;
59 static void signal_alrm (int signo)
64 static char childptyname[LINE_MAX];
67 static void print_child_error (const char *reason, char **argv)
71 fprintf (stderr, "%s: %d %s:", progname, (int) child, reason);
72 for (sp = argv; *sp != NULL; sp++)
80 static int read_out (int amaster)
85 buf_got = read (amaster, buf, sizeof buf);
88 /* Weird but at least after POLLHUP we get EIO instead of just EOF. */
89 if (buf_got == -1 && errno == EIO)
91 if (buf_got == -1 && errno == EAGAIN)
95 perror ("read (amaster)");
98 if (write (STDOUT_FILENO, buf, buf_got) != buf_got)
106 static int spawn (char **argv, int timeout)
109 int status, amaster, i, rc;
110 struct sigaction act;
112 struct termios termios;
115 /* We do not use signal(2) to be sure we do not have SA_RESTART. */
116 memset (&act, 0, sizeof (act));
117 act.sa_handler = signal_chld;
118 i = sigemptyset (&act.sa_mask);
120 act.sa_flags = 0; /* !SA_RESTART */
121 i = sigaction (SIGCHLD, &act, NULL);
124 i = sigemptyset (&set);
126 i = sigaddset (&set, SIGCHLD);
128 i = sigprocmask (SIG_SETMASK, &set, NULL);
131 /* With TERMP passed as NULL we get "\n" -> "\r\n". */
132 termios.c_iflag = IGNBRK | IGNPAR;
134 termios.c_cflag = CS8 | CREAD | CLOCAL | HUPCL | B9600;
135 termios.c_lflag = IEXTEN | NOFLSH;
136 memset (termios.c_cc, _POSIX_VDISABLE, sizeof (termios.c_cc));
137 termios.c_cc[VTIME] = 0;
138 termios.c_cc[VMIN ] = 1;
139 cfmakeraw (&termios);
141 /* Workaround a readline deadlock bug in _get_tty_settings(). */
142 termios.c_lflag &= ~FLUSHO;
144 child = forkpty (&amaster, childptyname, &termios, NULL);
148 perror ("forkpty(3)");
151 /* Do not replace STDIN as inferiors query its termios. */
153 i = close (STDIN_FILENO);
155 i = open ("/dev/null", O_RDONLY);
156 assert (i == STDIN_FILENO);
159 /* Do not setpgrp(2) in the parent process as the process-group
160 is shared for the whole sh(1) pipeline we could be a part
161 of. The process-group is set according to PID of the first
162 command in the pipeline.
163 We would rip even vi(1) in the case of:
164 ./orphanripper sh -c 'sleep 1&' | vi -
166 /* Do not setpgrp(2) as our pty would not be ours and we would
167 get `SIGSTOP' later, particularly after spawning gdb(1).
168 setsid(3) was already executed by forkpty(3) and it would fail if
170 if (getpid() != getpgrp ())
172 perror ("getpgrp(2)");
175 execvp (argv[0], argv);
176 perror ("execvp(2)");
181 i = fcntl (amaster, F_SETFL, O_RDWR | O_NONBLOCK);
184 perror ("fcntl (amaster, F_SETFL, O_NONBLOCK)");
188 /* We do not use signal(2) to be sure we do not have SA_RESTART. */
189 act.sa_handler = signal_alrm;
190 i = sigaction (SIGALRM, &act, NULL);
193 alarm_orig = alarm (timeout);
194 assert (alarm_orig == 0);
196 i = sigemptyset (&set);
199 while (!signal_alrm_hit)
201 struct pollfd pollfd;
204 pollfd.events = POLLIN;
205 i = ppoll (&pollfd, 1, NULL, &set);
206 if (i == -1 && errno == EINTR && signal_chld_hit)
209 /* Data available? Process it first. */
210 if (pollfd.revents & POLLIN)
212 if (!read_out (amaster))
214 fprintf (stderr, "%s: Unexpected EOF\n", progname);
218 if (pollfd.revents & POLLHUP)
220 if ((pollfd.revents &= ~POLLIN) != 0)
222 fprintf (stderr, "%s: ppoll(2): revents 0x%x\n", progname,
223 (unsigned) pollfd.revents);
233 i = kill (child, SIGKILL);
239 /* WNOHANG still could fail. */
240 child_got = waitpid (child, &status, 0);
241 if (child != child_got)
243 fprintf (stderr, "waitpid (%d) = %d: %m\n", (int) child, (int) child_got);
250 if (asprintf (&buf, "Timed out after %d seconds", timeout) != -1)
252 print_child_error (buf, argv);
257 else if (WIFEXITED (status))
258 rc = WEXITSTATUS (status);
259 else if (WIFSIGNALED (status))
261 print_child_error (strsignal (WTERMSIG (status)), argv);
262 rc = 128 + WTERMSIG (status);
264 else if (WIFSTOPPED (status))
266 fprintf (stderr, "waitpid (%d): WIFSTOPPED - WSTOPSIG is %d\n",
267 (int) child, WSTOPSIG (status));
272 fprintf (stderr, "waitpid (%d): !WIFEXITED (%d)\n", (int) child, status);
276 /* In the POLLHUP case we may not have seen SIGCHLD so far. */
277 i = sigprocmask (SIG_SETMASK, &set, NULL);
280 assert (signal_chld_hit != 0);
282 /* Do not unset O_NONBLOCK as a stale child (the whole purpose of this
283 program) having open its output pty would block us in read_out. */
285 i = fcntl (amaster, F_SETFL, O_RDONLY /* !O_NONBLOCK */);
288 perror ("fcntl (amaster, F_SETFL, O_RDONLY /* !O_NONBLOCK */)");
293 while (read_out (amaster));
295 /* Do not close the master FD as the child would have `/dev/pts/23 (deleted)'
296 entries which are not expected (and expecting ` (deleted)' would be
302 perror ("close (forkpty ()'s amaster)");
310 /* Detected commandline may look weird due to a race:
312 ./orphanripper sh -c 'sleep 1&' &
315 ./orphanripper: Killed -9 orphan PID 29612 (PGID 29611): sleep 1
316 Raced output (sh(1) child still did not update its argv[]):
318 ./orphanripper: Killed -9 orphan PID 29615 (PGID 29614): sh -c sleep 1&
319 We could delay a bit before ripping the children. */
320 static const char *read_cmdline (pid_t pid)
322 char cmdline_fname[32];
323 static char cmdline[LINE_MAX];
328 if (snprintf (cmdline_fname, sizeof cmdline_fname, "/proc/%d/cmdline",
331 fd = open (cmdline_fname, O_RDONLY);
334 /* It may have already exited - ENOENT. */
336 fprintf (stderr, "%s: open (\"%s\"): %m\n", progname, cmdline_fname);
340 got = read (fd, cmdline, sizeof (cmdline) - 1);
342 fprintf (stderr, "%s: read (\"%s\"): %m\n", progname,
345 fprintf (stderr, "%s: close (\"%s\"): %m\n", progname,
349 /* Convert '\0' argument delimiters to spaces. */
350 for (s = cmdline; s < cmdline + got; s++)
353 /* Trim the trailing spaces (typically single '\0'->' '). */
354 while (s > cmdline && isspace (s[-1]))
360 static int dir_scan (const char *dirname,
361 int (*callback) (struct dirent *dirent, const char *pathname))
364 struct dirent *dirent;
367 dir = opendir (dirname);
370 if (errno == EACCES || errno == ENOENT)
372 fprintf (stderr, "%s: opendir (\"%s\"): %m\n", progname, dirname);
375 while ((errno = 0, dirent = readdir (dir)))
377 char pathname[LINE_MAX];
380 pathname_len = snprintf (pathname, sizeof pathname, "%s/%s",
381 dirname, dirent->d_name);
382 if (pathname_len <= 0 || pathname_len >= (int) sizeof pathname)
384 fprintf (stderr, "entry file name too long: `%s' / `%s'\n",
385 dirname, dirent->d_name);
388 /* RHEL-4.5 on s390x never fills in D_TYPE. */
389 if (dirent->d_type == DT_UNKNOWN)
394 /* We are not interested in the /proc/PID/fd/ links targets. */
395 i = lstat (pathname, &statbuf);
398 if (errno == EACCES || errno == ENOENT)
400 fprintf (stderr, "%s: stat (\"%s\"): %m\n", progname, pathname);
403 if (S_ISDIR (statbuf.st_mode))
404 dirent->d_type = DT_DIR;
405 if (S_ISLNK (statbuf.st_mode))
406 dirent->d_type = DT_LNK;
407 /* No other D_TYPE types used in this code. */
409 rc = (*callback) (dirent, pathname);
418 fprintf (stderr, "%s: readdir (\"%s\"): %m\n", progname, dirname);
421 if (closedir (dir) != 0)
423 fprintf (stderr, "%s: closedir (\"%s\"): %m\n", progname, dirname);
429 static int fd_fs_scan (pid_t pid, int (*func) (pid_t pid, const char *link))
433 if (snprintf (dirname, sizeof dirname, "/proc/%d/fd", (int) pid) < 0)
435 perror ("snprintf(3)");
439 int callback (struct dirent *dirent, const char *pathname)
444 if ((dirent->d_type != DT_DIR && dirent->d_type != DT_LNK)
445 || (dirent->d_type == DT_DIR && strcmp (dirent->d_name, ".") != 0
446 && strcmp (dirent->d_name, "..") != 0)
447 || (dirent->d_type == DT_LNK && strspn (dirent->d_name, "0123456789")
448 != strlen (dirent->d_name)))
450 fprintf (stderr, "Unexpected entry \"%s\" (d_type %u)"
451 " on readdir (\"%s\"): %m\n",
452 dirent->d_name, (unsigned) dirent->d_type, dirname);
455 if (dirent->d_type == DT_DIR)
457 buf_len = readlink (pathname, buf, sizeof buf - 1);
458 if (buf_len <= 0 || buf_len >= (ssize_t) sizeof buf - 1)
460 if (errno != ENOENT && errno != EACCES)
461 fprintf (stderr, "Error reading link \"%s\": %m\n", pathname);
465 return (*func) (pid, buf);
468 return dir_scan (dirname, callback);
471 static void pid_fs_scan (void (*func) (pid_t pid, void *data), void *data)
473 int callback (struct dirent *dirent, const char *pathname)
475 if (dirent->d_type != DT_DIR
476 || strspn (dirent->d_name, "0123456789") != strlen (dirent->d_name))
478 (*func) (atoi (dirent->d_name), data);
482 dir_scan ("/proc", callback);
485 static int rip_check_ptyname (pid_t pid, const char *link)
487 assert (pid != getpid ());
489 return strcmp (link, childptyname) == 0;
497 static struct pid *pid_list;
499 static int pid_found (pid_t pid)
503 for (entry = pid_list; entry != NULL; entry = entry->next)
504 if (entry->pid == pid)
509 /* Single pass is not enough, a (multithreaded) process was seen to survive.
510 Repeated killing of the same process is not enough, zombies can be killed.
512 static int cleanup_acted;
514 static void pid_record (pid_t pid)
522 entry = malloc (sizeof (*entry));
525 fprintf (stderr, "%s: malloc: %m\n", progname);
529 entry->next = pid_list;
533 static void pid_forall (void (*func) (pid_t pid))
537 for (entry = pid_list; entry != NULL; entry = entry->next)
538 (*func) (entry->pid);
541 /* Returns 0 on failure. */
542 static pid_t pid_get_parent (pid_t pid)
549 if (snprintf (fname, sizeof fname, "/proc/%d/status", (int) pid) < 0)
551 perror ("snprintf(3)");
554 f = fopen (fname, "r");
559 while (errno = 0, fgets (line, sizeof line, f) == line)
561 if (strncmp (line, "PPid:\t", sizeof "PPid:\t" - 1) != 0)
563 retval = atoi (line + sizeof "PPid:\t" - 1);
569 fprintf (stderr, "%s: fgets (\"%s\"): %m\n", progname, fname);
574 fprintf (stderr, "%s: fclose (\"%s\"): %m\n", progname, fname);
580 static void killtree (pid_t pid);
582 static void killtree_pid_fs_scan (pid_t pid, void *data)
584 pid_t parent_pid = *(pid_t *) data;
586 /* Do not optimize it as we could miss some newly spawned processes.
587 Always traverse all the leaves. */
594 if (pid_get_parent (pid) != parent_pid)
600 static void killtree (pid_t pid)
603 pid_fs_scan (killtree_pid_fs_scan, &pid);
606 static void rip_pid_fs_scan (pid_t pid, void *data)
610 /* Shouldn't happen. */
611 if (pid == getpid ())
614 /* Check both PGID and the stale file descriptors. */
615 pgid = getpgid (pid);
617 || fd_fs_scan (pid, rip_check_ptyname) != 0)
621 static void killproc (pid_t pid)
625 cmdline = read_cmdline (pid);
626 /* Avoid printing the message for already gone processes. */
627 if (kill (pid, 0) != 0 && errno == ESRCH)
631 fprintf (stderr, "%s: Killed -9 orphan PID %d: %s\n", progname, (int) pid, cmdline);
632 if (kill (pid, SIGKILL) == 0)
634 else if (errno != ESRCH)
635 fprintf (stderr, "%s: kill (%d, SIGKILL): %m\n", progname, (int) pid);
636 /* RHEL-3 kernels cannot SIGKILL a `T (stopped)' process. */
638 /* Do not waitpid(2) as it cannot be our direct descendant and it gets
639 cleaned up by init(8). */
642 pid_got = waitpid (pid, NULL, 0);
645 fprintf (stderr, "%s: waitpid (%d) != %d: %m\n", progname,
646 (int) pid, (int) pid_got);
652 static void rip (void)
658 usleep (1000000 / 10);
660 pid_fs_scan (rip_pid_fs_scan, NULL);
661 pid_forall (killproc);
663 while (cleanup_acted);
666 int main (int argc, char **argv)
674 if (argc < 1 || strcmp (*argv, "-h") == 0
675 || strcmp (*argv, "--help") == 0)
677 puts ("Syntax: orphanripper [-t <seconds>] <execvp(3) commandline>");
680 if ((*argv)[0] == '-' && (*argv)[1] == 't')
682 char *timeout_s = NULL;
686 else if (isdigit ((*argv)[2]))
687 timeout_s = (*argv) + 2;
688 if (timeout_s != NULL)
694 l = strtol (timeout_s, &endptr, 0);
696 if ((endptr != NULL && *endptr != 0) || timeout < 0 || timeout != l)
698 fprintf (stderr, "%s: Invalid timeout value: %s\n", progname,
705 rc = spawn (argv, timeout);