/*
- * Copyright 2006 Free Software Foundation, Inc.
+ * Copyright 2006-2007 Free Software Foundation, Inc.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
*
* Reap any leftover children possibly holding file descriptors.
- * 2006-12-12 Jan Kratochvil <jan.kratochvil@redhat.com>
+ * Children are identified by the open file descriptor.
+ * PGID or SID may be set by the children on their own.
+ * If we find a child by its file descriptor, we kill it will all its process
+ * tree (grandchildren).
+ * The child process is run with `2>&1' redirection (due to forkpty(3)).
+ * 2007-07-10 Jan Kratochvil <jan.kratochvil@redhat.com>
*/
-
#define _XOPEN_SOURCE 1
#define _XOPEN_SOURCE_EXTENDED 1
#define _BSD_SOURCE 1
#include <string.h>
#include <limits.h>
#include <fcntl.h>
-
+#include <assert.h>
+#include <pty.h>
static const char *progname;
-static int spawn (char **argv, pid_t *child_pointer)
-{
- pid_t child, child_got;
- int status;
-
- switch (child = fork ()) {
- case -1:
- perror ("fork(2)");
- exit (EXIT_FAILURE);
- case 0:
- /* Do not setpgrp(2) in the parent process as the process-group
- is shared for the whole sh(1) pipeline we could be a part
- of. The process-group is set according to PID of the first
- command in the pipeline.
- We would rip even vi(1) in the case of:
- ./orphanripper sh -c 'sleep 1&' | vi -
- */
- /* Do not setpgrp(2) as our pty would not be ours and we would
- get `SIGSTOP' later, particularly after spawning gdb(1). */
- if (getpid() != setsid ()) {
- perror ("setsid");
- exit (EXIT_FAILURE);
- }
- execvp (argv[1], argv + 1);
- perror ("execvp(2)");
- exit (EXIT_FAILURE);
- default:
- break;
- }
- if (child != (child_got = waitpid (child, &status, 0))) {
- fprintf (stderr, "waitpid(%d)=%d: %m\n", (int) child,
- (int) child_got);
- exit (EXIT_FAILURE);
+static int signal_child_hit = 0;
+
+static void signal_child (int signo)
+{
+ signal_child_hit = 1;
+}
+
+static char childptyname[LINE_MAX];
+
+static int spawn (char **argv)
+{
+ pid_t child, child_got;
+ int status, amaster, i;
+ char buf[LINE_MAX];
+ ssize_t buf_got;
+ struct sigaction act;
+
+ /* We do not use signal(2) to be sure we have SA_RESTART unset. */
+ memset (&act, 0, sizeof (act));
+ act.sa_handler = signal_child;
+ i = sigemptyset (&act.sa_mask);
+ if (i != 0)
+ {
+ perror ("sigemotyset(3)");
+ exit (EXIT_FAILURE);
+ }
+ act.sa_flags = 0;
+ i = sigaction (SIGCHLD, &act, NULL);
+ if (i != 0)
+ {
+ perror ("sigaction(2)");
+ exit (EXIT_FAILURE);
+ }
+
+ child = forkpty (&amaster, childptyname, NULL, NULL);
+ switch (child)
+ {
+ case -1:
+ perror ("forkpty(3)");
+ exit (EXIT_FAILURE);
+ case 0:
+ execvp (argv[1], argv + 1);
+ perror ("execvp(2)");
+ exit (EXIT_FAILURE);
+ default:
+ break;
+ }
+ for (;;)
+ {
+ buf_got = read (amaster, buf, sizeof buf);
+ if (buf_got == -1)
+ {
+ assert (signal_child_hit != 0);
+ assert (errno == EINTR || errno == EIO);
}
- if (!WIFEXITED (status)) {
- fprintf (stderr, "waitpid(2): !WIFEXITED(%d)\n", status);
- exit (EXIT_FAILURE);
+ if (buf_got <= 0)
+ break;
+ if (write (STDOUT_FILENO, buf, buf_got) != buf_got)
+ {
+ perror ("write(2)");
+ exit (EXIT_FAILURE);
}
+ }
+ child_got = waitpid (child, &status, 0);
+ if (child != child_got)
+ {
+ fprintf (stderr, "waitpid (%d) = %d: %m\n", (int) child, (int) child_got);
+ exit (EXIT_FAILURE);
+ }
+ if (!WIFEXITED (status))
+ {
+ fprintf (stderr, "waitpid (%d): !WIFEXITED (%d)\n", (int) child, status);
+ exit (EXIT_FAILURE);
+ }
+ /* Do not close the master FD as the child would have `/dev/pts/23 (deleted)'
+ entries which are not expected (and expecting ` (deleted)' would be
+ a race. */
+#if 0
+ i = close (amaster);
+ if (i != 0)
+ {
+ perror ("close (forkpty ()'s amaster)");
+ exit (EXIT_FAILURE);
+ }
+#endif
- if (child_pointer)
- *child_pointer = child;
- return WEXITSTATUS (status);
+ return WEXITSTATUS (status);
}
/* Detected commandline may look weird due to a race:
Original command:
- ./orphanripper sh -c 'sleep 1&' &
+ ./orphanripper sh -c 'sleep 1&' &
Correct output:
- [1] 29610
- ./orphanripper: Killed -9 orphan PID 29612 (PGID 29611): sleep 1
+ [1] 29610
+ ./orphanripper: Killed -9 orphan PID 29612 (PGID 29611): sleep 1
Raced output (sh(1) child still did not update its argv[]):
- [1] 29613
- ./orphanripper: Killed -9 orphan PID 29615 (PGID 29614): sh -c sleep 1&
+ [1] 29613
+ ./orphanripper: Killed -9 orphan PID 29615 (PGID 29614): sh -c sleep 1&
We could delay a bit before ripping the children. */
static const char *read_cmdline (pid_t pid)
{
- char cmdline_fname[32];
- static char cmdline[LINE_MAX];
- int fd;
- ssize_t got;
- char *s;
-
- if (snprintf (cmdline_fname, sizeof (cmdline_fname), "/proc/%d/cmdline",
- (int) pid) < 0)
- return NULL;
- fd = open (cmdline_fname, O_RDONLY);
- if (fd == -1) {
- fprintf (stderr, "%s: open (\"%s\"): %m\n", progname,
- cmdline_fname);
- return NULL;
- }
- got = read (fd, cmdline, sizeof (cmdline) - 1);
- if (got == -1)
- fprintf (stderr, "%s: read (\"%s\"): %m\n", progname,
- cmdline_fname);
- if (close (fd))
- fprintf (stderr, "%s: close (\"%s\"): %m\n", progname,
- cmdline_fname);
- if (got < 0)
- return NULL;
- /* Convert '\0' argument delimiters to spaces. */
- for (s = cmdline; s < cmdline + got; s++)
- if (!*s)
- *s = ' ';
- /* Trim the trailing spaces (typically single '\0'->' '). */
- while (s > cmdline && isspace (s[-1]))
- s--;
- *s = 0;
- return cmdline;
-}
-
-static void rip_pid (pid_t pid, pid_t pgid_child)
-{
- pid_t pgid_pids;
- const char *cmdline;
-
- /* Don't shoot ourselves. */
- if (pid == getpid())
- return;
- pgid_pids = getpgid (pid);
- /* Ignore errors (permissions? untested). */
- if (pgid_pids == -1)
- return;
- /* Not a process of ours. */
- if (pgid_pids != pgid_child)
- return;
-
- cmdline = read_cmdline (pid);
- if (!cmdline)
- cmdline = "<error>";
- fprintf (stderr, "%s: Killed -9 orphan PID %d (PGID %d): %s\n",
- progname, (int) pid, (int) pgid_pids, cmdline);
- if (kill (pid, SIGKILL)) {
- fprintf (stderr, "%s: kill (%d, SIGKILL): %m\n", progname,
- (int) pid);
- return;
- }
- /* Do not waitpid(2) as it cannot be our direct descendant and it gets
- cleaned up by init(8). */
+ char cmdline_fname[32];
+ static char cmdline[LINE_MAX];
+ int fd;
+ ssize_t got;
+ char *s;
+
+ if (snprintf (cmdline_fname, sizeof cmdline_fname, "/proc/%d/cmdline",
+ (int) pid) < 0)
+ return NULL;
+ fd = open (cmdline_fname, O_RDONLY);
+ if (fd == -1)
+ {
+ /* It may have already exited - ENOENT. */
#if 0
- pid_t pid_got;
- if (pid != (pid_got = waitpid (pid, NULL, 0))) {
- fprintf (stderr, "%s: waitpid (%d) != %d: %m\n", progname,
- (int) pid, (int) pid_got);
- return;
- }
+ fprintf (stderr, "%s: open (\"%s\"): %m\n", progname, cmdline_fname);
#endif
+ return NULL;
+ }
+ got = read (fd, cmdline, sizeof (cmdline) - 1);
+ if (got == -1)
+ fprintf (stderr, "%s: read (\"%s\"): %m\n", progname,
+ cmdline_fname);
+ if (close (fd) != 0)
+ fprintf (stderr, "%s: close (\"%s\"): %m\n", progname,
+ cmdline_fname);
+ if (got < 0)
+ return NULL;
+ /* Convert '\0' argument delimiters to spaces. */
+ for (s = cmdline; s < cmdline + got; s++)
+ if (!*s)
+ *s = ' ';
+ /* Trim the trailing spaces (typically single '\0'->' '). */
+ while (s > cmdline && isspace (s[-1]))
+ s--;
+ *s = 0;
+ return cmdline;
}
-static void rip (pid_t pgid_child)
+static int fd_fs_scan (pid_t pid, int (*func) (pid_t pid, const char *link))
{
- DIR *dir;
- struct dirent *dirent;
+ DIR *dir;
+ struct dirent *dirent;
+ char dirname[64];
+ int rc = 0;
- dir = opendir ("/proc");
- if (!dir) {
- perror ("opendir (\"/proc\")");
- exit (EXIT_FAILURE);
+ if (snprintf (dirname, sizeof dirname, "/proc/%d/fd", (int) pid) < 0)
+ {
+ perror ("snprintf(3)");
+ exit (EXIT_FAILURE);
+ }
+ dir = opendir (dirname);
+ if (dir == NULL)
+ {
+ if (errno == EACCES)
+ return 0;
+ fprintf (stderr, "%s: opendir (\"%s\"): %m\n", progname, dirname);
+ exit (EXIT_FAILURE);
+ }
+ while ((errno = 0, dirent = readdir (dir)))
+ {
+ char linkname[LINE_MAX], buf[LINE_MAX];
+ int linkname_len;
+ ssize_t buf_len;
+
+ /* FIXME: POSIX portability. */
+ if ((dirent->d_type != DT_DIR && dirent->d_type != DT_LNK)
+ || (dirent->d_type == DT_DIR && strcmp (dirent->d_name, ".") != 0
+ && strcmp (dirent->d_name, "..") != 0)
+ || (dirent->d_type == DT_LNK && strspn (dirent->d_name, "0123456789")
+ != strlen (dirent->d_name)))
+ {
+ fprintf (stderr, "Unexpected entry \"%s\" on readdir (\"%s\"): %m\n",
+ dirent->d_name, dirname);
+ continue;
}
- while ((errno = 0, dirent = readdir (dir))) {
- const char *cs;
-
- /* FIXME: POSIX portability. */
- if (dirent->d_type != DT_DIR)
- continue;
- /* Check /^\d+$/: */
- for (cs = dirent->d_name; *cs; cs++)
- if (!isdigit (*cs))
- break;
- if (cs == dirent->d_name || *cs)
- continue;
- rip_pid (atoi (dirent->d_name), pgid_child);
+ if (dirent->d_type == DT_DIR)
+ continue;
+ linkname_len = snprintf (linkname, sizeof linkname, "%s/%s", dirname,
+ dirent->d_name);
+ if (linkname_len <= 0 || linkname_len >= sizeof linkname)
+ {
+ fprintf (stderr, "Link content too long: `%s' / `%s'\n",
+ dirent->d_name, dirent->d_name);
+ continue;
}
- if (errno) {
- perror ("readdir (\"/proc\")");
- exit (EXIT_FAILURE);
+ buf_len = readlink (linkname, buf, sizeof buf - 1);
+ if (buf_len <= 0 || buf_len >= sizeof buf - 1)
+ {
+ fprintf (stderr, "Error reading link \"%s\": %m\n",
+ linkname);
+ continue;
}
- if (closedir (dir)) {
- perror ("closedir (\"/proc\")");
- exit (EXIT_FAILURE);
+ buf[buf_len] = 0;
+ rc = (*func) (pid, buf);
+ if (rc != 0)
+ {
+ errno = 0;
+ break;
}
+ }
+ if (errno != 0)
+ {
+ fprintf (stderr, "%s: readdir (\"%s\"): %m\n", progname, dirname);
+ exit (EXIT_FAILURE);
+ }
+ if (closedir (dir) != 0)
+ {
+ fprintf (stderr, "%s: closedir (\"%s\"): %m\n", progname, dirname);
+ exit (EXIT_FAILURE);
+ }
+ return rc;
+}
+static void pid_fs_scan (void (*func) (pid_t pid, void *data), void *data)
+{
+ DIR *dir;
+ struct dirent *dirent;
+
+ dir = opendir ("/proc");
+ if (dir == NULL)
+ {
+ perror ("opendir (\"/proc\")");
+ exit (EXIT_FAILURE);
+ }
+ while ((errno = 0, dirent = readdir (dir)))
+ {
+ /* FIXME: POSIX portability. */
+ if (dirent->d_type != DT_DIR
+ || strspn (dirent->d_name, "0123456789") != strlen (dirent->d_name))
+ continue;
+ (*func) (atoi (dirent->d_name), data);
+ }
+ if (errno != 0)
+ {
+ perror ("readdir (\"/proc\")");
+ exit (EXIT_FAILURE);
+ }
+ if (closedir (dir) != 0)
+ {
+ perror ("closedir (\"/proc\")");
+ exit (EXIT_FAILURE);
+ }
+}
+
+static int rip_check (pid_t pid, const char *link)
+{
+ assert (pid != getpid ());
+
+ return strcmp (link, childptyname) == 0;
+}
+
+struct pid
+ {
+ struct pid *next;
+ pid_t pid;
+ };
+static struct pid *pid_list;
+
+static int pid_found (pid_t pid)
+{
+ struct pid *entry;
+
+ for (entry = pid_list; entry != NULL; entry = entry->next)
+ if (entry->pid == pid)
+ return 1;
+ return 0;
+}
+
+static void pid_record (pid_t pid)
+{
+ struct pid *entry;
+
+ if (pid_found (pid))
+ return;
+
+ entry = malloc (sizeof (*entry));
+ if (entry == NULL)
+ {
+ fprintf (stderr, "%s: malloc: %m\n", progname);
+ exit (EXIT_FAILURE);
+ }
+ entry->pid = pid;
+ entry->next = pid_list;
+ pid_list = entry;
+}
+
+static void pid_forall (void (*func) (pid_t pid))
+{
+ struct pid *entry;
+
+ for (entry = pid_list; entry != NULL; entry = entry->next)
+ (*func) (entry->pid);
+}
+
+/* Returns 0 on failure. */
+static pid_t pid_get_parent (pid_t pid)
+{
+ char fname[64];
+ FILE *f;
+ char line[LINE_MAX];
+ pid_t retval = 0;
+
+ if (snprintf (fname, sizeof fname, "/proc/%d/status", (int) pid) < 0)
+ {
+ perror ("snprintf(3)");
+ exit (EXIT_FAILURE);
+ }
+ f = fopen (fname, "r");
+ if (f == NULL)
+ {
+ return 0;
+ }
+ while (errno = 0, fgets (line, sizeof line, f) == line)
+ {
+ if (strncmp (line, "PPid:\t", sizeof "PPid:\t" - 1) != 0)
+ continue;
+ retval = atoi (line + sizeof "PPid:\t" - 1);
+ errno = 0;
+ break;
+ }
+ if (errno != 0)
+ {
+ fprintf (stderr, "%s: fgets (\"%s\"): %m\n", progname, fname);
+ exit (EXIT_FAILURE);
+ }
+ if (fclose (f) != 0)
+ {
+ fprintf (stderr, "%s: fclose (\"%s\"): %m\n", progname, fname);
+ exit (EXIT_FAILURE);
+ }
+ return retval;
+}
+
+static void killtree (pid_t pid);
+
+static void killtree_pid_fs_scan (pid_t pid, void *data)
+{
+ pid_t parent_pid = *(pid_t *) data;
+
+ /* Optimization. */
+ if (pid_found (pid))
+ return;
+
+ if (pid_get_parent (pid) != parent_pid)
+ return;
+
+ killtree (pid);
+}
+
+static void killtree (pid_t pid)
+{
+ pid_record (pid);
+ pid_fs_scan (killtree_pid_fs_scan, &pid);
+}
+
+static void rip_pid_fs_scan (pid_t pid, void *data)
+{
+ /* Shouldn't happen. */
+ if (pid == getpid ())
+ return;
+
+ if (fd_fs_scan (pid, rip_check) != 0)
+ killtree (pid);
+}
+
+static void killproc (pid_t pid)
+{
+ const char *cmdline;
+
+ cmdline = read_cmdline (pid);
+ if (kill (pid, 0) != 0 && errno == ESRCH)
+ return;
+ if (cmdline == NULL)
+ cmdline = "<error>";
+ fprintf (stderr, "%s: Killed -9 orphan PID %d: %s\n", progname, (int) pid, cmdline);
+ if (kill (pid, SIGKILL)) {
+ fprintf (stderr, "%s: kill (%d, SIGKILL): %m\n", progname,
+ (int) pid);
+ return;
+ }
+ /* Do not waitpid(2) as it cannot be our direct descendant and it gets
+ cleaned up by init(8). */
+#if 0
+ pid_t pid_got;
+ pid_got = waitpid (pid, NULL, 0);
+ if (pid != pid_got)
+ {
+ fprintf (stderr, "%s: waitpid (%d) != %d: %m\n", progname,
+ (int) pid, (int) pid_got);
+ return;
+ }
+#endif
+}
+
+static void rip ()
+{
+ pid_fs_scan (rip_pid_fs_scan, NULL);
+ pid_forall (killproc);
}
int main (int argc, char **argv)
{
- int rc;
- pid_t child;
+ int rc;
- if (argc < 2
- || !strcmp (argv[1], "-h")
- || !strcmp (argv[1], "--help")) {
- fputs("Syntax: orphanripper <execvp(3) commandline>\n", stdout);
- exit (EXIT_FAILURE);
- }
- progname = argv[0];
- rc = spawn (argv, &child);
- rip (child);
- return rc;
+ if (argc < 2 || strcmp (argv[1], "-h") == 0
+ || strcmp (argv[1], "--help") == 0)
+ {
+ fputs ("Syntax: orphanripper <execvp(3) commandline>\n", stdout);
+ exit (EXIT_FAILURE);
+ }
+ progname = argv[0];
+ rc = spawn (argv);
+ rip ();
+ return rc;
}