1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 This file is part of systemd.
5 Copyright 2010 Lennart Poettering
7 systemd is free software; you can redistribute it and/or modify it
8 under the terms of the GNU Lesser General Public License as published by
9 the Free Software Foundation; either version 2.1 of the License, or
10 (at your option) any later version.
12 systemd is distributed in the hope that it will be useful, but
13 WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 Lesser General Public License for more details.
17 You should have received a copy of the GNU Lesser General Public License
18 along with systemd; If not, see <http://www.gnu.org/licenses/>.
24 #include <linux/oom.h>
29 //#include <stdio_ext.h>
33 #include <sys/personality.h>
34 #include <sys/prctl.h>
35 #include <sys/types.h>
39 #if HAVE_VALGRIND_VALGRIND_H
40 #include <valgrind/valgrind.h>
43 #include "alloc-util.h"
44 //#include "architecture.h"
53 #include "process-util.h"
54 //#include "raw-clone.h"
55 #include "signal-util.h"
56 //#include "stat-util.h"
57 #include "string-table.h"
58 #include "string-util.h"
59 //#include "terminal-util.h"
60 #include "user-util.h"
63 int get_process_state(pid_t pid) {
67 _cleanup_free_ char *line = NULL;
71 p = procfs_file_alloca(pid, "stat");
73 r = read_one_line_file(p, &line);
79 p = strrchr(line, ')');
85 if (sscanf(p, " %c", &state) != 1)
88 return (unsigned char) state;
91 int get_process_comm(pid_t pid, char **name) {
98 p = procfs_file_alloca(pid, "comm");
100 r = read_one_line_file(p, name);
107 int get_process_cmdline(pid_t pid, size_t max_length, bool comm_fallback, char **line) {
108 _cleanup_fclose_ FILE *f = NULL;
110 char *k, *ans = NULL;
117 /* Retrieves a process' command line. Replaces unprintable characters while doing so by whitespace (coalescing
118 * multiple sequential ones into one). If max_length is != 0 will return a string of the specified size at most
119 * (the trailing NUL byte does count towards the length here!), abbreviated with a "..." ellipsis. If
120 * comm_fallback is true and the process has no command line set (the case for kernel threads), or has a
121 * command line that resolves to the empty string will return the "comm" name of the process instead.
123 * Returns -ESRCH if the process doesn't exist, and -ENOENT if the process has no command line (and
124 * comm_fallback is false). Returns 0 and sets *line otherwise. */
126 p = procfs_file_alloca(pid, "cmdline");
135 (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
137 if (max_length == 1) {
139 /* If there's only room for one byte, return the empty string */
147 } else if (max_length == 0) {
148 size_t len = 0, allocated = 0;
150 while ((c = getc(f)) != EOF) {
152 if (!GREEDY_REALLOC(ans, allocated, len+3)) {
174 bool dotdotdot = false;
177 ans = new(char, max_length);
183 while ((c = getc(f)) != EOF) {
210 if (max_length <= 4) {
214 k = ans + max_length - 4;
217 /* Eat up final spaces */
218 while (k > ans && isspace(k[-1])) {
224 strncpy(k, "...", left-1);
230 /* Kernel threads have no argv[] */
232 _cleanup_free_ char *t = NULL;
240 h = get_process_comm(pid, &t);
245 ans = strjoin("[", t, "]");
251 if (l + 3 <= max_length)
252 ans = strjoin("[", t, "]");
253 else if (max_length <= 6) {
255 ans = new(char, max_length);
259 memcpy(ans, "[...]", max_length-1);
260 ans[max_length-1] = 0;
264 t[max_length - 6] = 0;
266 /* Chop off final spaces */
268 while (e > t && isspace(e[-1]))
272 ans = strjoin("[", t, "...]");
283 #if 0 /// UNNEEDED by elogind
284 int rename_process(const char name[]) {
285 static size_t mm_size = 0;
286 static char *mm = NULL;
287 bool truncated = false;
290 /* This is a like a poor man's setproctitle(). It changes the comm field, argv[0], and also the glibc's
291 * internally used name of the process. For the first one a limit of 16 chars applies; to the second one in
292 * many cases one of 10 (i.e. length of "/sbin/init") — however if we have CAP_SYS_RESOURCES it is unbounded;
293 * to the third one 7 (i.e. the length of "systemd". If you pass a longer string it will likely be
296 * Returns 0 if a name was set but truncated, > 0 if it was set but not truncated. */
299 return -EINVAL; /* let's not confuse users unnecessarily with an empty name */
301 if (!is_main_thread())
302 return -EPERM; /* Let's not allow setting the process name from other threads than the main one, as we
303 * cache things without locking, and we make assumptions that PR_SET_NAME sets the
304 * process name that isn't correct on any other threads */
308 /* First step, change the comm field. The main thread's comm is identical to the process comm. This means we
309 * can use PR_SET_NAME, which sets the thread name for the calling thread. */
310 if (prctl(PR_SET_NAME, name) < 0)
311 log_debug_errno(errno, "PR_SET_NAME failed: %m");
312 if (l > 15) /* Linux process names can be 15 chars at max */
315 /* Second step, change glibc's ID of the process name. */
316 if (program_invocation_name) {
319 k = strlen(program_invocation_name);
320 strncpy(program_invocation_name, name, k);
325 /* Third step, completely replace the argv[] array the kernel maintains for us. This requires privileges, but
326 * has the advantage that the argv[] array is exactly what we want it to be, and not filled up with zeros at
327 * the end. This is the best option for changing /proc/self/cmdline. */
329 /* Let's not bother with this if we don't have euid == 0. Strictly speaking we should check for the
330 * CAP_SYS_RESOURCE capability which is independent of the euid. In our own code the capability generally is
331 * present only for euid == 0, hence let's use this as quick bypass check, to avoid calling mmap() if
332 * PR_SET_MM_ARG_{START,END} fails with EPERM later on anyway. After all geteuid() is dead cheap to call, but
335 log_debug("Skipping PR_SET_MM, as we don't have privileges.");
336 else if (mm_size < l+1) {
340 nn_size = PAGE_ALIGN(l+1);
341 nn = mmap(NULL, nn_size, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
342 if (nn == MAP_FAILED) {
343 log_debug_errno(errno, "mmap() failed: %m");
347 strncpy(nn, name, nn_size);
349 /* Now, let's tell the kernel about this new memory */
350 if (prctl(PR_SET_MM, PR_SET_MM_ARG_START, (unsigned long) nn, 0, 0) < 0) {
351 log_debug_errno(errno, "PR_SET_MM_ARG_START failed, proceeding without: %m");
352 (void) munmap(nn, nn_size);
356 /* And update the end pointer to the new end, too. If this fails, we don't really know what to do, it's
357 * pretty unlikely that we can rollback, hence we'll just accept the failure, and continue. */
358 if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) nn + l + 1, 0, 0) < 0)
359 log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m");
362 (void) munmap(mm, mm_size);
367 strncpy(mm, name, mm_size);
369 /* Update the end pointer, continuing regardless of any failure. */
370 if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) mm + l + 1, 0, 0) < 0)
371 log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m");
375 /* Fourth step: in all cases we'll also update the original argv[], so that our own code gets it right too if
376 * it still looks here */
378 if (saved_argc > 0) {
384 k = strlen(saved_argv[0]);
385 strncpy(saved_argv[0], name, k);
390 for (i = 1; i < saved_argc; i++) {
394 memzero(saved_argv[i], strlen(saved_argv[i]));
402 int is_kernel_thread(pid_t pid) {
409 if (IN_SET(pid, 0, 1) || pid == getpid_cached()) /* pid 1, and we ourselves certainly aren't a kernel thread */
414 p = procfs_file_alloca(pid, "cmdline");
422 (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
424 count = fread(&c, 1, 1, f);
428 /* Kernel threads have an empty cmdline */
431 return eof ? 1 : -errno;
436 #if 0 /// UNNEEDED by elogind
437 int get_process_capeff(pid_t pid, char **capeff) {
444 p = procfs_file_alloca(pid, "status");
446 r = get_proc_field(p, "CapEff", WHITESPACE, capeff);
454 static int get_process_link_contents(const char *proc_file, char **name) {
460 r = readlink_malloc(proc_file, name);
469 int get_process_exe(pid_t pid, char **name) {
476 p = procfs_file_alloca(pid, "exe");
477 r = get_process_link_contents(p, name);
481 d = endswith(*name, " (deleted)");
488 #if 0 /// UNNEEDED by elogind
489 static int get_process_id(pid_t pid, const char *field, uid_t *uid) {
490 _cleanup_fclose_ FILE *f = NULL;
500 p = procfs_file_alloca(pid, "status");
508 (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
510 FOREACH_LINE(line, f, return -errno) {
515 if (startswith(l, field)) {
517 l += strspn(l, WHITESPACE);
519 l[strcspn(l, WHITESPACE)] = 0;
521 return parse_uid(l, uid);
528 int get_process_uid(pid_t pid, uid_t *uid) {
530 if (pid == 0 || pid == getpid_cached()) {
535 return get_process_id(pid, "Uid:", uid);
538 int get_process_gid(pid_t pid, gid_t *gid) {
540 if (pid == 0 || pid == getpid_cached()) {
545 assert_cc(sizeof(uid_t) == sizeof(gid_t));
546 return get_process_id(pid, "Gid:", gid);
549 int get_process_cwd(pid_t pid, char **cwd) {
554 p = procfs_file_alloca(pid, "cwd");
556 return get_process_link_contents(p, cwd);
559 int get_process_root(pid_t pid, char **root) {
564 p = procfs_file_alloca(pid, "root");
566 return get_process_link_contents(p, root);
569 int get_process_environ(pid_t pid, char **env) {
570 _cleanup_fclose_ FILE *f = NULL;
571 _cleanup_free_ char *outcome = NULL;
574 size_t allocated = 0, sz = 0;
579 p = procfs_file_alloca(pid, "environ");
588 (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
590 while ((c = fgetc(f)) != EOF) {
591 if (!GREEDY_REALLOC(outcome, allocated, sz + 5))
595 outcome[sz++] = '\n';
597 sz += cescape_char(c, outcome + sz);
601 outcome = strdup("");
613 int get_process_ppid(pid_t pid, pid_t *_ppid) {
615 _cleanup_free_ char *line = NULL;
622 if (pid == 0 || pid == getpid_cached()) {
627 p = procfs_file_alloca(pid, "stat");
628 r = read_one_line_file(p, &line);
634 /* Let's skip the pid and comm fields. The latter is enclosed
635 * in () but does not escape any () in its value, so let's
636 * skip over it manually */
638 p = strrchr(line, ')');
650 if ((long unsigned) (pid_t) ppid != ppid)
653 *_ppid = (pid_t) ppid;
659 int wait_for_terminate(pid_t pid, siginfo_t *status) {
670 if (waitid(P_PID, pid, status, WEXITED) < 0) {
675 return negative_errno();
684 * < 0 : wait_for_terminate() failed to get the state of the
685 * process, the process was terminated by a signal, or
686 * failed for an unknown reason.
687 * >=0 : The process terminated normally, and its exit code is
690 * That is, success is indicated by a return value of zero, and an
691 * error is indicated by a non-zero value.
693 * A warning is emitted if the process terminates abnormally,
694 * and also if it returns non-zero unless check_exit_code is true.
696 int wait_for_terminate_and_check(const char *name, pid_t pid, WaitFlags flags) {
697 _cleanup_free_ char *buffer = NULL;
704 r = get_process_comm(pid, &buffer);
706 log_debug_errno(r, "Failed to acquire process name of " PID_FMT ", ignoring: %m", pid);
711 prio = flags & WAIT_LOG_ABNORMAL ? LOG_ERR : LOG_DEBUG;
713 r = wait_for_terminate(pid, &status);
715 return log_full_errno(prio, r, "Failed to wait for %s: %m", strna(name));
717 if (status.si_code == CLD_EXITED) {
718 if (status.si_status != EXIT_SUCCESS)
719 log_full(flags & WAIT_LOG_NON_ZERO_EXIT_STATUS ? LOG_ERR : LOG_DEBUG,
720 "%s failed with exit status %i.", strna(name), status.si_status);
722 log_debug("%s succeeded.", name);
724 return status.si_status;
726 } else if (IN_SET(status.si_code, CLD_KILLED, CLD_DUMPED)) {
728 log_full(prio, "%s terminated by signal %s.", strna(name), signal_to_string(status.si_status));
732 log_full(prio, "%s failed due to unknown reason.", strna(name));
738 * < 0 : wait_for_terminate_with_timeout() failed to get the state of the
739 * process, the process timed out, the process was terminated by a
740 * signal, or failed for an unknown reason.
741 * >=0 : The process terminated normally with no failures.
743 * Success is indicated by a return value of zero, a timeout is indicated
744 * by ETIMEDOUT, and all other child failure states are indicated by error
745 * is indicated by a non-zero value.
747 int wait_for_terminate_with_timeout(pid_t pid, usec_t timeout) {
752 assert_se(sigemptyset(&mask) == 0);
753 assert_se(sigaddset(&mask, SIGCHLD) == 0);
755 /* Drop into a sigtimewait-based timeout. Waiting for the
757 until = now(CLOCK_MONOTONIC) + timeout;
760 siginfo_t status = {};
763 n = now(CLOCK_MONOTONIC);
767 r = sigtimedwait(&mask, NULL, timespec_store(&ts, until - n)) < 0 ? -errno : 0;
768 /* Assuming we woke due to the child exiting. */
769 if (waitid(P_PID, pid, &status, WEXITED|WNOHANG) == 0) {
770 if (status.si_pid == pid) {
771 /* This is the correct child.*/
772 if (status.si_code == CLD_EXITED)
773 return (status.si_status == 0) ? 0 : -EPROTO;
778 /* Not the child, check for errors and proceed appropriately */
782 /* Timed out, child is likely hung. */
785 /* Received a different signal and should retry */
788 /* Return any unexpected errors */
797 #if 0 /// UNNEEDED by elogind
798 void sigkill_wait(pid_t pid) {
801 if (kill(pid, SIGKILL) > 0)
802 (void) wait_for_terminate(pid, NULL);
805 void sigkill_waitp(pid_t *pid) {
816 void sigterm_wait(pid_t pid) {
819 if (kill_and_sigcont(pid, SIGTERM) > 0)
820 (void) wait_for_terminate(pid, NULL);
823 int kill_and_sigcont(pid_t pid, int sig) {
826 r = kill(pid, sig) < 0 ? -errno : 0;
828 /* If this worked, also send SIGCONT, unless we already just sent a SIGCONT, or SIGKILL was sent which isn't
829 * affected by a process being suspended anyway. */
830 if (r >= 0 && !IN_SET(sig, SIGCONT, SIGKILL))
831 (void) kill(pid, SIGCONT);
837 int getenv_for_pid(pid_t pid, const char *field, char **_value) {
838 _cleanup_fclose_ FILE *f = NULL;
849 path = procfs_file_alloca(pid, "environ");
851 f = fopen(path, "re");
858 (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
867 for (i = 0; i < sizeof(line)-1; i++) {
871 if (_unlikely_(c == EOF)) {
881 if (strneq(line, field, l) && line[l] == '=') {
882 value = strdup(line + l + 1);
896 bool pid_is_unwaited(pid_t pid) {
897 /* Checks whether a PID is still valid at all, including a zombie */
902 if (pid <= 1) /* If we or PID 1 would be dead and have been waited for, this code would not be running */
905 if (pid == getpid_cached())
908 if (kill(pid, 0) >= 0)
911 return errno != ESRCH;
914 bool pid_is_alive(pid_t pid) {
917 /* Checks whether a PID is still valid and not a zombie */
922 if (pid <= 1) /* If we or PID 1 would be a zombie, this code would not be running */
925 if (pid == getpid_cached())
928 r = get_process_state(pid);
929 if (IN_SET(r, -ESRCH, 'Z'))
935 #if 0 /// UNNEEDED by elogind
936 int pid_from_same_root_fs(pid_t pid) {
942 if (pid == 0 || pid == getpid_cached())
945 root = procfs_file_alloca(pid, "root");
947 return files_same(root, "/proc/1/root", 0);
951 bool is_main_thread(void) {
952 static thread_local int cached = 0;
954 if (_unlikely_(cached == 0))
955 cached = getpid_cached() == gettid() ? 1 : -1;
960 #if 0 /// UNNEEDED by elogind
961 noreturn void freeze(void) {
965 /* Make sure nobody waits for us on a socket anymore */
966 close_all_fds(NULL, 0);
970 /* Let's not freeze right away, but keep reaping zombies. */
975 r = waitid(P_ALL, 0, &si, WEXITED);
976 if (r < 0 && errno != EINTR)
980 /* waitid() failed with an unexpected error, things are really borked. Freeze now! */
985 bool oom_score_adjust_is_valid(int oa) {
986 return oa >= OOM_SCORE_ADJ_MIN && oa <= OOM_SCORE_ADJ_MAX;
989 unsigned long personality_from_string(const char *p) {
993 return PERSONALITY_INVALID;
995 /* Parse a personality specifier. We use our own identifiers that indicate specific ABIs, rather than just
996 * hints regarding the register size, since we want to keep things open for multiple locally supported ABIs for
997 * the same register size. */
999 architecture = architecture_from_string(p);
1000 if (architecture < 0)
1001 return PERSONALITY_INVALID;
1003 if (architecture == native_architecture())
1005 #ifdef SECONDARY_ARCHITECTURE
1006 if (architecture == SECONDARY_ARCHITECTURE)
1010 return PERSONALITY_INVALID;
1013 const char* personality_to_string(unsigned long p) {
1014 int architecture = _ARCHITECTURE_INVALID;
1017 architecture = native_architecture();
1018 #ifdef SECONDARY_ARCHITECTURE
1019 else if (p == PER_LINUX32)
1020 architecture = SECONDARY_ARCHITECTURE;
1023 if (architecture < 0)
1026 return architecture_to_string(architecture);
1029 int safe_personality(unsigned long p) {
1032 /* So here's the deal, personality() is weirdly defined by glibc. In some cases it returns a failure via errno,
1033 * and in others as negative return value containing an errno-like value. Let's work around this: this is a
1034 * wrapper that uses errno if it is set, and uses the return value otherwise. And then it sets both errno and
1035 * the return value indicating the same issue, so that we are definitely on the safe side.
1037 * See https://github.com/systemd/systemd/issues/6737 */
1040 ret = personality(p);
1051 int opinionated_personality(unsigned long *ret) {
1054 /* Returns the current personality, or PERSONALITY_INVALID if we can't determine it. This function is a bit
1055 * opinionated though, and ignores all the finer-grained bits and exotic personalities, only distinguishing the
1056 * two most relevant personalities: PER_LINUX and PER_LINUX32. */
1058 current = safe_personality(PERSONALITY_INVALID);
1062 if (((unsigned long) current & 0xffff) == PER_LINUX32)
1070 void valgrind_summary_hack(void) {
1071 #if HAVE_VALGRIND_VALGRIND_H
1072 if (getpid_cached() == 1 && RUNNING_ON_VALGRIND) {
1074 pid = raw_clone(SIGCHLD);
1076 log_emergency_errno(errno, "Failed to fork off valgrind helper: %m");
1080 log_info("Spawned valgrind helper as PID "PID_FMT".", pid);
1081 (void) wait_for_terminate(pid, NULL);
1087 int pid_compare_func(const void *a, const void *b) {
1088 const pid_t *p = a, *q = b;
1090 /* Suitable for usage in qsort() */
1099 int ioprio_parse_priority(const char *s, int *ret) {
1105 r = safe_atoi(s, &i);
1109 if (!ioprio_priority_is_valid(i))
1117 /* The cached PID, possible values:
1119 * == UNSET [0] → cache not initialized yet
1120 * == BUSY [-1] → some thread is initializing it at the moment
1121 * any other → the cached PID
1124 #define CACHED_PID_UNSET ((pid_t) 0)
1125 #define CACHED_PID_BUSY ((pid_t) -1)
1127 static pid_t cached_pid = CACHED_PID_UNSET;
1129 void reset_cached_pid(void) {
1130 /* Invoked in the child after a fork(), i.e. at the first moment the PID changed */
1131 cached_pid = CACHED_PID_UNSET;
1134 /* We use glibc __register_atfork() + __dso_handle directly here, as they are not included in the glibc
1135 * headers. __register_atfork() is mostly equivalent to pthread_atfork(), but doesn't require us to link against
1136 * libpthread, as it is part of glibc anyway. */
1138 extern int __register_atfork(void (*prepare) (void), void (*parent) (void), void (*child) (void), void * __dso_handle);
1139 extern void* __dso_handle __attribute__ ((__weak__));
1140 #endif // ifdef __GLIBC__
1142 pid_t getpid_cached(void) {
1143 pid_t current_value;
1145 /* getpid_cached() is much like getpid(), but caches the value in local memory, to avoid having to invoke a
1146 * system call each time. This restores glibc behaviour from before 2.24, when getpid() was unconditionally
1147 * cached. Starting with 2.24 getpid() started to become prohibitively expensive when used for detecting when
1148 * objects were used across fork()s. With this caching the old behaviour is somewhat restored.
1150 * https://bugzilla.redhat.com/show_bug.cgi?id=1443976
1151 * https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=c579f48edba88380635ab98cb612030e3ed8691e
1154 current_value = __sync_val_compare_and_swap(&cached_pid, CACHED_PID_UNSET, CACHED_PID_BUSY);
1156 switch (current_value) {
1158 case CACHED_PID_UNSET: { /* Not initialized yet, then do so now */
1161 new_pid = raw_getpid();
1163 if (__register_atfork(NULL, NULL, reset_cached_pid, __dso_handle) != 0) {
1164 /* OOM? Let's try again later */
1165 cached_pid = CACHED_PID_UNSET;
1169 cached_pid = new_pid;
1173 case CACHED_PID_BUSY: /* Somebody else is currently initializing */
1174 return raw_getpid();
1176 default: /* Properly initialized */
1177 return current_value;
1181 int must_be_root(void) {
1186 log_error("Need to be root.");
1192 const int except_fds[],
1193 size_t n_except_fds,
1197 pid_t original_pid, pid;
1198 sigset_t saved_ss, ss;
1199 bool block_signals = false;
1202 /* A wrapper around fork(), that does a couple of important initializations in addition to mere forking. Always
1203 * returns the child's PID in *ret_pid. Returns == 0 in the child, and > 0 in the parent. */
1205 prio = flags & FORK_LOG ? LOG_ERR : LOG_DEBUG;
1207 original_pid = getpid_cached();
1209 if (flags & (FORK_RESET_SIGNALS|FORK_DEATHSIG)) {
1211 /* We temporarily block all signals, so that the new child has them blocked initially. This way, we can
1212 * be sure that SIGTERMs are not lost we might send to the child. */
1214 if (sigfillset(&ss) < 0)
1215 return log_full_errno(prio, errno, "Failed to reset signal set: %m");
1217 block_signals = true;
1219 } else if (flags & FORK_WAIT) {
1221 /* Let's block SIGCHLD at least, so that we can safely watch for the child process */
1223 if (sigemptyset(&ss) < 0)
1224 return log_full_errno(prio, errno, "Failed to clear signal set: %m");
1226 if (sigaddset(&ss, SIGCHLD) < 0)
1227 return log_full_errno(prio, errno, "Failed to add SIGCHLD to signal set: %m");
1229 block_signals = true;
1233 if (sigprocmask(SIG_SETMASK, &ss, &saved_ss) < 0)
1234 return log_full_errno(prio, errno, "Failed to set signal mask: %m");
1236 if (flags & FORK_NEW_MOUNTNS)
1237 pid = raw_clone(SIGCHLD|CLONE_NEWNS);
1243 if (block_signals) /* undo what we did above */
1244 (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL);
1246 return log_full_errno(prio, r, "Failed to fork: %m");
1249 /* We are in the parent process */
1251 log_debug("Successfully forked off '%s' as PID " PID_FMT ".", strna(name), pid);
1253 if (flags & FORK_WAIT) {
1254 r = wait_for_terminate_and_check(name, pid, (flags & FORK_LOG ? WAIT_LOG : 0));
1257 if (r != EXIT_SUCCESS) /* exit status > 0 should be treated as failure, too */
1261 if (block_signals) /* undo what we did above */
1262 (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL);
1270 /* We are in the child process */
1272 if (flags & FORK_REOPEN_LOG) {
1273 /* Close the logs if requested, before we log anything. And make sure we reopen it if needed. */
1275 log_set_open_when_needed(true);
1279 r = rename_process(name);
1281 log_full_errno(flags & FORK_LOG ? LOG_WARNING : LOG_DEBUG,
1282 r, "Failed to rename process, ignoring: %m");
1285 if (flags & FORK_DEATHSIG)
1286 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0) {
1287 log_full_errno(prio, errno, "Failed to set death signal: %m");
1288 _exit(EXIT_FAILURE);
1291 if (flags & FORK_RESET_SIGNALS) {
1292 r = reset_all_signal_handlers();
1294 log_full_errno(prio, r, "Failed to reset signal handlers: %m");
1295 _exit(EXIT_FAILURE);
1298 /* This implicitly undoes the signal mask stuff we did before the fork()ing above */
1299 r = reset_signal_mask();
1301 log_full_errno(prio, r, "Failed to reset signal mask: %m");
1302 _exit(EXIT_FAILURE);
1304 } else if (block_signals) { /* undo what we did above */
1305 if (sigprocmask(SIG_SETMASK, &saved_ss, NULL) < 0) {
1306 log_full_errno(prio, errno, "Failed to restore signal mask: %m");
1307 _exit(EXIT_FAILURE);
1311 if (flags & FORK_DEATHSIG) {
1313 /* Let's see if the parent PID is still the one we started from? If not, then the parent
1314 * already died by the time we set PR_SET_PDEATHSIG, hence let's emulate the effect */
1318 /* Parent is in a differn't PID namespace. */;
1319 else if (ppid != original_pid) {
1320 log_debug("Parent died early, raising SIGTERM.");
1321 (void) raise(SIGTERM);
1322 _exit(EXIT_FAILURE);
1326 if (flags & FORK_CLOSE_ALL_FDS) {
1327 /* Close the logs here in case it got reopened above, as close_all_fds() would close them for us */
1330 r = close_all_fds(except_fds, n_except_fds);
1332 log_full_errno(prio, r, "Failed to close all file descriptors: %m");
1333 _exit(EXIT_FAILURE);
1337 /* When we were asked to reopen the logs, do so again now */
1338 if (flags & FORK_REOPEN_LOG) {
1340 log_set_open_when_needed(false);
1343 if (flags & FORK_NULL_STDIO) {
1344 r = make_null_stdio();
1346 log_full_errno(prio, r, "Failed to connect stdin/stdout to /dev/null: %m");
1347 _exit(EXIT_FAILURE);
1352 *ret_pid = getpid_cached();
1357 int fork_agent(const char *name, const int except[], unsigned n_except, pid_t *ret_pid, const char *path, ...) {
1358 bool stdout_is_tty, stderr_is_tty;
1366 /* Spawns a temporary TTY agent, making sure it goes away when we go away */
1368 r = safe_fork_full(name, except, n_except, FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_CLOSE_ALL_FDS, ret_pid);
1376 stdout_is_tty = isatty(STDOUT_FILENO);
1377 stderr_is_tty = isatty(STDERR_FILENO);
1379 if (!stdout_is_tty || !stderr_is_tty) {
1382 /* Detach from stdout/stderr. and reopen
1383 * /dev/tty for them. This is important to
1384 * ensure that when systemctl is started via
1385 * popen() or a similar call that expects to
1386 * read EOF we actually do generate EOF and
1387 * not delay this indefinitely by because we
1388 * keep an unused copy of stdin around. */
1389 fd = open("/dev/tty", O_WRONLY);
1391 log_error_errno(errno, "Failed to open /dev/tty: %m");
1392 _exit(EXIT_FAILURE);
1395 if (!stdout_is_tty && dup2(fd, STDOUT_FILENO) < 0) {
1396 log_error_errno(errno, "Failed to dup2 /dev/tty: %m");
1397 _exit(EXIT_FAILURE);
1400 if (!stderr_is_tty && dup2(fd, STDERR_FILENO) < 0) {
1401 log_error_errno(errno, "Failed to dup2 /dev/tty: %m");
1402 _exit(EXIT_FAILURE);
1405 if (fd > STDERR_FILENO)
1409 /* Count arguments */
1411 for (n = 0; va_arg(ap, char*); n++)
1416 l = alloca(sizeof(char *) * (n + 1));
1418 /* Fill in arguments */
1420 for (i = 0; i <= n; i++)
1421 l[i] = va_arg(ap, char*);
1425 _exit(EXIT_FAILURE);
1428 #if 0 /// UNNEEDED by elogind
1429 static const char *const ioprio_class_table[] = {
1430 [IOPRIO_CLASS_NONE] = "none",
1431 [IOPRIO_CLASS_RT] = "realtime",
1432 [IOPRIO_CLASS_BE] = "best-effort",
1433 [IOPRIO_CLASS_IDLE] = "idle"
1436 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ioprio_class, int, INT_MAX);
1438 static const char *const sigchld_code_table[] = {
1439 [CLD_EXITED] = "exited",
1440 [CLD_KILLED] = "killed",
1441 [CLD_DUMPED] = "dumped",
1442 [CLD_TRAPPED] = "trapped",
1443 [CLD_STOPPED] = "stopped",
1444 [CLD_CONTINUED] = "continued",
1447 DEFINE_STRING_TABLE_LOOKUP(sigchld_code, int);
1449 static const char* const sched_policy_table[] = {
1450 [SCHED_OTHER] = "other",
1451 [SCHED_BATCH] = "batch",
1452 [SCHED_IDLE] = "idle",
1453 [SCHED_FIFO] = "fifo",
1457 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(sched_policy, int, INT_MAX);