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 int rename_process(const char name[]) {
284 static size_t mm_size = 0;
285 static char *mm = NULL;
286 bool truncated = false;
289 /* This is a like a poor man's setproctitle(). It changes the comm field, argv[0], and also the glibc's
290 * internally used name of the process. For the first one a limit of 16 chars applies; to the second one in
291 * many cases one of 10 (i.e. length of "/sbin/init") — however if we have CAP_SYS_RESOURCES it is unbounded;
292 * to the third one 7 (i.e. the length of "systemd". If you pass a longer string it will likely be
295 * Returns 0 if a name was set but truncated, > 0 if it was set but not truncated. */
298 return -EINVAL; /* let's not confuse users unnecessarily with an empty name */
300 if (!is_main_thread())
301 return -EPERM; /* Let's not allow setting the process name from other threads than the main one, as we
302 * cache things without locking, and we make assumptions that PR_SET_NAME sets the
303 * process name that isn't correct on any other threads */
307 /* First step, change the comm field. The main thread's comm is identical to the process comm. This means we
308 * can use PR_SET_NAME, which sets the thread name for the calling thread. */
309 if (prctl(PR_SET_NAME, name) < 0)
310 log_debug_errno(errno, "PR_SET_NAME failed: %m");
311 if (l > 15) /* Linux process names can be 15 chars at max */
314 /* Second step, change glibc's ID of the process name. */
315 if (program_invocation_name) {
318 k = strlen(program_invocation_name);
319 strncpy(program_invocation_name, name, k);
324 /* Third step, completely replace the argv[] array the kernel maintains for us. This requires privileges, but
325 * has the advantage that the argv[] array is exactly what we want it to be, and not filled up with zeros at
326 * the end. This is the best option for changing /proc/self/cmdline. */
328 /* Let's not bother with this if we don't have euid == 0. Strictly speaking we should check for the
329 * CAP_SYS_RESOURCE capability which is independent of the euid. In our own code the capability generally is
330 * present only for euid == 0, hence let's use this as quick bypass check, to avoid calling mmap() if
331 * PR_SET_MM_ARG_{START,END} fails with EPERM later on anyway. After all geteuid() is dead cheap to call, but
334 log_debug("Skipping PR_SET_MM, as we don't have privileges.");
335 else if (mm_size < l+1) {
339 nn_size = PAGE_ALIGN(l+1);
340 nn = mmap(NULL, nn_size, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
341 if (nn == MAP_FAILED) {
342 log_debug_errno(errno, "mmap() failed: %m");
346 strncpy(nn, name, nn_size);
348 /* Now, let's tell the kernel about this new memory */
349 if (prctl(PR_SET_MM, PR_SET_MM_ARG_START, (unsigned long) nn, 0, 0) < 0) {
350 log_debug_errno(errno, "PR_SET_MM_ARG_START failed, proceeding without: %m");
351 (void) munmap(nn, nn_size);
355 /* And update the end pointer to the new end, too. If this fails, we don't really know what to do, it's
356 * pretty unlikely that we can rollback, hence we'll just accept the failure, and continue. */
357 if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) nn + l + 1, 0, 0) < 0)
358 log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m");
361 (void) munmap(mm, mm_size);
366 strncpy(mm, name, mm_size);
368 /* Update the end pointer, continuing regardless of any failure. */
369 if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) mm + l + 1, 0, 0) < 0)
370 log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m");
374 /* Fourth step: in all cases we'll also update the original argv[], so that our own code gets it right too if
375 * it still looks here */
377 if (saved_argc > 0) {
383 k = strlen(saved_argv[0]);
384 strncpy(saved_argv[0], name, k);
389 for (i = 1; i < saved_argc; i++) {
393 memzero(saved_argv[i], strlen(saved_argv[i]));
400 int is_kernel_thread(pid_t pid) {
401 _cleanup_free_ char *line = NULL;
402 unsigned long long flags;
408 if (IN_SET(pid, 0, 1) || pid == getpid_cached()) /* pid 1, and we ourselves certainly aren't a kernel thread */
410 if (!pid_is_valid(pid))
413 p = procfs_file_alloca(pid, "stat");
414 r = read_one_line_file(p, &line);
420 /* Skip past the comm field */
421 q = strrchr(line, ')');
426 /* Skip 6 fields to reach the flags field */
427 for (i = 0; i < 6; i++) {
428 l = strspn(q, WHITESPACE);
433 l = strcspn(q, WHITESPACE);
439 /* Skip preceeding whitespace */
440 l = strspn(q, WHITESPACE);
445 /* Truncate the rest */
446 l = strcspn(q, WHITESPACE);
451 r = safe_atollu(q, &flags);
455 return !!(flags & PF_KTHREAD);
458 #if 0 /// UNNEEDED by elogind
459 int get_process_capeff(pid_t pid, char **capeff) {
466 p = procfs_file_alloca(pid, "status");
468 r = get_proc_field(p, "CapEff", WHITESPACE, capeff);
476 static int get_process_link_contents(const char *proc_file, char **name) {
482 r = readlink_malloc(proc_file, name);
491 int get_process_exe(pid_t pid, char **name) {
498 p = procfs_file_alloca(pid, "exe");
499 r = get_process_link_contents(p, name);
503 d = endswith(*name, " (deleted)");
510 #if 0 /// UNNEEDED by elogind
511 static int get_process_id(pid_t pid, const char *field, uid_t *uid) {
512 _cleanup_fclose_ FILE *f = NULL;
522 p = procfs_file_alloca(pid, "status");
530 (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
532 FOREACH_LINE(line, f, return -errno) {
537 if (startswith(l, field)) {
539 l += strspn(l, WHITESPACE);
541 l[strcspn(l, WHITESPACE)] = 0;
543 return parse_uid(l, uid);
550 int get_process_uid(pid_t pid, uid_t *uid) {
552 if (pid == 0 || pid == getpid_cached()) {
557 return get_process_id(pid, "Uid:", uid);
560 int get_process_gid(pid_t pid, gid_t *gid) {
562 if (pid == 0 || pid == getpid_cached()) {
567 assert_cc(sizeof(uid_t) == sizeof(gid_t));
568 return get_process_id(pid, "Gid:", gid);
571 int get_process_cwd(pid_t pid, char **cwd) {
576 p = procfs_file_alloca(pid, "cwd");
578 return get_process_link_contents(p, cwd);
581 int get_process_root(pid_t pid, char **root) {
586 p = procfs_file_alloca(pid, "root");
588 return get_process_link_contents(p, root);
591 int get_process_environ(pid_t pid, char **env) {
592 _cleanup_fclose_ FILE *f = NULL;
593 _cleanup_free_ char *outcome = NULL;
596 size_t allocated = 0, sz = 0;
601 p = procfs_file_alloca(pid, "environ");
610 (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
612 while ((c = fgetc(f)) != EOF) {
613 if (!GREEDY_REALLOC(outcome, allocated, sz + 5))
617 outcome[sz++] = '\n';
619 sz += cescape_char(c, outcome + sz);
623 outcome = strdup("");
635 int get_process_ppid(pid_t pid, pid_t *_ppid) {
637 _cleanup_free_ char *line = NULL;
644 if (pid == 0 || pid == getpid_cached()) {
649 p = procfs_file_alloca(pid, "stat");
650 r = read_one_line_file(p, &line);
656 /* Let's skip the pid and comm fields. The latter is enclosed
657 * in () but does not escape any () in its value, so let's
658 * skip over it manually */
660 p = strrchr(line, ')');
672 if ((long unsigned) (pid_t) ppid != ppid)
675 *_ppid = (pid_t) ppid;
681 int wait_for_terminate(pid_t pid, siginfo_t *status) {
692 if (waitid(P_PID, pid, status, WEXITED) < 0) {
697 return negative_errno();
706 * < 0 : wait_for_terminate() failed to get the state of the
707 * process, the process was terminated by a signal, or
708 * failed for an unknown reason.
709 * >=0 : The process terminated normally, and its exit code is
712 * That is, success is indicated by a return value of zero, and an
713 * error is indicated by a non-zero value.
715 * A warning is emitted if the process terminates abnormally,
716 * and also if it returns non-zero unless check_exit_code is true.
718 int wait_for_terminate_and_check(const char *name, pid_t pid, WaitFlags flags) {
719 _cleanup_free_ char *buffer = NULL;
726 r = get_process_comm(pid, &buffer);
728 log_debug_errno(r, "Failed to acquire process name of " PID_FMT ", ignoring: %m", pid);
733 prio = flags & WAIT_LOG_ABNORMAL ? LOG_ERR : LOG_DEBUG;
735 r = wait_for_terminate(pid, &status);
737 return log_full_errno(prio, r, "Failed to wait for %s: %m", strna(name));
739 if (status.si_code == CLD_EXITED) {
740 if (status.si_status != EXIT_SUCCESS)
741 log_full(flags & WAIT_LOG_NON_ZERO_EXIT_STATUS ? LOG_ERR : LOG_DEBUG,
742 "%s failed with exit status %i.", strna(name), status.si_status);
744 log_debug("%s succeeded.", name);
746 return status.si_status;
748 } else if (IN_SET(status.si_code, CLD_KILLED, CLD_DUMPED)) {
750 log_full(prio, "%s terminated by signal %s.", strna(name), signal_to_string(status.si_status));
754 log_full(prio, "%s failed due to unknown reason.", strna(name));
760 * < 0 : wait_for_terminate_with_timeout() failed to get the state of the
761 * process, the process timed out, the process was terminated by a
762 * signal, or failed for an unknown reason.
763 * >=0 : The process terminated normally with no failures.
765 * Success is indicated by a return value of zero, a timeout is indicated
766 * by ETIMEDOUT, and all other child failure states are indicated by error
767 * is indicated by a non-zero value.
769 int wait_for_terminate_with_timeout(pid_t pid, usec_t timeout) {
774 assert_se(sigemptyset(&mask) == 0);
775 assert_se(sigaddset(&mask, SIGCHLD) == 0);
777 /* Drop into a sigtimewait-based timeout. Waiting for the
779 until = now(CLOCK_MONOTONIC) + timeout;
782 siginfo_t status = {};
785 n = now(CLOCK_MONOTONIC);
789 r = sigtimedwait(&mask, NULL, timespec_store(&ts, until - n)) < 0 ? -errno : 0;
790 /* Assuming we woke due to the child exiting. */
791 if (waitid(P_PID, pid, &status, WEXITED|WNOHANG) == 0) {
792 if (status.si_pid == pid) {
793 /* This is the correct child.*/
794 if (status.si_code == CLD_EXITED)
795 return (status.si_status == 0) ? 0 : -EPROTO;
800 /* Not the child, check for errors and proceed appropriately */
804 /* Timed out, child is likely hung. */
807 /* Received a different signal and should retry */
810 /* Return any unexpected errors */
819 #if 0 /// UNNEEDED by elogind
820 void sigkill_wait(pid_t pid) {
823 if (kill(pid, SIGKILL) > 0)
824 (void) wait_for_terminate(pid, NULL);
827 void sigkill_waitp(pid_t *pid) {
839 void sigterm_wait(pid_t pid) {
842 if (kill_and_sigcont(pid, SIGTERM) > 0)
843 (void) wait_for_terminate(pid, NULL);
846 int kill_and_sigcont(pid_t pid, int sig) {
849 r = kill(pid, sig) < 0 ? -errno : 0;
851 /* If this worked, also send SIGCONT, unless we already just sent a SIGCONT, or SIGKILL was sent which isn't
852 * affected by a process being suspended anyway. */
853 if (r >= 0 && !IN_SET(sig, SIGCONT, SIGKILL))
854 (void) kill(pid, SIGCONT);
859 int getenv_for_pid(pid_t pid, const char *field, char **ret) {
860 _cleanup_fclose_ FILE *f = NULL;
870 if (pid == 0 || pid == getpid_cached()) {
887 path = procfs_file_alloca(pid, "environ");
889 f = fopen(path, "re");
897 (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
905 for (i = 0; i < sizeof(line)-1; i++) {
909 if (_unlikely_(c == EOF)) {
919 if (strneq(line, field, l) && line[l] == '=') {
920 value = strdup(line + l + 1);
934 bool pid_is_unwaited(pid_t pid) {
935 /* Checks whether a PID is still valid at all, including a zombie */
940 if (pid <= 1) /* If we or PID 1 would be dead and have been waited for, this code would not be running */
943 if (pid == getpid_cached())
946 if (kill(pid, 0) >= 0)
949 return errno != ESRCH;
952 bool pid_is_alive(pid_t pid) {
955 /* Checks whether a PID is still valid and not a zombie */
960 if (pid <= 1) /* If we or PID 1 would be a zombie, this code would not be running */
963 if (pid == getpid_cached())
966 r = get_process_state(pid);
967 if (IN_SET(r, -ESRCH, 'Z'))
973 #if 0 /// UNNEEDED by elogind
974 int pid_from_same_root_fs(pid_t pid) {
980 if (pid == 0 || pid == getpid_cached())
983 root = procfs_file_alloca(pid, "root");
985 return files_same(root, "/proc/1/root", 0);
989 bool is_main_thread(void) {
990 static thread_local int cached = 0;
992 if (_unlikely_(cached == 0))
993 cached = getpid_cached() == gettid() ? 1 : -1;
998 #if 0 /// UNNEEDED by elogind
999 noreturn void freeze(void) {
1003 /* Make sure nobody waits for us on a socket anymore */
1004 close_all_fds(NULL, 0);
1008 /* Let's not freeze right away, but keep reaping zombies. */
1013 r = waitid(P_ALL, 0, &si, WEXITED);
1014 if (r < 0 && errno != EINTR)
1018 /* waitid() failed with an unexpected error, things are really borked. Freeze now! */
1023 bool oom_score_adjust_is_valid(int oa) {
1024 return oa >= OOM_SCORE_ADJ_MIN && oa <= OOM_SCORE_ADJ_MAX;
1027 unsigned long personality_from_string(const char *p) {
1031 return PERSONALITY_INVALID;
1033 /* Parse a personality specifier. We use our own identifiers that indicate specific ABIs, rather than just
1034 * hints regarding the register size, since we want to keep things open for multiple locally supported ABIs for
1035 * the same register size. */
1037 architecture = architecture_from_string(p);
1038 if (architecture < 0)
1039 return PERSONALITY_INVALID;
1041 if (architecture == native_architecture())
1043 #ifdef SECONDARY_ARCHITECTURE
1044 if (architecture == SECONDARY_ARCHITECTURE)
1048 return PERSONALITY_INVALID;
1051 const char* personality_to_string(unsigned long p) {
1052 int architecture = _ARCHITECTURE_INVALID;
1055 architecture = native_architecture();
1056 #ifdef SECONDARY_ARCHITECTURE
1057 else if (p == PER_LINUX32)
1058 architecture = SECONDARY_ARCHITECTURE;
1061 if (architecture < 0)
1064 return architecture_to_string(architecture);
1067 int safe_personality(unsigned long p) {
1070 /* So here's the deal, personality() is weirdly defined by glibc. In some cases it returns a failure via errno,
1071 * and in others as negative return value containing an errno-like value. Let's work around this: this is a
1072 * wrapper that uses errno if it is set, and uses the return value otherwise. And then it sets both errno and
1073 * the return value indicating the same issue, so that we are definitely on the safe side.
1075 * See https://github.com/systemd/systemd/issues/6737 */
1078 ret = personality(p);
1089 int opinionated_personality(unsigned long *ret) {
1092 /* Returns the current personality, or PERSONALITY_INVALID if we can't determine it. This function is a bit
1093 * opinionated though, and ignores all the finer-grained bits and exotic personalities, only distinguishing the
1094 * two most relevant personalities: PER_LINUX and PER_LINUX32. */
1096 current = safe_personality(PERSONALITY_INVALID);
1100 if (((unsigned long) current & 0xffff) == PER_LINUX32)
1108 void valgrind_summary_hack(void) {
1109 #if HAVE_VALGRIND_VALGRIND_H
1110 if (getpid_cached() == 1 && RUNNING_ON_VALGRIND) {
1112 pid = raw_clone(SIGCHLD);
1114 log_emergency_errno(errno, "Failed to fork off valgrind helper: %m");
1118 log_info("Spawned valgrind helper as PID "PID_FMT".", pid);
1119 (void) wait_for_terminate(pid, NULL);
1125 int pid_compare_func(const void *a, const void *b) {
1126 const pid_t *p = a, *q = b;
1128 /* Suitable for usage in qsort() */
1137 int ioprio_parse_priority(const char *s, int *ret) {
1143 r = safe_atoi(s, &i);
1147 if (!ioprio_priority_is_valid(i))
1155 /* The cached PID, possible values:
1157 * == UNSET [0] → cache not initialized yet
1158 * == BUSY [-1] → some thread is initializing it at the moment
1159 * any other → the cached PID
1162 #define CACHED_PID_UNSET ((pid_t) 0)
1163 #define CACHED_PID_BUSY ((pid_t) -1)
1165 static pid_t cached_pid = CACHED_PID_UNSET;
1167 void reset_cached_pid(void) {
1168 /* Invoked in the child after a fork(), i.e. at the first moment the PID changed */
1169 cached_pid = CACHED_PID_UNSET;
1172 /* We use glibc __register_atfork() + __dso_handle directly here, as they are not included in the glibc
1173 * headers. __register_atfork() is mostly equivalent to pthread_atfork(), but doesn't require us to link against
1174 * libpthread, as it is part of glibc anyway. */
1176 extern int __register_atfork(void (*prepare) (void), void (*parent) (void), void (*child) (void), void * __dso_handle);
1177 extern void* __dso_handle __attribute__ ((__weak__));
1178 #endif // ifdef __GLIBC__
1180 pid_t getpid_cached(void) {
1181 static bool installed = false;
1182 pid_t current_value;
1184 /* getpid_cached() is much like getpid(), but caches the value in local memory, to avoid having to invoke a
1185 * system call each time. This restores glibc behaviour from before 2.24, when getpid() was unconditionally
1186 * cached. Starting with 2.24 getpid() started to become prohibitively expensive when used for detecting when
1187 * objects were used across fork()s. With this caching the old behaviour is somewhat restored.
1189 * https://bugzilla.redhat.com/show_bug.cgi?id=1443976
1190 * https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=c579f48edba88380635ab98cb612030e3ed8691e
1193 current_value = __sync_val_compare_and_swap(&cached_pid, CACHED_PID_UNSET, CACHED_PID_BUSY);
1195 switch (current_value) {
1197 case CACHED_PID_UNSET: { /* Not initialized yet, then do so now */
1200 new_pid = raw_getpid();
1203 /* __register_atfork() either returns 0 or -ENOMEM, in its glibc implementation. Since it's
1204 * only half-documented (glibc doesn't document it but LSB does — though only superficially)
1205 * we'll check for errors only in the most generic fashion possible. */
1207 if (__register_atfork(NULL, NULL, reset_cached_pid, __dso_handle) != 0) {
1208 /* OOM? Let's try again later */
1209 cached_pid = CACHED_PID_UNSET;
1216 cached_pid = new_pid;
1220 case CACHED_PID_BUSY: /* Somebody else is currently initializing */
1221 return raw_getpid();
1223 default: /* Properly initialized */
1224 return current_value;
1228 int must_be_root(void) {
1233 log_error("Need to be root.");
1239 const int except_fds[],
1240 size_t n_except_fds,
1244 pid_t original_pid, pid;
1245 sigset_t saved_ss, ss;
1246 bool block_signals = false;
1249 /* A wrapper around fork(), that does a couple of important initializations in addition to mere forking. Always
1250 * returns the child's PID in *ret_pid. Returns == 0 in the child, and > 0 in the parent. */
1252 prio = flags & FORK_LOG ? LOG_ERR : LOG_DEBUG;
1254 original_pid = getpid_cached();
1256 if (flags & (FORK_RESET_SIGNALS|FORK_DEATHSIG)) {
1258 /* We temporarily block all signals, so that the new child has them blocked initially. This way, we can
1259 * be sure that SIGTERMs are not lost we might send to the child. */
1261 if (sigfillset(&ss) < 0)
1262 return log_full_errno(prio, errno, "Failed to reset signal set: %m");
1264 block_signals = true;
1266 } else if (flags & FORK_WAIT) {
1268 /* Let's block SIGCHLD at least, so that we can safely watch for the child process */
1270 if (sigemptyset(&ss) < 0)
1271 return log_full_errno(prio, errno, "Failed to clear signal set: %m");
1273 if (sigaddset(&ss, SIGCHLD) < 0)
1274 return log_full_errno(prio, errno, "Failed to add SIGCHLD to signal set: %m");
1276 block_signals = true;
1280 if (sigprocmask(SIG_SETMASK, &ss, &saved_ss) < 0)
1281 return log_full_errno(prio, errno, "Failed to set signal mask: %m");
1283 if (flags & FORK_NEW_MOUNTNS)
1284 pid = raw_clone(SIGCHLD|CLONE_NEWNS);
1290 if (block_signals) /* undo what we did above */
1291 (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL);
1293 return log_full_errno(prio, r, "Failed to fork: %m");
1296 /* We are in the parent process */
1298 log_debug("Successfully forked off '%s' as PID " PID_FMT ".", strna(name), pid);
1300 if (flags & FORK_WAIT) {
1301 r = wait_for_terminate_and_check(name, pid, (flags & FORK_LOG ? WAIT_LOG : 0));
1304 if (r != EXIT_SUCCESS) /* exit status > 0 should be treated as failure, too */
1308 if (block_signals) /* undo what we did above */
1309 (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL);
1317 /* We are in the child process */
1319 if (flags & FORK_REOPEN_LOG) {
1320 /* Close the logs if requested, before we log anything. And make sure we reopen it if needed. */
1322 log_set_open_when_needed(true);
1326 r = rename_process(name);
1328 log_full_errno(flags & FORK_LOG ? LOG_WARNING : LOG_DEBUG,
1329 r, "Failed to rename process, ignoring: %m");
1332 if (flags & FORK_DEATHSIG)
1333 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0) {
1334 log_full_errno(prio, errno, "Failed to set death signal: %m");
1335 _exit(EXIT_FAILURE);
1338 if (flags & FORK_RESET_SIGNALS) {
1339 r = reset_all_signal_handlers();
1341 log_full_errno(prio, r, "Failed to reset signal handlers: %m");
1342 _exit(EXIT_FAILURE);
1345 /* This implicitly undoes the signal mask stuff we did before the fork()ing above */
1346 r = reset_signal_mask();
1348 log_full_errno(prio, r, "Failed to reset signal mask: %m");
1349 _exit(EXIT_FAILURE);
1351 } else if (block_signals) { /* undo what we did above */
1352 if (sigprocmask(SIG_SETMASK, &saved_ss, NULL) < 0) {
1353 log_full_errno(prio, errno, "Failed to restore signal mask: %m");
1354 _exit(EXIT_FAILURE);
1358 if (flags & FORK_DEATHSIG) {
1360 /* Let's see if the parent PID is still the one we started from? If not, then the parent
1361 * already died by the time we set PR_SET_PDEATHSIG, hence let's emulate the effect */
1365 /* Parent is in a differn't PID namespace. */;
1366 else if (ppid != original_pid) {
1367 log_debug("Parent died early, raising SIGTERM.");
1368 (void) raise(SIGTERM);
1369 _exit(EXIT_FAILURE);
1373 if (flags & FORK_CLOSE_ALL_FDS) {
1374 /* Close the logs here in case it got reopened above, as close_all_fds() would close them for us */
1377 r = close_all_fds(except_fds, n_except_fds);
1379 log_full_errno(prio, r, "Failed to close all file descriptors: %m");
1380 _exit(EXIT_FAILURE);
1384 /* When we were asked to reopen the logs, do so again now */
1385 if (flags & FORK_REOPEN_LOG) {
1387 log_set_open_when_needed(false);
1390 if (flags & FORK_NULL_STDIO) {
1391 r = make_null_stdio();
1393 log_full_errno(prio, r, "Failed to connect stdin/stdout to /dev/null: %m");
1394 _exit(EXIT_FAILURE);
1399 *ret_pid = getpid_cached();
1404 int fork_agent(const char *name, const int except[], unsigned n_except, pid_t *ret_pid, const char *path, ...) {
1405 bool stdout_is_tty, stderr_is_tty;
1413 /* Spawns a temporary TTY agent, making sure it goes away when we go away */
1415 r = safe_fork_full(name, except, n_except, FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_CLOSE_ALL_FDS, ret_pid);
1423 stdout_is_tty = isatty(STDOUT_FILENO);
1424 stderr_is_tty = isatty(STDERR_FILENO);
1426 if (!stdout_is_tty || !stderr_is_tty) {
1429 /* Detach from stdout/stderr. and reopen
1430 * /dev/tty for them. This is important to
1431 * ensure that when systemctl is started via
1432 * popen() or a similar call that expects to
1433 * read EOF we actually do generate EOF and
1434 * not delay this indefinitely by because we
1435 * keep an unused copy of stdin around. */
1436 fd = open("/dev/tty", O_WRONLY);
1438 log_error_errno(errno, "Failed to open /dev/tty: %m");
1439 _exit(EXIT_FAILURE);
1442 if (!stdout_is_tty && dup2(fd, STDOUT_FILENO) < 0) {
1443 log_error_errno(errno, "Failed to dup2 /dev/tty: %m");
1444 _exit(EXIT_FAILURE);
1447 if (!stderr_is_tty && dup2(fd, STDERR_FILENO) < 0) {
1448 log_error_errno(errno, "Failed to dup2 /dev/tty: %m");
1449 _exit(EXIT_FAILURE);
1452 safe_close_above_stdio(fd);
1455 /* Count arguments */
1457 for (n = 0; va_arg(ap, char*); n++)
1462 l = alloca(sizeof(char *) * (n + 1));
1464 /* Fill in arguments */
1466 for (i = 0; i <= n; i++)
1467 l[i] = va_arg(ap, char*);
1471 _exit(EXIT_FAILURE);
1474 #if 0 /// UNNEEDED by elogind
1475 static const char *const ioprio_class_table[] = {
1476 [IOPRIO_CLASS_NONE] = "none",
1477 [IOPRIO_CLASS_RT] = "realtime",
1478 [IOPRIO_CLASS_BE] = "best-effort",
1479 [IOPRIO_CLASS_IDLE] = "idle"
1482 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ioprio_class, int, INT_MAX);
1484 static const char *const sigchld_code_table[] = {
1485 [CLD_EXITED] = "exited",
1486 [CLD_KILLED] = "killed",
1487 [CLD_DUMPED] = "dumped",
1488 [CLD_TRAPPED] = "trapped",
1489 [CLD_STOPPED] = "stopped",
1490 [CLD_CONTINUED] = "continued",
1493 DEFINE_STRING_TABLE_LOOKUP(sigchld_code, int);
1495 static const char* const sched_policy_table[] = {
1496 [SCHED_OTHER] = "other",
1497 [SCHED_BATCH] = "batch",
1498 [SCHED_IDLE] = "idle",
1499 [SCHED_FIFO] = "fifo",
1503 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(sched_policy, int, INT_MAX);