1 /* SPDX-License-Identifier: LGPL-2.1+ */
11 #include <stdio_ext.h>
15 //#include <sys/mount.h>
16 #include <sys/personality.h>
17 #include <sys/prctl.h>
18 #include <sys/types.h>
22 #if HAVE_VALGRIND_VALGRIND_H
23 #include <valgrind/valgrind.h>
26 #include "alloc-util.h"
27 //#include "architecture.h"
36 #include "process-util.h"
37 #include "raw-clone.h"
38 #include "signal-util.h"
39 //#include "stat-util.h"
40 #include "string-table.h"
41 #include "string-util.h"
42 //#include "terminal-util.h"
43 #include "user-util.h"
46 int get_process_state(pid_t pid) {
50 _cleanup_free_ char *line = NULL;
54 p = procfs_file_alloca(pid, "stat");
56 r = read_one_line_file(p, &line);
62 p = strrchr(line, ')');
68 if (sscanf(p, " %c", &state) != 1)
71 return (unsigned char) state;
74 int get_process_comm(pid_t pid, char **ret) {
75 _cleanup_free_ char *escaped = NULL, *comm = NULL;
82 escaped = new(char, TASK_COMM_LEN);
86 p = procfs_file_alloca(pid, "comm");
88 r = read_one_line_file(p, &comm);
94 /* Escape unprintable characters, just in case, but don't grow the string beyond the underlying size */
95 cellescape(escaped, TASK_COMM_LEN, comm);
97 *ret = TAKE_PTR(escaped);
101 int get_process_cmdline(pid_t pid, size_t max_length, bool comm_fallback, char **line) {
102 _cleanup_fclose_ FILE *f = NULL;
104 char *k, *ans = NULL;
111 /* Retrieves a process' command line. Replaces unprintable characters while doing so by whitespace (coalescing
112 * multiple sequential ones into one). If max_length is != 0 will return a string of the specified size at most
113 * (the trailing NUL byte does count towards the length here!), abbreviated with a "..." ellipsis. If
114 * comm_fallback is true and the process has no command line set (the case for kernel threads), or has a
115 * command line that resolves to the empty string will return the "comm" name of the process instead.
117 * Returns -ESRCH if the process doesn't exist, and -ENOENT if the process has no command line (and
118 * comm_fallback is false). Returns 0 and sets *line otherwise. */
120 p = procfs_file_alloca(pid, "cmdline");
129 (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
131 if (max_length == 1) {
133 /* If there's only room for one byte, return the empty string */
141 } else if (max_length == 0) {
142 size_t len = 0, allocated = 0;
144 while ((c = getc(f)) != EOF) {
146 if (!GREEDY_REALLOC(ans, allocated, len+3)) {
168 bool dotdotdot = false;
171 ans = new(char, max_length);
177 while ((c = getc(f)) != EOF) {
204 if (max_length <= 4) {
208 k = ans + max_length - 4;
211 /* Eat up final spaces */
212 while (k > ans && isspace(k[-1])) {
218 strncpy(k, "...", left-1);
224 /* Kernel threads have no argv[] */
226 _cleanup_free_ char *t = NULL;
234 h = get_process_comm(pid, &t);
239 ans = strjoin("[", t, "]");
245 if (l + 3 <= max_length)
246 ans = strjoin("[", t, "]");
247 else if (max_length <= 6) {
249 ans = new(char, max_length);
253 memcpy(ans, "[...]", max_length-1);
254 ans[max_length-1] = 0;
256 t[max_length - 6] = 0;
258 /* Chop off final spaces */
259 delete_trailing_chars(t, WHITESPACE);
261 ans = strjoin("[", t, "...]");
272 int rename_process(const char name[]) {
273 static size_t mm_size = 0;
274 static char *mm = NULL;
275 bool truncated = false;
278 /* This is a like a poor man's setproctitle(). It changes the comm field, argv[0], and also the glibc's
279 * internally used name of the process. For the first one a limit of 16 chars applies; to the second one in
280 * many cases one of 10 (i.e. length of "/sbin/init") — however if we have CAP_SYS_RESOURCES it is unbounded;
281 * to the third one 7 (i.e. the length of "systemd". If you pass a longer string it will likely be
284 * Returns 0 if a name was set but truncated, > 0 if it was set but not truncated. */
287 return -EINVAL; /* let's not confuse users unnecessarily with an empty name */
289 if (!is_main_thread())
290 return -EPERM; /* Let's not allow setting the process name from other threads than the main one, as we
291 * cache things without locking, and we make assumptions that PR_SET_NAME sets the
292 * process name that isn't correct on any other threads */
296 /* First step, change the comm field. The main thread's comm is identical to the process comm. This means we
297 * can use PR_SET_NAME, which sets the thread name for the calling thread. */
298 if (prctl(PR_SET_NAME, name) < 0)
299 log_debug_errno(errno, "PR_SET_NAME failed: %m");
300 if (l >= TASK_COMM_LEN) /* Linux process names can be 15 chars at max */
303 /* Second step, change glibc's ID of the process name. */
304 if (program_invocation_name) {
307 k = strlen(program_invocation_name);
308 strncpy(program_invocation_name, name, k);
313 /* Third step, completely replace the argv[] array the kernel maintains for us. This requires privileges, but
314 * has the advantage that the argv[] array is exactly what we want it to be, and not filled up with zeros at
315 * the end. This is the best option for changing /proc/self/cmdline. */
317 /* Let's not bother with this if we don't have euid == 0. Strictly speaking we should check for the
318 * CAP_SYS_RESOURCE capability which is independent of the euid. In our own code the capability generally is
319 * present only for euid == 0, hence let's use this as quick bypass check, to avoid calling mmap() if
320 * PR_SET_MM_ARG_{START,END} fails with EPERM later on anyway. After all geteuid() is dead cheap to call, but
323 log_debug("Skipping PR_SET_MM, as we don't have privileges.");
324 else if (mm_size < l+1) {
328 nn_size = PAGE_ALIGN(l+1);
329 nn = mmap(NULL, nn_size, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
330 if (nn == MAP_FAILED) {
331 log_debug_errno(errno, "mmap() failed: %m");
335 strncpy(nn, name, nn_size);
337 /* Now, let's tell the kernel about this new memory */
338 if (prctl(PR_SET_MM, PR_SET_MM_ARG_START, (unsigned long) nn, 0, 0) < 0) {
339 log_debug_errno(errno, "PR_SET_MM_ARG_START failed, proceeding without: %m");
340 (void) munmap(nn, nn_size);
344 /* And update the end pointer to the new end, too. If this fails, we don't really know what to do, it's
345 * pretty unlikely that we can rollback, hence we'll just accept the failure, and continue. */
346 if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) nn + l + 1, 0, 0) < 0)
347 log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m");
350 (void) munmap(mm, mm_size);
355 strncpy(mm, name, mm_size);
357 /* Update the end pointer, continuing regardless of any failure. */
358 if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) mm + l + 1, 0, 0) < 0)
359 log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m");
363 /* Fourth step: in all cases we'll also update the original argv[], so that our own code gets it right too if
364 * it still looks here */
366 if (saved_argc > 0) {
372 k = strlen(saved_argv[0]);
373 strncpy(saved_argv[0], name, k);
378 for (i = 1; i < saved_argc; i++) {
382 memzero(saved_argv[i], strlen(saved_argv[i]));
389 int is_kernel_thread(pid_t pid) {
390 _cleanup_free_ char *line = NULL;
391 unsigned long long flags;
397 if (IN_SET(pid, 0, 1) || pid == getpid_cached()) /* pid 1, and we ourselves certainly aren't a kernel thread */
399 if (!pid_is_valid(pid))
402 p = procfs_file_alloca(pid, "stat");
403 r = read_one_line_file(p, &line);
409 /* Skip past the comm field */
410 q = strrchr(line, ')');
415 /* Skip 6 fields to reach the flags field */
416 for (i = 0; i < 6; i++) {
417 l = strspn(q, WHITESPACE);
422 l = strcspn(q, WHITESPACE);
428 /* Skip preceeding whitespace */
429 l = strspn(q, WHITESPACE);
434 /* Truncate the rest */
435 l = strcspn(q, WHITESPACE);
440 r = safe_atollu(q, &flags);
444 return !!(flags & PF_KTHREAD);
447 #if 0 /// UNNEEDED by elogind
448 int get_process_capeff(pid_t pid, char **capeff) {
455 p = procfs_file_alloca(pid, "status");
457 r = get_proc_field(p, "CapEff", WHITESPACE, capeff);
465 static int get_process_link_contents(const char *proc_file, char **name) {
471 r = readlink_malloc(proc_file, name);
480 int get_process_exe(pid_t pid, char **name) {
487 p = procfs_file_alloca(pid, "exe");
488 r = get_process_link_contents(p, name);
492 d = endswith(*name, " (deleted)");
499 #if 0 /// UNNEEDED by elogind
500 static int get_process_id(pid_t pid, const char *field, uid_t *uid) {
501 _cleanup_fclose_ FILE *f = NULL;
511 p = procfs_file_alloca(pid, "status");
519 (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
521 FOREACH_LINE(line, f, return -errno) {
526 if (startswith(l, field)) {
528 l += strspn(l, WHITESPACE);
530 l[strcspn(l, WHITESPACE)] = 0;
532 return parse_uid(l, uid);
539 int get_process_uid(pid_t pid, uid_t *uid) {
541 if (pid == 0 || pid == getpid_cached()) {
546 return get_process_id(pid, "Uid:", uid);
549 int get_process_gid(pid_t pid, gid_t *gid) {
551 if (pid == 0 || pid == getpid_cached()) {
556 assert_cc(sizeof(uid_t) == sizeof(gid_t));
557 return get_process_id(pid, "Gid:", gid);
560 int get_process_cwd(pid_t pid, char **cwd) {
565 p = procfs_file_alloca(pid, "cwd");
567 return get_process_link_contents(p, cwd);
570 int get_process_root(pid_t pid, char **root) {
575 p = procfs_file_alloca(pid, "root");
577 return get_process_link_contents(p, root);
580 int get_process_environ(pid_t pid, char **env) {
581 _cleanup_fclose_ FILE *f = NULL;
582 _cleanup_free_ char *outcome = NULL;
585 size_t allocated = 0, sz = 0;
590 p = procfs_file_alloca(pid, "environ");
599 (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
601 while ((c = fgetc(f)) != EOF) {
602 if (!GREEDY_REALLOC(outcome, allocated, sz + 5))
606 outcome[sz++] = '\n';
608 sz += cescape_char(c, outcome + sz);
612 outcome = strdup("");
618 *env = TAKE_PTR(outcome);
623 int get_process_ppid(pid_t pid, pid_t *_ppid) {
625 _cleanup_free_ char *line = NULL;
632 if (pid == 0 || pid == getpid_cached()) {
637 p = procfs_file_alloca(pid, "stat");
638 r = read_one_line_file(p, &line);
644 /* Let's skip the pid and comm fields. The latter is enclosed
645 * in () but does not escape any () in its value, so let's
646 * skip over it manually */
648 p = strrchr(line, ')');
660 if ((long unsigned) (pid_t) ppid != ppid)
663 *_ppid = (pid_t) ppid;
669 int wait_for_terminate(pid_t pid, siginfo_t *status) {
680 if (waitid(P_PID, pid, status, WEXITED) < 0) {
685 return negative_errno();
694 * < 0 : wait_for_terminate() failed to get the state of the
695 * process, the process was terminated by a signal, or
696 * failed for an unknown reason.
697 * >=0 : The process terminated normally, and its exit code is
700 * That is, success is indicated by a return value of zero, and an
701 * error is indicated by a non-zero value.
703 * A warning is emitted if the process terminates abnormally,
704 * and also if it returns non-zero unless check_exit_code is true.
706 int wait_for_terminate_and_check(const char *name, pid_t pid, WaitFlags flags) {
707 _cleanup_free_ char *buffer = NULL;
714 r = get_process_comm(pid, &buffer);
716 log_debug_errno(r, "Failed to acquire process name of " PID_FMT ", ignoring: %m", pid);
721 prio = flags & WAIT_LOG_ABNORMAL ? LOG_ERR : LOG_DEBUG;
723 r = wait_for_terminate(pid, &status);
725 return log_full_errno(prio, r, "Failed to wait for %s: %m", strna(name));
727 if (status.si_code == CLD_EXITED) {
728 if (status.si_status != EXIT_SUCCESS)
729 log_full(flags & WAIT_LOG_NON_ZERO_EXIT_STATUS ? LOG_ERR : LOG_DEBUG,
730 "%s failed with exit status %i.", strna(name), status.si_status);
732 log_debug("%s succeeded.", name);
734 return status.si_status;
736 } else if (IN_SET(status.si_code, CLD_KILLED, CLD_DUMPED)) {
738 log_full(prio, "%s terminated by signal %s.", strna(name), signal_to_string(status.si_status));
742 log_full(prio, "%s failed due to unknown reason.", strna(name));
749 * < 0 : wait_for_terminate_with_timeout() failed to get the state of the process, the process timed out, the process
750 * was terminated by a signal, or failed for an unknown reason.
752 * >=0 : The process terminated normally with no failures.
754 * Success is indicated by a return value of zero, a timeout is indicated by ETIMEDOUT, and all other child failure
755 * states are indicated by error is indicated by a non-zero value.
757 * This call assumes SIGCHLD has been blocked already, in particular before the child to wait for has been forked off
758 * to remain entirely race-free.
760 int wait_for_terminate_with_timeout(pid_t pid, usec_t timeout) {
765 assert_se(sigemptyset(&mask) == 0);
766 assert_se(sigaddset(&mask, SIGCHLD) == 0);
768 /* Drop into a sigtimewait-based timeout. Waiting for the
770 until = now(CLOCK_MONOTONIC) + timeout;
773 siginfo_t status = {};
776 n = now(CLOCK_MONOTONIC);
780 r = sigtimedwait(&mask, NULL, timespec_store(&ts, until - n)) < 0 ? -errno : 0;
781 /* Assuming we woke due to the child exiting. */
782 if (waitid(P_PID, pid, &status, WEXITED|WNOHANG) == 0) {
783 if (status.si_pid == pid) {
784 /* This is the correct child.*/
785 if (status.si_code == CLD_EXITED)
786 return (status.si_status == 0) ? 0 : -EPROTO;
791 /* Not the child, check for errors and proceed appropriately */
795 /* Timed out, child is likely hung. */
798 /* Received a different signal and should retry */
801 /* Return any unexpected errors */
810 #if 0 /// UNNEEDED by elogind
811 void sigkill_wait(pid_t pid) {
814 if (kill(pid, SIGKILL) > 0)
815 (void) wait_for_terminate(pid, NULL);
818 void sigkill_waitp(pid_t *pid) {
830 void sigterm_wait(pid_t pid) {
833 if (kill_and_sigcont(pid, SIGTERM) > 0)
834 (void) wait_for_terminate(pid, NULL);
837 int kill_and_sigcont(pid_t pid, int sig) {
840 r = kill(pid, sig) < 0 ? -errno : 0;
842 /* If this worked, also send SIGCONT, unless we already just sent a SIGCONT, or SIGKILL was sent which isn't
843 * affected by a process being suspended anyway. */
844 if (r >= 0 && !IN_SET(sig, SIGCONT, SIGKILL))
845 (void) kill(pid, SIGCONT);
850 int getenv_for_pid(pid_t pid, const char *field, char **ret) {
851 _cleanup_fclose_ FILE *f = NULL;
861 if (pid == 0 || pid == getpid_cached()) {
878 path = procfs_file_alloca(pid, "environ");
880 f = fopen(path, "re");
888 (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
896 for (i = 0; i < sizeof(line)-1; i++) {
900 if (_unlikely_(c == EOF)) {
910 if (strneq(line, field, l) && line[l] == '=') {
911 value = strdup(line + l + 1);
925 bool pid_is_unwaited(pid_t pid) {
926 /* Checks whether a PID is still valid at all, including a zombie */
931 if (pid <= 1) /* If we or PID 1 would be dead and have been waited for, this code would not be running */
934 if (pid == getpid_cached())
937 if (kill(pid, 0) >= 0)
940 return errno != ESRCH;
943 bool pid_is_alive(pid_t pid) {
946 /* Checks whether a PID is still valid and not a zombie */
951 if (pid <= 1) /* If we or PID 1 would be a zombie, this code would not be running */
954 if (pid == getpid_cached())
957 r = get_process_state(pid);
958 if (IN_SET(r, -ESRCH, 'Z'))
964 #if 0 /// UNNEEDED by elogind
965 int pid_from_same_root_fs(pid_t pid) {
971 if (pid == 0 || pid == getpid_cached())
974 root = procfs_file_alloca(pid, "root");
976 return files_same(root, "/proc/1/root", 0);
980 bool is_main_thread(void) {
981 static thread_local int cached = 0;
983 if (_unlikely_(cached == 0))
984 cached = getpid_cached() == gettid() ? 1 : -1;
989 #if 0 /// UNNEEDED by elogind
990 _noreturn_ void freeze(void) {
994 /* Make sure nobody waits for us on a socket anymore */
995 close_all_fds(NULL, 0);
999 /* Let's not freeze right away, but keep reaping zombies. */
1004 r = waitid(P_ALL, 0, &si, WEXITED);
1005 if (r < 0 && errno != EINTR)
1009 /* waitid() failed with an unexpected error, things are really borked. Freeze now! */
1014 bool oom_score_adjust_is_valid(int oa) {
1015 return oa >= OOM_SCORE_ADJ_MIN && oa <= OOM_SCORE_ADJ_MAX;
1018 unsigned long personality_from_string(const char *p) {
1022 return PERSONALITY_INVALID;
1024 /* Parse a personality specifier. We use our own identifiers that indicate specific ABIs, rather than just
1025 * hints regarding the register size, since we want to keep things open for multiple locally supported ABIs for
1026 * the same register size. */
1028 architecture = architecture_from_string(p);
1029 if (architecture < 0)
1030 return PERSONALITY_INVALID;
1032 if (architecture == native_architecture())
1034 #ifdef SECONDARY_ARCHITECTURE
1035 if (architecture == SECONDARY_ARCHITECTURE)
1039 return PERSONALITY_INVALID;
1042 const char* personality_to_string(unsigned long p) {
1043 int architecture = _ARCHITECTURE_INVALID;
1046 architecture = native_architecture();
1047 #ifdef SECONDARY_ARCHITECTURE
1048 else if (p == PER_LINUX32)
1049 architecture = SECONDARY_ARCHITECTURE;
1052 if (architecture < 0)
1055 return architecture_to_string(architecture);
1058 int safe_personality(unsigned long p) {
1061 /* So here's the deal, personality() is weirdly defined by glibc. In some cases it returns a failure via errno,
1062 * and in others as negative return value containing an errno-like value. Let's work around this: this is a
1063 * wrapper that uses errno if it is set, and uses the return value otherwise. And then it sets both errno and
1064 * the return value indicating the same issue, so that we are definitely on the safe side.
1066 * See https://github.com/systemd/systemd/issues/6737 */
1069 ret = personality(p);
1080 int opinionated_personality(unsigned long *ret) {
1083 /* Returns the current personality, or PERSONALITY_INVALID if we can't determine it. This function is a bit
1084 * opinionated though, and ignores all the finer-grained bits and exotic personalities, only distinguishing the
1085 * two most relevant personalities: PER_LINUX and PER_LINUX32. */
1087 current = safe_personality(PERSONALITY_INVALID);
1091 if (((unsigned long) current & 0xffff) == PER_LINUX32)
1099 void valgrind_summary_hack(void) {
1100 #if HAVE_VALGRIND_VALGRIND_H
1101 if (getpid_cached() == 1 && RUNNING_ON_VALGRIND) {
1103 pid = raw_clone(SIGCHLD);
1105 log_emergency_errno(errno, "Failed to fork off valgrind helper: %m");
1109 log_info("Spawned valgrind helper as PID "PID_FMT".", pid);
1110 (void) wait_for_terminate(pid, NULL);
1116 int pid_compare_func(const void *a, const void *b) {
1117 const pid_t *p = a, *q = b;
1119 /* Suitable for usage in qsort() */
1128 int ioprio_parse_priority(const char *s, int *ret) {
1134 r = safe_atoi(s, &i);
1138 if (!ioprio_priority_is_valid(i))
1146 /* The cached PID, possible values:
1148 * == UNSET [0] → cache not initialized yet
1149 * == BUSY [-1] → some thread is initializing it at the moment
1150 * any other → the cached PID
1153 #define CACHED_PID_UNSET ((pid_t) 0)
1154 #define CACHED_PID_BUSY ((pid_t) -1)
1156 static pid_t cached_pid = CACHED_PID_UNSET;
1158 void reset_cached_pid(void) {
1159 /* Invoked in the child after a fork(), i.e. at the first moment the PID changed */
1160 cached_pid = CACHED_PID_UNSET;
1163 /* We use glibc __register_atfork() + __dso_handle directly here, as they are not included in the glibc
1164 * headers. __register_atfork() is mostly equivalent to pthread_atfork(), but doesn't require us to link against
1165 * libpthread, as it is part of glibc anyway. */
1167 extern int __register_atfork(void (*prepare) (void), void (*parent) (void), void (*child) (void), void * __dso_handle);
1168 extern void* __dso_handle __attribute__ ((__weak__));
1169 #endif // ifdef __GLIBC__
1171 pid_t getpid_cached(void) {
1172 static bool installed = false;
1173 pid_t current_value;
1175 /* getpid_cached() is much like getpid(), but caches the value in local memory, to avoid having to invoke a
1176 * system call each time. This restores glibc behaviour from before 2.24, when getpid() was unconditionally
1177 * cached. Starting with 2.24 getpid() started to become prohibitively expensive when used for detecting when
1178 * objects were used across fork()s. With this caching the old behaviour is somewhat restored.
1180 * https://bugzilla.redhat.com/show_bug.cgi?id=1443976
1181 * https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=c579f48edba88380635ab98cb612030e3ed8691e
1184 current_value = __sync_val_compare_and_swap(&cached_pid, CACHED_PID_UNSET, CACHED_PID_BUSY);
1186 switch (current_value) {
1188 case CACHED_PID_UNSET: { /* Not initialized yet, then do so now */
1191 new_pid = raw_getpid();
1194 /* __register_atfork() either returns 0 or -ENOMEM, in its glibc implementation. Since it's
1195 * only half-documented (glibc doesn't document it but LSB does — though only superficially)
1196 * we'll check for errors only in the most generic fashion possible. */
1198 if (__register_atfork(NULL, NULL, reset_cached_pid, __dso_handle) != 0) {
1199 /* OOM? Let's try again later */
1200 cached_pid = CACHED_PID_UNSET;
1207 cached_pid = new_pid;
1211 case CACHED_PID_BUSY: /* Somebody else is currently initializing */
1212 return raw_getpid();
1214 default: /* Properly initialized */
1215 return current_value;
1219 int must_be_root(void) {
1224 log_error("Need to be root.");
1230 const int except_fds[],
1231 size_t n_except_fds,
1235 pid_t original_pid, pid;
1236 sigset_t saved_ss, ss;
1237 bool block_signals = false;
1240 /* A wrapper around fork(), that does a couple of important initializations in addition to mere forking. Always
1241 * returns the child's PID in *ret_pid. Returns == 0 in the child, and > 0 in the parent. */
1243 prio = flags & FORK_LOG ? LOG_ERR : LOG_DEBUG;
1245 original_pid = getpid_cached();
1247 if (flags & (FORK_RESET_SIGNALS|FORK_DEATHSIG)) {
1249 /* We temporarily block all signals, so that the new child has them blocked initially. This way, we can
1250 * be sure that SIGTERMs are not lost we might send to the child. */
1252 if (sigfillset(&ss) < 0)
1253 return log_full_errno(prio, errno, "Failed to reset signal set: %m");
1255 block_signals = true;
1257 } else if (flags & FORK_WAIT) {
1259 /* Let's block SIGCHLD at least, so that we can safely watch for the child process */
1261 if (sigemptyset(&ss) < 0)
1262 return log_full_errno(prio, errno, "Failed to clear signal set: %m");
1264 if (sigaddset(&ss, SIGCHLD) < 0)
1265 return log_full_errno(prio, errno, "Failed to add SIGCHLD to signal set: %m");
1267 block_signals = true;
1271 if (sigprocmask(SIG_SETMASK, &ss, &saved_ss) < 0)
1272 return log_full_errno(prio, errno, "Failed to set signal mask: %m");
1274 if (flags & FORK_NEW_MOUNTNS)
1275 pid = raw_clone(SIGCHLD|CLONE_NEWNS);
1281 if (block_signals) /* undo what we did above */
1282 (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL);
1284 return log_full_errno(prio, r, "Failed to fork: %m");
1287 /* We are in the parent process */
1289 log_debug("Successfully forked off '%s' as PID " PID_FMT ".", strna(name), pid);
1291 if (flags & FORK_WAIT) {
1292 r = wait_for_terminate_and_check(name, pid, (flags & FORK_LOG ? WAIT_LOG : 0));
1295 if (r != EXIT_SUCCESS) /* exit status > 0 should be treated as failure, too */
1299 if (block_signals) /* undo what we did above */
1300 (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL);
1308 /* We are in the child process */
1310 if (flags & FORK_REOPEN_LOG) {
1311 /* Close the logs if requested, before we log anything. And make sure we reopen it if needed. */
1313 log_set_open_when_needed(true);
1317 r = rename_process(name);
1319 log_full_errno(flags & FORK_LOG ? LOG_WARNING : LOG_DEBUG,
1320 r, "Failed to rename process, ignoring: %m");
1323 if (flags & FORK_DEATHSIG)
1324 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0) {
1325 log_full_errno(prio, errno, "Failed to set death signal: %m");
1326 _exit(EXIT_FAILURE);
1329 if (flags & FORK_RESET_SIGNALS) {
1330 r = reset_all_signal_handlers();
1332 log_full_errno(prio, r, "Failed to reset signal handlers: %m");
1333 _exit(EXIT_FAILURE);
1336 /* This implicitly undoes the signal mask stuff we did before the fork()ing above */
1337 r = reset_signal_mask();
1339 log_full_errno(prio, r, "Failed to reset signal mask: %m");
1340 _exit(EXIT_FAILURE);
1342 } else if (block_signals) { /* undo what we did above */
1343 if (sigprocmask(SIG_SETMASK, &saved_ss, NULL) < 0) {
1344 log_full_errno(prio, errno, "Failed to restore signal mask: %m");
1345 _exit(EXIT_FAILURE);
1349 if (flags & FORK_DEATHSIG) {
1351 /* Let's see if the parent PID is still the one we started from? If not, then the parent
1352 * already died by the time we set PR_SET_PDEATHSIG, hence let's emulate the effect */
1356 /* Parent is in a differn't PID namespace. */;
1357 else if (ppid != original_pid) {
1358 log_debug("Parent died early, raising SIGTERM.");
1359 (void) raise(SIGTERM);
1360 _exit(EXIT_FAILURE);
1364 if (FLAGS_SET(flags, FORK_NEW_MOUNTNS | FORK_MOUNTNS_SLAVE)) {
1366 /* Optionally, make sure we never propagate mounts to the host. */
1368 if (mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL) < 0) {
1369 log_full_errno(prio, errno, "Failed to remount root directory as MS_SLAVE: %m");
1370 _exit(EXIT_FAILURE);
1374 if (flags & FORK_CLOSE_ALL_FDS) {
1375 /* Close the logs here in case it got reopened above, as close_all_fds() would close them for us */
1378 r = close_all_fds(except_fds, n_except_fds);
1380 log_full_errno(prio, r, "Failed to close all file descriptors: %m");
1381 _exit(EXIT_FAILURE);
1385 /* When we were asked to reopen the logs, do so again now */
1386 if (flags & FORK_REOPEN_LOG) {
1388 log_set_open_when_needed(false);
1391 if (flags & FORK_NULL_STDIO) {
1392 r = make_null_stdio();
1394 log_full_errno(prio, r, "Failed to connect stdin/stdout to /dev/null: %m");
1395 _exit(EXIT_FAILURE);
1400 *ret_pid = getpid_cached();
1405 int fork_agent(const char *name, const int except[], size_t n_except, pid_t *ret_pid, const char *path, ...) {
1406 bool stdout_is_tty, stderr_is_tty;
1414 /* Spawns a temporary TTY agent, making sure it goes away when we go away */
1416 r = safe_fork_full(name, except, n_except, FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_CLOSE_ALL_FDS, ret_pid);
1424 stdout_is_tty = isatty(STDOUT_FILENO);
1425 stderr_is_tty = isatty(STDERR_FILENO);
1427 if (!stdout_is_tty || !stderr_is_tty) {
1430 /* Detach from stdout/stderr. and reopen
1431 * /dev/tty for them. This is important to
1432 * ensure that when systemctl is started via
1433 * popen() or a similar call that expects to
1434 * read EOF we actually do generate EOF and
1435 * not delay this indefinitely by because we
1436 * keep an unused copy of stdin around. */
1437 fd = open("/dev/tty", O_WRONLY);
1439 log_error_errno(errno, "Failed to open /dev/tty: %m");
1440 _exit(EXIT_FAILURE);
1443 if (!stdout_is_tty && dup2(fd, STDOUT_FILENO) < 0) {
1444 log_error_errno(errno, "Failed to dup2 /dev/tty: %m");
1445 _exit(EXIT_FAILURE);
1448 if (!stderr_is_tty && dup2(fd, STDERR_FILENO) < 0) {
1449 log_error_errno(errno, "Failed to dup2 /dev/tty: %m");
1450 _exit(EXIT_FAILURE);
1453 safe_close_above_stdio(fd);
1456 /* Count arguments */
1458 for (n = 0; va_arg(ap, char*); n++)
1463 l = newa(char*, n + 1);
1465 /* Fill in arguments */
1467 for (i = 0; i <= n; i++)
1468 l[i] = va_arg(ap, char*);
1472 _exit(EXIT_FAILURE);
1475 int set_oom_score_adjust(int value) {
1476 char t[DECIMAL_STR_MAX(int)];
1478 sprintf(t, "%i", value);
1480 return write_string_file("/proc/self/oom_score_adj", t,
1481 WRITE_STRING_FILE_VERIFY_ON_FAILURE|WRITE_STRING_FILE_DISABLE_BUFFER);
1484 #if 0 /// UNNEEDED by elogind
1485 static const char *const ioprio_class_table[] = {
1486 [IOPRIO_CLASS_NONE] = "none",
1487 [IOPRIO_CLASS_RT] = "realtime",
1488 [IOPRIO_CLASS_BE] = "best-effort",
1489 [IOPRIO_CLASS_IDLE] = "idle"
1492 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ioprio_class, int, IOPRIO_N_CLASSES);
1494 static const char *const sigchld_code_table[] = {
1495 [CLD_EXITED] = "exited",
1496 [CLD_KILLED] = "killed",
1497 [CLD_DUMPED] = "dumped",
1498 [CLD_TRAPPED] = "trapped",
1499 [CLD_STOPPED] = "stopped",
1500 [CLD_CONTINUED] = "continued",
1503 DEFINE_STRING_TABLE_LOOKUP(sigchld_code, int);
1505 static const char* const sched_policy_table[] = {
1506 [SCHED_OTHER] = "other",
1507 [SCHED_BATCH] = "batch",
1508 [SCHED_IDLE] = "idle",
1509 [SCHED_FIFO] = "fifo",
1513 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(sched_policy, int, INT_MAX);