1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 This file is part of systemd.
5 Copyright 2010 Lennart Poettering
11 #include <linux/oom.h>
16 #include <stdio_ext.h>
20 //#include <sys/mount.h>
21 #include <sys/personality.h>
22 #include <sys/prctl.h>
23 #include <sys/types.h>
27 #if HAVE_VALGRIND_VALGRIND_H
28 #include <valgrind/valgrind.h>
31 #include "alloc-util.h"
32 //#include "architecture.h"
41 #include "process-util.h"
42 #include "raw-clone.h"
43 #include "signal-util.h"
44 //#include "stat-util.h"
45 #include "string-table.h"
46 #include "string-util.h"
47 //#include "terminal-util.h"
48 #include "user-util.h"
51 int get_process_state(pid_t pid) {
55 _cleanup_free_ char *line = NULL;
59 p = procfs_file_alloca(pid, "stat");
61 r = read_one_line_file(p, &line);
67 p = strrchr(line, ')');
73 if (sscanf(p, " %c", &state) != 1)
76 return (unsigned char) state;
79 int get_process_comm(pid_t pid, char **name) {
86 p = procfs_file_alloca(pid, "comm");
88 r = read_one_line_file(p, name);
95 int get_process_cmdline(pid_t pid, size_t max_length, bool comm_fallback, char **line) {
96 _cleanup_fclose_ FILE *f = NULL;
105 /* Retrieves a process' command line. Replaces unprintable characters while doing so by whitespace (coalescing
106 * multiple sequential ones into one). If max_length is != 0 will return a string of the specified size at most
107 * (the trailing NUL byte does count towards the length here!), abbreviated with a "..." ellipsis. If
108 * comm_fallback is true and the process has no command line set (the case for kernel threads), or has a
109 * command line that resolves to the empty string will return the "comm" name of the process instead.
111 * Returns -ESRCH if the process doesn't exist, and -ENOENT if the process has no command line (and
112 * comm_fallback is false). Returns 0 and sets *line otherwise. */
114 p = procfs_file_alloca(pid, "cmdline");
123 (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
125 if (max_length == 1) {
127 /* If there's only room for one byte, return the empty string */
135 } else if (max_length == 0) {
136 size_t len = 0, allocated = 0;
138 while ((c = getc(f)) != EOF) {
140 if (!GREEDY_REALLOC(ans, allocated, len+3)) {
162 bool dotdotdot = false;
165 ans = new(char, max_length);
171 while ((c = getc(f)) != EOF) {
198 if (max_length <= 4) {
202 k = ans + max_length - 4;
205 /* Eat up final spaces */
206 while (k > ans && isspace(k[-1])) {
212 strncpy(k, "...", left-1);
218 /* Kernel threads have no argv[] */
220 _cleanup_free_ char *t = NULL;
228 h = get_process_comm(pid, &t);
233 ans = strjoin("[", t, "]");
239 if (l + 3 <= max_length)
240 ans = strjoin("[", t, "]");
241 else if (max_length <= 6) {
243 ans = new(char, max_length);
247 memcpy(ans, "[...]", max_length-1);
248 ans[max_length-1] = 0;
252 t[max_length - 6] = 0;
254 /* Chop off final spaces */
256 while (e > t && isspace(e[-1]))
260 ans = strjoin("[", t, "...]");
271 int rename_process(const char name[]) {
272 static size_t mm_size = 0;
273 static char *mm = NULL;
274 bool truncated = false;
277 /* This is a like a poor man's setproctitle(). It changes the comm field, argv[0], and also the glibc's
278 * internally used name of the process. For the first one a limit of 16 chars applies; to the second one in
279 * many cases one of 10 (i.e. length of "/sbin/init") — however if we have CAP_SYS_RESOURCES it is unbounded;
280 * to the third one 7 (i.e. the length of "systemd". If you pass a longer string it will likely be
283 * Returns 0 if a name was set but truncated, > 0 if it was set but not truncated. */
286 return -EINVAL; /* let's not confuse users unnecessarily with an empty name */
288 if (!is_main_thread())
289 return -EPERM; /* Let's not allow setting the process name from other threads than the main one, as we
290 * cache things without locking, and we make assumptions that PR_SET_NAME sets the
291 * process name that isn't correct on any other threads */
295 /* First step, change the comm field. The main thread's comm is identical to the process comm. This means we
296 * can use PR_SET_NAME, which sets the thread name for the calling thread. */
297 if (prctl(PR_SET_NAME, name) < 0)
298 log_debug_errno(errno, "PR_SET_NAME failed: %m");
299 if (l > 15) /* Linux process names can be 15 chars at max */
302 /* Second step, change glibc's ID of the process name. */
303 if (program_invocation_name) {
306 k = strlen(program_invocation_name);
307 strncpy(program_invocation_name, name, k);
312 /* Third step, completely replace the argv[] array the kernel maintains for us. This requires privileges, but
313 * has the advantage that the argv[] array is exactly what we want it to be, and not filled up with zeros at
314 * the end. This is the best option for changing /proc/self/cmdline. */
316 /* Let's not bother with this if we don't have euid == 0. Strictly speaking we should check for the
317 * CAP_SYS_RESOURCE capability which is independent of the euid. In our own code the capability generally is
318 * present only for euid == 0, hence let's use this as quick bypass check, to avoid calling mmap() if
319 * PR_SET_MM_ARG_{START,END} fails with EPERM later on anyway. After all geteuid() is dead cheap to call, but
322 log_debug("Skipping PR_SET_MM, as we don't have privileges.");
323 else if (mm_size < l+1) {
327 nn_size = PAGE_ALIGN(l+1);
328 nn = mmap(NULL, nn_size, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
329 if (nn == MAP_FAILED) {
330 log_debug_errno(errno, "mmap() failed: %m");
334 strncpy(nn, name, nn_size);
336 /* Now, let's tell the kernel about this new memory */
337 if (prctl(PR_SET_MM, PR_SET_MM_ARG_START, (unsigned long) nn, 0, 0) < 0) {
338 log_debug_errno(errno, "PR_SET_MM_ARG_START failed, proceeding without: %m");
339 (void) munmap(nn, nn_size);
343 /* And update the end pointer to the new end, too. If this fails, we don't really know what to do, it's
344 * pretty unlikely that we can rollback, hence we'll just accept the failure, and continue. */
345 if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) nn + l + 1, 0, 0) < 0)
346 log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m");
349 (void) munmap(mm, mm_size);
354 strncpy(mm, name, mm_size);
356 /* Update the end pointer, continuing regardless of any failure. */
357 if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) mm + l + 1, 0, 0) < 0)
358 log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m");
362 /* Fourth step: in all cases we'll also update the original argv[], so that our own code gets it right too if
363 * it still looks here */
365 if (saved_argc > 0) {
371 k = strlen(saved_argv[0]);
372 strncpy(saved_argv[0], name, k);
377 for (i = 1; i < saved_argc; i++) {
381 memzero(saved_argv[i], strlen(saved_argv[i]));
388 int is_kernel_thread(pid_t pid) {
389 _cleanup_free_ char *line = NULL;
390 unsigned long long flags;
396 if (IN_SET(pid, 0, 1) || pid == getpid_cached()) /* pid 1, and we ourselves certainly aren't a kernel thread */
398 if (!pid_is_valid(pid))
401 p = procfs_file_alloca(pid, "stat");
402 r = read_one_line_file(p, &line);
408 /* Skip past the comm field */
409 q = strrchr(line, ')');
414 /* Skip 6 fields to reach the flags field */
415 for (i = 0; i < 6; i++) {
416 l = strspn(q, WHITESPACE);
421 l = strcspn(q, WHITESPACE);
427 /* Skip preceeding whitespace */
428 l = strspn(q, WHITESPACE);
433 /* Truncate the rest */
434 l = strcspn(q, WHITESPACE);
439 r = safe_atollu(q, &flags);
443 return !!(flags & PF_KTHREAD);
446 #if 0 /// UNNEEDED by elogind
447 int get_process_capeff(pid_t pid, char **capeff) {
454 p = procfs_file_alloca(pid, "status");
456 r = get_proc_field(p, "CapEff", WHITESPACE, capeff);
464 static int get_process_link_contents(const char *proc_file, char **name) {
470 r = readlink_malloc(proc_file, name);
479 int get_process_exe(pid_t pid, char **name) {
486 p = procfs_file_alloca(pid, "exe");
487 r = get_process_link_contents(p, name);
491 d = endswith(*name, " (deleted)");
498 #if 0 /// UNNEEDED by elogind
499 static int get_process_id(pid_t pid, const char *field, uid_t *uid) {
500 _cleanup_fclose_ FILE *f = NULL;
510 p = procfs_file_alloca(pid, "status");
518 (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
520 FOREACH_LINE(line, f, return -errno) {
525 if (startswith(l, field)) {
527 l += strspn(l, WHITESPACE);
529 l[strcspn(l, WHITESPACE)] = 0;
531 return parse_uid(l, uid);
538 int get_process_uid(pid_t pid, uid_t *uid) {
540 if (pid == 0 || pid == getpid_cached()) {
545 return get_process_id(pid, "Uid:", uid);
548 int get_process_gid(pid_t pid, gid_t *gid) {
550 if (pid == 0 || pid == getpid_cached()) {
555 assert_cc(sizeof(uid_t) == sizeof(gid_t));
556 return get_process_id(pid, "Gid:", gid);
559 int get_process_cwd(pid_t pid, char **cwd) {
564 p = procfs_file_alloca(pid, "cwd");
566 return get_process_link_contents(p, cwd);
569 int get_process_root(pid_t pid, char **root) {
574 p = procfs_file_alloca(pid, "root");
576 return get_process_link_contents(p, root);
579 int get_process_environ(pid_t pid, char **env) {
580 _cleanup_fclose_ FILE *f = NULL;
581 _cleanup_free_ char *outcome = NULL;
584 size_t allocated = 0, sz = 0;
589 p = procfs_file_alloca(pid, "environ");
598 (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
600 while ((c = fgetc(f)) != EOF) {
601 if (!GREEDY_REALLOC(outcome, allocated, sz + 5))
605 outcome[sz++] = '\n';
607 sz += cescape_char(c, outcome + sz);
611 outcome = strdup("");
617 *env = TAKE_PTR(outcome);
622 int get_process_ppid(pid_t pid, pid_t *_ppid) {
624 _cleanup_free_ char *line = NULL;
631 if (pid == 0 || pid == getpid_cached()) {
636 p = procfs_file_alloca(pid, "stat");
637 r = read_one_line_file(p, &line);
643 /* Let's skip the pid and comm fields. The latter is enclosed
644 * in () but does not escape any () in its value, so let's
645 * skip over it manually */
647 p = strrchr(line, ')');
659 if ((long unsigned) (pid_t) ppid != ppid)
662 *_ppid = (pid_t) ppid;
668 int wait_for_terminate(pid_t pid, siginfo_t *status) {
679 if (waitid(P_PID, pid, status, WEXITED) < 0) {
684 return negative_errno();
693 * < 0 : wait_for_terminate() failed to get the state of the
694 * process, the process was terminated by a signal, or
695 * failed for an unknown reason.
696 * >=0 : The process terminated normally, and its exit code is
699 * That is, success is indicated by a return value of zero, and an
700 * error is indicated by a non-zero value.
702 * A warning is emitted if the process terminates abnormally,
703 * and also if it returns non-zero unless check_exit_code is true.
705 int wait_for_terminate_and_check(const char *name, pid_t pid, WaitFlags flags) {
706 _cleanup_free_ char *buffer = NULL;
713 r = get_process_comm(pid, &buffer);
715 log_debug_errno(r, "Failed to acquire process name of " PID_FMT ", ignoring: %m", pid);
720 prio = flags & WAIT_LOG_ABNORMAL ? LOG_ERR : LOG_DEBUG;
722 r = wait_for_terminate(pid, &status);
724 return log_full_errno(prio, r, "Failed to wait for %s: %m", strna(name));
726 if (status.si_code == CLD_EXITED) {
727 if (status.si_status != EXIT_SUCCESS)
728 log_full(flags & WAIT_LOG_NON_ZERO_EXIT_STATUS ? LOG_ERR : LOG_DEBUG,
729 "%s failed with exit status %i.", strna(name), status.si_status);
731 log_debug("%s succeeded.", name);
733 return status.si_status;
735 } else if (IN_SET(status.si_code, CLD_KILLED, CLD_DUMPED)) {
737 log_full(prio, "%s terminated by signal %s.", strna(name), signal_to_string(status.si_status));
741 log_full(prio, "%s failed due to unknown reason.", strna(name));
747 * < 0 : wait_for_terminate_with_timeout() failed to get the state of the
748 * process, the process timed out, the process was terminated by a
749 * signal, or failed for an unknown reason.
750 * >=0 : The process terminated normally with no failures.
752 * Success is indicated by a return value of zero, a timeout is indicated
753 * by ETIMEDOUT, and all other child failure states are indicated by error
754 * is indicated by a non-zero value.
756 int wait_for_terminate_with_timeout(pid_t pid, usec_t timeout) {
761 assert_se(sigemptyset(&mask) == 0);
762 assert_se(sigaddset(&mask, SIGCHLD) == 0);
764 /* Drop into a sigtimewait-based timeout. Waiting for the
766 until = now(CLOCK_MONOTONIC) + timeout;
769 siginfo_t status = {};
772 n = now(CLOCK_MONOTONIC);
776 r = sigtimedwait(&mask, NULL, timespec_store(&ts, until - n)) < 0 ? -errno : 0;
777 /* Assuming we woke due to the child exiting. */
778 if (waitid(P_PID, pid, &status, WEXITED|WNOHANG) == 0) {
779 if (status.si_pid == pid) {
780 /* This is the correct child.*/
781 if (status.si_code == CLD_EXITED)
782 return (status.si_status == 0) ? 0 : -EPROTO;
787 /* Not the child, check for errors and proceed appropriately */
791 /* Timed out, child is likely hung. */
794 /* Received a different signal and should retry */
797 /* Return any unexpected errors */
806 #if 0 /// UNNEEDED by elogind
807 void sigkill_wait(pid_t pid) {
810 if (kill(pid, SIGKILL) > 0)
811 (void) wait_for_terminate(pid, NULL);
814 void sigkill_waitp(pid_t *pid) {
826 void sigterm_wait(pid_t pid) {
829 if (kill_and_sigcont(pid, SIGTERM) > 0)
830 (void) wait_for_terminate(pid, NULL);
833 int kill_and_sigcont(pid_t pid, int sig) {
836 r = kill(pid, sig) < 0 ? -errno : 0;
838 /* If this worked, also send SIGCONT, unless we already just sent a SIGCONT, or SIGKILL was sent which isn't
839 * affected by a process being suspended anyway. */
840 if (r >= 0 && !IN_SET(sig, SIGCONT, SIGKILL))
841 (void) kill(pid, SIGCONT);
846 int getenv_for_pid(pid_t pid, const char *field, char **ret) {
847 _cleanup_fclose_ FILE *f = NULL;
857 if (pid == 0 || pid == getpid_cached()) {
874 path = procfs_file_alloca(pid, "environ");
876 f = fopen(path, "re");
884 (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
892 for (i = 0; i < sizeof(line)-1; i++) {
896 if (_unlikely_(c == EOF)) {
906 if (strneq(line, field, l) && line[l] == '=') {
907 value = strdup(line + l + 1);
921 bool pid_is_unwaited(pid_t pid) {
922 /* Checks whether a PID is still valid at all, including a zombie */
927 if (pid <= 1) /* If we or PID 1 would be dead and have been waited for, this code would not be running */
930 if (pid == getpid_cached())
933 if (kill(pid, 0) >= 0)
936 return errno != ESRCH;
939 bool pid_is_alive(pid_t pid) {
942 /* Checks whether a PID is still valid and not a zombie */
947 if (pid <= 1) /* If we or PID 1 would be a zombie, this code would not be running */
950 if (pid == getpid_cached())
953 r = get_process_state(pid);
954 if (IN_SET(r, -ESRCH, 'Z'))
960 #if 0 /// UNNEEDED by elogind
961 int pid_from_same_root_fs(pid_t pid) {
967 if (pid == 0 || pid == getpid_cached())
970 root = procfs_file_alloca(pid, "root");
972 return files_same(root, "/proc/1/root", 0);
976 bool is_main_thread(void) {
977 static thread_local int cached = 0;
979 if (_unlikely_(cached == 0))
980 cached = getpid_cached() == gettid() ? 1 : -1;
985 #if 0 /// UNNEEDED by elogind
986 _noreturn_ void freeze(void) {
990 /* Make sure nobody waits for us on a socket anymore */
991 close_all_fds(NULL, 0);
995 /* Let's not freeze right away, but keep reaping zombies. */
1000 r = waitid(P_ALL, 0, &si, WEXITED);
1001 if (r < 0 && errno != EINTR)
1005 /* waitid() failed with an unexpected error, things are really borked. Freeze now! */
1010 bool oom_score_adjust_is_valid(int oa) {
1011 return oa >= OOM_SCORE_ADJ_MIN && oa <= OOM_SCORE_ADJ_MAX;
1014 unsigned long personality_from_string(const char *p) {
1018 return PERSONALITY_INVALID;
1020 /* Parse a personality specifier. We use our own identifiers that indicate specific ABIs, rather than just
1021 * hints regarding the register size, since we want to keep things open for multiple locally supported ABIs for
1022 * the same register size. */
1024 architecture = architecture_from_string(p);
1025 if (architecture < 0)
1026 return PERSONALITY_INVALID;
1028 if (architecture == native_architecture())
1030 #ifdef SECONDARY_ARCHITECTURE
1031 if (architecture == SECONDARY_ARCHITECTURE)
1035 return PERSONALITY_INVALID;
1038 const char* personality_to_string(unsigned long p) {
1039 int architecture = _ARCHITECTURE_INVALID;
1042 architecture = native_architecture();
1043 #ifdef SECONDARY_ARCHITECTURE
1044 else if (p == PER_LINUX32)
1045 architecture = SECONDARY_ARCHITECTURE;
1048 if (architecture < 0)
1051 return architecture_to_string(architecture);
1054 int safe_personality(unsigned long p) {
1057 /* So here's the deal, personality() is weirdly defined by glibc. In some cases it returns a failure via errno,
1058 * and in others as negative return value containing an errno-like value. Let's work around this: this is a
1059 * wrapper that uses errno if it is set, and uses the return value otherwise. And then it sets both errno and
1060 * the return value indicating the same issue, so that we are definitely on the safe side.
1062 * See https://github.com/systemd/systemd/issues/6737 */
1065 ret = personality(p);
1076 int opinionated_personality(unsigned long *ret) {
1079 /* Returns the current personality, or PERSONALITY_INVALID if we can't determine it. This function is a bit
1080 * opinionated though, and ignores all the finer-grained bits and exotic personalities, only distinguishing the
1081 * two most relevant personalities: PER_LINUX and PER_LINUX32. */
1083 current = safe_personality(PERSONALITY_INVALID);
1087 if (((unsigned long) current & 0xffff) == PER_LINUX32)
1095 void valgrind_summary_hack(void) {
1096 #if HAVE_VALGRIND_VALGRIND_H
1097 if (getpid_cached() == 1 && RUNNING_ON_VALGRIND) {
1099 pid = raw_clone(SIGCHLD);
1101 log_emergency_errno(errno, "Failed to fork off valgrind helper: %m");
1105 log_info("Spawned valgrind helper as PID "PID_FMT".", pid);
1106 (void) wait_for_terminate(pid, NULL);
1112 int pid_compare_func(const void *a, const void *b) {
1113 const pid_t *p = a, *q = b;
1115 /* Suitable for usage in qsort() */
1124 int ioprio_parse_priority(const char *s, int *ret) {
1130 r = safe_atoi(s, &i);
1134 if (!ioprio_priority_is_valid(i))
1142 /* The cached PID, possible values:
1144 * == UNSET [0] → cache not initialized yet
1145 * == BUSY [-1] → some thread is initializing it at the moment
1146 * any other → the cached PID
1149 #define CACHED_PID_UNSET ((pid_t) 0)
1150 #define CACHED_PID_BUSY ((pid_t) -1)
1152 static pid_t cached_pid = CACHED_PID_UNSET;
1154 void reset_cached_pid(void) {
1155 /* Invoked in the child after a fork(), i.e. at the first moment the PID changed */
1156 cached_pid = CACHED_PID_UNSET;
1159 /* We use glibc __register_atfork() + __dso_handle directly here, as they are not included in the glibc
1160 * headers. __register_atfork() is mostly equivalent to pthread_atfork(), but doesn't require us to link against
1161 * libpthread, as it is part of glibc anyway. */
1163 extern int __register_atfork(void (*prepare) (void), void (*parent) (void), void (*child) (void), void * __dso_handle);
1164 extern void* __dso_handle __attribute__ ((__weak__));
1165 #endif // ifdef __GLIBC__
1167 pid_t getpid_cached(void) {
1168 static bool installed = false;
1169 pid_t current_value;
1171 /* getpid_cached() is much like getpid(), but caches the value in local memory, to avoid having to invoke a
1172 * system call each time. This restores glibc behaviour from before 2.24, when getpid() was unconditionally
1173 * cached. Starting with 2.24 getpid() started to become prohibitively expensive when used for detecting when
1174 * objects were used across fork()s. With this caching the old behaviour is somewhat restored.
1176 * https://bugzilla.redhat.com/show_bug.cgi?id=1443976
1177 * https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=c579f48edba88380635ab98cb612030e3ed8691e
1180 current_value = __sync_val_compare_and_swap(&cached_pid, CACHED_PID_UNSET, CACHED_PID_BUSY);
1182 switch (current_value) {
1184 case CACHED_PID_UNSET: { /* Not initialized yet, then do so now */
1187 new_pid = raw_getpid();
1190 /* __register_atfork() either returns 0 or -ENOMEM, in its glibc implementation. Since it's
1191 * only half-documented (glibc doesn't document it but LSB does — though only superficially)
1192 * we'll check for errors only in the most generic fashion possible. */
1194 if (__register_atfork(NULL, NULL, reset_cached_pid, __dso_handle) != 0) {
1195 /* OOM? Let's try again later */
1196 cached_pid = CACHED_PID_UNSET;
1203 cached_pid = new_pid;
1207 case CACHED_PID_BUSY: /* Somebody else is currently initializing */
1208 return raw_getpid();
1210 default: /* Properly initialized */
1211 return current_value;
1215 int must_be_root(void) {
1220 log_error("Need to be root.");
1226 const int except_fds[],
1227 size_t n_except_fds,
1231 pid_t original_pid, pid;
1232 sigset_t saved_ss, ss;
1233 bool block_signals = false;
1236 /* A wrapper around fork(), that does a couple of important initializations in addition to mere forking. Always
1237 * returns the child's PID in *ret_pid. Returns == 0 in the child, and > 0 in the parent. */
1239 prio = flags & FORK_LOG ? LOG_ERR : LOG_DEBUG;
1241 original_pid = getpid_cached();
1243 if (flags & (FORK_RESET_SIGNALS|FORK_DEATHSIG)) {
1245 /* We temporarily block all signals, so that the new child has them blocked initially. This way, we can
1246 * be sure that SIGTERMs are not lost we might send to the child. */
1248 if (sigfillset(&ss) < 0)
1249 return log_full_errno(prio, errno, "Failed to reset signal set: %m");
1251 block_signals = true;
1253 } else if (flags & FORK_WAIT) {
1255 /* Let's block SIGCHLD at least, so that we can safely watch for the child process */
1257 if (sigemptyset(&ss) < 0)
1258 return log_full_errno(prio, errno, "Failed to clear signal set: %m");
1260 if (sigaddset(&ss, SIGCHLD) < 0)
1261 return log_full_errno(prio, errno, "Failed to add SIGCHLD to signal set: %m");
1263 block_signals = true;
1267 if (sigprocmask(SIG_SETMASK, &ss, &saved_ss) < 0)
1268 return log_full_errno(prio, errno, "Failed to set signal mask: %m");
1270 if (flags & FORK_NEW_MOUNTNS)
1271 pid = raw_clone(SIGCHLD|CLONE_NEWNS);
1277 if (block_signals) /* undo what we did above */
1278 (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL);
1280 return log_full_errno(prio, r, "Failed to fork: %m");
1283 /* We are in the parent process */
1285 log_debug("Successfully forked off '%s' as PID " PID_FMT ".", strna(name), pid);
1287 if (flags & FORK_WAIT) {
1288 r = wait_for_terminate_and_check(name, pid, (flags & FORK_LOG ? WAIT_LOG : 0));
1291 if (r != EXIT_SUCCESS) /* exit status > 0 should be treated as failure, too */
1295 if (block_signals) /* undo what we did above */
1296 (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL);
1304 /* We are in the child process */
1306 if (flags & FORK_REOPEN_LOG) {
1307 /* Close the logs if requested, before we log anything. And make sure we reopen it if needed. */
1309 log_set_open_when_needed(true);
1313 r = rename_process(name);
1315 log_full_errno(flags & FORK_LOG ? LOG_WARNING : LOG_DEBUG,
1316 r, "Failed to rename process, ignoring: %m");
1319 if (flags & FORK_DEATHSIG)
1320 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0) {
1321 log_full_errno(prio, errno, "Failed to set death signal: %m");
1322 _exit(EXIT_FAILURE);
1325 if (flags & FORK_RESET_SIGNALS) {
1326 r = reset_all_signal_handlers();
1328 log_full_errno(prio, r, "Failed to reset signal handlers: %m");
1329 _exit(EXIT_FAILURE);
1332 /* This implicitly undoes the signal mask stuff we did before the fork()ing above */
1333 r = reset_signal_mask();
1335 log_full_errno(prio, r, "Failed to reset signal mask: %m");
1336 _exit(EXIT_FAILURE);
1338 } else if (block_signals) { /* undo what we did above */
1339 if (sigprocmask(SIG_SETMASK, &saved_ss, NULL) < 0) {
1340 log_full_errno(prio, errno, "Failed to restore signal mask: %m");
1341 _exit(EXIT_FAILURE);
1345 if (flags & FORK_DEATHSIG) {
1347 /* Let's see if the parent PID is still the one we started from? If not, then the parent
1348 * already died by the time we set PR_SET_PDEATHSIG, hence let's emulate the effect */
1352 /* Parent is in a differn't PID namespace. */;
1353 else if (ppid != original_pid) {
1354 log_debug("Parent died early, raising SIGTERM.");
1355 (void) raise(SIGTERM);
1356 _exit(EXIT_FAILURE);
1360 if ((flags & (FORK_NEW_MOUNTNS|FORK_MOUNTNS_SLAVE)) == (FORK_NEW_MOUNTNS|FORK_MOUNTNS_SLAVE)) {
1362 /* Optionally, make sure we never propagate mounts to the host. */
1364 if (mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL) < 0) {
1365 log_full_errno(prio, errno, "Failed to remount root directory as MS_SLAVE: %m");
1366 _exit(EXIT_FAILURE);
1370 if (flags & FORK_CLOSE_ALL_FDS) {
1371 /* Close the logs here in case it got reopened above, as close_all_fds() would close them for us */
1374 r = close_all_fds(except_fds, n_except_fds);
1376 log_full_errno(prio, r, "Failed to close all file descriptors: %m");
1377 _exit(EXIT_FAILURE);
1381 /* When we were asked to reopen the logs, do so again now */
1382 if (flags & FORK_REOPEN_LOG) {
1384 log_set_open_when_needed(false);
1387 if (flags & FORK_NULL_STDIO) {
1388 r = make_null_stdio();
1390 log_full_errno(prio, r, "Failed to connect stdin/stdout to /dev/null: %m");
1391 _exit(EXIT_FAILURE);
1396 *ret_pid = getpid_cached();
1401 int fork_agent(const char *name, const int except[], size_t n_except, pid_t *ret_pid, const char *path, ...) {
1402 bool stdout_is_tty, stderr_is_tty;
1410 /* Spawns a temporary TTY agent, making sure it goes away when we go away */
1412 r = safe_fork_full(name, except, n_except, FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_CLOSE_ALL_FDS, ret_pid);
1420 stdout_is_tty = isatty(STDOUT_FILENO);
1421 stderr_is_tty = isatty(STDERR_FILENO);
1423 if (!stdout_is_tty || !stderr_is_tty) {
1426 /* Detach from stdout/stderr. and reopen
1427 * /dev/tty for them. This is important to
1428 * ensure that when systemctl is started via
1429 * popen() or a similar call that expects to
1430 * read EOF we actually do generate EOF and
1431 * not delay this indefinitely by because we
1432 * keep an unused copy of stdin around. */
1433 fd = open("/dev/tty", O_WRONLY);
1435 log_error_errno(errno, "Failed to open /dev/tty: %m");
1436 _exit(EXIT_FAILURE);
1439 if (!stdout_is_tty && dup2(fd, STDOUT_FILENO) < 0) {
1440 log_error_errno(errno, "Failed to dup2 /dev/tty: %m");
1441 _exit(EXIT_FAILURE);
1444 if (!stderr_is_tty && dup2(fd, STDERR_FILENO) < 0) {
1445 log_error_errno(errno, "Failed to dup2 /dev/tty: %m");
1446 _exit(EXIT_FAILURE);
1449 safe_close_above_stdio(fd);
1452 /* Count arguments */
1454 for (n = 0; va_arg(ap, char*); n++)
1459 l = newa(char*, n + 1);
1461 /* Fill in arguments */
1463 for (i = 0; i <= n; i++)
1464 l[i] = va_arg(ap, char*);
1468 _exit(EXIT_FAILURE);
1471 int set_oom_score_adjust(int value) {
1472 char t[DECIMAL_STR_MAX(int)];
1474 sprintf(t, "%i", value);
1476 return write_string_file("/proc/self/oom_score_adj", t,
1477 WRITE_STRING_FILE_VERIFY_ON_FAILURE|WRITE_STRING_FILE_DISABLE_BUFFER);
1480 #if 0 /// UNNEEDED by elogind
1481 static const char *const ioprio_class_table[] = {
1482 [IOPRIO_CLASS_NONE] = "none",
1483 [IOPRIO_CLASS_RT] = "realtime",
1484 [IOPRIO_CLASS_BE] = "best-effort",
1485 [IOPRIO_CLASS_IDLE] = "idle"
1488 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ioprio_class, int, IOPRIO_N_CLASSES);
1490 static const char *const sigchld_code_table[] = {
1491 [CLD_EXITED] = "exited",
1492 [CLD_KILLED] = "killed",
1493 [CLD_DUMPED] = "dumped",
1494 [CLD_TRAPPED] = "trapped",
1495 [CLD_STOPPED] = "stopped",
1496 [CLD_CONTINUED] = "continued",
1499 DEFINE_STRING_TABLE_LOOKUP(sigchld_code, int);
1501 static const char* const sched_policy_table[] = {
1502 [SCHED_OTHER] = "other",
1503 [SCHED_BATCH] = "batch",
1504 [SCHED_IDLE] = "idle",
1505 [SCHED_FIFO] = "fifo",
1509 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(sched_policy, int, INT_MAX);