1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 Copyright 2010 Lennart Poettering
14 #include <stdio_ext.h>
18 //#include <sys/mount.h>
19 #include <sys/personality.h>
20 #include <sys/prctl.h>
21 #include <sys/types.h>
25 #if HAVE_VALGRIND_VALGRIND_H
26 #include <valgrind/valgrind.h>
29 #include "alloc-util.h"
30 //#include "architecture.h"
39 #include "process-util.h"
40 #include "raw-clone.h"
41 #include "signal-util.h"
42 //#include "stat-util.h"
43 #include "string-table.h"
44 #include "string-util.h"
45 //#include "terminal-util.h"
46 #include "user-util.h"
49 int get_process_state(pid_t pid) {
53 _cleanup_free_ char *line = NULL;
57 p = procfs_file_alloca(pid, "stat");
59 r = read_one_line_file(p, &line);
65 p = strrchr(line, ')');
71 if (sscanf(p, " %c", &state) != 1)
74 return (unsigned char) state;
77 int get_process_comm(pid_t pid, char **ret) {
78 _cleanup_free_ char *escaped = NULL, *comm = NULL;
85 escaped = new(char, TASK_COMM_LEN);
89 p = procfs_file_alloca(pid, "comm");
91 r = read_one_line_file(p, &comm);
97 /* Escape unprintable characters, just in case, but don't grow the string beyond the underlying size */
98 cellescape(escaped, TASK_COMM_LEN, comm);
100 *ret = TAKE_PTR(escaped);
104 int get_process_cmdline(pid_t pid, size_t max_length, bool comm_fallback, char **line) {
105 _cleanup_fclose_ FILE *f = NULL;
107 char *k, *ans = NULL;
114 /* Retrieves a process' command line. Replaces unprintable characters while doing so by whitespace (coalescing
115 * multiple sequential ones into one). If max_length is != 0 will return a string of the specified size at most
116 * (the trailing NUL byte does count towards the length here!), abbreviated with a "..." ellipsis. If
117 * comm_fallback is true and the process has no command line set (the case for kernel threads), or has a
118 * command line that resolves to the empty string will return the "comm" name of the process instead.
120 * Returns -ESRCH if the process doesn't exist, and -ENOENT if the process has no command line (and
121 * comm_fallback is false). Returns 0 and sets *line otherwise. */
123 p = procfs_file_alloca(pid, "cmdline");
132 (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
134 if (max_length == 1) {
136 /* If there's only room for one byte, return the empty string */
144 } else if (max_length == 0) {
145 size_t len = 0, allocated = 0;
147 while ((c = getc(f)) != EOF) {
149 if (!GREEDY_REALLOC(ans, allocated, len+3)) {
171 bool dotdotdot = false;
174 ans = new(char, max_length);
180 while ((c = getc(f)) != EOF) {
207 if (max_length <= 4) {
211 k = ans + max_length - 4;
214 /* Eat up final spaces */
215 while (k > ans && isspace(k[-1])) {
221 strncpy(k, "...", left-1);
227 /* Kernel threads have no argv[] */
229 _cleanup_free_ char *t = NULL;
237 h = get_process_comm(pid, &t);
242 ans = strjoin("[", t, "]");
248 if (l + 3 <= max_length)
249 ans = strjoin("[", t, "]");
250 else if (max_length <= 6) {
252 ans = new(char, max_length);
256 memcpy(ans, "[...]", max_length-1);
257 ans[max_length-1] = 0;
259 t[max_length - 6] = 0;
261 /* Chop off final spaces */
262 delete_trailing_chars(t, WHITESPACE);
264 ans = strjoin("[", t, "...]");
275 int rename_process(const char name[]) {
276 static size_t mm_size = 0;
277 static char *mm = NULL;
278 bool truncated = false;
281 /* This is a like a poor man's setproctitle(). It changes the comm field, argv[0], and also the glibc's
282 * internally used name of the process. For the first one a limit of 16 chars applies; to the second one in
283 * many cases one of 10 (i.e. length of "/sbin/init") — however if we have CAP_SYS_RESOURCES it is unbounded;
284 * to the third one 7 (i.e. the length of "systemd". If you pass a longer string it will likely be
287 * Returns 0 if a name was set but truncated, > 0 if it was set but not truncated. */
290 return -EINVAL; /* let's not confuse users unnecessarily with an empty name */
292 if (!is_main_thread())
293 return -EPERM; /* Let's not allow setting the process name from other threads than the main one, as we
294 * cache things without locking, and we make assumptions that PR_SET_NAME sets the
295 * process name that isn't correct on any other threads */
299 /* First step, change the comm field. The main thread's comm is identical to the process comm. This means we
300 * can use PR_SET_NAME, which sets the thread name for the calling thread. */
301 if (prctl(PR_SET_NAME, name) < 0)
302 log_debug_errno(errno, "PR_SET_NAME failed: %m");
303 if (l >= TASK_COMM_LEN) /* Linux process names can be 15 chars at max */
306 /* Second step, change glibc's ID of the process name. */
307 if (program_invocation_name) {
310 k = strlen(program_invocation_name);
311 strncpy(program_invocation_name, name, k);
316 /* Third step, completely replace the argv[] array the kernel maintains for us. This requires privileges, but
317 * has the advantage that the argv[] array is exactly what we want it to be, and not filled up with zeros at
318 * the end. This is the best option for changing /proc/self/cmdline. */
320 /* Let's not bother with this if we don't have euid == 0. Strictly speaking we should check for the
321 * CAP_SYS_RESOURCE capability which is independent of the euid. In our own code the capability generally is
322 * present only for euid == 0, hence let's use this as quick bypass check, to avoid calling mmap() if
323 * PR_SET_MM_ARG_{START,END} fails with EPERM later on anyway. After all geteuid() is dead cheap to call, but
326 log_debug("Skipping PR_SET_MM, as we don't have privileges.");
327 else if (mm_size < l+1) {
331 nn_size = PAGE_ALIGN(l+1);
332 nn = mmap(NULL, nn_size, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
333 if (nn == MAP_FAILED) {
334 log_debug_errno(errno, "mmap() failed: %m");
338 strncpy(nn, name, nn_size);
340 /* Now, let's tell the kernel about this new memory */
341 if (prctl(PR_SET_MM, PR_SET_MM_ARG_START, (unsigned long) nn, 0, 0) < 0) {
342 log_debug_errno(errno, "PR_SET_MM_ARG_START failed, proceeding without: %m");
343 (void) munmap(nn, nn_size);
347 /* And update the end pointer to the new end, too. If this fails, we don't really know what to do, it's
348 * pretty unlikely that we can rollback, hence we'll just accept the failure, and continue. */
349 if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) nn + l + 1, 0, 0) < 0)
350 log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m");
353 (void) munmap(mm, mm_size);
358 strncpy(mm, name, mm_size);
360 /* Update the end pointer, continuing regardless of any failure. */
361 if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) mm + l + 1, 0, 0) < 0)
362 log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m");
366 /* Fourth step: in all cases we'll also update the original argv[], so that our own code gets it right too if
367 * it still looks here */
369 if (saved_argc > 0) {
375 k = strlen(saved_argv[0]);
376 strncpy(saved_argv[0], name, k);
381 for (i = 1; i < saved_argc; i++) {
385 memzero(saved_argv[i], strlen(saved_argv[i]));
392 int is_kernel_thread(pid_t pid) {
393 _cleanup_free_ char *line = NULL;
394 unsigned long long flags;
400 if (IN_SET(pid, 0, 1) || pid == getpid_cached()) /* pid 1, and we ourselves certainly aren't a kernel thread */
402 if (!pid_is_valid(pid))
405 p = procfs_file_alloca(pid, "stat");
406 r = read_one_line_file(p, &line);
412 /* Skip past the comm field */
413 q = strrchr(line, ')');
418 /* Skip 6 fields to reach the flags field */
419 for (i = 0; i < 6; i++) {
420 l = strspn(q, WHITESPACE);
425 l = strcspn(q, WHITESPACE);
431 /* Skip preceeding whitespace */
432 l = strspn(q, WHITESPACE);
437 /* Truncate the rest */
438 l = strcspn(q, WHITESPACE);
443 r = safe_atollu(q, &flags);
447 return !!(flags & PF_KTHREAD);
450 #if 0 /// UNNEEDED by elogind
451 int get_process_capeff(pid_t pid, char **capeff) {
458 p = procfs_file_alloca(pid, "status");
460 r = get_proc_field(p, "CapEff", WHITESPACE, capeff);
468 static int get_process_link_contents(const char *proc_file, char **name) {
474 r = readlink_malloc(proc_file, name);
483 int get_process_exe(pid_t pid, char **name) {
490 p = procfs_file_alloca(pid, "exe");
491 r = get_process_link_contents(p, name);
495 d = endswith(*name, " (deleted)");
502 #if 0 /// UNNEEDED by elogind
503 static int get_process_id(pid_t pid, const char *field, uid_t *uid) {
504 _cleanup_fclose_ FILE *f = NULL;
514 p = procfs_file_alloca(pid, "status");
522 (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
524 FOREACH_LINE(line, f, return -errno) {
529 if (startswith(l, field)) {
531 l += strspn(l, WHITESPACE);
533 l[strcspn(l, WHITESPACE)] = 0;
535 return parse_uid(l, uid);
542 int get_process_uid(pid_t pid, uid_t *uid) {
544 if (pid == 0 || pid == getpid_cached()) {
549 return get_process_id(pid, "Uid:", uid);
552 int get_process_gid(pid_t pid, gid_t *gid) {
554 if (pid == 0 || pid == getpid_cached()) {
559 assert_cc(sizeof(uid_t) == sizeof(gid_t));
560 return get_process_id(pid, "Gid:", gid);
563 int get_process_cwd(pid_t pid, char **cwd) {
568 p = procfs_file_alloca(pid, "cwd");
570 return get_process_link_contents(p, cwd);
573 int get_process_root(pid_t pid, char **root) {
578 p = procfs_file_alloca(pid, "root");
580 return get_process_link_contents(p, root);
583 int get_process_environ(pid_t pid, char **env) {
584 _cleanup_fclose_ FILE *f = NULL;
585 _cleanup_free_ char *outcome = NULL;
588 size_t allocated = 0, sz = 0;
593 p = procfs_file_alloca(pid, "environ");
602 (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
604 while ((c = fgetc(f)) != EOF) {
605 if (!GREEDY_REALLOC(outcome, allocated, sz + 5))
609 outcome[sz++] = '\n';
611 sz += cescape_char(c, outcome + sz);
615 outcome = strdup("");
621 *env = TAKE_PTR(outcome);
626 int get_process_ppid(pid_t pid, pid_t *_ppid) {
628 _cleanup_free_ char *line = NULL;
635 if (pid == 0 || pid == getpid_cached()) {
640 p = procfs_file_alloca(pid, "stat");
641 r = read_one_line_file(p, &line);
647 /* Let's skip the pid and comm fields. The latter is enclosed
648 * in () but does not escape any () in its value, so let's
649 * skip over it manually */
651 p = strrchr(line, ')');
663 if ((long unsigned) (pid_t) ppid != ppid)
666 *_ppid = (pid_t) ppid;
672 int wait_for_terminate(pid_t pid, siginfo_t *status) {
683 if (waitid(P_PID, pid, status, WEXITED) < 0) {
688 return negative_errno();
697 * < 0 : wait_for_terminate() failed to get the state of the
698 * process, the process was terminated by a signal, or
699 * failed for an unknown reason.
700 * >=0 : The process terminated normally, and its exit code is
703 * That is, success is indicated by a return value of zero, and an
704 * error is indicated by a non-zero value.
706 * A warning is emitted if the process terminates abnormally,
707 * and also if it returns non-zero unless check_exit_code is true.
709 int wait_for_terminate_and_check(const char *name, pid_t pid, WaitFlags flags) {
710 _cleanup_free_ char *buffer = NULL;
717 r = get_process_comm(pid, &buffer);
719 log_debug_errno(r, "Failed to acquire process name of " PID_FMT ", ignoring: %m", pid);
724 prio = flags & WAIT_LOG_ABNORMAL ? LOG_ERR : LOG_DEBUG;
726 r = wait_for_terminate(pid, &status);
728 return log_full_errno(prio, r, "Failed to wait for %s: %m", strna(name));
730 if (status.si_code == CLD_EXITED) {
731 if (status.si_status != EXIT_SUCCESS)
732 log_full(flags & WAIT_LOG_NON_ZERO_EXIT_STATUS ? LOG_ERR : LOG_DEBUG,
733 "%s failed with exit status %i.", strna(name), status.si_status);
735 log_debug("%s succeeded.", name);
737 return status.si_status;
739 } else if (IN_SET(status.si_code, CLD_KILLED, CLD_DUMPED)) {
741 log_full(prio, "%s terminated by signal %s.", strna(name), signal_to_string(status.si_status));
745 log_full(prio, "%s failed due to unknown reason.", strna(name));
752 * < 0 : wait_for_terminate_with_timeout() failed to get the state of the process, the process timed out, the process
753 * was terminated by a signal, or failed for an unknown reason.
755 * >=0 : The process terminated normally with no failures.
757 * Success is indicated by a return value of zero, a timeout is indicated by ETIMEDOUT, and all other child failure
758 * states are indicated by error is indicated by a non-zero value.
760 * This call assumes SIGCHLD has been blocked already, in particular before the child to wait for has been forked off
761 * to remain entirely race-free.
763 int wait_for_terminate_with_timeout(pid_t pid, usec_t timeout) {
768 assert_se(sigemptyset(&mask) == 0);
769 assert_se(sigaddset(&mask, SIGCHLD) == 0);
771 /* Drop into a sigtimewait-based timeout. Waiting for the
773 until = now(CLOCK_MONOTONIC) + timeout;
776 siginfo_t status = {};
779 n = now(CLOCK_MONOTONIC);
783 r = sigtimedwait(&mask, NULL, timespec_store(&ts, until - n)) < 0 ? -errno : 0;
784 /* Assuming we woke due to the child exiting. */
785 if (waitid(P_PID, pid, &status, WEXITED|WNOHANG) == 0) {
786 if (status.si_pid == pid) {
787 /* This is the correct child.*/
788 if (status.si_code == CLD_EXITED)
789 return (status.si_status == 0) ? 0 : -EPROTO;
794 /* Not the child, check for errors and proceed appropriately */
798 /* Timed out, child is likely hung. */
801 /* Received a different signal and should retry */
804 /* Return any unexpected errors */
813 #if 0 /// UNNEEDED by elogind
814 void sigkill_wait(pid_t pid) {
817 if (kill(pid, SIGKILL) > 0)
818 (void) wait_for_terminate(pid, NULL);
821 void sigkill_waitp(pid_t *pid) {
833 void sigterm_wait(pid_t pid) {
836 if (kill_and_sigcont(pid, SIGTERM) > 0)
837 (void) wait_for_terminate(pid, NULL);
840 int kill_and_sigcont(pid_t pid, int sig) {
843 r = kill(pid, sig) < 0 ? -errno : 0;
845 /* If this worked, also send SIGCONT, unless we already just sent a SIGCONT, or SIGKILL was sent which isn't
846 * affected by a process being suspended anyway. */
847 if (r >= 0 && !IN_SET(sig, SIGCONT, SIGKILL))
848 (void) kill(pid, SIGCONT);
853 int getenv_for_pid(pid_t pid, const char *field, char **ret) {
854 _cleanup_fclose_ FILE *f = NULL;
864 if (pid == 0 || pid == getpid_cached()) {
881 path = procfs_file_alloca(pid, "environ");
883 f = fopen(path, "re");
891 (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
899 for (i = 0; i < sizeof(line)-1; i++) {
903 if (_unlikely_(c == EOF)) {
913 if (strneq(line, field, l) && line[l] == '=') {
914 value = strdup(line + l + 1);
928 bool pid_is_unwaited(pid_t pid) {
929 /* Checks whether a PID is still valid at all, including a zombie */
934 if (pid <= 1) /* If we or PID 1 would be dead and have been waited for, this code would not be running */
937 if (pid == getpid_cached())
940 if (kill(pid, 0) >= 0)
943 return errno != ESRCH;
946 bool pid_is_alive(pid_t pid) {
949 /* Checks whether a PID is still valid and not a zombie */
954 if (pid <= 1) /* If we or PID 1 would be a zombie, this code would not be running */
957 if (pid == getpid_cached())
960 r = get_process_state(pid);
961 if (IN_SET(r, -ESRCH, 'Z'))
967 #if 0 /// UNNEEDED by elogind
968 int pid_from_same_root_fs(pid_t pid) {
974 if (pid == 0 || pid == getpid_cached())
977 root = procfs_file_alloca(pid, "root");
979 return files_same(root, "/proc/1/root", 0);
983 bool is_main_thread(void) {
984 static thread_local int cached = 0;
986 if (_unlikely_(cached == 0))
987 cached = getpid_cached() == gettid() ? 1 : -1;
992 #if 0 /// UNNEEDED by elogind
993 _noreturn_ void freeze(void) {
997 /* Make sure nobody waits for us on a socket anymore */
998 close_all_fds(NULL, 0);
1002 /* Let's not freeze right away, but keep reaping zombies. */
1007 r = waitid(P_ALL, 0, &si, WEXITED);
1008 if (r < 0 && errno != EINTR)
1012 /* waitid() failed with an unexpected error, things are really borked. Freeze now! */
1017 bool oom_score_adjust_is_valid(int oa) {
1018 return oa >= OOM_SCORE_ADJ_MIN && oa <= OOM_SCORE_ADJ_MAX;
1021 unsigned long personality_from_string(const char *p) {
1025 return PERSONALITY_INVALID;
1027 /* Parse a personality specifier. We use our own identifiers that indicate specific ABIs, rather than just
1028 * hints regarding the register size, since we want to keep things open for multiple locally supported ABIs for
1029 * the same register size. */
1031 architecture = architecture_from_string(p);
1032 if (architecture < 0)
1033 return PERSONALITY_INVALID;
1035 if (architecture == native_architecture())
1037 #ifdef SECONDARY_ARCHITECTURE
1038 if (architecture == SECONDARY_ARCHITECTURE)
1042 return PERSONALITY_INVALID;
1045 const char* personality_to_string(unsigned long p) {
1046 int architecture = _ARCHITECTURE_INVALID;
1049 architecture = native_architecture();
1050 #ifdef SECONDARY_ARCHITECTURE
1051 else if (p == PER_LINUX32)
1052 architecture = SECONDARY_ARCHITECTURE;
1055 if (architecture < 0)
1058 return architecture_to_string(architecture);
1061 int safe_personality(unsigned long p) {
1064 /* So here's the deal, personality() is weirdly defined by glibc. In some cases it returns a failure via errno,
1065 * and in others as negative return value containing an errno-like value. Let's work around this: this is a
1066 * wrapper that uses errno if it is set, and uses the return value otherwise. And then it sets both errno and
1067 * the return value indicating the same issue, so that we are definitely on the safe side.
1069 * See https://github.com/systemd/systemd/issues/6737 */
1072 ret = personality(p);
1083 int opinionated_personality(unsigned long *ret) {
1086 /* Returns the current personality, or PERSONALITY_INVALID if we can't determine it. This function is a bit
1087 * opinionated though, and ignores all the finer-grained bits and exotic personalities, only distinguishing the
1088 * two most relevant personalities: PER_LINUX and PER_LINUX32. */
1090 current = safe_personality(PERSONALITY_INVALID);
1094 if (((unsigned long) current & 0xffff) == PER_LINUX32)
1102 void valgrind_summary_hack(void) {
1103 #if HAVE_VALGRIND_VALGRIND_H
1104 if (getpid_cached() == 1 && RUNNING_ON_VALGRIND) {
1106 pid = raw_clone(SIGCHLD);
1108 log_emergency_errno(errno, "Failed to fork off valgrind helper: %m");
1112 log_info("Spawned valgrind helper as PID "PID_FMT".", pid);
1113 (void) wait_for_terminate(pid, NULL);
1119 int pid_compare_func(const void *a, const void *b) {
1120 const pid_t *p = a, *q = b;
1122 /* Suitable for usage in qsort() */
1131 int ioprio_parse_priority(const char *s, int *ret) {
1137 r = safe_atoi(s, &i);
1141 if (!ioprio_priority_is_valid(i))
1149 /* The cached PID, possible values:
1151 * == UNSET [0] → cache not initialized yet
1152 * == BUSY [-1] → some thread is initializing it at the moment
1153 * any other → the cached PID
1156 #define CACHED_PID_UNSET ((pid_t) 0)
1157 #define CACHED_PID_BUSY ((pid_t) -1)
1159 static pid_t cached_pid = CACHED_PID_UNSET;
1161 void reset_cached_pid(void) {
1162 /* Invoked in the child after a fork(), i.e. at the first moment the PID changed */
1163 cached_pid = CACHED_PID_UNSET;
1166 /* We use glibc __register_atfork() + __dso_handle directly here, as they are not included in the glibc
1167 * headers. __register_atfork() is mostly equivalent to pthread_atfork(), but doesn't require us to link against
1168 * libpthread, as it is part of glibc anyway. */
1170 extern int __register_atfork(void (*prepare) (void), void (*parent) (void), void (*child) (void), void * __dso_handle);
1171 extern void* __dso_handle __attribute__ ((__weak__));
1172 #endif // ifdef __GLIBC__
1174 pid_t getpid_cached(void) {
1175 static bool installed = false;
1176 pid_t current_value;
1178 /* getpid_cached() is much like getpid(), but caches the value in local memory, to avoid having to invoke a
1179 * system call each time. This restores glibc behaviour from before 2.24, when getpid() was unconditionally
1180 * cached. Starting with 2.24 getpid() started to become prohibitively expensive when used for detecting when
1181 * objects were used across fork()s. With this caching the old behaviour is somewhat restored.
1183 * https://bugzilla.redhat.com/show_bug.cgi?id=1443976
1184 * https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=c579f48edba88380635ab98cb612030e3ed8691e
1187 current_value = __sync_val_compare_and_swap(&cached_pid, CACHED_PID_UNSET, CACHED_PID_BUSY);
1189 switch (current_value) {
1191 case CACHED_PID_UNSET: { /* Not initialized yet, then do so now */
1194 new_pid = raw_getpid();
1197 /* __register_atfork() either returns 0 or -ENOMEM, in its glibc implementation. Since it's
1198 * only half-documented (glibc doesn't document it but LSB does — though only superficially)
1199 * we'll check for errors only in the most generic fashion possible. */
1201 if (__register_atfork(NULL, NULL, reset_cached_pid, __dso_handle) != 0) {
1202 /* OOM? Let's try again later */
1203 cached_pid = CACHED_PID_UNSET;
1210 cached_pid = new_pid;
1214 case CACHED_PID_BUSY: /* Somebody else is currently initializing */
1215 return raw_getpid();
1217 default: /* Properly initialized */
1218 return current_value;
1222 int must_be_root(void) {
1227 log_error("Need to be root.");
1233 const int except_fds[],
1234 size_t n_except_fds,
1238 pid_t original_pid, pid;
1239 sigset_t saved_ss, ss;
1240 bool block_signals = false;
1243 /* A wrapper around fork(), that does a couple of important initializations in addition to mere forking. Always
1244 * returns the child's PID in *ret_pid. Returns == 0 in the child, and > 0 in the parent. */
1246 prio = flags & FORK_LOG ? LOG_ERR : LOG_DEBUG;
1248 original_pid = getpid_cached();
1250 if (flags & (FORK_RESET_SIGNALS|FORK_DEATHSIG)) {
1252 /* We temporarily block all signals, so that the new child has them blocked initially. This way, we can
1253 * be sure that SIGTERMs are not lost we might send to the child. */
1255 if (sigfillset(&ss) < 0)
1256 return log_full_errno(prio, errno, "Failed to reset signal set: %m");
1258 block_signals = true;
1260 } else if (flags & FORK_WAIT) {
1262 /* Let's block SIGCHLD at least, so that we can safely watch for the child process */
1264 if (sigemptyset(&ss) < 0)
1265 return log_full_errno(prio, errno, "Failed to clear signal set: %m");
1267 if (sigaddset(&ss, SIGCHLD) < 0)
1268 return log_full_errno(prio, errno, "Failed to add SIGCHLD to signal set: %m");
1270 block_signals = true;
1274 if (sigprocmask(SIG_SETMASK, &ss, &saved_ss) < 0)
1275 return log_full_errno(prio, errno, "Failed to set signal mask: %m");
1277 if (flags & FORK_NEW_MOUNTNS)
1278 pid = raw_clone(SIGCHLD|CLONE_NEWNS);
1284 if (block_signals) /* undo what we did above */
1285 (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL);
1287 return log_full_errno(prio, r, "Failed to fork: %m");
1290 /* We are in the parent process */
1292 log_debug("Successfully forked off '%s' as PID " PID_FMT ".", strna(name), pid);
1294 if (flags & FORK_WAIT) {
1295 r = wait_for_terminate_and_check(name, pid, (flags & FORK_LOG ? WAIT_LOG : 0));
1298 if (r != EXIT_SUCCESS) /* exit status > 0 should be treated as failure, too */
1302 if (block_signals) /* undo what we did above */
1303 (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL);
1311 /* We are in the child process */
1313 if (flags & FORK_REOPEN_LOG) {
1314 /* Close the logs if requested, before we log anything. And make sure we reopen it if needed. */
1316 log_set_open_when_needed(true);
1320 r = rename_process(name);
1322 log_full_errno(flags & FORK_LOG ? LOG_WARNING : LOG_DEBUG,
1323 r, "Failed to rename process, ignoring: %m");
1326 if (flags & FORK_DEATHSIG)
1327 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0) {
1328 log_full_errno(prio, errno, "Failed to set death signal: %m");
1329 _exit(EXIT_FAILURE);
1332 if (flags & FORK_RESET_SIGNALS) {
1333 r = reset_all_signal_handlers();
1335 log_full_errno(prio, r, "Failed to reset signal handlers: %m");
1336 _exit(EXIT_FAILURE);
1339 /* This implicitly undoes the signal mask stuff we did before the fork()ing above */
1340 r = reset_signal_mask();
1342 log_full_errno(prio, r, "Failed to reset signal mask: %m");
1343 _exit(EXIT_FAILURE);
1345 } else if (block_signals) { /* undo what we did above */
1346 if (sigprocmask(SIG_SETMASK, &saved_ss, NULL) < 0) {
1347 log_full_errno(prio, errno, "Failed to restore signal mask: %m");
1348 _exit(EXIT_FAILURE);
1352 if (flags & FORK_DEATHSIG) {
1354 /* Let's see if the parent PID is still the one we started from? If not, then the parent
1355 * already died by the time we set PR_SET_PDEATHSIG, hence let's emulate the effect */
1359 /* Parent is in a differn't PID namespace. */;
1360 else if (ppid != original_pid) {
1361 log_debug("Parent died early, raising SIGTERM.");
1362 (void) raise(SIGTERM);
1363 _exit(EXIT_FAILURE);
1367 if (FLAGS_SET(flags, FORK_NEW_MOUNTNS | FORK_MOUNTNS_SLAVE)) {
1369 /* Optionally, make sure we never propagate mounts to the host. */
1371 if (mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL) < 0) {
1372 log_full_errno(prio, errno, "Failed to remount root directory as MS_SLAVE: %m");
1373 _exit(EXIT_FAILURE);
1377 if (flags & FORK_CLOSE_ALL_FDS) {
1378 /* Close the logs here in case it got reopened above, as close_all_fds() would close them for us */
1381 r = close_all_fds(except_fds, n_except_fds);
1383 log_full_errno(prio, r, "Failed to close all file descriptors: %m");
1384 _exit(EXIT_FAILURE);
1388 /* When we were asked to reopen the logs, do so again now */
1389 if (flags & FORK_REOPEN_LOG) {
1391 log_set_open_when_needed(false);
1394 if (flags & FORK_NULL_STDIO) {
1395 r = make_null_stdio();
1397 log_full_errno(prio, r, "Failed to connect stdin/stdout to /dev/null: %m");
1398 _exit(EXIT_FAILURE);
1403 *ret_pid = getpid_cached();
1408 int fork_agent(const char *name, const int except[], size_t n_except, pid_t *ret_pid, const char *path, ...) {
1409 bool stdout_is_tty, stderr_is_tty;
1417 /* Spawns a temporary TTY agent, making sure it goes away when we go away */
1419 r = safe_fork_full(name, except, n_except, FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_CLOSE_ALL_FDS, ret_pid);
1427 stdout_is_tty = isatty(STDOUT_FILENO);
1428 stderr_is_tty = isatty(STDERR_FILENO);
1430 if (!stdout_is_tty || !stderr_is_tty) {
1433 /* Detach from stdout/stderr. and reopen
1434 * /dev/tty for them. This is important to
1435 * ensure that when systemctl is started via
1436 * popen() or a similar call that expects to
1437 * read EOF we actually do generate EOF and
1438 * not delay this indefinitely by because we
1439 * keep an unused copy of stdin around. */
1440 fd = open("/dev/tty", O_WRONLY);
1442 log_error_errno(errno, "Failed to open /dev/tty: %m");
1443 _exit(EXIT_FAILURE);
1446 if (!stdout_is_tty && dup2(fd, STDOUT_FILENO) < 0) {
1447 log_error_errno(errno, "Failed to dup2 /dev/tty: %m");
1448 _exit(EXIT_FAILURE);
1451 if (!stderr_is_tty && dup2(fd, STDERR_FILENO) < 0) {
1452 log_error_errno(errno, "Failed to dup2 /dev/tty: %m");
1453 _exit(EXIT_FAILURE);
1456 safe_close_above_stdio(fd);
1459 /* Count arguments */
1461 for (n = 0; va_arg(ap, char*); n++)
1466 l = newa(char*, n + 1);
1468 /* Fill in arguments */
1470 for (i = 0; i <= n; i++)
1471 l[i] = va_arg(ap, char*);
1475 _exit(EXIT_FAILURE);
1478 int set_oom_score_adjust(int value) {
1479 char t[DECIMAL_STR_MAX(int)];
1481 sprintf(t, "%i", value);
1483 return write_string_file("/proc/self/oom_score_adj", t,
1484 WRITE_STRING_FILE_VERIFY_ON_FAILURE|WRITE_STRING_FILE_DISABLE_BUFFER);
1487 #if 0 /// UNNEEDED by elogind
1488 static const char *const ioprio_class_table[] = {
1489 [IOPRIO_CLASS_NONE] = "none",
1490 [IOPRIO_CLASS_RT] = "realtime",
1491 [IOPRIO_CLASS_BE] = "best-effort",
1492 [IOPRIO_CLASS_IDLE] = "idle"
1495 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ioprio_class, int, IOPRIO_N_CLASSES);
1497 static const char *const sigchld_code_table[] = {
1498 [CLD_EXITED] = "exited",
1499 [CLD_KILLED] = "killed",
1500 [CLD_DUMPED] = "dumped",
1501 [CLD_TRAPPED] = "trapped",
1502 [CLD_STOPPED] = "stopped",
1503 [CLD_CONTINUED] = "continued",
1506 DEFINE_STRING_TABLE_LOOKUP(sigchld_code, int);
1508 static const char* const sched_policy_table[] = {
1509 [SCHED_OTHER] = "other",
1510 [SCHED_BATCH] = "batch",
1511 [SCHED_IDLE] = "idle",
1512 [SCHED_FIFO] = "fifo",
1516 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(sched_policy, int, INT_MAX);