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
) {
407 if (IN_SET(pid
, 0, 1) || pid
== getpid_cached()) /* pid 1, and we ourselves certainly aren't a kernel thread */
412 p
= procfs_file_alloca(pid
, "cmdline");
420 (void) __fsetlocking(f
, FSETLOCKING_BYCALLER
);
422 count
= fread(&c
, 1, 1, f
);
426 /* Kernel threads have an empty cmdline */
429 return eof
? 1 : -errno
;
434 int get_process_capeff(pid_t pid
, char **capeff
) {
441 p
= procfs_file_alloca(pid
, "status");
443 r
= get_proc_field(p
, "CapEff", WHITESPACE
, capeff
);
450 static int get_process_link_contents(const char *proc_file
, char **name
) {
456 r
= readlink_malloc(proc_file
, name
);
465 int get_process_exe(pid_t pid
, char **name
) {
472 p
= procfs_file_alloca(pid
, "exe");
473 r
= get_process_link_contents(p
, name
);
477 d
= endswith(*name
, " (deleted)");
484 static int get_process_id(pid_t pid
, const char *field
, uid_t
*uid
) {
485 _cleanup_fclose_
FILE *f
= NULL
;
495 p
= procfs_file_alloca(pid
, "status");
503 (void) __fsetlocking(f
, FSETLOCKING_BYCALLER
);
505 FOREACH_LINE(line
, f
, return -errno
) {
510 if (startswith(l
, field
)) {
512 l
+= strspn(l
, WHITESPACE
);
514 l
[strcspn(l
, WHITESPACE
)] = 0;
516 return parse_uid(l
, uid
);
523 int get_process_uid(pid_t pid
, uid_t
*uid
) {
525 if (pid
== 0 || pid
== getpid_cached()) {
530 return get_process_id(pid
, "Uid:", uid
);
533 int get_process_gid(pid_t pid
, gid_t
*gid
) {
535 if (pid
== 0 || pid
== getpid_cached()) {
540 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
541 return get_process_id(pid
, "Gid:", gid
);
544 int get_process_cwd(pid_t pid
, char **cwd
) {
549 p
= procfs_file_alloca(pid
, "cwd");
551 return get_process_link_contents(p
, cwd
);
554 int get_process_root(pid_t pid
, char **root
) {
559 p
= procfs_file_alloca(pid
, "root");
561 return get_process_link_contents(p
, root
);
564 int get_process_environ(pid_t pid
, char **env
) {
565 _cleanup_fclose_
FILE *f
= NULL
;
566 _cleanup_free_
char *outcome
= NULL
;
569 size_t allocated
= 0, sz
= 0;
574 p
= procfs_file_alloca(pid
, "environ");
583 (void) __fsetlocking(f
, FSETLOCKING_BYCALLER
);
585 while ((c
= fgetc(f
)) != EOF
) {
586 if (!GREEDY_REALLOC(outcome
, allocated
, sz
+ 5))
590 outcome
[sz
++] = '\n';
592 sz
+= cescape_char(c
, outcome
+ sz
);
596 outcome
= strdup("");
608 int get_process_ppid(pid_t pid
, pid_t
*_ppid
) {
610 _cleanup_free_
char *line
= NULL
;
617 if (pid
== 0 || pid
== getpid_cached()) {
622 p
= procfs_file_alloca(pid
, "stat");
623 r
= read_one_line_file(p
, &line
);
629 /* Let's skip the pid and comm fields. The latter is enclosed
630 * in () but does not escape any () in its value, so let's
631 * skip over it manually */
633 p
= strrchr(line
, ')');
645 if ((long unsigned) (pid_t
) ppid
!= ppid
)
648 *_ppid
= (pid_t
) ppid
;
653 int wait_for_terminate(pid_t pid
, siginfo_t
*status
) {
664 if (waitid(P_PID
, pid
, status
, WEXITED
) < 0) {
669 return negative_errno();
678 * < 0 : wait_for_terminate() failed to get the state of the
679 * process, the process was terminated by a signal, or
680 * failed for an unknown reason.
681 * >=0 : The process terminated normally, and its exit code is
684 * That is, success is indicated by a return value of zero, and an
685 * error is indicated by a non-zero value.
687 * A warning is emitted if the process terminates abnormally,
688 * and also if it returns non-zero unless check_exit_code is true.
690 int wait_for_terminate_and_check(const char *name
, pid_t pid
, WaitFlags flags
) {
691 _cleanup_free_
char *buffer
= NULL
;
698 r
= get_process_comm(pid
, &buffer
);
700 log_debug_errno(r
, "Failed to acquire process name of " PID_FMT
", ignoring: %m", pid
);
705 prio
= flags
& WAIT_LOG_ABNORMAL
? LOG_ERR
: LOG_DEBUG
;
707 r
= wait_for_terminate(pid
, &status
);
709 return log_full_errno(prio
, r
, "Failed to wait for %s: %m", strna(name
));
711 if (status
.si_code
== CLD_EXITED
) {
712 if (status
.si_status
!= EXIT_SUCCESS
)
713 log_full(flags
& WAIT_LOG_NON_ZERO_EXIT_STATUS
? LOG_ERR
: LOG_DEBUG
,
714 "%s failed with exit status %i.", strna(name
), status
.si_status
);
716 log_debug("%s succeeded.", name
);
718 return status
.si_status
;
720 } else if (IN_SET(status
.si_code
, CLD_KILLED
, CLD_DUMPED
)) {
722 log_full(prio
, "%s terminated by signal %s.", strna(name
), signal_to_string(status
.si_status
));
726 log_full(prio
, "%s failed due to unknown reason.", strna(name
));
732 * < 0 : wait_for_terminate_with_timeout() failed to get the state of the
733 * process, the process timed out, the process was terminated by a
734 * signal, or failed for an unknown reason.
735 * >=0 : The process terminated normally with no failures.
737 * Success is indicated by a return value of zero, a timeout is indicated
738 * by ETIMEDOUT, and all other child failure states are indicated by error
739 * is indicated by a non-zero value.
741 int wait_for_terminate_with_timeout(pid_t pid
, usec_t timeout
) {
746 assert_se(sigemptyset(&mask
) == 0);
747 assert_se(sigaddset(&mask
, SIGCHLD
) == 0);
749 /* Drop into a sigtimewait-based timeout. Waiting for the
751 until
= now(CLOCK_MONOTONIC
) + timeout
;
754 siginfo_t status
= {};
757 n
= now(CLOCK_MONOTONIC
);
761 r
= sigtimedwait(&mask
, NULL
, timespec_store(&ts
, until
- n
)) < 0 ? -errno
: 0;
762 /* Assuming we woke due to the child exiting. */
763 if (waitid(P_PID
, pid
, &status
, WEXITED
|WNOHANG
) == 0) {
764 if (status
.si_pid
== pid
) {
765 /* This is the correct child.*/
766 if (status
.si_code
== CLD_EXITED
)
767 return (status
.si_status
== 0) ? 0 : -EPROTO
;
772 /* Not the child, check for errors and proceed appropriately */
776 /* Timed out, child is likely hung. */
779 /* Received a different signal and should retry */
782 /* Return any unexpected errors */
791 void sigkill_wait(pid_t pid
) {
794 if (kill(pid
, SIGKILL
) > 0)
795 (void) wait_for_terminate(pid
, NULL
);
798 void sigkill_waitp(pid_t
*pid
) {
809 int kill_and_sigcont(pid_t pid
, int sig
) {
812 r
= kill(pid
, sig
) < 0 ? -errno
: 0;
814 /* If this worked, also send SIGCONT, unless we already just sent a SIGCONT, or SIGKILL was sent which isn't
815 * affected by a process being suspended anyway. */
816 if (r
>= 0 && !IN_SET(sig
, SIGCONT
, SIGKILL
))
817 (void) kill(pid
, SIGCONT
);
822 int getenv_for_pid(pid_t pid
, const char *field
, char **_value
) {
823 _cleanup_fclose_
FILE *f
= NULL
;
834 path
= procfs_file_alloca(pid
, "environ");
836 f
= fopen(path
, "re");
843 (void) __fsetlocking(f
, FSETLOCKING_BYCALLER
);
852 for (i
= 0; i
< sizeof(line
)-1; i
++) {
856 if (_unlikely_(c
== EOF
)) {
866 if (strneq(line
, field
, l
) && line
[l
] == '=') {
867 value
= strdup(line
+ l
+ 1);
881 bool pid_is_unwaited(pid_t pid
) {
882 /* Checks whether a PID is still valid at all, including a zombie */
887 if (pid
<= 1) /* If we or PID 1 would be dead and have been waited for, this code would not be running */
890 if (pid
== getpid_cached())
893 if (kill(pid
, 0) >= 0)
896 return errno
!= ESRCH
;
899 bool pid_is_alive(pid_t pid
) {
902 /* Checks whether a PID is still valid and not a zombie */
907 if (pid
<= 1) /* If we or PID 1 would be a zombie, this code would not be running */
910 if (pid
== getpid_cached())
913 r
= get_process_state(pid
);
914 if (IN_SET(r
, -ESRCH
, 'Z'))
920 int pid_from_same_root_fs(pid_t pid
) {
926 if (pid
== 0 || pid
== getpid_cached())
929 root
= procfs_file_alloca(pid
, "root");
931 return files_same(root
, "/proc/1/root", 0);
934 bool is_main_thread(void) {
935 static thread_local
int cached
= 0;
937 if (_unlikely_(cached
== 0))
938 cached
= getpid_cached() == gettid() ? 1 : -1;
943 noreturn
void freeze(void) {
947 /* Make sure nobody waits for us on a socket anymore */
948 close_all_fds(NULL
, 0);
952 /* Let's not freeze right away, but keep reaping zombies. */
957 r
= waitid(P_ALL
, 0, &si
, WEXITED
);
958 if (r
< 0 && errno
!= EINTR
)
962 /* waitid() failed with an unexpected error, things are really borked. Freeze now! */
967 bool oom_score_adjust_is_valid(int oa
) {
968 return oa
>= OOM_SCORE_ADJ_MIN
&& oa
<= OOM_SCORE_ADJ_MAX
;
971 unsigned long personality_from_string(const char *p
) {
975 return PERSONALITY_INVALID
;
977 /* Parse a personality specifier. We use our own identifiers that indicate specific ABIs, rather than just
978 * hints regarding the register size, since we want to keep things open for multiple locally supported ABIs for
979 * the same register size. */
981 architecture
= architecture_from_string(p
);
982 if (architecture
< 0)
983 return PERSONALITY_INVALID
;
985 if (architecture
== native_architecture())
987 #ifdef SECONDARY_ARCHITECTURE
988 if (architecture
== SECONDARY_ARCHITECTURE
)
992 return PERSONALITY_INVALID
;
995 const char* personality_to_string(unsigned long p
) {
996 int architecture
= _ARCHITECTURE_INVALID
;
999 architecture
= native_architecture();
1000 #ifdef SECONDARY_ARCHITECTURE
1001 else if (p
== PER_LINUX32
)
1002 architecture
= SECONDARY_ARCHITECTURE
;
1005 if (architecture
< 0)
1008 return architecture_to_string(architecture
);
1011 int safe_personality(unsigned long p
) {
1014 /* So here's the deal, personality() is weirdly defined by glibc. In some cases it returns a failure via errno,
1015 * and in others as negative return value containing an errno-like value. Let's work around this: this is a
1016 * wrapper that uses errno if it is set, and uses the return value otherwise. And then it sets both errno and
1017 * the return value indicating the same issue, so that we are definitely on the safe side.
1019 * See https://github.com/systemd/systemd/issues/6737 */
1022 ret
= personality(p
);
1033 int opinionated_personality(unsigned long *ret
) {
1036 /* Returns the current personality, or PERSONALITY_INVALID if we can't determine it. This function is a bit
1037 * opinionated though, and ignores all the finer-grained bits and exotic personalities, only distinguishing the
1038 * two most relevant personalities: PER_LINUX and PER_LINUX32. */
1040 current
= safe_personality(PERSONALITY_INVALID
);
1044 if (((unsigned long) current
& 0xffff) == PER_LINUX32
)
1052 void valgrind_summary_hack(void) {
1053 #if HAVE_VALGRIND_VALGRIND_H
1054 if (getpid_cached() == 1 && RUNNING_ON_VALGRIND
) {
1056 pid
= raw_clone(SIGCHLD
);
1058 log_emergency_errno(errno
, "Failed to fork off valgrind helper: %m");
1062 log_info("Spawned valgrind helper as PID "PID_FMT
".", pid
);
1063 (void) wait_for_terminate(pid
, NULL
);
1069 int pid_compare_func(const void *a
, const void *b
) {
1070 const pid_t
*p
= a
, *q
= b
;
1072 /* Suitable for usage in qsort() */
1081 int ioprio_parse_priority(const char *s
, int *ret
) {
1087 r
= safe_atoi(s
, &i
);
1091 if (!ioprio_priority_is_valid(i
))
1098 /* The cached PID, possible values:
1100 * == UNSET [0] → cache not initialized yet
1101 * == BUSY [-1] → some thread is initializing it at the moment
1102 * any other → the cached PID
1105 #define CACHED_PID_UNSET ((pid_t) 0)
1106 #define CACHED_PID_BUSY ((pid_t) -1)
1108 static pid_t cached_pid
= CACHED_PID_UNSET
;
1110 void reset_cached_pid(void) {
1111 /* Invoked in the child after a fork(), i.e. at the first moment the PID changed */
1112 cached_pid
= CACHED_PID_UNSET
;
1115 /* We use glibc __register_atfork() + __dso_handle directly here, as they are not included in the glibc
1116 * headers. __register_atfork() is mostly equivalent to pthread_atfork(), but doesn't require us to link against
1117 * libpthread, as it is part of glibc anyway. */
1118 extern int __register_atfork(void (*prepare
) (void), void (*parent
) (void), void (*child
) (void), void * __dso_handle
);
1119 extern void* __dso_handle
__attribute__ ((__weak__
));
1121 pid_t
getpid_cached(void) {
1122 pid_t current_value
;
1124 /* getpid_cached() is much like getpid(), but caches the value in local memory, to avoid having to invoke a
1125 * system call each time. This restores glibc behaviour from before 2.24, when getpid() was unconditionally
1126 * cached. Starting with 2.24 getpid() started to become prohibitively expensive when used for detecting when
1127 * objects were used across fork()s. With this caching the old behaviour is somewhat restored.
1129 * https://bugzilla.redhat.com/show_bug.cgi?id=1443976
1130 * https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=c579f48edba88380635ab98cb612030e3ed8691e
1133 current_value
= __sync_val_compare_and_swap(&cached_pid
, CACHED_PID_UNSET
, CACHED_PID_BUSY
);
1135 switch (current_value
) {
1137 case CACHED_PID_UNSET
: { /* Not initialized yet, then do so now */
1140 new_pid
= raw_getpid();
1142 if (__register_atfork(NULL
, NULL
, reset_cached_pid
, __dso_handle
) != 0) {
1143 /* OOM? Let's try again later */
1144 cached_pid
= CACHED_PID_UNSET
;
1148 cached_pid
= new_pid
;
1152 case CACHED_PID_BUSY
: /* Somebody else is currently initializing */
1153 return raw_getpid();
1155 default: /* Properly initialized */
1156 return current_value
;
1160 int must_be_root(void) {
1165 log_error("Need to be root.");
1171 const int except_fds
[],
1172 size_t n_except_fds
,
1176 pid_t original_pid
, pid
;
1177 sigset_t saved_ss
, ss
;
1178 bool block_signals
= false;
1181 /* A wrapper around fork(), that does a couple of important initializations in addition to mere forking. Always
1182 * returns the child's PID in *ret_pid. Returns == 0 in the child, and > 0 in the parent. */
1184 prio
= flags
& FORK_LOG
? LOG_ERR
: LOG_DEBUG
;
1186 original_pid
= getpid_cached();
1188 if (flags
& (FORK_RESET_SIGNALS
|FORK_DEATHSIG
)) {
1190 /* We temporarily block all signals, so that the new child has them blocked initially. This way, we can
1191 * be sure that SIGTERMs are not lost we might send to the child. */
1193 if (sigfillset(&ss
) < 0)
1194 return log_full_errno(prio
, errno
, "Failed to reset signal set: %m");
1196 block_signals
= true;
1198 } else if (flags
& FORK_WAIT
) {
1200 /* Let's block SIGCHLD at least, so that we can safely watch for the child process */
1202 if (sigemptyset(&ss
) < 0)
1203 return log_full_errno(prio
, errno
, "Failed to clear signal set: %m");
1205 if (sigaddset(&ss
, SIGCHLD
) < 0)
1206 return log_full_errno(prio
, errno
, "Failed to add SIGCHLD to signal set: %m");
1208 block_signals
= true;
1212 if (sigprocmask(SIG_SETMASK
, &ss
, &saved_ss
) < 0)
1213 return log_full_errno(prio
, errno
, "Failed to set signal mask: %m");
1215 if (flags
& FORK_NEW_MOUNTNS
)
1216 pid
= raw_clone(SIGCHLD
|CLONE_NEWNS
);
1222 if (block_signals
) /* undo what we did above */
1223 (void) sigprocmask(SIG_SETMASK
, &saved_ss
, NULL
);
1225 return log_full_errno(prio
, r
, "Failed to fork: %m");
1228 /* We are in the parent process */
1230 log_debug("Successfully forked off '%s' as PID " PID_FMT
".", strna(name
), pid
);
1232 if (flags
& FORK_WAIT
) {
1233 r
= wait_for_terminate_and_check(name
, pid
, (flags
& FORK_LOG
? WAIT_LOG
: 0));
1236 if (r
!= EXIT_SUCCESS
) /* exit status > 0 should be treated as failure, too */
1240 if (block_signals
) /* undo what we did above */
1241 (void) sigprocmask(SIG_SETMASK
, &saved_ss
, NULL
);
1249 /* We are in the child process */
1251 if (flags
& FORK_REOPEN_LOG
) {
1252 /* Close the logs if requested, before we log anything. And make sure we reopen it if needed. */
1254 log_set_open_when_needed(true);
1258 r
= rename_process(name
);
1260 log_full_errno(flags
& FORK_LOG
? LOG_WARNING
: LOG_DEBUG
,
1261 r
, "Failed to rename process, ignoring: %m");
1264 if (flags
& FORK_DEATHSIG
)
1265 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0) {
1266 log_full_errno(prio
, errno
, "Failed to set death signal: %m");
1267 _exit(EXIT_FAILURE
);
1270 if (flags
& FORK_RESET_SIGNALS
) {
1271 r
= reset_all_signal_handlers();
1273 log_full_errno(prio
, r
, "Failed to reset signal handlers: %m");
1274 _exit(EXIT_FAILURE
);
1277 /* This implicitly undoes the signal mask stuff we did before the fork()ing above */
1278 r
= reset_signal_mask();
1280 log_full_errno(prio
, r
, "Failed to reset signal mask: %m");
1281 _exit(EXIT_FAILURE
);
1283 } else if (block_signals
) { /* undo what we did above */
1284 if (sigprocmask(SIG_SETMASK
, &saved_ss
, NULL
) < 0) {
1285 log_full_errno(prio
, errno
, "Failed to restore signal mask: %m");
1286 _exit(EXIT_FAILURE
);
1290 if (flags
& FORK_DEATHSIG
) {
1292 /* Let's see if the parent PID is still the one we started from? If not, then the parent
1293 * already died by the time we set PR_SET_PDEATHSIG, hence let's emulate the effect */
1297 /* Parent is in a differn't PID namespace. */;
1298 else if (ppid
!= original_pid
) {
1299 log_debug("Parent died early, raising SIGTERM.");
1300 (void) raise(SIGTERM
);
1301 _exit(EXIT_FAILURE
);
1305 if (flags
& FORK_CLOSE_ALL_FDS
) {
1306 /* Close the logs here in case it got reopened above, as close_all_fds() would close them for us */
1309 r
= close_all_fds(except_fds
, n_except_fds
);
1311 log_full_errno(prio
, r
, "Failed to close all file descriptors: %m");
1312 _exit(EXIT_FAILURE
);
1316 /* When we were asked to reopen the logs, do so again now */
1317 if (flags
& FORK_REOPEN_LOG
) {
1319 log_set_open_when_needed(false);
1322 if (flags
& FORK_NULL_STDIO
) {
1323 r
= make_null_stdio();
1325 log_full_errno(prio
, r
, "Failed to connect stdin/stdout to /dev/null: %m");
1326 _exit(EXIT_FAILURE
);
1331 *ret_pid
= getpid_cached();
1336 int fork_agent(const char *name
, const int except
[], unsigned n_except
, pid_t
*ret_pid
, const char *path
, ...) {
1337 bool stdout_is_tty
, stderr_is_tty
;
1345 /* Spawns a temporary TTY agent, making sure it goes away when we go away */
1347 r
= safe_fork_full(name
, except
, n_except
, FORK_RESET_SIGNALS
|FORK_DEATHSIG
|FORK_CLOSE_ALL_FDS
, ret_pid
);
1355 stdout_is_tty
= isatty(STDOUT_FILENO
);
1356 stderr_is_tty
= isatty(STDERR_FILENO
);
1358 if (!stdout_is_tty
|| !stderr_is_tty
) {
1361 /* Detach from stdout/stderr. and reopen
1362 * /dev/tty for them. This is important to
1363 * ensure that when systemctl is started via
1364 * popen() or a similar call that expects to
1365 * read EOF we actually do generate EOF and
1366 * not delay this indefinitely by because we
1367 * keep an unused copy of stdin around. */
1368 fd
= open("/dev/tty", O_WRONLY
);
1370 log_error_errno(errno
, "Failed to open /dev/tty: %m");
1371 _exit(EXIT_FAILURE
);
1374 if (!stdout_is_tty
&& dup2(fd
, STDOUT_FILENO
) < 0) {
1375 log_error_errno(errno
, "Failed to dup2 /dev/tty: %m");
1376 _exit(EXIT_FAILURE
);
1379 if (!stderr_is_tty
&& dup2(fd
, STDERR_FILENO
) < 0) {
1380 log_error_errno(errno
, "Failed to dup2 /dev/tty: %m");
1381 _exit(EXIT_FAILURE
);
1384 if (fd
> STDERR_FILENO
)
1388 /* Count arguments */
1390 for (n
= 0; va_arg(ap
, char*); n
++)
1395 l
= alloca(sizeof(char *) * (n
+ 1));
1397 /* Fill in arguments */
1399 for (i
= 0; i
<= n
; i
++)
1400 l
[i
] = va_arg(ap
, char*);
1404 _exit(EXIT_FAILURE
);
1407 static const char *const ioprio_class_table
[] = {
1408 [IOPRIO_CLASS_NONE
] = "none",
1409 [IOPRIO_CLASS_RT
] = "realtime",
1410 [IOPRIO_CLASS_BE
] = "best-effort",
1411 [IOPRIO_CLASS_IDLE
] = "idle"
1414 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ioprio_class
, int, INT_MAX
);
1416 static const char *const sigchld_code_table
[] = {
1417 [CLD_EXITED
] = "exited",
1418 [CLD_KILLED
] = "killed",
1419 [CLD_DUMPED
] = "dumped",
1420 [CLD_TRAPPED
] = "trapped",
1421 [CLD_STOPPED
] = "stopped",
1422 [CLD_CONTINUED
] = "continued",
1425 DEFINE_STRING_TABLE_LOOKUP(sigchld_code
, int);
1427 static const char* const sched_policy_table
[] = {
1428 [SCHED_OTHER
] = "other",
1429 [SCHED_BATCH
] = "batch",
1430 [SCHED_IDLE
] = "idle",
1431 [SCHED_FIFO
] = "fifo",
1435 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(sched_policy
, int, INT_MAX
);