2 This file is part of systemd.
4 Copyright 2010 Lennart Poettering
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
23 #include <linux/oom.h>
31 #include <sys/personality.h>
32 #include <sys/prctl.h>
33 #include <sys/types.h>
37 #if HAVE_VALGRIND_VALGRIND_H
38 #include <valgrind/valgrind.h>
41 #include "alloc-util.h"
42 //#include "architecture.h"
51 #include "process-util.h"
52 //#include "raw-clone.h"
53 #include "signal-util.h"
54 //#include "stat-util.h"
55 #include "string-table.h"
56 #include "string-util.h"
57 #include "user-util.h"
60 int get_process_state(pid_t pid) {
64 _cleanup_free_ char *line = NULL;
68 p = procfs_file_alloca(pid, "stat");
70 r = read_one_line_file(p, &line);
76 p = strrchr(line, ')');
82 if (sscanf(p, " %c", &state) != 1)
85 return (unsigned char) state;
88 int get_process_comm(pid_t pid, char **name) {
95 p = procfs_file_alloca(pid, "comm");
97 r = read_one_line_file(p, name);
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 if (max_length == 1) {
134 /* If there's only room for one byte, return the empty string */
142 } else if (max_length == 0) {
143 size_t len = 0, allocated = 0;
145 while ((c = getc(f)) != EOF) {
147 if (!GREEDY_REALLOC(ans, allocated, len+3)) {
169 bool dotdotdot = false;
172 ans = new(char, max_length);
178 while ((c = getc(f)) != EOF) {
205 if (max_length <= 4) {
209 k = ans + max_length - 4;
212 /* Eat up final spaces */
213 while (k > ans && isspace(k[-1])) {
219 strncpy(k, "...", left-1);
225 /* Kernel threads have no argv[] */
227 _cleanup_free_ char *t = NULL;
235 h = get_process_comm(pid, &t);
240 ans = strjoin("[", t, "]");
246 if (l + 3 <= max_length)
247 ans = strjoin("[", t, "]");
248 else if (max_length <= 6) {
250 ans = new(char, max_length);
254 memcpy(ans, "[...]", max_length-1);
255 ans[max_length-1] = 0;
259 t[max_length - 6] = 0;
261 /* Chop off final spaces */
263 while (e > t && isspace(e[-1]))
267 ans = strjoin("[", t, "...]");
278 #if 0 /// UNNEEDED by elogind
279 int rename_process(const char name[]) {
280 static size_t mm_size = 0;
281 static char *mm = NULL;
282 bool truncated = false;
285 /* This is a like a poor man's setproctitle(). It changes the comm field, argv[0], and also the glibc's
286 * internally used name of the process. For the first one a limit of 16 chars applies; to the second one in
287 * many cases one of 10 (i.e. length of "/sbin/init") — however if we have CAP_SYS_RESOURCES it is unbounded;
288 * to the third one 7 (i.e. the length of "systemd". If you pass a longer string it will likely be
291 * Returns 0 if a name was set but truncated, > 0 if it was set but not truncated. */
294 return -EINVAL; /* let's not confuse users unnecessarily with an empty name */
298 /* First step, change the comm field. */
299 (void) prctl(PR_SET_NAME, name);
300 if (l > 15) /* 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]));
390 int is_kernel_thread(pid_t pid) {
397 if (pid == 0 || pid == 1 || pid == getpid_cached()) /* pid 1, and we ourselves certainly aren't a kernel thread */
402 p = procfs_file_alloca(pid, "cmdline");
410 count = fread(&c, 1, 1, f);
414 /* Kernel threads have an empty cmdline */
417 return eof ? 1 : -errno;
422 #if 0 /// UNNEEDED by elogind
423 int get_process_capeff(pid_t pid, char **capeff) {
430 p = procfs_file_alloca(pid, "status");
432 r = get_proc_field(p, "CapEff", WHITESPACE, capeff);
440 static int get_process_link_contents(const char *proc_file, char **name) {
446 r = readlink_malloc(proc_file, name);
455 int get_process_exe(pid_t pid, char **name) {
462 p = procfs_file_alloca(pid, "exe");
463 r = get_process_link_contents(p, name);
467 d = endswith(*name, " (deleted)");
474 #if 0 /// UNNEEDED by elogind
475 static int get_process_id(pid_t pid, const char *field, uid_t *uid) {
476 _cleanup_fclose_ FILE *f = NULL;
483 if (!pid_is_valid(pid))
486 p = procfs_file_alloca(pid, "status");
494 FOREACH_LINE(line, f, return -errno) {
499 if (startswith(l, field)) {
501 l += strspn(l, WHITESPACE);
503 l[strcspn(l, WHITESPACE)] = 0;
505 return parse_uid(l, uid);
512 int get_process_uid(pid_t pid, uid_t *uid) {
514 if (pid == 0 || pid == getpid_cached()) {
519 return get_process_id(pid, "Uid:", uid);
522 int get_process_gid(pid_t pid, gid_t *gid) {
524 if (pid == 0 || pid == getpid_cached()) {
529 assert_cc(sizeof(uid_t) == sizeof(gid_t));
530 return get_process_id(pid, "Gid:", gid);
533 int get_process_cwd(pid_t pid, char **cwd) {
538 p = procfs_file_alloca(pid, "cwd");
540 return get_process_link_contents(p, cwd);
543 int get_process_root(pid_t pid, char **root) {
548 p = procfs_file_alloca(pid, "root");
550 return get_process_link_contents(p, root);
553 int get_process_environ(pid_t pid, char **env) {
554 _cleanup_fclose_ FILE *f = NULL;
555 _cleanup_free_ char *outcome = NULL;
558 size_t allocated = 0, sz = 0;
563 p = procfs_file_alloca(pid, "environ");
572 while ((c = fgetc(f)) != EOF) {
573 if (!GREEDY_REALLOC(outcome, allocated, sz + 5))
577 outcome[sz++] = '\n';
579 sz += cescape_char(c, outcome + sz);
583 outcome = strdup("");
595 int get_process_ppid(pid_t pid, pid_t *_ppid) {
597 _cleanup_free_ char *line = NULL;
604 if (pid == 0 || pid == getpid_cached()) {
609 p = procfs_file_alloca(pid, "stat");
610 r = read_one_line_file(p, &line);
616 /* Let's skip the pid and comm fields. The latter is enclosed
617 * in () but does not escape any () in its value, so let's
618 * skip over it manually */
620 p = strrchr(line, ')');
632 if ((long unsigned) (pid_t) ppid != ppid)
635 *_ppid = (pid_t) ppid;
641 int wait_for_terminate(pid_t pid, siginfo_t *status) {
652 if (waitid(P_PID, pid, status, WEXITED) < 0) {
657 return negative_errno();
666 * < 0 : wait_for_terminate() failed to get the state of the
667 * process, the process was terminated by a signal, or
668 * failed for an unknown reason.
669 * >=0 : The process terminated normally, and its exit code is
672 * That is, success is indicated by a return value of zero, and an
673 * error is indicated by a non-zero value.
675 * A warning is emitted if the process terminates abnormally,
676 * and also if it returns non-zero unless check_exit_code is true.
678 int wait_for_terminate_and_warn(const char *name, pid_t pid, bool check_exit_code) {
685 r = wait_for_terminate(pid, &status);
687 return log_warning_errno(r, "Failed to wait for %s: %m", name);
689 if (status.si_code == CLD_EXITED) {
690 if (status.si_status != 0)
691 log_full(check_exit_code ? LOG_WARNING : LOG_DEBUG,
692 "%s failed with error code %i.", name, status.si_status);
694 log_debug("%s succeeded.", name);
696 return status.si_status;
697 } else if (IN_SET(status.si_code, CLD_KILLED, CLD_DUMPED)) {
699 log_warning("%s terminated by signal %s.", name, signal_to_string(status.si_status));
703 log_warning("%s failed due to unknown reason.", name);
707 #if 0 /// UNNEEDED by elogind
708 void sigkill_wait(pid_t pid) {
711 if (kill(pid, SIGKILL) > 0)
712 (void) wait_for_terminate(pid, NULL);
715 void sigkill_waitp(pid_t *pid) {
724 int kill_and_sigcont(pid_t pid, int sig) {
727 r = kill(pid, sig) < 0 ? -errno : 0;
729 /* If this worked, also send SIGCONT, unless we already just sent a SIGCONT, or SIGKILL was sent which isn't
730 * affected by a process being suspended anyway. */
731 if (r >= 0 && !IN_SET(sig, SIGCONT, SIGKILL))
732 (void) kill(pid, SIGCONT);
738 int getenv_for_pid(pid_t pid, const char *field, char **_value) {
739 _cleanup_fclose_ FILE *f = NULL;
750 path = procfs_file_alloca(pid, "environ");
752 f = fopen(path, "re");
766 for (i = 0; i < sizeof(line)-1; i++) {
770 if (_unlikely_(c == EOF)) {
780 if (strneq(line, field, l) && line[l] == '=') {
781 value = strdup(line + l + 1);
795 bool pid_is_unwaited(pid_t pid) {
796 /* Checks whether a PID is still valid at all, including a zombie */
798 if (!pid_is_valid(pid))
801 if (pid <= 1) /* If we or PID 1 would be dead and have been waited for, this code would not be running */
804 if (pid == getpid_cached())
807 if (kill(pid, 0) >= 0)
810 return errno != ESRCH;
813 bool pid_is_alive(pid_t pid) {
816 /* Checks whether a PID is still valid and not a zombie */
818 if (!pid_is_valid(pid))
821 if (pid <= 1) /* If we or PID 1 would be a zombie, this code would not be running */
824 if (pid == getpid_cached())
827 r = get_process_state(pid);
828 if (r == -ESRCH || r == 'Z')
834 #if 0 /// UNNEEDED by elogind
835 int pid_from_same_root_fs(pid_t pid) {
838 if (!pid_is_valid(pid))
841 if (pid == 0 || pid == getpid_cached())
844 root = procfs_file_alloca(pid, "root");
846 return files_same(root, "/proc/1/root", 0);
850 bool is_main_thread(void) {
851 static thread_local int cached = 0;
853 if (_unlikely_(cached == 0))
854 cached = getpid_cached() == gettid() ? 1 : -1;
859 #if 0 /// UNNEEDED by elogind
860 noreturn void freeze(void) {
864 /* Make sure nobody waits for us on a socket anymore */
865 close_all_fds(NULL, 0);
873 bool oom_score_adjust_is_valid(int oa) {
874 return oa >= OOM_SCORE_ADJ_MIN && oa <= OOM_SCORE_ADJ_MAX;
877 unsigned long personality_from_string(const char *p) {
881 return PERSONALITY_INVALID;
883 /* Parse a personality specifier. We use our own identifiers that indicate specific ABIs, rather than just
884 * hints regarding the register size, since we want to keep things open for multiple locally supported ABIs for
885 * the same register size. */
887 architecture = architecture_from_string(p);
888 if (architecture < 0)
889 return PERSONALITY_INVALID;
891 if (architecture == native_architecture())
893 #ifdef SECONDARY_ARCHITECTURE
894 if (architecture == SECONDARY_ARCHITECTURE)
898 return PERSONALITY_INVALID;
901 const char* personality_to_string(unsigned long p) {
902 int architecture = _ARCHITECTURE_INVALID;
905 architecture = native_architecture();
906 #ifdef SECONDARY_ARCHITECTURE
907 else if (p == PER_LINUX32)
908 architecture = SECONDARY_ARCHITECTURE;
911 if (architecture < 0)
914 return architecture_to_string(architecture);
917 int safe_personality(unsigned long p) {
920 /* So here's the deal, personality() is weirdly defined by glibc. In some cases it returns a failure via errno,
921 * and in others as negative return value containing an errno-like value. Let's work around this: this is a
922 * wrapper that uses errno if it is set, and uses the return value otherwise. And then it sets both errno and
923 * the return value indicating the same issue, so that we are definitely on the safe side.
925 * See https://github.com/elogind/elogind/issues/6737 */
928 ret = personality(p);
939 int opinionated_personality(unsigned long *ret) {
942 /* Returns the current personality, or PERSONALITY_INVALID if we can't determine it. This function is a bit
943 * opinionated though, and ignores all the finer-grained bits and exotic personalities, only distinguishing the
944 * two most relevant personalities: PER_LINUX and PER_LINUX32. */
946 current = safe_personality(PERSONALITY_INVALID);
950 if (((unsigned long) current & 0xffff) == PER_LINUX32)
958 void valgrind_summary_hack(void) {
959 #if HAVE_VALGRIND_VALGRIND_H
960 if (getpid_cached() == 1 && RUNNING_ON_VALGRIND) {
962 pid = raw_clone(SIGCHLD);
964 log_emergency_errno(errno, "Failed to fork off valgrind helper: %m");
968 log_info("Spawned valgrind helper as PID "PID_FMT".", pid);
969 (void) wait_for_terminate(pid, NULL);
975 int pid_compare_func(const void *a, const void *b) {
976 const pid_t *p = a, *q = b;
978 /* Suitable for usage in qsort() */
987 int ioprio_parse_priority(const char *s, int *ret) {
993 r = safe_atoi(s, &i);
997 if (!ioprio_priority_is_valid(i))
1005 /* The cached PID, possible values:
1007 * == UNSET [0] → cache not initialized yet
1008 * == BUSY [-1] → some thread is initializing it at the moment
1009 * any other → the cached PID
1012 #define CACHED_PID_UNSET ((pid_t) 0)
1013 #define CACHED_PID_BUSY ((pid_t) -1)
1015 static pid_t cached_pid = CACHED_PID_UNSET;
1017 static void reset_cached_pid(void) {
1018 /* Invoked in the child after a fork(), i.e. at the first moment the PID changed */
1019 cached_pid = CACHED_PID_UNSET;
1022 /* We use glibc __register_atfork() + __dso_handle directly here, as they are not included in the glibc
1023 * headers. __register_atfork() is mostly equivalent to pthread_atfork(), but doesn't require us to link against
1024 * libpthread, as it is part of glibc anyway. */
1025 extern int __register_atfork(void (*prepare) (void), void (*parent) (void), void (*child) (void), void * __dso_handle);
1026 extern void* __dso_handle __attribute__ ((__weak__));
1028 pid_t getpid_cached(void) {
1029 pid_t current_value;
1031 /* getpid_cached() is much like getpid(), but caches the value in local memory, to avoid having to invoke a
1032 * system call each time. This restores glibc behaviour from before 2.24, when getpid() was unconditionally
1033 * cached. Starting with 2.24 getpid() started to become prohibitively expensive when used for detecting when
1034 * objects were used across fork()s. With this caching the old behaviour is somewhat restored.
1036 * https://bugzilla.redhat.com/show_bug.cgi?id=1443976
1037 * https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=c579f48edba88380635ab98cb612030e3ed8691e
1040 current_value = __sync_val_compare_and_swap(&cached_pid, CACHED_PID_UNSET, CACHED_PID_BUSY);
1042 switch (current_value) {
1044 case CACHED_PID_UNSET: { /* Not initialized yet, then do so now */
1049 if (__register_atfork(NULL, NULL, reset_cached_pid, __dso_handle) != 0) {
1050 /* OOM? Let's try again later */
1051 cached_pid = CACHED_PID_UNSET;
1055 cached_pid = new_pid;
1059 case CACHED_PID_BUSY: /* Somebody else is currently initializing */
1062 default: /* Properly initialized */
1063 return current_value;
1067 #if 0 /// UNNEEDED by elogind
1068 static const char *const ioprio_class_table[] = {
1069 [IOPRIO_CLASS_NONE] = "none",
1070 [IOPRIO_CLASS_RT] = "realtime",
1071 [IOPRIO_CLASS_BE] = "best-effort",
1072 [IOPRIO_CLASS_IDLE] = "idle"
1075 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ioprio_class, int, INT_MAX);
1077 static const char *const sigchld_code_table[] = {
1078 [CLD_EXITED] = "exited",
1079 [CLD_KILLED] = "killed",
1080 [CLD_DUMPED] = "dumped",
1081 [CLD_TRAPPED] = "trapped",
1082 [CLD_STOPPED] = "stopped",
1083 [CLD_CONTINUED] = "continued",
1086 DEFINE_STRING_TABLE_LOOKUP(sigchld_code, int);
1088 static const char* const sched_policy_table[] = {
1089 [SCHED_OTHER] = "other",
1090 [SCHED_BATCH] = "batch",
1091 [SCHED_IDLE] = "idle",
1092 [SCHED_FIFO] = "fifo",
1096 DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(sched_policy, int, INT_MAX);