1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
4 This file is part of systemd.
6 Copyright 2010 Lennart Poettering
8 systemd is free software; you can redistribute it and/or modify it
9 under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2 of the License, or
11 (at your option) any later version.
13 systemd is distributed in the hope that it will be useful, but
14 WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with systemd; If not, see <http://www.gnu.org/licenses/>.
29 #include <sys/socket.h>
31 #include <sys/prctl.h>
32 #include <linux/sched.h>
33 #include <sys/types.h>
37 #include <sys/mount.h>
39 #include <linux/oom.h>
42 #include <security/pam_appl.h>
51 #include "securebits.h"
53 #include "namespace.h"
55 #include "exit-status.h"
57 #include "utmp-wtmp.h"
60 /* This assumes there is a 'tty' group */
63 static int shift_fds(int fds[], unsigned n_fds) {
64 int start, restart_from;
69 /* Modifies the fds array! (sorts it) */
79 for (i = start; i < (int) n_fds; i++) {
82 /* Already at right index? */
86 if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
89 close_nointr_nofail(fds[i]);
92 /* Hmm, the fd we wanted isn't free? Then
93 * let's remember that and try again from here*/
94 if (nfd != i+3 && restart_from < 0)
101 start = restart_from;
107 static int flags_fds(const int fds[], unsigned n_fds, bool nonblock) {
116 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
118 for (i = 0; i < n_fds; i++) {
120 if ((r = fd_nonblock(fds[i], nonblock)) < 0)
123 /* We unconditionally drop FD_CLOEXEC from the fds,
124 * since after all we want to pass these fds to our
127 if ((r = fd_cloexec(fds[i], false)) < 0)
134 static const char *tty_path(const ExecContext *context) {
137 if (context->tty_path)
138 return context->tty_path;
140 return "/dev/console";
143 static int open_null_as(int flags, int nfd) {
148 if ((fd = open("/dev/null", flags|O_NOCTTY)) < 0)
152 r = dup2(fd, nfd) < 0 ? -errno : nfd;
153 close_nointr_nofail(fd);
160 static int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, int nfd) {
164 struct sockaddr_un un;
168 assert(output < _EXEC_OUTPUT_MAX);
172 if ((fd = socket(AF_UNIX, SOCK_STREAM, 0)) < 0)
176 sa.sa.sa_family = AF_UNIX;
177 strncpy(sa.un.sun_path, LOGGER_SOCKET, sizeof(sa.un.sun_path));
179 if (connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + sizeof(LOGGER_SOCKET) - 1) < 0) {
180 close_nointr_nofail(fd);
184 if (shutdown(fd, SHUT_RD) < 0) {
185 close_nointr_nofail(fd);
189 /* We speak a very simple protocol between log server
190 * and client: one line for the log destination (kmsg
191 * or syslog), followed by the priority field,
192 * followed by the process name. Since we replaced
193 * stdin/stderr we simple use stdio to write to
194 * it. Note that we use stderr, to minimize buffer
195 * flushing issues. */
202 output == EXEC_OUTPUT_KMSG ? "kmsg" :
203 output == EXEC_OUTPUT_KMSG_AND_CONSOLE ? "kmsg+console" :
204 output == EXEC_OUTPUT_SYSLOG ? "syslog" :
206 context->syslog_priority,
207 context->syslog_identifier ? context->syslog_identifier : ident,
208 context->syslog_level_prefix);
211 r = dup2(fd, nfd) < 0 ? -errno : nfd;
212 close_nointr_nofail(fd);
218 static int open_terminal_as(const char *path, mode_t mode, int nfd) {
224 if ((fd = open_terminal(path, mode | O_NOCTTY)) < 0)
228 r = dup2(fd, nfd) < 0 ? -errno : nfd;
229 close_nointr_nofail(fd);
236 static bool is_terminal_input(ExecInput i) {
238 i == EXEC_INPUT_TTY ||
239 i == EXEC_INPUT_TTY_FORCE ||
240 i == EXEC_INPUT_TTY_FAIL;
243 static int fixup_input(ExecInput std_input, int socket_fd, bool apply_tty_stdin) {
245 if (is_terminal_input(std_input) && !apply_tty_stdin)
246 return EXEC_INPUT_NULL;
248 if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0)
249 return EXEC_INPUT_NULL;
254 static int fixup_output(ExecOutput std_output, int socket_fd) {
256 if (std_output == EXEC_OUTPUT_SOCKET && socket_fd < 0)
257 return EXEC_OUTPUT_INHERIT;
262 static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty_stdin) {
267 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
271 case EXEC_INPUT_NULL:
272 return open_null_as(O_RDONLY, STDIN_FILENO);
275 case EXEC_INPUT_TTY_FORCE:
276 case EXEC_INPUT_TTY_FAIL: {
279 if ((fd = acquire_terminal(
281 i == EXEC_INPUT_TTY_FAIL,
282 i == EXEC_INPUT_TTY_FORCE,
286 if (fd != STDIN_FILENO) {
287 r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
288 close_nointr_nofail(fd);
295 case EXEC_INPUT_SOCKET:
296 return dup2(socket_fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
299 assert_not_reached("Unknown input type");
303 static int setup_output(const ExecContext *context, int socket_fd, const char *ident, bool apply_tty_stdin) {
310 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
311 o = fixup_output(context->std_output, socket_fd);
313 /* This expects the input is already set up */
317 case EXEC_OUTPUT_INHERIT:
319 /* If input got downgraded, inherit the original value */
320 if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input))
321 return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO);
323 /* If the input is connected to anything that's not a /dev/null, inherit that... */
324 if (i != EXEC_INPUT_NULL)
325 return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
327 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
329 return STDOUT_FILENO;
331 /* We need to open /dev/null here anew, to get the
332 * right access mode. So we fall through */
334 case EXEC_OUTPUT_NULL:
335 return open_null_as(O_WRONLY, STDOUT_FILENO);
337 case EXEC_OUTPUT_TTY:
338 if (is_terminal_input(i))
339 return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
341 /* We don't reset the terminal if this is just about output */
342 return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO);
344 case EXEC_OUTPUT_SYSLOG:
345 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
346 case EXEC_OUTPUT_KMSG:
347 case EXEC_OUTPUT_KMSG_AND_CONSOLE:
348 return connect_logger_as(context, o, ident, STDOUT_FILENO);
350 case EXEC_OUTPUT_SOCKET:
351 assert(socket_fd >= 0);
352 return dup2(socket_fd, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
355 assert_not_reached("Unknown output type");
359 static int setup_error(const ExecContext *context, int socket_fd, const char *ident, bool apply_tty_stdin) {
366 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
367 o = fixup_output(context->std_output, socket_fd);
368 e = fixup_output(context->std_error, socket_fd);
370 /* This expects the input and output are already set up */
372 /* Don't change the stderr file descriptor if we inherit all
373 * the way and are not on a tty */
374 if (e == EXEC_OUTPUT_INHERIT &&
375 o == EXEC_OUTPUT_INHERIT &&
376 i == EXEC_INPUT_NULL &&
377 !is_terminal_input(context->std_input) &&
379 return STDERR_FILENO;
381 /* Duplicate from stdout if possible */
382 if (e == o || e == EXEC_OUTPUT_INHERIT)
383 return dup2(STDOUT_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
387 case EXEC_OUTPUT_NULL:
388 return open_null_as(O_WRONLY, STDERR_FILENO);
390 case EXEC_OUTPUT_TTY:
391 if (is_terminal_input(i))
392 return dup2(STDIN_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
394 /* We don't reset the terminal if this is just about output */
395 return open_terminal_as(tty_path(context), O_WRONLY, STDERR_FILENO);
397 case EXEC_OUTPUT_SYSLOG:
398 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
399 case EXEC_OUTPUT_KMSG:
400 case EXEC_OUTPUT_KMSG_AND_CONSOLE:
401 return connect_logger_as(context, e, ident, STDERR_FILENO);
403 case EXEC_OUTPUT_SOCKET:
404 assert(socket_fd >= 0);
405 return dup2(socket_fd, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
408 assert_not_reached("Unknown error type");
412 static int chown_terminal(int fd, uid_t uid) {
417 /* This might fail. What matters are the results. */
418 (void) fchown(fd, uid, -1);
419 (void) fchmod(fd, TTY_MODE);
421 if (fstat(fd, &st) < 0)
424 if (st.st_uid != uid || (st.st_mode & 0777) != TTY_MODE)
430 static int setup_confirm_stdio(const ExecContext *context,
432 int *_saved_stdout) {
433 int fd = -1, saved_stdin, saved_stdout = -1, r;
436 assert(_saved_stdin);
437 assert(_saved_stdout);
439 /* This returns positive EXIT_xxx return values instead of
440 * negative errno style values! */
442 if ((saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3)) < 0)
445 if ((saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3)) < 0) {
450 if ((fd = acquire_terminal(
452 context->std_input == EXEC_INPUT_TTY_FAIL,
453 context->std_input == EXEC_INPUT_TTY_FORCE,
459 if (chown_terminal(fd, getuid()) < 0) {
464 if (dup2(fd, STDIN_FILENO) < 0) {
469 if (dup2(fd, STDOUT_FILENO) < 0) {
475 close_nointr_nofail(fd);
477 *_saved_stdin = saved_stdin;
478 *_saved_stdout = saved_stdout;
483 if (saved_stdout >= 0)
484 close_nointr_nofail(saved_stdout);
486 if (saved_stdin >= 0)
487 close_nointr_nofail(saved_stdin);
490 close_nointr_nofail(fd);
495 static int restore_confirm_stdio(const ExecContext *context,
503 assert(*saved_stdin >= 0);
504 assert(saved_stdout);
505 assert(*saved_stdout >= 0);
507 /* This returns positive EXIT_xxx return values instead of
508 * negative errno style values! */
510 if (is_terminal_input(context->std_input)) {
512 /* The service wants terminal input. */
516 context->std_output == EXEC_OUTPUT_INHERIT ||
517 context->std_output == EXEC_OUTPUT_TTY;
520 /* If the service doesn't want a controlling terminal,
521 * then we need to get rid entirely of what we have
524 if (release_terminal() < 0)
527 if (dup2(*saved_stdin, STDIN_FILENO) < 0)
530 if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
533 *keep_stdout = *keep_stdin = false;
539 static int get_group_creds(const char *groupname, gid_t *gid) {
546 /* We enforce some special rules for gid=0: in order to avoid
547 * NSS lookups for root we hardcode its data. */
549 if (streq(groupname, "root") || streq(groupname, "0")) {
554 if (safe_atolu(groupname, &lu) >= 0) {
556 g = getgrgid((gid_t) lu);
559 g = getgrnam(groupname);
563 return errno != 0 ? -errno : -ESRCH;
569 static int get_user_creds(const char **username, uid_t *uid, gid_t *gid, const char **home) {
579 /* We enforce some special rules for uid=0: in order to avoid
580 * NSS lookups for root we hardcode its data. */
582 if (streq(*username, "root") || streq(*username, "0")) {
590 if (safe_atolu(*username, &lu) >= 0) {
592 p = getpwuid((uid_t) lu);
594 /* If there are multiple users with the same id, make
595 * sure to leave $USER to the configured value instead
596 * of the first occurrence in the database. However if
597 * the uid was configured by a numeric uid, then let's
598 * pick the real username from /etc/passwd. */
600 *username = p->pw_name;
603 p = getpwnam(*username);
607 return errno != 0 ? -errno : -ESRCH;
615 static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
616 bool keep_groups = false;
621 /* Lookup and set GID and supplementary group list. Here too
622 * we avoid NSS lookups for gid=0. */
624 if (context->group || username) {
627 if ((r = get_group_creds(context->group, &gid)) < 0)
630 /* First step, initialize groups from /etc/groups */
631 if (username && gid != 0) {
632 if (initgroups(username, gid) < 0)
638 /* Second step, set our gids */
639 if (setresgid(gid, gid, gid) < 0)
643 if (context->supplementary_groups) {
648 /* Final step, initialize any manually set supplementary groups */
649 assert_se((ngroups_max = (int) sysconf(_SC_NGROUPS_MAX)) > 0);
651 if (!(gids = new(gid_t, ngroups_max)))
655 if ((k = getgroups(ngroups_max, gids)) < 0) {
662 STRV_FOREACH(i, context->supplementary_groups) {
664 if (k >= ngroups_max) {
669 if ((r = get_group_creds(*i, gids+k)) < 0) {
677 if (setgroups(k, gids) < 0) {
688 static int enforce_user(const ExecContext *context, uid_t uid) {
692 /* Sets (but doesn't lookup) the uid and make sure we keep the
693 * capabilities while doing so. */
695 if (context->capabilities) {
697 static const cap_value_t bits[] = {
698 CAP_SETUID, /* Necessary so that we can run setresuid() below */
699 CAP_SETPCAP /* Necessary so that we can set PR_SET_SECUREBITS later on */
702 /* First step: If we need to keep capabilities but
703 * drop privileges we need to make sure we keep our
704 * caps, whiel we drop privileges. */
706 int sb = context->secure_bits|SECURE_KEEP_CAPS;
708 if (prctl(PR_GET_SECUREBITS) != sb)
709 if (prctl(PR_SET_SECUREBITS, sb) < 0)
713 /* Second step: set the capabilities. This will reduce
714 * the capabilities to the minimum we need. */
716 if (!(d = cap_dup(context->capabilities)))
719 if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
720 cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0) {
726 if (cap_set_proc(d) < 0) {
735 /* Third step: actually set the uids */
736 if (setresuid(uid, uid, uid) < 0)
739 /* At this point we should have all necessary capabilities but
740 are otherwise a normal user. However, the caps might got
741 corrupted due to the setresuid() so we need clean them up
742 later. This is done outside of this call. */
749 static int null_conv(
751 const struct pam_message **msg,
752 struct pam_response **resp,
755 /* We don't support conversations */
760 static int setup_pam(
765 int fds[], unsigned n_fds) {
767 static const struct pam_conv conv = {
772 pam_handle_t *handle = NULL;
774 int pam_code = PAM_SUCCESS;
776 bool close_session = false;
777 pid_t pam_pid = 0, parent_pid;
783 /* We set up PAM in the parent process, then fork. The child
784 * will then stay around until killed via PR_GET_PDEATHSIG or
785 * systemd via the cgroup logic. It will then remove the PAM
786 * session again. The parent process will exec() the actual
787 * daemon. We do things this way to ensure that the main PID
788 * of the daemon is the one we initially fork()ed. */
790 if ((pam_code = pam_start(name, user, &conv, &handle)) != PAM_SUCCESS) {
796 if ((pam_code = pam_set_item(handle, PAM_TTY, tty)) != PAM_SUCCESS)
799 if ((pam_code = pam_acct_mgmt(handle, PAM_SILENT)) != PAM_SUCCESS)
802 if ((pam_code = pam_open_session(handle, PAM_SILENT)) != PAM_SUCCESS)
805 close_session = true;
807 if ((pam_code = pam_setcred(handle, PAM_ESTABLISH_CRED | PAM_SILENT)) != PAM_SUCCESS)
810 if ((!(e = pam_getenvlist(handle)))) {
811 pam_code = PAM_BUF_ERR;
815 /* Block SIGTERM, so that we know that it won't get lost in
817 if (sigemptyset(&ss) < 0 ||
818 sigaddset(&ss, SIGTERM) < 0 ||
819 sigprocmask(SIG_BLOCK, &ss, &old_ss) < 0)
822 parent_pid = getpid();
824 if ((pam_pid = fork()) < 0)
831 /* The child's job is to reset the PAM session on
834 /* This string must fit in 10 chars (i.e. the length
835 * of "/sbin/init") */
836 rename_process("sd:pam");
838 /* Make sure we don't keep open the passed fds in this
839 child. We assume that otherwise only those fds are
840 open here that have been opened by PAM. */
841 close_many(fds, n_fds);
843 /* Wait until our parent died. This will most likely
844 * not work since the kernel does not allow
845 * unprivileged parents kill their privileged children
846 * this way. We rely on the control groups kill logic
847 * to do the rest for us. */
848 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
851 /* Check if our parent process might already have
853 if (getppid() == parent_pid) {
854 if (sigwait(&ss, &sig) < 0)
857 assert(sig == SIGTERM);
860 /* Only if our parent died we'll end the session */
861 if (getppid() != parent_pid)
862 if ((pam_code = pam_close_session(handle, PAM_DATA_SILENT)) != PAM_SUCCESS)
868 pam_end(handle, pam_code | PAM_DATA_SILENT);
872 /* If the child was forked off successfully it will do all the
873 * cleanups, so forget about the handle here. */
876 /* Unblock SIGSUR1 again in the parent */
877 if (sigprocmask(SIG_SETMASK, &old_ss, NULL) < 0)
880 /* We close the log explicitly here, since the PAM modules
881 * might have opened it, but we don't want this fd around. */
889 pam_code = pam_close_session(handle, PAM_DATA_SILENT);
891 pam_end(handle, pam_code | PAM_DATA_SILENT);
899 kill(pam_pid, SIGTERM);
900 kill(pam_pid, SIGCONT);
907 int exec_spawn(ExecCommand *command,
909 const ExecContext *context,
910 int fds[], unsigned n_fds,
912 bool apply_permissions,
914 bool apply_tty_stdin,
916 CGroupBonding *cgroup_bondings,
923 char **files_env = NULL;
928 assert(fds || n_fds <= 0);
930 if (context->std_input == EXEC_INPUT_SOCKET ||
931 context->std_output == EXEC_OUTPUT_SOCKET ||
932 context->std_error == EXEC_OUTPUT_SOCKET) {
944 if ((r = exec_context_load_environment(context, &files_env)) < 0) {
945 log_error("Failed to load environment files: %s", strerror(-r));
950 argv = command->argv;
952 if (!(line = exec_command_line(argv))) {
957 log_debug("About to execute: %s", line);
961 if ((r = cgroup_bonding_realize_list(cgroup_bondings)))
964 if ((pid = fork()) < 0) {
972 const char *username = NULL, *home = NULL;
973 uid_t uid = (uid_t) -1;
974 gid_t gid = (gid_t) -1;
975 char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
977 int saved_stdout = -1, saved_stdin = -1;
978 bool keep_stdout = false, keep_stdin = false;
982 /* This string must fit in 10 chars (i.e. the length
983 * of "/sbin/init") */
984 rename_process("sd.exec");
986 /* We reset exactly these signals, since they are the
987 * only ones we set to SIG_IGN in the main daemon. All
988 * others we leave untouched because we set them to
989 * SIG_DFL or a valid handler initially, both of which
990 * will be demoted to SIG_DFL. */
991 default_signals(SIGNALS_CRASH_HANDLER,
994 if (sigemptyset(&ss) < 0 ||
995 sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
996 r = EXIT_SIGNAL_MASK;
1000 /* Close sockets very early to make sure we don't
1001 * block init reexecution because it cannot bind its
1003 if (close_all_fds(socket_fd >= 0 ? &socket_fd : fds,
1004 socket_fd >= 0 ? 1 : n_fds) < 0) {
1009 if (!context->same_pgrp)
1015 if (context->tcpwrap_name) {
1017 if (!socket_tcpwrap(socket_fd, context->tcpwrap_name)) {
1022 for (i = 0; i < (int) n_fds; i++) {
1023 if (!socket_tcpwrap(fds[i], context->tcpwrap_name)) {
1030 /* We skip the confirmation step if we shall not apply the TTY */
1031 if (confirm_spawn &&
1032 (!is_terminal_input(context->std_input) || apply_tty_stdin)) {
1035 /* Set up terminal for the question */
1036 if ((r = setup_confirm_stdio(context,
1037 &saved_stdin, &saved_stdout)))
1040 /* Now ask the question. */
1041 if (!(line = exec_command_line(argv))) {
1046 r = ask(&response, "yns", "Execute %s? [Yes, No, Skip] ", line);
1049 if (r < 0 || response == 'n') {
1052 } else if (response == 's') {
1057 /* Release terminal for the question */
1058 if ((r = restore_confirm_stdio(context,
1059 &saved_stdin, &saved_stdout,
1060 &keep_stdin, &keep_stdout)))
1064 /* If a socket is connected to STDIN/STDOUT/STDERR, we
1065 * must sure to drop O_NONBLOCK */
1067 fd_nonblock(socket_fd, false);
1070 if (setup_input(context, socket_fd, apply_tty_stdin) < 0) {
1076 if (setup_output(context, socket_fd, file_name_from_path(command->path), apply_tty_stdin) < 0) {
1081 if (setup_error(context, socket_fd, file_name_from_path(command->path), apply_tty_stdin) < 0) {
1086 if (cgroup_bondings)
1087 if (cgroup_bonding_install_list(cgroup_bondings, 0) < 0) {
1092 if (context->oom_score_adjust_set) {
1095 snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
1098 if (write_one_line_file("/proc/self/oom_score_adj", t) < 0) {
1099 /* Compatibility with Linux <= 2.6.35 */
1103 adj = (context->oom_score_adjust * -OOM_DISABLE) / OOM_SCORE_ADJ_MAX;
1104 adj = CLAMP(adj, OOM_DISABLE, OOM_ADJUST_MAX);
1106 snprintf(t, sizeof(t), "%i", adj);
1109 if (write_one_line_file("/proc/self/oom_adj", t) < 0
1110 && errno != EACCES) {
1111 r = EXIT_OOM_ADJUST;
1117 if (context->nice_set)
1118 if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
1123 if (context->cpu_sched_set) {
1124 struct sched_param param;
1127 param.sched_priority = context->cpu_sched_priority;
1129 if (sched_setscheduler(0, context->cpu_sched_policy |
1130 (context->cpu_sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0), ¶m) < 0) {
1131 r = EXIT_SETSCHEDULER;
1136 if (context->cpuset)
1137 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
1138 r = EXIT_CPUAFFINITY;
1142 if (context->ioprio_set)
1143 if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
1148 if (context->timer_slack_nsec_set)
1149 if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
1150 r = EXIT_TIMERSLACK;
1154 if (context->utmp_id)
1155 utmp_put_init_process(0, context->utmp_id, getpid(), getsid(0), context->tty_path);
1157 if (context->user) {
1158 username = context->user;
1159 if (get_user_creds(&username, &uid, &gid, &home) < 0) {
1164 if (is_terminal_input(context->std_input))
1165 if (chown_terminal(STDIN_FILENO, uid) < 0) {
1172 if (context->pam_name && username) {
1173 if (setup_pam(context->pam_name, username, context->tty_path, &pam_env, fds, n_fds) < 0) {
1180 if (apply_permissions)
1181 if (enforce_groups(context, username, uid) < 0) {
1186 umask(context->umask);
1188 if (strv_length(context->read_write_dirs) > 0 ||
1189 strv_length(context->read_only_dirs) > 0 ||
1190 strv_length(context->inaccessible_dirs) > 0 ||
1191 context->mount_flags != MS_SHARED ||
1192 context->private_tmp)
1193 if ((r = setup_namespace(
1194 context->read_write_dirs,
1195 context->read_only_dirs,
1196 context->inaccessible_dirs,
1197 context->private_tmp,
1198 context->mount_flags)) < 0)
1202 if (context->root_directory)
1203 if (chroot(context->root_directory) < 0) {
1208 if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
1216 if (asprintf(&d, "%s/%s",
1217 context->root_directory ? context->root_directory : "",
1218 context->working_directory ? context->working_directory : "") < 0) {
1232 /* We repeat the fd closing here, to make sure that
1233 * nothing is leaked from the PAM modules */
1234 if (close_all_fds(fds, n_fds) < 0 ||
1235 shift_fds(fds, n_fds) < 0 ||
1236 flags_fds(fds, n_fds, context->non_blocking) < 0) {
1241 if (apply_permissions) {
1243 for (i = 0; i < RLIMIT_NLIMITS; i++) {
1244 if (!context->rlimit[i])
1247 if (setrlimit(i, context->rlimit[i]) < 0) {
1253 if (context->capability_bounding_set_drop)
1254 for (i = 0; i <= CAP_LAST_CAP; i++)
1255 if (context->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) i)) {
1256 if (prctl(PR_CAPBSET_DROP, i) < 0) {
1257 r = EXIT_CAPABILITIES;
1263 if (enforce_user(context, uid) < 0) {
1268 /* PR_GET_SECUREBITS is not privileged, while
1269 * PR_SET_SECUREBITS is. So to suppress
1270 * potential EPERMs we'll try not to call
1271 * PR_SET_SECUREBITS unless necessary. */
1272 if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
1273 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
1274 r = EXIT_SECUREBITS;
1278 if (context->capabilities)
1279 if (cap_set_proc(context->capabilities) < 0) {
1280 r = EXIT_CAPABILITIES;
1285 if (!(our_env = new0(char*, 7))) {
1291 if (asprintf(our_env + n_env++, "LISTEN_PID=%lu", (unsigned long) getpid()) < 0 ||
1292 asprintf(our_env + n_env++, "LISTEN_FDS=%u", n_fds) < 0) {
1298 if (asprintf(our_env + n_env++, "HOME=%s", home) < 0) {
1304 if (asprintf(our_env + n_env++, "LOGNAME=%s", username) < 0 ||
1305 asprintf(our_env + n_env++, "USER=%s", username) < 0) {
1310 if (is_terminal_input(context->std_input) ||
1311 context->std_output == EXEC_OUTPUT_TTY ||
1312 context->std_error == EXEC_OUTPUT_TTY)
1313 if (!(our_env[n_env++] = strdup(default_term_for_tty(tty_path(context))))) {
1320 if (!(final_env = strv_env_merge(
1324 context->environment,
1332 if (!(final_argv = replace_env_argv(argv, final_env))) {
1337 final_env = strv_env_clean(final_env);
1339 execve(command->path, final_argv, final_env);
1344 strv_free(final_env);
1346 strv_free(files_env);
1347 strv_free(final_argv);
1349 if (saved_stdin >= 0)
1350 close_nointr_nofail(saved_stdin);
1352 if (saved_stdout >= 0)
1353 close_nointr_nofail(saved_stdout);
1358 strv_free(files_env);
1360 /* We add the new process to the cgroup both in the child (so
1361 * that we can be sure that no user code is ever executed
1362 * outside of the cgroup) and in the parent (so that we can be
1363 * sure that when we kill the cgroup the process will be
1365 if (cgroup_bondings)
1366 cgroup_bonding_install_list(cgroup_bondings, pid);
1368 log_debug("Forked %s as %lu", command->path, (unsigned long) pid);
1370 exec_status_start(&command->exec_status, pid);
1376 strv_free(files_env);
1381 void exec_context_init(ExecContext *c) {
1385 c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
1386 c->cpu_sched_policy = SCHED_OTHER;
1387 c->syslog_priority = LOG_DAEMON|LOG_INFO;
1388 c->syslog_level_prefix = true;
1389 c->mount_flags = MS_SHARED;
1390 c->kill_signal = SIGTERM;
1391 c->send_sigkill = true;
1394 void exec_context_done(ExecContext *c) {
1399 strv_free(c->environment);
1400 c->environment = NULL;
1402 strv_free(c->environment_files);
1403 c->environment_files = NULL;
1405 for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1407 c->rlimit[l] = NULL;
1410 free(c->working_directory);
1411 c->working_directory = NULL;
1412 free(c->root_directory);
1413 c->root_directory = NULL;
1418 free(c->tcpwrap_name);
1419 c->tcpwrap_name = NULL;
1421 free(c->syslog_identifier);
1422 c->syslog_identifier = NULL;
1430 strv_free(c->supplementary_groups);
1431 c->supplementary_groups = NULL;
1436 if (c->capabilities) {
1437 cap_free(c->capabilities);
1438 c->capabilities = NULL;
1441 strv_free(c->read_only_dirs);
1442 c->read_only_dirs = NULL;
1444 strv_free(c->read_write_dirs);
1445 c->read_write_dirs = NULL;
1447 strv_free(c->inaccessible_dirs);
1448 c->inaccessible_dirs = NULL;
1451 CPU_FREE(c->cpuset);
1457 void exec_command_done(ExecCommand *c) {
1467 void exec_command_done_array(ExecCommand *c, unsigned n) {
1470 for (i = 0; i < n; i++)
1471 exec_command_done(c+i);
1474 void exec_command_free_list(ExecCommand *c) {
1478 LIST_REMOVE(ExecCommand, command, c, i);
1479 exec_command_done(i);
1484 void exec_command_free_array(ExecCommand **c, unsigned n) {
1487 for (i = 0; i < n; i++) {
1488 exec_command_free_list(c[i]);
1493 int exec_context_load_environment(const ExecContext *c, char ***l) {
1494 char **i, **r = NULL;
1499 STRV_FOREACH(i, c->environment_files) {
1502 bool ignore = false;
1512 if (!path_is_absolute(fn)) {
1521 if ((k = load_env_file(fn, &p)) < 0) {
1535 m = strv_env_merge(2, r, p);
1551 static void strv_fprintf(FILE *f, char **l) {
1557 fprintf(f, " %s", *g);
1560 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
1572 "%sWorkingDirectory: %s\n"
1573 "%sRootDirectory: %s\n"
1574 "%sNonBlocking: %s\n"
1575 "%sPrivateTmp: %s\n",
1577 prefix, c->working_directory ? c->working_directory : "/",
1578 prefix, c->root_directory ? c->root_directory : "/",
1579 prefix, yes_no(c->non_blocking),
1580 prefix, yes_no(c->private_tmp));
1582 STRV_FOREACH(e, c->environment)
1583 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
1585 STRV_FOREACH(e, c->environment_files)
1586 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
1588 if (c->tcpwrap_name)
1590 "%sTCPWrapName: %s\n",
1591 prefix, c->tcpwrap_name);
1598 if (c->oom_score_adjust_set)
1600 "%sOOMScoreAdjust: %i\n",
1601 prefix, c->oom_score_adjust);
1603 for (i = 0; i < RLIM_NLIMITS; i++)
1605 fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
1609 "%sIOSchedulingClass: %s\n"
1610 "%sIOPriority: %i\n",
1611 prefix, ioprio_class_to_string(IOPRIO_PRIO_CLASS(c->ioprio)),
1612 prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
1614 if (c->cpu_sched_set)
1616 "%sCPUSchedulingPolicy: %s\n"
1617 "%sCPUSchedulingPriority: %i\n"
1618 "%sCPUSchedulingResetOnFork: %s\n",
1619 prefix, sched_policy_to_string(c->cpu_sched_policy),
1620 prefix, c->cpu_sched_priority,
1621 prefix, yes_no(c->cpu_sched_reset_on_fork));
1624 fprintf(f, "%sCPUAffinity:", prefix);
1625 for (i = 0; i < c->cpuset_ncpus; i++)
1626 if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
1627 fprintf(f, " %i", i);
1631 if (c->timer_slack_nsec_set)
1632 fprintf(f, "%sTimerSlackNSec: %lu\n", prefix, c->timer_slack_nsec);
1635 "%sStandardInput: %s\n"
1636 "%sStandardOutput: %s\n"
1637 "%sStandardError: %s\n",
1638 prefix, exec_input_to_string(c->std_input),
1639 prefix, exec_output_to_string(c->std_output),
1640 prefix, exec_output_to_string(c->std_error));
1645 prefix, c->tty_path);
1647 if (c->std_output == EXEC_OUTPUT_SYSLOG || c->std_output == EXEC_OUTPUT_KMSG ||
1648 c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
1649 c->std_error == EXEC_OUTPUT_SYSLOG || c->std_error == EXEC_OUTPUT_KMSG ||
1650 c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE)
1652 "%sSyslogFacility: %s\n"
1653 "%sSyslogLevel: %s\n",
1654 prefix, log_facility_unshifted_to_string(c->syslog_priority >> 3),
1655 prefix, log_level_to_string(LOG_PRI(c->syslog_priority)));
1657 if (c->capabilities) {
1659 if ((t = cap_to_text(c->capabilities, NULL))) {
1660 fprintf(f, "%sCapabilities: %s\n",
1667 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
1669 (c->secure_bits & SECURE_KEEP_CAPS) ? " keep-caps" : "",
1670 (c->secure_bits & SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
1671 (c->secure_bits & SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
1672 (c->secure_bits & SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
1673 (c->secure_bits & SECURE_NOROOT) ? " noroot" : "",
1674 (c->secure_bits & SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
1676 if (c->capability_bounding_set_drop) {
1677 fprintf(f, "%sCapabilityBoundingSet:", prefix);
1679 for (i = 0; i <= CAP_LAST_CAP; i++)
1680 if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) i))) {
1683 if ((t = cap_to_name(i))) {
1684 fprintf(f, " %s", t);
1693 fprintf(f, "%sUser: %s\n", prefix, c->user);
1695 fprintf(f, "%sGroup: %s\n", prefix, c->group);
1697 if (strv_length(c->supplementary_groups) > 0) {
1698 fprintf(f, "%sSupplementaryGroups:", prefix);
1699 strv_fprintf(f, c->supplementary_groups);
1704 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
1706 if (strv_length(c->read_write_dirs) > 0) {
1707 fprintf(f, "%sReadWriteDirs:", prefix);
1708 strv_fprintf(f, c->read_write_dirs);
1712 if (strv_length(c->read_only_dirs) > 0) {
1713 fprintf(f, "%sReadOnlyDirs:", prefix);
1714 strv_fprintf(f, c->read_only_dirs);
1718 if (strv_length(c->inaccessible_dirs) > 0) {
1719 fprintf(f, "%sInaccessibleDirs:", prefix);
1720 strv_fprintf(f, c->inaccessible_dirs);
1726 "%sKillSignal: SIG%s\n"
1727 "%sSendSIGKILL: %s\n",
1728 prefix, kill_mode_to_string(c->kill_mode),
1729 prefix, signal_to_string(c->kill_signal),
1730 prefix, yes_no(c->send_sigkill));
1734 "%sUtmpIdentifier: %s\n",
1735 prefix, c->utmp_id);
1738 void exec_status_start(ExecStatus *s, pid_t pid) {
1743 dual_timestamp_get(&s->start_timestamp);
1746 void exec_status_exit(ExecStatus *s, pid_t pid, int code, int status, const char *utmp_id) {
1749 if ((s->pid && s->pid != pid) ||
1750 !s->start_timestamp.realtime <= 0)
1754 dual_timestamp_get(&s->exit_timestamp);
1760 utmp_put_dead_process(utmp_id, pid, code, status);
1763 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
1764 char buf[FORMAT_TIMESTAMP_MAX];
1777 prefix, (unsigned long) s->pid);
1779 if (s->start_timestamp.realtime > 0)
1781 "%sStart Timestamp: %s\n",
1782 prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
1784 if (s->exit_timestamp.realtime > 0)
1786 "%sExit Timestamp: %s\n"
1788 "%sExit Status: %i\n",
1789 prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
1790 prefix, sigchld_code_to_string(s->code),
1794 char *exec_command_line(char **argv) {
1802 STRV_FOREACH(a, argv)
1805 if (!(n = new(char, k)))
1809 STRV_FOREACH(a, argv) {
1816 if (strpbrk(*a, WHITESPACE)) {
1827 /* FIXME: this doesn't really handle arguments that have
1828 * spaces and ticks in them */
1833 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
1835 const char *prefix2;
1844 p2 = strappend(prefix, "\t");
1845 prefix2 = p2 ? p2 : prefix;
1847 cmd = exec_command_line(c->argv);
1850 "%sCommand Line: %s\n",
1851 prefix, cmd ? cmd : strerror(ENOMEM));
1855 exec_status_dump(&c->exec_status, f, prefix2);
1860 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
1866 LIST_FOREACH(command, c, c)
1867 exec_command_dump(c, f, prefix);
1870 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
1877 /* It's kind of important, that we keep the order here */
1878 LIST_FIND_TAIL(ExecCommand, command, *l, end);
1879 LIST_INSERT_AFTER(ExecCommand, command, *l, end, e);
1884 int exec_command_set(ExecCommand *c, const char *path, ...) {
1892 l = strv_new_ap(path, ap);
1898 if (!(p = strdup(path))) {
1912 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
1913 [EXEC_INPUT_NULL] = "null",
1914 [EXEC_INPUT_TTY] = "tty",
1915 [EXEC_INPUT_TTY_FORCE] = "tty-force",
1916 [EXEC_INPUT_TTY_FAIL] = "tty-fail",
1917 [EXEC_INPUT_SOCKET] = "socket"
1920 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
1922 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
1923 [EXEC_OUTPUT_INHERIT] = "inherit",
1924 [EXEC_OUTPUT_NULL] = "null",
1925 [EXEC_OUTPUT_TTY] = "tty",
1926 [EXEC_OUTPUT_SYSLOG] = "syslog",
1927 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
1928 [EXEC_OUTPUT_KMSG] = "kmsg",
1929 [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
1930 [EXEC_OUTPUT_SOCKET] = "socket"
1933 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
1935 static const char* const kill_mode_table[_KILL_MODE_MAX] = {
1936 [KILL_CONTROL_GROUP] = "control-group",
1937 [KILL_PROCESS] = "process",
1938 [KILL_NONE] = "none"
1941 DEFINE_STRING_TABLE_LOOKUP(kill_mode, KillMode);
1943 static const char* const kill_who_table[_KILL_WHO_MAX] = {
1944 [KILL_MAIN] = "main",
1945 [KILL_CONTROL] = "control",
1949 DEFINE_STRING_TABLE_LOOKUP(kill_who, KillWho);