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 Lesser General Public License as published by
10 the Free Software Foundation; either version 2.1 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 Lesser General Public License for more details.
18 You should have received a copy of the GNU Lesser 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>
43 #include <security/pam_appl.h>
49 #include "capability.h"
53 #include "securebits.h"
55 #include "namespace.h"
57 #include "exit-status.h"
59 #include "utmp-wtmp.h"
61 #include "loopback-setup.h"
62 #include "path-util.h"
64 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
66 /* This assumes there is a 'tty' group */
69 static int shift_fds(int fds[], unsigned n_fds) {
70 int start, restart_from;
75 /* Modifies the fds array! (sorts it) */
85 for (i = start; i < (int) n_fds; i++) {
88 /* Already at right index? */
92 if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
95 close_nointr_nofail(fds[i]);
98 /* Hmm, the fd we wanted isn't free? Then
99 * let's remember that and try again from here*/
100 if (nfd != i+3 && restart_from < 0)
104 if (restart_from < 0)
107 start = restart_from;
113 static int flags_fds(const int fds[], unsigned n_fds, bool nonblock) {
122 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
124 for (i = 0; i < n_fds; i++) {
126 if ((r = fd_nonblock(fds[i], nonblock)) < 0)
129 /* We unconditionally drop FD_CLOEXEC from the fds,
130 * since after all we want to pass these fds to our
133 if ((r = fd_cloexec(fds[i], false)) < 0)
140 static const char *tty_path(const ExecContext *context) {
143 if (context->tty_path)
144 return context->tty_path;
146 return "/dev/console";
149 void exec_context_tty_reset(const ExecContext *context) {
152 if (context->tty_vhangup)
153 terminal_vhangup(tty_path(context));
155 if (context->tty_reset)
156 reset_terminal(tty_path(context));
158 if (context->tty_vt_disallocate && context->tty_path)
159 vt_disallocate(context->tty_path);
162 static int open_null_as(int flags, int nfd) {
167 if ((fd = open("/dev/null", flags|O_NOCTTY)) < 0)
171 r = dup2(fd, nfd) < 0 ? -errno : nfd;
172 close_nointr_nofail(fd);
179 static int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, int nfd) {
181 union sockaddr_union sa;
184 assert(output < _EXEC_OUTPUT_MAX);
188 fd = socket(AF_UNIX, SOCK_STREAM, 0);
193 sa.un.sun_family = AF_UNIX;
194 strncpy(sa.un.sun_path, "/run/systemd/journal/stdout", sizeof(sa.un.sun_path));
196 r = connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path));
198 close_nointr_nofail(fd);
202 if (shutdown(fd, SHUT_RD) < 0) {
203 close_nointr_nofail(fd);
214 context->syslog_identifier ? context->syslog_identifier : ident,
215 context->syslog_priority,
216 !!context->syslog_level_prefix,
217 output == EXEC_OUTPUT_SYSLOG || output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE,
218 output == EXEC_OUTPUT_KMSG || output == EXEC_OUTPUT_KMSG_AND_CONSOLE,
219 output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || output == EXEC_OUTPUT_KMSG_AND_CONSOLE || output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE);
222 r = dup2(fd, nfd) < 0 ? -errno : nfd;
223 close_nointr_nofail(fd);
229 static int open_terminal_as(const char *path, mode_t mode, int nfd) {
235 if ((fd = open_terminal(path, mode | O_NOCTTY)) < 0)
239 r = dup2(fd, nfd) < 0 ? -errno : nfd;
240 close_nointr_nofail(fd);
247 static bool is_terminal_input(ExecInput i) {
249 i == EXEC_INPUT_TTY ||
250 i == EXEC_INPUT_TTY_FORCE ||
251 i == EXEC_INPUT_TTY_FAIL;
254 static int fixup_input(ExecInput std_input, int socket_fd, bool apply_tty_stdin) {
256 if (is_terminal_input(std_input) && !apply_tty_stdin)
257 return EXEC_INPUT_NULL;
259 if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0)
260 return EXEC_INPUT_NULL;
265 static int fixup_output(ExecOutput std_output, int socket_fd) {
267 if (std_output == EXEC_OUTPUT_SOCKET && socket_fd < 0)
268 return EXEC_OUTPUT_INHERIT;
273 static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty_stdin) {
278 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
282 case EXEC_INPUT_NULL:
283 return open_null_as(O_RDONLY, STDIN_FILENO);
286 case EXEC_INPUT_TTY_FORCE:
287 case EXEC_INPUT_TTY_FAIL: {
290 if ((fd = acquire_terminal(
292 i == EXEC_INPUT_TTY_FAIL,
293 i == EXEC_INPUT_TTY_FORCE,
297 if (fd != STDIN_FILENO) {
298 r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
299 close_nointr_nofail(fd);
306 case EXEC_INPUT_SOCKET:
307 return dup2(socket_fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
310 assert_not_reached("Unknown input type");
314 static int setup_output(const ExecContext *context, int socket_fd, const char *ident, bool apply_tty_stdin) {
321 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
322 o = fixup_output(context->std_output, socket_fd);
324 /* This expects the input is already set up */
328 case EXEC_OUTPUT_INHERIT:
330 /* If input got downgraded, inherit the original value */
331 if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input))
332 return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO);
334 /* If the input is connected to anything that's not a /dev/null, inherit that... */
335 if (i != EXEC_INPUT_NULL)
336 return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
338 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
340 return STDOUT_FILENO;
342 /* We need to open /dev/null here anew, to get the
343 * right access mode. So we fall through */
345 case EXEC_OUTPUT_NULL:
346 return open_null_as(O_WRONLY, STDOUT_FILENO);
348 case EXEC_OUTPUT_TTY:
349 if (is_terminal_input(i))
350 return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
352 /* We don't reset the terminal if this is just about output */
353 return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO);
355 case EXEC_OUTPUT_SYSLOG:
356 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
357 case EXEC_OUTPUT_KMSG:
358 case EXEC_OUTPUT_KMSG_AND_CONSOLE:
359 case EXEC_OUTPUT_JOURNAL:
360 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE:
361 return connect_logger_as(context, o, ident, STDOUT_FILENO);
363 case EXEC_OUTPUT_SOCKET:
364 assert(socket_fd >= 0);
365 return dup2(socket_fd, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
368 assert_not_reached("Unknown output type");
372 static int setup_error(const ExecContext *context, int socket_fd, const char *ident, bool apply_tty_stdin) {
379 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
380 o = fixup_output(context->std_output, socket_fd);
381 e = fixup_output(context->std_error, socket_fd);
383 /* This expects the input and output are already set up */
385 /* Don't change the stderr file descriptor if we inherit all
386 * the way and are not on a tty */
387 if (e == EXEC_OUTPUT_INHERIT &&
388 o == EXEC_OUTPUT_INHERIT &&
389 i == EXEC_INPUT_NULL &&
390 !is_terminal_input(context->std_input) &&
392 return STDERR_FILENO;
394 /* Duplicate from stdout if possible */
395 if (e == o || e == EXEC_OUTPUT_INHERIT)
396 return dup2(STDOUT_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
400 case EXEC_OUTPUT_NULL:
401 return open_null_as(O_WRONLY, STDERR_FILENO);
403 case EXEC_OUTPUT_TTY:
404 if (is_terminal_input(i))
405 return dup2(STDIN_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
407 /* We don't reset the terminal if this is just about output */
408 return open_terminal_as(tty_path(context), O_WRONLY, STDERR_FILENO);
410 case EXEC_OUTPUT_SYSLOG:
411 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
412 case EXEC_OUTPUT_KMSG:
413 case EXEC_OUTPUT_KMSG_AND_CONSOLE:
414 case EXEC_OUTPUT_JOURNAL:
415 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE:
416 return connect_logger_as(context, e, ident, STDERR_FILENO);
418 case EXEC_OUTPUT_SOCKET:
419 assert(socket_fd >= 0);
420 return dup2(socket_fd, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
423 assert_not_reached("Unknown error type");
427 static int chown_terminal(int fd, uid_t uid) {
432 /* This might fail. What matters are the results. */
433 (void) fchown(fd, uid, -1);
434 (void) fchmod(fd, TTY_MODE);
436 if (fstat(fd, &st) < 0)
439 if (st.st_uid != uid || (st.st_mode & 0777) != TTY_MODE)
445 static int setup_confirm_stdio(const ExecContext *context,
447 int *_saved_stdout) {
448 int fd = -1, saved_stdin, saved_stdout = -1, r;
451 assert(_saved_stdin);
452 assert(_saved_stdout);
454 /* This returns positive EXIT_xxx return values instead of
455 * negative errno style values! */
457 if ((saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3)) < 0)
460 if ((saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3)) < 0) {
465 if ((fd = acquire_terminal(
467 context->std_input == EXEC_INPUT_TTY_FAIL,
468 context->std_input == EXEC_INPUT_TTY_FORCE,
474 if (chown_terminal(fd, getuid()) < 0) {
479 if (dup2(fd, STDIN_FILENO) < 0) {
484 if (dup2(fd, STDOUT_FILENO) < 0) {
490 close_nointr_nofail(fd);
492 *_saved_stdin = saved_stdin;
493 *_saved_stdout = saved_stdout;
498 if (saved_stdout >= 0)
499 close_nointr_nofail(saved_stdout);
501 if (saved_stdin >= 0)
502 close_nointr_nofail(saved_stdin);
505 close_nointr_nofail(fd);
510 static int restore_confirm_stdio(const ExecContext *context,
518 assert(*saved_stdin >= 0);
519 assert(saved_stdout);
520 assert(*saved_stdout >= 0);
522 /* This returns positive EXIT_xxx return values instead of
523 * negative errno style values! */
525 if (is_terminal_input(context->std_input)) {
527 /* The service wants terminal input. */
531 context->std_output == EXEC_OUTPUT_INHERIT ||
532 context->std_output == EXEC_OUTPUT_TTY;
535 /* If the service doesn't want a controlling terminal,
536 * then we need to get rid entirely of what we have
539 if (release_terminal() < 0)
542 if (dup2(*saved_stdin, STDIN_FILENO) < 0)
545 if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
548 *keep_stdout = *keep_stdin = false;
554 static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
555 bool keep_groups = false;
560 /* Lookup and set GID and supplementary group list. Here too
561 * we avoid NSS lookups for gid=0. */
563 if (context->group || username) {
565 if (context->group) {
566 const char *g = context->group;
568 if ((r = get_group_creds(&g, &gid)) < 0)
572 /* First step, initialize groups from /etc/groups */
573 if (username && gid != 0) {
574 if (initgroups(username, gid) < 0)
580 /* Second step, set our gids */
581 if (setresgid(gid, gid, gid) < 0)
585 if (context->supplementary_groups) {
590 /* Final step, initialize any manually set supplementary groups */
591 assert_se((ngroups_max = (int) sysconf(_SC_NGROUPS_MAX)) > 0);
593 if (!(gids = new(gid_t, ngroups_max)))
597 if ((k = getgroups(ngroups_max, gids)) < 0) {
604 STRV_FOREACH(i, context->supplementary_groups) {
607 if (k >= ngroups_max) {
613 r = get_group_creds(&g, gids+k);
622 if (setgroups(k, gids) < 0) {
633 static int enforce_user(const ExecContext *context, uid_t uid) {
637 /* Sets (but doesn't lookup) the uid and make sure we keep the
638 * capabilities while doing so. */
640 if (context->capabilities) {
642 static const cap_value_t bits[] = {
643 CAP_SETUID, /* Necessary so that we can run setresuid() below */
644 CAP_SETPCAP /* Necessary so that we can set PR_SET_SECUREBITS later on */
647 /* First step: If we need to keep capabilities but
648 * drop privileges we need to make sure we keep our
649 * caps, whiel we drop privileges. */
651 int sb = context->secure_bits|SECURE_KEEP_CAPS;
653 if (prctl(PR_GET_SECUREBITS) != sb)
654 if (prctl(PR_SET_SECUREBITS, sb) < 0)
658 /* Second step: set the capabilities. This will reduce
659 * the capabilities to the minimum we need. */
661 if (!(d = cap_dup(context->capabilities)))
664 if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
665 cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0) {
671 if (cap_set_proc(d) < 0) {
680 /* Third step: actually set the uids */
681 if (setresuid(uid, uid, uid) < 0)
684 /* At this point we should have all necessary capabilities but
685 are otherwise a normal user. However, the caps might got
686 corrupted due to the setresuid() so we need clean them up
687 later. This is done outside of this call. */
694 static int null_conv(
696 const struct pam_message **msg,
697 struct pam_response **resp,
700 /* We don't support conversations */
705 static int setup_pam(
711 int fds[], unsigned n_fds) {
713 static const struct pam_conv conv = {
718 pam_handle_t *handle = NULL;
720 int pam_code = PAM_SUCCESS;
723 bool close_session = false;
724 pid_t pam_pid = 0, parent_pid;
730 /* We set up PAM in the parent process, then fork. The child
731 * will then stay around until killed via PR_GET_PDEATHSIG or
732 * systemd via the cgroup logic. It will then remove the PAM
733 * session again. The parent process will exec() the actual
734 * daemon. We do things this way to ensure that the main PID
735 * of the daemon is the one we initially fork()ed. */
737 if ((pam_code = pam_start(name, user, &conv, &handle)) != PAM_SUCCESS) {
743 if ((pam_code = pam_set_item(handle, PAM_TTY, tty)) != PAM_SUCCESS)
746 if ((pam_code = pam_acct_mgmt(handle, PAM_SILENT)) != PAM_SUCCESS)
749 if ((pam_code = pam_open_session(handle, PAM_SILENT)) != PAM_SUCCESS)
752 close_session = true;
754 if ((!(e = pam_getenvlist(handle)))) {
755 pam_code = PAM_BUF_ERR;
759 /* Block SIGTERM, so that we know that it won't get lost in
761 if (sigemptyset(&ss) < 0 ||
762 sigaddset(&ss, SIGTERM) < 0 ||
763 sigprocmask(SIG_BLOCK, &ss, &old_ss) < 0)
766 parent_pid = getpid();
768 if ((pam_pid = fork()) < 0)
775 /* The child's job is to reset the PAM session on
778 /* This string must fit in 10 chars (i.e. the length
779 * of "/sbin/init"), to look pretty in /bin/ps */
780 rename_process("(sd-pam)");
782 /* Make sure we don't keep open the passed fds in this
783 child. We assume that otherwise only those fds are
784 open here that have been opened by PAM. */
785 close_many(fds, n_fds);
787 /* Drop privileges - we don't need any to pam_close_session
788 * and this will make PR_SET_PDEATHSIG work in most cases.
789 * If this fails, ignore the error - but expect sd-pam threads
790 * to fail to exit normally */
791 if (setresuid(uid, uid, uid) < 0)
792 log_error("Error: Failed to setresuid() in sd-pam: %s", strerror(-r));
794 /* Wait until our parent died. This will only work if
795 * the above setresuid() succeeds, otherwise the kernel
796 * will not allow unprivileged parents kill their privileged
797 * children this way. We rely on the control groups kill logic
798 * to do the rest for us. */
799 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
802 /* Check if our parent process might already have
804 if (getppid() == parent_pid) {
806 if (sigwait(&ss, &sig) < 0) {
813 assert(sig == SIGTERM);
818 /* If our parent died we'll end the session */
819 if (getppid() != parent_pid)
820 if ((pam_code = pam_close_session(handle, PAM_DATA_SILENT)) != PAM_SUCCESS)
826 pam_end(handle, pam_code | PAM_DATA_SILENT);
830 /* If the child was forked off successfully it will do all the
831 * cleanups, so forget about the handle here. */
834 /* Unblock SIGTERM again in the parent */
835 if (sigprocmask(SIG_SETMASK, &old_ss, NULL) < 0)
838 /* We close the log explicitly here, since the PAM modules
839 * might have opened it, but we don't want this fd around. */
848 if (pam_code != PAM_SUCCESS)
849 err = -EPERM; /* PAM errors do not map to errno */
855 pam_code = pam_close_session(handle, PAM_DATA_SILENT);
857 pam_end(handle, pam_code | PAM_DATA_SILENT);
865 kill(pam_pid, SIGTERM);
866 kill(pam_pid, SIGCONT);
873 static void rename_process_from_path(const char *path) {
874 char process_name[11];
878 /* This resulting string must fit in 10 chars (i.e. the length
879 * of "/sbin/init") to look pretty in /bin/ps */
881 p = path_get_file_name(path);
883 rename_process("(...)");
889 /* The end of the process name is usually more
890 * interesting, since the first bit might just be
896 process_name[0] = '(';
897 memcpy(process_name+1, p, l);
898 process_name[1+l] = ')';
899 process_name[1+l+1] = 0;
901 rename_process(process_name);
904 int exec_spawn(ExecCommand *command,
906 const ExecContext *context,
907 int fds[], unsigned n_fds,
909 bool apply_permissions,
911 bool apply_tty_stdin,
913 CGroupBonding *cgroup_bondings,
914 CGroupAttribute *cgroup_attributes,
915 const char *cgroup_suffix,
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);
960 r = cgroup_bonding_realize_list(cgroup_bondings);
964 cgroup_attribute_apply_list(cgroup_attributes, cgroup_bondings);
966 if ((pid = fork()) < 0) {
974 const char *username = NULL, *home = NULL;
975 uid_t uid = (uid_t) -1;
976 gid_t gid = (gid_t) -1;
977 char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
979 int saved_stdout = -1, saved_stdin = -1;
980 bool keep_stdout = false, keep_stdin = false, set_access = false;
984 rename_process_from_path(command->path);
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 (context->ignore_sigpipe)
995 ignore_signals(SIGPIPE, -1);
997 assert_se(sigemptyset(&ss) == 0);
998 if (sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
1000 r = EXIT_SIGNAL_MASK;
1005 if (idle_pipe[1] >= 0)
1006 close_nointr_nofail(idle_pipe[1]);
1007 if (idle_pipe[0] >= 0) {
1008 fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT_USEC);
1009 close_nointr_nofail(idle_pipe[0]);
1013 /* Close sockets very early to make sure we don't
1014 * block init reexecution because it cannot bind its
1017 err = close_all_fds(socket_fd >= 0 ? &socket_fd : fds,
1018 socket_fd >= 0 ? 1 : n_fds);
1024 if (!context->same_pgrp)
1031 if (context->tcpwrap_name) {
1033 if (!socket_tcpwrap(socket_fd, context->tcpwrap_name)) {
1039 for (i = 0; i < (int) n_fds; i++) {
1040 if (!socket_tcpwrap(fds[i], context->tcpwrap_name)) {
1048 exec_context_tty_reset(context);
1050 /* We skip the confirmation step if we shall not apply the TTY */
1051 if (confirm_spawn &&
1052 (!is_terminal_input(context->std_input) || apply_tty_stdin)) {
1055 /* Set up terminal for the question */
1056 if ((r = setup_confirm_stdio(context,
1057 &saved_stdin, &saved_stdout))) {
1062 /* Now ask the question. */
1063 if (!(line = exec_command_line(argv))) {
1069 r = ask(&response, "yns", "Execute %s? [Yes, No, Skip] ", line);
1072 if (r < 0 || response == 'n') {
1076 } else if (response == 's') {
1081 /* Release terminal for the question */
1082 if ((r = restore_confirm_stdio(context,
1083 &saved_stdin, &saved_stdout,
1084 &keep_stdin, &keep_stdout))) {
1090 /* If a socket is connected to STDIN/STDOUT/STDERR, we
1091 * must sure to drop O_NONBLOCK */
1093 fd_nonblock(socket_fd, false);
1096 err = setup_input(context, socket_fd, apply_tty_stdin);
1104 err = setup_output(context, socket_fd, path_get_file_name(command->path), apply_tty_stdin);
1111 err = setup_error(context, socket_fd, path_get_file_name(command->path), apply_tty_stdin);
1117 if (cgroup_bondings) {
1118 err = cgroup_bonding_install_list(cgroup_bondings, 0, cgroup_suffix);
1125 if (context->oom_score_adjust_set) {
1128 snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
1131 if (write_one_line_file("/proc/self/oom_score_adj", t) < 0) {
1132 /* Compatibility with Linux <= 2.6.35 */
1136 adj = (context->oom_score_adjust * -OOM_DISABLE) / OOM_SCORE_ADJ_MAX;
1137 adj = CLAMP(adj, OOM_DISABLE, OOM_ADJUST_MAX);
1139 snprintf(t, sizeof(t), "%i", adj);
1142 if (write_one_line_file("/proc/self/oom_adj", t) < 0
1143 && errno != EACCES) {
1145 r = EXIT_OOM_ADJUST;
1151 if (context->nice_set)
1152 if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
1158 if (context->cpu_sched_set) {
1159 struct sched_param param;
1162 param.sched_priority = context->cpu_sched_priority;
1164 if (sched_setscheduler(0, context->cpu_sched_policy |
1165 (context->cpu_sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0), ¶m) < 0) {
1167 r = EXIT_SETSCHEDULER;
1172 if (context->cpuset)
1173 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
1175 r = EXIT_CPUAFFINITY;
1179 if (context->ioprio_set)
1180 if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
1186 if (context->timer_slack_nsec_set)
1187 if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
1189 r = EXIT_TIMERSLACK;
1193 if (context->utmp_id)
1194 utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path);
1196 if (context->user) {
1197 username = context->user;
1198 err = get_user_creds(&username, &uid, &gid, &home);
1204 if (is_terminal_input(context->std_input)) {
1205 err = chown_terminal(STDIN_FILENO, uid);
1212 if (cgroup_bondings && context->control_group_modify) {
1213 err = cgroup_bonding_set_group_access_list(cgroup_bondings, 0755, uid, gid);
1215 err = cgroup_bonding_set_task_access_list(cgroup_bondings, 0644, uid, gid, context->control_group_persistent);
1225 if (cgroup_bondings && !set_access && context->control_group_persistent >= 0) {
1226 err = cgroup_bonding_set_task_access_list(cgroup_bondings, (mode_t) -1, (uid_t) -1, (uid_t) -1, context->control_group_persistent);
1233 if (apply_permissions) {
1234 err = enforce_groups(context, username, gid);
1241 umask(context->umask);
1244 if (context->pam_name && username) {
1245 err = setup_pam(context->pam_name, username, uid, context->tty_path, &pam_env, fds, n_fds);
1252 if (context->private_network) {
1253 if (unshare(CLONE_NEWNET) < 0) {
1262 if (strv_length(context->read_write_dirs) > 0 ||
1263 strv_length(context->read_only_dirs) > 0 ||
1264 strv_length(context->inaccessible_dirs) > 0 ||
1265 context->mount_flags != MS_SHARED ||
1266 context->private_tmp) {
1267 err = setup_namespace(context->read_write_dirs,
1268 context->read_only_dirs,
1269 context->inaccessible_dirs,
1270 context->private_tmp,
1271 context->mount_flags);
1279 if (context->root_directory)
1280 if (chroot(context->root_directory) < 0) {
1286 if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
1295 if (asprintf(&d, "%s/%s",
1296 context->root_directory ? context->root_directory : "",
1297 context->working_directory ? context->working_directory : "") < 0) {
1313 /* We repeat the fd closing here, to make sure that
1314 * nothing is leaked from the PAM modules */
1315 err = close_all_fds(fds, n_fds);
1317 err = shift_fds(fds, n_fds);
1319 err = flags_fds(fds, n_fds, context->non_blocking);
1325 if (apply_permissions) {
1327 for (i = 0; i < RLIMIT_NLIMITS; i++) {
1328 if (!context->rlimit[i])
1331 if (setrlimit_closest(i, context->rlimit[i]) < 0) {
1338 if (context->capability_bounding_set_drop) {
1339 err = capability_bounding_set_drop(context->capability_bounding_set_drop, false);
1341 r = EXIT_CAPABILITIES;
1346 if (context->user) {
1347 err = enforce_user(context, uid);
1354 /* PR_GET_SECUREBITS is not privileged, while
1355 * PR_SET_SECUREBITS is. So to suppress
1356 * potential EPERMs we'll try not to call
1357 * PR_SET_SECUREBITS unless necessary. */
1358 if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
1359 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
1361 r = EXIT_SECUREBITS;
1365 if (context->capabilities)
1366 if (cap_set_proc(context->capabilities) < 0) {
1368 r = EXIT_CAPABILITIES;
1373 if (!(our_env = new0(char*, 7))) {
1380 if (asprintf(our_env + n_env++, "LISTEN_PID=%lu", (unsigned long) getpid()) < 0 ||
1381 asprintf(our_env + n_env++, "LISTEN_FDS=%u", n_fds) < 0) {
1388 if (asprintf(our_env + n_env++, "HOME=%s", home) < 0) {
1395 if (asprintf(our_env + n_env++, "LOGNAME=%s", username) < 0 ||
1396 asprintf(our_env + n_env++, "USER=%s", username) < 0) {
1402 if (is_terminal_input(context->std_input) ||
1403 context->std_output == EXEC_OUTPUT_TTY ||
1404 context->std_error == EXEC_OUTPUT_TTY)
1405 if (!(our_env[n_env++] = strdup(default_term_for_tty(tty_path(context))))) {
1413 if (!(final_env = strv_env_merge(
1417 context->environment,
1426 if (!(final_argv = replace_env_argv(argv, final_env))) {
1432 final_env = strv_env_clean(final_env);
1434 execve(command->path, final_argv, final_env);
1441 log_warning("Failed at step %s spawning %s: %s",
1442 exit_status_to_string(r, EXIT_STATUS_SYSTEMD),
1443 command->path, strerror(-err));
1447 strv_free(final_env);
1449 strv_free(files_env);
1450 strv_free(final_argv);
1452 if (saved_stdin >= 0)
1453 close_nointr_nofail(saved_stdin);
1455 if (saved_stdout >= 0)
1456 close_nointr_nofail(saved_stdout);
1461 strv_free(files_env);
1463 /* We add the new process to the cgroup both in the child (so
1464 * that we can be sure that no user code is ever executed
1465 * outside of the cgroup) and in the parent (so that we can be
1466 * sure that when we kill the cgroup the process will be
1468 if (cgroup_bondings)
1469 cgroup_bonding_install_list(cgroup_bondings, pid, cgroup_suffix);
1471 log_debug("Forked %s as %lu", command->path, (unsigned long) pid);
1473 exec_status_start(&command->exec_status, pid);
1479 strv_free(files_env);
1484 void exec_context_init(ExecContext *c) {
1488 c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
1489 c->cpu_sched_policy = SCHED_OTHER;
1490 c->syslog_priority = LOG_DAEMON|LOG_INFO;
1491 c->syslog_level_prefix = true;
1492 c->mount_flags = MS_SHARED;
1493 c->kill_signal = SIGTERM;
1494 c->send_sigkill = true;
1495 c->control_group_persistent = -1;
1496 c->ignore_sigpipe = true;
1499 void exec_context_done(ExecContext *c) {
1504 strv_free(c->environment);
1505 c->environment = NULL;
1507 strv_free(c->environment_files);
1508 c->environment_files = NULL;
1510 for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1512 c->rlimit[l] = NULL;
1515 free(c->working_directory);
1516 c->working_directory = NULL;
1517 free(c->root_directory);
1518 c->root_directory = NULL;
1523 free(c->tcpwrap_name);
1524 c->tcpwrap_name = NULL;
1526 free(c->syslog_identifier);
1527 c->syslog_identifier = NULL;
1535 strv_free(c->supplementary_groups);
1536 c->supplementary_groups = NULL;
1541 if (c->capabilities) {
1542 cap_free(c->capabilities);
1543 c->capabilities = NULL;
1546 strv_free(c->read_only_dirs);
1547 c->read_only_dirs = NULL;
1549 strv_free(c->read_write_dirs);
1550 c->read_write_dirs = NULL;
1552 strv_free(c->inaccessible_dirs);
1553 c->inaccessible_dirs = NULL;
1556 CPU_FREE(c->cpuset);
1562 void exec_command_done(ExecCommand *c) {
1572 void exec_command_done_array(ExecCommand *c, unsigned n) {
1575 for (i = 0; i < n; i++)
1576 exec_command_done(c+i);
1579 void exec_command_free_list(ExecCommand *c) {
1583 LIST_REMOVE(ExecCommand, command, c, i);
1584 exec_command_done(i);
1589 void exec_command_free_array(ExecCommand **c, unsigned n) {
1592 for (i = 0; i < n; i++) {
1593 exec_command_free_list(c[i]);
1598 int exec_context_load_environment(const ExecContext *c, char ***l) {
1599 char **i, **r = NULL;
1604 STRV_FOREACH(i, c->environment_files) {
1607 bool ignore = false;
1617 if (!path_is_absolute(fn)) {
1626 if ((k = load_env_file(fn, &p)) < 0) {
1640 m = strv_env_merge(2, r, p);
1656 static void strv_fprintf(FILE *f, char **l) {
1662 fprintf(f, " %s", *g);
1665 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
1677 "%sWorkingDirectory: %s\n"
1678 "%sRootDirectory: %s\n"
1679 "%sNonBlocking: %s\n"
1680 "%sPrivateTmp: %s\n"
1681 "%sControlGroupModify: %s\n"
1682 "%sControlGroupPersistent: %s\n"
1683 "%sPrivateNetwork: %s\n",
1685 prefix, c->working_directory ? c->working_directory : "/",
1686 prefix, c->root_directory ? c->root_directory : "/",
1687 prefix, yes_no(c->non_blocking),
1688 prefix, yes_no(c->private_tmp),
1689 prefix, yes_no(c->control_group_modify),
1690 prefix, yes_no(c->control_group_persistent),
1691 prefix, yes_no(c->private_network));
1693 STRV_FOREACH(e, c->environment)
1694 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
1696 STRV_FOREACH(e, c->environment_files)
1697 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
1699 if (c->tcpwrap_name)
1701 "%sTCPWrapName: %s\n",
1702 prefix, c->tcpwrap_name);
1709 if (c->oom_score_adjust_set)
1711 "%sOOMScoreAdjust: %i\n",
1712 prefix, c->oom_score_adjust);
1714 for (i = 0; i < RLIM_NLIMITS; i++)
1716 fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
1720 "%sIOSchedulingClass: %s\n"
1721 "%sIOPriority: %i\n",
1722 prefix, ioprio_class_to_string(IOPRIO_PRIO_CLASS(c->ioprio)),
1723 prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
1725 if (c->cpu_sched_set)
1727 "%sCPUSchedulingPolicy: %s\n"
1728 "%sCPUSchedulingPriority: %i\n"
1729 "%sCPUSchedulingResetOnFork: %s\n",
1730 prefix, sched_policy_to_string(c->cpu_sched_policy),
1731 prefix, c->cpu_sched_priority,
1732 prefix, yes_no(c->cpu_sched_reset_on_fork));
1735 fprintf(f, "%sCPUAffinity:", prefix);
1736 for (i = 0; i < c->cpuset_ncpus; i++)
1737 if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
1738 fprintf(f, " %i", i);
1742 if (c->timer_slack_nsec_set)
1743 fprintf(f, "%sTimerSlackNSec: %lu\n", prefix, c->timer_slack_nsec);
1746 "%sStandardInput: %s\n"
1747 "%sStandardOutput: %s\n"
1748 "%sStandardError: %s\n",
1749 prefix, exec_input_to_string(c->std_input),
1750 prefix, exec_output_to_string(c->std_output),
1751 prefix, exec_output_to_string(c->std_error));
1757 "%sTTYVHangup: %s\n"
1758 "%sTTYVTDisallocate: %s\n",
1759 prefix, c->tty_path,
1760 prefix, yes_no(c->tty_reset),
1761 prefix, yes_no(c->tty_vhangup),
1762 prefix, yes_no(c->tty_vt_disallocate));
1764 if (c->std_output == EXEC_OUTPUT_SYSLOG || c->std_output == EXEC_OUTPUT_KMSG || c->std_output == EXEC_OUTPUT_JOURNAL ||
1765 c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE ||
1766 c->std_error == EXEC_OUTPUT_SYSLOG || c->std_error == EXEC_OUTPUT_KMSG || c->std_error == EXEC_OUTPUT_JOURNAL ||
1767 c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE)
1769 "%sSyslogFacility: %s\n"
1770 "%sSyslogLevel: %s\n",
1771 prefix, log_facility_unshifted_to_string(c->syslog_priority >> 3),
1772 prefix, log_level_to_string(LOG_PRI(c->syslog_priority)));
1774 if (c->capabilities) {
1776 if ((t = cap_to_text(c->capabilities, NULL))) {
1777 fprintf(f, "%sCapabilities: %s\n",
1784 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
1786 (c->secure_bits & SECURE_KEEP_CAPS) ? " keep-caps" : "",
1787 (c->secure_bits & SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
1788 (c->secure_bits & SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
1789 (c->secure_bits & SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
1790 (c->secure_bits & SECURE_NOROOT) ? " noroot" : "",
1791 (c->secure_bits & SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
1793 if (c->capability_bounding_set_drop) {
1795 fprintf(f, "%sCapabilityBoundingSet:", prefix);
1797 for (l = 0; l <= cap_last_cap(); l++)
1798 if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) {
1801 if ((t = cap_to_name(l))) {
1802 fprintf(f, " %s", t);
1811 fprintf(f, "%sUser: %s\n", prefix, c->user);
1813 fprintf(f, "%sGroup: %s\n", prefix, c->group);
1815 if (strv_length(c->supplementary_groups) > 0) {
1816 fprintf(f, "%sSupplementaryGroups:", prefix);
1817 strv_fprintf(f, c->supplementary_groups);
1822 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
1824 if (strv_length(c->read_write_dirs) > 0) {
1825 fprintf(f, "%sReadWriteDirs:", prefix);
1826 strv_fprintf(f, c->read_write_dirs);
1830 if (strv_length(c->read_only_dirs) > 0) {
1831 fprintf(f, "%sReadOnlyDirs:", prefix);
1832 strv_fprintf(f, c->read_only_dirs);
1836 if (strv_length(c->inaccessible_dirs) > 0) {
1837 fprintf(f, "%sInaccessibleDirs:", prefix);
1838 strv_fprintf(f, c->inaccessible_dirs);
1844 "%sKillSignal: SIG%s\n"
1845 "%sSendSIGKILL: %s\n"
1846 "%sIgnoreSIGPIPE: %s\n",
1847 prefix, kill_mode_to_string(c->kill_mode),
1848 prefix, signal_to_string(c->kill_signal),
1849 prefix, yes_no(c->send_sigkill),
1850 prefix, yes_no(c->ignore_sigpipe));
1854 "%sUtmpIdentifier: %s\n",
1855 prefix, c->utmp_id);
1858 void exec_status_start(ExecStatus *s, pid_t pid) {
1863 dual_timestamp_get(&s->start_timestamp);
1866 void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
1869 if (s->pid && s->pid != pid)
1873 dual_timestamp_get(&s->exit_timestamp);
1879 if (context->utmp_id)
1880 utmp_put_dead_process(context->utmp_id, pid, code, status);
1882 exec_context_tty_reset(context);
1886 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
1887 char buf[FORMAT_TIMESTAMP_MAX];
1900 prefix, (unsigned long) s->pid);
1902 if (s->start_timestamp.realtime > 0)
1904 "%sStart Timestamp: %s\n",
1905 prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
1907 if (s->exit_timestamp.realtime > 0)
1909 "%sExit Timestamp: %s\n"
1911 "%sExit Status: %i\n",
1912 prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
1913 prefix, sigchld_code_to_string(s->code),
1917 char *exec_command_line(char **argv) {
1925 STRV_FOREACH(a, argv)
1928 if (!(n = new(char, k)))
1932 STRV_FOREACH(a, argv) {
1939 if (strpbrk(*a, WHITESPACE)) {
1950 /* FIXME: this doesn't really handle arguments that have
1951 * spaces and ticks in them */
1956 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
1958 const char *prefix2;
1967 p2 = strappend(prefix, "\t");
1968 prefix2 = p2 ? p2 : prefix;
1970 cmd = exec_command_line(c->argv);
1973 "%sCommand Line: %s\n",
1974 prefix, cmd ? cmd : strerror(ENOMEM));
1978 exec_status_dump(&c->exec_status, f, prefix2);
1983 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
1989 LIST_FOREACH(command, c, c)
1990 exec_command_dump(c, f, prefix);
1993 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
2000 /* It's kind of important, that we keep the order here */
2001 LIST_FIND_TAIL(ExecCommand, command, *l, end);
2002 LIST_INSERT_AFTER(ExecCommand, command, *l, end, e);
2007 int exec_command_set(ExecCommand *c, const char *path, ...) {
2015 l = strv_new_ap(path, ap);
2021 if (!(p = strdup(path))) {
2035 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
2036 [EXEC_INPUT_NULL] = "null",
2037 [EXEC_INPUT_TTY] = "tty",
2038 [EXEC_INPUT_TTY_FORCE] = "tty-force",
2039 [EXEC_INPUT_TTY_FAIL] = "tty-fail",
2040 [EXEC_INPUT_SOCKET] = "socket"
2043 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
2045 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
2046 [EXEC_OUTPUT_INHERIT] = "inherit",
2047 [EXEC_OUTPUT_NULL] = "null",
2048 [EXEC_OUTPUT_TTY] = "tty",
2049 [EXEC_OUTPUT_SYSLOG] = "syslog",
2050 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
2051 [EXEC_OUTPUT_KMSG] = "kmsg",
2052 [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
2053 [EXEC_OUTPUT_JOURNAL] = "journal",
2054 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE] = "journal+console",
2055 [EXEC_OUTPUT_SOCKET] = "socket"
2058 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
2060 static const char* const kill_mode_table[_KILL_MODE_MAX] = {
2061 [KILL_CONTROL_GROUP] = "control-group",
2062 [KILL_PROCESS] = "process",
2063 [KILL_NONE] = "none"
2066 DEFINE_STRING_TABLE_LOOKUP(kill_mode, KillMode);
2068 static const char* const kill_who_table[_KILL_WHO_MAX] = {
2069 [KILL_MAIN] = "main",
2070 [KILL_CONTROL] = "control",
2074 DEFINE_STRING_TABLE_LOOKUP(kill_who, KillWho);