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"
59 /* This assumes there is a 'tty' group */
62 static int shift_fds(int fds[], unsigned n_fds) {
63 int start, restart_from;
68 /* Modifies the fds array! (sorts it) */
78 for (i = start; i < (int) n_fds; i++) {
81 /* Already at right index? */
85 if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
88 close_nointr_nofail(fds[i]);
91 /* Hmm, the fd we wanted isn't free? Then
92 * let's remember that and try again from here*/
93 if (nfd != i+3 && restart_from < 0)
100 start = restart_from;
106 static int flags_fds(const int fds[], unsigned n_fds, bool nonblock) {
115 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
117 for (i = 0; i < n_fds; i++) {
119 if ((r = fd_nonblock(fds[i], nonblock)) < 0)
122 /* We unconditionally drop FD_CLOEXEC from the fds,
123 * since after all we want to pass these fds to our
126 if ((r = fd_cloexec(fds[i], false)) < 0)
133 static const char *tty_path(const ExecContext *context) {
136 if (context->tty_path)
137 return context->tty_path;
139 return "/dev/console";
142 static int open_null_as(int flags, int nfd) {
147 if ((fd = open("/dev/null", flags|O_NOCTTY)) < 0)
151 r = dup2(fd, nfd) < 0 ? -errno : nfd;
152 close_nointr_nofail(fd);
159 static int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, int nfd) {
163 struct sockaddr_un un;
167 assert(output < _EXEC_OUTPUT_MAX);
171 if ((fd = socket(AF_UNIX, SOCK_STREAM, 0)) < 0)
175 sa.sa.sa_family = AF_UNIX;
176 strncpy(sa.un.sun_path, LOGGER_SOCKET, sizeof(sa.un.sun_path));
178 if (connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + sizeof(LOGGER_SOCKET) - 1) < 0) {
179 close_nointr_nofail(fd);
183 if (shutdown(fd, SHUT_RD) < 0) {
184 close_nointr_nofail(fd);
188 /* We speak a very simple protocol between log server
189 * and client: one line for the log destination (kmsg
190 * or syslog), followed by the priority field,
191 * followed by the process name. Since we replaced
192 * stdin/stderr we simple use stdio to write to
193 * it. Note that we use stderr, to minimize buffer
194 * flushing issues. */
201 output == EXEC_OUTPUT_KMSG ? "kmsg" :
202 output == EXEC_OUTPUT_KMSG_AND_CONSOLE ? "kmsg+console" :
203 output == EXEC_OUTPUT_SYSLOG ? "syslog" :
205 context->syslog_priority,
206 context->syslog_identifier ? context->syslog_identifier : ident,
207 context->syslog_level_prefix);
210 r = dup2(fd, nfd) < 0 ? -errno : nfd;
211 close_nointr_nofail(fd);
217 static int open_terminal_as(const char *path, mode_t mode, int nfd) {
223 if ((fd = open_terminal(path, mode | O_NOCTTY)) < 0)
227 r = dup2(fd, nfd) < 0 ? -errno : nfd;
228 close_nointr_nofail(fd);
235 static bool is_terminal_input(ExecInput i) {
237 i == EXEC_INPUT_TTY ||
238 i == EXEC_INPUT_TTY_FORCE ||
239 i == EXEC_INPUT_TTY_FAIL;
242 static int fixup_input(ExecInput std_input, int socket_fd, bool apply_tty_stdin) {
244 if (is_terminal_input(std_input) && !apply_tty_stdin)
245 return EXEC_INPUT_NULL;
247 if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0)
248 return EXEC_INPUT_NULL;
253 static int fixup_output(ExecOutput std_output, int socket_fd) {
255 if (std_output == EXEC_OUTPUT_SOCKET && socket_fd < 0)
256 return EXEC_OUTPUT_INHERIT;
261 static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty_stdin) {
266 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
270 case EXEC_INPUT_NULL:
271 return open_null_as(O_RDONLY, STDIN_FILENO);
274 case EXEC_INPUT_TTY_FORCE:
275 case EXEC_INPUT_TTY_FAIL: {
278 if ((fd = acquire_terminal(
280 i == EXEC_INPUT_TTY_FAIL,
281 i == EXEC_INPUT_TTY_FORCE,
285 if (fd != STDIN_FILENO) {
286 r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
287 close_nointr_nofail(fd);
294 case EXEC_INPUT_SOCKET:
295 return dup2(socket_fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
298 assert_not_reached("Unknown input type");
302 static int setup_output(const ExecContext *context, int socket_fd, const char *ident, bool apply_tty_stdin) {
309 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
310 o = fixup_output(context->std_output, socket_fd);
312 /* This expects the input is already set up */
316 case EXEC_OUTPUT_INHERIT:
318 /* If input got downgraded, inherit the original value */
319 if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input))
320 return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO);
322 /* If the input is connected to anything that's not a /dev/null, inherit that... */
323 if (i != EXEC_INPUT_NULL)
324 return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
326 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
328 return STDOUT_FILENO;
330 /* We need to open /dev/null here anew, to get the
331 * right access mode. So we fall through */
333 case EXEC_OUTPUT_NULL:
334 return open_null_as(O_WRONLY, STDOUT_FILENO);
336 case EXEC_OUTPUT_TTY:
337 if (is_terminal_input(i))
338 return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
340 /* We don't reset the terminal if this is just about output */
341 return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO);
343 case EXEC_OUTPUT_SYSLOG:
344 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
345 case EXEC_OUTPUT_KMSG:
346 case EXEC_OUTPUT_KMSG_AND_CONSOLE:
347 return connect_logger_as(context, o, ident, STDOUT_FILENO);
349 case EXEC_OUTPUT_SOCKET:
350 assert(socket_fd >= 0);
351 return dup2(socket_fd, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
354 assert_not_reached("Unknown output type");
358 static int setup_error(const ExecContext *context, int socket_fd, const char *ident, bool apply_tty_stdin) {
365 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
366 o = fixup_output(context->std_output, socket_fd);
367 e = fixup_output(context->std_error, socket_fd);
369 /* This expects the input and output are already set up */
371 /* Don't change the stderr file descriptor if we inherit all
372 * the way and are not on a tty */
373 if (e == EXEC_OUTPUT_INHERIT &&
374 o == EXEC_OUTPUT_INHERIT &&
375 i == EXEC_INPUT_NULL &&
376 !is_terminal_input(context->std_input) &&
378 return STDERR_FILENO;
380 /* Duplicate from stdout if possible */
381 if (e == o || e == EXEC_OUTPUT_INHERIT)
382 return dup2(STDOUT_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
386 case EXEC_OUTPUT_NULL:
387 return open_null_as(O_WRONLY, STDERR_FILENO);
389 case EXEC_OUTPUT_TTY:
390 if (is_terminal_input(i))
391 return dup2(STDIN_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
393 /* We don't reset the terminal if this is just about output */
394 return open_terminal_as(tty_path(context), O_WRONLY, STDERR_FILENO);
396 case EXEC_OUTPUT_SYSLOG:
397 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
398 case EXEC_OUTPUT_KMSG:
399 case EXEC_OUTPUT_KMSG_AND_CONSOLE:
400 return connect_logger_as(context, e, ident, STDERR_FILENO);
402 case EXEC_OUTPUT_SOCKET:
403 assert(socket_fd >= 0);
404 return dup2(socket_fd, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
407 assert_not_reached("Unknown error type");
411 static int chown_terminal(int fd, uid_t uid) {
416 /* This might fail. What matters are the results. */
417 (void) fchown(fd, uid, -1);
418 (void) fchmod(fd, TTY_MODE);
420 if (fstat(fd, &st) < 0)
423 if (st.st_uid != uid || (st.st_mode & 0777) != TTY_MODE)
429 static int setup_confirm_stdio(const ExecContext *context,
431 int *_saved_stdout) {
432 int fd = -1, saved_stdin, saved_stdout = -1, r;
435 assert(_saved_stdin);
436 assert(_saved_stdout);
438 /* This returns positive EXIT_xxx return values instead of
439 * negative errno style values! */
441 if ((saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3)) < 0)
444 if ((saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3)) < 0) {
449 if ((fd = acquire_terminal(
451 context->std_input == EXEC_INPUT_TTY_FAIL,
452 context->std_input == EXEC_INPUT_TTY_FORCE,
458 if (chown_terminal(fd, getuid()) < 0) {
463 if (dup2(fd, STDIN_FILENO) < 0) {
468 if (dup2(fd, STDOUT_FILENO) < 0) {
474 close_nointr_nofail(fd);
476 *_saved_stdin = saved_stdin;
477 *_saved_stdout = saved_stdout;
482 if (saved_stdout >= 0)
483 close_nointr_nofail(saved_stdout);
485 if (saved_stdin >= 0)
486 close_nointr_nofail(saved_stdin);
489 close_nointr_nofail(fd);
494 static int restore_confirm_stdio(const ExecContext *context,
502 assert(*saved_stdin >= 0);
503 assert(saved_stdout);
504 assert(*saved_stdout >= 0);
506 /* This returns positive EXIT_xxx return values instead of
507 * negative errno style values! */
509 if (is_terminal_input(context->std_input)) {
511 /* The service wants terminal input. */
515 context->std_output == EXEC_OUTPUT_INHERIT ||
516 context->std_output == EXEC_OUTPUT_TTY;
519 /* If the service doesn't want a controlling terminal,
520 * then we need to get rid entirely of what we have
523 if (release_terminal() < 0)
526 if (dup2(*saved_stdin, STDIN_FILENO) < 0)
529 if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
532 *keep_stdout = *keep_stdin = false;
538 static int get_group_creds(const char *groupname, gid_t *gid) {
545 /* We enforce some special rules for gid=0: in order to avoid
546 * NSS lookups for root we hardcode its data. */
548 if (streq(groupname, "root") || streq(groupname, "0")) {
553 if (safe_atolu(groupname, &lu) >= 0) {
555 g = getgrgid((gid_t) lu);
558 g = getgrnam(groupname);
562 return errno != 0 ? -errno : -ESRCH;
568 static int get_user_creds(const char **username, uid_t *uid, gid_t *gid, const char **home) {
578 /* We enforce some special rules for uid=0: in order to avoid
579 * NSS lookups for root we hardcode its data. */
581 if (streq(*username, "root") || streq(*username, "0")) {
589 if (safe_atolu(*username, &lu) >= 0) {
591 p = getpwuid((uid_t) lu);
593 /* If there are multiple users with the same id, make
594 * sure to leave $USER to the configured value instead
595 * of the first occurrence in the database. However if
596 * the uid was configured by a numeric uid, then let's
597 * pick the real username from /etc/passwd. */
599 *username = p->pw_name;
602 p = getpwnam(*username);
606 return errno != 0 ? -errno : -ESRCH;
614 static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
615 bool keep_groups = false;
620 /* Lookup and set GID and supplementary group list. Here too
621 * we avoid NSS lookups for gid=0. */
623 if (context->group || username) {
626 if ((r = get_group_creds(context->group, &gid)) < 0)
629 /* First step, initialize groups from /etc/groups */
630 if (username && gid != 0) {
631 if (initgroups(username, gid) < 0)
637 /* Second step, set our gids */
638 if (setresgid(gid, gid, gid) < 0)
642 if (context->supplementary_groups) {
647 /* Final step, initialize any manually set supplementary groups */
648 ngroups_max = (int) sysconf(_SC_NGROUPS_MAX);
650 if (!(gids = new(gid_t, ngroups_max)))
654 if ((k = getgroups(ngroups_max, gids)) < 0) {
661 STRV_FOREACH(i, context->supplementary_groups) {
663 if (k >= ngroups_max) {
668 if ((r = get_group_creds(*i, gids+k)) < 0) {
676 if (setgroups(k, gids) < 0) {
687 static int enforce_user(const ExecContext *context, uid_t uid) {
691 /* Sets (but doesn't lookup) the uid and make sure we keep the
692 * capabilities while doing so. */
694 if (context->capabilities) {
696 static const cap_value_t bits[] = {
697 CAP_SETUID, /* Necessary so that we can run setresuid() below */
698 CAP_SETPCAP /* Necessary so that we can set PR_SET_SECUREBITS later on */
701 /* First step: If we need to keep capabilities but
702 * drop privileges we need to make sure we keep our
703 * caps, whiel we drop privileges. */
705 int sb = context->secure_bits|SECURE_KEEP_CAPS;
707 if (prctl(PR_GET_SECUREBITS) != sb)
708 if (prctl(PR_SET_SECUREBITS, sb) < 0)
712 /* Second step: set the capabilities. This will reduce
713 * the capabilities to the minimum we need. */
715 if (!(d = cap_dup(context->capabilities)))
718 if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
719 cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0) {
725 if (cap_set_proc(d) < 0) {
734 /* Third step: actually set the uids */
735 if (setresuid(uid, uid, uid) < 0)
738 /* At this point we should have all necessary capabilities but
739 are otherwise a normal user. However, the caps might got
740 corrupted due to the setresuid() so we need clean them up
741 later. This is done outside of this call. */
748 static int null_conv(
750 const struct pam_message **msg,
751 struct pam_response **resp,
754 /* We don't support conversations */
759 static int setup_pam(
764 int fds[], unsigned n_fds) {
766 static const struct pam_conv conv = {
771 pam_handle_t *handle = NULL;
773 int pam_code = PAM_SUCCESS;
775 bool close_session = false;
776 pid_t pam_pid = 0, parent_pid;
782 /* We set up PAM in the parent process, then fork. The child
783 * will then stay around until killed via PR_GET_PDEATHSIG or
784 * systemd via the cgroup logic. It will then remove the PAM
785 * session again. The parent process will exec() the actual
786 * daemon. We do things this way to ensure that the main PID
787 * of the daemon is the one we initially fork()ed. */
789 if ((pam_code = pam_start(name, user, &conv, &handle)) != PAM_SUCCESS) {
795 if ((pam_code = pam_set_item(handle, PAM_TTY, tty)) != PAM_SUCCESS)
798 if ((pam_code = pam_acct_mgmt(handle, PAM_SILENT)) != PAM_SUCCESS)
801 if ((pam_code = pam_open_session(handle, PAM_SILENT)) != PAM_SUCCESS)
804 close_session = true;
806 if ((pam_code = pam_setcred(handle, PAM_ESTABLISH_CRED | PAM_SILENT)) != PAM_SUCCESS)
809 if ((!(e = pam_getenvlist(handle)))) {
810 pam_code = PAM_BUF_ERR;
814 /* Block SIGTERM, so that we know that it won't get lost in
816 if (sigemptyset(&ss) < 0 ||
817 sigaddset(&ss, SIGTERM) < 0 ||
818 sigprocmask(SIG_BLOCK, &ss, &old_ss) < 0)
821 parent_pid = getpid();
823 if ((pam_pid = fork()) < 0)
830 /* The child's job is to reset the PAM session on
833 /* This string must fit in 10 chars (i.e. the length
834 * of "/sbin/init") */
835 rename_process("sd:pam");
837 /* Make sure we don't keep open the passed fds in this
838 child. We assume that otherwise only those fds are
839 open here that have been opened by PAM. */
840 close_many(fds, n_fds);
842 /* Wait until our parent died. This will most likely
843 * not work since the kernel does not allow
844 * unprivileged parents kill their privileged children
845 * this way. We rely on the control groups kill logic
846 * to do the rest for us. */
847 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
850 /* Check if our parent process might already have
852 if (getppid() == parent_pid) {
853 if (sigwait(&ss, &sig) < 0)
856 assert(sig == SIGTERM);
859 /* Only if our parent died we'll end the session */
860 if (getppid() != parent_pid)
861 if ((pam_code = pam_close_session(handle, PAM_DATA_SILENT)) != PAM_SUCCESS)
867 pam_end(handle, pam_code | PAM_DATA_SILENT);
871 /* If the child was forked off successfully it will do all the
872 * cleanups, so forget about the handle here. */
875 /* Unblock SIGSUR1 again in the parent */
876 if (sigprocmask(SIG_SETMASK, &old_ss, NULL) < 0)
879 /* We close the log explicitly here, since the PAM modules
880 * might have opened it, but we don't want this fd around. */
888 pam_code = pam_close_session(handle, PAM_DATA_SILENT);
890 pam_end(handle, pam_code | PAM_DATA_SILENT);
898 kill(pam_pid, SIGTERM);
899 kill(pam_pid, SIGCONT);
906 int exec_spawn(ExecCommand *command,
908 const ExecContext *context,
909 int fds[], unsigned n_fds,
911 bool apply_permissions,
913 bool apply_tty_stdin,
915 CGroupBonding *cgroup_bondings,
922 char **files_env = NULL;
927 assert(fds || n_fds <= 0);
929 if (context->std_input == EXEC_INPUT_SOCKET ||
930 context->std_output == EXEC_OUTPUT_SOCKET ||
931 context->std_error == EXEC_OUTPUT_SOCKET) {
943 if ((r = exec_context_load_environment(context, &files_env)) < 0) {
944 log_error("Failed to load environment files: %s", strerror(-r));
949 argv = command->argv;
951 if (!(line = exec_command_line(argv))) {
956 log_debug("About to execute: %s", line);
960 if ((r = cgroup_bonding_realize_list(cgroup_bondings)))
963 if ((pid = fork()) < 0) {
971 const char *username = NULL, *home = NULL;
972 uid_t uid = (uid_t) -1;
973 gid_t gid = (gid_t) -1;
974 char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
976 int saved_stdout = -1, saved_stdin = -1;
977 bool keep_stdout = false, keep_stdin = false;
981 /* This string must fit in 10 chars (i.e. the length
982 * of "/sbin/init") */
983 rename_process("sd:exec");
985 /* We reset exactly these signals, since they are the
986 * only ones we set to SIG_IGN in the main daemon. All
987 * others we leave untouched because we set them to
988 * SIG_DFL or a valid handler initially, both of which
989 * will be demoted to SIG_DFL. */
990 default_signals(SIGNALS_CRASH_HANDLER,
993 if (sigemptyset(&ss) < 0 ||
994 sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
995 r = EXIT_SIGNAL_MASK;
999 /* Close sockets very early to make sure we don't
1000 * block init reexecution because it cannot bind its
1002 if (close_all_fds(socket_fd >= 0 ? &socket_fd : fds,
1003 socket_fd >= 0 ? 1 : n_fds) < 0) {
1008 if (!context->same_pgrp)
1014 if (context->tcpwrap_name) {
1016 if (!socket_tcpwrap(socket_fd, context->tcpwrap_name)) {
1021 for (i = 0; i < (int) n_fds; i++) {
1022 if (!socket_tcpwrap(fds[i], context->tcpwrap_name)) {
1029 /* We skip the confirmation step if we shall not apply the TTY */
1030 if (confirm_spawn &&
1031 (!is_terminal_input(context->std_input) || apply_tty_stdin)) {
1034 /* Set up terminal for the question */
1035 if ((r = setup_confirm_stdio(context,
1036 &saved_stdin, &saved_stdout)))
1039 /* Now ask the question. */
1040 if (!(line = exec_command_line(argv))) {
1045 r = ask(&response, "yns", "Execute %s? [Yes, No, Skip] ", line);
1048 if (r < 0 || response == 'n') {
1051 } else if (response == 's') {
1056 /* Release terminal for the question */
1057 if ((r = restore_confirm_stdio(context,
1058 &saved_stdin, &saved_stdout,
1059 &keep_stdin, &keep_stdout)))
1063 /* If a socket is connected to STDIN/STDOUT/STDERR, we
1064 * must sure to drop O_NONBLOCK */
1066 fd_nonblock(socket_fd, false);
1069 if (setup_input(context, socket_fd, apply_tty_stdin) < 0) {
1075 if (setup_output(context, socket_fd, file_name_from_path(command->path), apply_tty_stdin) < 0) {
1080 if (setup_error(context, socket_fd, file_name_from_path(command->path), apply_tty_stdin) < 0) {
1085 if (cgroup_bondings)
1086 if (cgroup_bonding_install_list(cgroup_bondings, 0) < 0) {
1091 if (context->oom_score_adjust_set) {
1094 snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
1097 if (write_one_line_file("/proc/self/oom_score_adj", t) < 0) {
1098 /* Compatibility with Linux <= 2.6.35 */
1102 adj = (context->oom_score_adjust * -OOM_DISABLE) / OOM_SCORE_ADJ_MAX;
1103 adj = CLAMP(adj, OOM_DISABLE, OOM_ADJUST_MAX);
1105 snprintf(t, sizeof(t), "%i", adj);
1108 if (write_one_line_file("/proc/self/oom_adj", t) < 0) {
1109 r = EXIT_OOM_ADJUST;
1115 if (context->nice_set)
1116 if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
1121 if (context->cpu_sched_set) {
1122 struct sched_param param;
1125 param.sched_priority = context->cpu_sched_priority;
1127 if (sched_setscheduler(0, context->cpu_sched_policy |
1128 (context->cpu_sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0), ¶m) < 0) {
1129 r = EXIT_SETSCHEDULER;
1134 if (context->cpuset)
1135 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
1136 r = EXIT_CPUAFFINITY;
1140 if (context->ioprio_set)
1141 if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
1146 if (context->timer_slack_nsec_set)
1147 if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
1148 r = EXIT_TIMERSLACK;
1152 if (context->utmp_id)
1153 utmp_put_init_process(0, context->utmp_id, getpid(), getsid(0), context->tty_path);
1155 if (context->user) {
1156 username = context->user;
1157 if (get_user_creds(&username, &uid, &gid, &home) < 0) {
1162 if (is_terminal_input(context->std_input))
1163 if (chown_terminal(STDIN_FILENO, uid) < 0) {
1170 if (context->pam_name && username) {
1171 if (setup_pam(context->pam_name, username, context->tty_path, &pam_env, fds, n_fds) < 0) {
1178 if (apply_permissions)
1179 if (enforce_groups(context, username, uid) < 0) {
1184 umask(context->umask);
1186 if (strv_length(context->read_write_dirs) > 0 ||
1187 strv_length(context->read_only_dirs) > 0 ||
1188 strv_length(context->inaccessible_dirs) > 0 ||
1189 context->mount_flags != MS_SHARED ||
1190 context->private_tmp)
1191 if ((r = setup_namespace(
1192 context->read_write_dirs,
1193 context->read_only_dirs,
1194 context->inaccessible_dirs,
1195 context->private_tmp,
1196 context->mount_flags)) < 0)
1200 if (context->root_directory)
1201 if (chroot(context->root_directory) < 0) {
1206 if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
1214 if (asprintf(&d, "%s/%s",
1215 context->root_directory ? context->root_directory : "",
1216 context->working_directory ? context->working_directory : "") < 0) {
1230 /* We repeat the fd closing here, to make sure that
1231 * nothing is leaked from the PAM modules */
1232 if (close_all_fds(fds, n_fds) < 0 ||
1233 shift_fds(fds, n_fds) < 0 ||
1234 flags_fds(fds, n_fds, context->non_blocking) < 0) {
1239 if (apply_permissions) {
1241 for (i = 0; i < RLIMIT_NLIMITS; i++) {
1242 if (!context->rlimit[i])
1245 if (setrlimit(i, context->rlimit[i]) < 0) {
1252 if (enforce_user(context, uid) < 0) {
1257 /* PR_GET_SECUREBITS is not privileged, while
1258 * PR_SET_SECUREBITS is. So to suppress
1259 * potential EPERMs we'll try not to call
1260 * PR_SET_SECUREBITS unless necessary. */
1261 if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
1262 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
1263 r = EXIT_SECUREBITS;
1267 if (context->capabilities)
1268 if (cap_set_proc(context->capabilities) < 0) {
1269 r = EXIT_CAPABILITIES;
1274 if (!(our_env = new0(char*, 7))) {
1280 if (asprintf(our_env + n_env++, "LISTEN_PID=%lu", (unsigned long) getpid()) < 0 ||
1281 asprintf(our_env + n_env++, "LISTEN_FDS=%u", n_fds) < 0) {
1287 if (asprintf(our_env + n_env++, "HOME=%s", home) < 0) {
1293 if (asprintf(our_env + n_env++, "LOGNAME=%s", username) < 0 ||
1294 asprintf(our_env + n_env++, "USER=%s", username) < 0) {
1299 if (is_terminal_input(context->std_input) ||
1300 context->std_output == EXEC_OUTPUT_TTY ||
1301 context->std_error == EXEC_OUTPUT_TTY)
1302 if (!(our_env[n_env++] = strdup(default_term_for_tty(tty_path(context))))) {
1309 if (!(final_env = strv_env_merge(
1313 context->environment,
1321 if (!(final_argv = replace_env_argv(argv, final_env))) {
1326 final_env = strv_env_clean(final_env);
1328 execve(command->path, final_argv, final_env);
1333 strv_free(final_env);
1335 strv_free(files_env);
1336 strv_free(final_argv);
1338 if (saved_stdin >= 0)
1339 close_nointr_nofail(saved_stdin);
1341 if (saved_stdout >= 0)
1342 close_nointr_nofail(saved_stdout);
1347 strv_free(files_env);
1349 /* We add the new process to the cgroup both in the child (so
1350 * that we can be sure that no user code is ever executed
1351 * outside of the cgroup) and in the parent (so that we can be
1352 * sure that when we kill the cgroup the process will be
1354 if (cgroup_bondings)
1355 cgroup_bonding_install_list(cgroup_bondings, pid);
1357 log_debug("Forked %s as %lu", command->path, (unsigned long) pid);
1359 exec_status_start(&command->exec_status, pid);
1365 strv_free(files_env);
1370 void exec_context_init(ExecContext *c) {
1374 c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
1375 c->cpu_sched_policy = SCHED_OTHER;
1376 c->syslog_priority = LOG_DAEMON|LOG_INFO;
1377 c->syslog_level_prefix = true;
1378 c->mount_flags = MS_SHARED;
1379 c->kill_signal = SIGTERM;
1380 c->send_sigkill = true;
1383 void exec_context_done(ExecContext *c) {
1388 strv_free(c->environment);
1389 c->environment = NULL;
1391 strv_free(c->environment_files);
1392 c->environment_files = NULL;
1394 for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1396 c->rlimit[l] = NULL;
1399 free(c->working_directory);
1400 c->working_directory = NULL;
1401 free(c->root_directory);
1402 c->root_directory = NULL;
1407 free(c->tcpwrap_name);
1408 c->tcpwrap_name = NULL;
1410 free(c->syslog_identifier);
1411 c->syslog_identifier = NULL;
1419 strv_free(c->supplementary_groups);
1420 c->supplementary_groups = NULL;
1425 if (c->capabilities) {
1426 cap_free(c->capabilities);
1427 c->capabilities = NULL;
1430 strv_free(c->read_only_dirs);
1431 c->read_only_dirs = NULL;
1433 strv_free(c->read_write_dirs);
1434 c->read_write_dirs = NULL;
1436 strv_free(c->inaccessible_dirs);
1437 c->inaccessible_dirs = NULL;
1440 CPU_FREE(c->cpuset);
1446 void exec_command_done(ExecCommand *c) {
1456 void exec_command_done_array(ExecCommand *c, unsigned n) {
1459 for (i = 0; i < n; i++)
1460 exec_command_done(c+i);
1463 void exec_command_free_list(ExecCommand *c) {
1467 LIST_REMOVE(ExecCommand, command, c, i);
1468 exec_command_done(i);
1473 void exec_command_free_array(ExecCommand **c, unsigned n) {
1476 for (i = 0; i < n; i++) {
1477 exec_command_free_list(c[i]);
1482 int exec_context_load_environment(const ExecContext *c, char ***l) {
1483 char **i, **r = NULL;
1488 STRV_FOREACH(i, c->environment_files) {
1491 bool ignore = false;
1501 if (!path_is_absolute(fn)) {
1510 if ((k = load_env_file(fn, &p)) < 0) {
1524 m = strv_env_merge(2, r, p);
1540 static void strv_fprintf(FILE *f, char **l) {
1546 fprintf(f, " %s", *g);
1549 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
1561 "%sWorkingDirectory: %s\n"
1562 "%sRootDirectory: %s\n"
1563 "%sNonBlocking: %s\n"
1564 "%sPrivateTmp: %s\n",
1566 prefix, c->working_directory ? c->working_directory : "/",
1567 prefix, c->root_directory ? c->root_directory : "/",
1568 prefix, yes_no(c->non_blocking),
1569 prefix, yes_no(c->private_tmp));
1571 STRV_FOREACH(e, c->environment)
1572 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
1574 STRV_FOREACH(e, c->environment_files)
1575 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
1577 if (c->tcpwrap_name)
1579 "%sTCPWrapName: %s\n",
1580 prefix, c->tcpwrap_name);
1587 if (c->oom_score_adjust_set)
1589 "%sOOMScoreAdjust: %i\n",
1590 prefix, c->oom_score_adjust);
1592 for (i = 0; i < RLIM_NLIMITS; i++)
1594 fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
1598 "%sIOSchedulingClass: %s\n"
1599 "%sIOPriority: %i\n",
1600 prefix, ioprio_class_to_string(IOPRIO_PRIO_CLASS(c->ioprio)),
1601 prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
1603 if (c->cpu_sched_set)
1605 "%sCPUSchedulingPolicy: %s\n"
1606 "%sCPUSchedulingPriority: %i\n"
1607 "%sCPUSchedulingResetOnFork: %s\n",
1608 prefix, sched_policy_to_string(c->cpu_sched_policy),
1609 prefix, c->cpu_sched_priority,
1610 prefix, yes_no(c->cpu_sched_reset_on_fork));
1613 fprintf(f, "%sCPUAffinity:", prefix);
1614 for (i = 0; i < c->cpuset_ncpus; i++)
1615 if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
1616 fprintf(f, " %i", i);
1620 if (c->timer_slack_nsec_set)
1621 fprintf(f, "%sTimerSlackNSec: %lu\n", prefix, c->timer_slack_nsec);
1624 "%sStandardInput: %s\n"
1625 "%sStandardOutput: %s\n"
1626 "%sStandardError: %s\n",
1627 prefix, exec_input_to_string(c->std_input),
1628 prefix, exec_output_to_string(c->std_output),
1629 prefix, exec_output_to_string(c->std_error));
1634 prefix, c->tty_path);
1636 if (c->std_output == EXEC_OUTPUT_SYSLOG || c->std_output == EXEC_OUTPUT_KMSG ||
1637 c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
1638 c->std_error == EXEC_OUTPUT_SYSLOG || c->std_error == EXEC_OUTPUT_KMSG ||
1639 c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE)
1641 "%sSyslogFacility: %s\n"
1642 "%sSyslogLevel: %s\n",
1643 prefix, log_facility_to_string(LOG_FAC(c->syslog_priority)),
1644 prefix, log_level_to_string(LOG_PRI(c->syslog_priority)));
1646 if (c->capabilities) {
1648 if ((t = cap_to_text(c->capabilities, NULL))) {
1649 fprintf(f, "%sCapabilities: %s\n",
1656 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
1658 (c->secure_bits & SECURE_KEEP_CAPS) ? " keep-caps" : "",
1659 (c->secure_bits & SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
1660 (c->secure_bits & SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
1661 (c->secure_bits & SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
1662 (c->secure_bits & SECURE_NOROOT) ? " noroot" : "",
1663 (c->secure_bits & SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
1665 if (c->capability_bounding_set_drop) {
1666 fprintf(f, "%sCapabilityBoundingSetDrop:", prefix);
1668 for (i = 0; i <= CAP_LAST_CAP; i++)
1669 if (c->capability_bounding_set_drop & (1 << i)) {
1672 if ((t = cap_to_name(i))) {
1673 fprintf(f, " %s", t);
1682 fprintf(f, "%sUser: %s\n", prefix, c->user);
1684 fprintf(f, "%sGroup: %s\n", prefix, c->group);
1686 if (strv_length(c->supplementary_groups) > 0) {
1687 fprintf(f, "%sSupplementaryGroups:", prefix);
1688 strv_fprintf(f, c->supplementary_groups);
1693 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
1695 if (strv_length(c->read_write_dirs) > 0) {
1696 fprintf(f, "%sReadWriteDirs:", prefix);
1697 strv_fprintf(f, c->read_write_dirs);
1701 if (strv_length(c->read_only_dirs) > 0) {
1702 fprintf(f, "%sReadOnlyDirs:", prefix);
1703 strv_fprintf(f, c->read_only_dirs);
1707 if (strv_length(c->inaccessible_dirs) > 0) {
1708 fprintf(f, "%sInaccessibleDirs:", prefix);
1709 strv_fprintf(f, c->inaccessible_dirs);
1715 "%sKillSignal: SIG%s\n"
1716 "%sSendSIGKILL: %s\n",
1717 prefix, kill_mode_to_string(c->kill_mode),
1718 prefix, signal_to_string(c->kill_signal),
1719 prefix, yes_no(c->send_sigkill));
1723 "%sUtmpIdentifier: %s\n",
1724 prefix, c->utmp_id);
1727 void exec_status_start(ExecStatus *s, pid_t pid) {
1732 dual_timestamp_get(&s->start_timestamp);
1735 void exec_status_exit(ExecStatus *s, pid_t pid, int code, int status, const char *utmp_id) {
1738 if ((s->pid && s->pid != pid) ||
1739 !s->start_timestamp.realtime <= 0)
1743 dual_timestamp_get(&s->exit_timestamp);
1749 utmp_put_dead_process(utmp_id, pid, code, status);
1752 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
1753 char buf[FORMAT_TIMESTAMP_MAX];
1766 prefix, (unsigned long) s->pid);
1768 if (s->start_timestamp.realtime > 0)
1770 "%sStart Timestamp: %s\n",
1771 prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
1773 if (s->exit_timestamp.realtime > 0)
1775 "%sExit Timestamp: %s\n"
1777 "%sExit Status: %i\n",
1778 prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
1779 prefix, sigchld_code_to_string(s->code),
1783 char *exec_command_line(char **argv) {
1791 STRV_FOREACH(a, argv)
1794 if (!(n = new(char, k)))
1798 STRV_FOREACH(a, argv) {
1805 if (strpbrk(*a, WHITESPACE)) {
1816 /* FIXME: this doesn't really handle arguments that have
1817 * spaces and ticks in them */
1822 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
1824 const char *prefix2;
1833 p2 = strappend(prefix, "\t");
1834 prefix2 = p2 ? p2 : prefix;
1836 cmd = exec_command_line(c->argv);
1839 "%sCommand Line: %s\n",
1840 prefix, cmd ? cmd : strerror(ENOMEM));
1844 exec_status_dump(&c->exec_status, f, prefix2);
1849 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
1855 LIST_FOREACH(command, c, c)
1856 exec_command_dump(c, f, prefix);
1859 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
1866 /* It's kind of important, that we keep the order here */
1867 LIST_FIND_TAIL(ExecCommand, command, *l, end);
1868 LIST_INSERT_AFTER(ExecCommand, command, *l, end, e);
1873 int exec_command_set(ExecCommand *c, const char *path, ...) {
1881 l = strv_new_ap(path, ap);
1887 if (!(p = strdup(path))) {
1901 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
1902 [EXEC_INPUT_NULL] = "null",
1903 [EXEC_INPUT_TTY] = "tty",
1904 [EXEC_INPUT_TTY_FORCE] = "tty-force",
1905 [EXEC_INPUT_TTY_FAIL] = "tty-fail",
1906 [EXEC_INPUT_SOCKET] = "socket"
1909 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
1911 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
1912 [EXEC_OUTPUT_INHERIT] = "inherit",
1913 [EXEC_OUTPUT_NULL] = "null",
1914 [EXEC_OUTPUT_TTY] = "tty",
1915 [EXEC_OUTPUT_SYSLOG] = "syslog",
1916 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
1917 [EXEC_OUTPUT_KMSG] = "kmsg",
1918 [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
1919 [EXEC_OUTPUT_SOCKET] = "socket"
1922 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
1924 static const char* const kill_mode_table[_KILL_MODE_MAX] = {
1925 [KILL_CONTROL_GROUP] = "control-group",
1926 [KILL_PROCESS_GROUP] = "process-group",
1927 [KILL_PROCESS] = "process",
1928 [KILL_NONE] = "none"
1931 DEFINE_STRING_TABLE_LOOKUP(kill_mode, KillMode);
1933 static const char* const kill_who_table[_KILL_WHO_MAX] = {
1934 [KILL_MAIN] = "main",
1935 [KILL_CONTROL] = "control",
1939 DEFINE_STRING_TABLE_LOOKUP(kill_who, KillWho);