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 void exec_context_tty_reset(const ExecContext *context) {
146 if (context->tty_vhangup)
147 terminal_vhangup(tty_path(context));
149 if (context->tty_reset)
150 reset_terminal(tty_path(context));
152 if (context->tty_vt_disallocate && context->tty_path)
153 vt_disallocate(context->tty_path);
156 static int open_null_as(int flags, int nfd) {
161 if ((fd = open("/dev/null", flags|O_NOCTTY)) < 0)
165 r = dup2(fd, nfd) < 0 ? -errno : nfd;
166 close_nointr_nofail(fd);
173 static int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, int nfd) {
177 struct sockaddr_un un;
181 assert(output < _EXEC_OUTPUT_MAX);
185 if ((fd = socket(AF_UNIX, SOCK_STREAM, 0)) < 0)
189 sa.sa.sa_family = AF_UNIX;
190 strncpy(sa.un.sun_path, LOGGER_SOCKET, sizeof(sa.un.sun_path));
192 if (connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + sizeof(LOGGER_SOCKET) - 1) < 0) {
193 close_nointr_nofail(fd);
197 if (shutdown(fd, SHUT_RD) < 0) {
198 close_nointr_nofail(fd);
202 /* We speak a very simple protocol between log server
203 * and client: one line for the log destination (kmsg
204 * or syslog), followed by the priority field,
205 * followed by the process name. Since we replaced
206 * stdin/stderr we simple use stdio to write to
207 * it. Note that we use stderr, to minimize buffer
208 * flushing issues. */
215 output == EXEC_OUTPUT_KMSG ? "kmsg" :
216 output == EXEC_OUTPUT_KMSG_AND_CONSOLE ? "kmsg+console" :
217 output == EXEC_OUTPUT_SYSLOG ? "syslog" :
219 context->syslog_priority,
220 context->syslog_identifier ? context->syslog_identifier : ident,
221 context->syslog_level_prefix);
224 r = dup2(fd, nfd) < 0 ? -errno : nfd;
225 close_nointr_nofail(fd);
231 static int open_terminal_as(const char *path, mode_t mode, int nfd) {
237 if ((fd = open_terminal(path, mode | O_NOCTTY)) < 0)
241 r = dup2(fd, nfd) < 0 ? -errno : nfd;
242 close_nointr_nofail(fd);
249 static bool is_terminal_input(ExecInput i) {
251 i == EXEC_INPUT_TTY ||
252 i == EXEC_INPUT_TTY_FORCE ||
253 i == EXEC_INPUT_TTY_FAIL;
256 static int fixup_input(ExecInput std_input, int socket_fd, bool apply_tty_stdin) {
258 if (is_terminal_input(std_input) && !apply_tty_stdin)
259 return EXEC_INPUT_NULL;
261 if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0)
262 return EXEC_INPUT_NULL;
267 static int fixup_output(ExecOutput std_output, int socket_fd) {
269 if (std_output == EXEC_OUTPUT_SOCKET && socket_fd < 0)
270 return EXEC_OUTPUT_INHERIT;
275 static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty_stdin) {
280 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
284 case EXEC_INPUT_NULL:
285 return open_null_as(O_RDONLY, STDIN_FILENO);
288 case EXEC_INPUT_TTY_FORCE:
289 case EXEC_INPUT_TTY_FAIL: {
292 if ((fd = acquire_terminal(
294 i == EXEC_INPUT_TTY_FAIL,
295 i == EXEC_INPUT_TTY_FORCE,
299 if (fd != STDIN_FILENO) {
300 r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
301 close_nointr_nofail(fd);
308 case EXEC_INPUT_SOCKET:
309 return dup2(socket_fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
312 assert_not_reached("Unknown input type");
316 static int setup_output(const ExecContext *context, int socket_fd, const char *ident, bool apply_tty_stdin) {
323 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
324 o = fixup_output(context->std_output, socket_fd);
326 /* This expects the input is already set up */
330 case EXEC_OUTPUT_INHERIT:
332 /* If input got downgraded, inherit the original value */
333 if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input))
334 return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO);
336 /* If the input is connected to anything that's not a /dev/null, inherit that... */
337 if (i != EXEC_INPUT_NULL)
338 return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
340 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
342 return STDOUT_FILENO;
344 /* We need to open /dev/null here anew, to get the
345 * right access mode. So we fall through */
347 case EXEC_OUTPUT_NULL:
348 return open_null_as(O_WRONLY, STDOUT_FILENO);
350 case EXEC_OUTPUT_TTY:
351 if (is_terminal_input(i))
352 return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
354 /* We don't reset the terminal if this is just about output */
355 return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO);
357 case EXEC_OUTPUT_SYSLOG:
358 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
359 case EXEC_OUTPUT_KMSG:
360 case EXEC_OUTPUT_KMSG_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 return connect_logger_as(context, e, ident, STDERR_FILENO);
416 case EXEC_OUTPUT_SOCKET:
417 assert(socket_fd >= 0);
418 return dup2(socket_fd, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
421 assert_not_reached("Unknown error type");
425 static int chown_terminal(int fd, uid_t uid) {
430 /* This might fail. What matters are the results. */
431 (void) fchown(fd, uid, -1);
432 (void) fchmod(fd, TTY_MODE);
434 if (fstat(fd, &st) < 0)
437 if (st.st_uid != uid || (st.st_mode & 0777) != TTY_MODE)
443 static int setup_confirm_stdio(const ExecContext *context,
445 int *_saved_stdout) {
446 int fd = -1, saved_stdin, saved_stdout = -1, r;
449 assert(_saved_stdin);
450 assert(_saved_stdout);
452 /* This returns positive EXIT_xxx return values instead of
453 * negative errno style values! */
455 if ((saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3)) < 0)
458 if ((saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3)) < 0) {
463 if ((fd = acquire_terminal(
465 context->std_input == EXEC_INPUT_TTY_FAIL,
466 context->std_input == EXEC_INPUT_TTY_FORCE,
472 if (chown_terminal(fd, getuid()) < 0) {
477 if (dup2(fd, STDIN_FILENO) < 0) {
482 if (dup2(fd, STDOUT_FILENO) < 0) {
488 close_nointr_nofail(fd);
490 *_saved_stdin = saved_stdin;
491 *_saved_stdout = saved_stdout;
496 if (saved_stdout >= 0)
497 close_nointr_nofail(saved_stdout);
499 if (saved_stdin >= 0)
500 close_nointr_nofail(saved_stdin);
503 close_nointr_nofail(fd);
508 static int restore_confirm_stdio(const ExecContext *context,
516 assert(*saved_stdin >= 0);
517 assert(saved_stdout);
518 assert(*saved_stdout >= 0);
520 /* This returns positive EXIT_xxx return values instead of
521 * negative errno style values! */
523 if (is_terminal_input(context->std_input)) {
525 /* The service wants terminal input. */
529 context->std_output == EXEC_OUTPUT_INHERIT ||
530 context->std_output == EXEC_OUTPUT_TTY;
533 /* If the service doesn't want a controlling terminal,
534 * then we need to get rid entirely of what we have
537 if (release_terminal() < 0)
540 if (dup2(*saved_stdin, STDIN_FILENO) < 0)
543 if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
546 *keep_stdout = *keep_stdin = false;
552 static int get_group_creds(const char *groupname, gid_t *gid) {
559 /* We enforce some special rules for gid=0: in order to avoid
560 * NSS lookups for root we hardcode its data. */
562 if (streq(groupname, "root") || streq(groupname, "0")) {
567 if (safe_atolu(groupname, &lu) >= 0) {
569 g = getgrgid((gid_t) lu);
572 g = getgrnam(groupname);
576 return errno != 0 ? -errno : -ESRCH;
582 static int get_user_creds(const char **username, uid_t *uid, gid_t *gid, const char **home) {
592 /* We enforce some special rules for uid=0: in order to avoid
593 * NSS lookups for root we hardcode its data. */
595 if (streq(*username, "root") || streq(*username, "0")) {
603 if (safe_atolu(*username, &lu) >= 0) {
605 p = getpwuid((uid_t) lu);
607 /* If there are multiple users with the same id, make
608 * sure to leave $USER to the configured value instead
609 * of the first occurrence in the database. However if
610 * the uid was configured by a numeric uid, then let's
611 * pick the real username from /etc/passwd. */
613 *username = p->pw_name;
616 p = getpwnam(*username);
620 return errno != 0 ? -errno : -ESRCH;
628 static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
629 bool keep_groups = false;
634 /* Lookup and set GID and supplementary group list. Here too
635 * we avoid NSS lookups for gid=0. */
637 if (context->group || username) {
640 if ((r = get_group_creds(context->group, &gid)) < 0)
643 /* First step, initialize groups from /etc/groups */
644 if (username && gid != 0) {
645 if (initgroups(username, gid) < 0)
651 /* Second step, set our gids */
652 if (setresgid(gid, gid, gid) < 0)
656 if (context->supplementary_groups) {
661 /* Final step, initialize any manually set supplementary groups */
662 assert_se((ngroups_max = (int) sysconf(_SC_NGROUPS_MAX)) > 0);
664 if (!(gids = new(gid_t, ngroups_max)))
668 if ((k = getgroups(ngroups_max, gids)) < 0) {
675 STRV_FOREACH(i, context->supplementary_groups) {
677 if (k >= ngroups_max) {
682 if ((r = get_group_creds(*i, gids+k)) < 0) {
690 if (setgroups(k, gids) < 0) {
701 static int enforce_user(const ExecContext *context, uid_t uid) {
705 /* Sets (but doesn't lookup) the uid and make sure we keep the
706 * capabilities while doing so. */
708 if (context->capabilities) {
710 static const cap_value_t bits[] = {
711 CAP_SETUID, /* Necessary so that we can run setresuid() below */
712 CAP_SETPCAP /* Necessary so that we can set PR_SET_SECUREBITS later on */
715 /* First step: If we need to keep capabilities but
716 * drop privileges we need to make sure we keep our
717 * caps, whiel we drop privileges. */
719 int sb = context->secure_bits|SECURE_KEEP_CAPS;
721 if (prctl(PR_GET_SECUREBITS) != sb)
722 if (prctl(PR_SET_SECUREBITS, sb) < 0)
726 /* Second step: set the capabilities. This will reduce
727 * the capabilities to the minimum we need. */
729 if (!(d = cap_dup(context->capabilities)))
732 if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
733 cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0) {
739 if (cap_set_proc(d) < 0) {
748 /* Third step: actually set the uids */
749 if (setresuid(uid, uid, uid) < 0)
752 /* At this point we should have all necessary capabilities but
753 are otherwise a normal user. However, the caps might got
754 corrupted due to the setresuid() so we need clean them up
755 later. This is done outside of this call. */
762 static int null_conv(
764 const struct pam_message **msg,
765 struct pam_response **resp,
768 /* We don't support conversations */
773 static int setup_pam(
778 int fds[], unsigned n_fds) {
780 static const struct pam_conv conv = {
785 pam_handle_t *handle = NULL;
787 int pam_code = PAM_SUCCESS;
789 bool close_session = false;
790 pid_t pam_pid = 0, parent_pid;
796 /* We set up PAM in the parent process, then fork. The child
797 * will then stay around until killed via PR_GET_PDEATHSIG or
798 * systemd via the cgroup logic. It will then remove the PAM
799 * session again. The parent process will exec() the actual
800 * daemon. We do things this way to ensure that the main PID
801 * of the daemon is the one we initially fork()ed. */
803 if ((pam_code = pam_start(name, user, &conv, &handle)) != PAM_SUCCESS) {
809 if ((pam_code = pam_set_item(handle, PAM_TTY, tty)) != PAM_SUCCESS)
812 if ((pam_code = pam_acct_mgmt(handle, PAM_SILENT)) != PAM_SUCCESS)
815 if ((pam_code = pam_open_session(handle, PAM_SILENT)) != PAM_SUCCESS)
818 close_session = true;
820 if ((!(e = pam_getenvlist(handle)))) {
821 pam_code = PAM_BUF_ERR;
825 /* Block SIGTERM, so that we know that it won't get lost in
827 if (sigemptyset(&ss) < 0 ||
828 sigaddset(&ss, SIGTERM) < 0 ||
829 sigprocmask(SIG_BLOCK, &ss, &old_ss) < 0)
832 parent_pid = getpid();
834 if ((pam_pid = fork()) < 0)
841 /* The child's job is to reset the PAM session on
844 /* This string must fit in 10 chars (i.e. the length
845 * of "/sbin/init") */
846 rename_process("sd(PAM)");
848 /* Make sure we don't keep open the passed fds in this
849 child. We assume that otherwise only those fds are
850 open here that have been opened by PAM. */
851 close_many(fds, n_fds);
853 /* Wait until our parent died. This will most likely
854 * not work since the kernel does not allow
855 * unprivileged parents kill their privileged children
856 * this way. We rely on the control groups kill logic
857 * to do the rest for us. */
858 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
861 /* Check if our parent process might already have
863 if (getppid() == parent_pid) {
865 if (sigwait(&ss, &sig) < 0) {
872 assert(sig == SIGTERM);
877 /* If our parent died we'll end the session */
878 if (getppid() != parent_pid)
879 if ((pam_code = pam_close_session(handle, PAM_DATA_SILENT)) != PAM_SUCCESS)
885 pam_end(handle, pam_code | PAM_DATA_SILENT);
889 /* If the child was forked off successfully it will do all the
890 * cleanups, so forget about the handle here. */
893 /* Unblock SIGTERM again in the parent */
894 if (sigprocmask(SIG_SETMASK, &old_ss, NULL) < 0)
897 /* We close the log explicitly here, since the PAM modules
898 * might have opened it, but we don't want this fd around. */
906 pam_code = pam_close_session(handle, PAM_DATA_SILENT);
908 pam_end(handle, pam_code | PAM_DATA_SILENT);
916 kill(pam_pid, SIGTERM);
917 kill(pam_pid, SIGCONT);
924 static int do_capability_bounding_set_drop(uint64_t drop) {
926 cap_t old_cap = NULL, new_cap = NULL;
930 /* If we are run as PID 1 we will lack CAP_SETPCAP by default
931 * in the effective set (yes, the kernel drops that when
932 * executing init!), so get it back temporarily so that we can
933 * call PR_CAPBSET_DROP. */
935 old_cap = cap_get_proc();
939 if (cap_get_flag(old_cap, CAP_SETPCAP, CAP_EFFECTIVE, &fv) < 0) {
945 static const cap_value_t v = CAP_SETPCAP;
947 new_cap = cap_dup(old_cap);
953 if (cap_set_flag(new_cap, CAP_EFFECTIVE, 1, &v, CAP_SET) < 0) {
958 if (cap_set_proc(new_cap) < 0) {
964 for (i = 0; i <= MAX(63LU, (unsigned long) CAP_LAST_CAP); i++)
965 if (drop & ((uint64_t) 1ULL << (uint64_t) i)) {
966 if (prctl(PR_CAPBSET_DROP, i) < 0) {
982 cap_set_proc(old_cap);
989 int exec_spawn(ExecCommand *command,
991 const ExecContext *context,
992 int fds[], unsigned n_fds,
994 bool apply_permissions,
996 bool apply_tty_stdin,
998 CGroupBonding *cgroup_bondings,
1005 char **files_env = NULL;
1010 assert(fds || n_fds <= 0);
1012 if (context->std_input == EXEC_INPUT_SOCKET ||
1013 context->std_output == EXEC_OUTPUT_SOCKET ||
1014 context->std_error == EXEC_OUTPUT_SOCKET) {
1026 if ((r = exec_context_load_environment(context, &files_env)) < 0) {
1027 log_error("Failed to load environment files: %s", strerror(-r));
1032 argv = command->argv;
1034 if (!(line = exec_command_line(argv))) {
1039 log_debug("About to execute: %s", line);
1042 if (cgroup_bondings)
1043 if ((r = cgroup_bonding_realize_list(cgroup_bondings)))
1046 if ((pid = fork()) < 0) {
1054 const char *username = NULL, *home = NULL;
1055 uid_t uid = (uid_t) -1;
1056 gid_t gid = (gid_t) -1;
1057 char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
1059 int saved_stdout = -1, saved_stdin = -1;
1060 bool keep_stdout = false, keep_stdin = false;
1064 /* This string must fit in 10 chars (i.e. the length
1065 * of "/sbin/init") */
1066 rename_process("sd.exec");
1068 /* We reset exactly these signals, since they are the
1069 * only ones we set to SIG_IGN in the main daemon. All
1070 * others we leave untouched because we set them to
1071 * SIG_DFL or a valid handler initially, both of which
1072 * will be demoted to SIG_DFL. */
1073 default_signals(SIGNALS_CRASH_HANDLER,
1074 SIGNALS_IGNORE, -1);
1076 if (sigemptyset(&ss) < 0 ||
1077 sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
1078 r = EXIT_SIGNAL_MASK;
1082 /* Close sockets very early to make sure we don't
1083 * block init reexecution because it cannot bind its
1085 if (close_all_fds(socket_fd >= 0 ? &socket_fd : fds,
1086 socket_fd >= 0 ? 1 : n_fds) < 0) {
1091 if (!context->same_pgrp)
1097 if (context->tcpwrap_name) {
1099 if (!socket_tcpwrap(socket_fd, context->tcpwrap_name)) {
1104 for (i = 0; i < (int) n_fds; i++) {
1105 if (!socket_tcpwrap(fds[i], context->tcpwrap_name)) {
1112 exec_context_tty_reset(context);
1114 /* We skip the confirmation step if we shall not apply the TTY */
1115 if (confirm_spawn &&
1116 (!is_terminal_input(context->std_input) || apply_tty_stdin)) {
1119 /* Set up terminal for the question */
1120 if ((r = setup_confirm_stdio(context,
1121 &saved_stdin, &saved_stdout)))
1124 /* Now ask the question. */
1125 if (!(line = exec_command_line(argv))) {
1130 r = ask(&response, "yns", "Execute %s? [Yes, No, Skip] ", line);
1133 if (r < 0 || response == 'n') {
1136 } else if (response == 's') {
1141 /* Release terminal for the question */
1142 if ((r = restore_confirm_stdio(context,
1143 &saved_stdin, &saved_stdout,
1144 &keep_stdin, &keep_stdout)))
1148 /* If a socket is connected to STDIN/STDOUT/STDERR, we
1149 * must sure to drop O_NONBLOCK */
1151 fd_nonblock(socket_fd, false);
1154 if (setup_input(context, socket_fd, apply_tty_stdin) < 0) {
1160 if (setup_output(context, socket_fd, file_name_from_path(command->path), apply_tty_stdin) < 0) {
1165 if (setup_error(context, socket_fd, file_name_from_path(command->path), apply_tty_stdin) < 0) {
1170 if (cgroup_bondings)
1171 if (cgroup_bonding_install_list(cgroup_bondings, 0) < 0) {
1176 if (context->oom_score_adjust_set) {
1179 snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
1182 if (write_one_line_file("/proc/self/oom_score_adj", t) < 0) {
1183 /* Compatibility with Linux <= 2.6.35 */
1187 adj = (context->oom_score_adjust * -OOM_DISABLE) / OOM_SCORE_ADJ_MAX;
1188 adj = CLAMP(adj, OOM_DISABLE, OOM_ADJUST_MAX);
1190 snprintf(t, sizeof(t), "%i", adj);
1193 if (write_one_line_file("/proc/self/oom_adj", t) < 0
1194 && errno != EACCES) {
1195 r = EXIT_OOM_ADJUST;
1201 if (context->nice_set)
1202 if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
1207 if (context->cpu_sched_set) {
1208 struct sched_param param;
1211 param.sched_priority = context->cpu_sched_priority;
1213 if (sched_setscheduler(0, context->cpu_sched_policy |
1214 (context->cpu_sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0), ¶m) < 0) {
1215 r = EXIT_SETSCHEDULER;
1220 if (context->cpuset)
1221 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
1222 r = EXIT_CPUAFFINITY;
1226 if (context->ioprio_set)
1227 if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
1232 if (context->timer_slack_nsec_set)
1233 if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
1234 r = EXIT_TIMERSLACK;
1238 if (context->utmp_id)
1239 utmp_put_init_process(0, context->utmp_id, getpid(), getsid(0), context->tty_path);
1241 if (context->user) {
1242 username = context->user;
1243 if (get_user_creds(&username, &uid, &gid, &home) < 0) {
1248 if (is_terminal_input(context->std_input))
1249 if (chown_terminal(STDIN_FILENO, uid) < 0) {
1254 if (cgroup_bondings && context->control_group_modify)
1255 if (cgroup_bonding_set_group_access_list(cgroup_bondings, 0755, uid, gid) < 0 ||
1256 cgroup_bonding_set_task_access_list(cgroup_bondings, 0644, uid, gid) < 0) {
1262 if (apply_permissions)
1263 if (enforce_groups(context, username, uid) < 0) {
1268 umask(context->umask);
1271 if (context->pam_name && username) {
1272 if (setup_pam(context->pam_name, username, context->tty_path, &pam_env, fds, n_fds) != 0) {
1279 if (strv_length(context->read_write_dirs) > 0 ||
1280 strv_length(context->read_only_dirs) > 0 ||
1281 strv_length(context->inaccessible_dirs) > 0 ||
1282 context->mount_flags != MS_SHARED ||
1283 context->private_tmp)
1284 if ((r = setup_namespace(
1285 context->read_write_dirs,
1286 context->read_only_dirs,
1287 context->inaccessible_dirs,
1288 context->private_tmp,
1289 context->mount_flags)) < 0)
1293 if (context->root_directory)
1294 if (chroot(context->root_directory) < 0) {
1299 if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
1307 if (asprintf(&d, "%s/%s",
1308 context->root_directory ? context->root_directory : "",
1309 context->working_directory ? context->working_directory : "") < 0) {
1323 /* We repeat the fd closing here, to make sure that
1324 * nothing is leaked from the PAM modules */
1325 if (close_all_fds(fds, n_fds) < 0 ||
1326 shift_fds(fds, n_fds) < 0 ||
1327 flags_fds(fds, n_fds, context->non_blocking) < 0) {
1332 if (apply_permissions) {
1334 for (i = 0; i < RLIMIT_NLIMITS; i++) {
1335 if (!context->rlimit[i])
1338 if (setrlimit(i, context->rlimit[i]) < 0) {
1344 if (context->capability_bounding_set_drop)
1345 if (do_capability_bounding_set_drop(context->capability_bounding_set_drop) < 0) {
1346 r = EXIT_CAPABILITIES;
1351 if (enforce_user(context, uid) < 0) {
1356 /* PR_GET_SECUREBITS is not privileged, while
1357 * PR_SET_SECUREBITS is. So to suppress
1358 * potential EPERMs we'll try not to call
1359 * PR_SET_SECUREBITS unless necessary. */
1360 if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
1361 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
1362 r = EXIT_SECUREBITS;
1366 if (context->capabilities)
1367 if (cap_set_proc(context->capabilities) < 0) {
1368 r = EXIT_CAPABILITIES;
1373 if (!(our_env = new0(char*, 7))) {
1379 if (asprintf(our_env + n_env++, "LISTEN_PID=%lu", (unsigned long) getpid()) < 0 ||
1380 asprintf(our_env + n_env++, "LISTEN_FDS=%u", n_fds) < 0) {
1386 if (asprintf(our_env + n_env++, "HOME=%s", home) < 0) {
1392 if (asprintf(our_env + n_env++, "LOGNAME=%s", username) < 0 ||
1393 asprintf(our_env + n_env++, "USER=%s", username) < 0) {
1398 if (is_terminal_input(context->std_input) ||
1399 context->std_output == EXEC_OUTPUT_TTY ||
1400 context->std_error == EXEC_OUTPUT_TTY)
1401 if (!(our_env[n_env++] = strdup(default_term_for_tty(tty_path(context))))) {
1408 if (!(final_env = strv_env_merge(
1412 context->environment,
1420 if (!(final_argv = replace_env_argv(argv, final_env))) {
1425 final_env = strv_env_clean(final_env);
1427 execve(command->path, final_argv, final_env);
1432 strv_free(final_env);
1434 strv_free(files_env);
1435 strv_free(final_argv);
1437 if (saved_stdin >= 0)
1438 close_nointr_nofail(saved_stdin);
1440 if (saved_stdout >= 0)
1441 close_nointr_nofail(saved_stdout);
1446 strv_free(files_env);
1448 /* We add the new process to the cgroup both in the child (so
1449 * that we can be sure that no user code is ever executed
1450 * outside of the cgroup) and in the parent (so that we can be
1451 * sure that when we kill the cgroup the process will be
1453 if (cgroup_bondings)
1454 cgroup_bonding_install_list(cgroup_bondings, pid);
1456 log_debug("Forked %s as %lu", command->path, (unsigned long) pid);
1458 exec_status_start(&command->exec_status, pid);
1464 strv_free(files_env);
1469 void exec_context_init(ExecContext *c) {
1473 c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
1474 c->cpu_sched_policy = SCHED_OTHER;
1475 c->syslog_priority = LOG_DAEMON|LOG_INFO;
1476 c->syslog_level_prefix = true;
1477 c->mount_flags = MS_SHARED;
1478 c->kill_signal = SIGTERM;
1479 c->send_sigkill = true;
1482 void exec_context_done(ExecContext *c) {
1487 strv_free(c->environment);
1488 c->environment = NULL;
1490 strv_free(c->environment_files);
1491 c->environment_files = NULL;
1493 for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1495 c->rlimit[l] = NULL;
1498 free(c->working_directory);
1499 c->working_directory = NULL;
1500 free(c->root_directory);
1501 c->root_directory = NULL;
1506 free(c->tcpwrap_name);
1507 c->tcpwrap_name = NULL;
1509 free(c->syslog_identifier);
1510 c->syslog_identifier = NULL;
1518 strv_free(c->supplementary_groups);
1519 c->supplementary_groups = NULL;
1524 if (c->capabilities) {
1525 cap_free(c->capabilities);
1526 c->capabilities = NULL;
1529 strv_free(c->read_only_dirs);
1530 c->read_only_dirs = NULL;
1532 strv_free(c->read_write_dirs);
1533 c->read_write_dirs = NULL;
1535 strv_free(c->inaccessible_dirs);
1536 c->inaccessible_dirs = NULL;
1539 CPU_FREE(c->cpuset);
1545 void exec_command_done(ExecCommand *c) {
1555 void exec_command_done_array(ExecCommand *c, unsigned n) {
1558 for (i = 0; i < n; i++)
1559 exec_command_done(c+i);
1562 void exec_command_free_list(ExecCommand *c) {
1566 LIST_REMOVE(ExecCommand, command, c, i);
1567 exec_command_done(i);
1572 void exec_command_free_array(ExecCommand **c, unsigned n) {
1575 for (i = 0; i < n; i++) {
1576 exec_command_free_list(c[i]);
1581 int exec_context_load_environment(const ExecContext *c, char ***l) {
1582 char **i, **r = NULL;
1587 STRV_FOREACH(i, c->environment_files) {
1590 bool ignore = false;
1600 if (!path_is_absolute(fn)) {
1609 if ((k = load_env_file(fn, &p)) < 0) {
1623 m = strv_env_merge(2, r, p);
1639 static void strv_fprintf(FILE *f, char **l) {
1645 fprintf(f, " %s", *g);
1648 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
1660 "%sWorkingDirectory: %s\n"
1661 "%sRootDirectory: %s\n"
1662 "%sNonBlocking: %s\n"
1663 "%sPrivateTmp: %s\n"
1664 "%sControlGroupModify: %s\n",
1666 prefix, c->working_directory ? c->working_directory : "/",
1667 prefix, c->root_directory ? c->root_directory : "/",
1668 prefix, yes_no(c->non_blocking),
1669 prefix, yes_no(c->private_tmp),
1670 prefix, yes_no(c->control_group_modify));
1672 STRV_FOREACH(e, c->environment)
1673 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
1675 STRV_FOREACH(e, c->environment_files)
1676 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
1678 if (c->tcpwrap_name)
1680 "%sTCPWrapName: %s\n",
1681 prefix, c->tcpwrap_name);
1688 if (c->oom_score_adjust_set)
1690 "%sOOMScoreAdjust: %i\n",
1691 prefix, c->oom_score_adjust);
1693 for (i = 0; i < RLIM_NLIMITS; i++)
1695 fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
1699 "%sIOSchedulingClass: %s\n"
1700 "%sIOPriority: %i\n",
1701 prefix, ioprio_class_to_string(IOPRIO_PRIO_CLASS(c->ioprio)),
1702 prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
1704 if (c->cpu_sched_set)
1706 "%sCPUSchedulingPolicy: %s\n"
1707 "%sCPUSchedulingPriority: %i\n"
1708 "%sCPUSchedulingResetOnFork: %s\n",
1709 prefix, sched_policy_to_string(c->cpu_sched_policy),
1710 prefix, c->cpu_sched_priority,
1711 prefix, yes_no(c->cpu_sched_reset_on_fork));
1714 fprintf(f, "%sCPUAffinity:", prefix);
1715 for (i = 0; i < c->cpuset_ncpus; i++)
1716 if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
1717 fprintf(f, " %i", i);
1721 if (c->timer_slack_nsec_set)
1722 fprintf(f, "%sTimerSlackNSec: %lu\n", prefix, c->timer_slack_nsec);
1725 "%sStandardInput: %s\n"
1726 "%sStandardOutput: %s\n"
1727 "%sStandardError: %s\n",
1728 prefix, exec_input_to_string(c->std_input),
1729 prefix, exec_output_to_string(c->std_output),
1730 prefix, exec_output_to_string(c->std_error));
1736 "%sTTYVHangup: %s\n"
1737 "%sTTYVTDisallocate: %s\n",
1738 prefix, c->tty_path,
1739 prefix, yes_no(c->tty_reset),
1740 prefix, yes_no(c->tty_vhangup),
1741 prefix, yes_no(c->tty_vt_disallocate));
1743 if (c->std_output == EXEC_OUTPUT_SYSLOG || c->std_output == EXEC_OUTPUT_KMSG ||
1744 c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
1745 c->std_error == EXEC_OUTPUT_SYSLOG || c->std_error == EXEC_OUTPUT_KMSG ||
1746 c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE)
1748 "%sSyslogFacility: %s\n"
1749 "%sSyslogLevel: %s\n",
1750 prefix, log_facility_unshifted_to_string(c->syslog_priority >> 3),
1751 prefix, log_level_to_string(LOG_PRI(c->syslog_priority)));
1753 if (c->capabilities) {
1755 if ((t = cap_to_text(c->capabilities, NULL))) {
1756 fprintf(f, "%sCapabilities: %s\n",
1763 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
1765 (c->secure_bits & SECURE_KEEP_CAPS) ? " keep-caps" : "",
1766 (c->secure_bits & SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
1767 (c->secure_bits & SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
1768 (c->secure_bits & SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
1769 (c->secure_bits & SECURE_NOROOT) ? " noroot" : "",
1770 (c->secure_bits & SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
1772 if (c->capability_bounding_set_drop) {
1774 fprintf(f, "%sCapabilityBoundingSet:", prefix);
1776 for (l = 0; l <= (unsigned long) CAP_LAST_CAP; l++)
1777 if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) {
1780 if ((t = cap_to_name(l))) {
1781 fprintf(f, " %s", t);
1790 fprintf(f, "%sUser: %s\n", prefix, c->user);
1792 fprintf(f, "%sGroup: %s\n", prefix, c->group);
1794 if (strv_length(c->supplementary_groups) > 0) {
1795 fprintf(f, "%sSupplementaryGroups:", prefix);
1796 strv_fprintf(f, c->supplementary_groups);
1801 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
1803 if (strv_length(c->read_write_dirs) > 0) {
1804 fprintf(f, "%sReadWriteDirs:", prefix);
1805 strv_fprintf(f, c->read_write_dirs);
1809 if (strv_length(c->read_only_dirs) > 0) {
1810 fprintf(f, "%sReadOnlyDirs:", prefix);
1811 strv_fprintf(f, c->read_only_dirs);
1815 if (strv_length(c->inaccessible_dirs) > 0) {
1816 fprintf(f, "%sInaccessibleDirs:", prefix);
1817 strv_fprintf(f, c->inaccessible_dirs);
1823 "%sKillSignal: SIG%s\n"
1824 "%sSendSIGKILL: %s\n",
1825 prefix, kill_mode_to_string(c->kill_mode),
1826 prefix, signal_to_string(c->kill_signal),
1827 prefix, yes_no(c->send_sigkill));
1831 "%sUtmpIdentifier: %s\n",
1832 prefix, c->utmp_id);
1835 void exec_status_start(ExecStatus *s, pid_t pid) {
1840 dual_timestamp_get(&s->start_timestamp);
1843 void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
1846 if ((s->pid && s->pid != pid) ||
1847 !s->start_timestamp.realtime <= 0)
1851 dual_timestamp_get(&s->exit_timestamp);
1857 if (context->utmp_id)
1858 utmp_put_dead_process(context->utmp_id, pid, code, status);
1860 exec_context_tty_reset(context);
1864 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
1865 char buf[FORMAT_TIMESTAMP_MAX];
1878 prefix, (unsigned long) s->pid);
1880 if (s->start_timestamp.realtime > 0)
1882 "%sStart Timestamp: %s\n",
1883 prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
1885 if (s->exit_timestamp.realtime > 0)
1887 "%sExit Timestamp: %s\n"
1889 "%sExit Status: %i\n",
1890 prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
1891 prefix, sigchld_code_to_string(s->code),
1895 char *exec_command_line(char **argv) {
1903 STRV_FOREACH(a, argv)
1906 if (!(n = new(char, k)))
1910 STRV_FOREACH(a, argv) {
1917 if (strpbrk(*a, WHITESPACE)) {
1928 /* FIXME: this doesn't really handle arguments that have
1929 * spaces and ticks in them */
1934 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
1936 const char *prefix2;
1945 p2 = strappend(prefix, "\t");
1946 prefix2 = p2 ? p2 : prefix;
1948 cmd = exec_command_line(c->argv);
1951 "%sCommand Line: %s\n",
1952 prefix, cmd ? cmd : strerror(ENOMEM));
1956 exec_status_dump(&c->exec_status, f, prefix2);
1961 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
1967 LIST_FOREACH(command, c, c)
1968 exec_command_dump(c, f, prefix);
1971 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
1978 /* It's kind of important, that we keep the order here */
1979 LIST_FIND_TAIL(ExecCommand, command, *l, end);
1980 LIST_INSERT_AFTER(ExecCommand, command, *l, end, e);
1985 int exec_command_set(ExecCommand *c, const char *path, ...) {
1993 l = strv_new_ap(path, ap);
1999 if (!(p = strdup(path))) {
2013 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
2014 [EXEC_INPUT_NULL] = "null",
2015 [EXEC_INPUT_TTY] = "tty",
2016 [EXEC_INPUT_TTY_FORCE] = "tty-force",
2017 [EXEC_INPUT_TTY_FAIL] = "tty-fail",
2018 [EXEC_INPUT_SOCKET] = "socket"
2021 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
2023 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
2024 [EXEC_OUTPUT_INHERIT] = "inherit",
2025 [EXEC_OUTPUT_NULL] = "null",
2026 [EXEC_OUTPUT_TTY] = "tty",
2027 [EXEC_OUTPUT_SYSLOG] = "syslog",
2028 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
2029 [EXEC_OUTPUT_KMSG] = "kmsg",
2030 [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
2031 [EXEC_OUTPUT_SOCKET] = "socket"
2034 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
2036 static const char* const kill_mode_table[_KILL_MODE_MAX] = {
2037 [KILL_CONTROL_GROUP] = "control-group",
2038 [KILL_PROCESS] = "process",
2039 [KILL_NONE] = "none"
2042 DEFINE_STRING_TABLE_LOOKUP(kill_mode, KillMode);
2044 static const char* const kill_who_table[_KILL_WHO_MAX] = {
2045 [KILL_MAIN] = "main",
2046 [KILL_CONTROL] = "control",
2050 DEFINE_STRING_TABLE_LOOKUP(kill_who, KillWho);