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>
51 #include <security/pam_appl.h>
55 #include <selinux/selinux.h>
61 #include "capability.h"
64 #include "sd-messages.h"
66 #include "securebits.h"
67 #include "namespace.h"
69 #include "exit-status.h"
71 #include "utmp-wtmp.h"
73 #include "path-util.h"
78 #include "selinux-util.h"
80 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
81 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
83 /* This assumes there is a 'tty' group */
86 #define SNDBUF_SIZE (8*1024*1024)
88 static int shift_fds(int fds[], unsigned n_fds) {
89 int start, restart_from;
94 /* Modifies the fds array! (sorts it) */
104 for (i = start; i < (int) n_fds; i++) {
107 /* Already at right index? */
111 if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
114 close_nointr_nofail(fds[i]);
117 /* Hmm, the fd we wanted isn't free? Then
118 * let's remember that and try again from here*/
119 if (nfd != i+3 && restart_from < 0)
123 if (restart_from < 0)
126 start = restart_from;
132 static int flags_fds(const int fds[], unsigned n_fds, bool nonblock) {
141 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
143 for (i = 0; i < n_fds; i++) {
145 if ((r = fd_nonblock(fds[i], nonblock)) < 0)
148 /* We unconditionally drop FD_CLOEXEC from the fds,
149 * since after all we want to pass these fds to our
152 if ((r = fd_cloexec(fds[i], false)) < 0)
159 _pure_ static const char *tty_path(const ExecContext *context) {
162 if (context->tty_path)
163 return context->tty_path;
165 return "/dev/console";
168 static void exec_context_tty_reset(const ExecContext *context) {
171 if (context->tty_vhangup)
172 terminal_vhangup(tty_path(context));
174 if (context->tty_reset)
175 reset_terminal(tty_path(context));
177 if (context->tty_vt_disallocate && context->tty_path)
178 vt_disallocate(context->tty_path);
181 static bool is_terminal_output(ExecOutput o) {
183 o == EXEC_OUTPUT_TTY ||
184 o == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
185 o == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
186 o == EXEC_OUTPUT_JOURNAL_AND_CONSOLE;
189 static int open_null_as(int flags, int nfd) {
194 fd = open("/dev/null", flags|O_NOCTTY);
199 r = dup2(fd, nfd) < 0 ? -errno : nfd;
200 close_nointr_nofail(fd);
207 static int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, const char *unit_id, int nfd) {
209 union sockaddr_union sa = {
210 .un.sun_family = AF_UNIX,
211 .un.sun_path = "/run/systemd/journal/stdout",
215 assert(output < _EXEC_OUTPUT_MAX);
219 fd = socket(AF_UNIX, SOCK_STREAM, 0);
223 r = connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path));
225 close_nointr_nofail(fd);
229 if (shutdown(fd, SHUT_RD) < 0) {
230 close_nointr_nofail(fd);
234 fd_inc_sndbuf(fd, SNDBUF_SIZE);
244 context->syslog_identifier ? context->syslog_identifier : ident,
246 context->syslog_priority,
247 !!context->syslog_level_prefix,
248 output == EXEC_OUTPUT_SYSLOG || output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE,
249 output == EXEC_OUTPUT_KMSG || output == EXEC_OUTPUT_KMSG_AND_CONSOLE,
250 is_terminal_output(output));
253 r = dup2(fd, nfd) < 0 ? -errno : nfd;
254 close_nointr_nofail(fd);
260 static int open_terminal_as(const char *path, mode_t mode, int nfd) {
266 if ((fd = open_terminal(path, mode | O_NOCTTY)) < 0)
270 r = dup2(fd, nfd) < 0 ? -errno : nfd;
271 close_nointr_nofail(fd);
278 static bool is_terminal_input(ExecInput i) {
280 i == EXEC_INPUT_TTY ||
281 i == EXEC_INPUT_TTY_FORCE ||
282 i == EXEC_INPUT_TTY_FAIL;
285 static int fixup_input(ExecInput std_input, int socket_fd, bool apply_tty_stdin) {
287 if (is_terminal_input(std_input) && !apply_tty_stdin)
288 return EXEC_INPUT_NULL;
290 if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0)
291 return EXEC_INPUT_NULL;
296 static int fixup_output(ExecOutput std_output, int socket_fd) {
298 if (std_output == EXEC_OUTPUT_SOCKET && socket_fd < 0)
299 return EXEC_OUTPUT_INHERIT;
304 static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty_stdin) {
309 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
313 case EXEC_INPUT_NULL:
314 return open_null_as(O_RDONLY, STDIN_FILENO);
317 case EXEC_INPUT_TTY_FORCE:
318 case EXEC_INPUT_TTY_FAIL: {
321 fd = acquire_terminal(tty_path(context),
322 i == EXEC_INPUT_TTY_FAIL,
323 i == EXEC_INPUT_TTY_FORCE,
329 if (fd != STDIN_FILENO) {
330 r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
331 close_nointr_nofail(fd);
338 case EXEC_INPUT_SOCKET:
339 return dup2(socket_fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
342 assert_not_reached("Unknown input type");
346 static int setup_output(const ExecContext *context, int fileno, int socket_fd, const char *ident, const char *unit_id, bool apply_tty_stdin) {
354 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
355 o = fixup_output(context->std_output, socket_fd);
357 if (fileno == STDERR_FILENO) {
359 e = fixup_output(context->std_error, socket_fd);
361 /* This expects the input and output are already set up */
363 /* Don't change the stderr file descriptor if we inherit all
364 * the way and are not on a tty */
365 if (e == EXEC_OUTPUT_INHERIT &&
366 o == EXEC_OUTPUT_INHERIT &&
367 i == EXEC_INPUT_NULL &&
368 !is_terminal_input(context->std_input) &&
372 /* Duplicate from stdout if possible */
373 if (e == o || e == EXEC_OUTPUT_INHERIT)
374 return dup2(STDOUT_FILENO, fileno) < 0 ? -errno : fileno;
378 } else if (o == EXEC_OUTPUT_INHERIT) {
379 /* If input got downgraded, inherit the original value */
380 if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input))
381 return open_terminal_as(tty_path(context), O_WRONLY, fileno);
383 /* If the input is connected to anything that's not a /dev/null, inherit that... */
384 if (i != EXEC_INPUT_NULL)
385 return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
387 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
391 /* We need to open /dev/null here anew, to get the right access mode. */
392 return open_null_as(O_WRONLY, fileno);
397 case EXEC_OUTPUT_NULL:
398 return open_null_as(O_WRONLY, fileno);
400 case EXEC_OUTPUT_TTY:
401 if (is_terminal_input(i))
402 return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
404 /* We don't reset the terminal if this is just about output */
405 return open_terminal_as(tty_path(context), O_WRONLY, fileno);
407 case EXEC_OUTPUT_SYSLOG:
408 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
409 case EXEC_OUTPUT_KMSG:
410 case EXEC_OUTPUT_KMSG_AND_CONSOLE:
411 case EXEC_OUTPUT_JOURNAL:
412 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE:
413 r = connect_logger_as(context, o, ident, unit_id, fileno);
415 log_struct_unit(LOG_CRIT, unit_id,
416 "MESSAGE=Failed to connect std%s of %s to the journal socket: %s",
417 fileno == STDOUT_FILENO ? "out" : "err",
418 unit_id, strerror(-r),
421 r = open_null_as(O_WRONLY, fileno);
425 case EXEC_OUTPUT_SOCKET:
426 assert(socket_fd >= 0);
427 return dup2(socket_fd, fileno) < 0 ? -errno : fileno;
430 assert_not_reached("Unknown error type");
434 static int chown_terminal(int fd, uid_t uid) {
439 /* This might fail. What matters are the results. */
440 (void) fchown(fd, uid, -1);
441 (void) fchmod(fd, TTY_MODE);
443 if (fstat(fd, &st) < 0)
446 if (st.st_uid != uid || (st.st_mode & 0777) != TTY_MODE)
452 static int setup_confirm_stdio(int *_saved_stdin,
453 int *_saved_stdout) {
454 int fd = -1, saved_stdin, saved_stdout = -1, r;
456 assert(_saved_stdin);
457 assert(_saved_stdout);
459 saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3);
463 saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3);
464 if (saved_stdout < 0) {
469 fd = acquire_terminal(
474 DEFAULT_CONFIRM_USEC);
480 r = chown_terminal(fd, getuid());
484 if (dup2(fd, STDIN_FILENO) < 0) {
489 if (dup2(fd, STDOUT_FILENO) < 0) {
495 close_nointr_nofail(fd);
497 *_saved_stdin = saved_stdin;
498 *_saved_stdout = saved_stdout;
503 if (saved_stdout >= 0)
504 close_nointr_nofail(saved_stdout);
506 if (saved_stdin >= 0)
507 close_nointr_nofail(saved_stdin);
510 close_nointr_nofail(fd);
515 _printf_(1, 2) static int write_confirm_message(const char *format, ...) {
521 fd = open_terminal("/dev/console", O_WRONLY|O_NOCTTY|O_CLOEXEC);
525 va_start(ap, format);
526 vdprintf(fd, format, ap);
529 close_nointr_nofail(fd);
534 static int restore_confirm_stdio(int *saved_stdin,
540 assert(saved_stdout);
544 if (*saved_stdin >= 0)
545 if (dup2(*saved_stdin, STDIN_FILENO) < 0)
548 if (*saved_stdout >= 0)
549 if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
552 if (*saved_stdin >= 0)
553 close_nointr_nofail(*saved_stdin);
555 if (*saved_stdout >= 0)
556 close_nointr_nofail(*saved_stdout);
561 static int ask_for_confirmation(char *response, char **argv) {
562 int saved_stdout = -1, saved_stdin = -1, r;
565 r = setup_confirm_stdio(&saved_stdin, &saved_stdout);
569 line = exec_command_line(argv);
573 r = ask(response, "yns", "Execute %s? [Yes, No, Skip] ", line);
576 restore_confirm_stdio(&saved_stdin, &saved_stdout);
581 static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
582 bool keep_groups = false;
587 /* Lookup and set GID and supplementary group list. Here too
588 * we avoid NSS lookups for gid=0. */
590 if (context->group || username) {
592 if (context->group) {
593 const char *g = context->group;
595 if ((r = get_group_creds(&g, &gid)) < 0)
599 /* First step, initialize groups from /etc/groups */
600 if (username && gid != 0) {
601 if (initgroups(username, gid) < 0)
607 /* Second step, set our gids */
608 if (setresgid(gid, gid, gid) < 0)
612 if (context->supplementary_groups) {
617 /* Final step, initialize any manually set supplementary groups */
618 assert_se((ngroups_max = (int) sysconf(_SC_NGROUPS_MAX)) > 0);
620 if (!(gids = new(gid_t, ngroups_max)))
624 if ((k = getgroups(ngroups_max, gids)) < 0) {
631 STRV_FOREACH(i, context->supplementary_groups) {
634 if (k >= ngroups_max) {
640 r = get_group_creds(&g, gids+k);
649 if (setgroups(k, gids) < 0) {
660 static int enforce_user(const ExecContext *context, uid_t uid) {
663 /* Sets (but doesn't lookup) the uid and make sure we keep the
664 * capabilities while doing so. */
666 if (context->capabilities) {
667 _cleanup_cap_free_ cap_t d = NULL;
668 static const cap_value_t bits[] = {
669 CAP_SETUID, /* Necessary so that we can run setresuid() below */
670 CAP_SETPCAP /* Necessary so that we can set PR_SET_SECUREBITS later on */
673 /* First step: If we need to keep capabilities but
674 * drop privileges we need to make sure we keep our
675 * caps, while we drop privileges. */
677 int sb = context->secure_bits | 1<<SECURE_KEEP_CAPS;
679 if (prctl(PR_GET_SECUREBITS) != sb)
680 if (prctl(PR_SET_SECUREBITS, sb) < 0)
684 /* Second step: set the capabilities. This will reduce
685 * the capabilities to the minimum we need. */
687 d = cap_dup(context->capabilities);
691 if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
692 cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0)
695 if (cap_set_proc(d) < 0)
699 /* Third step: actually set the uids */
700 if (setresuid(uid, uid, uid) < 0)
703 /* At this point we should have all necessary capabilities but
704 are otherwise a normal user. However, the caps might got
705 corrupted due to the setresuid() so we need clean them up
706 later. This is done outside of this call. */
713 static int null_conv(
715 const struct pam_message **msg,
716 struct pam_response **resp,
719 /* We don't support conversations */
724 static int setup_pam(
730 int fds[], unsigned n_fds) {
732 static const struct pam_conv conv = {
737 pam_handle_t *handle = NULL;
739 int pam_code = PAM_SUCCESS;
742 bool close_session = false;
743 pid_t pam_pid = 0, parent_pid;
750 /* We set up PAM in the parent process, then fork. The child
751 * will then stay around until killed via PR_GET_PDEATHSIG or
752 * systemd via the cgroup logic. It will then remove the PAM
753 * session again. The parent process will exec() the actual
754 * daemon. We do things this way to ensure that the main PID
755 * of the daemon is the one we initially fork()ed. */
757 if (log_get_max_level() < LOG_PRI(LOG_DEBUG))
760 pam_code = pam_start(name, user, &conv, &handle);
761 if (pam_code != PAM_SUCCESS) {
767 pam_code = pam_set_item(handle, PAM_TTY, tty);
768 if (pam_code != PAM_SUCCESS)
772 pam_code = pam_acct_mgmt(handle, flags);
773 if (pam_code != PAM_SUCCESS)
776 pam_code = pam_open_session(handle, flags);
777 if (pam_code != PAM_SUCCESS)
780 close_session = true;
782 e = pam_getenvlist(handle);
784 pam_code = PAM_BUF_ERR;
788 /* Block SIGTERM, so that we know that it won't get lost in
790 if (sigemptyset(&ss) < 0 ||
791 sigaddset(&ss, SIGTERM) < 0 ||
792 sigprocmask(SIG_BLOCK, &ss, &old_ss) < 0)
795 parent_pid = getpid();
805 /* The child's job is to reset the PAM session on
808 /* This string must fit in 10 chars (i.e. the length
809 * of "/sbin/init"), to look pretty in /bin/ps */
810 rename_process("(sd-pam)");
812 /* Make sure we don't keep open the passed fds in this
813 child. We assume that otherwise only those fds are
814 open here that have been opened by PAM. */
815 close_many(fds, n_fds);
817 /* Drop privileges - we don't need any to pam_close_session
818 * and this will make PR_SET_PDEATHSIG work in most cases.
819 * If this fails, ignore the error - but expect sd-pam threads
820 * to fail to exit normally */
821 if (setresuid(uid, uid, uid) < 0)
822 log_error("Error: Failed to setresuid() in sd-pam: %s", strerror(-r));
824 /* Wait until our parent died. This will only work if
825 * the above setresuid() succeeds, otherwise the kernel
826 * will not allow unprivileged parents kill their privileged
827 * children this way. We rely on the control groups kill logic
828 * to do the rest for us. */
829 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
832 /* Check if our parent process might already have
834 if (getppid() == parent_pid) {
836 if (sigwait(&ss, &sig) < 0) {
843 assert(sig == SIGTERM);
848 /* If our parent died we'll end the session */
849 if (getppid() != parent_pid) {
850 pam_code = pam_close_session(handle, flags);
851 if (pam_code != PAM_SUCCESS)
858 pam_end(handle, pam_code | flags);
862 /* If the child was forked off successfully it will do all the
863 * cleanups, so forget about the handle here. */
866 /* Unblock SIGTERM again in the parent */
867 if (sigprocmask(SIG_SETMASK, &old_ss, NULL) < 0)
870 /* We close the log explicitly here, since the PAM modules
871 * might have opened it, but we don't want this fd around. */
880 if (pam_code != PAM_SUCCESS) {
881 log_error("PAM failed: %s", pam_strerror(handle, pam_code));
882 err = -EPERM; /* PAM errors do not map to errno */
884 log_error("PAM failed: %m");
890 pam_code = pam_close_session(handle, flags);
892 pam_end(handle, pam_code | flags);
900 kill(pam_pid, SIGTERM);
901 kill(pam_pid, SIGCONT);
908 static void rename_process_from_path(const char *path) {
909 char process_name[11];
913 /* This resulting string must fit in 10 chars (i.e. the length
914 * of "/sbin/init") to look pretty in /bin/ps */
918 rename_process("(...)");
924 /* The end of the process name is usually more
925 * interesting, since the first bit might just be
931 process_name[0] = '(';
932 memcpy(process_name+1, p, l);
933 process_name[1+l] = ')';
934 process_name[1+l+1] = 0;
936 rename_process(process_name);
940 static int apply_seccomp(ExecContext *c) {
941 uint32_t action = SCMP_ACT_ALLOW;
947 c->syscall_filter = seccomp_init(c->syscall_filter_default_action);
948 if (!c->syscall_filter)
951 if (c->syscall_filter_default_action == SCMP_ACT_ALLOW)
952 action = SCMP_ACT_KILL;
954 SET_FOREACH(id, c->filtered_syscalls, i) {
955 int r = seccomp_rule_add(c->syscall_filter, action, PTR_TO_INT(id) - 1, 0);
957 log_error("Failed to add syscall filter");
962 return seccomp_load(c->syscall_filter);
966 static void do_idle_pipe_dance(int idle_pipe[4]) {
969 if (idle_pipe[1] >= 0)
970 close_nointr_nofail(idle_pipe[1]);
971 if (idle_pipe[2] >= 0)
972 close_nointr_nofail(idle_pipe[2]);
974 if (idle_pipe[0] >= 0) {
977 r = fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT_USEC);
979 if (idle_pipe[3] >= 0 && r == 0 /* timeout */) {
980 /* Signal systemd that we are bored and want to continue. */
981 write(idle_pipe[3], "x", 1);
983 /* Wait for systemd to react to the signal above. */
984 fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT2_USEC);
987 close_nointr_nofail(idle_pipe[0]);
991 if (idle_pipe[3] >= 0)
992 close_nointr_nofail(idle_pipe[3]);
995 static int build_environment(
998 usec_t watchdog_usec,
1000 const char *username,
1004 _cleanup_strv_free_ char **our_env = NULL;
1011 our_env = new0(char*, 10);
1016 if (asprintf(&x, "LISTEN_PID="PID_FMT, getpid()) < 0)
1018 our_env[n_env++] = x;
1020 if (asprintf(&x, "LISTEN_FDS=%u", n_fds) < 0)
1022 our_env[n_env++] = x;
1025 if (watchdog_usec > 0) {
1026 if (asprintf(&x, "WATCHDOG_PID="PID_FMT, getpid()) < 0)
1028 our_env[n_env++] = x;
1030 if (asprintf(&x, "WATCHDOG_USEC=%llu", (unsigned long long) watchdog_usec) < 0)
1032 our_env[n_env++] = x;
1036 x = strappend("HOME=", home);
1039 our_env[n_env++] = x;
1043 x = strappend("LOGNAME=", username);
1046 our_env[n_env++] = x;
1048 x = strappend("USER=", username);
1051 our_env[n_env++] = x;
1055 x = strappend("SHELL=", shell);
1058 our_env[n_env++] = x;
1061 if (is_terminal_input(c->std_input) ||
1062 c->std_output == EXEC_OUTPUT_TTY ||
1063 c->std_error == EXEC_OUTPUT_TTY ||
1066 x = strdup(default_term_for_tty(tty_path(c)));
1069 our_env[n_env++] = x;
1072 our_env[n_env++] = NULL;
1073 assert(n_env <= 10);
1081 int exec_spawn(ExecCommand *command,
1083 ExecContext *context,
1084 int fds[], unsigned n_fds,
1086 bool apply_permissions,
1088 bool apply_tty_stdin,
1090 CGroupControllerMask cgroup_supported,
1091 const char *cgroup_path,
1092 const char *unit_id,
1093 usec_t watchdog_usec,
1095 ExecRuntime *runtime,
1098 _cleanup_strv_free_ char **files_env = NULL;
1107 assert(fds || n_fds <= 0);
1109 if (context->std_input == EXEC_INPUT_SOCKET ||
1110 context->std_output == EXEC_OUTPUT_SOCKET ||
1111 context->std_error == EXEC_OUTPUT_SOCKET) {
1123 r = exec_context_load_environment(context, &files_env);
1125 log_struct_unit(LOG_ERR,
1127 "MESSAGE=Failed to load environment files: %s", strerror(-r),
1134 argv = command->argv;
1136 line = exec_command_line(argv);
1140 log_struct_unit(LOG_DEBUG,
1142 "EXECUTABLE=%s", command->path,
1143 "MESSAGE=About to execute: %s", line,
1152 _cleanup_strv_free_ char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
1153 const char *username = NULL, *home = NULL, *shell = NULL;
1154 unsigned n_dont_close = 0;
1155 int dont_close[n_fds + 3];
1156 uid_t uid = (uid_t) -1;
1157 gid_t gid = (gid_t) -1;
1163 rename_process_from_path(command->path);
1165 /* We reset exactly these signals, since they are the
1166 * only ones we set to SIG_IGN in the main daemon. All
1167 * others we leave untouched because we set them to
1168 * SIG_DFL or a valid handler initially, both of which
1169 * will be demoted to SIG_DFL. */
1170 default_signals(SIGNALS_CRASH_HANDLER,
1171 SIGNALS_IGNORE, -1);
1173 if (context->ignore_sigpipe)
1174 ignore_signals(SIGPIPE, -1);
1176 assert_se(sigemptyset(&ss) == 0);
1177 if (sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
1179 r = EXIT_SIGNAL_MASK;
1184 do_idle_pipe_dance(idle_pipe);
1186 /* Close sockets very early to make sure we don't
1187 * block init reexecution because it cannot bind its
1192 dont_close[n_dont_close++] = socket_fd;
1194 memcpy(dont_close + n_dont_close, fds, sizeof(int) * n_fds);
1195 n_dont_close += n_fds;
1198 if (runtime->netns_storage_socket[0] >= 0)
1199 dont_close[n_dont_close++] = runtime->netns_storage_socket[0];
1200 if (runtime->netns_storage_socket[1] >= 0)
1201 dont_close[n_dont_close++] = runtime->netns_storage_socket[1];
1204 err = close_all_fds(dont_close, n_dont_close);
1210 if (!context->same_pgrp)
1217 if (context->tcpwrap_name) {
1219 if (!socket_tcpwrap(socket_fd, context->tcpwrap_name)) {
1225 for (i = 0; i < (int) n_fds; i++) {
1226 if (!socket_tcpwrap(fds[i], context->tcpwrap_name)) {
1234 exec_context_tty_reset(context);
1236 if (confirm_spawn) {
1239 err = ask_for_confirmation(&response, argv);
1240 if (err == -ETIMEDOUT)
1241 write_confirm_message("Confirmation question timed out, assuming positive response.\n");
1243 write_confirm_message("Couldn't ask confirmation question, assuming positive response: %s\n", strerror(-err));
1244 else if (response == 's') {
1245 write_confirm_message("Skipping execution.\n");
1249 } else if (response == 'n') {
1250 write_confirm_message("Failing execution.\n");
1256 /* If a socket is connected to STDIN/STDOUT/STDERR, we
1257 * must sure to drop O_NONBLOCK */
1259 fd_nonblock(socket_fd, false);
1261 err = setup_input(context, socket_fd, apply_tty_stdin);
1267 err = setup_output(context, STDOUT_FILENO, socket_fd, basename(command->path), unit_id, apply_tty_stdin);
1273 err = setup_output(context, STDERR_FILENO, socket_fd, basename(command->path), unit_id, apply_tty_stdin);
1280 err = cg_attach_everywhere(cgroup_supported, cgroup_path, 0);
1287 if (context->oom_score_adjust_set) {
1290 snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
1293 if (write_string_file("/proc/self/oom_score_adj", t) < 0) {
1295 r = EXIT_OOM_ADJUST;
1300 if (context->nice_set)
1301 if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
1307 if (context->cpu_sched_set) {
1308 struct sched_param param = {
1309 .sched_priority = context->cpu_sched_priority,
1312 r = sched_setscheduler(0,
1313 context->cpu_sched_policy |
1314 (context->cpu_sched_reset_on_fork ?
1315 SCHED_RESET_ON_FORK : 0),
1319 r = EXIT_SETSCHEDULER;
1324 if (context->cpuset)
1325 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
1327 r = EXIT_CPUAFFINITY;
1331 if (context->ioprio_set)
1332 if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
1338 if (context->timer_slack_nsec != (nsec_t) -1)
1339 if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
1341 r = EXIT_TIMERSLACK;
1345 if (context->utmp_id)
1346 utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path);
1348 if (context->user) {
1349 username = context->user;
1350 err = get_user_creds(&username, &uid, &gid, &home, &shell);
1356 if (is_terminal_input(context->std_input)) {
1357 err = chown_terminal(STDIN_FILENO, uid);
1366 if (cgroup_path && context->user && context->pam_name) {
1367 err = cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, 0644, uid, gid);
1374 err = cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, 0755, uid, gid);
1382 if (apply_permissions) {
1383 err = enforce_groups(context, username, gid);
1390 umask(context->umask);
1393 if (apply_permissions && context->pam_name && username) {
1394 err = setup_pam(context->pam_name, username, uid, context->tty_path, &pam_env, fds, n_fds);
1401 if (context->private_network && runtime && runtime->netns_storage_socket[0] >= 0) {
1402 err = setup_netns(runtime->netns_storage_socket);
1409 if (!strv_isempty(context->read_write_dirs) ||
1410 !strv_isempty(context->read_only_dirs) ||
1411 !strv_isempty(context->inaccessible_dirs) ||
1412 context->mount_flags != 0 ||
1413 (context->private_tmp && runtime && (runtime->tmp_dir || runtime->var_tmp_dir)) ||
1414 context->private_devices) {
1416 char *tmp = NULL, *var = NULL;
1418 /* The runtime struct only contains the parent
1419 * of the private /tmp, which is
1420 * non-accessible to world users. Inside of it
1421 * there's a /tmp that is sticky, and that's
1422 * the one we want to use here. */
1424 if (context->private_tmp && runtime) {
1425 if (runtime->tmp_dir)
1426 tmp = strappenda(runtime->tmp_dir, "/tmp");
1427 if (runtime->var_tmp_dir)
1428 var = strappenda(runtime->var_tmp_dir, "/tmp");
1431 err = setup_namespace(
1432 context->read_write_dirs,
1433 context->read_only_dirs,
1434 context->inaccessible_dirs,
1437 context->private_devices,
1438 context->mount_flags);
1447 if (context->root_directory)
1448 if (chroot(context->root_directory) < 0) {
1454 if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
1460 _cleanup_free_ char *d = NULL;
1462 if (asprintf(&d, "%s/%s",
1463 context->root_directory ? context->root_directory : "",
1464 context->working_directory ? context->working_directory : "") < 0) {
1477 /* We repeat the fd closing here, to make sure that
1478 * nothing is leaked from the PAM modules */
1479 err = close_all_fds(fds, n_fds);
1481 err = shift_fds(fds, n_fds);
1483 err = flags_fds(fds, n_fds, context->non_blocking);
1489 if (apply_permissions) {
1491 for (i = 0; i < RLIMIT_NLIMITS; i++) {
1492 if (!context->rlimit[i])
1495 if (setrlimit_closest(i, context->rlimit[i]) < 0) {
1502 if (context->capability_bounding_set_drop) {
1503 err = capability_bounding_set_drop(context->capability_bounding_set_drop, false);
1505 r = EXIT_CAPABILITIES;
1510 if (context->user) {
1511 err = enforce_user(context, uid);
1518 /* PR_GET_SECUREBITS is not privileged, while
1519 * PR_SET_SECUREBITS is. So to suppress
1520 * potential EPERMs we'll try not to call
1521 * PR_SET_SECUREBITS unless necessary. */
1522 if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
1523 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
1525 r = EXIT_SECUREBITS;
1529 if (context->capabilities)
1530 if (cap_set_proc(context->capabilities) < 0) {
1532 r = EXIT_CAPABILITIES;
1536 if (context->no_new_privileges)
1537 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
1539 r = EXIT_NO_NEW_PRIVILEGES;
1544 if (context->filtered_syscalls) {
1545 err = apply_seccomp(context);
1553 if (context->selinux_context && use_selinux()) {
1557 c = context->selinux_context;
1564 err = setexeccon(c);
1565 if (err < 0 && !ignore) {
1566 r = EXIT_SELINUX_CONTEXT;
1573 err = build_environment(context, n_fds, watchdog_usec, home, username, shell, &our_env);
1579 final_env = strv_env_merge(5,
1582 context->environment,
1592 final_argv = replace_env_argv(argv, final_env);
1599 final_env = strv_env_clean(final_env);
1601 if (_unlikely_(log_get_max_level() >= LOG_PRI(LOG_DEBUG))) {
1602 line = exec_command_line(final_argv);
1605 log_struct_unit(LOG_DEBUG,
1607 "EXECUTABLE=%s", command->path,
1608 "MESSAGE=Executing: %s", line,
1615 execve(command->path, final_argv, final_env);
1622 log_struct(LOG_ERR, MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED),
1623 "EXECUTABLE=%s", command->path,
1624 "MESSAGE=Failed at step %s spawning %s: %s",
1625 exit_status_to_string(r, EXIT_STATUS_SYSTEMD),
1626 command->path, strerror(-err),
1635 log_struct_unit(LOG_DEBUG,
1637 "MESSAGE=Forked %s as "PID_FMT,
1641 /* We add the new process to the cgroup both in the child (so
1642 * that we can be sure that no user code is ever executed
1643 * outside of the cgroup) and in the parent (so that we can be
1644 * sure that when we kill the cgroup the process will be
1647 cg_attach(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, pid);
1649 exec_status_start(&command->exec_status, pid);
1655 void exec_context_init(ExecContext *c) {
1659 c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
1660 c->cpu_sched_policy = SCHED_OTHER;
1661 c->syslog_priority = LOG_DAEMON|LOG_INFO;
1662 c->syslog_level_prefix = true;
1663 c->ignore_sigpipe = true;
1664 c->timer_slack_nsec = (nsec_t) -1;
1667 void exec_context_done(ExecContext *c) {
1672 strv_free(c->environment);
1673 c->environment = NULL;
1675 strv_free(c->environment_files);
1676 c->environment_files = NULL;
1678 for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1680 c->rlimit[l] = NULL;
1683 free(c->working_directory);
1684 c->working_directory = NULL;
1685 free(c->root_directory);
1686 c->root_directory = NULL;
1691 free(c->tcpwrap_name);
1692 c->tcpwrap_name = NULL;
1694 free(c->syslog_identifier);
1695 c->syslog_identifier = NULL;
1703 strv_free(c->supplementary_groups);
1704 c->supplementary_groups = NULL;
1709 if (c->capabilities) {
1710 cap_free(c->capabilities);
1711 c->capabilities = NULL;
1714 strv_free(c->read_only_dirs);
1715 c->read_only_dirs = NULL;
1717 strv_free(c->read_write_dirs);
1718 c->read_write_dirs = NULL;
1720 strv_free(c->inaccessible_dirs);
1721 c->inaccessible_dirs = NULL;
1724 CPU_FREE(c->cpuset);
1729 free(c->selinux_context);
1730 c->selinux_context = NULL;
1732 free(c->syscall_filter);
1733 c->syscall_filter = NULL;
1735 free(c->syscall_filter_string);
1736 c->syscall_filter_string = NULL;
1739 if (c->syscall_filter) {
1740 seccomp_release(c->syscall_filter);
1741 c->syscall_filter = NULL;
1743 set_free(c->filtered_syscalls);
1744 c->filtered_syscalls = NULL;
1748 void exec_command_done(ExecCommand *c) {
1758 void exec_command_done_array(ExecCommand *c, unsigned n) {
1761 for (i = 0; i < n; i++)
1762 exec_command_done(c+i);
1765 void exec_command_free_list(ExecCommand *c) {
1769 LIST_REMOVE(command, c, i);
1770 exec_command_done(i);
1775 void exec_command_free_array(ExecCommand **c, unsigned n) {
1778 for (i = 0; i < n; i++) {
1779 exec_command_free_list(c[i]);
1784 int exec_context_load_environment(const ExecContext *c, char ***l) {
1785 char **i, **r = NULL;
1790 STRV_FOREACH(i, c->environment_files) {
1793 bool ignore = false;
1795 _cleanup_globfree_ glob_t pglob = {};
1805 if (!path_is_absolute(fn)) {
1813 /* Filename supports globbing, take all matching files */
1815 if (glob(fn, 0, NULL, &pglob) != 0) {
1820 return errno ? -errno : -EINVAL;
1822 count = pglob.gl_pathc;
1830 for (n = 0; n < count; n++) {
1831 k = load_env_file(pglob.gl_pathv[n], NULL, &p);
1839 /* Log invalid environment variables with filename */
1841 p = strv_env_clean_log(p, pglob.gl_pathv[n]);
1848 m = strv_env_merge(2, r, p);
1864 static bool tty_may_match_dev_console(const char *tty) {
1865 char *active = NULL, *console;
1868 if (startswith(tty, "/dev/"))
1871 /* trivial identity? */
1872 if (streq(tty, "console"))
1875 console = resolve_dev_console(&active);
1876 /* if we could not resolve, assume it may */
1880 /* "tty0" means the active VC, so it may be the same sometimes */
1881 b = streq(console, tty) || (streq(console, "tty0") && tty_is_vc(tty));
1887 bool exec_context_may_touch_console(ExecContext *ec) {
1888 return (ec->tty_reset || ec->tty_vhangup || ec->tty_vt_disallocate ||
1889 is_terminal_input(ec->std_input) ||
1890 is_terminal_output(ec->std_output) ||
1891 is_terminal_output(ec->std_error)) &&
1892 tty_may_match_dev_console(tty_path(ec));
1895 static void strv_fprintf(FILE *f, char **l) {
1901 fprintf(f, " %s", *g);
1904 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
1911 prefix = strempty(prefix);
1915 "%sWorkingDirectory: %s\n"
1916 "%sRootDirectory: %s\n"
1917 "%sNonBlocking: %s\n"
1918 "%sPrivateTmp: %s\n"
1919 "%sPrivateNetwork: %s\n"
1920 "%sPrivateDevices: %s\n"
1921 "%sIgnoreSIGPIPE: %s\n",
1923 prefix, c->working_directory ? c->working_directory : "/",
1924 prefix, c->root_directory ? c->root_directory : "/",
1925 prefix, yes_no(c->non_blocking),
1926 prefix, yes_no(c->private_tmp),
1927 prefix, yes_no(c->private_network),
1928 prefix, yes_no(c->private_devices),
1929 prefix, yes_no(c->ignore_sigpipe));
1931 STRV_FOREACH(e, c->environment)
1932 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
1934 STRV_FOREACH(e, c->environment_files)
1935 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
1937 if (c->tcpwrap_name)
1939 "%sTCPWrapName: %s\n",
1940 prefix, c->tcpwrap_name);
1947 if (c->oom_score_adjust_set)
1949 "%sOOMScoreAdjust: %i\n",
1950 prefix, c->oom_score_adjust);
1952 for (i = 0; i < RLIM_NLIMITS; i++)
1954 fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
1956 if (c->ioprio_set) {
1960 r = ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c->ioprio), &class_str);
1964 "%sIOSchedulingClass: %s\n"
1965 "%sIOPriority: %i\n",
1966 prefix, strna(class_str),
1967 prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
1971 if (c->cpu_sched_set) {
1975 r = sched_policy_to_string_alloc(c->cpu_sched_policy, &policy_str);
1979 "%sCPUSchedulingPolicy: %s\n"
1980 "%sCPUSchedulingPriority: %i\n"
1981 "%sCPUSchedulingResetOnFork: %s\n",
1982 prefix, strna(policy_str),
1983 prefix, c->cpu_sched_priority,
1984 prefix, yes_no(c->cpu_sched_reset_on_fork));
1989 fprintf(f, "%sCPUAffinity:", prefix);
1990 for (i = 0; i < c->cpuset_ncpus; i++)
1991 if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
1992 fprintf(f, " %u", i);
1996 if (c->timer_slack_nsec != (nsec_t) -1)
1997 fprintf(f, "%sTimerSlackNSec: "NSEC_FMT "\n", prefix, c->timer_slack_nsec);
2000 "%sStandardInput: %s\n"
2001 "%sStandardOutput: %s\n"
2002 "%sStandardError: %s\n",
2003 prefix, exec_input_to_string(c->std_input),
2004 prefix, exec_output_to_string(c->std_output),
2005 prefix, exec_output_to_string(c->std_error));
2011 "%sTTYVHangup: %s\n"
2012 "%sTTYVTDisallocate: %s\n",
2013 prefix, c->tty_path,
2014 prefix, yes_no(c->tty_reset),
2015 prefix, yes_no(c->tty_vhangup),
2016 prefix, yes_no(c->tty_vt_disallocate));
2018 if (c->std_output == EXEC_OUTPUT_SYSLOG ||
2019 c->std_output == EXEC_OUTPUT_KMSG ||
2020 c->std_output == EXEC_OUTPUT_JOURNAL ||
2021 c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
2022 c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
2023 c->std_output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE ||
2024 c->std_error == EXEC_OUTPUT_SYSLOG ||
2025 c->std_error == EXEC_OUTPUT_KMSG ||
2026 c->std_error == EXEC_OUTPUT_JOURNAL ||
2027 c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
2028 c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
2029 c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE) {
2031 _cleanup_free_ char *fac_str = NULL, *lvl_str = NULL;
2033 log_facility_unshifted_to_string_alloc(c->syslog_priority >> 3, &fac_str);
2034 log_level_to_string_alloc(LOG_PRI(c->syslog_priority), &lvl_str);
2037 "%sSyslogFacility: %s\n"
2038 "%sSyslogLevel: %s\n",
2039 prefix, strna(fac_str),
2040 prefix, strna(lvl_str));
2043 if (c->capabilities) {
2044 _cleanup_cap_free_charp_ char *t;
2046 t = cap_to_text(c->capabilities, NULL);
2048 fprintf(f, "%sCapabilities: %s\n", prefix, t);
2052 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
2054 (c->secure_bits & 1<<SECURE_KEEP_CAPS) ? " keep-caps" : "",
2055 (c->secure_bits & 1<<SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
2056 (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
2057 (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
2058 (c->secure_bits & 1<<SECURE_NOROOT) ? " noroot" : "",
2059 (c->secure_bits & 1<<SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
2061 if (c->capability_bounding_set_drop) {
2063 fprintf(f, "%sCapabilityBoundingSet:", prefix);
2065 for (l = 0; l <= cap_last_cap(); l++)
2066 if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) {
2067 _cleanup_cap_free_charp_ char *t;
2071 fprintf(f, " %s", t);
2078 fprintf(f, "%sUser: %s\n", prefix, c->user);
2080 fprintf(f, "%sGroup: %s\n", prefix, c->group);
2082 if (strv_length(c->supplementary_groups) > 0) {
2083 fprintf(f, "%sSupplementaryGroups:", prefix);
2084 strv_fprintf(f, c->supplementary_groups);
2089 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
2091 if (strv_length(c->read_write_dirs) > 0) {
2092 fprintf(f, "%sReadWriteDirs:", prefix);
2093 strv_fprintf(f, c->read_write_dirs);
2097 if (strv_length(c->read_only_dirs) > 0) {
2098 fprintf(f, "%sReadOnlyDirs:", prefix);
2099 strv_fprintf(f, c->read_only_dirs);
2103 if (strv_length(c->inaccessible_dirs) > 0) {
2104 fprintf(f, "%sInaccessibleDirs:", prefix);
2105 strv_fprintf(f, c->inaccessible_dirs);
2111 "%sUtmpIdentifier: %s\n",
2112 prefix, c->utmp_id);
2114 if (c->selinux_context)
2116 "%sSELinuxContext: %s\n",
2117 prefix, c->selinux_context);
2120 void exec_status_start(ExecStatus *s, pid_t pid) {
2125 dual_timestamp_get(&s->start_timestamp);
2128 void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
2131 if (s->pid && s->pid != pid)
2135 dual_timestamp_get(&s->exit_timestamp);
2141 if (context->utmp_id)
2142 utmp_put_dead_process(context->utmp_id, pid, code, status);
2144 exec_context_tty_reset(context);
2148 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
2149 char buf[FORMAT_TIMESTAMP_MAX];
2161 "%sPID: "PID_FMT"\n",
2164 if (s->start_timestamp.realtime > 0)
2166 "%sStart Timestamp: %s\n",
2167 prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
2169 if (s->exit_timestamp.realtime > 0)
2171 "%sExit Timestamp: %s\n"
2173 "%sExit Status: %i\n",
2174 prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
2175 prefix, sigchld_code_to_string(s->code),
2179 char *exec_command_line(char **argv) {
2187 STRV_FOREACH(a, argv)
2190 if (!(n = new(char, k)))
2194 STRV_FOREACH(a, argv) {
2201 if (strpbrk(*a, WHITESPACE)) {
2212 /* FIXME: this doesn't really handle arguments that have
2213 * spaces and ticks in them */
2218 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
2220 const char *prefix2;
2229 p2 = strappend(prefix, "\t");
2230 prefix2 = p2 ? p2 : prefix;
2232 cmd = exec_command_line(c->argv);
2235 "%sCommand Line: %s\n",
2236 prefix, cmd ? cmd : strerror(ENOMEM));
2240 exec_status_dump(&c->exec_status, f, prefix2);
2245 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
2251 LIST_FOREACH(command, c, c)
2252 exec_command_dump(c, f, prefix);
2255 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
2262 /* It's kind of important, that we keep the order here */
2263 LIST_FIND_TAIL(command, *l, end);
2264 LIST_INSERT_AFTER(command, *l, end, e);
2269 int exec_command_set(ExecCommand *c, const char *path, ...) {
2277 l = strv_new_ap(path, ap);
2298 static int exec_runtime_allocate(ExecRuntime **rt) {
2303 *rt = new0(ExecRuntime, 1);
2308 (*rt)->netns_storage_socket[0] = (*rt)->netns_storage_socket[1] = -1;
2313 int exec_runtime_make(ExecRuntime **rt, ExecContext *c, const char *id) {
2323 if (!c->private_network && !c->private_tmp)
2326 r = exec_runtime_allocate(rt);
2330 if (c->private_network && (*rt)->netns_storage_socket[0] < 0) {
2331 if (socketpair(AF_UNIX, SOCK_DGRAM, 0, (*rt)->netns_storage_socket) < 0)
2335 if (c->private_tmp && !(*rt)->tmp_dir) {
2336 r = setup_tmp_dirs(id, &(*rt)->tmp_dir, &(*rt)->var_tmp_dir);
2344 ExecRuntime *exec_runtime_ref(ExecRuntime *r) {
2346 assert(r->n_ref > 0);
2352 ExecRuntime *exec_runtime_unref(ExecRuntime *r) {
2357 assert(r->n_ref > 0);
2360 if (r->n_ref <= 0) {
2362 free(r->var_tmp_dir);
2363 close_pipe(r->netns_storage_socket);
2370 int exec_runtime_serialize(ExecRuntime *rt, Unit *u, FILE *f, FDSet *fds) {
2379 unit_serialize_item(u, f, "tmp-dir", rt->tmp_dir);
2381 if (rt->var_tmp_dir)
2382 unit_serialize_item(u, f, "var-tmp-dir", rt->var_tmp_dir);
2384 if (rt->netns_storage_socket[0] >= 0) {
2387 copy = fdset_put_dup(fds, rt->netns_storage_socket[0]);
2391 unit_serialize_item_format(u, f, "netns-socket-0", "%i", copy);
2394 if (rt->netns_storage_socket[1] >= 0) {
2397 copy = fdset_put_dup(fds, rt->netns_storage_socket[1]);
2401 unit_serialize_item_format(u, f, "netns-socket-1", "%i", copy);
2407 int exec_runtime_deserialize_item(ExecRuntime **rt, Unit *u, const char *key, const char *value, FDSet *fds) {
2414 if (streq(key, "tmp-dir")) {
2417 r = exec_runtime_allocate(rt);
2421 copy = strdup(value);
2425 free((*rt)->tmp_dir);
2426 (*rt)->tmp_dir = copy;
2428 } else if (streq(key, "var-tmp-dir")) {
2431 r = exec_runtime_allocate(rt);
2435 copy = strdup(value);
2439 free((*rt)->var_tmp_dir);
2440 (*rt)->var_tmp_dir = copy;
2442 } else if (streq(key, "netns-socket-0")) {
2445 r = exec_runtime_allocate(rt);
2449 if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
2450 log_debug_unit(u->id, "Failed to parse netns socket value %s", value);
2452 if ((*rt)->netns_storage_socket[0] >= 0)
2453 close_nointr_nofail((*rt)->netns_storage_socket[0]);
2455 (*rt)->netns_storage_socket[0] = fdset_remove(fds, fd);
2457 } else if (streq(key, "netns-socket-1")) {
2460 r = exec_runtime_allocate(rt);
2464 if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
2465 log_debug_unit(u->id, "Failed to parse netns socket value %s", value);
2467 if ((*rt)->netns_storage_socket[1] >= 0)
2468 close_nointr_nofail((*rt)->netns_storage_socket[1]);
2470 (*rt)->netns_storage_socket[1] = fdset_remove(fds, fd);
2478 static void *remove_tmpdir_thread(void *p) {
2479 _cleanup_free_ char *path = p;
2481 rm_rf_dangerous(path, false, true, false);
2485 void exec_runtime_destroy(ExecRuntime *rt) {
2489 /* If there are multiple users of this, let's leave the stuff around */
2494 log_debug("Spawning thread to nuke %s", rt->tmp_dir);
2495 asynchronous_job(remove_tmpdir_thread, rt->tmp_dir);
2499 if (rt->var_tmp_dir) {
2500 log_debug("Spawning thread to nuke %s", rt->var_tmp_dir);
2501 asynchronous_job(remove_tmpdir_thread, rt->var_tmp_dir);
2502 rt->var_tmp_dir = NULL;
2505 close_pipe(rt->netns_storage_socket);
2508 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
2509 [EXEC_INPUT_NULL] = "null",
2510 [EXEC_INPUT_TTY] = "tty",
2511 [EXEC_INPUT_TTY_FORCE] = "tty-force",
2512 [EXEC_INPUT_TTY_FAIL] = "tty-fail",
2513 [EXEC_INPUT_SOCKET] = "socket"
2516 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
2518 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
2519 [EXEC_OUTPUT_INHERIT] = "inherit",
2520 [EXEC_OUTPUT_NULL] = "null",
2521 [EXEC_OUTPUT_TTY] = "tty",
2522 [EXEC_OUTPUT_SYSLOG] = "syslog",
2523 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
2524 [EXEC_OUTPUT_KMSG] = "kmsg",
2525 [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
2526 [EXEC_OUTPUT_JOURNAL] = "journal",
2527 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE] = "journal+console",
2528 [EXEC_OUTPUT_SOCKET] = "socket"
2531 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);