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>
41 #include <linux/seccomp-bpf.h>
47 #include <security/pam_appl.h>
53 #include "capability.h"
56 #include "sd-messages.h"
58 #include "securebits.h"
59 #include "namespace.h"
61 #include "exit-status.h"
63 #include "utmp-wtmp.h"
65 #include "path-util.h"
66 #include "syscall-list.h"
72 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
73 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
75 /* This assumes there is a 'tty' group */
78 #define SNDBUF_SIZE (8*1024*1024)
80 static int shift_fds(int fds[], unsigned n_fds) {
81 int start, restart_from;
86 /* Modifies the fds array! (sorts it) */
96 for (i = start; i < (int) n_fds; i++) {
99 /* Already at right index? */
103 if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
106 close_nointr_nofail(fds[i]);
109 /* Hmm, the fd we wanted isn't free? Then
110 * let's remember that and try again from here*/
111 if (nfd != i+3 && restart_from < 0)
115 if (restart_from < 0)
118 start = restart_from;
124 static int flags_fds(const int fds[], unsigned n_fds, bool nonblock) {
133 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
135 for (i = 0; i < n_fds; i++) {
137 if ((r = fd_nonblock(fds[i], nonblock)) < 0)
140 /* We unconditionally drop FD_CLOEXEC from the fds,
141 * since after all we want to pass these fds to our
144 if ((r = fd_cloexec(fds[i], false)) < 0)
151 _pure_ static const char *tty_path(const ExecContext *context) {
154 if (context->tty_path)
155 return context->tty_path;
157 return "/dev/console";
160 static void exec_context_tty_reset(const ExecContext *context) {
163 if (context->tty_vhangup)
164 terminal_vhangup(tty_path(context));
166 if (context->tty_reset)
167 reset_terminal(tty_path(context));
169 if (context->tty_vt_disallocate && context->tty_path)
170 vt_disallocate(context->tty_path);
173 static bool is_terminal_output(ExecOutput o) {
175 o == EXEC_OUTPUT_TTY ||
176 o == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
177 o == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
178 o == EXEC_OUTPUT_JOURNAL_AND_CONSOLE;
181 static int open_null_as(int flags, int nfd) {
186 fd = open("/dev/null", flags|O_NOCTTY);
191 r = dup2(fd, nfd) < 0 ? -errno : nfd;
192 close_nointr_nofail(fd);
199 static int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, const char *unit_id, int nfd) {
201 union sockaddr_union sa = {
202 .un.sun_family = AF_UNIX,
203 .un.sun_path = "/run/systemd/journal/stdout",
207 assert(output < _EXEC_OUTPUT_MAX);
211 fd = socket(AF_UNIX, SOCK_STREAM, 0);
215 r = connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path));
217 close_nointr_nofail(fd);
221 if (shutdown(fd, SHUT_RD) < 0) {
222 close_nointr_nofail(fd);
226 fd_inc_sndbuf(fd, SNDBUF_SIZE);
236 context->syslog_identifier ? context->syslog_identifier : ident,
238 context->syslog_priority,
239 !!context->syslog_level_prefix,
240 output == EXEC_OUTPUT_SYSLOG || output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE,
241 output == EXEC_OUTPUT_KMSG || output == EXEC_OUTPUT_KMSG_AND_CONSOLE,
242 is_terminal_output(output));
245 r = dup2(fd, nfd) < 0 ? -errno : nfd;
246 close_nointr_nofail(fd);
252 static int open_terminal_as(const char *path, mode_t mode, int nfd) {
258 if ((fd = open_terminal(path, mode | O_NOCTTY)) < 0)
262 r = dup2(fd, nfd) < 0 ? -errno : nfd;
263 close_nointr_nofail(fd);
270 static bool is_terminal_input(ExecInput i) {
272 i == EXEC_INPUT_TTY ||
273 i == EXEC_INPUT_TTY_FORCE ||
274 i == EXEC_INPUT_TTY_FAIL;
277 static int fixup_input(ExecInput std_input, int socket_fd, bool apply_tty_stdin) {
279 if (is_terminal_input(std_input) && !apply_tty_stdin)
280 return EXEC_INPUT_NULL;
282 if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0)
283 return EXEC_INPUT_NULL;
288 static int fixup_output(ExecOutput std_output, int socket_fd) {
290 if (std_output == EXEC_OUTPUT_SOCKET && socket_fd < 0)
291 return EXEC_OUTPUT_INHERIT;
296 static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty_stdin) {
301 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
305 case EXEC_INPUT_NULL:
306 return open_null_as(O_RDONLY, STDIN_FILENO);
309 case EXEC_INPUT_TTY_FORCE:
310 case EXEC_INPUT_TTY_FAIL: {
313 fd = acquire_terminal(tty_path(context),
314 i == EXEC_INPUT_TTY_FAIL,
315 i == EXEC_INPUT_TTY_FORCE,
321 if (fd != STDIN_FILENO) {
322 r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
323 close_nointr_nofail(fd);
330 case EXEC_INPUT_SOCKET:
331 return dup2(socket_fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
334 assert_not_reached("Unknown input type");
338 static int setup_output(const ExecContext *context, int fileno, int socket_fd, const char *ident, const char *unit_id, bool apply_tty_stdin) {
346 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
347 o = fixup_output(context->std_output, socket_fd);
349 if (fileno == STDERR_FILENO) {
351 e = fixup_output(context->std_error, socket_fd);
353 /* This expects the input and output are already set up */
355 /* Don't change the stderr file descriptor if we inherit all
356 * the way and are not on a tty */
357 if (e == EXEC_OUTPUT_INHERIT &&
358 o == EXEC_OUTPUT_INHERIT &&
359 i == EXEC_INPUT_NULL &&
360 !is_terminal_input(context->std_input) &&
364 /* Duplicate from stdout if possible */
365 if (e == o || e == EXEC_OUTPUT_INHERIT)
366 return dup2(STDOUT_FILENO, fileno) < 0 ? -errno : fileno;
370 } else if (o == EXEC_OUTPUT_INHERIT) {
371 /* If input got downgraded, inherit the original value */
372 if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input))
373 return open_terminal_as(tty_path(context), O_WRONLY, fileno);
375 /* If the input is connected to anything that's not a /dev/null, inherit that... */
376 if (i != EXEC_INPUT_NULL)
377 return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
379 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
383 /* We need to open /dev/null here anew, to get the right access mode. */
384 return open_null_as(O_WRONLY, fileno);
389 case EXEC_OUTPUT_NULL:
390 return open_null_as(O_WRONLY, fileno);
392 case EXEC_OUTPUT_TTY:
393 if (is_terminal_input(i))
394 return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
396 /* We don't reset the terminal if this is just about output */
397 return open_terminal_as(tty_path(context), O_WRONLY, fileno);
399 case EXEC_OUTPUT_SYSLOG:
400 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
401 case EXEC_OUTPUT_KMSG:
402 case EXEC_OUTPUT_KMSG_AND_CONSOLE:
403 case EXEC_OUTPUT_JOURNAL:
404 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE:
405 r = connect_logger_as(context, o, ident, unit_id, fileno);
407 log_struct_unit(LOG_CRIT, unit_id,
408 "MESSAGE=Failed to connect std%s of %s to the journal socket: %s",
409 fileno == STDOUT_FILENO ? "out" : "err",
410 unit_id, strerror(-r),
413 r = open_null_as(O_WRONLY, fileno);
417 case EXEC_OUTPUT_SOCKET:
418 assert(socket_fd >= 0);
419 return dup2(socket_fd, fileno) < 0 ? -errno : fileno;
422 assert_not_reached("Unknown error type");
426 static int chown_terminal(int fd, uid_t uid) {
431 /* This might fail. What matters are the results. */
432 (void) fchown(fd, uid, -1);
433 (void) fchmod(fd, TTY_MODE);
435 if (fstat(fd, &st) < 0)
438 if (st.st_uid != uid || (st.st_mode & 0777) != TTY_MODE)
444 static int setup_confirm_stdio(int *_saved_stdin,
445 int *_saved_stdout) {
446 int fd = -1, saved_stdin, saved_stdout = -1, r;
448 assert(_saved_stdin);
449 assert(_saved_stdout);
451 saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3);
455 saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3);
456 if (saved_stdout < 0) {
461 fd = acquire_terminal(
466 DEFAULT_CONFIRM_USEC);
472 r = chown_terminal(fd, getuid());
476 if (dup2(fd, STDIN_FILENO) < 0) {
481 if (dup2(fd, STDOUT_FILENO) < 0) {
487 close_nointr_nofail(fd);
489 *_saved_stdin = saved_stdin;
490 *_saved_stdout = saved_stdout;
495 if (saved_stdout >= 0)
496 close_nointr_nofail(saved_stdout);
498 if (saved_stdin >= 0)
499 close_nointr_nofail(saved_stdin);
502 close_nointr_nofail(fd);
507 _printf_(1, 2) static int write_confirm_message(const char *format, ...) {
513 fd = open_terminal("/dev/console", O_WRONLY|O_NOCTTY|O_CLOEXEC);
517 va_start(ap, format);
518 vdprintf(fd, format, ap);
521 close_nointr_nofail(fd);
526 static int restore_confirm_stdio(int *saved_stdin,
532 assert(saved_stdout);
536 if (*saved_stdin >= 0)
537 if (dup2(*saved_stdin, STDIN_FILENO) < 0)
540 if (*saved_stdout >= 0)
541 if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
544 if (*saved_stdin >= 0)
545 close_nointr_nofail(*saved_stdin);
547 if (*saved_stdout >= 0)
548 close_nointr_nofail(*saved_stdout);
553 static int ask_for_confirmation(char *response, char **argv) {
554 int saved_stdout = -1, saved_stdin = -1, r;
557 r = setup_confirm_stdio(&saved_stdin, &saved_stdout);
561 line = exec_command_line(argv);
565 r = ask(response, "yns", "Execute %s? [Yes, No, Skip] ", line);
568 restore_confirm_stdio(&saved_stdin, &saved_stdout);
573 static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
574 bool keep_groups = false;
579 /* Lookup and set GID and supplementary group list. Here too
580 * we avoid NSS lookups for gid=0. */
582 if (context->group || username) {
584 if (context->group) {
585 const char *g = context->group;
587 if ((r = get_group_creds(&g, &gid)) < 0)
591 /* First step, initialize groups from /etc/groups */
592 if (username && gid != 0) {
593 if (initgroups(username, gid) < 0)
599 /* Second step, set our gids */
600 if (setresgid(gid, gid, gid) < 0)
604 if (context->supplementary_groups) {
609 /* Final step, initialize any manually set supplementary groups */
610 assert_se((ngroups_max = (int) sysconf(_SC_NGROUPS_MAX)) > 0);
612 if (!(gids = new(gid_t, ngroups_max)))
616 if ((k = getgroups(ngroups_max, gids)) < 0) {
623 STRV_FOREACH(i, context->supplementary_groups) {
626 if (k >= ngroups_max) {
632 r = get_group_creds(&g, gids+k);
641 if (setgroups(k, gids) < 0) {
652 static int enforce_user(const ExecContext *context, uid_t uid) {
656 /* Sets (but doesn't lookup) the uid and make sure we keep the
657 * capabilities while doing so. */
659 if (context->capabilities) {
661 static const cap_value_t bits[] = {
662 CAP_SETUID, /* Necessary so that we can run setresuid() below */
663 CAP_SETPCAP /* Necessary so that we can set PR_SET_SECUREBITS later on */
666 /* First step: If we need to keep capabilities but
667 * drop privileges we need to make sure we keep our
668 * caps, while we drop privileges. */
670 int sb = context->secure_bits | 1<<SECURE_KEEP_CAPS;
672 if (prctl(PR_GET_SECUREBITS) != sb)
673 if (prctl(PR_SET_SECUREBITS, sb) < 0)
677 /* Second step: set the capabilities. This will reduce
678 * the capabilities to the minimum we need. */
680 if (!(d = cap_dup(context->capabilities)))
683 if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
684 cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0) {
690 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);
939 static int apply_seccomp(uint32_t *syscall_filter) {
940 static const struct sock_filter header[] = {
941 VALIDATE_ARCHITECTURE,
944 static const struct sock_filter footer[] = {
950 struct sock_filter *f;
951 struct sock_fprog prog = {};
953 assert(syscall_filter);
955 /* First: count the syscalls to check for */
956 for (i = 0, n = 0; i < syscall_max(); i++)
957 if (syscall_filter[i >> 4] & (1 << (i & 31)))
960 /* Second: build the filter program from a header the syscall
961 * matches and the footer */
962 f = alloca(sizeof(struct sock_filter) * (ELEMENTSOF(header) + 2*n + ELEMENTSOF(footer)));
963 memcpy(f, header, sizeof(header));
965 for (i = 0, n = 0; i < syscall_max(); i++)
966 if (syscall_filter[i >> 4] & (1 << (i & 31))) {
967 struct sock_filter item[] = {
968 BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, INDEX_TO_SYSCALL(i), 0, 1),
969 BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW)
972 assert_cc(ELEMENTSOF(item) == 2);
974 f[ELEMENTSOF(header) + 2*n] = item[0];
975 f[ELEMENTSOF(header) + 2*n+1] = item[1];
980 memcpy(f + (ELEMENTSOF(header) + 2*n), footer, sizeof(footer));
982 /* Third: install the filter */
983 prog.len = ELEMENTSOF(header) + ELEMENTSOF(footer) + 2*n;
985 if (prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog) < 0)
991 static void do_idle_pipe_dance(int idle_pipe[4]) {
994 if (idle_pipe[1] >= 0)
995 close_nointr_nofail(idle_pipe[1]);
996 if (idle_pipe[2] >= 0)
997 close_nointr_nofail(idle_pipe[2]);
999 if (idle_pipe[0] >= 0) {
1002 r = fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT_USEC);
1004 if (idle_pipe[3] >= 0 && r == 0 /* timeout */) {
1005 /* Signal systemd that we are bored and want to continue. */
1006 write(idle_pipe[3], "x", 1);
1008 /* Wait for systemd to react to the signal above. */
1009 fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT2_USEC);
1012 close_nointr_nofail(idle_pipe[0]);
1016 if (idle_pipe[3] >= 0)
1017 close_nointr_nofail(idle_pipe[3]);
1020 static int build_environment(
1024 const char *username,
1028 _cleanup_strv_free_ char **our_env = NULL;
1035 our_env = new(char*, 8);
1040 if (asprintf(&x, "LISTEN_PID=%lu", (unsigned long) getpid()) < 0)
1042 our_env[n_env++] = x;
1044 if (asprintf(&x, "LISTEN_FDS=%u", n_fds) < 0)
1046 our_env[n_env++] = x;
1050 x = strappend("HOME=", home);
1053 our_env[n_env++] = x;
1057 x = strappend("LOGNAME=", username);
1060 our_env[n_env++] = x;
1062 x = strappend("USER=", username);
1065 our_env[n_env++] = x;
1069 x = strappend("SHELL=", shell);
1072 our_env[n_env++] = x;
1075 if (is_terminal_input(c->std_input) ||
1076 c->std_output == EXEC_OUTPUT_TTY ||
1077 c->std_error == EXEC_OUTPUT_TTY ||
1080 x = strdup(default_term_for_tty(tty_path(c)));
1083 our_env[n_env++] = x;
1086 our_env[n_env++] = NULL;
1095 int exec_spawn(ExecCommand *command,
1097 ExecContext *context,
1098 int fds[], unsigned n_fds,
1100 bool apply_permissions,
1102 bool apply_tty_stdin,
1104 CGroupControllerMask cgroup_supported,
1105 const char *cgroup_path,
1106 const char *unit_id,
1108 ExecRuntime *runtime,
1111 _cleanup_strv_free_ char **files_env = NULL;
1120 assert(fds || n_fds <= 0);
1122 if (context->std_input == EXEC_INPUT_SOCKET ||
1123 context->std_output == EXEC_OUTPUT_SOCKET ||
1124 context->std_error == EXEC_OUTPUT_SOCKET) {
1136 r = exec_context_load_environment(context, &files_env);
1138 log_struct_unit(LOG_ERR,
1140 "MESSAGE=Failed to load environment files: %s", strerror(-r),
1147 argv = command->argv;
1149 line = exec_command_line(argv);
1153 log_struct_unit(LOG_DEBUG,
1155 "EXECUTABLE=%s", command->path,
1156 "MESSAGE=About to execute: %s", line,
1165 _cleanup_strv_free_ char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
1166 const char *username = NULL, *home = NULL, *shell = NULL;
1167 unsigned n_dont_close = 0;
1168 int dont_close[n_fds + 3];
1169 uid_t uid = (uid_t) -1;
1170 gid_t gid = (gid_t) -1;
1176 rename_process_from_path(command->path);
1178 /* We reset exactly these signals, since they are the
1179 * only ones we set to SIG_IGN in the main daemon. All
1180 * others we leave untouched because we set them to
1181 * SIG_DFL or a valid handler initially, both of which
1182 * will be demoted to SIG_DFL. */
1183 default_signals(SIGNALS_CRASH_HANDLER,
1184 SIGNALS_IGNORE, -1);
1186 if (context->ignore_sigpipe)
1187 ignore_signals(SIGPIPE, -1);
1189 assert_se(sigemptyset(&ss) == 0);
1190 if (sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
1192 r = EXIT_SIGNAL_MASK;
1197 do_idle_pipe_dance(idle_pipe);
1199 /* Close sockets very early to make sure we don't
1200 * block init reexecution because it cannot bind its
1205 dont_close[n_dont_close++] = socket_fd;
1207 memcpy(dont_close + n_dont_close, fds, sizeof(int) * n_fds);
1208 n_dont_close += n_fds;
1211 if (runtime->netns_storage_socket[0] >= 0)
1212 dont_close[n_dont_close++] = runtime->netns_storage_socket[0];
1213 if (runtime->netns_storage_socket[1] >= 0)
1214 dont_close[n_dont_close++] = runtime->netns_storage_socket[1];
1217 err = close_all_fds(dont_close, n_dont_close);
1223 if (!context->same_pgrp)
1230 if (context->tcpwrap_name) {
1232 if (!socket_tcpwrap(socket_fd, context->tcpwrap_name)) {
1238 for (i = 0; i < (int) n_fds; i++) {
1239 if (!socket_tcpwrap(fds[i], context->tcpwrap_name)) {
1247 exec_context_tty_reset(context);
1249 if (confirm_spawn) {
1252 err = ask_for_confirmation(&response, argv);
1253 if (err == -ETIMEDOUT)
1254 write_confirm_message("Confirmation question timed out, assuming positive response.\n");
1256 write_confirm_message("Couldn't ask confirmation question, assuming positive response: %s\n", strerror(-err));
1257 else if (response == 's') {
1258 write_confirm_message("Skipping execution.\n");
1262 } else if (response == 'n') {
1263 write_confirm_message("Failing execution.\n");
1269 /* If a socket is connected to STDIN/STDOUT/STDERR, we
1270 * must sure to drop O_NONBLOCK */
1272 fd_nonblock(socket_fd, false);
1274 err = setup_input(context, socket_fd, apply_tty_stdin);
1280 err = setup_output(context, STDOUT_FILENO, socket_fd, basename(command->path), unit_id, apply_tty_stdin);
1286 err = setup_output(context, STDERR_FILENO, socket_fd, basename(command->path), unit_id, apply_tty_stdin);
1293 err = cg_attach_everywhere(cgroup_supported, cgroup_path, 0);
1300 if (context->oom_score_adjust_set) {
1303 snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
1306 if (write_string_file("/proc/self/oom_score_adj", t) < 0) {
1308 r = EXIT_OOM_ADJUST;
1313 if (context->nice_set)
1314 if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
1320 if (context->cpu_sched_set) {
1321 struct sched_param param = {
1322 .sched_priority = context->cpu_sched_priority,
1325 r = sched_setscheduler(0,
1326 context->cpu_sched_policy |
1327 (context->cpu_sched_reset_on_fork ?
1328 SCHED_RESET_ON_FORK : 0),
1332 r = EXIT_SETSCHEDULER;
1337 if (context->cpuset)
1338 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
1340 r = EXIT_CPUAFFINITY;
1344 if (context->ioprio_set)
1345 if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
1351 if (context->timer_slack_nsec != (nsec_t) -1)
1352 if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
1354 r = EXIT_TIMERSLACK;
1358 if (context->utmp_id)
1359 utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path);
1361 if (context->user) {
1362 username = context->user;
1363 err = get_user_creds(&username, &uid, &gid, &home, &shell);
1369 if (is_terminal_input(context->std_input)) {
1370 err = chown_terminal(STDIN_FILENO, uid);
1379 if (cgroup_path && context->user && context->pam_name) {
1380 err = cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, 0644, uid, gid);
1387 err = cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, 0755, uid, gid);
1395 if (apply_permissions) {
1396 err = enforce_groups(context, username, gid);
1403 umask(context->umask);
1406 if (apply_permissions && context->pam_name && username) {
1407 err = setup_pam(context->pam_name, username, uid, context->tty_path, &pam_env, fds, n_fds);
1414 if (context->private_network && runtime && runtime->netns_storage_socket[0] >= 0) {
1415 err = setup_netns(runtime->netns_storage_socket);
1422 if (!strv_isempty(context->read_write_dirs) ||
1423 !strv_isempty(context->read_only_dirs) ||
1424 !strv_isempty(context->inaccessible_dirs) ||
1425 context->mount_flags != 0 ||
1426 (context->private_tmp && runtime && (runtime->tmp_dir || runtime->var_tmp_dir))) {
1428 char *tmp = NULL, *var = NULL;
1430 /* The runtime struct only contains the parent
1431 * of the private /tmp, which is
1432 * non-accessible to world users. Inside of it
1433 * there's a /tmp that is sticky, and that's
1434 * the one we want to use here. */
1436 if (context->private_tmp && runtime) {
1437 if (runtime->tmp_dir)
1438 tmp = strappenda(runtime->tmp_dir, "/tmp");
1439 if (runtime->var_tmp_dir)
1440 var = strappenda(runtime->var_tmp_dir, "/tmp");
1443 err = setup_namespace(
1444 context->read_write_dirs,
1445 context->read_only_dirs,
1446 context->inaccessible_dirs,
1449 context->mount_flags);
1458 if (context->root_directory)
1459 if (chroot(context->root_directory) < 0) {
1465 if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
1471 _cleanup_free_ char *d = NULL;
1473 if (asprintf(&d, "%s/%s",
1474 context->root_directory ? context->root_directory : "",
1475 context->working_directory ? context->working_directory : "") < 0) {
1488 /* We repeat the fd closing here, to make sure that
1489 * nothing is leaked from the PAM modules */
1490 err = close_all_fds(fds, n_fds);
1492 err = shift_fds(fds, n_fds);
1494 err = flags_fds(fds, n_fds, context->non_blocking);
1500 if (apply_permissions) {
1502 for (i = 0; i < RLIMIT_NLIMITS; i++) {
1503 if (!context->rlimit[i])
1506 if (setrlimit_closest(i, context->rlimit[i]) < 0) {
1513 if (context->capability_bounding_set_drop) {
1514 err = capability_bounding_set_drop(context->capability_bounding_set_drop, false);
1516 r = EXIT_CAPABILITIES;
1521 if (context->user) {
1522 err = enforce_user(context, uid);
1529 /* PR_GET_SECUREBITS is not privileged, while
1530 * PR_SET_SECUREBITS is. So to suppress
1531 * potential EPERMs we'll try not to call
1532 * PR_SET_SECUREBITS unless necessary. */
1533 if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
1534 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
1536 r = EXIT_SECUREBITS;
1540 if (context->capabilities)
1541 if (cap_set_proc(context->capabilities) < 0) {
1543 r = EXIT_CAPABILITIES;
1547 if (context->no_new_privileges)
1548 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
1550 r = EXIT_NO_NEW_PRIVILEGES;
1554 if (context->syscall_filter) {
1555 err = apply_seccomp(context->syscall_filter);
1563 err = build_environment(context, n_fds, home, username, shell, &our_env);
1569 final_env = strv_env_merge(5,
1572 context->environment,
1582 final_argv = replace_env_argv(argv, final_env);
1589 final_env = strv_env_clean(final_env);
1591 if (_unlikely_(log_get_max_level() >= LOG_PRI(LOG_DEBUG))) {
1592 line = exec_command_line(final_argv);
1595 log_struct_unit(LOG_DEBUG,
1597 "EXECUTABLE=%s", command->path,
1598 "MESSAGE=Executing: %s", line,
1605 execve(command->path, final_argv, final_env);
1612 log_struct(LOG_ERR, MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED),
1613 "EXECUTABLE=%s", command->path,
1614 "MESSAGE=Failed at step %s spawning %s: %s",
1615 exit_status_to_string(r, EXIT_STATUS_SYSTEMD),
1616 command->path, strerror(-err),
1625 log_struct_unit(LOG_DEBUG,
1627 "MESSAGE=Forked %s as %lu",
1628 command->path, (unsigned long) pid,
1631 /* We add the new process to the cgroup both in the child (so
1632 * that we can be sure that no user code is ever executed
1633 * outside of the cgroup) and in the parent (so that we can be
1634 * sure that when we kill the cgroup the process will be
1637 cg_attach(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, pid);
1639 exec_status_start(&command->exec_status, pid);
1645 void exec_context_init(ExecContext *c) {
1649 c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
1650 c->cpu_sched_policy = SCHED_OTHER;
1651 c->syslog_priority = LOG_DAEMON|LOG_INFO;
1652 c->syslog_level_prefix = true;
1653 c->ignore_sigpipe = true;
1654 c->timer_slack_nsec = (nsec_t) -1;
1657 void exec_context_done(ExecContext *c) {
1662 strv_free(c->environment);
1663 c->environment = NULL;
1665 strv_free(c->environment_files);
1666 c->environment_files = NULL;
1668 for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1670 c->rlimit[l] = NULL;
1673 free(c->working_directory);
1674 c->working_directory = NULL;
1675 free(c->root_directory);
1676 c->root_directory = NULL;
1681 free(c->tcpwrap_name);
1682 c->tcpwrap_name = NULL;
1684 free(c->syslog_identifier);
1685 c->syslog_identifier = NULL;
1693 strv_free(c->supplementary_groups);
1694 c->supplementary_groups = NULL;
1699 if (c->capabilities) {
1700 cap_free(c->capabilities);
1701 c->capabilities = NULL;
1704 strv_free(c->read_only_dirs);
1705 c->read_only_dirs = NULL;
1707 strv_free(c->read_write_dirs);
1708 c->read_write_dirs = NULL;
1710 strv_free(c->inaccessible_dirs);
1711 c->inaccessible_dirs = NULL;
1714 CPU_FREE(c->cpuset);
1719 free(c->syscall_filter);
1720 c->syscall_filter = NULL;
1723 void exec_command_done(ExecCommand *c) {
1733 void exec_command_done_array(ExecCommand *c, unsigned n) {
1736 for (i = 0; i < n; i++)
1737 exec_command_done(c+i);
1740 void exec_command_free_list(ExecCommand *c) {
1744 LIST_REMOVE(command, c, i);
1745 exec_command_done(i);
1750 void exec_command_free_array(ExecCommand **c, unsigned n) {
1753 for (i = 0; i < n; i++) {
1754 exec_command_free_list(c[i]);
1759 int exec_context_load_environment(const ExecContext *c, char ***l) {
1760 char **i, **r = NULL;
1765 STRV_FOREACH(i, c->environment_files) {
1768 bool ignore = false;
1770 _cleanup_globfree_ glob_t pglob = {};
1780 if (!path_is_absolute(fn)) {
1788 /* Filename supports globbing, take all matching files */
1790 if (glob(fn, 0, NULL, &pglob) != 0) {
1795 return errno ? -errno : -EINVAL;
1797 count = pglob.gl_pathc;
1805 for (n = 0; n < count; n++) {
1806 k = load_env_file(pglob.gl_pathv[n], NULL, &p);
1814 /* Log invalid environment variables with filename */
1816 p = strv_env_clean_log(p, pglob.gl_pathv[n]);
1823 m = strv_env_merge(2, r, p);
1839 static bool tty_may_match_dev_console(const char *tty) {
1840 char *active = NULL, *console;
1843 if (startswith(tty, "/dev/"))
1846 /* trivial identity? */
1847 if (streq(tty, "console"))
1850 console = resolve_dev_console(&active);
1851 /* if we could not resolve, assume it may */
1855 /* "tty0" means the active VC, so it may be the same sometimes */
1856 b = streq(console, tty) || (streq(console, "tty0") && tty_is_vc(tty));
1862 bool exec_context_may_touch_console(ExecContext *ec) {
1863 return (ec->tty_reset || ec->tty_vhangup || ec->tty_vt_disallocate ||
1864 is_terminal_input(ec->std_input) ||
1865 is_terminal_output(ec->std_output) ||
1866 is_terminal_output(ec->std_error)) &&
1867 tty_may_match_dev_console(tty_path(ec));
1870 static void strv_fprintf(FILE *f, char **l) {
1876 fprintf(f, " %s", *g);
1879 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
1886 prefix = strempty(prefix);
1890 "%sWorkingDirectory: %s\n"
1891 "%sRootDirectory: %s\n"
1892 "%sNonBlocking: %s\n"
1893 "%sPrivateTmp: %s\n"
1894 "%sPrivateNetwork: %s\n"
1895 "%sIgnoreSIGPIPE: %s\n",
1897 prefix, c->working_directory ? c->working_directory : "/",
1898 prefix, c->root_directory ? c->root_directory : "/",
1899 prefix, yes_no(c->non_blocking),
1900 prefix, yes_no(c->private_tmp),
1901 prefix, yes_no(c->private_network),
1902 prefix, yes_no(c->ignore_sigpipe));
1904 STRV_FOREACH(e, c->environment)
1905 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
1907 STRV_FOREACH(e, c->environment_files)
1908 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
1910 if (c->tcpwrap_name)
1912 "%sTCPWrapName: %s\n",
1913 prefix, c->tcpwrap_name);
1920 if (c->oom_score_adjust_set)
1922 "%sOOMScoreAdjust: %i\n",
1923 prefix, c->oom_score_adjust);
1925 for (i = 0; i < RLIM_NLIMITS; i++)
1927 fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
1929 if (c->ioprio_set) {
1933 r = ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c->ioprio), &class_str);
1937 "%sIOSchedulingClass: %s\n"
1938 "%sIOPriority: %i\n",
1939 prefix, strna(class_str),
1940 prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
1944 if (c->cpu_sched_set) {
1948 r = sched_policy_to_string_alloc(c->cpu_sched_policy, &policy_str);
1952 "%sCPUSchedulingPolicy: %s\n"
1953 "%sCPUSchedulingPriority: %i\n"
1954 "%sCPUSchedulingResetOnFork: %s\n",
1955 prefix, strna(policy_str),
1956 prefix, c->cpu_sched_priority,
1957 prefix, yes_no(c->cpu_sched_reset_on_fork));
1962 fprintf(f, "%sCPUAffinity:", prefix);
1963 for (i = 0; i < c->cpuset_ncpus; i++)
1964 if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
1965 fprintf(f, " %i", i);
1969 if (c->timer_slack_nsec != (nsec_t) -1)
1970 fprintf(f, "%sTimerSlackNSec: %lu\n", prefix, (unsigned long)c->timer_slack_nsec);
1973 "%sStandardInput: %s\n"
1974 "%sStandardOutput: %s\n"
1975 "%sStandardError: %s\n",
1976 prefix, exec_input_to_string(c->std_input),
1977 prefix, exec_output_to_string(c->std_output),
1978 prefix, exec_output_to_string(c->std_error));
1984 "%sTTYVHangup: %s\n"
1985 "%sTTYVTDisallocate: %s\n",
1986 prefix, c->tty_path,
1987 prefix, yes_no(c->tty_reset),
1988 prefix, yes_no(c->tty_vhangup),
1989 prefix, yes_no(c->tty_vt_disallocate));
1991 if (c->std_output == EXEC_OUTPUT_SYSLOG || c->std_output == EXEC_OUTPUT_KMSG || c->std_output == EXEC_OUTPUT_JOURNAL ||
1992 c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE ||
1993 c->std_error == EXEC_OUTPUT_SYSLOG || c->std_error == EXEC_OUTPUT_KMSG || c->std_error == EXEC_OUTPUT_JOURNAL ||
1994 c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE) {
1995 char *fac_str, *lvl_str;
1998 r = log_facility_unshifted_to_string_alloc(c->syslog_priority >> 3, &fac_str);
2002 r = log_level_to_string_alloc(LOG_PRI(c->syslog_priority), &lvl_str);
2007 "%sSyslogFacility: %s\n"
2008 "%sSyslogLevel: %s\n",
2009 prefix, strna(fac_str),
2010 prefix, strna(lvl_str));
2015 if (c->capabilities) {
2017 if ((t = cap_to_text(c->capabilities, NULL))) {
2018 fprintf(f, "%sCapabilities: %s\n",
2025 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
2027 (c->secure_bits & 1<<SECURE_KEEP_CAPS) ? " keep-caps" : "",
2028 (c->secure_bits & 1<<SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
2029 (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
2030 (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
2031 (c->secure_bits & 1<<SECURE_NOROOT) ? " noroot" : "",
2032 (c->secure_bits & 1<<SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
2034 if (c->capability_bounding_set_drop) {
2036 fprintf(f, "%sCapabilityBoundingSet:", prefix);
2038 for (l = 0; l <= cap_last_cap(); l++)
2039 if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) {
2042 if ((t = cap_to_name(l))) {
2043 fprintf(f, " %s", t);
2052 fprintf(f, "%sUser: %s\n", prefix, c->user);
2054 fprintf(f, "%sGroup: %s\n", prefix, c->group);
2056 if (strv_length(c->supplementary_groups) > 0) {
2057 fprintf(f, "%sSupplementaryGroups:", prefix);
2058 strv_fprintf(f, c->supplementary_groups);
2063 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
2065 if (strv_length(c->read_write_dirs) > 0) {
2066 fprintf(f, "%sReadWriteDirs:", prefix);
2067 strv_fprintf(f, c->read_write_dirs);
2071 if (strv_length(c->read_only_dirs) > 0) {
2072 fprintf(f, "%sReadOnlyDirs:", prefix);
2073 strv_fprintf(f, c->read_only_dirs);
2077 if (strv_length(c->inaccessible_dirs) > 0) {
2078 fprintf(f, "%sInaccessibleDirs:", prefix);
2079 strv_fprintf(f, c->inaccessible_dirs);
2085 "%sUtmpIdentifier: %s\n",
2086 prefix, c->utmp_id);
2089 void exec_status_start(ExecStatus *s, pid_t pid) {
2094 dual_timestamp_get(&s->start_timestamp);
2097 void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
2100 if (s->pid && s->pid != pid)
2104 dual_timestamp_get(&s->exit_timestamp);
2110 if (context->utmp_id)
2111 utmp_put_dead_process(context->utmp_id, pid, code, status);
2113 exec_context_tty_reset(context);
2117 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
2118 char buf[FORMAT_TIMESTAMP_MAX];
2131 prefix, (unsigned long) s->pid);
2133 if (s->start_timestamp.realtime > 0)
2135 "%sStart Timestamp: %s\n",
2136 prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
2138 if (s->exit_timestamp.realtime > 0)
2140 "%sExit Timestamp: %s\n"
2142 "%sExit Status: %i\n",
2143 prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
2144 prefix, sigchld_code_to_string(s->code),
2148 char *exec_command_line(char **argv) {
2156 STRV_FOREACH(a, argv)
2159 if (!(n = new(char, k)))
2163 STRV_FOREACH(a, argv) {
2170 if (strpbrk(*a, WHITESPACE)) {
2181 /* FIXME: this doesn't really handle arguments that have
2182 * spaces and ticks in them */
2187 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
2189 const char *prefix2;
2198 p2 = strappend(prefix, "\t");
2199 prefix2 = p2 ? p2 : prefix;
2201 cmd = exec_command_line(c->argv);
2204 "%sCommand Line: %s\n",
2205 prefix, cmd ? cmd : strerror(ENOMEM));
2209 exec_status_dump(&c->exec_status, f, prefix2);
2214 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
2220 LIST_FOREACH(command, c, c)
2221 exec_command_dump(c, f, prefix);
2224 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
2231 /* It's kind of important, that we keep the order here */
2232 LIST_FIND_TAIL(command, *l, end);
2233 LIST_INSERT_AFTER(command, *l, end, e);
2238 int exec_command_set(ExecCommand *c, const char *path, ...) {
2246 l = strv_new_ap(path, ap);
2267 static int exec_runtime_allocate(ExecRuntime **rt) {
2272 *rt = new0(ExecRuntime, 1);
2277 (*rt)->netns_storage_socket[0] = (*rt)->netns_storage_socket[1] = -1;
2282 int exec_runtime_make(ExecRuntime **rt, ExecContext *c, const char *id) {
2292 if (!c->private_network && !c->private_tmp)
2295 r = exec_runtime_allocate(rt);
2299 if (c->private_network && (*rt)->netns_storage_socket[0] < 0) {
2300 if (socketpair(AF_UNIX, SOCK_DGRAM, 0, (*rt)->netns_storage_socket) < 0)
2304 if (c->private_tmp && !(*rt)->tmp_dir) {
2305 r = setup_tmp_dirs(id, &(*rt)->tmp_dir, &(*rt)->var_tmp_dir);
2313 ExecRuntime *exec_runtime_ref(ExecRuntime *r) {
2315 assert(r->n_ref > 0);
2321 ExecRuntime *exec_runtime_unref(ExecRuntime *r) {
2326 assert(r->n_ref > 0);
2329 if (r->n_ref <= 0) {
2331 free(r->var_tmp_dir);
2332 close_pipe(r->netns_storage_socket);
2339 int exec_runtime_serialize(ExecRuntime *rt, Unit *u, FILE *f, FDSet *fds) {
2348 unit_serialize_item(u, f, "tmp-dir", rt->tmp_dir);
2350 if (rt->var_tmp_dir)
2351 unit_serialize_item(u, f, "var-tmp-dir", rt->var_tmp_dir);
2353 if (rt->netns_storage_socket[0] >= 0) {
2356 copy = fdset_put_dup(fds, rt->netns_storage_socket[0]);
2360 unit_serialize_item_format(u, f, "netns-socket-0", "%i", copy);
2363 if (rt->netns_storage_socket[1] >= 0) {
2366 copy = fdset_put_dup(fds, rt->netns_storage_socket[1]);
2370 unit_serialize_item_format(u, f, "netns-socket-1", "%i", copy);
2376 int exec_runtime_deserialize_item(ExecRuntime **rt, Unit *u, const char *key, const char *value, FDSet *fds) {
2383 if (streq(key, "tmp-dir")) {
2386 r = exec_runtime_allocate(rt);
2390 copy = strdup(value);
2394 free((*rt)->tmp_dir);
2395 (*rt)->tmp_dir = copy;
2397 } else if (streq(key, "var-tmp-dir")) {
2400 r = exec_runtime_allocate(rt);
2404 copy = strdup(value);
2408 free((*rt)->var_tmp_dir);
2409 (*rt)->var_tmp_dir = copy;
2411 } else if (streq(key, "netns-socket-0")) {
2414 r = exec_runtime_allocate(rt);
2418 if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
2419 log_debug_unit(u->id, "Failed to parse netns socket value %s", value);
2421 if ((*rt)->netns_storage_socket[0] >= 0)
2422 close_nointr_nofail((*rt)->netns_storage_socket[0]);
2424 (*rt)->netns_storage_socket[0] = fdset_remove(fds, fd);
2426 } else if (streq(key, "netns-socket-1")) {
2429 r = exec_runtime_allocate(rt);
2433 if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
2434 log_debug_unit(u->id, "Failed to parse netns socket value %s", value);
2436 if ((*rt)->netns_storage_socket[1] >= 0)
2437 close_nointr_nofail((*rt)->netns_storage_socket[1]);
2439 (*rt)->netns_storage_socket[1] = fdset_remove(fds, fd);
2447 static void *remove_tmpdir_thread(void *p) {
2448 _cleanup_free_ char *path = p;
2450 rm_rf_dangerous(path, false, true, false);
2454 void exec_runtime_destroy(ExecRuntime *rt) {
2458 /* If there are multiple users of this, let's leave the stuff around */
2463 log_debug("Spawning thread to nuke %s", rt->tmp_dir);
2464 asynchronous_job(remove_tmpdir_thread, rt->tmp_dir);
2468 if (rt->var_tmp_dir) {
2469 log_debug("Spawning thread to nuke %s", rt->var_tmp_dir);
2470 asynchronous_job(remove_tmpdir_thread, rt->var_tmp_dir);
2471 rt->var_tmp_dir = NULL;
2474 close_pipe(rt->netns_storage_socket);
2477 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
2478 [EXEC_INPUT_NULL] = "null",
2479 [EXEC_INPUT_TTY] = "tty",
2480 [EXEC_INPUT_TTY_FORCE] = "tty-force",
2481 [EXEC_INPUT_TTY_FAIL] = "tty-fail",
2482 [EXEC_INPUT_SOCKET] = "socket"
2485 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
2487 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
2488 [EXEC_OUTPUT_INHERIT] = "inherit",
2489 [EXEC_OUTPUT_NULL] = "null",
2490 [EXEC_OUTPUT_TTY] = "tty",
2491 [EXEC_OUTPUT_SYSLOG] = "syslog",
2492 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
2493 [EXEC_OUTPUT_KMSG] = "kmsg",
2494 [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
2495 [EXEC_OUTPUT_JOURNAL] = "journal",
2496 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE] = "journal+console",
2497 [EXEC_OUTPUT_SOCKET] = "socket"
2500 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);