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>
42 #include <sys/personality.h>
47 #include <security/pam_appl.h>
51 #include <selinux/selinux.h>
59 #include <sys/apparmor.h>
65 #include "capability.h"
68 #include "sd-messages.h"
70 #include "securebits.h"
71 #include "namespace.h"
72 #include "exit-status.h"
74 #include "utmp-wtmp.h"
76 #include "path-util.h"
81 #include "selinux-util.h"
82 #include "errno-list.h"
85 #include "apparmor-util.h"
88 #include "seccomp-util.h"
91 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
92 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
94 /* This assumes there is a 'tty' group */
97 #define SNDBUF_SIZE (8*1024*1024)
99 static int shift_fds(int fds[], unsigned n_fds) {
100 int start, restart_from;
105 /* Modifies the fds array! (sorts it) */
115 for (i = start; i < (int) n_fds; i++) {
118 /* Already at right index? */
122 if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
128 /* Hmm, the fd we wanted isn't free? Then
129 * let's remember that and try again from here*/
130 if (nfd != i+3 && restart_from < 0)
134 if (restart_from < 0)
137 start = restart_from;
143 static int flags_fds(const int fds[], unsigned n_fds, bool nonblock) {
152 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
154 for (i = 0; i < n_fds; i++) {
156 if ((r = fd_nonblock(fds[i], nonblock)) < 0)
159 /* We unconditionally drop FD_CLOEXEC from the fds,
160 * since after all we want to pass these fds to our
163 if ((r = fd_cloexec(fds[i], false)) < 0)
170 _pure_ static const char *tty_path(const ExecContext *context) {
173 if (context->tty_path)
174 return context->tty_path;
176 return "/dev/console";
179 static void exec_context_tty_reset(const ExecContext *context) {
182 if (context->tty_vhangup)
183 terminal_vhangup(tty_path(context));
185 if (context->tty_reset)
186 reset_terminal(tty_path(context));
188 if (context->tty_vt_disallocate && context->tty_path)
189 vt_disallocate(context->tty_path);
192 static bool is_terminal_output(ExecOutput o) {
194 o == EXEC_OUTPUT_TTY ||
195 o == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
196 o == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
197 o == EXEC_OUTPUT_JOURNAL_AND_CONSOLE;
200 static int open_null_as(int flags, int nfd) {
205 fd = open("/dev/null", flags|O_NOCTTY);
210 r = dup2(fd, nfd) < 0 ? -errno : nfd;
218 static int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, const char *unit_id, int nfd) {
220 union sockaddr_union sa = {
221 .un.sun_family = AF_UNIX,
222 .un.sun_path = "/run/systemd/journal/stdout",
226 assert(output < _EXEC_OUTPUT_MAX);
230 fd = socket(AF_UNIX, SOCK_STREAM, 0);
234 r = connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path));
240 if (shutdown(fd, SHUT_RD) < 0) {
245 fd_inc_sndbuf(fd, SNDBUF_SIZE);
255 context->syslog_identifier ? context->syslog_identifier : ident,
257 context->syslog_priority,
258 !!context->syslog_level_prefix,
259 output == EXEC_OUTPUT_SYSLOG || output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE,
260 output == EXEC_OUTPUT_KMSG || output == EXEC_OUTPUT_KMSG_AND_CONSOLE,
261 is_terminal_output(output));
264 r = dup2(fd, nfd) < 0 ? -errno : nfd;
271 static int open_terminal_as(const char *path, mode_t mode, int nfd) {
277 if ((fd = open_terminal(path, mode | O_NOCTTY)) < 0)
281 r = dup2(fd, nfd) < 0 ? -errno : nfd;
289 static bool is_terminal_input(ExecInput i) {
291 i == EXEC_INPUT_TTY ||
292 i == EXEC_INPUT_TTY_FORCE ||
293 i == EXEC_INPUT_TTY_FAIL;
296 static int fixup_input(ExecInput std_input, int socket_fd, bool apply_tty_stdin) {
298 if (is_terminal_input(std_input) && !apply_tty_stdin)
299 return EXEC_INPUT_NULL;
301 if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0)
302 return EXEC_INPUT_NULL;
307 static int fixup_output(ExecOutput std_output, int socket_fd) {
309 if (std_output == EXEC_OUTPUT_SOCKET && socket_fd < 0)
310 return EXEC_OUTPUT_INHERIT;
315 static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty_stdin) {
320 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
324 case EXEC_INPUT_NULL:
325 return open_null_as(O_RDONLY, STDIN_FILENO);
328 case EXEC_INPUT_TTY_FORCE:
329 case EXEC_INPUT_TTY_FAIL: {
332 fd = acquire_terminal(tty_path(context),
333 i == EXEC_INPUT_TTY_FAIL,
334 i == EXEC_INPUT_TTY_FORCE,
340 if (fd != STDIN_FILENO) {
341 r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
349 case EXEC_INPUT_SOCKET:
350 return dup2(socket_fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
353 assert_not_reached("Unknown input type");
357 static int setup_output(const ExecContext *context, int fileno, int socket_fd, const char *ident, const char *unit_id, bool apply_tty_stdin) {
365 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
366 o = fixup_output(context->std_output, socket_fd);
368 if (fileno == STDERR_FILENO) {
370 e = fixup_output(context->std_error, socket_fd);
372 /* This expects the input and output are already set up */
374 /* Don't change the stderr file descriptor if we inherit all
375 * the way and are not on a tty */
376 if (e == EXEC_OUTPUT_INHERIT &&
377 o == EXEC_OUTPUT_INHERIT &&
378 i == EXEC_INPUT_NULL &&
379 !is_terminal_input(context->std_input) &&
383 /* Duplicate from stdout if possible */
384 if (e == o || e == EXEC_OUTPUT_INHERIT)
385 return dup2(STDOUT_FILENO, fileno) < 0 ? -errno : fileno;
389 } else if (o == EXEC_OUTPUT_INHERIT) {
390 /* If input got downgraded, inherit the original value */
391 if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input))
392 return open_terminal_as(tty_path(context), O_WRONLY, fileno);
394 /* If the input is connected to anything that's not a /dev/null, inherit that... */
395 if (i != EXEC_INPUT_NULL)
396 return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
398 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
402 /* We need to open /dev/null here anew, to get the right access mode. */
403 return open_null_as(O_WRONLY, fileno);
408 case EXEC_OUTPUT_NULL:
409 return open_null_as(O_WRONLY, fileno);
411 case EXEC_OUTPUT_TTY:
412 if (is_terminal_input(i))
413 return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
415 /* We don't reset the terminal if this is just about output */
416 return open_terminal_as(tty_path(context), O_WRONLY, fileno);
418 case EXEC_OUTPUT_SYSLOG:
419 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
420 case EXEC_OUTPUT_KMSG:
421 case EXEC_OUTPUT_KMSG_AND_CONSOLE:
422 case EXEC_OUTPUT_JOURNAL:
423 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE:
424 r = connect_logger_as(context, o, ident, unit_id, fileno);
426 log_struct_unit(LOG_CRIT, unit_id,
427 "MESSAGE=Failed to connect std%s of %s to the journal socket: %s",
428 fileno == STDOUT_FILENO ? "out" : "err",
429 unit_id, strerror(-r),
432 r = open_null_as(O_WRONLY, fileno);
436 case EXEC_OUTPUT_SOCKET:
437 assert(socket_fd >= 0);
438 return dup2(socket_fd, fileno) < 0 ? -errno : fileno;
441 assert_not_reached("Unknown error type");
445 static int chown_terminal(int fd, uid_t uid) {
450 /* This might fail. What matters are the results. */
451 (void) fchown(fd, uid, -1);
452 (void) fchmod(fd, TTY_MODE);
454 if (fstat(fd, &st) < 0)
457 if (st.st_uid != uid || (st.st_mode & 0777) != TTY_MODE)
463 static int setup_confirm_stdio(int *_saved_stdin,
464 int *_saved_stdout) {
465 int fd = -1, saved_stdin, saved_stdout = -1, r;
467 assert(_saved_stdin);
468 assert(_saved_stdout);
470 saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3);
474 saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3);
475 if (saved_stdout < 0) {
480 fd = acquire_terminal(
485 DEFAULT_CONFIRM_USEC);
491 r = chown_terminal(fd, getuid());
495 if (dup2(fd, STDIN_FILENO) < 0) {
500 if (dup2(fd, STDOUT_FILENO) < 0) {
508 *_saved_stdin = saved_stdin;
509 *_saved_stdout = saved_stdout;
514 safe_close(saved_stdout);
515 safe_close(saved_stdin);
521 _printf_(1, 2) static int write_confirm_message(const char *format, ...) {
522 _cleanup_close_ int fd = -1;
527 fd = open_terminal("/dev/console", O_WRONLY|O_NOCTTY|O_CLOEXEC);
531 va_start(ap, format);
532 vdprintf(fd, format, ap);
538 static int restore_confirm_stdio(int *saved_stdin,
544 assert(saved_stdout);
548 if (*saved_stdin >= 0)
549 if (dup2(*saved_stdin, STDIN_FILENO) < 0)
552 if (*saved_stdout >= 0)
553 if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
556 safe_close(*saved_stdin);
557 safe_close(*saved_stdout);
562 static int ask_for_confirmation(char *response, char **argv) {
563 int saved_stdout = -1, saved_stdin = -1, r;
564 _cleanup_free_ char *line = NULL;
566 r = setup_confirm_stdio(&saved_stdin, &saved_stdout);
570 line = exec_command_line(argv);
574 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);
941 static int apply_seccomp(ExecContext *c) {
942 uint32_t negative_action, action;
943 scmp_filter_ctx *seccomp;
950 negative_action = c->syscall_errno == 0 ? SCMP_ACT_KILL : SCMP_ACT_ERRNO(c->syscall_errno);
952 seccomp = seccomp_init(c->syscall_whitelist ? negative_action : SCMP_ACT_ALLOW);
956 if (c->syscall_archs) {
958 SET_FOREACH(id, c->syscall_archs, i) {
959 r = seccomp_arch_add(seccomp, PTR_TO_UINT32(id) - 1);
967 r = seccomp_add_secondary_archs(seccomp);
972 action = c->syscall_whitelist ? SCMP_ACT_ALLOW : negative_action;
973 SET_FOREACH(id, c->syscall_filter, i) {
974 r = seccomp_rule_add(seccomp, action, PTR_TO_INT(id) - 1, 0);
979 r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0);
983 r = seccomp_load(seccomp);
986 seccomp_release(seccomp);
990 static int apply_address_families(ExecContext *c) {
991 scmp_filter_ctx *seccomp;
997 seccomp = seccomp_init(SCMP_ACT_ALLOW);
1001 r = seccomp_add_secondary_archs(seccomp);
1005 if (c->address_families_whitelist) {
1006 int af, first = 0, last = 0;
1009 /* If this is a whitelist, we first block the address
1010 * families that are out of range and then everything
1011 * that is not in the set. First, we find the lowest
1012 * and highest address family in the set. */
1014 SET_FOREACH(afp, c->address_families, i) {
1015 af = PTR_TO_INT(afp);
1017 if (af <= 0 || af >= af_max())
1020 if (first == 0 || af < first)
1023 if (last == 0 || af > last)
1027 assert((first == 0) == (last == 0));
1031 /* No entries in the valid range, block everything */
1032 r = seccomp_rule_add(
1034 SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1042 /* Block everything below the first entry */
1043 r = seccomp_rule_add(
1045 SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1048 SCMP_A0(SCMP_CMP_LT, first));
1052 /* Block everything above the last entry */
1053 r = seccomp_rule_add(
1055 SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1058 SCMP_A0(SCMP_CMP_GT, last));
1062 /* Block everything between the first and last
1064 for (af = 1; af < af_max(); af++) {
1066 if (set_contains(c->address_families, INT_TO_PTR(af)))
1069 r = seccomp_rule_add(
1071 SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1074 SCMP_A0(SCMP_CMP_EQ, af));
1083 /* If this is a blacklist, then generate one rule for
1084 * each address family that are then combined in OR
1087 SET_FOREACH(af, c->address_families, i) {
1089 r = seccomp_rule_add(
1091 SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1094 SCMP_A0(SCMP_CMP_EQ, PTR_TO_INT(af)));
1100 r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0);
1104 r = seccomp_load(seccomp);
1107 seccomp_release(seccomp);
1113 static void do_idle_pipe_dance(int idle_pipe[4]) {
1117 safe_close(idle_pipe[1]);
1118 safe_close(idle_pipe[2]);
1120 if (idle_pipe[0] >= 0) {
1123 r = fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT_USEC);
1125 if (idle_pipe[3] >= 0 && r == 0 /* timeout */) {
1126 /* Signal systemd that we are bored and want to continue. */
1127 write(idle_pipe[3], "x", 1);
1129 /* Wait for systemd to react to the signal above. */
1130 fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT2_USEC);
1133 safe_close(idle_pipe[0]);
1137 safe_close(idle_pipe[3]);
1140 static int build_environment(
1143 usec_t watchdog_usec,
1145 const char *username,
1149 _cleanup_strv_free_ char **our_env = NULL;
1156 our_env = new0(char*, 10);
1161 if (asprintf(&x, "LISTEN_PID="PID_FMT, getpid()) < 0)
1163 our_env[n_env++] = x;
1165 if (asprintf(&x, "LISTEN_FDS=%u", n_fds) < 0)
1167 our_env[n_env++] = x;
1170 if (watchdog_usec > 0) {
1171 if (asprintf(&x, "WATCHDOG_PID="PID_FMT, getpid()) < 0)
1173 our_env[n_env++] = x;
1175 if (asprintf(&x, "WATCHDOG_USEC="USEC_FMT, watchdog_usec) < 0)
1177 our_env[n_env++] = x;
1181 x = strappend("HOME=", home);
1184 our_env[n_env++] = x;
1188 x = strappend("LOGNAME=", username);
1191 our_env[n_env++] = x;
1193 x = strappend("USER=", username);
1196 our_env[n_env++] = x;
1200 x = strappend("SHELL=", shell);
1203 our_env[n_env++] = x;
1206 if (is_terminal_input(c->std_input) ||
1207 c->std_output == EXEC_OUTPUT_TTY ||
1208 c->std_error == EXEC_OUTPUT_TTY ||
1211 x = strdup(default_term_for_tty(tty_path(c)));
1214 our_env[n_env++] = x;
1217 our_env[n_env++] = NULL;
1218 assert(n_env <= 10);
1226 int exec_spawn(ExecCommand *command,
1228 ExecContext *context,
1229 int fds[], unsigned n_fds,
1231 bool apply_permissions,
1233 bool apply_tty_stdin,
1235 CGroupControllerMask cgroup_supported,
1236 const char *cgroup_path,
1237 const char *runtime_prefix,
1238 const char *unit_id,
1239 usec_t watchdog_usec,
1241 ExecRuntime *runtime,
1244 _cleanup_strv_free_ char **files_env = NULL;
1253 assert(fds || n_fds <= 0);
1255 if (context->std_input == EXEC_INPUT_SOCKET ||
1256 context->std_output == EXEC_OUTPUT_SOCKET ||
1257 context->std_error == EXEC_OUTPUT_SOCKET) {
1269 r = exec_context_load_environment(context, &files_env);
1271 log_struct_unit(LOG_ERR,
1273 "MESSAGE=Failed to load environment files: %s", strerror(-r),
1280 argv = command->argv;
1282 line = exec_command_line(argv);
1286 log_struct_unit(LOG_DEBUG,
1288 "EXECUTABLE=%s", command->path,
1289 "MESSAGE=About to execute: %s", line,
1298 _cleanup_strv_free_ char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
1299 const char *username = NULL, *home = NULL, *shell = NULL;
1300 unsigned n_dont_close = 0;
1301 int dont_close[n_fds + 3];
1302 uid_t uid = (uid_t) -1;
1303 gid_t gid = (gid_t) -1;
1309 rename_process_from_path(command->path);
1311 /* We reset exactly these signals, since they are the
1312 * only ones we set to SIG_IGN in the main daemon. All
1313 * others we leave untouched because we set them to
1314 * SIG_DFL or a valid handler initially, both of which
1315 * will be demoted to SIG_DFL. */
1316 default_signals(SIGNALS_CRASH_HANDLER,
1317 SIGNALS_IGNORE, -1);
1319 if (context->ignore_sigpipe)
1320 ignore_signals(SIGPIPE, -1);
1322 assert_se(sigemptyset(&ss) == 0);
1323 if (sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
1325 r = EXIT_SIGNAL_MASK;
1330 do_idle_pipe_dance(idle_pipe);
1332 /* Close sockets very early to make sure we don't
1333 * block init reexecution because it cannot bind its
1338 dont_close[n_dont_close++] = socket_fd;
1340 memcpy(dont_close + n_dont_close, fds, sizeof(int) * n_fds);
1341 n_dont_close += n_fds;
1344 if (runtime->netns_storage_socket[0] >= 0)
1345 dont_close[n_dont_close++] = runtime->netns_storage_socket[0];
1346 if (runtime->netns_storage_socket[1] >= 0)
1347 dont_close[n_dont_close++] = runtime->netns_storage_socket[1];
1350 err = close_all_fds(dont_close, n_dont_close);
1356 if (!context->same_pgrp)
1363 exec_context_tty_reset(context);
1365 if (confirm_spawn) {
1368 err = ask_for_confirmation(&response, argv);
1369 if (err == -ETIMEDOUT)
1370 write_confirm_message("Confirmation question timed out, assuming positive response.\n");
1372 write_confirm_message("Couldn't ask confirmation question, assuming positive response: %s\n", strerror(-err));
1373 else if (response == 's') {
1374 write_confirm_message("Skipping execution.\n");
1378 } else if (response == 'n') {
1379 write_confirm_message("Failing execution.\n");
1385 /* If a socket is connected to STDIN/STDOUT/STDERR, we
1386 * must sure to drop O_NONBLOCK */
1388 fd_nonblock(socket_fd, false);
1390 err = setup_input(context, socket_fd, apply_tty_stdin);
1396 err = setup_output(context, STDOUT_FILENO, socket_fd, basename(command->path), unit_id, apply_tty_stdin);
1402 err = setup_output(context, STDERR_FILENO, socket_fd, basename(command->path), unit_id, apply_tty_stdin);
1409 err = cg_attach_everywhere(cgroup_supported, cgroup_path, 0);
1416 if (context->oom_score_adjust_set) {
1419 snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
1422 if (write_string_file("/proc/self/oom_score_adj", t) < 0) {
1424 r = EXIT_OOM_ADJUST;
1429 if (context->nice_set)
1430 if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
1436 if (context->cpu_sched_set) {
1437 struct sched_param param = {
1438 .sched_priority = context->cpu_sched_priority,
1441 r = sched_setscheduler(0,
1442 context->cpu_sched_policy |
1443 (context->cpu_sched_reset_on_fork ?
1444 SCHED_RESET_ON_FORK : 0),
1448 r = EXIT_SETSCHEDULER;
1453 if (context->cpuset)
1454 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
1456 r = EXIT_CPUAFFINITY;
1460 if (context->ioprio_set)
1461 if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
1467 if (context->timer_slack_nsec != (nsec_t) -1)
1468 if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
1470 r = EXIT_TIMERSLACK;
1474 if (context->personality != 0xffffffffUL)
1475 if (personality(context->personality) < 0) {
1477 r = EXIT_PERSONALITY;
1481 if (context->utmp_id)
1482 utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path);
1484 if (context->user) {
1485 username = context->user;
1486 err = get_user_creds(&username, &uid, &gid, &home, &shell);
1492 if (is_terminal_input(context->std_input)) {
1493 err = chown_terminal(STDIN_FILENO, uid);
1502 if (cgroup_path && context->user && context->pam_name) {
1503 err = cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, 0644, uid, gid);
1510 err = cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, 0755, uid, gid);
1518 if (!strv_isempty(context->runtime_directory) && runtime_prefix) {
1521 STRV_FOREACH(rt, context->runtime_directory) {
1522 _cleanup_free_ char *p;
1524 p = strjoin(runtime_prefix, "/", *rt, NULL);
1526 r = EXIT_RUNTIME_DIRECTORY;
1531 err = mkdir_safe(p, context->runtime_directory_mode, uid, gid);
1533 r = EXIT_RUNTIME_DIRECTORY;
1539 if (apply_permissions) {
1540 err = enforce_groups(context, username, gid);
1547 umask(context->umask);
1550 if (apply_permissions && context->pam_name && username) {
1551 err = setup_pam(context->pam_name, username, uid, context->tty_path, &pam_env, fds, n_fds);
1558 if (context->private_network && runtime && runtime->netns_storage_socket[0] >= 0) {
1559 err = setup_netns(runtime->netns_storage_socket);
1566 if (!strv_isempty(context->read_write_dirs) ||
1567 !strv_isempty(context->read_only_dirs) ||
1568 !strv_isempty(context->inaccessible_dirs) ||
1569 context->mount_flags != 0 ||
1570 (context->private_tmp && runtime && (runtime->tmp_dir || runtime->var_tmp_dir)) ||
1571 context->private_devices ||
1572 context->protect_system != PROTECT_SYSTEM_NO ||
1573 context->protect_home != PROTECT_HOME_NO) {
1575 char *tmp = NULL, *var = NULL;
1577 /* The runtime struct only contains the parent
1578 * of the private /tmp, which is
1579 * non-accessible to world users. Inside of it
1580 * there's a /tmp that is sticky, and that's
1581 * the one we want to use here. */
1583 if (context->private_tmp && runtime) {
1584 if (runtime->tmp_dir)
1585 tmp = strappenda(runtime->tmp_dir, "/tmp");
1586 if (runtime->var_tmp_dir)
1587 var = strappenda(runtime->var_tmp_dir, "/tmp");
1590 err = setup_namespace(
1591 context->read_write_dirs,
1592 context->read_only_dirs,
1593 context->inaccessible_dirs,
1596 context->private_devices,
1597 context->protect_home,
1598 context->protect_system,
1599 context->mount_flags);
1607 if (context->root_directory)
1608 if (chroot(context->root_directory) < 0) {
1614 if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
1620 _cleanup_free_ char *d = NULL;
1622 if (asprintf(&d, "%s/%s",
1623 context->root_directory ? context->root_directory : "",
1624 context->working_directory ? context->working_directory : "") < 0) {
1637 /* We repeat the fd closing here, to make sure that
1638 * nothing is leaked from the PAM modules */
1639 err = close_all_fds(fds, n_fds);
1641 err = shift_fds(fds, n_fds);
1643 err = flags_fds(fds, n_fds, context->non_blocking);
1649 if (apply_permissions) {
1651 for (i = 0; i < _RLIMIT_MAX; i++) {
1652 if (!context->rlimit[i])
1655 if (setrlimit_closest(i, context->rlimit[i]) < 0) {
1662 if (context->capability_bounding_set_drop) {
1663 err = capability_bounding_set_drop(context->capability_bounding_set_drop, false);
1665 r = EXIT_CAPABILITIES;
1670 if (context->user) {
1671 err = enforce_user(context, uid);
1678 /* PR_GET_SECUREBITS is not privileged, while
1679 * PR_SET_SECUREBITS is. So to suppress
1680 * potential EPERMs we'll try not to call
1681 * PR_SET_SECUREBITS unless necessary. */
1682 if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
1683 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
1685 r = EXIT_SECUREBITS;
1689 if (context->capabilities)
1690 if (cap_set_proc(context->capabilities) < 0) {
1692 r = EXIT_CAPABILITIES;
1696 if (context->no_new_privileges)
1697 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
1699 r = EXIT_NO_NEW_PRIVILEGES;
1704 if (context->address_families_whitelist ||
1705 !set_isempty(context->address_families)) {
1706 err = apply_address_families(context);
1708 r = EXIT_ADDRESS_FAMILIES;
1713 if (context->syscall_whitelist ||
1714 !set_isempty(context->syscall_filter) ||
1715 !set_isempty(context->syscall_archs)) {
1716 err = apply_seccomp(context);
1725 if (context->selinux_context && use_selinux()) {
1726 err = setexeccon(context->selinux_context);
1727 if (err < 0 && !context->selinux_context_ignore) {
1728 r = EXIT_SELINUX_CONTEXT;
1734 #ifdef HAVE_APPARMOR
1735 if (context->apparmor_profile && use_apparmor()) {
1736 err = aa_change_onexec(context->apparmor_profile);
1737 if (err < 0 && !context->apparmor_profile_ignore) {
1738 r = EXIT_APPARMOR_PROFILE;
1745 err = build_environment(context, n_fds, watchdog_usec, home, username, shell, &our_env);
1751 final_env = strv_env_merge(5,
1754 context->environment,
1764 final_argv = replace_env_argv(argv, final_env);
1771 final_env = strv_env_clean(final_env);
1773 if (_unlikely_(log_get_max_level() >= LOG_PRI(LOG_DEBUG))) {
1774 line = exec_command_line(final_argv);
1777 log_struct_unit(LOG_DEBUG,
1779 "EXECUTABLE=%s", command->path,
1780 "MESSAGE=Executing: %s", line,
1787 execve(command->path, final_argv, final_env);
1794 log_struct(LOG_ERR, MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED),
1795 "EXECUTABLE=%s", command->path,
1796 "MESSAGE=Failed at step %s spawning %s: %s",
1797 exit_status_to_string(r, EXIT_STATUS_SYSTEMD),
1798 command->path, strerror(-err),
1807 log_struct_unit(LOG_DEBUG,
1809 "MESSAGE=Forked %s as "PID_FMT,
1813 /* We add the new process to the cgroup both in the child (so
1814 * that we can be sure that no user code is ever executed
1815 * outside of the cgroup) and in the parent (so that we can be
1816 * sure that when we kill the cgroup the process will be
1819 cg_attach(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, pid);
1821 exec_status_start(&command->exec_status, pid);
1827 void exec_context_init(ExecContext *c) {
1831 c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
1832 c->cpu_sched_policy = SCHED_OTHER;
1833 c->syslog_priority = LOG_DAEMON|LOG_INFO;
1834 c->syslog_level_prefix = true;
1835 c->ignore_sigpipe = true;
1836 c->timer_slack_nsec = (nsec_t) -1;
1837 c->personality = 0xffffffffUL;
1838 c->runtime_directory_mode = 0755;
1841 void exec_context_done(ExecContext *c) {
1846 strv_free(c->environment);
1847 c->environment = NULL;
1849 strv_free(c->environment_files);
1850 c->environment_files = NULL;
1852 for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1854 c->rlimit[l] = NULL;
1857 free(c->working_directory);
1858 c->working_directory = NULL;
1859 free(c->root_directory);
1860 c->root_directory = NULL;
1865 free(c->syslog_identifier);
1866 c->syslog_identifier = NULL;
1874 strv_free(c->supplementary_groups);
1875 c->supplementary_groups = NULL;
1880 if (c->capabilities) {
1881 cap_free(c->capabilities);
1882 c->capabilities = NULL;
1885 strv_free(c->read_only_dirs);
1886 c->read_only_dirs = NULL;
1888 strv_free(c->read_write_dirs);
1889 c->read_write_dirs = NULL;
1891 strv_free(c->inaccessible_dirs);
1892 c->inaccessible_dirs = NULL;
1895 CPU_FREE(c->cpuset);
1900 free(c->selinux_context);
1901 c->selinux_context = NULL;
1903 free(c->apparmor_profile);
1904 c->apparmor_profile = NULL;
1906 set_free(c->syscall_filter);
1907 c->syscall_filter = NULL;
1909 set_free(c->syscall_archs);
1910 c->syscall_archs = NULL;
1912 set_free(c->address_families);
1913 c->address_families = NULL;
1915 strv_free(c->runtime_directory);
1916 c->runtime_directory = NULL;
1919 int exec_context_destroy_runtime_directory(ExecContext *c, const char *runtime_prefix) {
1924 if (!runtime_prefix)
1927 STRV_FOREACH(i, c->runtime_directory) {
1928 _cleanup_free_ char *p;
1930 p = strjoin(runtime_prefix, "/", *i, NULL);
1934 /* We execute this synchronously, since we need to be
1935 * sure this is gone when we start the service
1937 rm_rf_dangerous(p, false, true, false);
1943 void exec_command_done(ExecCommand *c) {
1953 void exec_command_done_array(ExecCommand *c, unsigned n) {
1956 for (i = 0; i < n; i++)
1957 exec_command_done(c+i);
1960 void exec_command_free_list(ExecCommand *c) {
1964 LIST_REMOVE(command, c, i);
1965 exec_command_done(i);
1970 void exec_command_free_array(ExecCommand **c, unsigned n) {
1973 for (i = 0; i < n; i++) {
1974 exec_command_free_list(c[i]);
1979 int exec_context_load_environment(const ExecContext *c, char ***l) {
1980 char **i, **r = NULL;
1985 STRV_FOREACH(i, c->environment_files) {
1988 bool ignore = false;
1990 _cleanup_globfree_ glob_t pglob = {};
2000 if (!path_is_absolute(fn)) {
2008 /* Filename supports globbing, take all matching files */
2010 if (glob(fn, 0, NULL, &pglob) != 0) {
2015 return errno ? -errno : -EINVAL;
2017 count = pglob.gl_pathc;
2025 for (n = 0; n < count; n++) {
2026 k = load_env_file(pglob.gl_pathv[n], NULL, &p);
2034 /* Log invalid environment variables with filename */
2036 p = strv_env_clean_log(p, pglob.gl_pathv[n]);
2043 m = strv_env_merge(2, r, p);
2059 static bool tty_may_match_dev_console(const char *tty) {
2060 _cleanup_free_ char *active = NULL;
2063 if (startswith(tty, "/dev/"))
2066 /* trivial identity? */
2067 if (streq(tty, "console"))
2070 console = resolve_dev_console(&active);
2071 /* if we could not resolve, assume it may */
2075 /* "tty0" means the active VC, so it may be the same sometimes */
2076 return streq(console, tty) || (streq(console, "tty0") && tty_is_vc(tty));
2079 bool exec_context_may_touch_console(ExecContext *ec) {
2080 return (ec->tty_reset || ec->tty_vhangup || ec->tty_vt_disallocate ||
2081 is_terminal_input(ec->std_input) ||
2082 is_terminal_output(ec->std_output) ||
2083 is_terminal_output(ec->std_error)) &&
2084 tty_may_match_dev_console(tty_path(ec));
2087 static void strv_fprintf(FILE *f, char **l) {
2093 fprintf(f, " %s", *g);
2096 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
2103 prefix = strempty(prefix);
2107 "%sWorkingDirectory: %s\n"
2108 "%sRootDirectory: %s\n"
2109 "%sNonBlocking: %s\n"
2110 "%sPrivateTmp: %s\n"
2111 "%sPrivateNetwork: %s\n"
2112 "%sPrivateDevices: %s\n"
2113 "%sProtectHome: %s\n"
2114 "%sProtectSystem: %s\n"
2115 "%sIgnoreSIGPIPE: %s\n",
2117 prefix, c->working_directory ? c->working_directory : "/",
2118 prefix, c->root_directory ? c->root_directory : "/",
2119 prefix, yes_no(c->non_blocking),
2120 prefix, yes_no(c->private_tmp),
2121 prefix, yes_no(c->private_network),
2122 prefix, yes_no(c->private_devices),
2123 prefix, protect_home_to_string(c->protect_home),
2124 prefix, protect_system_to_string(c->protect_system),
2125 prefix, yes_no(c->ignore_sigpipe));
2127 STRV_FOREACH(e, c->environment)
2128 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
2130 STRV_FOREACH(e, c->environment_files)
2131 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
2138 if (c->oom_score_adjust_set)
2140 "%sOOMScoreAdjust: %i\n",
2141 prefix, c->oom_score_adjust);
2143 for (i = 0; i < RLIM_NLIMITS; i++)
2145 fprintf(f, "%s%s: "RLIM_FMT"\n",
2146 prefix, rlimit_to_string(i), c->rlimit[i]->rlim_max);
2148 if (c->ioprio_set) {
2149 _cleanup_free_ char *class_str = NULL;
2151 ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c->ioprio), &class_str);
2153 "%sIOSchedulingClass: %s\n"
2154 "%sIOPriority: %i\n",
2155 prefix, strna(class_str),
2156 prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
2159 if (c->cpu_sched_set) {
2160 _cleanup_free_ char *policy_str = NULL;
2162 sched_policy_to_string_alloc(c->cpu_sched_policy, &policy_str);
2164 "%sCPUSchedulingPolicy: %s\n"
2165 "%sCPUSchedulingPriority: %i\n"
2166 "%sCPUSchedulingResetOnFork: %s\n",
2167 prefix, strna(policy_str),
2168 prefix, c->cpu_sched_priority,
2169 prefix, yes_no(c->cpu_sched_reset_on_fork));
2173 fprintf(f, "%sCPUAffinity:", prefix);
2174 for (i = 0; i < c->cpuset_ncpus; i++)
2175 if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
2176 fprintf(f, " %u", i);
2180 if (c->timer_slack_nsec != (nsec_t) -1)
2181 fprintf(f, "%sTimerSlackNSec: "NSEC_FMT "\n", prefix, c->timer_slack_nsec);
2184 "%sStandardInput: %s\n"
2185 "%sStandardOutput: %s\n"
2186 "%sStandardError: %s\n",
2187 prefix, exec_input_to_string(c->std_input),
2188 prefix, exec_output_to_string(c->std_output),
2189 prefix, exec_output_to_string(c->std_error));
2195 "%sTTYVHangup: %s\n"
2196 "%sTTYVTDisallocate: %s\n",
2197 prefix, c->tty_path,
2198 prefix, yes_no(c->tty_reset),
2199 prefix, yes_no(c->tty_vhangup),
2200 prefix, yes_no(c->tty_vt_disallocate));
2202 if (c->std_output == EXEC_OUTPUT_SYSLOG ||
2203 c->std_output == EXEC_OUTPUT_KMSG ||
2204 c->std_output == EXEC_OUTPUT_JOURNAL ||
2205 c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
2206 c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
2207 c->std_output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE ||
2208 c->std_error == EXEC_OUTPUT_SYSLOG ||
2209 c->std_error == EXEC_OUTPUT_KMSG ||
2210 c->std_error == EXEC_OUTPUT_JOURNAL ||
2211 c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
2212 c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
2213 c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE) {
2215 _cleanup_free_ char *fac_str = NULL, *lvl_str = NULL;
2217 log_facility_unshifted_to_string_alloc(c->syslog_priority >> 3, &fac_str);
2218 log_level_to_string_alloc(LOG_PRI(c->syslog_priority), &lvl_str);
2221 "%sSyslogFacility: %s\n"
2222 "%sSyslogLevel: %s\n",
2223 prefix, strna(fac_str),
2224 prefix, strna(lvl_str));
2227 if (c->capabilities) {
2228 _cleanup_cap_free_charp_ char *t;
2230 t = cap_to_text(c->capabilities, NULL);
2232 fprintf(f, "%sCapabilities: %s\n", prefix, t);
2236 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
2238 (c->secure_bits & 1<<SECURE_KEEP_CAPS) ? " keep-caps" : "",
2239 (c->secure_bits & 1<<SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
2240 (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
2241 (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
2242 (c->secure_bits & 1<<SECURE_NOROOT) ? " noroot" : "",
2243 (c->secure_bits & 1<<SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
2245 if (c->capability_bounding_set_drop) {
2247 fprintf(f, "%sCapabilityBoundingSet:", prefix);
2249 for (l = 0; l <= cap_last_cap(); l++)
2250 if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) {
2251 _cleanup_cap_free_charp_ char *t;
2255 fprintf(f, " %s", t);
2262 fprintf(f, "%sUser: %s\n", prefix, c->user);
2264 fprintf(f, "%sGroup: %s\n", prefix, c->group);
2266 if (strv_length(c->supplementary_groups) > 0) {
2267 fprintf(f, "%sSupplementaryGroups:", prefix);
2268 strv_fprintf(f, c->supplementary_groups);
2273 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
2275 if (strv_length(c->read_write_dirs) > 0) {
2276 fprintf(f, "%sReadWriteDirs:", prefix);
2277 strv_fprintf(f, c->read_write_dirs);
2281 if (strv_length(c->read_only_dirs) > 0) {
2282 fprintf(f, "%sReadOnlyDirs:", prefix);
2283 strv_fprintf(f, c->read_only_dirs);
2287 if (strv_length(c->inaccessible_dirs) > 0) {
2288 fprintf(f, "%sInaccessibleDirs:", prefix);
2289 strv_fprintf(f, c->inaccessible_dirs);
2295 "%sUtmpIdentifier: %s\n",
2296 prefix, c->utmp_id);
2298 if (c->selinux_context)
2300 "%sSELinuxContext: %s%s\n",
2301 prefix, c->selinux_context_ignore ? "-" : "", c->selinux_context);
2303 if (c->personality != 0xffffffffUL)
2305 "%sPersonality: %s\n",
2306 prefix, strna(personality_to_string(c->personality)));
2308 if (c->syscall_filter) {
2316 "%sSystemCallFilter: ",
2319 if (!c->syscall_whitelist)
2323 SET_FOREACH(id, c->syscall_filter, j) {
2324 _cleanup_free_ char *name = NULL;
2331 name = seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE, PTR_TO_INT(id) - 1);
2332 fputs(strna(name), f);
2339 if (c->syscall_archs) {
2346 "%sSystemCallArchitectures:",
2350 SET_FOREACH(id, c->syscall_archs, j)
2351 fprintf(f, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id) - 1)));
2356 if (c->syscall_errno != 0)
2358 "%sSystemCallErrorNumber: %s\n",
2359 prefix, strna(errno_to_name(c->syscall_errno)));
2361 if (c->apparmor_profile)
2363 "%sAppArmorProfile: %s%s\n",
2364 prefix, c->apparmor_profile_ignore ? "-" : "", c->apparmor_profile);
2367 void exec_status_start(ExecStatus *s, pid_t pid) {
2372 dual_timestamp_get(&s->start_timestamp);
2375 void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
2378 if (s->pid && s->pid != pid)
2382 dual_timestamp_get(&s->exit_timestamp);
2388 if (context->utmp_id)
2389 utmp_put_dead_process(context->utmp_id, pid, code, status);
2391 exec_context_tty_reset(context);
2395 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
2396 char buf[FORMAT_TIMESTAMP_MAX];
2408 "%sPID: "PID_FMT"\n",
2411 if (s->start_timestamp.realtime > 0)
2413 "%sStart Timestamp: %s\n",
2414 prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
2416 if (s->exit_timestamp.realtime > 0)
2418 "%sExit Timestamp: %s\n"
2420 "%sExit Status: %i\n",
2421 prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
2422 prefix, sigchld_code_to_string(s->code),
2426 char *exec_command_line(char **argv) {
2434 STRV_FOREACH(a, argv)
2437 if (!(n = new(char, k)))
2441 STRV_FOREACH(a, argv) {
2448 if (strpbrk(*a, WHITESPACE)) {
2459 /* FIXME: this doesn't really handle arguments that have
2460 * spaces and ticks in them */
2465 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
2466 _cleanup_free_ char *p2 = NULL;
2467 const char *prefix2;
2469 _cleanup_free_ char *cmd = NULL;
2476 p2 = strappend(prefix, "\t");
2477 prefix2 = p2 ? p2 : prefix;
2479 cmd = exec_command_line(c->argv);
2482 "%sCommand Line: %s\n",
2483 prefix, cmd ? cmd : strerror(ENOMEM));
2485 exec_status_dump(&c->exec_status, f, prefix2);
2488 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
2494 LIST_FOREACH(command, c, c)
2495 exec_command_dump(c, f, prefix);
2498 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
2505 /* It's kind of important, that we keep the order here */
2506 LIST_FIND_TAIL(command, *l, end);
2507 LIST_INSERT_AFTER(command, *l, end, e);
2512 int exec_command_set(ExecCommand *c, const char *path, ...) {
2520 l = strv_new_ap(path, ap);
2541 static int exec_runtime_allocate(ExecRuntime **rt) {
2546 *rt = new0(ExecRuntime, 1);
2551 (*rt)->netns_storage_socket[0] = (*rt)->netns_storage_socket[1] = -1;
2556 int exec_runtime_make(ExecRuntime **rt, ExecContext *c, const char *id) {
2566 if (!c->private_network && !c->private_tmp)
2569 r = exec_runtime_allocate(rt);
2573 if (c->private_network && (*rt)->netns_storage_socket[0] < 0) {
2574 if (socketpair(AF_UNIX, SOCK_DGRAM, 0, (*rt)->netns_storage_socket) < 0)
2578 if (c->private_tmp && !(*rt)->tmp_dir) {
2579 r = setup_tmp_dirs(id, &(*rt)->tmp_dir, &(*rt)->var_tmp_dir);
2587 ExecRuntime *exec_runtime_ref(ExecRuntime *r) {
2589 assert(r->n_ref > 0);
2595 ExecRuntime *exec_runtime_unref(ExecRuntime *r) {
2600 assert(r->n_ref > 0);
2603 if (r->n_ref <= 0) {
2605 free(r->var_tmp_dir);
2606 safe_close_pair(r->netns_storage_socket);
2613 int exec_runtime_serialize(ExecRuntime *rt, Unit *u, FILE *f, FDSet *fds) {
2622 unit_serialize_item(u, f, "tmp-dir", rt->tmp_dir);
2624 if (rt->var_tmp_dir)
2625 unit_serialize_item(u, f, "var-tmp-dir", rt->var_tmp_dir);
2627 if (rt->netns_storage_socket[0] >= 0) {
2630 copy = fdset_put_dup(fds, rt->netns_storage_socket[0]);
2634 unit_serialize_item_format(u, f, "netns-socket-0", "%i", copy);
2637 if (rt->netns_storage_socket[1] >= 0) {
2640 copy = fdset_put_dup(fds, rt->netns_storage_socket[1]);
2644 unit_serialize_item_format(u, f, "netns-socket-1", "%i", copy);
2650 int exec_runtime_deserialize_item(ExecRuntime **rt, Unit *u, const char *key, const char *value, FDSet *fds) {
2657 if (streq(key, "tmp-dir")) {
2660 r = exec_runtime_allocate(rt);
2664 copy = strdup(value);
2668 free((*rt)->tmp_dir);
2669 (*rt)->tmp_dir = copy;
2671 } else if (streq(key, "var-tmp-dir")) {
2674 r = exec_runtime_allocate(rt);
2678 copy = strdup(value);
2682 free((*rt)->var_tmp_dir);
2683 (*rt)->var_tmp_dir = copy;
2685 } else if (streq(key, "netns-socket-0")) {
2688 r = exec_runtime_allocate(rt);
2692 if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
2693 log_debug_unit(u->id, "Failed to parse netns socket value %s", value);
2695 safe_close((*rt)->netns_storage_socket[0]);
2696 (*rt)->netns_storage_socket[0] = fdset_remove(fds, fd);
2698 } else if (streq(key, "netns-socket-1")) {
2701 r = exec_runtime_allocate(rt);
2705 if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
2706 log_debug_unit(u->id, "Failed to parse netns socket value %s", value);
2708 safe_close((*rt)->netns_storage_socket[1]);
2709 (*rt)->netns_storage_socket[1] = fdset_remove(fds, fd);
2717 static void *remove_tmpdir_thread(void *p) {
2718 _cleanup_free_ char *path = p;
2720 rm_rf_dangerous(path, false, true, false);
2724 void exec_runtime_destroy(ExecRuntime *rt) {
2730 /* If there are multiple users of this, let's leave the stuff around */
2735 log_debug("Spawning thread to nuke %s", rt->tmp_dir);
2737 r = asynchronous_job(remove_tmpdir_thread, rt->tmp_dir);
2739 log_warning("Failed to nuke %s: %s", rt->tmp_dir, strerror(-r));
2746 if (rt->var_tmp_dir) {
2747 log_debug("Spawning thread to nuke %s", rt->var_tmp_dir);
2749 r = asynchronous_job(remove_tmpdir_thread, rt->var_tmp_dir);
2751 log_warning("Failed to nuke %s: %s", rt->var_tmp_dir, strerror(-r));
2752 free(rt->var_tmp_dir);
2755 rt->var_tmp_dir = NULL;
2758 safe_close_pair(rt->netns_storage_socket);
2761 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
2762 [EXEC_INPUT_NULL] = "null",
2763 [EXEC_INPUT_TTY] = "tty",
2764 [EXEC_INPUT_TTY_FORCE] = "tty-force",
2765 [EXEC_INPUT_TTY_FAIL] = "tty-fail",
2766 [EXEC_INPUT_SOCKET] = "socket"
2769 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
2771 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
2772 [EXEC_OUTPUT_INHERIT] = "inherit",
2773 [EXEC_OUTPUT_NULL] = "null",
2774 [EXEC_OUTPUT_TTY] = "tty",
2775 [EXEC_OUTPUT_SYSLOG] = "syslog",
2776 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
2777 [EXEC_OUTPUT_KMSG] = "kmsg",
2778 [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
2779 [EXEC_OUTPUT_JOURNAL] = "journal",
2780 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE] = "journal+console",
2781 [EXEC_OUTPUT_SOCKET] = "socket"
2784 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);