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_char(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(
1141 const ExecContext *c,
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,
1227 const ExecContext *context,
1228 const ExecParameters *exec_params,
1229 ExecRuntime *runtime,
1232 _cleanup_strv_free_ char **files_env = NULL;
1233 int *fds = NULL; unsigned n_fds = 0;
1242 assert(exec_params);
1243 assert(exec_params->fds || exec_params->n_fds <= 0);
1245 if (context->std_input == EXEC_INPUT_SOCKET ||
1246 context->std_output == EXEC_OUTPUT_SOCKET ||
1247 context->std_error == EXEC_OUTPUT_SOCKET) {
1249 if (exec_params->n_fds != 1)
1252 socket_fd = exec_params->fds[0];
1255 fds = exec_params->fds;
1256 n_fds = exec_params->n_fds;
1259 r = exec_context_load_environment(context, &files_env);
1261 log_struct_unit(LOG_ERR,
1262 exec_params->unit_id,
1263 "MESSAGE=Failed to load environment files: %s", strerror(-r),
1269 argv = exec_params->argv ?: command->argv;
1271 line = exec_command_line(argv);
1275 log_struct_unit(LOG_DEBUG,
1276 exec_params->unit_id,
1277 "EXECUTABLE=%s", command->path,
1278 "MESSAGE=About to execute: %s", line,
1287 _cleanup_strv_free_ char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
1288 const char *username = NULL, *home = NULL, *shell = NULL;
1289 unsigned n_dont_close = 0;
1290 int dont_close[n_fds + 3];
1291 uid_t uid = (uid_t) -1;
1292 gid_t gid = (gid_t) -1;
1297 rename_process_from_path(command->path);
1299 /* We reset exactly these signals, since they are the
1300 * only ones we set to SIG_IGN in the main daemon. All
1301 * others we leave untouched because we set them to
1302 * SIG_DFL or a valid handler initially, both of which
1303 * will be demoted to SIG_DFL. */
1304 default_signals(SIGNALS_CRASH_HANDLER,
1305 SIGNALS_IGNORE, -1);
1307 if (context->ignore_sigpipe)
1308 ignore_signals(SIGPIPE, -1);
1310 err = reset_signal_mask();
1312 r = EXIT_SIGNAL_MASK;
1316 if (exec_params->idle_pipe)
1317 do_idle_pipe_dance(exec_params->idle_pipe);
1319 /* Close sockets very early to make sure we don't
1320 * block init reexecution because it cannot bind its
1325 dont_close[n_dont_close++] = socket_fd;
1327 memcpy(dont_close + n_dont_close, fds, sizeof(int) * n_fds);
1328 n_dont_close += n_fds;
1331 if (runtime->netns_storage_socket[0] >= 0)
1332 dont_close[n_dont_close++] = runtime->netns_storage_socket[0];
1333 if (runtime->netns_storage_socket[1] >= 0)
1334 dont_close[n_dont_close++] = runtime->netns_storage_socket[1];
1337 err = close_all_fds(dont_close, n_dont_close);
1343 if (!context->same_pgrp)
1350 exec_context_tty_reset(context);
1352 if (exec_params->confirm_spawn) {
1355 err = ask_for_confirmation(&response, argv);
1356 if (err == -ETIMEDOUT)
1357 write_confirm_message("Confirmation question timed out, assuming positive response.\n");
1359 write_confirm_message("Couldn't ask confirmation question, assuming positive response: %s\n", strerror(-err));
1360 else if (response == 's') {
1361 write_confirm_message("Skipping execution.\n");
1365 } else if (response == 'n') {
1366 write_confirm_message("Failing execution.\n");
1372 /* If a socket is connected to STDIN/STDOUT/STDERR, we
1373 * must sure to drop O_NONBLOCK */
1375 fd_nonblock(socket_fd, false);
1377 err = setup_input(context, socket_fd, exec_params->apply_tty_stdin);
1383 err = setup_output(context, STDOUT_FILENO, socket_fd, basename(command->path), exec_params->unit_id, exec_params->apply_tty_stdin);
1389 err = setup_output(context, STDERR_FILENO, socket_fd, basename(command->path), exec_params->unit_id, exec_params->apply_tty_stdin);
1395 if (exec_params->cgroup_path) {
1396 err = cg_attach_everywhere(exec_params->cgroup_supported, exec_params->cgroup_path, 0);
1403 if (context->oom_score_adjust_set) {
1406 snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
1409 if (write_string_file("/proc/self/oom_score_adj", t) < 0) {
1411 r = EXIT_OOM_ADJUST;
1416 if (context->nice_set)
1417 if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
1423 if (context->cpu_sched_set) {
1424 struct sched_param param = {
1425 .sched_priority = context->cpu_sched_priority,
1428 r = sched_setscheduler(0,
1429 context->cpu_sched_policy |
1430 (context->cpu_sched_reset_on_fork ?
1431 SCHED_RESET_ON_FORK : 0),
1435 r = EXIT_SETSCHEDULER;
1440 if (context->cpuset)
1441 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
1443 r = EXIT_CPUAFFINITY;
1447 if (context->ioprio_set)
1448 if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
1454 if (context->timer_slack_nsec != NSEC_INFINITY)
1455 if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
1457 r = EXIT_TIMERSLACK;
1461 if (context->personality != 0xffffffffUL)
1462 if (personality(context->personality) < 0) {
1464 r = EXIT_PERSONALITY;
1468 if (context->utmp_id)
1469 utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path);
1471 if (context->user) {
1472 username = context->user;
1473 err = get_user_creds(&username, &uid, &gid, &home, &shell);
1479 if (is_terminal_input(context->std_input)) {
1480 err = chown_terminal(STDIN_FILENO, uid);
1489 if (exec_params->cgroup_path && context->user && context->pam_name) {
1490 err = cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER, exec_params->cgroup_path, 0644, uid, gid);
1497 err = cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER, exec_params->cgroup_path, 0755, uid, gid);
1505 if (!strv_isempty(context->runtime_directory) && exec_params->runtime_prefix) {
1508 STRV_FOREACH(rt, context->runtime_directory) {
1509 _cleanup_free_ char *p;
1511 p = strjoin(exec_params->runtime_prefix, "/", *rt, NULL);
1513 r = EXIT_RUNTIME_DIRECTORY;
1518 err = mkdir_safe(p, context->runtime_directory_mode, uid, gid);
1520 r = EXIT_RUNTIME_DIRECTORY;
1526 if (exec_params->apply_permissions) {
1527 err = enforce_groups(context, username, gid);
1534 umask(context->umask);
1537 if (exec_params->apply_permissions && context->pam_name && username) {
1538 err = setup_pam(context->pam_name, username, uid, context->tty_path, &pam_env, fds, n_fds);
1545 if (context->private_network && runtime && runtime->netns_storage_socket[0] >= 0) {
1546 err = setup_netns(runtime->netns_storage_socket);
1553 if (!strv_isempty(context->read_write_dirs) ||
1554 !strv_isempty(context->read_only_dirs) ||
1555 !strv_isempty(context->inaccessible_dirs) ||
1556 context->mount_flags != 0 ||
1557 (context->private_tmp && runtime && (runtime->tmp_dir || runtime->var_tmp_dir)) ||
1558 context->private_devices ||
1559 context->protect_system != PROTECT_SYSTEM_NO ||
1560 context->protect_home != PROTECT_HOME_NO) {
1562 char *tmp = NULL, *var = NULL;
1564 /* The runtime struct only contains the parent
1565 * of the private /tmp, which is
1566 * non-accessible to world users. Inside of it
1567 * there's a /tmp that is sticky, and that's
1568 * the one we want to use here. */
1570 if (context->private_tmp && runtime) {
1571 if (runtime->tmp_dir)
1572 tmp = strappenda(runtime->tmp_dir, "/tmp");
1573 if (runtime->var_tmp_dir)
1574 var = strappenda(runtime->var_tmp_dir, "/tmp");
1577 err = setup_namespace(
1578 context->read_write_dirs,
1579 context->read_only_dirs,
1580 context->inaccessible_dirs,
1583 context->private_devices,
1584 context->protect_home,
1585 context->protect_system,
1586 context->mount_flags);
1593 if (exec_params->apply_chroot) {
1594 if (context->root_directory)
1595 if (chroot(context->root_directory) < 0) {
1601 if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
1607 _cleanup_free_ char *d = NULL;
1609 if (asprintf(&d, "%s/%s",
1610 context->root_directory ? context->root_directory : "",
1611 context->working_directory ? context->working_directory : "") < 0) {
1624 /* We repeat the fd closing here, to make sure that
1625 * nothing is leaked from the PAM modules. Note that
1626 * we are more aggressive this time since socket_fd
1627 * and the netns fds we don#t need anymore. */
1628 err = close_all_fds(fds, n_fds);
1630 err = shift_fds(fds, n_fds);
1632 err = flags_fds(fds, n_fds, context->non_blocking);
1638 if (exec_params->apply_permissions) {
1640 for (i = 0; i < _RLIMIT_MAX; i++) {
1641 if (!context->rlimit[i])
1644 if (setrlimit_closest(i, context->rlimit[i]) < 0) {
1651 if (context->capability_bounding_set_drop) {
1652 err = capability_bounding_set_drop(context->capability_bounding_set_drop, false);
1654 r = EXIT_CAPABILITIES;
1659 if (context->user) {
1660 err = enforce_user(context, uid);
1667 /* PR_GET_SECUREBITS is not privileged, while
1668 * PR_SET_SECUREBITS is. So to suppress
1669 * potential EPERMs we'll try not to call
1670 * PR_SET_SECUREBITS unless necessary. */
1671 if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
1672 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
1674 r = EXIT_SECUREBITS;
1678 if (context->capabilities)
1679 if (cap_set_proc(context->capabilities) < 0) {
1681 r = EXIT_CAPABILITIES;
1685 if (context->no_new_privileges)
1686 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
1688 r = EXIT_NO_NEW_PRIVILEGES;
1693 if (context->address_families_whitelist ||
1694 !set_isempty(context->address_families)) {
1695 err = apply_address_families(context);
1697 r = EXIT_ADDRESS_FAMILIES;
1702 if (context->syscall_whitelist ||
1703 !set_isempty(context->syscall_filter) ||
1704 !set_isempty(context->syscall_archs)) {
1705 err = apply_seccomp(context);
1714 if (context->selinux_context && use_selinux()) {
1715 err = setexeccon(context->selinux_context);
1716 if (err < 0 && !context->selinux_context_ignore) {
1717 r = EXIT_SELINUX_CONTEXT;
1723 #ifdef HAVE_APPARMOR
1724 if (context->apparmor_profile && use_apparmor()) {
1725 err = aa_change_onexec(context->apparmor_profile);
1726 if (err < 0 && !context->apparmor_profile_ignore) {
1727 r = EXIT_APPARMOR_PROFILE;
1734 err = build_environment(context, n_fds, exec_params->watchdog_usec, home, username, shell, &our_env);
1740 final_env = strv_env_merge(5,
1741 exec_params->environment,
1743 context->environment,
1753 final_argv = replace_env_argv(argv, final_env);
1760 final_env = strv_env_clean(final_env);
1762 if (_unlikely_(log_get_max_level() >= LOG_PRI(LOG_DEBUG))) {
1763 line = exec_command_line(final_argv);
1766 log_struct_unit(LOG_DEBUG,
1767 exec_params->unit_id,
1768 "EXECUTABLE=%s", command->path,
1769 "MESSAGE=Executing: %s", line,
1776 execve(command->path, final_argv, final_env);
1783 log_struct(LOG_ERR, MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED),
1784 "EXECUTABLE=%s", command->path,
1785 "MESSAGE=Failed at step %s spawning %s: %s",
1786 exit_status_to_string(r, EXIT_STATUS_SYSTEMD),
1787 command->path, strerror(-err),
1796 log_struct_unit(LOG_DEBUG,
1797 exec_params->unit_id,
1798 "MESSAGE=Forked %s as "PID_FMT,
1802 /* We add the new process to the cgroup both in the child (so
1803 * that we can be sure that no user code is ever executed
1804 * outside of the cgroup) and in the parent (so that we can be
1805 * sure that when we kill the cgroup the process will be
1807 if (exec_params->cgroup_path)
1808 cg_attach(SYSTEMD_CGROUP_CONTROLLER, exec_params->cgroup_path, pid);
1810 exec_status_start(&command->exec_status, pid);
1816 void exec_context_init(ExecContext *c) {
1820 c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
1821 c->cpu_sched_policy = SCHED_OTHER;
1822 c->syslog_priority = LOG_DAEMON|LOG_INFO;
1823 c->syslog_level_prefix = true;
1824 c->ignore_sigpipe = true;
1825 c->timer_slack_nsec = NSEC_INFINITY;
1826 c->personality = 0xffffffffUL;
1827 c->runtime_directory_mode = 0755;
1830 void exec_context_done(ExecContext *c) {
1835 strv_free(c->environment);
1836 c->environment = NULL;
1838 strv_free(c->environment_files);
1839 c->environment_files = NULL;
1841 for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1843 c->rlimit[l] = NULL;
1846 free(c->working_directory);
1847 c->working_directory = NULL;
1848 free(c->root_directory);
1849 c->root_directory = NULL;
1854 free(c->syslog_identifier);
1855 c->syslog_identifier = NULL;
1863 strv_free(c->supplementary_groups);
1864 c->supplementary_groups = NULL;
1869 if (c->capabilities) {
1870 cap_free(c->capabilities);
1871 c->capabilities = NULL;
1874 strv_free(c->read_only_dirs);
1875 c->read_only_dirs = NULL;
1877 strv_free(c->read_write_dirs);
1878 c->read_write_dirs = NULL;
1880 strv_free(c->inaccessible_dirs);
1881 c->inaccessible_dirs = NULL;
1884 CPU_FREE(c->cpuset);
1889 free(c->selinux_context);
1890 c->selinux_context = NULL;
1892 free(c->apparmor_profile);
1893 c->apparmor_profile = NULL;
1895 set_free(c->syscall_filter);
1896 c->syscall_filter = NULL;
1898 set_free(c->syscall_archs);
1899 c->syscall_archs = NULL;
1901 set_free(c->address_families);
1902 c->address_families = NULL;
1904 strv_free(c->runtime_directory);
1905 c->runtime_directory = NULL;
1908 int exec_context_destroy_runtime_directory(ExecContext *c, const char *runtime_prefix) {
1913 if (!runtime_prefix)
1916 STRV_FOREACH(i, c->runtime_directory) {
1917 _cleanup_free_ char *p;
1919 p = strjoin(runtime_prefix, "/", *i, NULL);
1923 /* We execute this synchronously, since we need to be
1924 * sure this is gone when we start the service
1926 rm_rf_dangerous(p, false, true, false);
1932 void exec_command_done(ExecCommand *c) {
1942 void exec_command_done_array(ExecCommand *c, unsigned n) {
1945 for (i = 0; i < n; i++)
1946 exec_command_done(c+i);
1949 void exec_command_free_list(ExecCommand *c) {
1953 LIST_REMOVE(command, c, i);
1954 exec_command_done(i);
1959 void exec_command_free_array(ExecCommand **c, unsigned n) {
1962 for (i = 0; i < n; i++) {
1963 exec_command_free_list(c[i]);
1968 int exec_context_load_environment(const ExecContext *c, char ***l) {
1969 char **i, **r = NULL;
1974 STRV_FOREACH(i, c->environment_files) {
1977 bool ignore = false;
1979 _cleanup_globfree_ glob_t pglob = {};
1989 if (!path_is_absolute(fn)) {
1997 /* Filename supports globbing, take all matching files */
1999 if (glob(fn, 0, NULL, &pglob) != 0) {
2004 return errno ? -errno : -EINVAL;
2006 count = pglob.gl_pathc;
2014 for (n = 0; n < count; n++) {
2015 k = load_env_file(NULL, pglob.gl_pathv[n], NULL, &p);
2023 /* Log invalid environment variables with filename */
2025 p = strv_env_clean_log(p, pglob.gl_pathv[n]);
2032 m = strv_env_merge(2, r, p);
2048 static bool tty_may_match_dev_console(const char *tty) {
2049 _cleanup_free_ char *active = NULL;
2052 if (startswith(tty, "/dev/"))
2055 /* trivial identity? */
2056 if (streq(tty, "console"))
2059 console = resolve_dev_console(&active);
2060 /* if we could not resolve, assume it may */
2064 /* "tty0" means the active VC, so it may be the same sometimes */
2065 return streq(console, tty) || (streq(console, "tty0") && tty_is_vc(tty));
2068 bool exec_context_may_touch_console(ExecContext *ec) {
2069 return (ec->tty_reset || ec->tty_vhangup || ec->tty_vt_disallocate ||
2070 is_terminal_input(ec->std_input) ||
2071 is_terminal_output(ec->std_output) ||
2072 is_terminal_output(ec->std_error)) &&
2073 tty_may_match_dev_console(tty_path(ec));
2076 static void strv_fprintf(FILE *f, char **l) {
2082 fprintf(f, " %s", *g);
2085 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
2092 prefix = strempty(prefix);
2096 "%sWorkingDirectory: %s\n"
2097 "%sRootDirectory: %s\n"
2098 "%sNonBlocking: %s\n"
2099 "%sPrivateTmp: %s\n"
2100 "%sPrivateNetwork: %s\n"
2101 "%sPrivateDevices: %s\n"
2102 "%sProtectHome: %s\n"
2103 "%sProtectSystem: %s\n"
2104 "%sIgnoreSIGPIPE: %s\n",
2106 prefix, c->working_directory ? c->working_directory : "/",
2107 prefix, c->root_directory ? c->root_directory : "/",
2108 prefix, yes_no(c->non_blocking),
2109 prefix, yes_no(c->private_tmp),
2110 prefix, yes_no(c->private_network),
2111 prefix, yes_no(c->private_devices),
2112 prefix, protect_home_to_string(c->protect_home),
2113 prefix, protect_system_to_string(c->protect_system),
2114 prefix, yes_no(c->ignore_sigpipe));
2116 STRV_FOREACH(e, c->environment)
2117 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
2119 STRV_FOREACH(e, c->environment_files)
2120 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
2127 if (c->oom_score_adjust_set)
2129 "%sOOMScoreAdjust: %i\n",
2130 prefix, c->oom_score_adjust);
2132 for (i = 0; i < RLIM_NLIMITS; i++)
2134 fprintf(f, "%s%s: "RLIM_FMT"\n",
2135 prefix, rlimit_to_string(i), c->rlimit[i]->rlim_max);
2137 if (c->ioprio_set) {
2138 _cleanup_free_ char *class_str = NULL;
2140 ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c->ioprio), &class_str);
2142 "%sIOSchedulingClass: %s\n"
2143 "%sIOPriority: %i\n",
2144 prefix, strna(class_str),
2145 prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
2148 if (c->cpu_sched_set) {
2149 _cleanup_free_ char *policy_str = NULL;
2151 sched_policy_to_string_alloc(c->cpu_sched_policy, &policy_str);
2153 "%sCPUSchedulingPolicy: %s\n"
2154 "%sCPUSchedulingPriority: %i\n"
2155 "%sCPUSchedulingResetOnFork: %s\n",
2156 prefix, strna(policy_str),
2157 prefix, c->cpu_sched_priority,
2158 prefix, yes_no(c->cpu_sched_reset_on_fork));
2162 fprintf(f, "%sCPUAffinity:", prefix);
2163 for (i = 0; i < c->cpuset_ncpus; i++)
2164 if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
2165 fprintf(f, " %u", i);
2169 if (c->timer_slack_nsec != NSEC_INFINITY)
2170 fprintf(f, "%sTimerSlackNSec: "NSEC_FMT "\n", prefix, c->timer_slack_nsec);
2173 "%sStandardInput: %s\n"
2174 "%sStandardOutput: %s\n"
2175 "%sStandardError: %s\n",
2176 prefix, exec_input_to_string(c->std_input),
2177 prefix, exec_output_to_string(c->std_output),
2178 prefix, exec_output_to_string(c->std_error));
2184 "%sTTYVHangup: %s\n"
2185 "%sTTYVTDisallocate: %s\n",
2186 prefix, c->tty_path,
2187 prefix, yes_no(c->tty_reset),
2188 prefix, yes_no(c->tty_vhangup),
2189 prefix, yes_no(c->tty_vt_disallocate));
2191 if (c->std_output == EXEC_OUTPUT_SYSLOG ||
2192 c->std_output == EXEC_OUTPUT_KMSG ||
2193 c->std_output == EXEC_OUTPUT_JOURNAL ||
2194 c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
2195 c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
2196 c->std_output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE ||
2197 c->std_error == EXEC_OUTPUT_SYSLOG ||
2198 c->std_error == EXEC_OUTPUT_KMSG ||
2199 c->std_error == EXEC_OUTPUT_JOURNAL ||
2200 c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
2201 c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
2202 c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE) {
2204 _cleanup_free_ char *fac_str = NULL, *lvl_str = NULL;
2206 log_facility_unshifted_to_string_alloc(c->syslog_priority >> 3, &fac_str);
2207 log_level_to_string_alloc(LOG_PRI(c->syslog_priority), &lvl_str);
2210 "%sSyslogFacility: %s\n"
2211 "%sSyslogLevel: %s\n",
2212 prefix, strna(fac_str),
2213 prefix, strna(lvl_str));
2216 if (c->capabilities) {
2217 _cleanup_cap_free_charp_ char *t;
2219 t = cap_to_text(c->capabilities, NULL);
2221 fprintf(f, "%sCapabilities: %s\n", prefix, t);
2225 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
2227 (c->secure_bits & 1<<SECURE_KEEP_CAPS) ? " keep-caps" : "",
2228 (c->secure_bits & 1<<SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
2229 (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
2230 (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
2231 (c->secure_bits & 1<<SECURE_NOROOT) ? " noroot" : "",
2232 (c->secure_bits & 1<<SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
2234 if (c->capability_bounding_set_drop) {
2236 fprintf(f, "%sCapabilityBoundingSet:", prefix);
2238 for (l = 0; l <= cap_last_cap(); l++)
2239 if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) {
2240 _cleanup_cap_free_charp_ char *t;
2244 fprintf(f, " %s", t);
2251 fprintf(f, "%sUser: %s\n", prefix, c->user);
2253 fprintf(f, "%sGroup: %s\n", prefix, c->group);
2255 if (strv_length(c->supplementary_groups) > 0) {
2256 fprintf(f, "%sSupplementaryGroups:", prefix);
2257 strv_fprintf(f, c->supplementary_groups);
2262 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
2264 if (strv_length(c->read_write_dirs) > 0) {
2265 fprintf(f, "%sReadWriteDirs:", prefix);
2266 strv_fprintf(f, c->read_write_dirs);
2270 if (strv_length(c->read_only_dirs) > 0) {
2271 fprintf(f, "%sReadOnlyDirs:", prefix);
2272 strv_fprintf(f, c->read_only_dirs);
2276 if (strv_length(c->inaccessible_dirs) > 0) {
2277 fprintf(f, "%sInaccessibleDirs:", prefix);
2278 strv_fprintf(f, c->inaccessible_dirs);
2284 "%sUtmpIdentifier: %s\n",
2285 prefix, c->utmp_id);
2287 if (c->selinux_context)
2289 "%sSELinuxContext: %s%s\n",
2290 prefix, c->selinux_context_ignore ? "-" : "", c->selinux_context);
2292 if (c->personality != 0xffffffffUL)
2294 "%sPersonality: %s\n",
2295 prefix, strna(personality_to_string(c->personality)));
2297 if (c->syscall_filter) {
2305 "%sSystemCallFilter: ",
2308 if (!c->syscall_whitelist)
2312 SET_FOREACH(id, c->syscall_filter, j) {
2313 _cleanup_free_ char *name = NULL;
2320 name = seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE, PTR_TO_INT(id) - 1);
2321 fputs(strna(name), f);
2328 if (c->syscall_archs) {
2335 "%sSystemCallArchitectures:",
2339 SET_FOREACH(id, c->syscall_archs, j)
2340 fprintf(f, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id) - 1)));
2345 if (c->syscall_errno != 0)
2347 "%sSystemCallErrorNumber: %s\n",
2348 prefix, strna(errno_to_name(c->syscall_errno)));
2350 if (c->apparmor_profile)
2352 "%sAppArmorProfile: %s%s\n",
2353 prefix, c->apparmor_profile_ignore ? "-" : "", c->apparmor_profile);
2356 void exec_status_start(ExecStatus *s, pid_t pid) {
2361 dual_timestamp_get(&s->start_timestamp);
2364 void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
2367 if (s->pid && s->pid != pid)
2371 dual_timestamp_get(&s->exit_timestamp);
2377 if (context->utmp_id)
2378 utmp_put_dead_process(context->utmp_id, pid, code, status);
2380 exec_context_tty_reset(context);
2384 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
2385 char buf[FORMAT_TIMESTAMP_MAX];
2393 prefix = strempty(prefix);
2396 "%sPID: "PID_FMT"\n",
2399 if (s->start_timestamp.realtime > 0)
2401 "%sStart Timestamp: %s\n",
2402 prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
2404 if (s->exit_timestamp.realtime > 0)
2406 "%sExit Timestamp: %s\n"
2408 "%sExit Status: %i\n",
2409 prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
2410 prefix, sigchld_code_to_string(s->code),
2414 char *exec_command_line(char **argv) {
2422 STRV_FOREACH(a, argv)
2425 if (!(n = new(char, k)))
2429 STRV_FOREACH(a, argv) {
2436 if (strpbrk(*a, WHITESPACE)) {
2447 /* FIXME: this doesn't really handle arguments that have
2448 * spaces and ticks in them */
2453 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
2454 _cleanup_free_ char *cmd = NULL;
2455 const char *prefix2;
2460 prefix = strempty(prefix);
2461 prefix2 = strappenda(prefix, "\t");
2463 cmd = exec_command_line(c->argv);
2465 "%sCommand Line: %s\n",
2466 prefix, cmd ? cmd : strerror(ENOMEM));
2468 exec_status_dump(&c->exec_status, f, prefix2);
2471 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
2474 prefix = strempty(prefix);
2476 LIST_FOREACH(command, c, c)
2477 exec_command_dump(c, f, prefix);
2480 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
2487 /* It's kind of important, that we keep the order here */
2488 LIST_FIND_TAIL(command, *l, end);
2489 LIST_INSERT_AFTER(command, *l, end, e);
2494 int exec_command_set(ExecCommand *c, const char *path, ...) {
2502 l = strv_new_ap(path, ap);
2523 static int exec_runtime_allocate(ExecRuntime **rt) {
2528 *rt = new0(ExecRuntime, 1);
2533 (*rt)->netns_storage_socket[0] = (*rt)->netns_storage_socket[1] = -1;
2538 int exec_runtime_make(ExecRuntime **rt, ExecContext *c, const char *id) {
2548 if (!c->private_network && !c->private_tmp)
2551 r = exec_runtime_allocate(rt);
2555 if (c->private_network && (*rt)->netns_storage_socket[0] < 0) {
2556 if (socketpair(AF_UNIX, SOCK_DGRAM, 0, (*rt)->netns_storage_socket) < 0)
2560 if (c->private_tmp && !(*rt)->tmp_dir) {
2561 r = setup_tmp_dirs(id, &(*rt)->tmp_dir, &(*rt)->var_tmp_dir);
2569 ExecRuntime *exec_runtime_ref(ExecRuntime *r) {
2571 assert(r->n_ref > 0);
2577 ExecRuntime *exec_runtime_unref(ExecRuntime *r) {
2582 assert(r->n_ref > 0);
2585 if (r->n_ref <= 0) {
2587 free(r->var_tmp_dir);
2588 safe_close_pair(r->netns_storage_socket);
2595 int exec_runtime_serialize(ExecRuntime *rt, Unit *u, FILE *f, FDSet *fds) {
2604 unit_serialize_item(u, f, "tmp-dir", rt->tmp_dir);
2606 if (rt->var_tmp_dir)
2607 unit_serialize_item(u, f, "var-tmp-dir", rt->var_tmp_dir);
2609 if (rt->netns_storage_socket[0] >= 0) {
2612 copy = fdset_put_dup(fds, rt->netns_storage_socket[0]);
2616 unit_serialize_item_format(u, f, "netns-socket-0", "%i", copy);
2619 if (rt->netns_storage_socket[1] >= 0) {
2622 copy = fdset_put_dup(fds, rt->netns_storage_socket[1]);
2626 unit_serialize_item_format(u, f, "netns-socket-1", "%i", copy);
2632 int exec_runtime_deserialize_item(ExecRuntime **rt, Unit *u, const char *key, const char *value, FDSet *fds) {
2639 if (streq(key, "tmp-dir")) {
2642 r = exec_runtime_allocate(rt);
2646 copy = strdup(value);
2650 free((*rt)->tmp_dir);
2651 (*rt)->tmp_dir = copy;
2653 } else if (streq(key, "var-tmp-dir")) {
2656 r = exec_runtime_allocate(rt);
2660 copy = strdup(value);
2664 free((*rt)->var_tmp_dir);
2665 (*rt)->var_tmp_dir = copy;
2667 } else if (streq(key, "netns-socket-0")) {
2670 r = exec_runtime_allocate(rt);
2674 if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
2675 log_debug_unit(u->id, "Failed to parse netns socket value %s", value);
2677 safe_close((*rt)->netns_storage_socket[0]);
2678 (*rt)->netns_storage_socket[0] = fdset_remove(fds, fd);
2680 } else if (streq(key, "netns-socket-1")) {
2683 r = exec_runtime_allocate(rt);
2687 if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
2688 log_debug_unit(u->id, "Failed to parse netns socket value %s", value);
2690 safe_close((*rt)->netns_storage_socket[1]);
2691 (*rt)->netns_storage_socket[1] = fdset_remove(fds, fd);
2699 static void *remove_tmpdir_thread(void *p) {
2700 _cleanup_free_ char *path = p;
2702 rm_rf_dangerous(path, false, true, false);
2706 void exec_runtime_destroy(ExecRuntime *rt) {
2712 /* If there are multiple users of this, let's leave the stuff around */
2717 log_debug("Spawning thread to nuke %s", rt->tmp_dir);
2719 r = asynchronous_job(remove_tmpdir_thread, rt->tmp_dir);
2721 log_warning("Failed to nuke %s: %s", rt->tmp_dir, strerror(-r));
2728 if (rt->var_tmp_dir) {
2729 log_debug("Spawning thread to nuke %s", rt->var_tmp_dir);
2731 r = asynchronous_job(remove_tmpdir_thread, rt->var_tmp_dir);
2733 log_warning("Failed to nuke %s: %s", rt->var_tmp_dir, strerror(-r));
2734 free(rt->var_tmp_dir);
2737 rt->var_tmp_dir = NULL;
2740 safe_close_pair(rt->netns_storage_socket);
2743 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
2744 [EXEC_INPUT_NULL] = "null",
2745 [EXEC_INPUT_TTY] = "tty",
2746 [EXEC_INPUT_TTY_FORCE] = "tty-force",
2747 [EXEC_INPUT_TTY_FAIL] = "tty-fail",
2748 [EXEC_INPUT_SOCKET] = "socket"
2751 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
2753 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
2754 [EXEC_OUTPUT_INHERIT] = "inherit",
2755 [EXEC_OUTPUT_NULL] = "null",
2756 [EXEC_OUTPUT_TTY] = "tty",
2757 [EXEC_OUTPUT_SYSLOG] = "syslog",
2758 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
2759 [EXEC_OUTPUT_KMSG] = "kmsg",
2760 [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
2761 [EXEC_OUTPUT_JOURNAL] = "journal",
2762 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE] = "journal+console",
2763 [EXEC_OUTPUT_SOCKET] = "socket"
2766 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);