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"
73 #include "exit-status.h"
75 #include "utmp-wtmp.h"
77 #include "path-util.h"
82 #include "selinux-util.h"
83 #include "errno-list.h"
86 #include "apparmor-util.h"
89 #include "seccomp-util.h"
92 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
93 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
95 /* This assumes there is a 'tty' group */
98 #define SNDBUF_SIZE (8*1024*1024)
100 static int shift_fds(int fds[], unsigned n_fds) {
101 int start, restart_from;
106 /* Modifies the fds array! (sorts it) */
116 for (i = start; i < (int) n_fds; i++) {
119 /* Already at right index? */
123 if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
126 close_nointr_nofail(fds[i]);
129 /* Hmm, the fd we wanted isn't free? Then
130 * let's remember that and try again from here*/
131 if (nfd != i+3 && restart_from < 0)
135 if (restart_from < 0)
138 start = restart_from;
144 static int flags_fds(const int fds[], unsigned n_fds, bool nonblock) {
153 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
155 for (i = 0; i < n_fds; i++) {
157 if ((r = fd_nonblock(fds[i], nonblock)) < 0)
160 /* We unconditionally drop FD_CLOEXEC from the fds,
161 * since after all we want to pass these fds to our
164 if ((r = fd_cloexec(fds[i], false)) < 0)
171 _pure_ static const char *tty_path(const ExecContext *context) {
174 if (context->tty_path)
175 return context->tty_path;
177 return "/dev/console";
180 static void exec_context_tty_reset(const ExecContext *context) {
183 if (context->tty_vhangup)
184 terminal_vhangup(tty_path(context));
186 if (context->tty_reset)
187 reset_terminal(tty_path(context));
189 if (context->tty_vt_disallocate && context->tty_path)
190 vt_disallocate(context->tty_path);
193 static bool is_terminal_output(ExecOutput o) {
195 o == EXEC_OUTPUT_TTY ||
196 o == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
197 o == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
198 o == EXEC_OUTPUT_JOURNAL_AND_CONSOLE;
201 static int open_null_as(int flags, int nfd) {
206 fd = open("/dev/null", flags|O_NOCTTY);
211 r = dup2(fd, nfd) < 0 ? -errno : nfd;
212 close_nointr_nofail(fd);
219 static int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, const char *unit_id, int nfd) {
221 union sockaddr_union sa = {
222 .un.sun_family = AF_UNIX,
223 .un.sun_path = "/run/systemd/journal/stdout",
227 assert(output < _EXEC_OUTPUT_MAX);
231 fd = socket(AF_UNIX, SOCK_STREAM, 0);
235 r = connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path));
237 close_nointr_nofail(fd);
241 if (shutdown(fd, SHUT_RD) < 0) {
242 close_nointr_nofail(fd);
246 fd_inc_sndbuf(fd, SNDBUF_SIZE);
256 context->syslog_identifier ? context->syslog_identifier : ident,
258 context->syslog_priority,
259 !!context->syslog_level_prefix,
260 output == EXEC_OUTPUT_SYSLOG || output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE,
261 output == EXEC_OUTPUT_KMSG || output == EXEC_OUTPUT_KMSG_AND_CONSOLE,
262 is_terminal_output(output));
265 r = dup2(fd, nfd) < 0 ? -errno : nfd;
266 close_nointr_nofail(fd);
272 static int open_terminal_as(const char *path, mode_t mode, int nfd) {
278 if ((fd = open_terminal(path, mode | O_NOCTTY)) < 0)
282 r = dup2(fd, nfd) < 0 ? -errno : nfd;
283 close_nointr_nofail(fd);
290 static bool is_terminal_input(ExecInput i) {
292 i == EXEC_INPUT_TTY ||
293 i == EXEC_INPUT_TTY_FORCE ||
294 i == EXEC_INPUT_TTY_FAIL;
297 static int fixup_input(ExecInput std_input, int socket_fd, bool apply_tty_stdin) {
299 if (is_terminal_input(std_input) && !apply_tty_stdin)
300 return EXEC_INPUT_NULL;
302 if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0)
303 return EXEC_INPUT_NULL;
308 static int fixup_output(ExecOutput std_output, int socket_fd) {
310 if (std_output == EXEC_OUTPUT_SOCKET && socket_fd < 0)
311 return EXEC_OUTPUT_INHERIT;
316 static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty_stdin) {
321 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
325 case EXEC_INPUT_NULL:
326 return open_null_as(O_RDONLY, STDIN_FILENO);
329 case EXEC_INPUT_TTY_FORCE:
330 case EXEC_INPUT_TTY_FAIL: {
333 fd = acquire_terminal(tty_path(context),
334 i == EXEC_INPUT_TTY_FAIL,
335 i == EXEC_INPUT_TTY_FORCE,
341 if (fd != STDIN_FILENO) {
342 r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
343 close_nointr_nofail(fd);
350 case EXEC_INPUT_SOCKET:
351 return dup2(socket_fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
354 assert_not_reached("Unknown input type");
358 static int setup_output(const ExecContext *context, int fileno, int socket_fd, const char *ident, const char *unit_id, bool apply_tty_stdin) {
366 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
367 o = fixup_output(context->std_output, socket_fd);
369 if (fileno == STDERR_FILENO) {
371 e = fixup_output(context->std_error, socket_fd);
373 /* This expects the input and output are already set up */
375 /* Don't change the stderr file descriptor if we inherit all
376 * the way and are not on a tty */
377 if (e == EXEC_OUTPUT_INHERIT &&
378 o == EXEC_OUTPUT_INHERIT &&
379 i == EXEC_INPUT_NULL &&
380 !is_terminal_input(context->std_input) &&
384 /* Duplicate from stdout if possible */
385 if (e == o || e == EXEC_OUTPUT_INHERIT)
386 return dup2(STDOUT_FILENO, fileno) < 0 ? -errno : fileno;
390 } else if (o == EXEC_OUTPUT_INHERIT) {
391 /* If input got downgraded, inherit the original value */
392 if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input))
393 return open_terminal_as(tty_path(context), O_WRONLY, fileno);
395 /* If the input is connected to anything that's not a /dev/null, inherit that... */
396 if (i != EXEC_INPUT_NULL)
397 return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
399 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
403 /* We need to open /dev/null here anew, to get the right access mode. */
404 return open_null_as(O_WRONLY, fileno);
409 case EXEC_OUTPUT_NULL:
410 return open_null_as(O_WRONLY, fileno);
412 case EXEC_OUTPUT_TTY:
413 if (is_terminal_input(i))
414 return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
416 /* We don't reset the terminal if this is just about output */
417 return open_terminal_as(tty_path(context), O_WRONLY, fileno);
419 case EXEC_OUTPUT_SYSLOG:
420 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
421 case EXEC_OUTPUT_KMSG:
422 case EXEC_OUTPUT_KMSG_AND_CONSOLE:
423 case EXEC_OUTPUT_JOURNAL:
424 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE:
425 r = connect_logger_as(context, o, ident, unit_id, fileno);
427 log_struct_unit(LOG_CRIT, unit_id,
428 "MESSAGE=Failed to connect std%s of %s to the journal socket: %s",
429 fileno == STDOUT_FILENO ? "out" : "err",
430 unit_id, strerror(-r),
433 r = open_null_as(O_WRONLY, fileno);
437 case EXEC_OUTPUT_SOCKET:
438 assert(socket_fd >= 0);
439 return dup2(socket_fd, fileno) < 0 ? -errno : fileno;
442 assert_not_reached("Unknown error type");
446 static int chown_terminal(int fd, uid_t uid) {
451 /* This might fail. What matters are the results. */
452 (void) fchown(fd, uid, -1);
453 (void) fchmod(fd, TTY_MODE);
455 if (fstat(fd, &st) < 0)
458 if (st.st_uid != uid || (st.st_mode & 0777) != TTY_MODE)
464 static int setup_confirm_stdio(int *_saved_stdin,
465 int *_saved_stdout) {
466 int fd = -1, saved_stdin, saved_stdout = -1, r;
468 assert(_saved_stdin);
469 assert(_saved_stdout);
471 saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3);
475 saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3);
476 if (saved_stdout < 0) {
481 fd = acquire_terminal(
486 DEFAULT_CONFIRM_USEC);
492 r = chown_terminal(fd, getuid());
496 if (dup2(fd, STDIN_FILENO) < 0) {
501 if (dup2(fd, STDOUT_FILENO) < 0) {
507 close_nointr_nofail(fd);
509 *_saved_stdin = saved_stdin;
510 *_saved_stdout = saved_stdout;
515 if (saved_stdout >= 0)
516 close_nointr_nofail(saved_stdout);
518 if (saved_stdin >= 0)
519 close_nointr_nofail(saved_stdin);
522 close_nointr_nofail(fd);
527 _printf_(1, 2) static int write_confirm_message(const char *format, ...) {
533 fd = open_terminal("/dev/console", O_WRONLY|O_NOCTTY|O_CLOEXEC);
537 va_start(ap, format);
538 vdprintf(fd, format, ap);
541 close_nointr_nofail(fd);
546 static int restore_confirm_stdio(int *saved_stdin,
552 assert(saved_stdout);
556 if (*saved_stdin >= 0)
557 if (dup2(*saved_stdin, STDIN_FILENO) < 0)
560 if (*saved_stdout >= 0)
561 if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
564 if (*saved_stdin >= 0)
565 close_nointr_nofail(*saved_stdin);
567 if (*saved_stdout >= 0)
568 close_nointr_nofail(*saved_stdout);
573 static int ask_for_confirmation(char *response, char **argv) {
574 int saved_stdout = -1, saved_stdin = -1, r;
577 r = setup_confirm_stdio(&saved_stdin, &saved_stdout);
581 line = exec_command_line(argv);
585 r = ask(response, "yns", "Execute %s? [Yes, No, Skip] ", line);
588 restore_confirm_stdio(&saved_stdin, &saved_stdout);
593 static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
594 bool keep_groups = false;
599 /* Lookup and set GID and supplementary group list. Here too
600 * we avoid NSS lookups for gid=0. */
602 if (context->group || username) {
604 if (context->group) {
605 const char *g = context->group;
607 if ((r = get_group_creds(&g, &gid)) < 0)
611 /* First step, initialize groups from /etc/groups */
612 if (username && gid != 0) {
613 if (initgroups(username, gid) < 0)
619 /* Second step, set our gids */
620 if (setresgid(gid, gid, gid) < 0)
624 if (context->supplementary_groups) {
629 /* Final step, initialize any manually set supplementary groups */
630 assert_se((ngroups_max = (int) sysconf(_SC_NGROUPS_MAX)) > 0);
632 if (!(gids = new(gid_t, ngroups_max)))
636 if ((k = getgroups(ngroups_max, gids)) < 0) {
643 STRV_FOREACH(i, context->supplementary_groups) {
646 if (k >= ngroups_max) {
652 r = get_group_creds(&g, gids+k);
661 if (setgroups(k, gids) < 0) {
672 static int enforce_user(const ExecContext *context, uid_t uid) {
675 /* Sets (but doesn't lookup) the uid and make sure we keep the
676 * capabilities while doing so. */
678 if (context->capabilities) {
679 _cleanup_cap_free_ cap_t d = NULL;
680 static const cap_value_t bits[] = {
681 CAP_SETUID, /* Necessary so that we can run setresuid() below */
682 CAP_SETPCAP /* Necessary so that we can set PR_SET_SECUREBITS later on */
685 /* First step: If we need to keep capabilities but
686 * drop privileges we need to make sure we keep our
687 * caps, while we drop privileges. */
689 int sb = context->secure_bits | 1<<SECURE_KEEP_CAPS;
691 if (prctl(PR_GET_SECUREBITS) != sb)
692 if (prctl(PR_SET_SECUREBITS, sb) < 0)
696 /* Second step: set the capabilities. This will reduce
697 * the capabilities to the minimum we need. */
699 d = cap_dup(context->capabilities);
703 if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
704 cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0)
707 if (cap_set_proc(d) < 0)
711 /* Third step: actually set the uids */
712 if (setresuid(uid, uid, uid) < 0)
715 /* At this point we should have all necessary capabilities but
716 are otherwise a normal user. However, the caps might got
717 corrupted due to the setresuid() so we need clean them up
718 later. This is done outside of this call. */
725 static int null_conv(
727 const struct pam_message **msg,
728 struct pam_response **resp,
731 /* We don't support conversations */
736 static int setup_pam(
742 int fds[], unsigned n_fds) {
744 static const struct pam_conv conv = {
749 pam_handle_t *handle = NULL;
751 int pam_code = PAM_SUCCESS;
754 bool close_session = false;
755 pid_t pam_pid = 0, parent_pid;
762 /* We set up PAM in the parent process, then fork. The child
763 * will then stay around until killed via PR_GET_PDEATHSIG or
764 * systemd via the cgroup logic. It will then remove the PAM
765 * session again. The parent process will exec() the actual
766 * daemon. We do things this way to ensure that the main PID
767 * of the daemon is the one we initially fork()ed. */
769 if (log_get_max_level() < LOG_PRI(LOG_DEBUG))
772 pam_code = pam_start(name, user, &conv, &handle);
773 if (pam_code != PAM_SUCCESS) {
779 pam_code = pam_set_item(handle, PAM_TTY, tty);
780 if (pam_code != PAM_SUCCESS)
784 pam_code = pam_acct_mgmt(handle, flags);
785 if (pam_code != PAM_SUCCESS)
788 pam_code = pam_open_session(handle, flags);
789 if (pam_code != PAM_SUCCESS)
792 close_session = true;
794 e = pam_getenvlist(handle);
796 pam_code = PAM_BUF_ERR;
800 /* Block SIGTERM, so that we know that it won't get lost in
802 if (sigemptyset(&ss) < 0 ||
803 sigaddset(&ss, SIGTERM) < 0 ||
804 sigprocmask(SIG_BLOCK, &ss, &old_ss) < 0)
807 parent_pid = getpid();
817 /* The child's job is to reset the PAM session on
820 /* This string must fit in 10 chars (i.e. the length
821 * of "/sbin/init"), to look pretty in /bin/ps */
822 rename_process("(sd-pam)");
824 /* Make sure we don't keep open the passed fds in this
825 child. We assume that otherwise only those fds are
826 open here that have been opened by PAM. */
827 close_many(fds, n_fds);
829 /* Drop privileges - we don't need any to pam_close_session
830 * and this will make PR_SET_PDEATHSIG work in most cases.
831 * If this fails, ignore the error - but expect sd-pam threads
832 * to fail to exit normally */
833 if (setresuid(uid, uid, uid) < 0)
834 log_error("Error: Failed to setresuid() in sd-pam: %s", strerror(-r));
836 /* Wait until our parent died. This will only work if
837 * the above setresuid() succeeds, otherwise the kernel
838 * will not allow unprivileged parents kill their privileged
839 * children this way. We rely on the control groups kill logic
840 * to do the rest for us. */
841 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
844 /* Check if our parent process might already have
846 if (getppid() == parent_pid) {
848 if (sigwait(&ss, &sig) < 0) {
855 assert(sig == SIGTERM);
860 /* If our parent died we'll end the session */
861 if (getppid() != parent_pid) {
862 pam_code = pam_close_session(handle, flags);
863 if (pam_code != PAM_SUCCESS)
870 pam_end(handle, pam_code | flags);
874 /* If the child was forked off successfully it will do all the
875 * cleanups, so forget about the handle here. */
878 /* Unblock SIGTERM again in the parent */
879 if (sigprocmask(SIG_SETMASK, &old_ss, NULL) < 0)
882 /* We close the log explicitly here, since the PAM modules
883 * might have opened it, but we don't want this fd around. */
892 if (pam_code != PAM_SUCCESS) {
893 log_error("PAM failed: %s", pam_strerror(handle, pam_code));
894 err = -EPERM; /* PAM errors do not map to errno */
896 log_error("PAM failed: %m");
902 pam_code = pam_close_session(handle, flags);
904 pam_end(handle, pam_code | flags);
912 kill(pam_pid, SIGTERM);
913 kill(pam_pid, SIGCONT);
920 static void rename_process_from_path(const char *path) {
921 char process_name[11];
925 /* This resulting string must fit in 10 chars (i.e. the length
926 * of "/sbin/init") to look pretty in /bin/ps */
930 rename_process("(...)");
936 /* The end of the process name is usually more
937 * interesting, since the first bit might just be
943 process_name[0] = '(';
944 memcpy(process_name+1, p, l);
945 process_name[1+l] = ')';
946 process_name[1+l+1] = 0;
948 rename_process(process_name);
953 static int apply_seccomp(ExecContext *c) {
954 uint32_t negative_action, action;
955 scmp_filter_ctx *seccomp;
962 negative_action = c->syscall_errno == 0 ? SCMP_ACT_KILL : SCMP_ACT_ERRNO(c->syscall_errno);
964 seccomp = seccomp_init(c->syscall_whitelist ? negative_action : SCMP_ACT_ALLOW);
968 if (c->syscall_archs) {
970 SET_FOREACH(id, c->syscall_archs, i) {
971 r = seccomp_arch_add(seccomp, PTR_TO_UINT32(id) - 1);
979 r = seccomp_add_secondary_archs(seccomp);
984 action = c->syscall_whitelist ? SCMP_ACT_ALLOW : negative_action;
985 SET_FOREACH(id, c->syscall_filter, i) {
986 r = seccomp_rule_add(seccomp, action, PTR_TO_INT(id) - 1, 0);
991 r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0);
995 r = seccomp_load(seccomp);
998 seccomp_release(seccomp);
1002 static int apply_address_families(ExecContext *c) {
1003 scmp_filter_ctx *seccomp;
1009 seccomp = seccomp_init(SCMP_ACT_ALLOW);
1013 r = seccomp_add_secondary_archs(seccomp);
1017 if (c->address_families_whitelist) {
1018 int af, first = 0, last = 0;
1021 /* If this is a whitelist, we first block the address
1022 * families that are out of range and then everything
1023 * that is not in the set. First, we find the lowest
1024 * and highest address family in the set. */
1026 SET_FOREACH(afp, c->address_families, i) {
1027 af = PTR_TO_INT(afp);
1029 if (af <= 0 || af >= af_max())
1032 if (first == 0 || af < first)
1035 if (last == 0 || af > last)
1039 assert((first == 0) == (last == 0));
1043 /* No entries in the valid range, block everything */
1044 r = seccomp_rule_add(
1046 SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1054 /* Block everything below the first entry */
1055 r = seccomp_rule_add(
1057 SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1060 SCMP_A0(SCMP_CMP_LT, first));
1064 /* Block everything above the last entry */
1065 r = seccomp_rule_add(
1067 SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1070 SCMP_A0(SCMP_CMP_GT, last));
1074 /* Block everything between the first and last
1076 for (af = 1; af < af_max(); af++) {
1078 if (set_contains(c->address_families, INT_TO_PTR(af)))
1081 r = seccomp_rule_add(
1083 SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1086 SCMP_A0(SCMP_CMP_EQ, af));
1095 /* If this is a blacklist, then generate one rule for
1096 * each address family that are then combined in OR
1099 SET_FOREACH(af, c->address_families, i) {
1101 r = seccomp_rule_add(
1103 SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1106 SCMP_A0(SCMP_CMP_EQ, PTR_TO_INT(af)));
1112 r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0);
1116 r = seccomp_load(seccomp);
1119 seccomp_release(seccomp);
1125 static void do_idle_pipe_dance(int idle_pipe[4]) {
1128 if (idle_pipe[1] >= 0)
1129 close_nointr_nofail(idle_pipe[1]);
1130 if (idle_pipe[2] >= 0)
1131 close_nointr_nofail(idle_pipe[2]);
1133 if (idle_pipe[0] >= 0) {
1136 r = fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT_USEC);
1138 if (idle_pipe[3] >= 0 && r == 0 /* timeout */) {
1139 /* Signal systemd that we are bored and want to continue. */
1140 write(idle_pipe[3], "x", 1);
1142 /* Wait for systemd to react to the signal above. */
1143 fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT2_USEC);
1146 close_nointr_nofail(idle_pipe[0]);
1150 if (idle_pipe[3] >= 0)
1151 close_nointr_nofail(idle_pipe[3]);
1154 static int build_environment(
1157 usec_t watchdog_usec,
1159 const char *username,
1163 _cleanup_strv_free_ char **our_env = NULL;
1170 our_env = new0(char*, 10);
1175 if (asprintf(&x, "LISTEN_PID="PID_FMT, getpid()) < 0)
1177 our_env[n_env++] = x;
1179 if (asprintf(&x, "LISTEN_FDS=%u", n_fds) < 0)
1181 our_env[n_env++] = x;
1184 if (watchdog_usec > 0) {
1185 if (asprintf(&x, "WATCHDOG_PID="PID_FMT, getpid()) < 0)
1187 our_env[n_env++] = x;
1189 if (asprintf(&x, "WATCHDOG_USEC=%llu", (unsigned long long) watchdog_usec) < 0)
1191 our_env[n_env++] = x;
1195 x = strappend("HOME=", home);
1198 our_env[n_env++] = x;
1202 x = strappend("LOGNAME=", username);
1205 our_env[n_env++] = x;
1207 x = strappend("USER=", username);
1210 our_env[n_env++] = x;
1214 x = strappend("SHELL=", shell);
1217 our_env[n_env++] = x;
1220 if (is_terminal_input(c->std_input) ||
1221 c->std_output == EXEC_OUTPUT_TTY ||
1222 c->std_error == EXEC_OUTPUT_TTY ||
1225 x = strdup(default_term_for_tty(tty_path(c)));
1228 our_env[n_env++] = x;
1231 our_env[n_env++] = NULL;
1232 assert(n_env <= 10);
1240 int exec_spawn(ExecCommand *command,
1242 ExecContext *context,
1243 int fds[], unsigned n_fds,
1245 bool apply_permissions,
1247 bool apply_tty_stdin,
1249 CGroupControllerMask cgroup_supported,
1250 const char *cgroup_path,
1251 const char *runtime_prefix,
1252 const char *unit_id,
1253 usec_t watchdog_usec,
1255 ExecRuntime *runtime,
1258 _cleanup_strv_free_ char **files_env = NULL;
1267 assert(fds || n_fds <= 0);
1269 if (context->std_input == EXEC_INPUT_SOCKET ||
1270 context->std_output == EXEC_OUTPUT_SOCKET ||
1271 context->std_error == EXEC_OUTPUT_SOCKET) {
1283 r = exec_context_load_environment(context, &files_env);
1285 log_struct_unit(LOG_ERR,
1287 "MESSAGE=Failed to load environment files: %s", strerror(-r),
1294 argv = command->argv;
1296 line = exec_command_line(argv);
1300 log_struct_unit(LOG_DEBUG,
1302 "EXECUTABLE=%s", command->path,
1303 "MESSAGE=About to execute: %s", line,
1312 _cleanup_strv_free_ char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
1313 const char *username = NULL, *home = NULL, *shell = NULL;
1314 unsigned n_dont_close = 0;
1315 int dont_close[n_fds + 3];
1316 uid_t uid = (uid_t) -1;
1317 gid_t gid = (gid_t) -1;
1323 rename_process_from_path(command->path);
1325 /* We reset exactly these signals, since they are the
1326 * only ones we set to SIG_IGN in the main daemon. All
1327 * others we leave untouched because we set them to
1328 * SIG_DFL or a valid handler initially, both of which
1329 * will be demoted to SIG_DFL. */
1330 default_signals(SIGNALS_CRASH_HANDLER,
1331 SIGNALS_IGNORE, -1);
1333 if (context->ignore_sigpipe)
1334 ignore_signals(SIGPIPE, -1);
1336 assert_se(sigemptyset(&ss) == 0);
1337 if (sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
1339 r = EXIT_SIGNAL_MASK;
1344 do_idle_pipe_dance(idle_pipe);
1346 /* Close sockets very early to make sure we don't
1347 * block init reexecution because it cannot bind its
1352 dont_close[n_dont_close++] = socket_fd;
1354 memcpy(dont_close + n_dont_close, fds, sizeof(int) * n_fds);
1355 n_dont_close += n_fds;
1358 if (runtime->netns_storage_socket[0] >= 0)
1359 dont_close[n_dont_close++] = runtime->netns_storage_socket[0];
1360 if (runtime->netns_storage_socket[1] >= 0)
1361 dont_close[n_dont_close++] = runtime->netns_storage_socket[1];
1364 err = close_all_fds(dont_close, n_dont_close);
1370 if (!context->same_pgrp)
1377 if (context->tcpwrap_name) {
1379 if (!socket_tcpwrap(socket_fd, context->tcpwrap_name)) {
1385 for (i = 0; i < (int) n_fds; i++) {
1386 if (!socket_tcpwrap(fds[i], context->tcpwrap_name)) {
1394 exec_context_tty_reset(context);
1396 if (confirm_spawn) {
1399 err = ask_for_confirmation(&response, argv);
1400 if (err == -ETIMEDOUT)
1401 write_confirm_message("Confirmation question timed out, assuming positive response.\n");
1403 write_confirm_message("Couldn't ask confirmation question, assuming positive response: %s\n", strerror(-err));
1404 else if (response == 's') {
1405 write_confirm_message("Skipping execution.\n");
1409 } else if (response == 'n') {
1410 write_confirm_message("Failing execution.\n");
1416 /* If a socket is connected to STDIN/STDOUT/STDERR, we
1417 * must sure to drop O_NONBLOCK */
1419 fd_nonblock(socket_fd, false);
1421 err = setup_input(context, socket_fd, apply_tty_stdin);
1427 err = setup_output(context, STDOUT_FILENO, socket_fd, basename(command->path), unit_id, apply_tty_stdin);
1433 err = setup_output(context, STDERR_FILENO, socket_fd, basename(command->path), unit_id, apply_tty_stdin);
1440 err = cg_attach_everywhere(cgroup_supported, cgroup_path, 0);
1447 if (context->oom_score_adjust_set) {
1450 snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
1453 if (write_string_file("/proc/self/oom_score_adj", t) < 0) {
1455 r = EXIT_OOM_ADJUST;
1460 if (context->nice_set)
1461 if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
1467 if (context->cpu_sched_set) {
1468 struct sched_param param = {
1469 .sched_priority = context->cpu_sched_priority,
1472 r = sched_setscheduler(0,
1473 context->cpu_sched_policy |
1474 (context->cpu_sched_reset_on_fork ?
1475 SCHED_RESET_ON_FORK : 0),
1479 r = EXIT_SETSCHEDULER;
1484 if (context->cpuset)
1485 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
1487 r = EXIT_CPUAFFINITY;
1491 if (context->ioprio_set)
1492 if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
1498 if (context->timer_slack_nsec != (nsec_t) -1)
1499 if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
1501 r = EXIT_TIMERSLACK;
1505 if (context->personality != 0xffffffffUL)
1506 if (personality(context->personality) < 0) {
1508 r = EXIT_PERSONALITY;
1512 if (context->utmp_id)
1513 utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path);
1515 if (context->user) {
1516 username = context->user;
1517 err = get_user_creds(&username, &uid, &gid, &home, &shell);
1523 if (is_terminal_input(context->std_input)) {
1524 err = chown_terminal(STDIN_FILENO, uid);
1533 if (cgroup_path && context->user && context->pam_name) {
1534 err = cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, 0644, uid, gid);
1541 err = cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, 0755, uid, gid);
1549 if (!strv_isempty(context->runtime_directory) && runtime_prefix) {
1552 STRV_FOREACH(rt, context->runtime_directory) {
1553 _cleanup_free_ char *p;
1555 p = strjoin(runtime_prefix, "/", *rt, NULL);
1557 r = EXIT_RUNTIME_DIRECTORY;
1562 err = mkdir_safe(p, context->runtime_directory_mode, uid, gid);
1564 r = EXIT_RUNTIME_DIRECTORY;
1570 if (apply_permissions) {
1571 err = enforce_groups(context, username, gid);
1578 umask(context->umask);
1581 if (apply_permissions && context->pam_name && username) {
1582 err = setup_pam(context->pam_name, username, uid, context->tty_path, &pam_env, fds, n_fds);
1589 if (context->private_network && runtime && runtime->netns_storage_socket[0] >= 0) {
1590 err = setup_netns(runtime->netns_storage_socket);
1597 if (!strv_isempty(context->read_write_dirs) ||
1598 !strv_isempty(context->read_only_dirs) ||
1599 !strv_isempty(context->inaccessible_dirs) ||
1600 context->mount_flags != 0 ||
1601 (context->private_tmp && runtime && (runtime->tmp_dir || runtime->var_tmp_dir)) ||
1602 context->private_devices) {
1604 char *tmp = NULL, *var = NULL;
1606 /* The runtime struct only contains the parent
1607 * of the private /tmp, which is
1608 * non-accessible to world users. Inside of it
1609 * there's a /tmp that is sticky, and that's
1610 * the one we want to use here. */
1612 if (context->private_tmp && runtime) {
1613 if (runtime->tmp_dir)
1614 tmp = strappenda(runtime->tmp_dir, "/tmp");
1615 if (runtime->var_tmp_dir)
1616 var = strappenda(runtime->var_tmp_dir, "/tmp");
1619 err = setup_namespace(
1620 context->read_write_dirs,
1621 context->read_only_dirs,
1622 context->inaccessible_dirs,
1625 context->private_devices,
1626 context->mount_flags);
1635 if (context->root_directory)
1636 if (chroot(context->root_directory) < 0) {
1642 if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
1648 _cleanup_free_ char *d = NULL;
1650 if (asprintf(&d, "%s/%s",
1651 context->root_directory ? context->root_directory : "",
1652 context->working_directory ? context->working_directory : "") < 0) {
1665 /* We repeat the fd closing here, to make sure that
1666 * nothing is leaked from the PAM modules */
1667 err = close_all_fds(fds, n_fds);
1669 err = shift_fds(fds, n_fds);
1671 err = flags_fds(fds, n_fds, context->non_blocking);
1677 if (apply_permissions) {
1679 for (i = 0; i < _RLIMIT_MAX; i++) {
1680 if (!context->rlimit[i])
1683 if (setrlimit_closest(i, context->rlimit[i]) < 0) {
1690 if (context->capability_bounding_set_drop) {
1691 err = capability_bounding_set_drop(context->capability_bounding_set_drop, false);
1693 r = EXIT_CAPABILITIES;
1698 if (context->user) {
1699 err = enforce_user(context, uid);
1706 /* PR_GET_SECUREBITS is not privileged, while
1707 * PR_SET_SECUREBITS is. So to suppress
1708 * potential EPERMs we'll try not to call
1709 * PR_SET_SECUREBITS unless necessary. */
1710 if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
1711 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
1713 r = EXIT_SECUREBITS;
1717 if (context->capabilities)
1718 if (cap_set_proc(context->capabilities) < 0) {
1720 r = EXIT_CAPABILITIES;
1724 if (context->no_new_privileges)
1725 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
1727 r = EXIT_NO_NEW_PRIVILEGES;
1732 if (context->address_families_whitelist ||
1733 !set_isempty(context->address_families)) {
1734 err = apply_address_families(context);
1736 r = EXIT_ADDRESS_FAMILIES;
1741 if (context->syscall_whitelist ||
1742 !set_isempty(context->syscall_filter) ||
1743 !set_isempty(context->syscall_archs)) {
1744 err = apply_seccomp(context);
1753 if (context->selinux_context && use_selinux()) {
1754 err = setexeccon(context->selinux_context);
1755 if (err < 0 && !context->selinux_context_ignore) {
1756 r = EXIT_SELINUX_CONTEXT;
1762 #ifdef HAVE_APPARMOR
1763 if (context->apparmor_profile && use_apparmor()) {
1764 err = aa_change_onexec(context->apparmor_profile);
1765 if (err < 0 && !context->apparmor_profile_ignore) {
1766 r = EXIT_APPARMOR_PROFILE;
1773 err = build_environment(context, n_fds, watchdog_usec, home, username, shell, &our_env);
1779 final_env = strv_env_merge(5,
1782 context->environment,
1792 final_argv = replace_env_argv(argv, final_env);
1799 final_env = strv_env_clean(final_env);
1801 if (_unlikely_(log_get_max_level() >= LOG_PRI(LOG_DEBUG))) {
1802 line = exec_command_line(final_argv);
1805 log_struct_unit(LOG_DEBUG,
1807 "EXECUTABLE=%s", command->path,
1808 "MESSAGE=Executing: %s", line,
1815 execve(command->path, final_argv, final_env);
1822 log_struct(LOG_ERR, MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED),
1823 "EXECUTABLE=%s", command->path,
1824 "MESSAGE=Failed at step %s spawning %s: %s",
1825 exit_status_to_string(r, EXIT_STATUS_SYSTEMD),
1826 command->path, strerror(-err),
1835 log_struct_unit(LOG_DEBUG,
1837 "MESSAGE=Forked %s as "PID_FMT,
1841 /* We add the new process to the cgroup both in the child (so
1842 * that we can be sure that no user code is ever executed
1843 * outside of the cgroup) and in the parent (so that we can be
1844 * sure that when we kill the cgroup the process will be
1847 cg_attach(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, pid);
1849 exec_status_start(&command->exec_status, pid);
1855 void exec_context_init(ExecContext *c) {
1859 c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
1860 c->cpu_sched_policy = SCHED_OTHER;
1861 c->syslog_priority = LOG_DAEMON|LOG_INFO;
1862 c->syslog_level_prefix = true;
1863 c->ignore_sigpipe = true;
1864 c->timer_slack_nsec = (nsec_t) -1;
1865 c->personality = 0xffffffffUL;
1866 c->runtime_directory_mode = 0755;
1869 void exec_context_done(ExecContext *c) {
1874 strv_free(c->environment);
1875 c->environment = NULL;
1877 strv_free(c->environment_files);
1878 c->environment_files = NULL;
1880 for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1882 c->rlimit[l] = NULL;
1885 free(c->working_directory);
1886 c->working_directory = NULL;
1887 free(c->root_directory);
1888 c->root_directory = NULL;
1893 free(c->tcpwrap_name);
1894 c->tcpwrap_name = NULL;
1896 free(c->syslog_identifier);
1897 c->syslog_identifier = NULL;
1905 strv_free(c->supplementary_groups);
1906 c->supplementary_groups = NULL;
1911 if (c->capabilities) {
1912 cap_free(c->capabilities);
1913 c->capabilities = NULL;
1916 strv_free(c->read_only_dirs);
1917 c->read_only_dirs = NULL;
1919 strv_free(c->read_write_dirs);
1920 c->read_write_dirs = NULL;
1922 strv_free(c->inaccessible_dirs);
1923 c->inaccessible_dirs = NULL;
1926 CPU_FREE(c->cpuset);
1931 free(c->selinux_context);
1932 c->selinux_context = NULL;
1934 free(c->apparmor_profile);
1935 c->apparmor_profile = NULL;
1937 set_free(c->syscall_filter);
1938 c->syscall_filter = NULL;
1940 set_free(c->syscall_archs);
1941 c->syscall_archs = NULL;
1943 set_free(c->address_families);
1944 c->address_families = NULL;
1946 strv_free(c->runtime_directory);
1947 c->runtime_directory = NULL;
1950 int exec_context_destroy_runtime_directory(ExecContext *c, const char *runtime_prefix) {
1955 if (!runtime_prefix)
1958 STRV_FOREACH(i, c->runtime_directory) {
1959 _cleanup_free_ char *p;
1961 p = strjoin(runtime_prefix, "/", *i, NULL);
1965 /* We execute this synchronously, since we need to be
1966 * sure this is gone when we start the service
1968 rm_rf_dangerous(p, false, true, false);
1974 void exec_command_done(ExecCommand *c) {
1984 void exec_command_done_array(ExecCommand *c, unsigned n) {
1987 for (i = 0; i < n; i++)
1988 exec_command_done(c+i);
1991 void exec_command_free_list(ExecCommand *c) {
1995 LIST_REMOVE(command, c, i);
1996 exec_command_done(i);
2001 void exec_command_free_array(ExecCommand **c, unsigned n) {
2004 for (i = 0; i < n; i++) {
2005 exec_command_free_list(c[i]);
2010 int exec_context_load_environment(const ExecContext *c, char ***l) {
2011 char **i, **r = NULL;
2016 STRV_FOREACH(i, c->environment_files) {
2019 bool ignore = false;
2021 _cleanup_globfree_ glob_t pglob = {};
2031 if (!path_is_absolute(fn)) {
2039 /* Filename supports globbing, take all matching files */
2041 if (glob(fn, 0, NULL, &pglob) != 0) {
2046 return errno ? -errno : -EINVAL;
2048 count = pglob.gl_pathc;
2056 for (n = 0; n < count; n++) {
2057 k = load_env_file(pglob.gl_pathv[n], NULL, &p);
2065 /* Log invalid environment variables with filename */
2067 p = strv_env_clean_log(p, pglob.gl_pathv[n]);
2074 m = strv_env_merge(2, r, p);
2090 static bool tty_may_match_dev_console(const char *tty) {
2091 char *active = NULL, *console;
2094 if (startswith(tty, "/dev/"))
2097 /* trivial identity? */
2098 if (streq(tty, "console"))
2101 console = resolve_dev_console(&active);
2102 /* if we could not resolve, assume it may */
2106 /* "tty0" means the active VC, so it may be the same sometimes */
2107 b = streq(console, tty) || (streq(console, "tty0") && tty_is_vc(tty));
2113 bool exec_context_may_touch_console(ExecContext *ec) {
2114 return (ec->tty_reset || ec->tty_vhangup || ec->tty_vt_disallocate ||
2115 is_terminal_input(ec->std_input) ||
2116 is_terminal_output(ec->std_output) ||
2117 is_terminal_output(ec->std_error)) &&
2118 tty_may_match_dev_console(tty_path(ec));
2121 static void strv_fprintf(FILE *f, char **l) {
2127 fprintf(f, " %s", *g);
2130 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
2137 prefix = strempty(prefix);
2141 "%sWorkingDirectory: %s\n"
2142 "%sRootDirectory: %s\n"
2143 "%sNonBlocking: %s\n"
2144 "%sPrivateTmp: %s\n"
2145 "%sPrivateNetwork: %s\n"
2146 "%sPrivateDevices: %s\n"
2147 "%sIgnoreSIGPIPE: %s\n",
2149 prefix, c->working_directory ? c->working_directory : "/",
2150 prefix, c->root_directory ? c->root_directory : "/",
2151 prefix, yes_no(c->non_blocking),
2152 prefix, yes_no(c->private_tmp),
2153 prefix, yes_no(c->private_network),
2154 prefix, yes_no(c->private_devices),
2155 prefix, yes_no(c->ignore_sigpipe));
2157 STRV_FOREACH(e, c->environment)
2158 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
2160 STRV_FOREACH(e, c->environment_files)
2161 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
2163 if (c->tcpwrap_name)
2165 "%sTCPWrapName: %s\n",
2166 prefix, c->tcpwrap_name);
2173 if (c->oom_score_adjust_set)
2175 "%sOOMScoreAdjust: %i\n",
2176 prefix, c->oom_score_adjust);
2178 for (i = 0; i < RLIM_NLIMITS; i++)
2180 fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
2182 if (c->ioprio_set) {
2183 _cleanup_free_ char *class_str = NULL;
2185 ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c->ioprio), &class_str);
2187 "%sIOSchedulingClass: %s\n"
2188 "%sIOPriority: %i\n",
2189 prefix, strna(class_str),
2190 prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
2193 if (c->cpu_sched_set) {
2194 _cleanup_free_ char *policy_str = NULL;
2196 sched_policy_to_string_alloc(c->cpu_sched_policy, &policy_str);
2198 "%sCPUSchedulingPolicy: %s\n"
2199 "%sCPUSchedulingPriority: %i\n"
2200 "%sCPUSchedulingResetOnFork: %s\n",
2201 prefix, strna(policy_str),
2202 prefix, c->cpu_sched_priority,
2203 prefix, yes_no(c->cpu_sched_reset_on_fork));
2207 fprintf(f, "%sCPUAffinity:", prefix);
2208 for (i = 0; i < c->cpuset_ncpus; i++)
2209 if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
2210 fprintf(f, " %u", i);
2214 if (c->timer_slack_nsec != (nsec_t) -1)
2215 fprintf(f, "%sTimerSlackNSec: "NSEC_FMT "\n", prefix, c->timer_slack_nsec);
2218 "%sStandardInput: %s\n"
2219 "%sStandardOutput: %s\n"
2220 "%sStandardError: %s\n",
2221 prefix, exec_input_to_string(c->std_input),
2222 prefix, exec_output_to_string(c->std_output),
2223 prefix, exec_output_to_string(c->std_error));
2229 "%sTTYVHangup: %s\n"
2230 "%sTTYVTDisallocate: %s\n",
2231 prefix, c->tty_path,
2232 prefix, yes_no(c->tty_reset),
2233 prefix, yes_no(c->tty_vhangup),
2234 prefix, yes_no(c->tty_vt_disallocate));
2236 if (c->std_output == EXEC_OUTPUT_SYSLOG ||
2237 c->std_output == EXEC_OUTPUT_KMSG ||
2238 c->std_output == EXEC_OUTPUT_JOURNAL ||
2239 c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
2240 c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
2241 c->std_output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE ||
2242 c->std_error == EXEC_OUTPUT_SYSLOG ||
2243 c->std_error == EXEC_OUTPUT_KMSG ||
2244 c->std_error == EXEC_OUTPUT_JOURNAL ||
2245 c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
2246 c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
2247 c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE) {
2249 _cleanup_free_ char *fac_str = NULL, *lvl_str = NULL;
2251 log_facility_unshifted_to_string_alloc(c->syslog_priority >> 3, &fac_str);
2252 log_level_to_string_alloc(LOG_PRI(c->syslog_priority), &lvl_str);
2255 "%sSyslogFacility: %s\n"
2256 "%sSyslogLevel: %s\n",
2257 prefix, strna(fac_str),
2258 prefix, strna(lvl_str));
2261 if (c->capabilities) {
2262 _cleanup_cap_free_charp_ char *t;
2264 t = cap_to_text(c->capabilities, NULL);
2266 fprintf(f, "%sCapabilities: %s\n", prefix, t);
2270 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
2272 (c->secure_bits & 1<<SECURE_KEEP_CAPS) ? " keep-caps" : "",
2273 (c->secure_bits & 1<<SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
2274 (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
2275 (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
2276 (c->secure_bits & 1<<SECURE_NOROOT) ? " noroot" : "",
2277 (c->secure_bits & 1<<SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
2279 if (c->capability_bounding_set_drop) {
2281 fprintf(f, "%sCapabilityBoundingSet:", prefix);
2283 for (l = 0; l <= cap_last_cap(); l++)
2284 if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) {
2285 _cleanup_cap_free_charp_ char *t;
2289 fprintf(f, " %s", t);
2296 fprintf(f, "%sUser: %s\n", prefix, c->user);
2298 fprintf(f, "%sGroup: %s\n", prefix, c->group);
2300 if (strv_length(c->supplementary_groups) > 0) {
2301 fprintf(f, "%sSupplementaryGroups:", prefix);
2302 strv_fprintf(f, c->supplementary_groups);
2307 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
2309 if (strv_length(c->read_write_dirs) > 0) {
2310 fprintf(f, "%sReadWriteDirs:", prefix);
2311 strv_fprintf(f, c->read_write_dirs);
2315 if (strv_length(c->read_only_dirs) > 0) {
2316 fprintf(f, "%sReadOnlyDirs:", prefix);
2317 strv_fprintf(f, c->read_only_dirs);
2321 if (strv_length(c->inaccessible_dirs) > 0) {
2322 fprintf(f, "%sInaccessibleDirs:", prefix);
2323 strv_fprintf(f, c->inaccessible_dirs);
2329 "%sUtmpIdentifier: %s\n",
2330 prefix, c->utmp_id);
2332 if (c->selinux_context)
2334 "%sSELinuxContext: %s%s\n",
2335 prefix, c->selinux_context_ignore ? "-" : "", c->selinux_context);
2337 if (c->personality != 0xffffffffUL)
2339 "%sPersonality: %s\n",
2340 prefix, strna(personality_to_string(c->personality)));
2342 if (c->syscall_filter) {
2350 "%sSystemCallFilter: ",
2353 if (!c->syscall_whitelist)
2357 SET_FOREACH(id, c->syscall_filter, j) {
2358 _cleanup_free_ char *name = NULL;
2365 name = seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE, PTR_TO_INT(id) - 1);
2366 fputs(strna(name), f);
2373 if (c->syscall_archs) {
2380 "%sSystemCallArchitectures:",
2384 SET_FOREACH(id, c->syscall_archs, j)
2385 fprintf(f, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id) - 1)));
2390 if (c->syscall_errno != 0)
2392 "%sSystemCallErrorNumber: %s\n",
2393 prefix, strna(errno_to_name(c->syscall_errno)));
2395 if (c->apparmor_profile)
2397 "%sAppArmorProfile: %s%s\n",
2398 prefix, c->apparmor_profile_ignore ? "-" : "", c->apparmor_profile);
2401 void exec_status_start(ExecStatus *s, pid_t pid) {
2406 dual_timestamp_get(&s->start_timestamp);
2409 void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
2412 if (s->pid && s->pid != pid)
2416 dual_timestamp_get(&s->exit_timestamp);
2422 if (context->utmp_id)
2423 utmp_put_dead_process(context->utmp_id, pid, code, status);
2425 exec_context_tty_reset(context);
2429 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
2430 char buf[FORMAT_TIMESTAMP_MAX];
2442 "%sPID: "PID_FMT"\n",
2445 if (s->start_timestamp.realtime > 0)
2447 "%sStart Timestamp: %s\n",
2448 prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
2450 if (s->exit_timestamp.realtime > 0)
2452 "%sExit Timestamp: %s\n"
2454 "%sExit Status: %i\n",
2455 prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
2456 prefix, sigchld_code_to_string(s->code),
2460 char *exec_command_line(char **argv) {
2468 STRV_FOREACH(a, argv)
2471 if (!(n = new(char, k)))
2475 STRV_FOREACH(a, argv) {
2482 if (strpbrk(*a, WHITESPACE)) {
2493 /* FIXME: this doesn't really handle arguments that have
2494 * spaces and ticks in them */
2499 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
2501 const char *prefix2;
2510 p2 = strappend(prefix, "\t");
2511 prefix2 = p2 ? p2 : prefix;
2513 cmd = exec_command_line(c->argv);
2516 "%sCommand Line: %s\n",
2517 prefix, cmd ? cmd : strerror(ENOMEM));
2521 exec_status_dump(&c->exec_status, f, prefix2);
2526 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
2532 LIST_FOREACH(command, c, c)
2533 exec_command_dump(c, f, prefix);
2536 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
2543 /* It's kind of important, that we keep the order here */
2544 LIST_FIND_TAIL(command, *l, end);
2545 LIST_INSERT_AFTER(command, *l, end, e);
2550 int exec_command_set(ExecCommand *c, const char *path, ...) {
2558 l = strv_new_ap(path, ap);
2579 static int exec_runtime_allocate(ExecRuntime **rt) {
2584 *rt = new0(ExecRuntime, 1);
2589 (*rt)->netns_storage_socket[0] = (*rt)->netns_storage_socket[1] = -1;
2594 int exec_runtime_make(ExecRuntime **rt, ExecContext *c, const char *id) {
2604 if (!c->private_network && !c->private_tmp)
2607 r = exec_runtime_allocate(rt);
2611 if (c->private_network && (*rt)->netns_storage_socket[0] < 0) {
2612 if (socketpair(AF_UNIX, SOCK_DGRAM, 0, (*rt)->netns_storage_socket) < 0)
2616 if (c->private_tmp && !(*rt)->tmp_dir) {
2617 r = setup_tmp_dirs(id, &(*rt)->tmp_dir, &(*rt)->var_tmp_dir);
2625 ExecRuntime *exec_runtime_ref(ExecRuntime *r) {
2627 assert(r->n_ref > 0);
2633 ExecRuntime *exec_runtime_unref(ExecRuntime *r) {
2638 assert(r->n_ref > 0);
2641 if (r->n_ref <= 0) {
2643 free(r->var_tmp_dir);
2644 close_pipe(r->netns_storage_socket);
2651 int exec_runtime_serialize(ExecRuntime *rt, Unit *u, FILE *f, FDSet *fds) {
2660 unit_serialize_item(u, f, "tmp-dir", rt->tmp_dir);
2662 if (rt->var_tmp_dir)
2663 unit_serialize_item(u, f, "var-tmp-dir", rt->var_tmp_dir);
2665 if (rt->netns_storage_socket[0] >= 0) {
2668 copy = fdset_put_dup(fds, rt->netns_storage_socket[0]);
2672 unit_serialize_item_format(u, f, "netns-socket-0", "%i", copy);
2675 if (rt->netns_storage_socket[1] >= 0) {
2678 copy = fdset_put_dup(fds, rt->netns_storage_socket[1]);
2682 unit_serialize_item_format(u, f, "netns-socket-1", "%i", copy);
2688 int exec_runtime_deserialize_item(ExecRuntime **rt, Unit *u, const char *key, const char *value, FDSet *fds) {
2695 if (streq(key, "tmp-dir")) {
2698 r = exec_runtime_allocate(rt);
2702 copy = strdup(value);
2706 free((*rt)->tmp_dir);
2707 (*rt)->tmp_dir = copy;
2709 } else if (streq(key, "var-tmp-dir")) {
2712 r = exec_runtime_allocate(rt);
2716 copy = strdup(value);
2720 free((*rt)->var_tmp_dir);
2721 (*rt)->var_tmp_dir = copy;
2723 } else if (streq(key, "netns-socket-0")) {
2726 r = exec_runtime_allocate(rt);
2730 if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
2731 log_debug_unit(u->id, "Failed to parse netns socket value %s", value);
2733 if ((*rt)->netns_storage_socket[0] >= 0)
2734 close_nointr_nofail((*rt)->netns_storage_socket[0]);
2736 (*rt)->netns_storage_socket[0] = fdset_remove(fds, fd);
2738 } else if (streq(key, "netns-socket-1")) {
2741 r = exec_runtime_allocate(rt);
2745 if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
2746 log_debug_unit(u->id, "Failed to parse netns socket value %s", value);
2748 if ((*rt)->netns_storage_socket[1] >= 0)
2749 close_nointr_nofail((*rt)->netns_storage_socket[1]);
2751 (*rt)->netns_storage_socket[1] = fdset_remove(fds, fd);
2759 static void *remove_tmpdir_thread(void *p) {
2760 _cleanup_free_ char *path = p;
2762 rm_rf_dangerous(path, false, true, false);
2766 void exec_runtime_destroy(ExecRuntime *rt) {
2772 /* If there are multiple users of this, let's leave the stuff around */
2777 log_debug("Spawning thread to nuke %s", rt->tmp_dir);
2779 r = asynchronous_job(remove_tmpdir_thread, rt->tmp_dir);
2781 log_warning("Failed to nuke %s: %s", rt->tmp_dir, strerror(-r));
2788 if (rt->var_tmp_dir) {
2789 log_debug("Spawning thread to nuke %s", rt->var_tmp_dir);
2791 r = asynchronous_job(remove_tmpdir_thread, rt->var_tmp_dir);
2793 log_warning("Failed to nuke %s: %s", rt->var_tmp_dir, strerror(-r));
2794 free(rt->var_tmp_dir);
2797 rt->var_tmp_dir = NULL;
2800 close_pipe(rt->netns_storage_socket);
2803 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
2804 [EXEC_INPUT_NULL] = "null",
2805 [EXEC_INPUT_TTY] = "tty",
2806 [EXEC_INPUT_TTY_FORCE] = "tty-force",
2807 [EXEC_INPUT_TTY_FAIL] = "tty-fail",
2808 [EXEC_INPUT_SOCKET] = "socket"
2811 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
2813 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
2814 [EXEC_OUTPUT_INHERIT] = "inherit",
2815 [EXEC_OUTPUT_NULL] = "null",
2816 [EXEC_OUTPUT_TTY] = "tty",
2817 [EXEC_OUTPUT_SYSLOG] = "syslog",
2818 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
2819 [EXEC_OUTPUT_KMSG] = "kmsg",
2820 [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
2821 [EXEC_OUTPUT_JOURNAL] = "journal",
2822 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE] = "journal+console",
2823 [EXEC_OUTPUT_SOCKET] = "socket"
2826 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);