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"
86 #include "smack-util.h"
87 #include "bus-kernel.h"
91 #include "seccomp-util.h"
94 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
95 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
97 /* This assumes there is a 'tty' group */
100 #define SNDBUF_SIZE (8*1024*1024)
102 static int shift_fds(int fds[], unsigned n_fds) {
103 int start, restart_from;
108 /* Modifies the fds array! (sorts it) */
118 for (i = start; i < (int) n_fds; i++) {
121 /* Already at right index? */
125 if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
131 /* Hmm, the fd we wanted isn't free? Then
132 * let's remember that and try again from here*/
133 if (nfd != i+3 && restart_from < 0)
137 if (restart_from < 0)
140 start = restart_from;
146 static int flags_fds(const int fds[], unsigned n_fds, bool nonblock) {
155 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
157 for (i = 0; i < n_fds; i++) {
159 if ((r = fd_nonblock(fds[i], nonblock)) < 0)
162 /* We unconditionally drop FD_CLOEXEC from the fds,
163 * since after all we want to pass these fds to our
166 if ((r = fd_cloexec(fds[i], false)) < 0)
173 _pure_ static const char *tty_path(const ExecContext *context) {
176 if (context->tty_path)
177 return context->tty_path;
179 return "/dev/console";
182 static void exec_context_tty_reset(const ExecContext *context) {
185 if (context->tty_vhangup)
186 terminal_vhangup(tty_path(context));
188 if (context->tty_reset)
189 reset_terminal(tty_path(context));
191 if (context->tty_vt_disallocate && context->tty_path)
192 vt_disallocate(context->tty_path);
195 static bool is_terminal_output(ExecOutput o) {
197 o == EXEC_OUTPUT_TTY ||
198 o == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
199 o == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
200 o == EXEC_OUTPUT_JOURNAL_AND_CONSOLE;
203 static int open_null_as(int flags, int nfd) {
208 fd = open("/dev/null", flags|O_NOCTTY);
213 r = dup2(fd, nfd) < 0 ? -errno : nfd;
221 static int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, const char *unit_id, int nfd) {
223 union sockaddr_union sa = {
224 .un.sun_family = AF_UNIX,
225 .un.sun_path = "/run/systemd/journal/stdout",
229 assert(output < _EXEC_OUTPUT_MAX);
233 fd = socket(AF_UNIX, SOCK_STREAM, 0);
237 r = connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path));
243 if (shutdown(fd, SHUT_RD) < 0) {
248 fd_inc_sndbuf(fd, SNDBUF_SIZE);
258 context->syslog_identifier ? context->syslog_identifier : ident,
260 context->syslog_priority,
261 !!context->syslog_level_prefix,
262 output == EXEC_OUTPUT_SYSLOG || output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE,
263 output == EXEC_OUTPUT_KMSG || output == EXEC_OUTPUT_KMSG_AND_CONSOLE,
264 is_terminal_output(output));
267 r = dup2(fd, nfd) < 0 ? -errno : nfd;
274 static int open_terminal_as(const char *path, mode_t mode, int nfd) {
280 if ((fd = open_terminal(path, mode | O_NOCTTY)) < 0)
284 r = dup2(fd, nfd) < 0 ? -errno : nfd;
292 static bool is_terminal_input(ExecInput i) {
294 i == EXEC_INPUT_TTY ||
295 i == EXEC_INPUT_TTY_FORCE ||
296 i == EXEC_INPUT_TTY_FAIL;
299 static int fixup_input(ExecInput std_input, int socket_fd, bool apply_tty_stdin) {
301 if (is_terminal_input(std_input) && !apply_tty_stdin)
302 return EXEC_INPUT_NULL;
304 if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0)
305 return EXEC_INPUT_NULL;
310 static int fixup_output(ExecOutput std_output, int socket_fd) {
312 if (std_output == EXEC_OUTPUT_SOCKET && socket_fd < 0)
313 return EXEC_OUTPUT_INHERIT;
318 static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty_stdin) {
323 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
327 case EXEC_INPUT_NULL:
328 return open_null_as(O_RDONLY, STDIN_FILENO);
331 case EXEC_INPUT_TTY_FORCE:
332 case EXEC_INPUT_TTY_FAIL: {
335 fd = acquire_terminal(tty_path(context),
336 i == EXEC_INPUT_TTY_FAIL,
337 i == EXEC_INPUT_TTY_FORCE,
343 if (fd != STDIN_FILENO) {
344 r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
352 case EXEC_INPUT_SOCKET:
353 return dup2(socket_fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
356 assert_not_reached("Unknown input type");
360 static int setup_output(const ExecContext *context, int fileno, int socket_fd, const char *ident, const char *unit_id, bool apply_tty_stdin) {
368 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
369 o = fixup_output(context->std_output, socket_fd);
371 if (fileno == STDERR_FILENO) {
373 e = fixup_output(context->std_error, socket_fd);
375 /* This expects the input and output are already set up */
377 /* Don't change the stderr file descriptor if we inherit all
378 * the way and are not on a tty */
379 if (e == EXEC_OUTPUT_INHERIT &&
380 o == EXEC_OUTPUT_INHERIT &&
381 i == EXEC_INPUT_NULL &&
382 !is_terminal_input(context->std_input) &&
386 /* Duplicate from stdout if possible */
387 if (e == o || e == EXEC_OUTPUT_INHERIT)
388 return dup2(STDOUT_FILENO, fileno) < 0 ? -errno : fileno;
392 } else if (o == EXEC_OUTPUT_INHERIT) {
393 /* If input got downgraded, inherit the original value */
394 if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input))
395 return open_terminal_as(tty_path(context), O_WRONLY, fileno);
397 /* If the input is connected to anything that's not a /dev/null, inherit that... */
398 if (i != EXEC_INPUT_NULL)
399 return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
401 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
405 /* We need to open /dev/null here anew, to get the right access mode. */
406 return open_null_as(O_WRONLY, fileno);
411 case EXEC_OUTPUT_NULL:
412 return open_null_as(O_WRONLY, fileno);
414 case EXEC_OUTPUT_TTY:
415 if (is_terminal_input(i))
416 return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
418 /* We don't reset the terminal if this is just about output */
419 return open_terminal_as(tty_path(context), O_WRONLY, fileno);
421 case EXEC_OUTPUT_SYSLOG:
422 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
423 case EXEC_OUTPUT_KMSG:
424 case EXEC_OUTPUT_KMSG_AND_CONSOLE:
425 case EXEC_OUTPUT_JOURNAL:
426 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE:
427 r = connect_logger_as(context, o, ident, unit_id, fileno);
429 log_unit_struct(unit_id,
431 LOG_MESSAGE("Failed to connect %s of %s to the journal socket: %s",
432 fileno == STDOUT_FILENO ? "stdout" : "stderr",
433 unit_id, strerror(-r)),
436 r = open_null_as(O_WRONLY, fileno);
440 case EXEC_OUTPUT_SOCKET:
441 assert(socket_fd >= 0);
442 return dup2(socket_fd, fileno) < 0 ? -errno : fileno;
445 assert_not_reached("Unknown error type");
449 static int chown_terminal(int fd, uid_t uid) {
454 /* This might fail. What matters are the results. */
455 (void) fchown(fd, uid, -1);
456 (void) fchmod(fd, TTY_MODE);
458 if (fstat(fd, &st) < 0)
461 if (st.st_uid != uid || (st.st_mode & 0777) != TTY_MODE)
467 static int setup_confirm_stdio(int *_saved_stdin,
468 int *_saved_stdout) {
469 int fd = -1, saved_stdin, saved_stdout = -1, r;
471 assert(_saved_stdin);
472 assert(_saved_stdout);
474 saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3);
478 saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3);
479 if (saved_stdout < 0) {
484 fd = acquire_terminal(
489 DEFAULT_CONFIRM_USEC);
495 r = chown_terminal(fd, getuid());
499 if (dup2(fd, STDIN_FILENO) < 0) {
504 if (dup2(fd, STDOUT_FILENO) < 0) {
512 *_saved_stdin = saved_stdin;
513 *_saved_stdout = saved_stdout;
518 safe_close(saved_stdout);
519 safe_close(saved_stdin);
525 _printf_(1, 2) static int write_confirm_message(const char *format, ...) {
526 _cleanup_close_ int fd = -1;
531 fd = open_terminal("/dev/console", O_WRONLY|O_NOCTTY|O_CLOEXEC);
535 va_start(ap, format);
536 vdprintf(fd, format, ap);
542 static int restore_confirm_stdio(int *saved_stdin,
548 assert(saved_stdout);
552 if (*saved_stdin >= 0)
553 if (dup2(*saved_stdin, STDIN_FILENO) < 0)
556 if (*saved_stdout >= 0)
557 if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
560 safe_close(*saved_stdin);
561 safe_close(*saved_stdout);
566 static int ask_for_confirmation(char *response, char **argv) {
567 int saved_stdout = -1, saved_stdin = -1, r;
568 _cleanup_free_ char *line = NULL;
570 r = setup_confirm_stdio(&saved_stdin, &saved_stdout);
574 line = exec_command_line(argv);
578 r = ask_char(response, "yns", "Execute %s? [Yes, No, Skip] ", line);
580 restore_confirm_stdio(&saved_stdin, &saved_stdout);
585 static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
586 bool keep_groups = false;
591 /* Lookup and set GID and supplementary group list. Here too
592 * we avoid NSS lookups for gid=0. */
594 if (context->group || username) {
596 if (context->group) {
597 const char *g = context->group;
599 if ((r = get_group_creds(&g, &gid)) < 0)
603 /* First step, initialize groups from /etc/groups */
604 if (username && gid != 0) {
605 if (initgroups(username, gid) < 0)
611 /* Second step, set our gids */
612 if (setresgid(gid, gid, gid) < 0)
616 if (context->supplementary_groups) {
621 /* Final step, initialize any manually set supplementary groups */
622 assert_se((ngroups_max = (int) sysconf(_SC_NGROUPS_MAX)) > 0);
624 if (!(gids = new(gid_t, ngroups_max)))
628 if ((k = getgroups(ngroups_max, gids)) < 0) {
635 STRV_FOREACH(i, context->supplementary_groups) {
638 if (k >= ngroups_max) {
644 r = get_group_creds(&g, gids+k);
653 if (setgroups(k, gids) < 0) {
664 static int enforce_user(const ExecContext *context, uid_t uid) {
667 /* Sets (but doesn't lookup) the uid and make sure we keep the
668 * capabilities while doing so. */
670 if (context->capabilities) {
671 _cleanup_cap_free_ cap_t d = NULL;
672 static const cap_value_t bits[] = {
673 CAP_SETUID, /* Necessary so that we can run setresuid() below */
674 CAP_SETPCAP /* Necessary so that we can set PR_SET_SECUREBITS later on */
677 /* First step: If we need to keep capabilities but
678 * drop privileges we need to make sure we keep our
679 * caps, while we drop privileges. */
681 int sb = context->secure_bits | 1<<SECURE_KEEP_CAPS;
683 if (prctl(PR_GET_SECUREBITS) != sb)
684 if (prctl(PR_SET_SECUREBITS, sb) < 0)
688 /* Second step: set the capabilities. This will reduce
689 * the capabilities to the minimum we need. */
691 d = cap_dup(context->capabilities);
695 if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
696 cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0)
699 if (cap_set_proc(d) < 0)
703 /* Third step: actually set the uids */
704 if (setresuid(uid, uid, uid) < 0)
707 /* At this point we should have all necessary capabilities but
708 are otherwise a normal user. However, the caps might got
709 corrupted due to the setresuid() so we need clean them up
710 later. This is done outside of this call. */
717 static int null_conv(
719 const struct pam_message **msg,
720 struct pam_response **resp,
723 /* We don't support conversations */
728 static int setup_pam(
734 int fds[], unsigned n_fds) {
736 static const struct pam_conv conv = {
741 pam_handle_t *handle = NULL;
743 int pam_code = PAM_SUCCESS;
746 bool close_session = false;
747 pid_t pam_pid = 0, parent_pid;
754 /* We set up PAM in the parent process, then fork. The child
755 * will then stay around until killed via PR_GET_PDEATHSIG or
756 * systemd via the cgroup logic. It will then remove the PAM
757 * session again. The parent process will exec() the actual
758 * daemon. We do things this way to ensure that the main PID
759 * of the daemon is the one we initially fork()ed. */
761 if (log_get_max_level() < LOG_PRI(LOG_DEBUG))
764 pam_code = pam_start(name, user, &conv, &handle);
765 if (pam_code != PAM_SUCCESS) {
771 pam_code = pam_set_item(handle, PAM_TTY, tty);
772 if (pam_code != PAM_SUCCESS)
776 pam_code = pam_acct_mgmt(handle, flags);
777 if (pam_code != PAM_SUCCESS)
780 pam_code = pam_open_session(handle, flags);
781 if (pam_code != PAM_SUCCESS)
784 close_session = true;
786 e = pam_getenvlist(handle);
788 pam_code = PAM_BUF_ERR;
792 /* Block SIGTERM, so that we know that it won't get lost in
794 if (sigemptyset(&ss) < 0 ||
795 sigaddset(&ss, SIGTERM) < 0 ||
796 sigprocmask(SIG_BLOCK, &ss, &old_ss) < 0)
799 parent_pid = getpid();
809 /* The child's job is to reset the PAM session on
812 /* This string must fit in 10 chars (i.e. the length
813 * of "/sbin/init"), to look pretty in /bin/ps */
814 rename_process("(sd-pam)");
816 /* Make sure we don't keep open the passed fds in this
817 child. We assume that otherwise only those fds are
818 open here that have been opened by PAM. */
819 close_many(fds, n_fds);
821 /* Drop privileges - we don't need any to pam_close_session
822 * and this will make PR_SET_PDEATHSIG work in most cases.
823 * If this fails, ignore the error - but expect sd-pam threads
824 * to fail to exit normally */
825 if (setresuid(uid, uid, uid) < 0)
826 log_error_errno(r, "Error: Failed to setresuid() in sd-pam: %m");
828 /* Wait until our parent died. This will only work if
829 * the above setresuid() succeeds, otherwise the kernel
830 * will not allow unprivileged parents kill their privileged
831 * children this way. We rely on the control groups kill logic
832 * to do the rest for us. */
833 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
836 /* Check if our parent process might already have
838 if (getppid() == parent_pid) {
840 if (sigwait(&ss, &sig) < 0) {
847 assert(sig == SIGTERM);
852 /* If our parent died we'll end the session */
853 if (getppid() != parent_pid) {
854 pam_code = pam_close_session(handle, flags);
855 if (pam_code != PAM_SUCCESS)
862 pam_end(handle, pam_code | flags);
866 /* If the child was forked off successfully it will do all the
867 * cleanups, so forget about the handle here. */
870 /* Unblock SIGTERM again in the parent */
871 if (sigprocmask(SIG_SETMASK, &old_ss, NULL) < 0)
874 /* We close the log explicitly here, since the PAM modules
875 * might have opened it, but we don't want this fd around. */
884 if (pam_code != PAM_SUCCESS) {
885 log_error("PAM failed: %s", pam_strerror(handle, pam_code));
886 err = -EPERM; /* PAM errors do not map to errno */
888 log_error_errno(errno, "PAM failed: %m");
894 pam_code = pam_close_session(handle, flags);
896 pam_end(handle, pam_code | flags);
904 kill(pam_pid, SIGTERM);
905 kill(pam_pid, SIGCONT);
912 static void rename_process_from_path(const char *path) {
913 char process_name[11];
917 /* This resulting string must fit in 10 chars (i.e. the length
918 * of "/sbin/init") to look pretty in /bin/ps */
922 rename_process("(...)");
928 /* The end of the process name is usually more
929 * interesting, since the first bit might just be
935 process_name[0] = '(';
936 memcpy(process_name+1, p, l);
937 process_name[1+l] = ')';
938 process_name[1+l+1] = 0;
940 rename_process(process_name);
945 static int apply_seccomp(const ExecContext *c) {
946 uint32_t negative_action, action;
947 scmp_filter_ctx *seccomp;
954 negative_action = c->syscall_errno == 0 ? SCMP_ACT_KILL : SCMP_ACT_ERRNO(c->syscall_errno);
956 seccomp = seccomp_init(c->syscall_whitelist ? negative_action : SCMP_ACT_ALLOW);
960 if (c->syscall_archs) {
962 SET_FOREACH(id, c->syscall_archs, i) {
963 r = seccomp_arch_add(seccomp, PTR_TO_UINT32(id) - 1);
971 r = seccomp_add_secondary_archs(seccomp);
976 action = c->syscall_whitelist ? SCMP_ACT_ALLOW : negative_action;
977 SET_FOREACH(id, c->syscall_filter, i) {
978 r = seccomp_rule_add(seccomp, action, PTR_TO_INT(id) - 1, 0);
983 r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0);
987 r = seccomp_load(seccomp);
990 seccomp_release(seccomp);
994 static int apply_address_families(const ExecContext *c) {
995 scmp_filter_ctx *seccomp;
1001 seccomp = seccomp_init(SCMP_ACT_ALLOW);
1005 r = seccomp_add_secondary_archs(seccomp);
1009 if (c->address_families_whitelist) {
1010 int af, first = 0, last = 0;
1013 /* If this is a whitelist, we first block the address
1014 * families that are out of range and then everything
1015 * that is not in the set. First, we find the lowest
1016 * and highest address family in the set. */
1018 SET_FOREACH(afp, c->address_families, i) {
1019 af = PTR_TO_INT(afp);
1021 if (af <= 0 || af >= af_max())
1024 if (first == 0 || af < first)
1027 if (last == 0 || af > last)
1031 assert((first == 0) == (last == 0));
1035 /* No entries in the valid range, block everything */
1036 r = seccomp_rule_add(
1038 SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1046 /* Block everything below the first entry */
1047 r = seccomp_rule_add(
1049 SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1052 SCMP_A0(SCMP_CMP_LT, first));
1056 /* Block everything above the last entry */
1057 r = seccomp_rule_add(
1059 SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1062 SCMP_A0(SCMP_CMP_GT, last));
1066 /* Block everything between the first and last
1068 for (af = 1; af < af_max(); af++) {
1070 if (set_contains(c->address_families, INT_TO_PTR(af)))
1073 r = seccomp_rule_add(
1075 SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1078 SCMP_A0(SCMP_CMP_EQ, af));
1087 /* If this is a blacklist, then generate one rule for
1088 * each address family that are then combined in OR
1091 SET_FOREACH(af, c->address_families, i) {
1093 r = seccomp_rule_add(
1095 SCMP_ACT_ERRNO(EPROTONOSUPPORT),
1098 SCMP_A0(SCMP_CMP_EQ, PTR_TO_INT(af)));
1104 r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0);
1108 r = seccomp_load(seccomp);
1111 seccomp_release(seccomp);
1117 static void do_idle_pipe_dance(int idle_pipe[4]) {
1121 safe_close(idle_pipe[1]);
1122 safe_close(idle_pipe[2]);
1124 if (idle_pipe[0] >= 0) {
1127 r = fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT_USEC);
1129 if (idle_pipe[3] >= 0 && r == 0 /* timeout */) {
1130 /* Signal systemd that we are bored and want to continue. */
1131 write(idle_pipe[3], "x", 1);
1133 /* Wait for systemd to react to the signal above. */
1134 fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT2_USEC);
1137 safe_close(idle_pipe[0]);
1141 safe_close(idle_pipe[3]);
1144 static int build_environment(
1145 const ExecContext *c,
1147 usec_t watchdog_usec,
1149 const char *username,
1153 _cleanup_strv_free_ char **our_env = NULL;
1160 our_env = new0(char*, 10);
1165 if (asprintf(&x, "LISTEN_PID="PID_FMT, getpid()) < 0)
1167 our_env[n_env++] = x;
1169 if (asprintf(&x, "LISTEN_FDS=%u", n_fds) < 0)
1171 our_env[n_env++] = x;
1174 if (watchdog_usec > 0) {
1175 if (asprintf(&x, "WATCHDOG_PID="PID_FMT, getpid()) < 0)
1177 our_env[n_env++] = x;
1179 if (asprintf(&x, "WATCHDOG_USEC="USEC_FMT, watchdog_usec) < 0)
1181 our_env[n_env++] = x;
1185 x = strappend("HOME=", home);
1188 our_env[n_env++] = x;
1192 x = strappend("LOGNAME=", username);
1195 our_env[n_env++] = x;
1197 x = strappend("USER=", username);
1200 our_env[n_env++] = x;
1204 x = strappend("SHELL=", shell);
1207 our_env[n_env++] = x;
1210 if (is_terminal_input(c->std_input) ||
1211 c->std_output == EXEC_OUTPUT_TTY ||
1212 c->std_error == EXEC_OUTPUT_TTY ||
1215 x = strdup(default_term_for_tty(tty_path(c)));
1218 our_env[n_env++] = x;
1221 our_env[n_env++] = NULL;
1222 assert(n_env <= 10);
1230 static int exec_child(ExecCommand *command,
1231 const ExecContext *context,
1232 const ExecParameters *params,
1233 ExecRuntime *runtime,
1236 int *fds, unsigned n_fds,
1240 _cleanup_strv_free_ char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
1241 _cleanup_free_ char *mac_selinux_context_net = NULL;
1242 const char *username = NULL, *home = NULL, *shell = NULL;
1243 unsigned n_dont_close = 0;
1244 int dont_close[n_fds + 4];
1245 uid_t uid = UID_INVALID;
1246 gid_t gid = GID_INVALID;
1254 rename_process_from_path(command->path);
1256 /* We reset exactly these signals, since they are the
1257 * only ones we set to SIG_IGN in the main daemon. All
1258 * others we leave untouched because we set them to
1259 * SIG_DFL or a valid handler initially, both of which
1260 * will be demoted to SIG_DFL. */
1261 default_signals(SIGNALS_CRASH_HANDLER,
1262 SIGNALS_IGNORE, -1);
1264 if (context->ignore_sigpipe)
1265 ignore_signals(SIGPIPE, -1);
1267 err = reset_signal_mask();
1269 *error = EXIT_SIGNAL_MASK;
1273 if (params->idle_pipe)
1274 do_idle_pipe_dance(params->idle_pipe);
1276 /* Close sockets very early to make sure we don't
1277 * block init reexecution because it cannot bind its
1282 dont_close[n_dont_close++] = socket_fd;
1284 memcpy(dont_close + n_dont_close, fds, sizeof(int) * n_fds);
1285 n_dont_close += n_fds;
1287 if (params->bus_endpoint_fd >= 0)
1288 dont_close[n_dont_close++] = params->bus_endpoint_fd;
1290 if (runtime->netns_storage_socket[0] >= 0)
1291 dont_close[n_dont_close++] = runtime->netns_storage_socket[0];
1292 if (runtime->netns_storage_socket[1] >= 0)
1293 dont_close[n_dont_close++] = runtime->netns_storage_socket[1];
1296 err = close_all_fds(dont_close, n_dont_close);
1302 if (!context->same_pgrp)
1304 *error = EXIT_SETSID;
1308 exec_context_tty_reset(context);
1310 if (params->confirm_spawn) {
1313 err = ask_for_confirmation(&response, argv);
1314 if (err == -ETIMEDOUT)
1315 write_confirm_message("Confirmation question timed out, assuming positive response.\n");
1317 write_confirm_message("Couldn't ask confirmation question, assuming positive response: %s\n", strerror(-err));
1318 else if (response == 's') {
1319 write_confirm_message("Skipping execution.\n");
1320 *error = EXIT_CONFIRM;
1322 } else if (response == 'n') {
1323 write_confirm_message("Failing execution.\n");
1329 /* If a socket is connected to STDIN/STDOUT/STDERR, we
1330 * must sure to drop O_NONBLOCK */
1332 fd_nonblock(socket_fd, false);
1334 err = setup_input(context, socket_fd, params->apply_tty_stdin);
1336 *error = EXIT_STDIN;
1340 err = setup_output(context, STDOUT_FILENO, socket_fd, basename(command->path), params->unit_id, params->apply_tty_stdin);
1342 *error = EXIT_STDOUT;
1346 err = setup_output(context, STDERR_FILENO, socket_fd, basename(command->path), params->unit_id, params->apply_tty_stdin);
1348 *error = EXIT_STDERR;
1352 if (params->cgroup_path) {
1353 err = cg_attach_everywhere(params->cgroup_supported, params->cgroup_path, 0);
1355 *error = EXIT_CGROUP;
1360 if (context->oom_score_adjust_set) {
1363 snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
1366 if (write_string_file("/proc/self/oom_score_adj", t) < 0) {
1367 *error = EXIT_OOM_ADJUST;
1372 if (context->nice_set)
1373 if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
1378 if (context->cpu_sched_set) {
1379 struct sched_param param = {
1380 .sched_priority = context->cpu_sched_priority,
1383 err = sched_setscheduler(0,
1384 context->cpu_sched_policy |
1385 (context->cpu_sched_reset_on_fork ?
1386 SCHED_RESET_ON_FORK : 0),
1389 *error = EXIT_SETSCHEDULER;
1394 if (context->cpuset)
1395 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
1396 *error = EXIT_CPUAFFINITY;
1400 if (context->ioprio_set)
1401 if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
1402 *error = EXIT_IOPRIO;
1406 if (context->timer_slack_nsec != NSEC_INFINITY)
1407 if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
1408 *error = EXIT_TIMERSLACK;
1412 if (context->personality != 0xffffffffUL)
1413 if (personality(context->personality) < 0) {
1414 *error = EXIT_PERSONALITY;
1418 if (context->utmp_id)
1419 utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path);
1421 if (context->user) {
1422 username = context->user;
1423 err = get_user_creds(&username, &uid, &gid, &home, &shell);
1429 if (is_terminal_input(context->std_input)) {
1430 err = chown_terminal(STDIN_FILENO, uid);
1432 *error = EXIT_STDIN;
1439 if (params->bus_endpoint_fd >= 0 && context->bus_endpoint) {
1440 uid_t ep_uid = (uid == UID_INVALID) ? 0 : uid;
1442 err = bus_kernel_set_endpoint_policy(params->bus_endpoint_fd, ep_uid, context->bus_endpoint);
1444 *error = EXIT_BUS_ENDPOINT;
1450 /* If delegation is enabled we'll pass ownership of the cgroup
1451 * (but only in systemd's own controller hierarchy!) to the
1452 * user of the new process. */
1453 if (params->cgroup_path && context->user && params->cgroup_delegate) {
1454 err = cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, 0644, uid, gid);
1456 *error = EXIT_CGROUP;
1461 err = cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, 0755, uid, gid);
1463 *error = EXIT_CGROUP;
1468 if (!strv_isempty(context->runtime_directory) && params->runtime_prefix) {
1471 STRV_FOREACH(rt, context->runtime_directory) {
1472 _cleanup_free_ char *p;
1474 p = strjoin(params->runtime_prefix, "/", *rt, NULL);
1476 *error = EXIT_RUNTIME_DIRECTORY;
1480 err = mkdir_safe(p, context->runtime_directory_mode, uid, gid);
1482 *error = EXIT_RUNTIME_DIRECTORY;
1488 if (params->apply_permissions) {
1489 err = enforce_groups(context, username, gid);
1491 *error = EXIT_GROUP;
1496 umask(context->umask);
1499 if (params->apply_permissions && context->pam_name && username) {
1500 err = setup_pam(context->pam_name, username, uid, context->tty_path, &pam_env, fds, n_fds);
1508 if (context->private_network && runtime && runtime->netns_storage_socket[0] >= 0) {
1509 err = setup_netns(runtime->netns_storage_socket);
1511 *error = EXIT_NETWORK;
1516 if (!strv_isempty(context->read_write_dirs) ||
1517 !strv_isempty(context->read_only_dirs) ||
1518 !strv_isempty(context->inaccessible_dirs) ||
1519 context->mount_flags != 0 ||
1520 (context->private_tmp && runtime && (runtime->tmp_dir || runtime->var_tmp_dir)) ||
1521 params->bus_endpoint_path ||
1522 context->private_devices ||
1523 context->protect_system != PROTECT_SYSTEM_NO ||
1524 context->protect_home != PROTECT_HOME_NO) {
1526 char *tmp = NULL, *var = NULL;
1528 /* The runtime struct only contains the parent
1529 * of the private /tmp, which is
1530 * non-accessible to world users. Inside of it
1531 * there's a /tmp that is sticky, and that's
1532 * the one we want to use here. */
1534 if (context->private_tmp && runtime) {
1535 if (runtime->tmp_dir)
1536 tmp = strappenda(runtime->tmp_dir, "/tmp");
1537 if (runtime->var_tmp_dir)
1538 var = strappenda(runtime->var_tmp_dir, "/tmp");
1541 err = setup_namespace(
1542 context->read_write_dirs,
1543 context->read_only_dirs,
1544 context->inaccessible_dirs,
1547 params->bus_endpoint_path,
1548 context->private_devices,
1549 context->protect_home,
1550 context->protect_system,
1551 context->mount_flags);
1554 log_unit_warning_errno(params->unit_id, err, "Failed to set up file system namespace due to lack of privileges. Execution sandbox will not be in effect: %m");
1556 *error = EXIT_NAMESPACE;
1561 if (params->apply_chroot) {
1562 if (context->root_directory)
1563 if (chroot(context->root_directory) < 0) {
1564 *error = EXIT_CHROOT;
1568 if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
1569 *error = EXIT_CHDIR;
1573 _cleanup_free_ char *d = NULL;
1575 if (asprintf(&d, "%s/%s",
1576 context->root_directory ? context->root_directory : "",
1577 context->working_directory ? context->working_directory : "") < 0) {
1578 *error = EXIT_MEMORY;
1583 *error = EXIT_CHDIR;
1589 if (params->apply_permissions && mac_selinux_use() && params->selinux_context_net && socket_fd >= 0) {
1590 err = mac_selinux_get_child_mls_label(socket_fd, command->path, context->selinux_context, &mac_selinux_context_net);
1592 *error = EXIT_SELINUX_CONTEXT;
1598 /* We repeat the fd closing here, to make sure that
1599 * nothing is leaked from the PAM modules. Note that
1600 * we are more aggressive this time since socket_fd
1601 * and the netns fds we don't need anymore. The custom
1602 * endpoint fd was needed to upload the policy and can
1603 * now be closed as well. */
1604 err = close_all_fds(fds, n_fds);
1606 err = shift_fds(fds, n_fds);
1608 err = flags_fds(fds, n_fds, context->non_blocking);
1614 if (params->apply_permissions) {
1616 for (i = 0; i < _RLIMIT_MAX; i++) {
1617 if (!context->rlimit[i])
1620 if (setrlimit_closest(i, context->rlimit[i]) < 0) {
1621 *error = EXIT_LIMITS;
1626 if (context->capability_bounding_set_drop) {
1627 err = capability_bounding_set_drop(context->capability_bounding_set_drop, false);
1629 *error = EXIT_CAPABILITIES;
1635 if (context->smack_process_label) {
1636 err = mac_smack_apply_pid(0, context->smack_process_label);
1638 *error = EXIT_SMACK_PROCESS_LABEL;
1644 if (context->user) {
1645 err = enforce_user(context, uid);
1652 /* PR_GET_SECUREBITS is not privileged, while
1653 * PR_SET_SECUREBITS is. So to suppress
1654 * potential EPERMs we'll try not to call
1655 * PR_SET_SECUREBITS unless necessary. */
1656 if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
1657 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
1658 *error = EXIT_SECUREBITS;
1662 if (context->capabilities)
1663 if (cap_set_proc(context->capabilities) < 0) {
1664 *error = EXIT_CAPABILITIES;
1668 if (context->no_new_privileges)
1669 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
1670 *error = EXIT_NO_NEW_PRIVILEGES;
1675 if (context->address_families_whitelist ||
1676 !set_isempty(context->address_families)) {
1677 err = apply_address_families(context);
1679 *error = EXIT_ADDRESS_FAMILIES;
1684 if (context->syscall_whitelist ||
1685 !set_isempty(context->syscall_filter) ||
1686 !set_isempty(context->syscall_archs)) {
1687 err = apply_seccomp(context);
1689 *error = EXIT_SECCOMP;
1696 if (mac_selinux_use()) {
1697 char *exec_context = mac_selinux_context_net ?: context->selinux_context;
1700 err = setexeccon(exec_context);
1702 *error = EXIT_SELINUX_CONTEXT;
1709 #ifdef HAVE_APPARMOR
1710 if (context->apparmor_profile && mac_apparmor_use()) {
1711 err = aa_change_onexec(context->apparmor_profile);
1712 if (err < 0 && !context->apparmor_profile_ignore) {
1713 *error = EXIT_APPARMOR_PROFILE;
1720 err = build_environment(context, n_fds, params->watchdog_usec, home, username, shell, &our_env);
1722 *error = EXIT_MEMORY;
1726 final_env = strv_env_merge(5,
1727 params->environment,
1729 context->environment,
1734 *error = EXIT_MEMORY;
1738 final_argv = replace_env_argv(argv, final_env);
1740 *error = EXIT_MEMORY;
1744 final_env = strv_env_clean(final_env);
1746 if (_unlikely_(log_get_max_level() >= LOG_PRI(LOG_DEBUG))) {
1747 _cleanup_free_ char *line;
1749 line = exec_command_line(final_argv);
1752 log_unit_struct(params->unit_id,
1754 "EXECUTABLE=%s", command->path,
1755 LOG_MESSAGE("Executing: %s", line),
1760 execve(command->path, final_argv, final_env);
1765 int exec_spawn(ExecCommand *command,
1766 const ExecContext *context,
1767 const ExecParameters *params,
1768 ExecRuntime *runtime,
1771 _cleanup_strv_free_ char **files_env = NULL;
1772 int *fds = NULL; unsigned n_fds = 0;
1782 assert(params->fds || params->n_fds <= 0);
1784 if (context->std_input == EXEC_INPUT_SOCKET ||
1785 context->std_output == EXEC_OUTPUT_SOCKET ||
1786 context->std_error == EXEC_OUTPUT_SOCKET) {
1788 if (params->n_fds != 1)
1791 socket_fd = params->fds[0];
1795 n_fds = params->n_fds;
1798 err = exec_context_load_environment(context, params->unit_id, &files_env);
1800 log_unit_struct(params->unit_id,
1802 LOG_MESSAGE("Failed to load environment files: %s", strerror(-err)),
1808 argv = params->argv ?: command->argv;
1810 line = exec_command_line(argv);
1814 log_unit_struct(params->unit_id,
1816 "EXECUTABLE=%s", command->path,
1817 LOG_MESSAGE("About to execute: %s", line),
1828 err = exec_child(command,
1840 LOG_MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED),
1841 "EXECUTABLE=%s", command->path,
1842 LOG_MESSAGE("Failed at step %s spawning %s: %s",
1843 exit_status_to_string(r, EXIT_STATUS_SYSTEMD),
1844 command->path, strerror(-err)),
1853 log_unit_struct(params->unit_id,
1855 LOG_MESSAGE("Forked %s as "PID_FMT,
1856 command->path, pid),
1859 /* We add the new process to the cgroup both in the child (so
1860 * that we can be sure that no user code is ever executed
1861 * outside of the cgroup) and in the parent (so that we can be
1862 * sure that when we kill the cgroup the process will be
1864 if (params->cgroup_path)
1865 cg_attach(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, pid);
1867 exec_status_start(&command->exec_status, pid);
1873 void exec_context_init(ExecContext *c) {
1877 c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
1878 c->cpu_sched_policy = SCHED_OTHER;
1879 c->syslog_priority = LOG_DAEMON|LOG_INFO;
1880 c->syslog_level_prefix = true;
1881 c->ignore_sigpipe = true;
1882 c->timer_slack_nsec = NSEC_INFINITY;
1883 c->personality = 0xffffffffUL;
1884 c->runtime_directory_mode = 0755;
1887 void exec_context_done(ExecContext *c) {
1892 strv_free(c->environment);
1893 c->environment = NULL;
1895 strv_free(c->environment_files);
1896 c->environment_files = NULL;
1898 for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1900 c->rlimit[l] = NULL;
1903 free(c->working_directory);
1904 c->working_directory = NULL;
1905 free(c->root_directory);
1906 c->root_directory = NULL;
1911 free(c->syslog_identifier);
1912 c->syslog_identifier = NULL;
1920 strv_free(c->supplementary_groups);
1921 c->supplementary_groups = NULL;
1926 if (c->capabilities) {
1927 cap_free(c->capabilities);
1928 c->capabilities = NULL;
1931 strv_free(c->read_only_dirs);
1932 c->read_only_dirs = NULL;
1934 strv_free(c->read_write_dirs);
1935 c->read_write_dirs = NULL;
1937 strv_free(c->inaccessible_dirs);
1938 c->inaccessible_dirs = NULL;
1941 CPU_FREE(c->cpuset);
1946 free(c->selinux_context);
1947 c->selinux_context = NULL;
1949 free(c->apparmor_profile);
1950 c->apparmor_profile = NULL;
1952 set_free(c->syscall_filter);
1953 c->syscall_filter = NULL;
1955 set_free(c->syscall_archs);
1956 c->syscall_archs = NULL;
1958 set_free(c->address_families);
1959 c->address_families = NULL;
1961 strv_free(c->runtime_directory);
1962 c->runtime_directory = NULL;
1964 bus_endpoint_free(c->bus_endpoint);
1965 c->bus_endpoint = NULL;
1968 int exec_context_destroy_runtime_directory(ExecContext *c, const char *runtime_prefix) {
1973 if (!runtime_prefix)
1976 STRV_FOREACH(i, c->runtime_directory) {
1977 _cleanup_free_ char *p;
1979 p = strjoin(runtime_prefix, "/", *i, NULL);
1983 /* We execute this synchronously, since we need to be
1984 * sure this is gone when we start the service
1986 rm_rf_dangerous(p, false, true, false);
1992 void exec_command_done(ExecCommand *c) {
2002 void exec_command_done_array(ExecCommand *c, unsigned n) {
2005 for (i = 0; i < n; i++)
2006 exec_command_done(c+i);
2009 void exec_command_free_list(ExecCommand *c) {
2013 LIST_REMOVE(command, c, i);
2014 exec_command_done(i);
2019 void exec_command_free_array(ExecCommand **c, unsigned n) {
2022 for (i = 0; i < n; i++) {
2023 exec_command_free_list(c[i]);
2028 int exec_context_load_environment(const ExecContext *c, const char *unit_id, char ***l) {
2029 char **i, **r = NULL;
2034 STRV_FOREACH(i, c->environment_files) {
2037 bool ignore = false;
2039 _cleanup_globfree_ glob_t pglob = {};
2049 if (!path_is_absolute(fn)) {
2057 /* Filename supports globbing, take all matching files */
2059 if (glob(fn, 0, NULL, &pglob) != 0) {
2064 return errno ? -errno : -EINVAL;
2066 count = pglob.gl_pathc;
2074 for (n = 0; n < count; n++) {
2075 k = load_env_file(NULL, pglob.gl_pathv[n], NULL, &p);
2083 /* Log invalid environment variables with filename */
2085 p = strv_env_clean_log(p, unit_id, pglob.gl_pathv[n]);
2092 m = strv_env_merge(2, r, p);
2108 static bool tty_may_match_dev_console(const char *tty) {
2109 _cleanup_free_ char *active = NULL;
2112 if (startswith(tty, "/dev/"))
2115 /* trivial identity? */
2116 if (streq(tty, "console"))
2119 console = resolve_dev_console(&active);
2120 /* if we could not resolve, assume it may */
2124 /* "tty0" means the active VC, so it may be the same sometimes */
2125 return streq(console, tty) || (streq(console, "tty0") && tty_is_vc(tty));
2128 bool exec_context_may_touch_console(ExecContext *ec) {
2129 return (ec->tty_reset || ec->tty_vhangup || ec->tty_vt_disallocate ||
2130 is_terminal_input(ec->std_input) ||
2131 is_terminal_output(ec->std_output) ||
2132 is_terminal_output(ec->std_error)) &&
2133 tty_may_match_dev_console(tty_path(ec));
2136 static void strv_fprintf(FILE *f, char **l) {
2142 fprintf(f, " %s", *g);
2145 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
2152 prefix = strempty(prefix);
2156 "%sWorkingDirectory: %s\n"
2157 "%sRootDirectory: %s\n"
2158 "%sNonBlocking: %s\n"
2159 "%sPrivateTmp: %s\n"
2160 "%sPrivateNetwork: %s\n"
2161 "%sPrivateDevices: %s\n"
2162 "%sProtectHome: %s\n"
2163 "%sProtectSystem: %s\n"
2164 "%sIgnoreSIGPIPE: %s\n",
2166 prefix, c->working_directory ? c->working_directory : "/",
2167 prefix, c->root_directory ? c->root_directory : "/",
2168 prefix, yes_no(c->non_blocking),
2169 prefix, yes_no(c->private_tmp),
2170 prefix, yes_no(c->private_network),
2171 prefix, yes_no(c->private_devices),
2172 prefix, protect_home_to_string(c->protect_home),
2173 prefix, protect_system_to_string(c->protect_system),
2174 prefix, yes_no(c->ignore_sigpipe));
2176 STRV_FOREACH(e, c->environment)
2177 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
2179 STRV_FOREACH(e, c->environment_files)
2180 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
2187 if (c->oom_score_adjust_set)
2189 "%sOOMScoreAdjust: %i\n",
2190 prefix, c->oom_score_adjust);
2192 for (i = 0; i < RLIM_NLIMITS; i++)
2194 fprintf(f, "%s%s: "RLIM_FMT"\n",
2195 prefix, rlimit_to_string(i), c->rlimit[i]->rlim_max);
2197 if (c->ioprio_set) {
2198 _cleanup_free_ char *class_str = NULL;
2200 ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c->ioprio), &class_str);
2202 "%sIOSchedulingClass: %s\n"
2203 "%sIOPriority: %i\n",
2204 prefix, strna(class_str),
2205 prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
2208 if (c->cpu_sched_set) {
2209 _cleanup_free_ char *policy_str = NULL;
2211 sched_policy_to_string_alloc(c->cpu_sched_policy, &policy_str);
2213 "%sCPUSchedulingPolicy: %s\n"
2214 "%sCPUSchedulingPriority: %i\n"
2215 "%sCPUSchedulingResetOnFork: %s\n",
2216 prefix, strna(policy_str),
2217 prefix, c->cpu_sched_priority,
2218 prefix, yes_no(c->cpu_sched_reset_on_fork));
2222 fprintf(f, "%sCPUAffinity:", prefix);
2223 for (i = 0; i < c->cpuset_ncpus; i++)
2224 if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
2225 fprintf(f, " %u", i);
2229 if (c->timer_slack_nsec != NSEC_INFINITY)
2230 fprintf(f, "%sTimerSlackNSec: "NSEC_FMT "\n", prefix, c->timer_slack_nsec);
2233 "%sStandardInput: %s\n"
2234 "%sStandardOutput: %s\n"
2235 "%sStandardError: %s\n",
2236 prefix, exec_input_to_string(c->std_input),
2237 prefix, exec_output_to_string(c->std_output),
2238 prefix, exec_output_to_string(c->std_error));
2244 "%sTTYVHangup: %s\n"
2245 "%sTTYVTDisallocate: %s\n",
2246 prefix, c->tty_path,
2247 prefix, yes_no(c->tty_reset),
2248 prefix, yes_no(c->tty_vhangup),
2249 prefix, yes_no(c->tty_vt_disallocate));
2251 if (c->std_output == EXEC_OUTPUT_SYSLOG ||
2252 c->std_output == EXEC_OUTPUT_KMSG ||
2253 c->std_output == EXEC_OUTPUT_JOURNAL ||
2254 c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
2255 c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
2256 c->std_output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE ||
2257 c->std_error == EXEC_OUTPUT_SYSLOG ||
2258 c->std_error == EXEC_OUTPUT_KMSG ||
2259 c->std_error == EXEC_OUTPUT_JOURNAL ||
2260 c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
2261 c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
2262 c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE) {
2264 _cleanup_free_ char *fac_str = NULL, *lvl_str = NULL;
2266 log_facility_unshifted_to_string_alloc(c->syslog_priority >> 3, &fac_str);
2267 log_level_to_string_alloc(LOG_PRI(c->syslog_priority), &lvl_str);
2270 "%sSyslogFacility: %s\n"
2271 "%sSyslogLevel: %s\n",
2272 prefix, strna(fac_str),
2273 prefix, strna(lvl_str));
2276 if (c->capabilities) {
2277 _cleanup_cap_free_charp_ char *t;
2279 t = cap_to_text(c->capabilities, NULL);
2281 fprintf(f, "%sCapabilities: %s\n", prefix, t);
2285 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
2287 (c->secure_bits & 1<<SECURE_KEEP_CAPS) ? " keep-caps" : "",
2288 (c->secure_bits & 1<<SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
2289 (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
2290 (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
2291 (c->secure_bits & 1<<SECURE_NOROOT) ? " noroot" : "",
2292 (c->secure_bits & 1<<SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
2294 if (c->capability_bounding_set_drop) {
2296 fprintf(f, "%sCapabilityBoundingSet:", prefix);
2298 for (l = 0; l <= cap_last_cap(); l++)
2299 if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) {
2300 _cleanup_cap_free_charp_ char *t;
2304 fprintf(f, " %s", t);
2311 fprintf(f, "%sUser: %s\n", prefix, c->user);
2313 fprintf(f, "%sGroup: %s\n", prefix, c->group);
2315 if (strv_length(c->supplementary_groups) > 0) {
2316 fprintf(f, "%sSupplementaryGroups:", prefix);
2317 strv_fprintf(f, c->supplementary_groups);
2322 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
2324 if (strv_length(c->read_write_dirs) > 0) {
2325 fprintf(f, "%sReadWriteDirs:", prefix);
2326 strv_fprintf(f, c->read_write_dirs);
2330 if (strv_length(c->read_only_dirs) > 0) {
2331 fprintf(f, "%sReadOnlyDirs:", prefix);
2332 strv_fprintf(f, c->read_only_dirs);
2336 if (strv_length(c->inaccessible_dirs) > 0) {
2337 fprintf(f, "%sInaccessibleDirs:", prefix);
2338 strv_fprintf(f, c->inaccessible_dirs);
2344 "%sUtmpIdentifier: %s\n",
2345 prefix, c->utmp_id);
2347 if (c->selinux_context)
2349 "%sSELinuxContext: %s%s\n",
2350 prefix, c->selinux_context_ignore ? "-" : "", c->selinux_context);
2352 if (c->personality != 0xffffffffUL)
2354 "%sPersonality: %s\n",
2355 prefix, strna(personality_to_string(c->personality)));
2357 if (c->syscall_filter) {
2365 "%sSystemCallFilter: ",
2368 if (!c->syscall_whitelist)
2372 SET_FOREACH(id, c->syscall_filter, j) {
2373 _cleanup_free_ char *name = NULL;
2380 name = seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE, PTR_TO_INT(id) - 1);
2381 fputs(strna(name), f);
2388 if (c->syscall_archs) {
2395 "%sSystemCallArchitectures:",
2399 SET_FOREACH(id, c->syscall_archs, j)
2400 fprintf(f, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id) - 1)));
2405 if (c->syscall_errno != 0)
2407 "%sSystemCallErrorNumber: %s\n",
2408 prefix, strna(errno_to_name(c->syscall_errno)));
2410 if (c->apparmor_profile)
2412 "%sAppArmorProfile: %s%s\n",
2413 prefix, c->apparmor_profile_ignore ? "-" : "", c->apparmor_profile);
2416 bool exec_context_maintains_privileges(ExecContext *c) {
2419 /* Returns true if the process forked off would run run under
2420 * an unchanged UID or as root. */
2425 if (streq(c->user, "root") || streq(c->user, "0"))
2431 void exec_status_start(ExecStatus *s, pid_t pid) {
2436 dual_timestamp_get(&s->start_timestamp);
2439 void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
2442 if (s->pid && s->pid != pid)
2446 dual_timestamp_get(&s->exit_timestamp);
2452 if (context->utmp_id)
2453 utmp_put_dead_process(context->utmp_id, pid, code, status);
2455 exec_context_tty_reset(context);
2459 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
2460 char buf[FORMAT_TIMESTAMP_MAX];
2468 prefix = strempty(prefix);
2471 "%sPID: "PID_FMT"\n",
2474 if (s->start_timestamp.realtime > 0)
2476 "%sStart Timestamp: %s\n",
2477 prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
2479 if (s->exit_timestamp.realtime > 0)
2481 "%sExit Timestamp: %s\n"
2483 "%sExit Status: %i\n",
2484 prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
2485 prefix, sigchld_code_to_string(s->code),
2489 char *exec_command_line(char **argv) {
2497 STRV_FOREACH(a, argv)
2500 if (!(n = new(char, k)))
2504 STRV_FOREACH(a, argv) {
2511 if (strpbrk(*a, WHITESPACE)) {
2522 /* FIXME: this doesn't really handle arguments that have
2523 * spaces and ticks in them */
2528 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
2529 _cleanup_free_ char *cmd = NULL;
2530 const char *prefix2;
2535 prefix = strempty(prefix);
2536 prefix2 = strappenda(prefix, "\t");
2538 cmd = exec_command_line(c->argv);
2540 "%sCommand Line: %s\n",
2541 prefix, cmd ? cmd : strerror(ENOMEM));
2543 exec_status_dump(&c->exec_status, f, prefix2);
2546 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
2549 prefix = strempty(prefix);
2551 LIST_FOREACH(command, c, c)
2552 exec_command_dump(c, f, prefix);
2555 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
2562 /* It's kind of important, that we keep the order here */
2563 LIST_FIND_TAIL(command, *l, end);
2564 LIST_INSERT_AFTER(command, *l, end, e);
2569 int exec_command_set(ExecCommand *c, const char *path, ...) {
2577 l = strv_new_ap(path, ap);
2598 int exec_command_append(ExecCommand *c, const char *path, ...) {
2599 _cleanup_strv_free_ char **l = NULL;
2607 l = strv_new_ap(path, ap);
2613 r = strv_extend_strv(&c->argv, l);
2621 static int exec_runtime_allocate(ExecRuntime **rt) {
2626 *rt = new0(ExecRuntime, 1);
2631 (*rt)->netns_storage_socket[0] = (*rt)->netns_storage_socket[1] = -1;
2636 int exec_runtime_make(ExecRuntime **rt, ExecContext *c, const char *id) {
2646 if (!c->private_network && !c->private_tmp)
2649 r = exec_runtime_allocate(rt);
2653 if (c->private_network && (*rt)->netns_storage_socket[0] < 0) {
2654 if (socketpair(AF_UNIX, SOCK_DGRAM, 0, (*rt)->netns_storage_socket) < 0)
2658 if (c->private_tmp && !(*rt)->tmp_dir) {
2659 r = setup_tmp_dirs(id, &(*rt)->tmp_dir, &(*rt)->var_tmp_dir);
2667 ExecRuntime *exec_runtime_ref(ExecRuntime *r) {
2669 assert(r->n_ref > 0);
2675 ExecRuntime *exec_runtime_unref(ExecRuntime *r) {
2680 assert(r->n_ref > 0);
2683 if (r->n_ref <= 0) {
2685 free(r->var_tmp_dir);
2686 safe_close_pair(r->netns_storage_socket);
2693 int exec_runtime_serialize(ExecRuntime *rt, Unit *u, FILE *f, FDSet *fds) {
2702 unit_serialize_item(u, f, "tmp-dir", rt->tmp_dir);
2704 if (rt->var_tmp_dir)
2705 unit_serialize_item(u, f, "var-tmp-dir", rt->var_tmp_dir);
2707 if (rt->netns_storage_socket[0] >= 0) {
2710 copy = fdset_put_dup(fds, rt->netns_storage_socket[0]);
2714 unit_serialize_item_format(u, f, "netns-socket-0", "%i", copy);
2717 if (rt->netns_storage_socket[1] >= 0) {
2720 copy = fdset_put_dup(fds, rt->netns_storage_socket[1]);
2724 unit_serialize_item_format(u, f, "netns-socket-1", "%i", copy);
2730 int exec_runtime_deserialize_item(ExecRuntime **rt, Unit *u, const char *key, const char *value, FDSet *fds) {
2737 if (streq(key, "tmp-dir")) {
2740 r = exec_runtime_allocate(rt);
2744 copy = strdup(value);
2748 free((*rt)->tmp_dir);
2749 (*rt)->tmp_dir = copy;
2751 } else if (streq(key, "var-tmp-dir")) {
2754 r = exec_runtime_allocate(rt);
2758 copy = strdup(value);
2762 free((*rt)->var_tmp_dir);
2763 (*rt)->var_tmp_dir = copy;
2765 } else if (streq(key, "netns-socket-0")) {
2768 r = exec_runtime_allocate(rt);
2772 if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
2773 log_unit_debug(u->id, "Failed to parse netns socket value %s", value);
2775 safe_close((*rt)->netns_storage_socket[0]);
2776 (*rt)->netns_storage_socket[0] = fdset_remove(fds, fd);
2778 } else if (streq(key, "netns-socket-1")) {
2781 r = exec_runtime_allocate(rt);
2785 if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
2786 log_unit_debug(u->id, "Failed to parse netns socket value %s", value);
2788 safe_close((*rt)->netns_storage_socket[1]);
2789 (*rt)->netns_storage_socket[1] = fdset_remove(fds, fd);
2797 static void *remove_tmpdir_thread(void *p) {
2798 _cleanup_free_ char *path = p;
2800 rm_rf_dangerous(path, false, true, false);
2804 void exec_runtime_destroy(ExecRuntime *rt) {
2810 /* If there are multiple users of this, let's leave the stuff around */
2815 log_debug("Spawning thread to nuke %s", rt->tmp_dir);
2817 r = asynchronous_job(remove_tmpdir_thread, rt->tmp_dir);
2819 log_warning_errno(r, "Failed to nuke %s: %m", rt->tmp_dir);
2826 if (rt->var_tmp_dir) {
2827 log_debug("Spawning thread to nuke %s", rt->var_tmp_dir);
2829 r = asynchronous_job(remove_tmpdir_thread, rt->var_tmp_dir);
2831 log_warning_errno(r, "Failed to nuke %s: %m", rt->var_tmp_dir);
2832 free(rt->var_tmp_dir);
2835 rt->var_tmp_dir = NULL;
2838 safe_close_pair(rt->netns_storage_socket);
2841 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
2842 [EXEC_INPUT_NULL] = "null",
2843 [EXEC_INPUT_TTY] = "tty",
2844 [EXEC_INPUT_TTY_FORCE] = "tty-force",
2845 [EXEC_INPUT_TTY_FAIL] = "tty-fail",
2846 [EXEC_INPUT_SOCKET] = "socket"
2849 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
2851 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
2852 [EXEC_OUTPUT_INHERIT] = "inherit",
2853 [EXEC_OUTPUT_NULL] = "null",
2854 [EXEC_OUTPUT_TTY] = "tty",
2855 [EXEC_OUTPUT_SYSLOG] = "syslog",
2856 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
2857 [EXEC_OUTPUT_KMSG] = "kmsg",
2858 [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
2859 [EXEC_OUTPUT_JOURNAL] = "journal",
2860 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE] = "journal+console",
2861 [EXEC_OUTPUT_SOCKET] = "socket"
2864 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);