1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
4 This file is part of systemd.
6 Copyright 2010 Lennart Poettering
8 systemd is free software; you can redistribute it and/or modify it
9 under the terms of the GNU Lesser General Public License as published by
10 the Free Software Foundation; either version 2.1 of the License, or
11 (at your option) any later version.
13 systemd is distributed in the hope that it will be useful, but
14 WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 Lesser General Public License for more details.
18 You should have received a copy of the GNU Lesser General Public License
19 along with systemd; If not, see <http://www.gnu.org/licenses/>.
29 #include <sys/socket.h>
31 #include <sys/prctl.h>
32 #include <linux/sched.h>
33 #include <sys/types.h>
37 #include <sys/mount.h>
39 #include <linux/oom.h>
41 #include <linux/seccomp-bpf.h>
46 #include <security/pam_appl.h>
52 #include "capability.h"
55 #include "sd-messages.h"
57 #include "securebits.h"
59 #include "namespace.h"
61 #include "exit-status.h"
63 #include "utmp-wtmp.h"
65 #include "loopback-setup.h"
66 #include "path-util.h"
67 #include "syscall-list.h"
71 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
73 /* This assumes there is a 'tty' group */
76 static int shift_fds(int fds[], unsigned n_fds) {
77 int start, restart_from;
82 /* Modifies the fds array! (sorts it) */
92 for (i = start; i < (int) n_fds; i++) {
95 /* Already at right index? */
99 if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
102 close_nointr_nofail(fds[i]);
105 /* Hmm, the fd we wanted isn't free? Then
106 * let's remember that and try again from here*/
107 if (nfd != i+3 && restart_from < 0)
111 if (restart_from < 0)
114 start = restart_from;
120 static int flags_fds(const int fds[], unsigned n_fds, bool nonblock) {
129 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
131 for (i = 0; i < n_fds; i++) {
133 if ((r = fd_nonblock(fds[i], nonblock)) < 0)
136 /* We unconditionally drop FD_CLOEXEC from the fds,
137 * since after all we want to pass these fds to our
140 if ((r = fd_cloexec(fds[i], false)) < 0)
147 static const char *tty_path(const ExecContext *context) {
150 if (context->tty_path)
151 return context->tty_path;
153 return "/dev/console";
156 void exec_context_tty_reset(const ExecContext *context) {
159 if (context->tty_vhangup)
160 terminal_vhangup(tty_path(context));
162 if (context->tty_reset)
163 reset_terminal(tty_path(context));
165 if (context->tty_vt_disallocate && context->tty_path)
166 vt_disallocate(context->tty_path);
169 static bool is_terminal_output(ExecOutput o) {
171 o == EXEC_OUTPUT_TTY ||
172 o == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
173 o == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
174 o == EXEC_OUTPUT_JOURNAL_AND_CONSOLE;
177 void exec_context_serialize(const ExecContext *context, Unit *u, FILE *f) {
182 if (context->tmp_dir)
183 unit_serialize_item(u, f, "tmp-dir", context->tmp_dir);
185 if (context->var_tmp_dir)
186 unit_serialize_item(u, f, "var-tmp-dir", context->var_tmp_dir);
189 static int open_null_as(int flags, int nfd) {
194 if ((fd = open("/dev/null", flags|O_NOCTTY)) < 0)
198 r = dup2(fd, nfd) < 0 ? -errno : nfd;
199 close_nointr_nofail(fd);
206 static int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, const char *unit_id, int nfd) {
208 union sockaddr_union sa;
211 assert(output < _EXEC_OUTPUT_MAX);
215 fd = socket(AF_UNIX, SOCK_STREAM, 0);
220 sa.un.sun_family = AF_UNIX;
221 strncpy(sa.un.sun_path, "/run/systemd/journal/stdout", sizeof(sa.un.sun_path));
223 r = connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path));
225 close_nointr_nofail(fd);
229 if (shutdown(fd, SHUT_RD) < 0) {
230 close_nointr_nofail(fd);
242 context->syslog_identifier ? context->syslog_identifier : ident,
244 context->syslog_priority,
245 !!context->syslog_level_prefix,
246 output == EXEC_OUTPUT_SYSLOG || output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE,
247 output == EXEC_OUTPUT_KMSG || output == EXEC_OUTPUT_KMSG_AND_CONSOLE,
248 is_terminal_output(output));
251 r = dup2(fd, nfd) < 0 ? -errno : nfd;
252 close_nointr_nofail(fd);
258 static int open_terminal_as(const char *path, mode_t mode, int nfd) {
264 if ((fd = open_terminal(path, mode | O_NOCTTY)) < 0)
268 r = dup2(fd, nfd) < 0 ? -errno : nfd;
269 close_nointr_nofail(fd);
276 static bool is_terminal_input(ExecInput i) {
278 i == EXEC_INPUT_TTY ||
279 i == EXEC_INPUT_TTY_FORCE ||
280 i == EXEC_INPUT_TTY_FAIL;
283 static int fixup_input(ExecInput std_input, int socket_fd, bool apply_tty_stdin) {
285 if (is_terminal_input(std_input) && !apply_tty_stdin)
286 return EXEC_INPUT_NULL;
288 if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0)
289 return EXEC_INPUT_NULL;
294 static int fixup_output(ExecOutput std_output, int socket_fd) {
296 if (std_output == EXEC_OUTPUT_SOCKET && socket_fd < 0)
297 return EXEC_OUTPUT_INHERIT;
302 static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty_stdin) {
307 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
311 case EXEC_INPUT_NULL:
312 return open_null_as(O_RDONLY, STDIN_FILENO);
315 case EXEC_INPUT_TTY_FORCE:
316 case EXEC_INPUT_TTY_FAIL: {
319 if ((fd = acquire_terminal(
321 i == EXEC_INPUT_TTY_FAIL,
322 i == EXEC_INPUT_TTY_FORCE,
327 if (fd != STDIN_FILENO) {
328 r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
329 close_nointr_nofail(fd);
336 case EXEC_INPUT_SOCKET:
337 return dup2(socket_fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
340 assert_not_reached("Unknown input type");
344 static int setup_output(const ExecContext *context, int fileno, int socket_fd, const char *ident, const char *unit_id, bool apply_tty_stdin) {
352 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
353 o = fixup_output(context->std_output, socket_fd);
355 if (fileno == STDERR_FILENO) {
357 e = fixup_output(context->std_error, socket_fd);
359 /* This expects the input and output are already set up */
361 /* Don't change the stderr file descriptor if we inherit all
362 * the way and are not on a tty */
363 if (e == EXEC_OUTPUT_INHERIT &&
364 o == EXEC_OUTPUT_INHERIT &&
365 i == EXEC_INPUT_NULL &&
366 !is_terminal_input(context->std_input) &&
370 /* Duplicate from stdout if possible */
371 if (e == o || e == EXEC_OUTPUT_INHERIT)
372 return dup2(STDOUT_FILENO, fileno) < 0 ? -errno : fileno;
376 } else if (o == EXEC_OUTPUT_INHERIT) {
377 /* If input got downgraded, inherit the original value */
378 if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input))
379 return open_terminal_as(tty_path(context), O_WRONLY, fileno);
381 /* If the input is connected to anything that's not a /dev/null, inherit that... */
382 if (i != EXEC_INPUT_NULL)
383 return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
385 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
389 /* We need to open /dev/null here anew, to get the right access mode. */
390 return open_null_as(O_WRONLY, fileno);
395 case EXEC_OUTPUT_NULL:
396 return open_null_as(O_WRONLY, fileno);
398 case EXEC_OUTPUT_TTY:
399 if (is_terminal_input(i))
400 return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
402 /* We don't reset the terminal if this is just about output */
403 return open_terminal_as(tty_path(context), O_WRONLY, fileno);
405 case EXEC_OUTPUT_SYSLOG:
406 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
407 case EXEC_OUTPUT_KMSG:
408 case EXEC_OUTPUT_KMSG_AND_CONSOLE:
409 case EXEC_OUTPUT_JOURNAL:
410 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE:
411 r = connect_logger_as(context, o, ident, unit_id, fileno);
413 log_struct_unit(LOG_CRIT, unit_id,
414 "MESSAGE=Failed to connect std%s of %s to the journal socket: %s",
415 fileno == STDOUT_FILENO ? "out" : "err",
416 unit_id, strerror(-r),
419 r = open_null_as(O_WRONLY, fileno);
423 case EXEC_OUTPUT_SOCKET:
424 assert(socket_fd >= 0);
425 return dup2(socket_fd, fileno) < 0 ? -errno : fileno;
428 assert_not_reached("Unknown error type");
432 static int chown_terminal(int fd, uid_t uid) {
437 /* This might fail. What matters are the results. */
438 (void) fchown(fd, uid, -1);
439 (void) fchmod(fd, TTY_MODE);
441 if (fstat(fd, &st) < 0)
444 if (st.st_uid != uid || (st.st_mode & 0777) != TTY_MODE)
450 static int setup_confirm_stdio(int *_saved_stdin,
451 int *_saved_stdout) {
452 int fd = -1, saved_stdin, saved_stdout = -1, r;
454 assert(_saved_stdin);
455 assert(_saved_stdout);
457 saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3);
461 saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3);
462 if (saved_stdout < 0) {
467 fd = acquire_terminal(
472 DEFAULT_CONFIRM_USEC);
478 r = chown_terminal(fd, getuid());
482 if (dup2(fd, STDIN_FILENO) < 0) {
487 if (dup2(fd, STDOUT_FILENO) < 0) {
493 close_nointr_nofail(fd);
495 *_saved_stdin = saved_stdin;
496 *_saved_stdout = saved_stdout;
501 if (saved_stdout >= 0)
502 close_nointr_nofail(saved_stdout);
504 if (saved_stdin >= 0)
505 close_nointr_nofail(saved_stdin);
508 close_nointr_nofail(fd);
513 static int write_confirm_message(const char *format, ...) {
519 fd = open_terminal("/dev/console", O_WRONLY|O_NOCTTY|O_CLOEXEC);
523 va_start(ap, format);
524 vdprintf(fd, format, ap);
527 close_nointr_nofail(fd);
532 static int restore_confirm_stdio(int *saved_stdin,
538 assert(saved_stdout);
542 if (*saved_stdin >= 0)
543 if (dup2(*saved_stdin, STDIN_FILENO) < 0)
546 if (*saved_stdout >= 0)
547 if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
550 if (*saved_stdin >= 0)
551 close_nointr_nofail(*saved_stdin);
553 if (*saved_stdout >= 0)
554 close_nointr_nofail(*saved_stdout);
559 static int ask_for_confirmation(char *response, char **argv) {
560 int saved_stdout = -1, saved_stdin = -1, r;
563 r = setup_confirm_stdio(&saved_stdin, &saved_stdout);
567 line = exec_command_line(argv);
571 r = ask(response, "yns", "Execute %s? [Yes, No, Skip] ", line);
574 restore_confirm_stdio(&saved_stdin, &saved_stdout);
579 static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
580 bool keep_groups = false;
585 /* Lookup and set GID and supplementary group list. Here too
586 * we avoid NSS lookups for gid=0. */
588 if (context->group || username) {
590 if (context->group) {
591 const char *g = context->group;
593 if ((r = get_group_creds(&g, &gid)) < 0)
597 /* First step, initialize groups from /etc/groups */
598 if (username && gid != 0) {
599 if (initgroups(username, gid) < 0)
605 /* Second step, set our gids */
606 if (setresgid(gid, gid, gid) < 0)
610 if (context->supplementary_groups) {
615 /* Final step, initialize any manually set supplementary groups */
616 assert_se((ngroups_max = (int) sysconf(_SC_NGROUPS_MAX)) > 0);
618 if (!(gids = new(gid_t, ngroups_max)))
622 if ((k = getgroups(ngroups_max, gids)) < 0) {
629 STRV_FOREACH(i, context->supplementary_groups) {
632 if (k >= ngroups_max) {
638 r = get_group_creds(&g, gids+k);
647 if (setgroups(k, gids) < 0) {
658 static int enforce_user(const ExecContext *context, uid_t uid) {
662 /* Sets (but doesn't lookup) the uid and make sure we keep the
663 * capabilities while doing so. */
665 if (context->capabilities) {
667 static const cap_value_t bits[] = {
668 CAP_SETUID, /* Necessary so that we can run setresuid() below */
669 CAP_SETPCAP /* Necessary so that we can set PR_SET_SECUREBITS later on */
672 /* First step: If we need to keep capabilities but
673 * drop privileges we need to make sure we keep our
674 * caps, while we drop privileges. */
676 int sb = context->secure_bits | 1<<SECURE_KEEP_CAPS;
678 if (prctl(PR_GET_SECUREBITS) != sb)
679 if (prctl(PR_SET_SECUREBITS, sb) < 0)
683 /* Second step: set the capabilities. This will reduce
684 * the capabilities to the minimum we need. */
686 if (!(d = cap_dup(context->capabilities)))
689 if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
690 cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0) {
696 if (cap_set_proc(d) < 0) {
705 /* Third step: actually set the uids */
706 if (setresuid(uid, uid, uid) < 0)
709 /* At this point we should have all necessary capabilities but
710 are otherwise a normal user. However, the caps might got
711 corrupted due to the setresuid() so we need clean them up
712 later. This is done outside of this call. */
719 static int null_conv(
721 const struct pam_message **msg,
722 struct pam_response **resp,
725 /* We don't support conversations */
730 static int setup_pam(
736 int fds[], unsigned n_fds) {
738 static const struct pam_conv conv = {
743 pam_handle_t *handle = NULL;
745 int pam_code = PAM_SUCCESS;
748 bool close_session = false;
749 pid_t pam_pid = 0, parent_pid;
755 /* We set up PAM in the parent process, then fork. The child
756 * will then stay around until killed via PR_GET_PDEATHSIG or
757 * systemd via the cgroup logic. It will then remove the PAM
758 * session again. The parent process will exec() the actual
759 * daemon. We do things this way to ensure that the main PID
760 * of the daemon is the one we initially fork()ed. */
762 if ((pam_code = pam_start(name, user, &conv, &handle)) != PAM_SUCCESS) {
768 if ((pam_code = pam_set_item(handle, PAM_TTY, tty)) != PAM_SUCCESS)
771 if ((pam_code = pam_acct_mgmt(handle, PAM_SILENT)) != PAM_SUCCESS)
774 if ((pam_code = pam_open_session(handle, PAM_SILENT)) != PAM_SUCCESS)
777 close_session = true;
779 if ((!(e = pam_getenvlist(handle)))) {
780 pam_code = PAM_BUF_ERR;
784 /* Block SIGTERM, so that we know that it won't get lost in
786 if (sigemptyset(&ss) < 0 ||
787 sigaddset(&ss, SIGTERM) < 0 ||
788 sigprocmask(SIG_BLOCK, &ss, &old_ss) < 0)
791 parent_pid = getpid();
793 if ((pam_pid = fork()) < 0)
800 /* The child's job is to reset the PAM session on
803 /* This string must fit in 10 chars (i.e. the length
804 * of "/sbin/init"), to look pretty in /bin/ps */
805 rename_process("(sd-pam)");
807 /* Make sure we don't keep open the passed fds in this
808 child. We assume that otherwise only those fds are
809 open here that have been opened by PAM. */
810 close_many(fds, n_fds);
812 /* Drop privileges - we don't need any to pam_close_session
813 * and this will make PR_SET_PDEATHSIG work in most cases.
814 * If this fails, ignore the error - but expect sd-pam threads
815 * to fail to exit normally */
816 if (setresuid(uid, uid, uid) < 0)
817 log_error("Error: Failed to setresuid() in sd-pam: %s", strerror(-r));
819 /* Wait until our parent died. This will only work if
820 * the above setresuid() succeeds, otherwise the kernel
821 * will not allow unprivileged parents kill their privileged
822 * children this way. We rely on the control groups kill logic
823 * to do the rest for us. */
824 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
827 /* Check if our parent process might already have
829 if (getppid() == parent_pid) {
831 if (sigwait(&ss, &sig) < 0) {
838 assert(sig == SIGTERM);
843 /* If our parent died we'll end the session */
844 if (getppid() != parent_pid)
845 if ((pam_code = pam_close_session(handle, PAM_DATA_SILENT)) != PAM_SUCCESS)
851 pam_end(handle, pam_code | PAM_DATA_SILENT);
855 /* If the child was forked off successfully it will do all the
856 * cleanups, so forget about the handle here. */
859 /* Unblock SIGTERM again in the parent */
860 if (sigprocmask(SIG_SETMASK, &old_ss, NULL) < 0)
863 /* We close the log explicitly here, since the PAM modules
864 * might have opened it, but we don't want this fd around. */
873 if (pam_code != PAM_SUCCESS)
874 err = -EPERM; /* PAM errors do not map to errno */
880 pam_code = pam_close_session(handle, PAM_DATA_SILENT);
882 pam_end(handle, pam_code | PAM_DATA_SILENT);
890 kill(pam_pid, SIGTERM);
891 kill(pam_pid, SIGCONT);
898 static void rename_process_from_path(const char *path) {
899 char process_name[11];
903 /* This resulting string must fit in 10 chars (i.e. the length
904 * of "/sbin/init") to look pretty in /bin/ps */
906 p = path_get_file_name(path);
908 rename_process("(...)");
914 /* The end of the process name is usually more
915 * interesting, since the first bit might just be
921 process_name[0] = '(';
922 memcpy(process_name+1, p, l);
923 process_name[1+l] = ')';
924 process_name[1+l+1] = 0;
926 rename_process(process_name);
929 static int apply_seccomp(uint32_t *syscall_filter) {
930 static const struct sock_filter header[] = {
931 VALIDATE_ARCHITECTURE,
934 static const struct sock_filter footer[] = {
940 struct sock_filter *f;
941 struct sock_fprog prog;
943 assert(syscall_filter);
945 /* First: count the syscalls to check for */
946 for (i = 0, n = 0; i < syscall_max(); i++)
947 if (syscall_filter[i >> 4] & (1 << (i & 31)))
950 /* Second: build the filter program from a header the syscall
951 * matches and the footer */
952 f = alloca(sizeof(struct sock_filter) * (ELEMENTSOF(header) + 2*n + ELEMENTSOF(footer)));
953 memcpy(f, header, sizeof(header));
955 for (i = 0, n = 0; i < syscall_max(); i++)
956 if (syscall_filter[i >> 4] & (1 << (i & 31))) {
957 struct sock_filter item[] = {
958 BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, INDEX_TO_SYSCALL(i), 0, 1),
959 BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW)
962 assert_cc(ELEMENTSOF(item) == 2);
964 f[ELEMENTSOF(header) + 2*n] = item[0];
965 f[ELEMENTSOF(header) + 2*n+1] = item[1];
970 memcpy(f + (ELEMENTSOF(header) + 2*n), footer, sizeof(footer));
972 /* Third: install the filter */
974 prog.len = ELEMENTSOF(header) + ELEMENTSOF(footer) + 2*n;
976 if (prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog) < 0)
982 int exec_spawn(ExecCommand *command,
984 ExecContext *context,
985 int fds[], unsigned n_fds,
987 bool apply_permissions,
989 bool apply_tty_stdin,
991 CGroupBonding *cgroup_bondings,
992 CGroupAttribute *cgroup_attributes,
993 const char *cgroup_suffix,
1002 char _cleanup_strv_free_ **files_env = NULL;
1007 assert(fds || n_fds <= 0);
1009 if (context->std_input == EXEC_INPUT_SOCKET ||
1010 context->std_output == EXEC_OUTPUT_SOCKET ||
1011 context->std_error == EXEC_OUTPUT_SOCKET) {
1023 r = exec_context_load_environment(context, &files_env);
1025 log_struct_unit(LOG_ERR,
1027 "MESSAGE=Failed to load environment files: %s", strerror(-r),
1034 argv = command->argv;
1036 line = exec_command_line(argv);
1040 log_struct_unit(LOG_DEBUG,
1042 "MESSAGE=About to execute %s", line,
1046 r = cgroup_bonding_realize_list(cgroup_bondings);
1050 cgroup_attribute_apply_list(cgroup_attributes, cgroup_bondings);
1052 if (context->private_tmp && !context->tmp_dir && !context->var_tmp_dir) {
1053 r = setup_tmpdirs(&context->tmp_dir, &context->var_tmp_dir);
1065 const char *username = NULL, *home = NULL;
1066 uid_t uid = (uid_t) -1;
1067 gid_t gid = (gid_t) -1;
1068 char _cleanup_strv_free_ **our_env = NULL, **pam_env = NULL,
1069 **final_env = NULL, **final_argv = NULL;
1071 bool set_access = false;
1075 rename_process_from_path(command->path);
1077 /* We reset exactly these signals, since they are the
1078 * only ones we set to SIG_IGN in the main daemon. All
1079 * others we leave untouched because we set them to
1080 * SIG_DFL or a valid handler initially, both of which
1081 * will be demoted to SIG_DFL. */
1082 default_signals(SIGNALS_CRASH_HANDLER,
1083 SIGNALS_IGNORE, -1);
1085 if (context->ignore_sigpipe)
1086 ignore_signals(SIGPIPE, -1);
1088 assert_se(sigemptyset(&ss) == 0);
1089 if (sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
1091 r = EXIT_SIGNAL_MASK;
1096 if (idle_pipe[1] >= 0)
1097 close_nointr_nofail(idle_pipe[1]);
1098 if (idle_pipe[0] >= 0) {
1099 fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT_USEC);
1100 close_nointr_nofail(idle_pipe[0]);
1104 /* Close sockets very early to make sure we don't
1105 * block init reexecution because it cannot bind its
1108 err = close_all_fds(socket_fd >= 0 ? &socket_fd : fds,
1109 socket_fd >= 0 ? 1 : n_fds);
1115 if (!context->same_pgrp)
1122 if (context->tcpwrap_name) {
1124 if (!socket_tcpwrap(socket_fd, context->tcpwrap_name)) {
1130 for (i = 0; i < (int) n_fds; i++) {
1131 if (!socket_tcpwrap(fds[i], context->tcpwrap_name)) {
1139 exec_context_tty_reset(context);
1141 if (confirm_spawn) {
1144 err = ask_for_confirmation(&response, argv);
1145 if (err == -ETIMEDOUT)
1146 write_confirm_message("Confirmation question timed out, assuming positive response.\n");
1148 write_confirm_message("Couldn't ask confirmation question, assuming positive response: %s\n", strerror(-err));
1149 else if (response == 's') {
1150 write_confirm_message("Skipping execution.\n");
1154 } else if (response == 'n') {
1155 write_confirm_message("Failing execution.\n");
1161 /* If a socket is connected to STDIN/STDOUT/STDERR, we
1162 * must sure to drop O_NONBLOCK */
1164 fd_nonblock(socket_fd, false);
1166 err = setup_input(context, socket_fd, apply_tty_stdin);
1172 err = setup_output(context, STDOUT_FILENO, socket_fd, path_get_file_name(command->path), unit_id, apply_tty_stdin);
1178 err = setup_output(context, STDERR_FILENO, socket_fd, path_get_file_name(command->path), unit_id, apply_tty_stdin);
1184 if (cgroup_bondings) {
1185 err = cgroup_bonding_install_list(cgroup_bondings, 0, cgroup_suffix);
1192 if (context->oom_score_adjust_set) {
1195 snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
1198 if (write_string_file("/proc/self/oom_score_adj", t) < 0) {
1200 r = EXIT_OOM_ADJUST;
1205 if (context->nice_set)
1206 if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
1212 if (context->cpu_sched_set) {
1213 struct sched_param param;
1216 param.sched_priority = context->cpu_sched_priority;
1218 r = sched_setscheduler(0,
1219 context->cpu_sched_policy |
1220 (context->cpu_sched_reset_on_fork ?
1221 SCHED_RESET_ON_FORK : 0),
1225 r = EXIT_SETSCHEDULER;
1230 if (context->cpuset)
1231 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
1233 r = EXIT_CPUAFFINITY;
1237 if (context->ioprio_set)
1238 if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
1244 if (context->timer_slack_nsec != (nsec_t) -1)
1245 if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
1247 r = EXIT_TIMERSLACK;
1251 if (context->utmp_id)
1252 utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path);
1254 if (context->user) {
1255 username = context->user;
1256 err = get_user_creds(&username, &uid, &gid, &home, NULL);
1262 if (is_terminal_input(context->std_input)) {
1263 err = chown_terminal(STDIN_FILENO, uid);
1270 if (cgroup_bondings && context->control_group_modify) {
1271 err = cgroup_bonding_set_group_access_list(cgroup_bondings, 0755, uid, gid);
1273 err = cgroup_bonding_set_task_access_list(cgroup_bondings, 0644, uid, gid, context->control_group_persistent);
1283 if (cgroup_bondings && !set_access && context->control_group_persistent >= 0) {
1284 err = cgroup_bonding_set_task_access_list(cgroup_bondings, (mode_t) -1, (uid_t) -1, (uid_t) -1, context->control_group_persistent);
1291 if (apply_permissions) {
1292 err = enforce_groups(context, username, gid);
1299 umask(context->umask);
1302 if (apply_permissions && context->pam_name && username) {
1303 err = setup_pam(context->pam_name, username, uid, context->tty_path, &pam_env, fds, n_fds);
1310 if (context->private_network) {
1311 if (unshare(CLONE_NEWNET) < 0) {
1320 if (strv_length(context->read_write_dirs) > 0 ||
1321 strv_length(context->read_only_dirs) > 0 ||
1322 strv_length(context->inaccessible_dirs) > 0 ||
1323 context->mount_flags != 0 ||
1324 context->private_tmp) {
1325 err = setup_namespace(context->read_write_dirs,
1326 context->read_only_dirs,
1327 context->inaccessible_dirs,
1329 context->var_tmp_dir,
1330 context->private_tmp,
1331 context->mount_flags);
1339 if (context->root_directory)
1340 if (chroot(context->root_directory) < 0) {
1346 if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
1352 char _cleanup_free_ *d = NULL;
1354 if (asprintf(&d, "%s/%s",
1355 context->root_directory ? context->root_directory : "",
1356 context->working_directory ? context->working_directory : "") < 0) {
1369 /* We repeat the fd closing here, to make sure that
1370 * nothing is leaked from the PAM modules */
1371 err = close_all_fds(fds, n_fds);
1373 err = shift_fds(fds, n_fds);
1375 err = flags_fds(fds, n_fds, context->non_blocking);
1381 if (apply_permissions) {
1383 for (i = 0; i < RLIMIT_NLIMITS; i++) {
1384 if (!context->rlimit[i])
1387 if (setrlimit_closest(i, context->rlimit[i]) < 0) {
1394 if (context->capability_bounding_set_drop) {
1395 err = capability_bounding_set_drop(context->capability_bounding_set_drop, false);
1397 r = EXIT_CAPABILITIES;
1402 if (context->user) {
1403 err = enforce_user(context, uid);
1410 /* PR_GET_SECUREBITS is not privileged, while
1411 * PR_SET_SECUREBITS is. So to suppress
1412 * potential EPERMs we'll try not to call
1413 * PR_SET_SECUREBITS unless necessary. */
1414 if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
1415 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
1417 r = EXIT_SECUREBITS;
1421 if (context->capabilities)
1422 if (cap_set_proc(context->capabilities) < 0) {
1424 r = EXIT_CAPABILITIES;
1428 if (context->no_new_privileges)
1429 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
1431 r = EXIT_NO_NEW_PRIVILEGES;
1435 if (context->syscall_filter) {
1436 err = apply_seccomp(context->syscall_filter);
1444 our_env = new0(char*, 7);
1452 if (asprintf(our_env + n_env++, "LISTEN_PID=%lu", (unsigned long) getpid()) < 0 ||
1453 asprintf(our_env + n_env++, "LISTEN_FDS=%u", n_fds) < 0) {
1460 if (asprintf(our_env + n_env++, "HOME=%s", home) < 0) {
1467 if (asprintf(our_env + n_env++, "LOGNAME=%s", username) < 0 ||
1468 asprintf(our_env + n_env++, "USER=%s", username) < 0) {
1474 if (is_terminal_input(context->std_input) ||
1475 context->std_output == EXEC_OUTPUT_TTY ||
1476 context->std_error == EXEC_OUTPUT_TTY)
1477 if (!(our_env[n_env++] = strdup(default_term_for_tty(tty_path(context))))) {
1485 final_env = strv_env_merge(5,
1488 context->environment,
1498 final_argv = replace_env_argv(argv, final_env);
1505 final_env = strv_env_clean(final_env);
1507 execve(command->path, final_argv, final_env);
1514 log_struct(LOG_ERR, MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED),
1515 "EXECUTABLE=%s", command->path,
1516 "MESSAGE=Failed at step %s spawning %s: %s",
1517 exit_status_to_string(r, EXIT_STATUS_SYSTEMD),
1518 command->path, strerror(-err),
1527 log_struct_unit(LOG_DEBUG,
1529 "MESSAGE=Forked %s as %lu",
1530 command->path, (unsigned long) pid,
1533 /* We add the new process to the cgroup both in the child (so
1534 * that we can be sure that no user code is ever executed
1535 * outside of the cgroup) and in the parent (so that we can be
1536 * sure that when we kill the cgroup the process will be
1538 if (cgroup_bondings)
1539 cgroup_bonding_install_list(cgroup_bondings, pid, cgroup_suffix);
1541 exec_status_start(&command->exec_status, pid);
1547 void exec_context_init(ExecContext *c) {
1551 c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
1552 c->cpu_sched_policy = SCHED_OTHER;
1553 c->syslog_priority = LOG_DAEMON|LOG_INFO;
1554 c->syslog_level_prefix = true;
1555 c->control_group_persistent = -1;
1556 c->ignore_sigpipe = true;
1557 c->timer_slack_nsec = (nsec_t) -1;
1560 void exec_context_tmp_dirs_done(ExecContext *c) {
1561 char* dirs[] = {c->tmp_dir ? c->tmp_dir : c->var_tmp_dir,
1562 c->tmp_dir ? c->var_tmp_dir : NULL,
1566 for(dirp = dirs; *dirp; dirp++) {
1570 r = rm_rf_dangerous(*dirp, false, true, false);
1571 dir = dirname(*dirp);
1573 log_warning("Failed to remove content of temporary directory %s: %s",
1578 log_warning("Failed to remove temporary directory %s: %s",
1585 c->tmp_dir = c->var_tmp_dir = NULL;
1588 void exec_context_done(ExecContext *c, bool reloading_or_reexecuting) {
1593 strv_free(c->environment);
1594 c->environment = NULL;
1596 strv_free(c->environment_files);
1597 c->environment_files = NULL;
1599 for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1601 c->rlimit[l] = NULL;
1604 free(c->working_directory);
1605 c->working_directory = NULL;
1606 free(c->root_directory);
1607 c->root_directory = NULL;
1612 free(c->tcpwrap_name);
1613 c->tcpwrap_name = NULL;
1615 free(c->syslog_identifier);
1616 c->syslog_identifier = NULL;
1624 strv_free(c->supplementary_groups);
1625 c->supplementary_groups = NULL;
1630 if (c->capabilities) {
1631 cap_free(c->capabilities);
1632 c->capabilities = NULL;
1635 strv_free(c->read_only_dirs);
1636 c->read_only_dirs = NULL;
1638 strv_free(c->read_write_dirs);
1639 c->read_write_dirs = NULL;
1641 strv_free(c->inaccessible_dirs);
1642 c->inaccessible_dirs = NULL;
1645 CPU_FREE(c->cpuset);
1650 free(c->syscall_filter);
1651 c->syscall_filter = NULL;
1653 if (!reloading_or_reexecuting)
1654 exec_context_tmp_dirs_done(c);
1657 void exec_command_done(ExecCommand *c) {
1667 void exec_command_done_array(ExecCommand *c, unsigned n) {
1670 for (i = 0; i < n; i++)
1671 exec_command_done(c+i);
1674 void exec_command_free_list(ExecCommand *c) {
1678 LIST_REMOVE(ExecCommand, command, c, i);
1679 exec_command_done(i);
1684 void exec_command_free_array(ExecCommand **c, unsigned n) {
1687 for (i = 0; i < n; i++) {
1688 exec_command_free_list(c[i]);
1693 int exec_context_load_environment(const ExecContext *c, char ***l) {
1694 char **i, **r = NULL;
1699 STRV_FOREACH(i, c->environment_files) {
1702 bool ignore = false;
1714 if (!path_is_absolute(fn)) {
1723 /* Filename supports globbing, take all matching files */
1726 if (glob(fn, 0, NULL, &pglob) != 0) {
1732 return errno ? -errno : -EINVAL;
1734 count = pglob.gl_pathc;
1743 for (n = 0; n < count; n++) {
1744 k = load_env_file(pglob.gl_pathv[n], NULL, &p);
1759 m = strv_env_merge(2, r, p);
1779 static bool tty_may_match_dev_console(const char *tty) {
1780 char *active = NULL, *console;
1783 if (startswith(tty, "/dev/"))
1786 /* trivial identity? */
1787 if (streq(tty, "console"))
1790 console = resolve_dev_console(&active);
1791 /* if we could not resolve, assume it may */
1795 /* "tty0" means the active VC, so it may be the same sometimes */
1796 b = streq(console, tty) || (streq(console, "tty0") && tty_is_vc(tty));
1802 bool exec_context_may_touch_console(ExecContext *ec) {
1803 return (ec->tty_reset || ec->tty_vhangup || ec->tty_vt_disallocate ||
1804 is_terminal_input(ec->std_input) ||
1805 is_terminal_output(ec->std_output) ||
1806 is_terminal_output(ec->std_error)) &&
1807 tty_may_match_dev_console(tty_path(ec));
1810 static void strv_fprintf(FILE *f, char **l) {
1816 fprintf(f, " %s", *g);
1819 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
1831 "%sWorkingDirectory: %s\n"
1832 "%sRootDirectory: %s\n"
1833 "%sNonBlocking: %s\n"
1834 "%sPrivateTmp: %s\n"
1835 "%sControlGroupModify: %s\n"
1836 "%sControlGroupPersistent: %s\n"
1837 "%sPrivateNetwork: %s\n"
1838 "%sIgnoreSIGPIPE: %s\n",
1840 prefix, c->working_directory ? c->working_directory : "/",
1841 prefix, c->root_directory ? c->root_directory : "/",
1842 prefix, yes_no(c->non_blocking),
1843 prefix, yes_no(c->private_tmp),
1844 prefix, yes_no(c->control_group_modify),
1845 prefix, yes_no(c->control_group_persistent),
1846 prefix, yes_no(c->private_network),
1847 prefix, yes_no(c->ignore_sigpipe));
1849 STRV_FOREACH(e, c->environment)
1850 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
1852 STRV_FOREACH(e, c->environment_files)
1853 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
1855 if (c->tcpwrap_name)
1857 "%sTCPWrapName: %s\n",
1858 prefix, c->tcpwrap_name);
1865 if (c->oom_score_adjust_set)
1867 "%sOOMScoreAdjust: %i\n",
1868 prefix, c->oom_score_adjust);
1870 for (i = 0; i < RLIM_NLIMITS; i++)
1872 fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
1874 if (c->ioprio_set) {
1878 r = ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c->ioprio), &class_str);
1882 "%sIOSchedulingClass: %s\n"
1883 "%sIOPriority: %i\n",
1884 prefix, strna(class_str),
1885 prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
1889 if (c->cpu_sched_set) {
1893 r = sched_policy_to_string_alloc(c->cpu_sched_policy, &policy_str);
1897 "%sCPUSchedulingPolicy: %s\n"
1898 "%sCPUSchedulingPriority: %i\n"
1899 "%sCPUSchedulingResetOnFork: %s\n",
1900 prefix, strna(policy_str),
1901 prefix, c->cpu_sched_priority,
1902 prefix, yes_no(c->cpu_sched_reset_on_fork));
1907 fprintf(f, "%sCPUAffinity:", prefix);
1908 for (i = 0; i < c->cpuset_ncpus; i++)
1909 if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
1910 fprintf(f, " %i", i);
1914 if (c->timer_slack_nsec != (nsec_t) -1)
1915 fprintf(f, "%sTimerSlackNSec: %lu\n", prefix, (unsigned long)c->timer_slack_nsec);
1918 "%sStandardInput: %s\n"
1919 "%sStandardOutput: %s\n"
1920 "%sStandardError: %s\n",
1921 prefix, exec_input_to_string(c->std_input),
1922 prefix, exec_output_to_string(c->std_output),
1923 prefix, exec_output_to_string(c->std_error));
1929 "%sTTYVHangup: %s\n"
1930 "%sTTYVTDisallocate: %s\n",
1931 prefix, c->tty_path,
1932 prefix, yes_no(c->tty_reset),
1933 prefix, yes_no(c->tty_vhangup),
1934 prefix, yes_no(c->tty_vt_disallocate));
1936 if (c->std_output == EXEC_OUTPUT_SYSLOG || c->std_output == EXEC_OUTPUT_KMSG || c->std_output == EXEC_OUTPUT_JOURNAL ||
1937 c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE ||
1938 c->std_error == EXEC_OUTPUT_SYSLOG || c->std_error == EXEC_OUTPUT_KMSG || c->std_error == EXEC_OUTPUT_JOURNAL ||
1939 c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE) {
1940 char *fac_str, *lvl_str;
1943 r = log_facility_unshifted_to_string_alloc(c->syslog_priority >> 3, &fac_str);
1947 r = log_level_to_string_alloc(LOG_PRI(c->syslog_priority), &lvl_str);
1952 "%sSyslogFacility: %s\n"
1953 "%sSyslogLevel: %s\n",
1954 prefix, strna(fac_str),
1955 prefix, strna(lvl_str));
1960 if (c->capabilities) {
1962 if ((t = cap_to_text(c->capabilities, NULL))) {
1963 fprintf(f, "%sCapabilities: %s\n",
1970 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
1972 (c->secure_bits & 1<<SECURE_KEEP_CAPS) ? " keep-caps" : "",
1973 (c->secure_bits & 1<<SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
1974 (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
1975 (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
1976 (c->secure_bits & 1<<SECURE_NOROOT) ? " noroot" : "",
1977 (c->secure_bits & 1<<SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
1979 if (c->capability_bounding_set_drop) {
1981 fprintf(f, "%sCapabilityBoundingSet:", prefix);
1983 for (l = 0; l <= cap_last_cap(); l++)
1984 if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) {
1987 if ((t = cap_to_name(l))) {
1988 fprintf(f, " %s", t);
1997 fprintf(f, "%sUser: %s\n", prefix, c->user);
1999 fprintf(f, "%sGroup: %s\n", prefix, c->group);
2001 if (strv_length(c->supplementary_groups) > 0) {
2002 fprintf(f, "%sSupplementaryGroups:", prefix);
2003 strv_fprintf(f, c->supplementary_groups);
2008 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
2010 if (strv_length(c->read_write_dirs) > 0) {
2011 fprintf(f, "%sReadWriteDirs:", prefix);
2012 strv_fprintf(f, c->read_write_dirs);
2016 if (strv_length(c->read_only_dirs) > 0) {
2017 fprintf(f, "%sReadOnlyDirs:", prefix);
2018 strv_fprintf(f, c->read_only_dirs);
2022 if (strv_length(c->inaccessible_dirs) > 0) {
2023 fprintf(f, "%sInaccessibleDirs:", prefix);
2024 strv_fprintf(f, c->inaccessible_dirs);
2030 "%sUtmpIdentifier: %s\n",
2031 prefix, c->utmp_id);
2034 void exec_status_start(ExecStatus *s, pid_t pid) {
2039 dual_timestamp_get(&s->start_timestamp);
2042 void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
2045 if (s->pid && s->pid != pid)
2049 dual_timestamp_get(&s->exit_timestamp);
2055 if (context->utmp_id)
2056 utmp_put_dead_process(context->utmp_id, pid, code, status);
2058 exec_context_tty_reset(context);
2062 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
2063 char buf[FORMAT_TIMESTAMP_MAX];
2076 prefix, (unsigned long) s->pid);
2078 if (s->start_timestamp.realtime > 0)
2080 "%sStart Timestamp: %s\n",
2081 prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
2083 if (s->exit_timestamp.realtime > 0)
2085 "%sExit Timestamp: %s\n"
2087 "%sExit Status: %i\n",
2088 prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
2089 prefix, sigchld_code_to_string(s->code),
2093 char *exec_command_line(char **argv) {
2101 STRV_FOREACH(a, argv)
2104 if (!(n = new(char, k)))
2108 STRV_FOREACH(a, argv) {
2115 if (strpbrk(*a, WHITESPACE)) {
2126 /* FIXME: this doesn't really handle arguments that have
2127 * spaces and ticks in them */
2132 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
2134 const char *prefix2;
2143 p2 = strappend(prefix, "\t");
2144 prefix2 = p2 ? p2 : prefix;
2146 cmd = exec_command_line(c->argv);
2149 "%sCommand Line: %s\n",
2150 prefix, cmd ? cmd : strerror(ENOMEM));
2154 exec_status_dump(&c->exec_status, f, prefix2);
2159 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
2165 LIST_FOREACH(command, c, c)
2166 exec_command_dump(c, f, prefix);
2169 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
2176 /* It's kind of important, that we keep the order here */
2177 LIST_FIND_TAIL(ExecCommand, command, *l, end);
2178 LIST_INSERT_AFTER(ExecCommand, command, *l, end, e);
2183 int exec_command_set(ExecCommand *c, const char *path, ...) {
2191 l = strv_new_ap(path, ap);
2197 if (!(p = strdup(path))) {
2211 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
2212 [EXEC_INPUT_NULL] = "null",
2213 [EXEC_INPUT_TTY] = "tty",
2214 [EXEC_INPUT_TTY_FORCE] = "tty-force",
2215 [EXEC_INPUT_TTY_FAIL] = "tty-fail",
2216 [EXEC_INPUT_SOCKET] = "socket"
2219 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
2221 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
2222 [EXEC_OUTPUT_INHERIT] = "inherit",
2223 [EXEC_OUTPUT_NULL] = "null",
2224 [EXEC_OUTPUT_TTY] = "tty",
2225 [EXEC_OUTPUT_SYSLOG] = "syslog",
2226 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
2227 [EXEC_OUTPUT_KMSG] = "kmsg",
2228 [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
2229 [EXEC_OUTPUT_JOURNAL] = "journal",
2230 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE] = "journal+console",
2231 [EXEC_OUTPUT_SOCKET] = "socket"
2234 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);