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 = {
209 .un.sun_family = AF_UNIX,
210 .un.sun_path = "/run/systemd/journal/stdout",
214 assert(output < _EXEC_OUTPUT_MAX);
218 fd = socket(AF_UNIX, SOCK_STREAM, 0);
222 r = connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path));
224 close_nointr_nofail(fd);
228 if (shutdown(fd, SHUT_RD) < 0) {
229 close_nointr_nofail(fd);
241 context->syslog_identifier ? context->syslog_identifier : ident,
243 context->syslog_priority,
244 !!context->syslog_level_prefix,
245 output == EXEC_OUTPUT_SYSLOG || output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE,
246 output == EXEC_OUTPUT_KMSG || output == EXEC_OUTPUT_KMSG_AND_CONSOLE,
247 is_terminal_output(output));
250 r = dup2(fd, nfd) < 0 ? -errno : nfd;
251 close_nointr_nofail(fd);
257 static int open_terminal_as(const char *path, mode_t mode, int nfd) {
263 if ((fd = open_terminal(path, mode | O_NOCTTY)) < 0)
267 r = dup2(fd, nfd) < 0 ? -errno : nfd;
268 close_nointr_nofail(fd);
275 static bool is_terminal_input(ExecInput i) {
277 i == EXEC_INPUT_TTY ||
278 i == EXEC_INPUT_TTY_FORCE ||
279 i == EXEC_INPUT_TTY_FAIL;
282 static int fixup_input(ExecInput std_input, int socket_fd, bool apply_tty_stdin) {
284 if (is_terminal_input(std_input) && !apply_tty_stdin)
285 return EXEC_INPUT_NULL;
287 if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0)
288 return EXEC_INPUT_NULL;
293 static int fixup_output(ExecOutput std_output, int socket_fd) {
295 if (std_output == EXEC_OUTPUT_SOCKET && socket_fd < 0)
296 return EXEC_OUTPUT_INHERIT;
301 static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty_stdin) {
306 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
310 case EXEC_INPUT_NULL:
311 return open_null_as(O_RDONLY, STDIN_FILENO);
314 case EXEC_INPUT_TTY_FORCE:
315 case EXEC_INPUT_TTY_FAIL: {
318 if ((fd = acquire_terminal(
320 i == EXEC_INPUT_TTY_FAIL,
321 i == EXEC_INPUT_TTY_FORCE,
326 if (fd != STDIN_FILENO) {
327 r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
328 close_nointr_nofail(fd);
335 case EXEC_INPUT_SOCKET:
336 return dup2(socket_fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
339 assert_not_reached("Unknown input type");
343 static int setup_output(const ExecContext *context, int fileno, int socket_fd, const char *ident, const char *unit_id, bool apply_tty_stdin) {
351 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
352 o = fixup_output(context->std_output, socket_fd);
354 if (fileno == STDERR_FILENO) {
356 e = fixup_output(context->std_error, socket_fd);
358 /* This expects the input and output are already set up */
360 /* Don't change the stderr file descriptor if we inherit all
361 * the way and are not on a tty */
362 if (e == EXEC_OUTPUT_INHERIT &&
363 o == EXEC_OUTPUT_INHERIT &&
364 i == EXEC_INPUT_NULL &&
365 !is_terminal_input(context->std_input) &&
369 /* Duplicate from stdout if possible */
370 if (e == o || e == EXEC_OUTPUT_INHERIT)
371 return dup2(STDOUT_FILENO, fileno) < 0 ? -errno : fileno;
375 } else if (o == EXEC_OUTPUT_INHERIT) {
376 /* If input got downgraded, inherit the original value */
377 if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input))
378 return open_terminal_as(tty_path(context), O_WRONLY, fileno);
380 /* If the input is connected to anything that's not a /dev/null, inherit that... */
381 if (i != EXEC_INPUT_NULL)
382 return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
384 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
388 /* We need to open /dev/null here anew, to get the right access mode. */
389 return open_null_as(O_WRONLY, fileno);
394 case EXEC_OUTPUT_NULL:
395 return open_null_as(O_WRONLY, fileno);
397 case EXEC_OUTPUT_TTY:
398 if (is_terminal_input(i))
399 return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
401 /* We don't reset the terminal if this is just about output */
402 return open_terminal_as(tty_path(context), O_WRONLY, fileno);
404 case EXEC_OUTPUT_SYSLOG:
405 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
406 case EXEC_OUTPUT_KMSG:
407 case EXEC_OUTPUT_KMSG_AND_CONSOLE:
408 case EXEC_OUTPUT_JOURNAL:
409 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE:
410 r = connect_logger_as(context, o, ident, unit_id, fileno);
412 log_struct_unit(LOG_CRIT, unit_id,
413 "MESSAGE=Failed to connect std%s of %s to the journal socket: %s",
414 fileno == STDOUT_FILENO ? "out" : "err",
415 unit_id, strerror(-r),
418 r = open_null_as(O_WRONLY, fileno);
422 case EXEC_OUTPUT_SOCKET:
423 assert(socket_fd >= 0);
424 return dup2(socket_fd, fileno) < 0 ? -errno : fileno;
427 assert_not_reached("Unknown error type");
431 static int chown_terminal(int fd, uid_t uid) {
436 /* This might fail. What matters are the results. */
437 (void) fchown(fd, uid, -1);
438 (void) fchmod(fd, TTY_MODE);
440 if (fstat(fd, &st) < 0)
443 if (st.st_uid != uid || (st.st_mode & 0777) != TTY_MODE)
449 static int setup_confirm_stdio(int *_saved_stdin,
450 int *_saved_stdout) {
451 int fd = -1, saved_stdin, saved_stdout = -1, r;
453 assert(_saved_stdin);
454 assert(_saved_stdout);
456 saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3);
460 saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3);
461 if (saved_stdout < 0) {
466 fd = acquire_terminal(
471 DEFAULT_CONFIRM_USEC);
477 r = chown_terminal(fd, getuid());
481 if (dup2(fd, STDIN_FILENO) < 0) {
486 if (dup2(fd, STDOUT_FILENO) < 0) {
492 close_nointr_nofail(fd);
494 *_saved_stdin = saved_stdin;
495 *_saved_stdout = saved_stdout;
500 if (saved_stdout >= 0)
501 close_nointr_nofail(saved_stdout);
503 if (saved_stdin >= 0)
504 close_nointr_nofail(saved_stdin);
507 close_nointr_nofail(fd);
512 static int write_confirm_message(const char *format, ...) {
518 fd = open_terminal("/dev/console", O_WRONLY|O_NOCTTY|O_CLOEXEC);
522 va_start(ap, format);
523 vdprintf(fd, format, ap);
526 close_nointr_nofail(fd);
531 static int restore_confirm_stdio(int *saved_stdin,
537 assert(saved_stdout);
541 if (*saved_stdin >= 0)
542 if (dup2(*saved_stdin, STDIN_FILENO) < 0)
545 if (*saved_stdout >= 0)
546 if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
549 if (*saved_stdin >= 0)
550 close_nointr_nofail(*saved_stdin);
552 if (*saved_stdout >= 0)
553 close_nointr_nofail(*saved_stdout);
558 static int ask_for_confirmation(char *response, char **argv) {
559 int saved_stdout = -1, saved_stdin = -1, r;
562 r = setup_confirm_stdio(&saved_stdin, &saved_stdout);
566 line = exec_command_line(argv);
570 r = ask(response, "yns", "Execute %s? [Yes, No, Skip] ", line);
573 restore_confirm_stdio(&saved_stdin, &saved_stdout);
578 static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
579 bool keep_groups = false;
584 /* Lookup and set GID and supplementary group list. Here too
585 * we avoid NSS lookups for gid=0. */
587 if (context->group || username) {
589 if (context->group) {
590 const char *g = context->group;
592 if ((r = get_group_creds(&g, &gid)) < 0)
596 /* First step, initialize groups from /etc/groups */
597 if (username && gid != 0) {
598 if (initgroups(username, gid) < 0)
604 /* Second step, set our gids */
605 if (setresgid(gid, gid, gid) < 0)
609 if (context->supplementary_groups) {
614 /* Final step, initialize any manually set supplementary groups */
615 assert_se((ngroups_max = (int) sysconf(_SC_NGROUPS_MAX)) > 0);
617 if (!(gids = new(gid_t, ngroups_max)))
621 if ((k = getgroups(ngroups_max, gids)) < 0) {
628 STRV_FOREACH(i, context->supplementary_groups) {
631 if (k >= ngroups_max) {
637 r = get_group_creds(&g, gids+k);
646 if (setgroups(k, gids) < 0) {
657 static int enforce_user(const ExecContext *context, uid_t uid) {
661 /* Sets (but doesn't lookup) the uid and make sure we keep the
662 * capabilities while doing so. */
664 if (context->capabilities) {
666 static const cap_value_t bits[] = {
667 CAP_SETUID, /* Necessary so that we can run setresuid() below */
668 CAP_SETPCAP /* Necessary so that we can set PR_SET_SECUREBITS later on */
671 /* First step: If we need to keep capabilities but
672 * drop privileges we need to make sure we keep our
673 * caps, while we drop privileges. */
675 int sb = context->secure_bits | 1<<SECURE_KEEP_CAPS;
677 if (prctl(PR_GET_SECUREBITS) != sb)
678 if (prctl(PR_SET_SECUREBITS, sb) < 0)
682 /* Second step: set the capabilities. This will reduce
683 * the capabilities to the minimum we need. */
685 if (!(d = cap_dup(context->capabilities)))
688 if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
689 cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0) {
695 if (cap_set_proc(d) < 0) {
704 /* Third step: actually set the uids */
705 if (setresuid(uid, uid, uid) < 0)
708 /* At this point we should have all necessary capabilities but
709 are otherwise a normal user. However, the caps might got
710 corrupted due to the setresuid() so we need clean them up
711 later. This is done outside of this call. */
718 static int null_conv(
720 const struct pam_message **msg,
721 struct pam_response **resp,
724 /* We don't support conversations */
729 static int setup_pam(
735 int fds[], unsigned n_fds) {
737 static const struct pam_conv conv = {
742 pam_handle_t *handle = NULL;
744 int pam_code = PAM_SUCCESS;
747 bool close_session = false;
748 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 ((pam_code = pam_start(name, user, &conv, &handle)) != PAM_SUCCESS) {
767 if ((pam_code = pam_set_item(handle, PAM_TTY, tty)) != PAM_SUCCESS)
770 if ((pam_code = pam_acct_mgmt(handle, PAM_SILENT)) != PAM_SUCCESS)
773 if ((pam_code = pam_open_session(handle, PAM_SILENT)) != PAM_SUCCESS)
776 close_session = true;
778 if ((!(e = pam_getenvlist(handle)))) {
779 pam_code = PAM_BUF_ERR;
783 /* Block SIGTERM, so that we know that it won't get lost in
785 if (sigemptyset(&ss) < 0 ||
786 sigaddset(&ss, SIGTERM) < 0 ||
787 sigprocmask(SIG_BLOCK, &ss, &old_ss) < 0)
790 parent_pid = getpid();
792 if ((pam_pid = fork()) < 0)
799 /* The child's job is to reset the PAM session on
802 /* This string must fit in 10 chars (i.e. the length
803 * of "/sbin/init"), to look pretty in /bin/ps */
804 rename_process("(sd-pam)");
806 /* Make sure we don't keep open the passed fds in this
807 child. We assume that otherwise only those fds are
808 open here that have been opened by PAM. */
809 close_many(fds, n_fds);
811 /* Drop privileges - we don't need any to pam_close_session
812 * and this will make PR_SET_PDEATHSIG work in most cases.
813 * If this fails, ignore the error - but expect sd-pam threads
814 * to fail to exit normally */
815 if (setresuid(uid, uid, uid) < 0)
816 log_error("Error: Failed to setresuid() in sd-pam: %s", strerror(-r));
818 /* Wait until our parent died. This will only work if
819 * the above setresuid() succeeds, otherwise the kernel
820 * will not allow unprivileged parents kill their privileged
821 * children this way. We rely on the control groups kill logic
822 * to do the rest for us. */
823 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
826 /* Check if our parent process might already have
828 if (getppid() == parent_pid) {
830 if (sigwait(&ss, &sig) < 0) {
837 assert(sig == SIGTERM);
842 /* If our parent died we'll end the session */
843 if (getppid() != parent_pid)
844 if ((pam_code = pam_close_session(handle, PAM_DATA_SILENT)) != PAM_SUCCESS)
850 pam_end(handle, pam_code | PAM_DATA_SILENT);
854 /* If the child was forked off successfully it will do all the
855 * cleanups, so forget about the handle here. */
858 /* Unblock SIGTERM again in the parent */
859 if (sigprocmask(SIG_SETMASK, &old_ss, NULL) < 0)
862 /* We close the log explicitly here, since the PAM modules
863 * might have opened it, but we don't want this fd around. */
872 if (pam_code != PAM_SUCCESS)
873 err = -EPERM; /* PAM errors do not map to errno */
879 pam_code = pam_close_session(handle, PAM_DATA_SILENT);
881 pam_end(handle, pam_code | PAM_DATA_SILENT);
889 kill(pam_pid, SIGTERM);
890 kill(pam_pid, SIGCONT);
897 static void rename_process_from_path(const char *path) {
898 char process_name[11];
902 /* This resulting string must fit in 10 chars (i.e. the length
903 * of "/sbin/init") to look pretty in /bin/ps */
905 p = path_get_file_name(path);
907 rename_process("(...)");
913 /* The end of the process name is usually more
914 * interesting, since the first bit might just be
920 process_name[0] = '(';
921 memcpy(process_name+1, p, l);
922 process_name[1+l] = ')';
923 process_name[1+l+1] = 0;
925 rename_process(process_name);
928 static int apply_seccomp(uint32_t *syscall_filter) {
929 static const struct sock_filter header[] = {
930 VALIDATE_ARCHITECTURE,
933 static const struct sock_filter footer[] = {
939 struct sock_filter *f;
940 struct sock_fprog prog = {};
942 assert(syscall_filter);
944 /* First: count the syscalls to check for */
945 for (i = 0, n = 0; i < syscall_max(); i++)
946 if (syscall_filter[i >> 4] & (1 << (i & 31)))
949 /* Second: build the filter program from a header the syscall
950 * matches and the footer */
951 f = alloca(sizeof(struct sock_filter) * (ELEMENTSOF(header) + 2*n + ELEMENTSOF(footer)));
952 memcpy(f, header, sizeof(header));
954 for (i = 0, n = 0; i < syscall_max(); i++)
955 if (syscall_filter[i >> 4] & (1 << (i & 31))) {
956 struct sock_filter item[] = {
957 BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, INDEX_TO_SYSCALL(i), 0, 1),
958 BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW)
961 assert_cc(ELEMENTSOF(item) == 2);
963 f[ELEMENTSOF(header) + 2*n] = item[0];
964 f[ELEMENTSOF(header) + 2*n+1] = item[1];
969 memcpy(f + (ELEMENTSOF(header) + 2*n), footer, sizeof(footer));
971 /* Third: install the filter */
972 prog.len = ELEMENTSOF(header) + ELEMENTSOF(footer) + 2*n;
974 if (prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog) < 0)
980 int exec_spawn(ExecCommand *command,
982 ExecContext *context,
983 int fds[], unsigned n_fds,
985 bool apply_permissions,
987 bool apply_tty_stdin,
989 CGroupBonding *cgroup_bondings,
990 CGroupAttribute *cgroup_attributes,
991 const char *cgroup_suffix,
1000 char _cleanup_strv_free_ **files_env = NULL;
1005 assert(fds || n_fds <= 0);
1007 if (context->std_input == EXEC_INPUT_SOCKET ||
1008 context->std_output == EXEC_OUTPUT_SOCKET ||
1009 context->std_error == EXEC_OUTPUT_SOCKET) {
1021 r = exec_context_load_environment(context, &files_env);
1023 log_struct_unit(LOG_ERR,
1025 "MESSAGE=Failed to load environment files: %s", strerror(-r),
1032 argv = command->argv;
1034 line = exec_command_line(argv);
1038 log_struct_unit(LOG_DEBUG,
1040 "MESSAGE=About to execute %s", line,
1044 r = cgroup_bonding_realize_list(cgroup_bondings);
1048 /* We must initialize the attributes in the parent, before we
1049 fork, because we really need them initialized before making
1050 the process a member of the group (which we do in both the
1051 child and the parent), and we cannot really apply them twice
1052 (due to 'append' style attributes) */
1053 cgroup_attribute_apply_list(cgroup_attributes, cgroup_bondings);
1055 if (context->private_tmp && !context->tmp_dir && !context->var_tmp_dir) {
1056 r = setup_tmpdirs(&context->tmp_dir, &context->var_tmp_dir);
1068 const char *username = NULL, *home = NULL;
1069 uid_t uid = (uid_t) -1;
1070 gid_t gid = (gid_t) -1;
1071 char _cleanup_strv_free_ **our_env = NULL, **pam_env = NULL,
1072 **final_env = NULL, **final_argv = NULL;
1074 bool set_access = false;
1078 rename_process_from_path(command->path);
1080 /* We reset exactly these signals, since they are the
1081 * only ones we set to SIG_IGN in the main daemon. All
1082 * others we leave untouched because we set them to
1083 * SIG_DFL or a valid handler initially, both of which
1084 * will be demoted to SIG_DFL. */
1085 default_signals(SIGNALS_CRASH_HANDLER,
1086 SIGNALS_IGNORE, -1);
1088 if (context->ignore_sigpipe)
1089 ignore_signals(SIGPIPE, -1);
1091 assert_se(sigemptyset(&ss) == 0);
1092 if (sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
1094 r = EXIT_SIGNAL_MASK;
1099 if (idle_pipe[1] >= 0)
1100 close_nointr_nofail(idle_pipe[1]);
1101 if (idle_pipe[0] >= 0) {
1102 fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT_USEC);
1103 close_nointr_nofail(idle_pipe[0]);
1107 /* Close sockets very early to make sure we don't
1108 * block init reexecution because it cannot bind its
1111 err = close_all_fds(socket_fd >= 0 ? &socket_fd : fds,
1112 socket_fd >= 0 ? 1 : n_fds);
1118 if (!context->same_pgrp)
1125 if (context->tcpwrap_name) {
1127 if (!socket_tcpwrap(socket_fd, context->tcpwrap_name)) {
1133 for (i = 0; i < (int) n_fds; i++) {
1134 if (!socket_tcpwrap(fds[i], context->tcpwrap_name)) {
1142 exec_context_tty_reset(context);
1144 if (confirm_spawn) {
1147 err = ask_for_confirmation(&response, argv);
1148 if (err == -ETIMEDOUT)
1149 write_confirm_message("Confirmation question timed out, assuming positive response.\n");
1151 write_confirm_message("Couldn't ask confirmation question, assuming positive response: %s\n", strerror(-err));
1152 else if (response == 's') {
1153 write_confirm_message("Skipping execution.\n");
1157 } else if (response == 'n') {
1158 write_confirm_message("Failing execution.\n");
1164 /* If a socket is connected to STDIN/STDOUT/STDERR, we
1165 * must sure to drop O_NONBLOCK */
1167 fd_nonblock(socket_fd, false);
1169 err = setup_input(context, socket_fd, apply_tty_stdin);
1175 err = setup_output(context, STDOUT_FILENO, socket_fd, path_get_file_name(command->path), unit_id, apply_tty_stdin);
1181 err = setup_output(context, STDERR_FILENO, socket_fd, path_get_file_name(command->path), unit_id, apply_tty_stdin);
1187 if (cgroup_bondings) {
1188 err = cgroup_bonding_install_list(cgroup_bondings, 0, cgroup_suffix);
1195 if (context->oom_score_adjust_set) {
1198 snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
1201 if (write_string_file("/proc/self/oom_score_adj", t) < 0) {
1203 r = EXIT_OOM_ADJUST;
1208 if (context->nice_set)
1209 if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
1215 if (context->cpu_sched_set) {
1216 struct sched_param param = {
1217 .sched_priority = context->cpu_sched_priority,
1220 r = sched_setscheduler(0,
1221 context->cpu_sched_policy |
1222 (context->cpu_sched_reset_on_fork ?
1223 SCHED_RESET_ON_FORK : 0),
1227 r = EXIT_SETSCHEDULER;
1232 if (context->cpuset)
1233 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
1235 r = EXIT_CPUAFFINITY;
1239 if (context->ioprio_set)
1240 if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
1246 if (context->timer_slack_nsec != (nsec_t) -1)
1247 if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
1249 r = EXIT_TIMERSLACK;
1253 if (context->utmp_id)
1254 utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path);
1256 if (context->user) {
1257 username = context->user;
1258 err = get_user_creds(&username, &uid, &gid, &home, NULL);
1264 if (is_terminal_input(context->std_input)) {
1265 err = chown_terminal(STDIN_FILENO, uid);
1272 if (cgroup_bondings && context->control_group_modify) {
1273 err = cgroup_bonding_set_group_access_list(cgroup_bondings, 0755, uid, gid);
1275 err = cgroup_bonding_set_task_access_list(
1280 context->control_group_persistent);
1290 if (cgroup_bondings && !set_access && context->control_group_persistent >= 0) {
1291 err = cgroup_bonding_set_task_access_list(
1296 context->control_group_persistent);
1303 if (apply_permissions) {
1304 err = enforce_groups(context, username, gid);
1311 umask(context->umask);
1314 if (apply_permissions && context->pam_name && username) {
1315 err = setup_pam(context->pam_name, username, uid, context->tty_path, &pam_env, fds, n_fds);
1322 if (context->private_network) {
1323 if (unshare(CLONE_NEWNET) < 0) {
1332 if (strv_length(context->read_write_dirs) > 0 ||
1333 strv_length(context->read_only_dirs) > 0 ||
1334 strv_length(context->inaccessible_dirs) > 0 ||
1335 context->mount_flags != 0 ||
1336 context->private_tmp) {
1337 err = setup_namespace(context->read_write_dirs,
1338 context->read_only_dirs,
1339 context->inaccessible_dirs,
1341 context->var_tmp_dir,
1342 context->private_tmp,
1343 context->mount_flags);
1351 if (context->root_directory)
1352 if (chroot(context->root_directory) < 0) {
1358 if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
1364 char _cleanup_free_ *d = NULL;
1366 if (asprintf(&d, "%s/%s",
1367 context->root_directory ? context->root_directory : "",
1368 context->working_directory ? context->working_directory : "") < 0) {
1381 /* We repeat the fd closing here, to make sure that
1382 * nothing is leaked from the PAM modules */
1383 err = close_all_fds(fds, n_fds);
1385 err = shift_fds(fds, n_fds);
1387 err = flags_fds(fds, n_fds, context->non_blocking);
1393 if (apply_permissions) {
1395 for (i = 0; i < RLIMIT_NLIMITS; i++) {
1396 if (!context->rlimit[i])
1399 if (setrlimit_closest(i, context->rlimit[i]) < 0) {
1406 if (context->capability_bounding_set_drop) {
1407 err = capability_bounding_set_drop(context->capability_bounding_set_drop, false);
1409 r = EXIT_CAPABILITIES;
1414 if (context->user) {
1415 err = enforce_user(context, uid);
1422 /* PR_GET_SECUREBITS is not privileged, while
1423 * PR_SET_SECUREBITS is. So to suppress
1424 * potential EPERMs we'll try not to call
1425 * PR_SET_SECUREBITS unless necessary. */
1426 if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
1427 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
1429 r = EXIT_SECUREBITS;
1433 if (context->capabilities)
1434 if (cap_set_proc(context->capabilities) < 0) {
1436 r = EXIT_CAPABILITIES;
1440 if (context->no_new_privileges)
1441 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
1443 r = EXIT_NO_NEW_PRIVILEGES;
1447 if (context->syscall_filter) {
1448 err = apply_seccomp(context->syscall_filter);
1456 our_env = new0(char*, 7);
1464 if (asprintf(our_env + n_env++, "LISTEN_PID=%lu", (unsigned long) getpid()) < 0 ||
1465 asprintf(our_env + n_env++, "LISTEN_FDS=%u", n_fds) < 0) {
1472 if (asprintf(our_env + n_env++, "HOME=%s", home) < 0) {
1479 if (asprintf(our_env + n_env++, "LOGNAME=%s", username) < 0 ||
1480 asprintf(our_env + n_env++, "USER=%s", username) < 0) {
1486 if (is_terminal_input(context->std_input) ||
1487 context->std_output == EXEC_OUTPUT_TTY ||
1488 context->std_error == EXEC_OUTPUT_TTY)
1489 if (!(our_env[n_env++] = strdup(default_term_for_tty(tty_path(context))))) {
1497 final_env = strv_env_merge(5,
1500 context->environment,
1510 final_argv = replace_env_argv(argv, final_env);
1517 final_env = strv_env_clean(final_env);
1519 execve(command->path, final_argv, final_env);
1526 log_struct(LOG_ERR, MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED),
1527 "EXECUTABLE=%s", command->path,
1528 "MESSAGE=Failed at step %s spawning %s: %s",
1529 exit_status_to_string(r, EXIT_STATUS_SYSTEMD),
1530 command->path, strerror(-err),
1539 log_struct_unit(LOG_DEBUG,
1541 "MESSAGE=Forked %s as %lu",
1542 command->path, (unsigned long) pid,
1545 /* We add the new process to the cgroup both in the child (so
1546 * that we can be sure that no user code is ever executed
1547 * outside of the cgroup) and in the parent (so that we can be
1548 * sure that when we kill the cgroup the process will be
1550 cgroup_bonding_install_list(cgroup_bondings, pid, cgroup_suffix);
1552 exec_status_start(&command->exec_status, pid);
1558 void exec_context_init(ExecContext *c) {
1562 c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
1563 c->cpu_sched_policy = SCHED_OTHER;
1564 c->syslog_priority = LOG_DAEMON|LOG_INFO;
1565 c->syslog_level_prefix = true;
1566 c->control_group_persistent = -1;
1567 c->ignore_sigpipe = true;
1568 c->timer_slack_nsec = (nsec_t) -1;
1571 void exec_context_tmp_dirs_done(ExecContext *c) {
1572 char* dirs[] = {c->tmp_dir ? c->tmp_dir : c->var_tmp_dir,
1573 c->tmp_dir ? c->var_tmp_dir : NULL,
1577 for(dirp = dirs; *dirp; dirp++) {
1581 r = rm_rf_dangerous(*dirp, false, true, false);
1582 dir = dirname(*dirp);
1584 log_warning("Failed to remove content of temporary directory %s: %s",
1589 log_warning("Failed to remove temporary directory %s: %s",
1596 c->tmp_dir = c->var_tmp_dir = NULL;
1599 void exec_context_done(ExecContext *c, bool reloading_or_reexecuting) {
1604 strv_free(c->environment);
1605 c->environment = NULL;
1607 strv_free(c->environment_files);
1608 c->environment_files = NULL;
1610 for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1612 c->rlimit[l] = NULL;
1615 free(c->working_directory);
1616 c->working_directory = NULL;
1617 free(c->root_directory);
1618 c->root_directory = NULL;
1623 free(c->tcpwrap_name);
1624 c->tcpwrap_name = NULL;
1626 free(c->syslog_identifier);
1627 c->syslog_identifier = NULL;
1635 strv_free(c->supplementary_groups);
1636 c->supplementary_groups = NULL;
1641 if (c->capabilities) {
1642 cap_free(c->capabilities);
1643 c->capabilities = NULL;
1646 strv_free(c->read_only_dirs);
1647 c->read_only_dirs = NULL;
1649 strv_free(c->read_write_dirs);
1650 c->read_write_dirs = NULL;
1652 strv_free(c->inaccessible_dirs);
1653 c->inaccessible_dirs = NULL;
1656 CPU_FREE(c->cpuset);
1661 free(c->syscall_filter);
1662 c->syscall_filter = NULL;
1664 if (!reloading_or_reexecuting)
1665 exec_context_tmp_dirs_done(c);
1668 void exec_command_done(ExecCommand *c) {
1678 void exec_command_done_array(ExecCommand *c, unsigned n) {
1681 for (i = 0; i < n; i++)
1682 exec_command_done(c+i);
1685 void exec_command_free_list(ExecCommand *c) {
1689 LIST_REMOVE(ExecCommand, command, c, i);
1690 exec_command_done(i);
1695 void exec_command_free_array(ExecCommand **c, unsigned n) {
1698 for (i = 0; i < n; i++) {
1699 exec_command_free_list(c[i]);
1704 int exec_context_load_environment(const ExecContext *c, char ***l) {
1705 char **i, **r = NULL;
1710 STRV_FOREACH(i, c->environment_files) {
1713 bool ignore = false;
1715 glob_t _cleanup_globfree_ pglob = {};
1725 if (!path_is_absolute(fn)) {
1733 /* Filename supports globbing, take all matching files */
1735 if (glob(fn, 0, NULL, &pglob) != 0) {
1740 return errno ? -errno : -EINVAL;
1742 count = pglob.gl_pathc;
1750 for (n = 0; n < count; n++) {
1751 k = load_env_file(pglob.gl_pathv[n], NULL, &p);
1765 m = strv_env_merge(2, r, p);
1781 static bool tty_may_match_dev_console(const char *tty) {
1782 char *active = NULL, *console;
1785 if (startswith(tty, "/dev/"))
1788 /* trivial identity? */
1789 if (streq(tty, "console"))
1792 console = resolve_dev_console(&active);
1793 /* if we could not resolve, assume it may */
1797 /* "tty0" means the active VC, so it may be the same sometimes */
1798 b = streq(console, tty) || (streq(console, "tty0") && tty_is_vc(tty));
1804 bool exec_context_may_touch_console(ExecContext *ec) {
1805 return (ec->tty_reset || ec->tty_vhangup || ec->tty_vt_disallocate ||
1806 is_terminal_input(ec->std_input) ||
1807 is_terminal_output(ec->std_output) ||
1808 is_terminal_output(ec->std_error)) &&
1809 tty_may_match_dev_console(tty_path(ec));
1812 static void strv_fprintf(FILE *f, char **l) {
1818 fprintf(f, " %s", *g);
1821 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
1833 "%sWorkingDirectory: %s\n"
1834 "%sRootDirectory: %s\n"
1835 "%sNonBlocking: %s\n"
1836 "%sPrivateTmp: %s\n"
1837 "%sControlGroupModify: %s\n"
1838 "%sControlGroupPersistent: %s\n"
1839 "%sPrivateNetwork: %s\n"
1840 "%sIgnoreSIGPIPE: %s\n",
1842 prefix, c->working_directory ? c->working_directory : "/",
1843 prefix, c->root_directory ? c->root_directory : "/",
1844 prefix, yes_no(c->non_blocking),
1845 prefix, yes_no(c->private_tmp),
1846 prefix, yes_no(c->control_group_modify),
1847 prefix, yes_no(c->control_group_persistent),
1848 prefix, yes_no(c->private_network),
1849 prefix, yes_no(c->ignore_sigpipe));
1851 STRV_FOREACH(e, c->environment)
1852 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
1854 STRV_FOREACH(e, c->environment_files)
1855 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
1857 if (c->tcpwrap_name)
1859 "%sTCPWrapName: %s\n",
1860 prefix, c->tcpwrap_name);
1867 if (c->oom_score_adjust_set)
1869 "%sOOMScoreAdjust: %i\n",
1870 prefix, c->oom_score_adjust);
1872 for (i = 0; i < RLIM_NLIMITS; i++)
1874 fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
1876 if (c->ioprio_set) {
1880 r = ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c->ioprio), &class_str);
1884 "%sIOSchedulingClass: %s\n"
1885 "%sIOPriority: %i\n",
1886 prefix, strna(class_str),
1887 prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
1891 if (c->cpu_sched_set) {
1895 r = sched_policy_to_string_alloc(c->cpu_sched_policy, &policy_str);
1899 "%sCPUSchedulingPolicy: %s\n"
1900 "%sCPUSchedulingPriority: %i\n"
1901 "%sCPUSchedulingResetOnFork: %s\n",
1902 prefix, strna(policy_str),
1903 prefix, c->cpu_sched_priority,
1904 prefix, yes_no(c->cpu_sched_reset_on_fork));
1909 fprintf(f, "%sCPUAffinity:", prefix);
1910 for (i = 0; i < c->cpuset_ncpus; i++)
1911 if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
1912 fprintf(f, " %i", i);
1916 if (c->timer_slack_nsec != (nsec_t) -1)
1917 fprintf(f, "%sTimerSlackNSec: %lu\n", prefix, (unsigned long)c->timer_slack_nsec);
1920 "%sStandardInput: %s\n"
1921 "%sStandardOutput: %s\n"
1922 "%sStandardError: %s\n",
1923 prefix, exec_input_to_string(c->std_input),
1924 prefix, exec_output_to_string(c->std_output),
1925 prefix, exec_output_to_string(c->std_error));
1931 "%sTTYVHangup: %s\n"
1932 "%sTTYVTDisallocate: %s\n",
1933 prefix, c->tty_path,
1934 prefix, yes_no(c->tty_reset),
1935 prefix, yes_no(c->tty_vhangup),
1936 prefix, yes_no(c->tty_vt_disallocate));
1938 if (c->std_output == EXEC_OUTPUT_SYSLOG || c->std_output == EXEC_OUTPUT_KMSG || c->std_output == EXEC_OUTPUT_JOURNAL ||
1939 c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE ||
1940 c->std_error == EXEC_OUTPUT_SYSLOG || c->std_error == EXEC_OUTPUT_KMSG || c->std_error == EXEC_OUTPUT_JOURNAL ||
1941 c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE) {
1942 char *fac_str, *lvl_str;
1945 r = log_facility_unshifted_to_string_alloc(c->syslog_priority >> 3, &fac_str);
1949 r = log_level_to_string_alloc(LOG_PRI(c->syslog_priority), &lvl_str);
1954 "%sSyslogFacility: %s\n"
1955 "%sSyslogLevel: %s\n",
1956 prefix, strna(fac_str),
1957 prefix, strna(lvl_str));
1962 if (c->capabilities) {
1964 if ((t = cap_to_text(c->capabilities, NULL))) {
1965 fprintf(f, "%sCapabilities: %s\n",
1972 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
1974 (c->secure_bits & 1<<SECURE_KEEP_CAPS) ? " keep-caps" : "",
1975 (c->secure_bits & 1<<SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
1976 (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
1977 (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
1978 (c->secure_bits & 1<<SECURE_NOROOT) ? " noroot" : "",
1979 (c->secure_bits & 1<<SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
1981 if (c->capability_bounding_set_drop) {
1983 fprintf(f, "%sCapabilityBoundingSet:", prefix);
1985 for (l = 0; l <= cap_last_cap(); l++)
1986 if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) {
1989 if ((t = cap_to_name(l))) {
1990 fprintf(f, " %s", t);
1999 fprintf(f, "%sUser: %s\n", prefix, c->user);
2001 fprintf(f, "%sGroup: %s\n", prefix, c->group);
2003 if (strv_length(c->supplementary_groups) > 0) {
2004 fprintf(f, "%sSupplementaryGroups:", prefix);
2005 strv_fprintf(f, c->supplementary_groups);
2010 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
2012 if (strv_length(c->read_write_dirs) > 0) {
2013 fprintf(f, "%sReadWriteDirs:", prefix);
2014 strv_fprintf(f, c->read_write_dirs);
2018 if (strv_length(c->read_only_dirs) > 0) {
2019 fprintf(f, "%sReadOnlyDirs:", prefix);
2020 strv_fprintf(f, c->read_only_dirs);
2024 if (strv_length(c->inaccessible_dirs) > 0) {
2025 fprintf(f, "%sInaccessibleDirs:", prefix);
2026 strv_fprintf(f, c->inaccessible_dirs);
2032 "%sUtmpIdentifier: %s\n",
2033 prefix, c->utmp_id);
2036 void exec_status_start(ExecStatus *s, pid_t pid) {
2041 dual_timestamp_get(&s->start_timestamp);
2044 void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
2047 if (s->pid && s->pid != pid)
2051 dual_timestamp_get(&s->exit_timestamp);
2057 if (context->utmp_id)
2058 utmp_put_dead_process(context->utmp_id, pid, code, status);
2060 exec_context_tty_reset(context);
2064 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
2065 char buf[FORMAT_TIMESTAMP_MAX];
2078 prefix, (unsigned long) s->pid);
2080 if (s->start_timestamp.realtime > 0)
2082 "%sStart Timestamp: %s\n",
2083 prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
2085 if (s->exit_timestamp.realtime > 0)
2087 "%sExit Timestamp: %s\n"
2089 "%sExit Status: %i\n",
2090 prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
2091 prefix, sigchld_code_to_string(s->code),
2095 char *exec_command_line(char **argv) {
2103 STRV_FOREACH(a, argv)
2106 if (!(n = new(char, k)))
2110 STRV_FOREACH(a, argv) {
2117 if (strpbrk(*a, WHITESPACE)) {
2128 /* FIXME: this doesn't really handle arguments that have
2129 * spaces and ticks in them */
2134 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
2136 const char *prefix2;
2145 p2 = strappend(prefix, "\t");
2146 prefix2 = p2 ? p2 : prefix;
2148 cmd = exec_command_line(c->argv);
2151 "%sCommand Line: %s\n",
2152 prefix, cmd ? cmd : strerror(ENOMEM));
2156 exec_status_dump(&c->exec_status, f, prefix2);
2161 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
2167 LIST_FOREACH(command, c, c)
2168 exec_command_dump(c, f, prefix);
2171 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
2178 /* It's kind of important, that we keep the order here */
2179 LIST_FIND_TAIL(ExecCommand, command, *l, end);
2180 LIST_INSERT_AFTER(ExecCommand, command, *l, end, e);
2185 int exec_command_set(ExecCommand *c, const char *path, ...) {
2193 l = strv_new_ap(path, ap);
2199 if (!(p = strdup(path))) {
2213 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
2214 [EXEC_INPUT_NULL] = "null",
2215 [EXEC_INPUT_TTY] = "tty",
2216 [EXEC_INPUT_TTY_FORCE] = "tty-force",
2217 [EXEC_INPUT_TTY_FAIL] = "tty-fail",
2218 [EXEC_INPUT_SOCKET] = "socket"
2221 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
2223 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
2224 [EXEC_OUTPUT_INHERIT] = "inherit",
2225 [EXEC_OUTPUT_NULL] = "null",
2226 [EXEC_OUTPUT_TTY] = "tty",
2227 [EXEC_OUTPUT_SYSLOG] = "syslog",
2228 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
2229 [EXEC_OUTPUT_KMSG] = "kmsg",
2230 [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
2231 [EXEC_OUTPUT_JOURNAL] = "journal",
2232 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE] = "journal+console",
2233 [EXEC_OUTPUT_SOCKET] = "socket"
2236 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);