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 General Public License as published by
10 the Free Software Foundation; either version 2 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 General Public License for more details.
18 You should have received a copy of the GNU 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 <security/pam_appl.h>
51 #include "securebits.h"
53 #include "namespace.h"
55 #include "exit-status.h"
57 #include "utmp-wtmp.h"
59 #include "loopback-setup.h"
61 /* This assumes there is a 'tty' group */
64 static int shift_fds(int fds[], unsigned n_fds) {
65 int start, restart_from;
70 /* Modifies the fds array! (sorts it) */
80 for (i = start; i < (int) n_fds; i++) {
83 /* Already at right index? */
87 if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
90 close_nointr_nofail(fds[i]);
93 /* Hmm, the fd we wanted isn't free? Then
94 * let's remember that and try again from here*/
95 if (nfd != i+3 && restart_from < 0)
102 start = restart_from;
108 static int flags_fds(const int fds[], unsigned n_fds, bool nonblock) {
117 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
119 for (i = 0; i < n_fds; i++) {
121 if ((r = fd_nonblock(fds[i], nonblock)) < 0)
124 /* We unconditionally drop FD_CLOEXEC from the fds,
125 * since after all we want to pass these fds to our
128 if ((r = fd_cloexec(fds[i], false)) < 0)
135 static const char *tty_path(const ExecContext *context) {
138 if (context->tty_path)
139 return context->tty_path;
141 return "/dev/console";
144 void exec_context_tty_reset(const ExecContext *context) {
147 if (context->tty_vhangup)
148 terminal_vhangup(tty_path(context));
150 if (context->tty_reset)
151 reset_terminal(tty_path(context));
153 if (context->tty_vt_disallocate && context->tty_path)
154 vt_disallocate(context->tty_path);
157 static int open_null_as(int flags, int nfd) {
162 if ((fd = open("/dev/null", flags|O_NOCTTY)) < 0)
166 r = dup2(fd, nfd) < 0 ? -errno : nfd;
167 close_nointr_nofail(fd);
174 static int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, int nfd) {
176 union sockaddr_union sa;
179 assert(output < _EXEC_OUTPUT_MAX);
183 fd = socket(AF_UNIX, SOCK_STREAM, 0);
188 sa.un.sun_family = AF_UNIX;
189 strncpy(sa.un.sun_path, "/run/systemd/journal/stdout", sizeof(sa.un.sun_path));
191 r = connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path));
193 close_nointr_nofail(fd);
197 if (shutdown(fd, SHUT_RD) < 0) {
198 close_nointr_nofail(fd);
209 context->syslog_identifier ? context->syslog_identifier : ident,
210 context->syslog_priority,
211 !!context->syslog_level_prefix,
212 output == EXEC_OUTPUT_SYSLOG || output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE,
213 output == EXEC_OUTPUT_KMSG || output == EXEC_OUTPUT_KMSG_AND_CONSOLE,
214 output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || output == EXEC_OUTPUT_KMSG_AND_CONSOLE || output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE);
217 r = dup2(fd, nfd) < 0 ? -errno : nfd;
218 close_nointr_nofail(fd);
224 static int open_terminal_as(const char *path, mode_t mode, int nfd) {
230 if ((fd = open_terminal(path, mode | O_NOCTTY)) < 0)
234 r = dup2(fd, nfd) < 0 ? -errno : nfd;
235 close_nointr_nofail(fd);
242 static bool is_terminal_input(ExecInput i) {
244 i == EXEC_INPUT_TTY ||
245 i == EXEC_INPUT_TTY_FORCE ||
246 i == EXEC_INPUT_TTY_FAIL;
249 static int fixup_input(ExecInput std_input, int socket_fd, bool apply_tty_stdin) {
251 if (is_terminal_input(std_input) && !apply_tty_stdin)
252 return EXEC_INPUT_NULL;
254 if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0)
255 return EXEC_INPUT_NULL;
260 static int fixup_output(ExecOutput std_output, int socket_fd) {
262 if (std_output == EXEC_OUTPUT_SOCKET && socket_fd < 0)
263 return EXEC_OUTPUT_INHERIT;
268 static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty_stdin) {
273 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
277 case EXEC_INPUT_NULL:
278 return open_null_as(O_RDONLY, STDIN_FILENO);
281 case EXEC_INPUT_TTY_FORCE:
282 case EXEC_INPUT_TTY_FAIL: {
285 if ((fd = acquire_terminal(
287 i == EXEC_INPUT_TTY_FAIL,
288 i == EXEC_INPUT_TTY_FORCE,
292 if (fd != STDIN_FILENO) {
293 r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
294 close_nointr_nofail(fd);
301 case EXEC_INPUT_SOCKET:
302 return dup2(socket_fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
305 assert_not_reached("Unknown input type");
309 static int setup_output(const ExecContext *context, int socket_fd, const char *ident, bool apply_tty_stdin) {
316 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
317 o = fixup_output(context->std_output, socket_fd);
319 /* This expects the input is already set up */
323 case EXEC_OUTPUT_INHERIT:
325 /* If input got downgraded, inherit the original value */
326 if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input))
327 return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO);
329 /* If the input is connected to anything that's not a /dev/null, inherit that... */
330 if (i != EXEC_INPUT_NULL)
331 return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
333 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
335 return STDOUT_FILENO;
337 /* We need to open /dev/null here anew, to get the
338 * right access mode. So we fall through */
340 case EXEC_OUTPUT_NULL:
341 return open_null_as(O_WRONLY, STDOUT_FILENO);
343 case EXEC_OUTPUT_TTY:
344 if (is_terminal_input(i))
345 return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
347 /* We don't reset the terminal if this is just about output */
348 return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO);
350 case EXEC_OUTPUT_SYSLOG:
351 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
352 case EXEC_OUTPUT_KMSG:
353 case EXEC_OUTPUT_KMSG_AND_CONSOLE:
354 case EXEC_OUTPUT_JOURNAL:
355 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE:
356 return connect_logger_as(context, o, ident, STDOUT_FILENO);
358 case EXEC_OUTPUT_SOCKET:
359 assert(socket_fd >= 0);
360 return dup2(socket_fd, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
363 assert_not_reached("Unknown output type");
367 static int setup_error(const ExecContext *context, int socket_fd, const char *ident, bool apply_tty_stdin) {
374 i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
375 o = fixup_output(context->std_output, socket_fd);
376 e = fixup_output(context->std_error, socket_fd);
378 /* This expects the input and output are already set up */
380 /* Don't change the stderr file descriptor if we inherit all
381 * the way and are not on a tty */
382 if (e == EXEC_OUTPUT_INHERIT &&
383 o == EXEC_OUTPUT_INHERIT &&
384 i == EXEC_INPUT_NULL &&
385 !is_terminal_input(context->std_input) &&
387 return STDERR_FILENO;
389 /* Duplicate from stdout if possible */
390 if (e == o || e == EXEC_OUTPUT_INHERIT)
391 return dup2(STDOUT_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
395 case EXEC_OUTPUT_NULL:
396 return open_null_as(O_WRONLY, STDERR_FILENO);
398 case EXEC_OUTPUT_TTY:
399 if (is_terminal_input(i))
400 return dup2(STDIN_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
402 /* We don't reset the terminal if this is just about output */
403 return open_terminal_as(tty_path(context), O_WRONLY, STDERR_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 return connect_logger_as(context, e, ident, STDERR_FILENO);
413 case EXEC_OUTPUT_SOCKET:
414 assert(socket_fd >= 0);
415 return dup2(socket_fd, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
418 assert_not_reached("Unknown error type");
422 static int chown_terminal(int fd, uid_t uid) {
427 /* This might fail. What matters are the results. */
428 (void) fchown(fd, uid, -1);
429 (void) fchmod(fd, TTY_MODE);
431 if (fstat(fd, &st) < 0)
434 if (st.st_uid != uid || (st.st_mode & 0777) != TTY_MODE)
440 static int setup_confirm_stdio(const ExecContext *context,
442 int *_saved_stdout) {
443 int fd = -1, saved_stdin, saved_stdout = -1, r;
446 assert(_saved_stdin);
447 assert(_saved_stdout);
449 /* This returns positive EXIT_xxx return values instead of
450 * negative errno style values! */
452 if ((saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3)) < 0)
455 if ((saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3)) < 0) {
460 if ((fd = acquire_terminal(
462 context->std_input == EXEC_INPUT_TTY_FAIL,
463 context->std_input == EXEC_INPUT_TTY_FORCE,
469 if (chown_terminal(fd, getuid()) < 0) {
474 if (dup2(fd, STDIN_FILENO) < 0) {
479 if (dup2(fd, STDOUT_FILENO) < 0) {
485 close_nointr_nofail(fd);
487 *_saved_stdin = saved_stdin;
488 *_saved_stdout = saved_stdout;
493 if (saved_stdout >= 0)
494 close_nointr_nofail(saved_stdout);
496 if (saved_stdin >= 0)
497 close_nointr_nofail(saved_stdin);
500 close_nointr_nofail(fd);
505 static int restore_confirm_stdio(const ExecContext *context,
513 assert(*saved_stdin >= 0);
514 assert(saved_stdout);
515 assert(*saved_stdout >= 0);
517 /* This returns positive EXIT_xxx return values instead of
518 * negative errno style values! */
520 if (is_terminal_input(context->std_input)) {
522 /* The service wants terminal input. */
526 context->std_output == EXEC_OUTPUT_INHERIT ||
527 context->std_output == EXEC_OUTPUT_TTY;
530 /* If the service doesn't want a controlling terminal,
531 * then we need to get rid entirely of what we have
534 if (release_terminal() < 0)
537 if (dup2(*saved_stdin, STDIN_FILENO) < 0)
540 if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
543 *keep_stdout = *keep_stdin = false;
549 static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
550 bool keep_groups = false;
555 /* Lookup and set GID and supplementary group list. Here too
556 * we avoid NSS lookups for gid=0. */
558 if (context->group || username) {
560 if (context->group) {
561 const char *g = context->group;
563 if ((r = get_group_creds(&g, &gid)) < 0)
567 /* First step, initialize groups from /etc/groups */
568 if (username && gid != 0) {
569 if (initgroups(username, gid) < 0)
575 /* Second step, set our gids */
576 if (setresgid(gid, gid, gid) < 0)
580 if (context->supplementary_groups) {
585 /* Final step, initialize any manually set supplementary groups */
586 assert_se((ngroups_max = (int) sysconf(_SC_NGROUPS_MAX)) > 0);
588 if (!(gids = new(gid_t, ngroups_max)))
592 if ((k = getgroups(ngroups_max, gids)) < 0) {
599 STRV_FOREACH(i, context->supplementary_groups) {
602 if (k >= ngroups_max) {
608 r = get_group_creds(&g, gids+k);
617 if (setgroups(k, gids) < 0) {
628 static int enforce_user(const ExecContext *context, uid_t uid) {
632 /* Sets (but doesn't lookup) the uid and make sure we keep the
633 * capabilities while doing so. */
635 if (context->capabilities) {
637 static const cap_value_t bits[] = {
638 CAP_SETUID, /* Necessary so that we can run setresuid() below */
639 CAP_SETPCAP /* Necessary so that we can set PR_SET_SECUREBITS later on */
642 /* First step: If we need to keep capabilities but
643 * drop privileges we need to make sure we keep our
644 * caps, whiel we drop privileges. */
646 int sb = context->secure_bits|SECURE_KEEP_CAPS;
648 if (prctl(PR_GET_SECUREBITS) != sb)
649 if (prctl(PR_SET_SECUREBITS, sb) < 0)
653 /* Second step: set the capabilities. This will reduce
654 * the capabilities to the minimum we need. */
656 if (!(d = cap_dup(context->capabilities)))
659 if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
660 cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0) {
666 if (cap_set_proc(d) < 0) {
675 /* Third step: actually set the uids */
676 if (setresuid(uid, uid, uid) < 0)
679 /* At this point we should have all necessary capabilities but
680 are otherwise a normal user. However, the caps might got
681 corrupted due to the setresuid() so we need clean them up
682 later. This is done outside of this call. */
689 static int null_conv(
691 const struct pam_message **msg,
692 struct pam_response **resp,
695 /* We don't support conversations */
700 static int setup_pam(
705 int fds[], unsigned n_fds) {
707 static const struct pam_conv conv = {
712 pam_handle_t *handle = NULL;
714 int pam_code = PAM_SUCCESS;
717 bool close_session = false;
718 pid_t pam_pid = 0, parent_pid;
724 /* We set up PAM in the parent process, then fork. The child
725 * will then stay around until killed via PR_GET_PDEATHSIG or
726 * systemd via the cgroup logic. It will then remove the PAM
727 * session again. The parent process will exec() the actual
728 * daemon. We do things this way to ensure that the main PID
729 * of the daemon is the one we initially fork()ed. */
731 if ((pam_code = pam_start(name, user, &conv, &handle)) != PAM_SUCCESS) {
737 if ((pam_code = pam_set_item(handle, PAM_TTY, tty)) != PAM_SUCCESS)
740 if ((pam_code = pam_acct_mgmt(handle, PAM_SILENT)) != PAM_SUCCESS)
743 if ((pam_code = pam_open_session(handle, PAM_SILENT)) != PAM_SUCCESS)
746 close_session = true;
748 if ((!(e = pam_getenvlist(handle)))) {
749 pam_code = PAM_BUF_ERR;
753 /* Block SIGTERM, so that we know that it won't get lost in
755 if (sigemptyset(&ss) < 0 ||
756 sigaddset(&ss, SIGTERM) < 0 ||
757 sigprocmask(SIG_BLOCK, &ss, &old_ss) < 0)
760 parent_pid = getpid();
762 if ((pam_pid = fork()) < 0)
769 /* The child's job is to reset the PAM session on
772 /* This string must fit in 10 chars (i.e. the length
773 * of "/sbin/init"), to look pretty in /bin/ps */
774 rename_process("(sd-pam)");
776 /* Make sure we don't keep open the passed fds in this
777 child. We assume that otherwise only those fds are
778 open here that have been opened by PAM. */
779 close_many(fds, n_fds);
781 /* Wait until our parent died. This will most likely
782 * not work since the kernel does not allow
783 * unprivileged parents kill their privileged children
784 * this way. We rely on the control groups kill logic
785 * to do the rest for us. */
786 if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
789 /* Check if our parent process might already have
791 if (getppid() == parent_pid) {
793 if (sigwait(&ss, &sig) < 0) {
800 assert(sig == SIGTERM);
805 /* If our parent died we'll end the session */
806 if (getppid() != parent_pid)
807 if ((pam_code = pam_close_session(handle, PAM_DATA_SILENT)) != PAM_SUCCESS)
813 pam_end(handle, pam_code | PAM_DATA_SILENT);
817 /* If the child was forked off successfully it will do all the
818 * cleanups, so forget about the handle here. */
821 /* Unblock SIGTERM again in the parent */
822 if (sigprocmask(SIG_SETMASK, &old_ss, NULL) < 0)
825 /* We close the log explicitly here, since the PAM modules
826 * might have opened it, but we don't want this fd around. */
835 if (pam_code != PAM_SUCCESS)
836 err = -EPERM; /* PAM errors do not map to errno */
842 pam_code = pam_close_session(handle, PAM_DATA_SILENT);
844 pam_end(handle, pam_code | PAM_DATA_SILENT);
852 kill(pam_pid, SIGTERM);
853 kill(pam_pid, SIGCONT);
860 static int do_capability_bounding_set_drop(uint64_t drop) {
862 cap_t old_cap = NULL, new_cap = NULL;
866 /* If we are run as PID 1 we will lack CAP_SETPCAP by default
867 * in the effective set (yes, the kernel drops that when
868 * executing init!), so get it back temporarily so that we can
869 * call PR_CAPBSET_DROP. */
871 old_cap = cap_get_proc();
875 if (cap_get_flag(old_cap, CAP_SETPCAP, CAP_EFFECTIVE, &fv) < 0) {
881 static const cap_value_t v = CAP_SETPCAP;
883 new_cap = cap_dup(old_cap);
889 if (cap_set_flag(new_cap, CAP_EFFECTIVE, 1, &v, CAP_SET) < 0) {
894 if (cap_set_proc(new_cap) < 0) {
900 for (i = 0; i <= cap_last_cap(); i++)
901 if (drop & ((uint64_t) 1ULL << (uint64_t) i)) {
902 if (prctl(PR_CAPBSET_DROP, i) < 0) {
915 cap_set_proc(old_cap);
922 static void rename_process_from_path(const char *path) {
923 char process_name[11];
927 /* This resulting string must fit in 10 chars (i.e. the length
928 * of "/sbin/init") to look pretty in /bin/ps */
930 p = file_name_from_path(path);
932 rename_process("(...)");
938 /* The end of the process name is usually more
939 * interesting, since the first bit might just be
945 process_name[0] = '(';
946 memcpy(process_name+1, p, l);
947 process_name[1+l] = ')';
948 process_name[1+l+1] = 0;
950 rename_process(process_name);
953 int exec_spawn(ExecCommand *command,
955 const ExecContext *context,
956 int fds[], unsigned n_fds,
958 bool apply_permissions,
960 bool apply_tty_stdin,
962 CGroupBonding *cgroup_bondings,
963 CGroupAttribute *cgroup_attributes,
970 char **files_env = NULL;
975 assert(fds || n_fds <= 0);
977 if (context->std_input == EXEC_INPUT_SOCKET ||
978 context->std_output == EXEC_OUTPUT_SOCKET ||
979 context->std_error == EXEC_OUTPUT_SOCKET) {
991 if ((r = exec_context_load_environment(context, &files_env)) < 0) {
992 log_error("Failed to load environment files: %s", strerror(-r));
997 argv = command->argv;
999 if (!(line = exec_command_line(argv))) {
1004 log_debug("About to execute: %s", line);
1007 r = cgroup_bonding_realize_list(cgroup_bondings);
1011 cgroup_attribute_apply_list(cgroup_attributes, cgroup_bondings);
1013 if ((pid = fork()) < 0) {
1021 const char *username = NULL, *home = NULL;
1022 uid_t uid = (uid_t) -1;
1023 gid_t gid = (gid_t) -1;
1024 char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
1026 int saved_stdout = -1, saved_stdin = -1;
1027 bool keep_stdout = false, keep_stdin = false, set_access = false;
1031 rename_process_from_path(command->path);
1033 /* We reset exactly these signals, since they are the
1034 * only ones we set to SIG_IGN in the main daemon. All
1035 * others we leave untouched because we set them to
1036 * SIG_DFL or a valid handler initially, both of which
1037 * will be demoted to SIG_DFL. */
1038 default_signals(SIGNALS_CRASH_HANDLER,
1039 SIGNALS_IGNORE, -1);
1041 if (sigemptyset(&ss) < 0 ||
1042 sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
1044 r = EXIT_SIGNAL_MASK;
1048 /* Close sockets very early to make sure we don't
1049 * block init reexecution because it cannot bind its
1052 err = close_all_fds(socket_fd >= 0 ? &socket_fd : fds,
1053 socket_fd >= 0 ? 1 : n_fds);
1059 if (!context->same_pgrp)
1066 if (context->tcpwrap_name) {
1068 if (!socket_tcpwrap(socket_fd, context->tcpwrap_name)) {
1074 for (i = 0; i < (int) n_fds; i++) {
1075 if (!socket_tcpwrap(fds[i], context->tcpwrap_name)) {
1083 exec_context_tty_reset(context);
1085 /* We skip the confirmation step if we shall not apply the TTY */
1086 if (confirm_spawn &&
1087 (!is_terminal_input(context->std_input) || apply_tty_stdin)) {
1090 /* Set up terminal for the question */
1091 if ((r = setup_confirm_stdio(context,
1092 &saved_stdin, &saved_stdout))) {
1097 /* Now ask the question. */
1098 if (!(line = exec_command_line(argv))) {
1104 r = ask(&response, "yns", "Execute %s? [Yes, No, Skip] ", line);
1107 if (r < 0 || response == 'n') {
1111 } else if (response == 's') {
1116 /* Release terminal for the question */
1117 if ((r = restore_confirm_stdio(context,
1118 &saved_stdin, &saved_stdout,
1119 &keep_stdin, &keep_stdout))) {
1125 /* If a socket is connected to STDIN/STDOUT/STDERR, we
1126 * must sure to drop O_NONBLOCK */
1128 fd_nonblock(socket_fd, false);
1131 err = setup_input(context, socket_fd, apply_tty_stdin);
1139 err = setup_output(context, socket_fd, file_name_from_path(command->path), apply_tty_stdin);
1146 err = setup_error(context, socket_fd, file_name_from_path(command->path), apply_tty_stdin);
1152 if (cgroup_bondings) {
1153 err = cgroup_bonding_install_list(cgroup_bondings, 0);
1160 if (context->oom_score_adjust_set) {
1163 snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
1166 if (write_one_line_file("/proc/self/oom_score_adj", t) < 0) {
1167 /* Compatibility with Linux <= 2.6.35 */
1171 adj = (context->oom_score_adjust * -OOM_DISABLE) / OOM_SCORE_ADJ_MAX;
1172 adj = CLAMP(adj, OOM_DISABLE, OOM_ADJUST_MAX);
1174 snprintf(t, sizeof(t), "%i", adj);
1177 if (write_one_line_file("/proc/self/oom_adj", t) < 0
1178 && errno != EACCES) {
1180 r = EXIT_OOM_ADJUST;
1186 if (context->nice_set)
1187 if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
1193 if (context->cpu_sched_set) {
1194 struct sched_param param;
1197 param.sched_priority = context->cpu_sched_priority;
1199 if (sched_setscheduler(0, context->cpu_sched_policy |
1200 (context->cpu_sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0), ¶m) < 0) {
1202 r = EXIT_SETSCHEDULER;
1207 if (context->cpuset)
1208 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
1210 r = EXIT_CPUAFFINITY;
1214 if (context->ioprio_set)
1215 if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
1221 if (context->timer_slack_nsec_set)
1222 if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
1224 r = EXIT_TIMERSLACK;
1228 if (context->utmp_id)
1229 utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path);
1231 if (context->user) {
1232 username = context->user;
1233 err = get_user_creds(&username, &uid, &gid, &home);
1239 if (is_terminal_input(context->std_input)) {
1240 err = chown_terminal(STDIN_FILENO, uid);
1247 if (cgroup_bondings && context->control_group_modify) {
1248 err = cgroup_bonding_set_group_access_list(cgroup_bondings, 0755, uid, gid);
1250 err = cgroup_bonding_set_task_access_list(cgroup_bondings, 0644, uid, gid, context->control_group_persistent);
1260 if (cgroup_bondings && !set_access && context->control_group_persistent >= 0) {
1261 err = cgroup_bonding_set_task_access_list(cgroup_bondings, (mode_t) -1, (uid_t) -1, (uid_t) -1, context->control_group_persistent);
1268 if (apply_permissions) {
1269 err = enforce_groups(context, username, gid);
1276 umask(context->umask);
1279 if (context->pam_name && username) {
1280 err = setup_pam(context->pam_name, username, context->tty_path, &pam_env, fds, n_fds);
1287 if (context->private_network) {
1288 if (unshare(CLONE_NEWNET) < 0) {
1297 if (strv_length(context->read_write_dirs) > 0 ||
1298 strv_length(context->read_only_dirs) > 0 ||
1299 strv_length(context->inaccessible_dirs) > 0 ||
1300 context->mount_flags != MS_SHARED ||
1301 context->private_tmp) {
1302 err = setup_namespace(context->read_write_dirs,
1303 context->read_only_dirs,
1304 context->inaccessible_dirs,
1305 context->private_tmp,
1306 context->mount_flags);
1314 if (context->root_directory)
1315 if (chroot(context->root_directory) < 0) {
1321 if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
1330 if (asprintf(&d, "%s/%s",
1331 context->root_directory ? context->root_directory : "",
1332 context->working_directory ? context->working_directory : "") < 0) {
1348 /* We repeat the fd closing here, to make sure that
1349 * nothing is leaked from the PAM modules */
1350 err = close_all_fds(fds, n_fds);
1352 err = shift_fds(fds, n_fds);
1354 err = flags_fds(fds, n_fds, context->non_blocking);
1360 if (apply_permissions) {
1362 for (i = 0; i < RLIMIT_NLIMITS; i++) {
1363 if (!context->rlimit[i])
1366 if (setrlimit(i, context->rlimit[i]) < 0) {
1373 if (context->capability_bounding_set_drop) {
1374 err = do_capability_bounding_set_drop(context->capability_bounding_set_drop);
1376 r = EXIT_CAPABILITIES;
1381 if (context->user) {
1382 err = enforce_user(context, uid);
1389 /* PR_GET_SECUREBITS is not privileged, while
1390 * PR_SET_SECUREBITS is. So to suppress
1391 * potential EPERMs we'll try not to call
1392 * PR_SET_SECUREBITS unless necessary. */
1393 if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
1394 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
1396 r = EXIT_SECUREBITS;
1400 if (context->capabilities)
1401 if (cap_set_proc(context->capabilities) < 0) {
1403 r = EXIT_CAPABILITIES;
1408 if (!(our_env = new0(char*, 7))) {
1415 if (asprintf(our_env + n_env++, "LISTEN_PID=%lu", (unsigned long) getpid()) < 0 ||
1416 asprintf(our_env + n_env++, "LISTEN_FDS=%u", n_fds) < 0) {
1423 if (asprintf(our_env + n_env++, "HOME=%s", home) < 0) {
1430 if (asprintf(our_env + n_env++, "LOGNAME=%s", username) < 0 ||
1431 asprintf(our_env + n_env++, "USER=%s", username) < 0) {
1437 if (is_terminal_input(context->std_input) ||
1438 context->std_output == EXEC_OUTPUT_TTY ||
1439 context->std_error == EXEC_OUTPUT_TTY)
1440 if (!(our_env[n_env++] = strdup(default_term_for_tty(tty_path(context))))) {
1448 if (!(final_env = strv_env_merge(
1452 context->environment,
1461 if (!(final_argv = replace_env_argv(argv, final_env))) {
1467 final_env = strv_env_clean(final_env);
1469 execve(command->path, final_argv, final_env);
1476 log_warning("Failed at step %s spawning %s: %s",
1477 exit_status_to_string(r, EXIT_STATUS_SYSTEMD),
1478 command->path, strerror(-err));
1482 strv_free(final_env);
1484 strv_free(files_env);
1485 strv_free(final_argv);
1487 if (saved_stdin >= 0)
1488 close_nointr_nofail(saved_stdin);
1490 if (saved_stdout >= 0)
1491 close_nointr_nofail(saved_stdout);
1496 strv_free(files_env);
1498 /* We add the new process to the cgroup both in the child (so
1499 * that we can be sure that no user code is ever executed
1500 * outside of the cgroup) and in the parent (so that we can be
1501 * sure that when we kill the cgroup the process will be
1503 if (cgroup_bondings)
1504 cgroup_bonding_install_list(cgroup_bondings, pid);
1506 log_debug("Forked %s as %lu", command->path, (unsigned long) pid);
1508 exec_status_start(&command->exec_status, pid);
1514 strv_free(files_env);
1519 void exec_context_init(ExecContext *c) {
1523 c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
1524 c->cpu_sched_policy = SCHED_OTHER;
1525 c->syslog_priority = LOG_DAEMON|LOG_INFO;
1526 c->syslog_level_prefix = true;
1527 c->mount_flags = MS_SHARED;
1528 c->kill_signal = SIGTERM;
1529 c->send_sigkill = true;
1530 c->control_group_persistent = -1;
1533 void exec_context_done(ExecContext *c) {
1538 strv_free(c->environment);
1539 c->environment = NULL;
1541 strv_free(c->environment_files);
1542 c->environment_files = NULL;
1544 for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
1546 c->rlimit[l] = NULL;
1549 free(c->working_directory);
1550 c->working_directory = NULL;
1551 free(c->root_directory);
1552 c->root_directory = NULL;
1557 free(c->tcpwrap_name);
1558 c->tcpwrap_name = NULL;
1560 free(c->syslog_identifier);
1561 c->syslog_identifier = NULL;
1569 strv_free(c->supplementary_groups);
1570 c->supplementary_groups = NULL;
1575 if (c->capabilities) {
1576 cap_free(c->capabilities);
1577 c->capabilities = NULL;
1580 strv_free(c->read_only_dirs);
1581 c->read_only_dirs = NULL;
1583 strv_free(c->read_write_dirs);
1584 c->read_write_dirs = NULL;
1586 strv_free(c->inaccessible_dirs);
1587 c->inaccessible_dirs = NULL;
1590 CPU_FREE(c->cpuset);
1596 void exec_command_done(ExecCommand *c) {
1606 void exec_command_done_array(ExecCommand *c, unsigned n) {
1609 for (i = 0; i < n; i++)
1610 exec_command_done(c+i);
1613 void exec_command_free_list(ExecCommand *c) {
1617 LIST_REMOVE(ExecCommand, command, c, i);
1618 exec_command_done(i);
1623 void exec_command_free_array(ExecCommand **c, unsigned n) {
1626 for (i = 0; i < n; i++) {
1627 exec_command_free_list(c[i]);
1632 int exec_context_load_environment(const ExecContext *c, char ***l) {
1633 char **i, **r = NULL;
1638 STRV_FOREACH(i, c->environment_files) {
1641 bool ignore = false;
1651 if (!path_is_absolute(fn)) {
1660 if ((k = load_env_file(fn, &p)) < 0) {
1674 m = strv_env_merge(2, r, p);
1690 static void strv_fprintf(FILE *f, char **l) {
1696 fprintf(f, " %s", *g);
1699 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
1711 "%sWorkingDirectory: %s\n"
1712 "%sRootDirectory: %s\n"
1713 "%sNonBlocking: %s\n"
1714 "%sPrivateTmp: %s\n"
1715 "%sControlGroupModify: %s\n"
1716 "%sControlGroupPersistent: %s\n"
1717 "%sPrivateNetwork: %s\n",
1719 prefix, c->working_directory ? c->working_directory : "/",
1720 prefix, c->root_directory ? c->root_directory : "/",
1721 prefix, yes_no(c->non_blocking),
1722 prefix, yes_no(c->private_tmp),
1723 prefix, yes_no(c->control_group_modify),
1724 prefix, yes_no(c->control_group_persistent),
1725 prefix, yes_no(c->private_network));
1727 STRV_FOREACH(e, c->environment)
1728 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
1730 STRV_FOREACH(e, c->environment_files)
1731 fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
1733 if (c->tcpwrap_name)
1735 "%sTCPWrapName: %s\n",
1736 prefix, c->tcpwrap_name);
1743 if (c->oom_score_adjust_set)
1745 "%sOOMScoreAdjust: %i\n",
1746 prefix, c->oom_score_adjust);
1748 for (i = 0; i < RLIM_NLIMITS; i++)
1750 fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
1754 "%sIOSchedulingClass: %s\n"
1755 "%sIOPriority: %i\n",
1756 prefix, ioprio_class_to_string(IOPRIO_PRIO_CLASS(c->ioprio)),
1757 prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
1759 if (c->cpu_sched_set)
1761 "%sCPUSchedulingPolicy: %s\n"
1762 "%sCPUSchedulingPriority: %i\n"
1763 "%sCPUSchedulingResetOnFork: %s\n",
1764 prefix, sched_policy_to_string(c->cpu_sched_policy),
1765 prefix, c->cpu_sched_priority,
1766 prefix, yes_no(c->cpu_sched_reset_on_fork));
1769 fprintf(f, "%sCPUAffinity:", prefix);
1770 for (i = 0; i < c->cpuset_ncpus; i++)
1771 if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
1772 fprintf(f, " %i", i);
1776 if (c->timer_slack_nsec_set)
1777 fprintf(f, "%sTimerSlackNSec: %lu\n", prefix, c->timer_slack_nsec);
1780 "%sStandardInput: %s\n"
1781 "%sStandardOutput: %s\n"
1782 "%sStandardError: %s\n",
1783 prefix, exec_input_to_string(c->std_input),
1784 prefix, exec_output_to_string(c->std_output),
1785 prefix, exec_output_to_string(c->std_error));
1791 "%sTTYVHangup: %s\n"
1792 "%sTTYVTDisallocate: %s\n",
1793 prefix, c->tty_path,
1794 prefix, yes_no(c->tty_reset),
1795 prefix, yes_no(c->tty_vhangup),
1796 prefix, yes_no(c->tty_vt_disallocate));
1798 if (c->std_output == EXEC_OUTPUT_SYSLOG || c->std_output == EXEC_OUTPUT_KMSG || c->std_output == EXEC_OUTPUT_JOURNAL ||
1799 c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE ||
1800 c->std_error == EXEC_OUTPUT_SYSLOG || c->std_error == EXEC_OUTPUT_KMSG || c->std_error == EXEC_OUTPUT_JOURNAL ||
1801 c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE)
1803 "%sSyslogFacility: %s\n"
1804 "%sSyslogLevel: %s\n",
1805 prefix, log_facility_unshifted_to_string(c->syslog_priority >> 3),
1806 prefix, log_level_to_string(LOG_PRI(c->syslog_priority)));
1808 if (c->capabilities) {
1810 if ((t = cap_to_text(c->capabilities, NULL))) {
1811 fprintf(f, "%sCapabilities: %s\n",
1818 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
1820 (c->secure_bits & SECURE_KEEP_CAPS) ? " keep-caps" : "",
1821 (c->secure_bits & SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
1822 (c->secure_bits & SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
1823 (c->secure_bits & SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
1824 (c->secure_bits & SECURE_NOROOT) ? " noroot" : "",
1825 (c->secure_bits & SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
1827 if (c->capability_bounding_set_drop) {
1829 fprintf(f, "%sCapabilityBoundingSet:", prefix);
1831 for (l = 0; l <= cap_last_cap(); l++)
1832 if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) {
1835 if ((t = cap_to_name(l))) {
1836 fprintf(f, " %s", t);
1845 fprintf(f, "%sUser: %s\n", prefix, c->user);
1847 fprintf(f, "%sGroup: %s\n", prefix, c->group);
1849 if (strv_length(c->supplementary_groups) > 0) {
1850 fprintf(f, "%sSupplementaryGroups:", prefix);
1851 strv_fprintf(f, c->supplementary_groups);
1856 fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
1858 if (strv_length(c->read_write_dirs) > 0) {
1859 fprintf(f, "%sReadWriteDirs:", prefix);
1860 strv_fprintf(f, c->read_write_dirs);
1864 if (strv_length(c->read_only_dirs) > 0) {
1865 fprintf(f, "%sReadOnlyDirs:", prefix);
1866 strv_fprintf(f, c->read_only_dirs);
1870 if (strv_length(c->inaccessible_dirs) > 0) {
1871 fprintf(f, "%sInaccessibleDirs:", prefix);
1872 strv_fprintf(f, c->inaccessible_dirs);
1878 "%sKillSignal: SIG%s\n"
1879 "%sSendSIGKILL: %s\n",
1880 prefix, kill_mode_to_string(c->kill_mode),
1881 prefix, signal_to_string(c->kill_signal),
1882 prefix, yes_no(c->send_sigkill));
1886 "%sUtmpIdentifier: %s\n",
1887 prefix, c->utmp_id);
1890 void exec_status_start(ExecStatus *s, pid_t pid) {
1895 dual_timestamp_get(&s->start_timestamp);
1898 void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
1901 if (s->pid && s->pid != pid)
1905 dual_timestamp_get(&s->exit_timestamp);
1911 if (context->utmp_id)
1912 utmp_put_dead_process(context->utmp_id, pid, code, status);
1914 exec_context_tty_reset(context);
1918 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
1919 char buf[FORMAT_TIMESTAMP_MAX];
1932 prefix, (unsigned long) s->pid);
1934 if (s->start_timestamp.realtime > 0)
1936 "%sStart Timestamp: %s\n",
1937 prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
1939 if (s->exit_timestamp.realtime > 0)
1941 "%sExit Timestamp: %s\n"
1943 "%sExit Status: %i\n",
1944 prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
1945 prefix, sigchld_code_to_string(s->code),
1949 char *exec_command_line(char **argv) {
1957 STRV_FOREACH(a, argv)
1960 if (!(n = new(char, k)))
1964 STRV_FOREACH(a, argv) {
1971 if (strpbrk(*a, WHITESPACE)) {
1982 /* FIXME: this doesn't really handle arguments that have
1983 * spaces and ticks in them */
1988 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
1990 const char *prefix2;
1999 p2 = strappend(prefix, "\t");
2000 prefix2 = p2 ? p2 : prefix;
2002 cmd = exec_command_line(c->argv);
2005 "%sCommand Line: %s\n",
2006 prefix, cmd ? cmd : strerror(ENOMEM));
2010 exec_status_dump(&c->exec_status, f, prefix2);
2015 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
2021 LIST_FOREACH(command, c, c)
2022 exec_command_dump(c, f, prefix);
2025 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
2032 /* It's kind of important, that we keep the order here */
2033 LIST_FIND_TAIL(ExecCommand, command, *l, end);
2034 LIST_INSERT_AFTER(ExecCommand, command, *l, end, e);
2039 int exec_command_set(ExecCommand *c, const char *path, ...) {
2047 l = strv_new_ap(path, ap);
2053 if (!(p = strdup(path))) {
2067 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
2068 [EXEC_INPUT_NULL] = "null",
2069 [EXEC_INPUT_TTY] = "tty",
2070 [EXEC_INPUT_TTY_FORCE] = "tty-force",
2071 [EXEC_INPUT_TTY_FAIL] = "tty-fail",
2072 [EXEC_INPUT_SOCKET] = "socket"
2075 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
2077 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
2078 [EXEC_OUTPUT_INHERIT] = "inherit",
2079 [EXEC_OUTPUT_NULL] = "null",
2080 [EXEC_OUTPUT_TTY] = "tty",
2081 [EXEC_OUTPUT_SYSLOG] = "syslog",
2082 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
2083 [EXEC_OUTPUT_KMSG] = "kmsg",
2084 [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
2085 [EXEC_OUTPUT_JOURNAL] = "journal",
2086 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE] = "journal+console",
2087 [EXEC_OUTPUT_SOCKET] = "socket"
2090 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
2092 static const char* const kill_mode_table[_KILL_MODE_MAX] = {
2093 [KILL_CONTROL_GROUP] = "control-group",
2094 [KILL_PROCESS] = "process",
2095 [KILL_NONE] = "none"
2098 DEFINE_STRING_TABLE_LOOKUP(kill_mode, KillMode);
2100 static const char* const kill_who_table[_KILL_WHO_MAX] = {
2101 [KILL_MAIN] = "main",
2102 [KILL_CONTROL] = "control",
2106 DEFINE_STRING_TABLE_LOOKUP(kill_who, KillWho);