1 /*-*- Mode: C; c-basic-offset: 8 -*-*/
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>
44 #include "securebits.h"
47 static int close_fds(int except[], unsigned n_except) {
52 /* Modifies the fds array! (sorts it) */
54 if (!(d = opendir("/proc/self/fd")))
57 while ((de = readdir(d))) {
60 if (de->d_name[0] == '.')
63 if ((r = safe_atoi(de->d_name, &fd)) < 0)
77 for (i = 0; i < n_except; i++)
78 if (except[i] == fd) {
87 if ((r = close_nointr(fd)) < 0)
96 static int shift_fds(int fds[], unsigned n_fds) {
97 int start, restart_from;
110 for (i = start; i < (int) n_fds; i++) {
113 /* Already at right index? */
117 if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
120 assert_se(close_nointr(fds[i]) == 0);
123 /* Hmm, the fd we wanted isn't free? Then
124 * let's remember that and try again from here*/
125 if (nfd != i+3 && restart_from < 0)
129 if (restart_from < 0)
132 start = restart_from;
138 static int flags_fds(int fds[], unsigned n_fds, bool nonblock) {
147 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
149 for (i = 0; i < n_fds; i++) {
151 if ((r = fd_nonblock(fds[i], nonblock)) < 0)
154 /* We unconditionally drop FD_CLOEXEC from the fds,
155 * since after all we want to pass these fds to our
158 if ((r = fd_cloexec(fds[i], false)) < 0)
165 static int replace_null_fd(int fd, int flags) {
171 if ((nfd = open("/dev/null", flags|O_NOCTTY)) < 0)
175 close_nointr_nofail(nfd);
182 static int setup_output(const ExecContext *context, const char *ident) {
187 switch (context->output) {
189 case EXEC_OUTPUT_CONSOLE:
192 case EXEC_OUTPUT_NULL:
194 if ((r = replace_null_fd(STDOUT_FILENO, O_WRONLY)) < 0 ||
195 (r = replace_null_fd(STDERR_FILENO, O_WRONLY)) < 0)
200 case EXEC_OUTPUT_KERNEL:
201 case EXEC_OUTPUT_SYSLOG: {
206 struct sockaddr_un un;
209 close_nointr(STDOUT_FILENO);
210 close_nointr(STDERR_FILENO);
212 if ((fd = socket(AF_UNIX, SOCK_STREAM, 0)) < 0)
215 if (fd != STDOUT_FILENO) {
216 close_nointr_nofail(fd);
221 sa.sa.sa_family = AF_UNIX;
222 strncpy(sa.un.sun_path+1, LOGGER_SOCKET, sizeof(sa.un.sun_path)-1);
224 if (connect(fd, &sa.sa, sizeof(sa)) < 0) {
225 close_nointr_nofail(fd);
229 if (shutdown(fd, SHUT_RD) < 0) {
230 close_nointr_nofail(fd);
234 if ((fd = dup(fd)) < 0) {
235 close_nointr_nofail(fd);
239 if (fd != STDERR_FILENO) {
240 close_nointr_nofail(fd);
244 /* We speak a very simple protocol between log server
245 * and client: one line for the log destination (kmsg
246 * or syslog), followed by the priority field,
247 * followed by the process name. Since we replaced
248 * stdin/stderr we simple use stdio to write to
249 * it. Note that we use stderr, to minimize buffer
250 * flushing issues. */
256 context->output == EXEC_OUTPUT_KERNEL ? "kmsg" : "syslog",
257 context->syslog_priority,
258 context->syslog_identifier ? context->syslog_identifier : ident);
264 assert_not_reached("Unknown output type");
268 static int setup_input(const ExecContext *context) {
273 switch (context->input) {
275 case EXEC_INPUT_CONSOLE:
278 case EXEC_INPUT_NULL:
279 if ((r = replace_null_fd(STDIN_FILENO, O_RDONLY)) < 0)
285 assert_not_reached("Unknown input type");
289 static int get_group_creds(const char *groupname, gid_t *gid) {
296 /* We enforce some special rules for gid=0: in order to avoid
297 * NSS lookups for root we hardcode its data. */
299 if (streq(groupname, "root") || streq(groupname, "0")) {
304 if (safe_atolu(groupname, &lu) >= 0) {
306 g = getgrgid((gid_t) lu);
309 g = getgrnam(groupname);
313 return errno != 0 ? -errno : -ESRCH;
319 static int get_user_creds(const char **username, uid_t *uid, gid_t *gid, const char **home) {
329 /* We enforce some special rules for uid=0: in order to avoid
330 * NSS lookups for root we hardcode its data. */
332 if (streq(*username, "root") || streq(*username, "0")) {
340 if (safe_atolu(*username, &lu) >= 0) {
342 p = getpwuid((uid_t) lu);
344 /* If there are multiple users with the same id, make
345 * sure to leave $USER to the configured value instead
346 * of the first occurence in the database. However if
347 * the uid was configured by a numeric uid, then let's
348 * pick the real username from /etc/passwd. */
350 *username = p->pw_name;
353 p = getpwnam(*username);
357 return errno != 0 ? -errno : -ESRCH;
365 static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
366 bool keep_groups = false;
371 /* Lookup and ser GID and supplementary group list. Here too
372 * we avoid NSS lookups for gid=0. */
374 if (context->group || username) {
377 if ((r = get_group_creds(context->group, &gid)) < 0)
380 /* First step, initialize groups from /etc/groups */
381 if (username && gid != 0) {
382 if (initgroups(username, gid) < 0)
388 /* Second step, set our gids */
389 if (setresgid(gid, gid, gid) < 0)
393 if (context->supplementary_groups) {
398 /* Final step, initialize any manually set supplementary groups */
399 ngroups_max = (int) sysconf(_SC_NGROUPS_MAX);
401 if (!(gids = new(gid_t, ngroups_max)))
405 if ((k = getgroups(ngroups_max, gids)) < 0) {
412 STRV_FOREACH(i, context->supplementary_groups) {
414 if (k >= ngroups_max) {
419 if ((r = get_group_creds(*i, gids+k)) < 0) {
427 if (setgroups(k, gids) < 0) {
438 static int enforce_user(const ExecContext *context, uid_t uid) {
442 /* Sets (but doesn't lookup) the uid and make sure we keep the
443 * capabilities while doing so. */
445 if (context->capabilities) {
447 static const cap_value_t bits[] = {
448 CAP_SETUID, /* Necessary so that we can run setresuid() below */
449 CAP_SETPCAP /* Necessary so that we can set PR_SET_SECUREBITS later on */
452 /* First step: If we need to keep capabilities but
453 * drop privileges we need to make sure we keep our
454 * caps, whiel we drop priviliges. */
456 int sb = context->secure_bits|SECURE_KEEP_CAPS;
458 if (prctl(PR_GET_SECUREBITS) != sb)
459 if (prctl(PR_SET_SECUREBITS, sb) < 0)
463 /* Second step: set the capabilites. This will reduce
464 * the capabilities to the minimum we need. */
466 if (!(d = cap_dup(context->capabilities)))
469 if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
470 cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0) {
476 if (cap_set_proc(d) < 0) {
485 /* Third step: actually set the uids */
486 if (setresuid(uid, uid, uid) < 0)
489 /* At this point we should have all necessary capabilities but
490 are otherwise a normal user. However, the caps might got
491 corrupted due to the setresuid() so we need clean them up
492 later. This is done outside of this call. */
497 int exec_spawn(const ExecCommand *command,
498 const ExecContext *context,
499 int *fds, unsigned n_fds,
500 bool apply_permissions,
502 CGroupBonding *cgroup_bondings,
511 assert(fds || n_fds <= 0);
513 log_debug("About to execute %s", command->path);
516 if ((r = cgroup_bonding_realize_list(cgroup_bondings)))
519 if ((pid = fork()) < 0)
525 const char *username = NULL, *home = NULL;
526 uid_t uid = (uid_t) -1;
527 gid_t gid = (gid_t) -1;
528 char **our_env = NULL, **final_env = NULL;
533 if (sigemptyset(&ss) < 0 ||
534 sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
535 r = EXIT_SIGNAL_MASK;
539 if (setpgid(0, 0) < 0) {
544 umask(context->umask);
546 if (setup_input(context) < 0) {
551 if (setup_output(context, file_name_from_path(command->path)) < 0) {
557 if ((r = cgroup_bonding_install_list(cgroup_bondings, 0)) < 0) {
562 if (context->oom_adjust_set) {
565 snprintf(t, sizeof(t), "%i", context->oom_adjust);
568 if (write_one_line_file("/proc/self/oom_adj", t) < 0) {
574 if (context->nice_set)
575 if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
580 if (context->cpu_sched_set) {
581 struct sched_param param;
584 param.sched_priority = context->cpu_sched_priority;
586 if (sched_setscheduler(0, context->cpu_sched_policy |
587 (context->cpu_sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0), ¶m) < 0) {
588 r = EXIT_SETSCHEDULER;
593 if (context->cpu_affinity_set)
594 if (sched_setaffinity(0, sizeof(context->cpu_affinity), &context->cpu_affinity) < 0) {
595 r = EXIT_CPUAFFINITY;
599 if (context->ioprio_set)
600 if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
605 if (context->timer_slack_ns_set)
606 if (prctl(PR_SET_TIMERSLACK, context->timer_slack_ns_set) < 0) {
612 username = context->user;
613 if (get_user_creds(&username, &uid, &gid, &home) < 0) {
619 if (apply_permissions)
620 if (enforce_groups(context, username, uid) < 0) {
626 if (context->root_directory)
627 if (chroot(context->root_directory) < 0) {
632 if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
640 if (asprintf(&d, "%s/%s",
641 context->root_directory ? context->root_directory : "",
642 context->working_directory ? context->working_directory : "") < 0) {
656 if (close_fds(fds, n_fds) < 0 ||
657 shift_fds(fds, n_fds) < 0 ||
658 flags_fds(fds, n_fds, context->non_blocking) < 0) {
663 if (apply_permissions) {
665 for (i = 0; i < RLIMIT_NLIMITS; i++) {
666 if (!context->rlimit[i])
669 if (setrlimit(i, context->rlimit[i]) < 0) {
676 if (enforce_user(context, uid) < 0) {
681 /* PR_GET_SECUREBITS is not priviliged, while
682 * PR_SET_SECUREBITS is. So to suppress
683 * potential EPERMs we'll try not to call
684 * PR_SET_SECUREBITS unless necessary. */
685 if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
686 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
691 if (context->capabilities)
692 if (cap_set_proc(context->capabilities) < 0) {
693 r = EXIT_CAPABILITIES;
698 if (!(our_env = new0(char*, 6))) {
704 if (asprintf(our_env + n_env++, "LISTEN_PID=%llu", (unsigned long long) getpid()) < 0 ||
705 asprintf(our_env + n_env++, "LISTEN_FDS=%u", n_fds) < 0) {
711 if (asprintf(our_env + n_env++, "HOME=%s", home) < 0) {
717 if (asprintf(our_env + n_env++, "LOGNAME=%s", username) < 0 ||
718 asprintf(our_env + n_env++, "USER=%s", username) < 0) {
723 if (!(final_env = strv_env_merge(environ, our_env, context->environment, NULL))) {
728 execve(command->path, command->argv, final_env);
733 strv_free(final_env);
739 log_debug("Forked %s as %llu", command->path, (unsigned long long) pid);
745 void exec_context_init(ExecContext *c) {
750 c->oom_adjust_set = false;
753 c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
754 c->ioprio_set = false;
755 c->cpu_sched_policy = SCHED_OTHER;
756 c->cpu_sched_priority = 0;
757 c->cpu_sched_set = false;
758 CPU_ZERO(&c->cpu_affinity);
759 c->cpu_affinity_set = false;
763 c->syslog_priority = LOG_DAEMON|LOG_INFO;
766 c->capability_bounding_set_drop = 0;
769 void exec_context_done(ExecContext *c) {
774 strv_free(c->environment);
775 c->environment = NULL;
777 for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
782 free(c->working_directory);
783 c->working_directory = NULL;
784 free(c->root_directory);
785 c->root_directory = NULL;
787 free(c->syslog_identifier);
788 c->syslog_identifier = NULL;
796 strv_free(c->supplementary_groups);
797 c->supplementary_groups = NULL;
799 if (c->capabilities) {
800 cap_free(c->capabilities);
801 c->capabilities = NULL;
805 void exec_command_free_list(ExecCommand *c) {
809 LIST_REMOVE(ExecCommand, command, c, i);
817 void exec_command_free_array(ExecCommand **c, unsigned n) {
820 for (i = 0; i < n; i++) {
821 exec_command_free_list(c[i]);
826 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
838 "%sWorkingDirectory: %s\n"
839 "%sRootDirectory: %s\n"
840 "%sNonBlocking: %s\n",
842 prefix, c->working_directory ? c->working_directory : "/",
843 prefix, c->root_directory ? c->root_directory : "/",
844 prefix, yes_no(c->non_blocking));
847 for (e = c->environment; *e; e++)
848 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
855 if (c->oom_adjust_set)
858 prefix, c->oom_adjust);
860 for (i = 0; i < RLIM_NLIMITS; i++)
862 fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
866 "%sIOSchedulingClass: %s\n"
867 "%sIOPriority: %i\n",
868 prefix, ioprio_class_to_string(IOPRIO_PRIO_CLASS(c->ioprio)),
869 prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
871 if (c->cpu_sched_set)
873 "%sCPUSchedulingPolicy: %s\n"
874 "%sCPUSchedulingPriority: %i\n"
875 "%sCPUSchedulingResetOnFork: %s\n",
876 prefix, sched_policy_to_string(c->cpu_sched_policy),
877 prefix, c->cpu_sched_priority,
878 prefix, yes_no(c->cpu_sched_reset_on_fork));
880 if (c->cpu_affinity_set) {
881 fprintf(f, "%sCPUAffinity:", prefix);
882 for (i = 0; i < CPU_SETSIZE; i++)
883 if (CPU_ISSET(i, &c->cpu_affinity))
884 fprintf(f, " %i", i);
888 if (c->timer_slack_ns_set)
889 fprintf(f, "%sTimerSlackNS: %lu\n", prefix, c->timer_slack_ns);
894 prefix, exec_input_to_string(c->input),
895 prefix, exec_output_to_string(c->output));
897 if (c->output == EXEC_OUTPUT_SYSLOG || c->output == EXEC_OUTPUT_KERNEL)
899 "%sSyslogFacility: %s\n"
900 "%sSyslogLevel: %s\n",
901 prefix, log_facility_to_string(LOG_FAC(c->syslog_priority)),
902 prefix, log_level_to_string(LOG_PRI(c->syslog_priority)));
904 if (c->capabilities) {
906 if ((t = cap_to_text(c->capabilities, NULL))) {
907 fprintf(f, "%sCapabilities: %s\n",
914 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
916 (c->secure_bits & SECURE_KEEP_CAPS) ? " keep-caps" : "",
917 (c->secure_bits & SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
918 (c->secure_bits & SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
919 (c->secure_bits & SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
920 (c->secure_bits & SECURE_NOROOT) ? " noroot" : "",
921 (c->secure_bits & SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
923 if (c->capability_bounding_set_drop) {
924 fprintf(f, "%sCapabilityBoundingSetDrop:", prefix);
926 for (i = 0; i <= CAP_LAST_CAP; i++)
927 if (c->capability_bounding_set_drop & (1 << i)) {
930 if ((t = cap_to_name(i))) {
931 fprintf(f, " %s", t);
940 fprintf(f, "%sUser: %s", prefix, c->user);
942 fprintf(f, "%sGroup: %s", prefix, c->group);
944 if (c->supplementary_groups) {
947 fprintf(f, "%sSupplementaryGroups:", prefix);
949 STRV_FOREACH(g, c->supplementary_groups)
950 fprintf(f, " %s", *g);
956 void exec_status_fill(ExecStatus *s, pid_t pid, int code, int status) {
962 s->timestamp = now(CLOCK_REALTIME);
965 char *exec_command_line(ExecCommand *c) {
974 STRV_FOREACH(a, c->argv)
977 if (!(n = new(char, k)))
981 STRV_FOREACH(a, c->argv) {
988 if (strpbrk(*a, WHITESPACE)) {
999 /* FIXME: this doesn't really handle arguments that have
1000 * spaces and ticks in them */
1005 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
1014 cmd = exec_command_line(c);
1017 "%sCommand Line: %s\n",
1018 prefix, cmd ? cmd : strerror(ENOMEM));
1023 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
1029 LIST_FOREACH(command, c, c)
1030 exec_command_dump(c, f, prefix);
1033 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
1040 /* It's kinda important that we keep the order here */
1041 LIST_FIND_TAIL(ExecCommand, command, *l, end);
1042 LIST_INSERT_AFTER(ExecCommand, command, *l, end, e);
1047 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
1048 [EXEC_OUTPUT_CONSOLE] = "console",
1049 [EXEC_OUTPUT_NULL] = "null",
1050 [EXEC_OUTPUT_SYSLOG] = "syslog",
1051 [EXEC_OUTPUT_KERNEL] = "kernel"
1054 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
1056 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
1057 [EXEC_INPUT_NULL] = "null",
1058 [EXEC_INPUT_CONSOLE] = "console"
1061 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);