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 shift_fds(int fds[], unsigned n_fds) {
48 int start, restart_from;
53 /* Modifies the fds array! (sorts it) */
63 for (i = start; i < (int) n_fds; i++) {
66 /* Already at right index? */
70 if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
73 assert_se(close_nointr(fds[i]) == 0);
76 /* Hmm, the fd we wanted isn't free? Then
77 * let's remember that and try again from here*/
78 if (nfd != i+3 && restart_from < 0)
91 static int flags_fds(int fds[], unsigned n_fds, bool nonblock) {
100 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
102 for (i = 0; i < n_fds; i++) {
104 if ((r = fd_nonblock(fds[i], nonblock)) < 0)
107 /* We unconditionally drop FD_CLOEXEC from the fds,
108 * since after all we want to pass these fds to our
111 if ((r = fd_cloexec(fds[i], false)) < 0)
118 static int replace_null_fd(int fd, int flags) {
124 if ((nfd = open("/dev/null", flags|O_NOCTTY)) < 0)
128 close_nointr_nofail(nfd);
135 static int setup_output(const ExecContext *context, const char *ident) {
140 switch (context->output) {
142 case EXEC_OUTPUT_CONSOLE:
145 case EXEC_OUTPUT_NULL:
147 if ((r = replace_null_fd(STDOUT_FILENO, O_WRONLY)) < 0 ||
148 (r = replace_null_fd(STDERR_FILENO, O_WRONLY)) < 0)
153 case EXEC_OUTPUT_KERNEL:
154 case EXEC_OUTPUT_SYSLOG: {
159 struct sockaddr_un un;
162 close_nointr(STDOUT_FILENO);
163 close_nointr(STDERR_FILENO);
165 if ((fd = socket(AF_UNIX, SOCK_STREAM, 0)) < 0)
168 if (fd != STDOUT_FILENO) {
169 close_nointr_nofail(fd);
174 sa.sa.sa_family = AF_UNIX;
175 strncpy(sa.un.sun_path+1, LOGGER_SOCKET, sizeof(sa.un.sun_path)-1);
177 if (connect(fd, &sa.sa, sizeof(sa)) < 0) {
178 close_nointr_nofail(fd);
182 if (shutdown(fd, SHUT_RD) < 0) {
183 close_nointr_nofail(fd);
187 if ((fd = dup(fd)) < 0) {
188 close_nointr_nofail(fd);
192 if (fd != STDERR_FILENO) {
193 close_nointr_nofail(fd);
197 /* We speak a very simple protocol between log server
198 * and client: one line for the log destination (kmsg
199 * or syslog), followed by the priority field,
200 * followed by the process name. Since we replaced
201 * stdin/stderr we simple use stdio to write to
202 * it. Note that we use stderr, to minimize buffer
203 * flushing issues. */
209 context->output == EXEC_OUTPUT_KERNEL ? "kmsg" : "syslog",
210 context->syslog_priority,
211 context->syslog_identifier ? context->syslog_identifier : ident);
217 assert_not_reached("Unknown output type");
221 static int setup_input(const ExecContext *context) {
226 switch (context->input) {
228 case EXEC_INPUT_CONSOLE:
231 case EXEC_INPUT_NULL:
232 if ((r = replace_null_fd(STDIN_FILENO, O_RDONLY)) < 0)
238 assert_not_reached("Unknown input type");
242 static int get_group_creds(const char *groupname, gid_t *gid) {
249 /* We enforce some special rules for gid=0: in order to avoid
250 * NSS lookups for root we hardcode its data. */
252 if (streq(groupname, "root") || streq(groupname, "0")) {
257 if (safe_atolu(groupname, &lu) >= 0) {
259 g = getgrgid((gid_t) lu);
262 g = getgrnam(groupname);
266 return errno != 0 ? -errno : -ESRCH;
272 static int get_user_creds(const char **username, uid_t *uid, gid_t *gid, const char **home) {
282 /* We enforce some special rules for uid=0: in order to avoid
283 * NSS lookups for root we hardcode its data. */
285 if (streq(*username, "root") || streq(*username, "0")) {
293 if (safe_atolu(*username, &lu) >= 0) {
295 p = getpwuid((uid_t) lu);
297 /* If there are multiple users with the same id, make
298 * sure to leave $USER to the configured value instead
299 * of the first occurence in the database. However if
300 * the uid was configured by a numeric uid, then let's
301 * pick the real username from /etc/passwd. */
303 *username = p->pw_name;
306 p = getpwnam(*username);
310 return errno != 0 ? -errno : -ESRCH;
318 static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
319 bool keep_groups = false;
324 /* Lookup and ser GID and supplementary group list. Here too
325 * we avoid NSS lookups for gid=0. */
327 if (context->group || username) {
330 if ((r = get_group_creds(context->group, &gid)) < 0)
333 /* First step, initialize groups from /etc/groups */
334 if (username && gid != 0) {
335 if (initgroups(username, gid) < 0)
341 /* Second step, set our gids */
342 if (setresgid(gid, gid, gid) < 0)
346 if (context->supplementary_groups) {
351 /* Final step, initialize any manually set supplementary groups */
352 ngroups_max = (int) sysconf(_SC_NGROUPS_MAX);
354 if (!(gids = new(gid_t, ngroups_max)))
358 if ((k = getgroups(ngroups_max, gids)) < 0) {
365 STRV_FOREACH(i, context->supplementary_groups) {
367 if (k >= ngroups_max) {
372 if ((r = get_group_creds(*i, gids+k)) < 0) {
380 if (setgroups(k, gids) < 0) {
391 static int enforce_user(const ExecContext *context, uid_t uid) {
395 /* Sets (but doesn't lookup) the uid and make sure we keep the
396 * capabilities while doing so. */
398 if (context->capabilities) {
400 static const cap_value_t bits[] = {
401 CAP_SETUID, /* Necessary so that we can run setresuid() below */
402 CAP_SETPCAP /* Necessary so that we can set PR_SET_SECUREBITS later on */
405 /* First step: If we need to keep capabilities but
406 * drop privileges we need to make sure we keep our
407 * caps, whiel we drop priviliges. */
409 int sb = context->secure_bits|SECURE_KEEP_CAPS;
411 if (prctl(PR_GET_SECUREBITS) != sb)
412 if (prctl(PR_SET_SECUREBITS, sb) < 0)
416 /* Second step: set the capabilites. This will reduce
417 * the capabilities to the minimum we need. */
419 if (!(d = cap_dup(context->capabilities)))
422 if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
423 cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0) {
429 if (cap_set_proc(d) < 0) {
438 /* Third step: actually set the uids */
439 if (setresuid(uid, uid, uid) < 0)
442 /* At this point we should have all necessary capabilities but
443 are otherwise a normal user. However, the caps might got
444 corrupted due to the setresuid() so we need clean them up
445 later. This is done outside of this call. */
450 int exec_spawn(const ExecCommand *command,
451 const ExecContext *context,
452 int *fds, unsigned n_fds,
453 bool apply_permissions,
455 CGroupBonding *cgroup_bondings,
464 assert(fds || n_fds <= 0);
466 log_debug("About to execute %s", command->path);
469 if ((r = cgroup_bonding_realize_list(cgroup_bondings)))
472 if ((pid = fork()) < 0)
478 const char *username = NULL, *home = NULL;
479 uid_t uid = (uid_t) -1;
480 gid_t gid = (gid_t) -1;
481 char **our_env = NULL, **final_env = NULL;
486 if (sigemptyset(&ss) < 0 ||
487 sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
488 r = EXIT_SIGNAL_MASK;
492 if (context->new_session) {
498 if (setpgid(0, 0) < 0) {
504 umask(context->umask);
506 if (setup_input(context) < 0) {
511 if (setup_output(context, file_name_from_path(command->path)) < 0) {
517 if ((r = cgroup_bonding_install_list(cgroup_bondings, 0)) < 0) {
522 if (context->oom_adjust_set) {
525 snprintf(t, sizeof(t), "%i", context->oom_adjust);
528 if (write_one_line_file("/proc/self/oom_adj", t) < 0) {
534 if (context->nice_set)
535 if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
540 if (context->cpu_sched_set) {
541 struct sched_param param;
544 param.sched_priority = context->cpu_sched_priority;
546 if (sched_setscheduler(0, context->cpu_sched_policy |
547 (context->cpu_sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0), ¶m) < 0) {
548 r = EXIT_SETSCHEDULER;
553 if (context->cpu_affinity_set)
554 if (sched_setaffinity(0, sizeof(context->cpu_affinity), &context->cpu_affinity) < 0) {
555 r = EXIT_CPUAFFINITY;
559 if (context->ioprio_set)
560 if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
565 if (context->timer_slack_ns_set)
566 if (prctl(PR_SET_TIMERSLACK, context->timer_slack_ns_set) < 0) {
572 username = context->user;
573 if (get_user_creds(&username, &uid, &gid, &home) < 0) {
579 if (apply_permissions)
580 if (enforce_groups(context, username, uid) < 0) {
586 if (context->root_directory)
587 if (chroot(context->root_directory) < 0) {
592 if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
600 if (asprintf(&d, "%s/%s",
601 context->root_directory ? context->root_directory : "",
602 context->working_directory ? context->working_directory : "") < 0) {
616 if (close_all_fds(fds, n_fds) < 0 ||
617 shift_fds(fds, n_fds) < 0 ||
618 flags_fds(fds, n_fds, context->non_blocking) < 0) {
623 if (apply_permissions) {
625 for (i = 0; i < RLIMIT_NLIMITS; i++) {
626 if (!context->rlimit[i])
629 if (setrlimit(i, context->rlimit[i]) < 0) {
636 if (enforce_user(context, uid) < 0) {
641 /* PR_GET_SECUREBITS is not priviliged, while
642 * PR_SET_SECUREBITS is. So to suppress
643 * potential EPERMs we'll try not to call
644 * PR_SET_SECUREBITS unless necessary. */
645 if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
646 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
651 if (context->capabilities)
652 if (cap_set_proc(context->capabilities) < 0) {
653 r = EXIT_CAPABILITIES;
658 if (!(our_env = new0(char*, 6))) {
664 if (asprintf(our_env + n_env++, "LISTEN_PID=%llu", (unsigned long long) getpid()) < 0 ||
665 asprintf(our_env + n_env++, "LISTEN_FDS=%u", n_fds) < 0) {
671 if (asprintf(our_env + n_env++, "HOME=%s", home) < 0) {
677 if (asprintf(our_env + n_env++, "LOGNAME=%s", username) < 0 ||
678 asprintf(our_env + n_env++, "USER=%s", username) < 0) {
683 if (!(final_env = strv_env_merge(environ, our_env, context->environment, NULL))) {
688 execve(command->path, command->argv, final_env);
693 strv_free(final_env);
699 log_debug("Forked %s as %llu", command->path, (unsigned long long) pid);
705 void exec_context_init(ExecContext *c) {
710 c->oom_adjust_set = false;
713 c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
714 c->ioprio_set = false;
715 c->cpu_sched_policy = SCHED_OTHER;
716 c->cpu_sched_priority = 0;
717 c->cpu_sched_set = false;
718 CPU_ZERO(&c->cpu_affinity);
719 c->cpu_affinity_set = false;
720 c->timer_slack_ns = 0;
721 c->timer_slack_ns_set = false;
723 c->cpu_sched_reset_on_fork = false;
724 c->non_blocking = false;
725 c->new_session = false;
729 c->syslog_priority = LOG_DAEMON|LOG_INFO;
732 c->capability_bounding_set_drop = 0;
735 void exec_context_done(ExecContext *c) {
740 strv_free(c->environment);
741 c->environment = NULL;
743 for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
748 free(c->working_directory);
749 c->working_directory = NULL;
750 free(c->root_directory);
751 c->root_directory = NULL;
753 free(c->syslog_identifier);
754 c->syslog_identifier = NULL;
762 strv_free(c->supplementary_groups);
763 c->supplementary_groups = NULL;
765 if (c->capabilities) {
766 cap_free(c->capabilities);
767 c->capabilities = NULL;
771 void exec_command_free_list(ExecCommand *c) {
775 LIST_REMOVE(ExecCommand, command, c, i);
783 void exec_command_free_array(ExecCommand **c, unsigned n) {
786 for (i = 0; i < n; i++) {
787 exec_command_free_list(c[i]);
792 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
804 "%sWorkingDirectory: %s\n"
805 "%sRootDirectory: %s\n"
806 "%sNonBlocking: %s\n"
807 "%sNewSession: %s\n",
809 prefix, c->working_directory ? c->working_directory : "/",
810 prefix, c->root_directory ? c->root_directory : "/",
811 prefix, yes_no(c->non_blocking),
812 prefix, yes_no(c->new_session));
815 for (e = c->environment; *e; e++)
816 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
823 if (c->oom_adjust_set)
826 prefix, c->oom_adjust);
828 for (i = 0; i < RLIM_NLIMITS; i++)
830 fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
834 "%sIOSchedulingClass: %s\n"
835 "%sIOPriority: %i\n",
836 prefix, ioprio_class_to_string(IOPRIO_PRIO_CLASS(c->ioprio)),
837 prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
839 if (c->cpu_sched_set)
841 "%sCPUSchedulingPolicy: %s\n"
842 "%sCPUSchedulingPriority: %i\n"
843 "%sCPUSchedulingResetOnFork: %s\n",
844 prefix, sched_policy_to_string(c->cpu_sched_policy),
845 prefix, c->cpu_sched_priority,
846 prefix, yes_no(c->cpu_sched_reset_on_fork));
848 if (c->cpu_affinity_set) {
849 fprintf(f, "%sCPUAffinity:", prefix);
850 for (i = 0; i < CPU_SETSIZE; i++)
851 if (CPU_ISSET(i, &c->cpu_affinity))
852 fprintf(f, " %i", i);
856 if (c->timer_slack_ns_set)
857 fprintf(f, "%sTimerSlackNS: %lu\n", prefix, c->timer_slack_ns);
862 prefix, exec_input_to_string(c->input),
863 prefix, exec_output_to_string(c->output));
865 if (c->output == EXEC_OUTPUT_SYSLOG || c->output == EXEC_OUTPUT_KERNEL)
867 "%sSyslogFacility: %s\n"
868 "%sSyslogLevel: %s\n",
869 prefix, log_facility_to_string(LOG_FAC(c->syslog_priority)),
870 prefix, log_level_to_string(LOG_PRI(c->syslog_priority)));
872 if (c->capabilities) {
874 if ((t = cap_to_text(c->capabilities, NULL))) {
875 fprintf(f, "%sCapabilities: %s\n",
882 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
884 (c->secure_bits & SECURE_KEEP_CAPS) ? " keep-caps" : "",
885 (c->secure_bits & SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
886 (c->secure_bits & SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
887 (c->secure_bits & SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
888 (c->secure_bits & SECURE_NOROOT) ? " noroot" : "",
889 (c->secure_bits & SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
891 if (c->capability_bounding_set_drop) {
892 fprintf(f, "%sCapabilityBoundingSetDrop:", prefix);
894 for (i = 0; i <= CAP_LAST_CAP; i++)
895 if (c->capability_bounding_set_drop & (1 << i)) {
898 if ((t = cap_to_name(i))) {
899 fprintf(f, " %s", t);
908 fprintf(f, "%sUser: %s", prefix, c->user);
910 fprintf(f, "%sGroup: %s", prefix, c->group);
912 if (c->supplementary_groups) {
915 fprintf(f, "%sSupplementaryGroups:", prefix);
917 STRV_FOREACH(g, c->supplementary_groups)
918 fprintf(f, " %s", *g);
924 void exec_status_fill(ExecStatus *s, pid_t pid, int code, int status) {
930 s->timestamp = now(CLOCK_REALTIME);
933 char *exec_command_line(ExecCommand *c) {
942 STRV_FOREACH(a, c->argv)
945 if (!(n = new(char, k)))
949 STRV_FOREACH(a, c->argv) {
956 if (strpbrk(*a, WHITESPACE)) {
967 /* FIXME: this doesn't really handle arguments that have
968 * spaces and ticks in them */
973 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
982 cmd = exec_command_line(c);
985 "%sCommand Line: %s\n",
986 prefix, cmd ? cmd : strerror(ENOMEM));
991 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
997 LIST_FOREACH(command, c, c)
998 exec_command_dump(c, f, prefix);
1001 void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
1008 /* It's kinda important that we keep the order here */
1009 LIST_FIND_TAIL(ExecCommand, command, *l, end);
1010 LIST_INSERT_AFTER(ExecCommand, command, *l, end, e);
1015 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
1016 [EXEC_OUTPUT_CONSOLE] = "console",
1017 [EXEC_OUTPUT_NULL] = "null",
1018 [EXEC_OUTPUT_SYSLOG] = "syslog",
1019 [EXEC_OUTPUT_KERNEL] = "kernel"
1022 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
1024 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
1025 [EXEC_INPUT_NULL] = "null",
1026 [EXEC_INPUT_CONSOLE] = "console"
1029 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);