1 /*-*- Mode: C; c-basic-offset: 8 -*-*/
3 #pragma GCC diagnostic ignored "-Wattributes"
12 #include <sys/socket.h>
14 #include <sys/prctl.h>
23 #include "securebits.h"
25 static int close_fds(int except[], unsigned n_except) {
30 /* Modifies the fds array! (sorts it) */
32 if (!(d = opendir("/proc/self/fd")))
35 while ((de = readdir(d))) {
38 if (de->d_name[0] == '.')
41 if ((r = safe_atoi(de->d_name, &fd)) < 0)
55 for (i = 0; i < n_except; i++)
56 if (except[i] == fd) {
65 if ((r = close_nointr(fd)) < 0)
74 static int shift_fds(int fds[], unsigned n_fds) {
75 int start, restart_from;
88 for (i = start; i < (int) n_fds; i++) {
91 /* Already at right index? */
95 if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
98 assert_se(close_nointr(fds[i]) == 0);
101 /* Hmm, the fd we wanted isn't free? Then
102 * let's remember that and try again from here*/
103 if (nfd != i+3 && restart_from < 0)
107 if (restart_from < 0)
110 start = restart_from;
116 static int flags_fds(int fds[], unsigned n_fds) {
124 /* Drops O_NONBLOCK and FD_CLOEXEC from the file flags */
126 for (i = 0; i < n_fds; i++) {
129 if ((flags = fcntl(fds[i], F_GETFL, 0)) < 0)
132 /* Since we are at it, let's make sure that nobody
133 * forgot setting O_NONBLOCK for all our fds */
135 if (fcntl(fds[i], F_SETFL, flags &~O_NONBLOCK) < 0)
138 if ((flags = fcntl(fds[i], F_GETFD, 0)) < 0)
141 /* Also make sure nobody forgot O_CLOEXEC for all our
143 if (fcntl(fds[i], F_SETFD, flags &~FD_CLOEXEC) < 0)
150 static int replace_null_fd(int fd, int flags) {
156 if ((nfd = open("/dev/null", flags|O_NOCTTY)) < 0)
160 close_nointr_nofail(nfd);
167 static int setup_output(const ExecContext *context, const char *ident) {
172 switch (context->output) {
174 case EXEC_OUTPUT_CONSOLE:
177 case EXEC_OUTPUT_NULL:
179 if ((r = replace_null_fd(STDOUT_FILENO, O_WRONLY)) < 0 ||
180 (r = replace_null_fd(STDERR_FILENO, O_WRONLY)) < 0)
185 case EXEC_OUTPUT_KERNEL:
186 case EXEC_OUTPUT_SYSLOG: {
191 struct sockaddr_un un;
194 close_nointr(STDOUT_FILENO);
195 close_nointr(STDERR_FILENO);
197 if ((fd = socket(AF_UNIX, SOCK_STREAM, 0)) < 0)
200 if (fd != STDOUT_FILENO) {
201 close_nointr_nofail(fd);
206 sa.sa.sa_family = AF_UNIX;
207 strncpy(sa.un.sun_path+1, LOGGER_SOCKET, sizeof(sa.un.sun_path)-1);
209 if (connect(fd, &sa.sa, sizeof(sa)) < 0) {
210 close_nointr_nofail(fd);
214 if (shutdown(fd, SHUT_RD) < 0) {
215 close_nointr_nofail(fd);
219 if ((fd = dup(fd)) < 0) {
220 close_nointr_nofail(fd);
224 if (fd != STDERR_FILENO) {
225 close_nointr_nofail(fd);
229 /* We speak a very simple protocol between log server
230 * and client: one line for the log destination (kmsg
231 * or syslog), followed by the priority field,
232 * followed by the process name. Since we replaced
233 * stdin/stderr we simple use stdio to write to
234 * it. Note that we use stderr, to minimize buffer
235 * flushing issues. */
241 context->output == EXEC_OUTPUT_KERNEL ? "kmsg" : "syslog",
242 context->syslog_priority,
243 context->syslog_identifier ? context->syslog_identifier : ident);
249 assert_not_reached("Unknown output type");
253 int setup_input(const ExecContext *context) {
258 switch (context->input) {
260 case EXEC_INPUT_CONSOLE:
263 case EXEC_INPUT_NULL:
264 if ((r = replace_null_fd(STDIN_FILENO, O_RDONLY)) < 0)
270 assert_not_reached("Unknown input type");
274 int exec_spawn(const ExecCommand *command, const ExecContext *context, int *fds, unsigned n_fds, pid_t *ret) {
280 assert(fds || n_fds <= 0);
282 log_debug("about to execute %s", command->path);
284 if ((pid = fork()) < 0)
288 char **e, **f = NULL;
294 if (sigemptyset(&ss) < 0 ||
295 sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
296 r = EXIT_SIGNAL_MASK;
300 if (setpgid(0, 0) < 0) {
305 umask(context->umask);
307 if (setup_input(context) < 0) {
312 if (setup_output(context, file_name_from_path(command->path)) < 0) {
317 if (context->oom_adjust_set) {
320 snprintf(t, sizeof(t), "%i", context->oom_adjust);
323 if (write_one_line_file("/proc/self/oom_adj", t) < 0) {
329 if (context->root_directory)
330 if (chroot(context->root_directory) < 0) {
335 if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
340 if (context->nice_set)
341 if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
346 if (context->cpu_sched_set) {
347 struct sched_param param;
350 param.sched_priority = context->cpu_sched_priority;
352 if (sched_setscheduler(0, context->cpu_sched_policy |
353 (context->cpu_sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0), ¶m) < 0) {
354 r = EXIT_SETSCHEDULER;
359 if (context->cpu_affinity_set)
360 if (sched_setaffinity(0, sizeof(context->cpu_affinity), &context->cpu_affinity) < 0) {
361 r = EXIT_CPUAFFINITY;
365 if (context->ioprio_set)
366 if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
371 if (context->timer_slack_ns_set)
372 if (prctl(PR_SET_TIMERSLACK, context->timer_slack_ns_set) < 0) {
377 if (close_fds(fds, n_fds) < 0 ||
378 shift_fds(fds, n_fds) < 0 ||
379 flags_fds(fds, n_fds) < 0) {
384 for (i = 0; i < RLIMIT_NLIMITS; i++) {
385 if (!context->rlimit[i])
388 if (setrlimit(i, context->rlimit[i]) < 0) {
394 if (context->secure_bits) {
395 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
403 char *listen_env[3] = {
409 snprintf(a, sizeof(a), "LISTEN_PID=%llu", (unsigned long long) getpid());
410 snprintf(b, sizeof(b), "LISTEN_FDS=%u", n_fds);
415 if (context->environment) {
416 if (!(f = strv_merge(listen_env, context->environment))) {
425 e = context->environment;
427 execve(command->path, command->argv, e);
436 log_debug("executed %s as %llu", command->path, (unsigned long long) pid);
442 void exec_context_init(ExecContext *c) {
447 c->oom_adjust_set = false;
450 c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
451 c->ioprio_set = false;
452 c->cpu_sched_policy = SCHED_OTHER;
453 c->cpu_sched_priority = 0;
454 c->cpu_sched_set = false;
455 CPU_ZERO(&c->cpu_affinity);
456 c->cpu_affinity_set = false;
460 c->syslog_priority = LOG_DAEMON|LOG_INFO;
463 c->capability_bounding_set_drop = 0;
466 void exec_context_done(ExecContext *c) {
471 strv_free(c->environment);
472 c->environment = NULL;
474 for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
479 free(c->working_directory);
480 c->working_directory = NULL;
481 free(c->root_directory);
482 c->root_directory = NULL;
484 free(c->syslog_identifier);
485 c->syslog_identifier = NULL;
493 strv_free(c->supplementary_groups);
494 c->supplementary_groups = NULL;
496 if (c->capabilities) {
497 cap_free(c->capabilities);
498 c->capabilities = NULL;
502 void exec_command_free_list(ExecCommand *c) {
506 LIST_REMOVE(ExecCommand, command, c, i);
514 void exec_command_free_array(ExecCommand **c, unsigned n) {
517 for (i = 0; i < n; i++) {
518 exec_command_free_list(c[i]);
523 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
535 "%sWorkingDirectory: %s\n"
536 "%sRootDirectory: %s\n",
538 prefix, c->working_directory ? c->working_directory : "/",
539 prefix, c->root_directory ? c->root_directory : "/");
542 for (e = c->environment; *e; e++)
543 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
550 if (c->oom_adjust_set)
553 prefix, c->oom_adjust);
555 for (i = 0; i < RLIM_NLIMITS; i++)
557 fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
561 "%sIOSchedulingClass: %s\n"
562 "%sIOPriority: %i\n",
563 prefix, ioprio_class_to_string(IOPRIO_PRIO_CLASS(c->ioprio)),
564 prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
566 if (c->cpu_sched_set)
568 "%sCPUSchedulingPolicy: %s\n"
569 "%sCPUSchedulingPriority: %i\n"
570 "%sCPUSchedulingResetOnFork: %s\n",
571 prefix, sched_policy_to_string(c->cpu_sched_policy),
572 prefix, c->cpu_sched_priority,
573 prefix, yes_no(c->cpu_sched_reset_on_fork));
575 if (c->cpu_affinity_set) {
576 fprintf(f, "%sCPUAffinity:", prefix);
577 for (i = 0; i < CPU_SETSIZE; i++)
578 if (CPU_ISSET(i, &c->cpu_affinity))
579 fprintf(f, " %i", i);
583 if (c->timer_slack_ns_set)
584 fprintf(f, "%sTimerSlackNS: %lu\n", prefix, c->timer_slack_ns);
589 prefix, exec_input_to_string(c->input),
590 prefix, exec_output_to_string(c->output));
592 if (c->output == EXEC_OUTPUT_SYSLOG || c->output == EXEC_OUTPUT_KERNEL)
594 "%sSyslogFacility: %s\n"
595 "%sSyslogLevel: %s\n",
596 prefix, log_facility_to_string(LOG_FAC(c->syslog_priority)),
597 prefix, log_level_to_string(LOG_PRI(c->syslog_priority)));
599 if (c->capabilities) {
601 if ((t = cap_to_text(c->capabilities, NULL))) {
602 fprintf(f, "%sCapabilities: %s\n",
609 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
611 (c->secure_bits & SECURE_KEEP_CAPS) ? " keep-caps" : "",
612 (c->secure_bits & SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
613 (c->secure_bits & SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
614 (c->secure_bits & SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
615 (c->secure_bits & SECURE_NOROOT) ? " noroot" : "",
616 (c->secure_bits & SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
618 if (c->capability_bounding_set_drop) {
619 fprintf(f, "%sCapabilityBoundingSetDrop:", prefix);
621 for (i = 0; i <= CAP_LAST_CAP; i++)
622 if (c->capability_bounding_set_drop & (1 << i)) {
625 if ((t = cap_to_name(i))) {
626 fprintf(f, " %s", t);
635 fprintf(f, "%sUser: %s", prefix, c->user);
637 fprintf(f, "%sGroup: %s", prefix, c->group);
639 if (c->supplementary_groups) {
642 fprintf(f, "%sSupplementaryGroups:", prefix);
644 STRV_FOREACH(g, c->supplementary_groups)
645 fprintf(f, " %s", *g);
651 void exec_status_fill(ExecStatus *s, pid_t pid, int code, int status) {
657 s->timestamp = now(CLOCK_REALTIME);
660 char *exec_command_line(ExecCommand *c) {
669 STRV_FOREACH(a, c->argv)
672 if (!(n = new(char, k)))
676 STRV_FOREACH(a, c->argv) {
683 if (strpbrk(*a, WHITESPACE)) {
694 /* FIXME: this doesn't really handle arguments that have
695 * spaces and ticks in them */
700 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
709 cmd = exec_command_line(c);
712 "%sCommand Line: %s\n",
713 prefix, cmd ? cmd : strerror(ENOMEM));
718 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
724 LIST_FOREACH(command, c, c)
725 exec_command_dump(c, f, prefix);
728 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
729 [EXEC_OUTPUT_CONSOLE] = "console",
730 [EXEC_OUTPUT_NULL] = "null",
731 [EXEC_OUTPUT_SYSLOG] = "syslog",
732 [EXEC_OUTPUT_KERNEL] = "kernel"
735 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
737 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
738 [EXEC_INPUT_NULL] = "null",
739 [EXEC_INPUT_CONSOLE] = "console"
742 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);