]> git.ipfire.org Git - people/ms/systemd.git/blob - execute.c
use linux/sched.h instead of sched.h for older glibc
[people/ms/systemd.git] / execute.c
1 /*-*- Mode: C; c-basic-offset: 8 -*-*/
2
3 #pragma GCC diagnostic ignored "-Wattributes"
4
5 #include <assert.h>
6 #include <dirent.h>
7 #include <errno.h>
8 #include <fcntl.h>
9 #include <unistd.h>
10 #include <string.h>
11 #include <signal.h>
12 #include <sys/socket.h>
13 #include <sys/un.h>
14 #include <sys/prctl.h>
15 #include <linux/sched.h>
16
17 #include "execute.h"
18 #include "strv.h"
19 #include "macro.h"
20 #include "util.h"
21 #include "log.h"
22 #include "ioprio.h"
23 #include "securebits.h"
24
25 static int close_fds(int except[], unsigned n_except) {
26 DIR *d;
27 struct dirent *de;
28 int r = 0;
29
30 /* Modifies the fds array! (sorts it) */
31
32 if (!(d = opendir("/proc/self/fd")))
33 return -errno;
34
35 while ((de = readdir(d))) {
36 int fd;
37
38 if (de->d_name[0] == '.')
39 continue;
40
41 if ((r = safe_atoi(de->d_name, &fd)) < 0)
42 goto finish;
43
44 if (fd < 3)
45 continue;
46
47 if (fd == dirfd(d))
48 continue;
49
50 if (except) {
51 bool found;
52 unsigned i;
53
54 found = false;
55 for (i = 0; i < n_except; i++)
56 if (except[i] == fd) {
57 found = true;
58 break;
59 }
60
61 if (found)
62 continue;
63 }
64
65 if ((r = close_nointr(fd)) < 0)
66 goto finish;
67 }
68
69 finish:
70 closedir(d);
71 return r;
72 }
73
74 static int shift_fds(int fds[], unsigned n_fds) {
75 int start, restart_from;
76
77 if (n_fds <= 0)
78 return 0;
79
80 assert(fds);
81
82 start = 0;
83 for (;;) {
84 int i;
85
86 restart_from = -1;
87
88 for (i = start; i < (int) n_fds; i++) {
89 int nfd;
90
91 /* Already at right index? */
92 if (fds[i] == i+3)
93 continue;
94
95 if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
96 return -errno;
97
98 assert_se(close_nointr(fds[i]) == 0);
99 fds[i] = nfd;
100
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)
104 restart_from = i;
105 }
106
107 if (restart_from < 0)
108 break;
109
110 start = restart_from;
111 }
112
113 return 0;
114 }
115
116 static int flags_fds(int fds[], unsigned n_fds) {
117 unsigned i;
118
119 if (n_fds <= 0)
120 return 0;
121
122 assert(fds);
123
124 /* Drops O_NONBLOCK and FD_CLOEXEC from the file flags */
125
126 for (i = 0; i < n_fds; i++) {
127 int flags;
128
129 if ((flags = fcntl(fds[i], F_GETFL, 0)) < 0)
130 return -errno;
131
132 /* Since we are at it, let's make sure that nobody
133 * forgot setting O_NONBLOCK for all our fds */
134
135 if (fcntl(fds[i], F_SETFL, flags &~O_NONBLOCK) < 0)
136 return -errno;
137
138 if ((flags = fcntl(fds[i], F_GETFD, 0)) < 0)
139 return -errno;
140
141 /* Also make sure nobody forgot O_CLOEXEC for all our
142 * fds */
143 if (fcntl(fds[i], F_SETFD, flags &~FD_CLOEXEC) < 0)
144 return -errno;
145 }
146
147 return 0;
148 }
149
150 static int replace_null_fd(int fd, int flags) {
151 int nfd;
152 assert(fd >= 0);
153
154 close_nointr(fd);
155
156 if ((nfd = open("/dev/null", flags|O_NOCTTY)) < 0)
157 return -errno;
158
159 if (nfd != fd) {
160 close_nointr_nofail(nfd);
161 return -EIO;
162 }
163
164 return 0;
165 }
166
167 static int setup_output(const ExecContext *context, const char *ident) {
168 int r;
169
170 assert(context);
171
172 switch (context->output) {
173
174 case EXEC_OUTPUT_CONSOLE:
175 return 0;
176
177 case EXEC_OUTPUT_NULL:
178
179 if ((r = replace_null_fd(STDOUT_FILENO, O_WRONLY)) < 0 ||
180 (r = replace_null_fd(STDERR_FILENO, O_WRONLY)) < 0)
181 return r;
182
183 return 0;
184
185 case EXEC_OUTPUT_KERNEL:
186 case EXEC_OUTPUT_SYSLOG: {
187
188 int fd;
189 union {
190 struct sockaddr sa;
191 struct sockaddr_un un;
192 } sa;
193
194 close_nointr(STDOUT_FILENO);
195 close_nointr(STDERR_FILENO);
196
197 if ((fd = socket(AF_UNIX, SOCK_STREAM, 0)) < 0)
198 return -errno;
199
200 if (fd != STDOUT_FILENO) {
201 close_nointr_nofail(fd);
202 return -EIO;
203 }
204
205 zero(sa);
206 sa.sa.sa_family = AF_UNIX;
207 strncpy(sa.un.sun_path+1, LOGGER_SOCKET, sizeof(sa.un.sun_path)-1);
208
209 if (connect(fd, &sa.sa, sizeof(sa)) < 0) {
210 close_nointr_nofail(fd);
211 return -errno;
212 }
213
214 if (shutdown(fd, SHUT_RD) < 0) {
215 close_nointr_nofail(fd);
216 return -errno;
217 }
218
219 if ((fd = dup(fd)) < 0) {
220 close_nointr_nofail(fd);
221 return -errno;
222 }
223
224 if (fd != STDERR_FILENO) {
225 close_nointr_nofail(fd);
226 return -EIO;
227 }
228
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. */
236
237 fprintf(stderr,
238 "%s\n"
239 "%i\n"
240 "%s\n",
241 context->output == EXEC_OUTPUT_KERNEL ? "kmsg" : "syslog",
242 context->syslog_priority,
243 context->syslog_identifier ? context->syslog_identifier : ident);
244
245 return 0;
246 }
247
248 default:
249 assert_not_reached("Unknown output type");
250 }
251 }
252
253 int setup_input(const ExecContext *context) {
254 int r;
255
256 assert(context);
257
258 switch (context->input) {
259
260 case EXEC_INPUT_CONSOLE:
261 return 0;
262
263 case EXEC_INPUT_NULL:
264 if ((r = replace_null_fd(STDIN_FILENO, O_RDONLY)) < 0)
265 return r;
266
267 return 0;
268
269 default:
270 assert_not_reached("Unknown input type");
271 }
272 }
273
274 int exec_spawn(const ExecCommand *command, const ExecContext *context, int *fds, unsigned n_fds, pid_t *ret) {
275 pid_t pid;
276
277 assert(command);
278 assert(context);
279 assert(ret);
280 assert(fds || n_fds <= 0);
281
282 log_debug("about to execute %s", command->path);
283
284 if ((pid = fork()) < 0)
285 return -errno;
286
287 if (pid == 0) {
288 char **e, **f = NULL;
289 int i, r;
290 sigset_t ss;
291
292 /* child */
293
294 if (sigemptyset(&ss) < 0 ||
295 sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
296 r = EXIT_SIGNAL_MASK;
297 goto fail;
298 }
299
300 if (setpgid(0, 0) < 0) {
301 r = EXIT_PGID;
302 goto fail;
303 }
304
305 umask(context->umask);
306
307 if (setup_input(context) < 0) {
308 r = EXIT_INPUT;
309 goto fail;
310 }
311
312 if (setup_output(context, file_name_from_path(command->path)) < 0) {
313 r = EXIT_OUTPUT;
314 goto fail;
315 }
316
317 if (context->oom_adjust_set) {
318 char t[16];
319
320 snprintf(t, sizeof(t), "%i", context->oom_adjust);
321 char_array_0(t);
322
323 if (write_one_line_file("/proc/self/oom_adj", t) < 0) {
324 r = EXIT_OOM_ADJUST;
325 goto fail;
326 }
327 }
328
329 if (context->root_directory)
330 if (chroot(context->root_directory) < 0) {
331 r = EXIT_CHROOT;
332 goto fail;
333 }
334
335 if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
336 r = EXIT_CHDIR;
337 goto fail;
338 }
339
340 if (context->nice_set)
341 if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
342 r = EXIT_NICE;
343 goto fail;
344 }
345
346 if (context->cpu_sched_set) {
347 struct sched_param param;
348
349 zero(param);
350 param.sched_priority = context->cpu_sched_priority;
351
352 if (sched_setscheduler(0, context->cpu_sched_policy |
353 (context->cpu_sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0), &param) < 0) {
354 r = EXIT_SETSCHEDULER;
355 goto fail;
356 }
357 }
358
359 if (context->cpu_affinity_set)
360 if (sched_setaffinity(0, sizeof(context->cpu_affinity), &context->cpu_affinity) < 0) {
361 r = EXIT_CPUAFFINITY;
362 goto fail;
363 }
364
365 if (context->ioprio_set)
366 if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
367 r = EXIT_IOPRIO;
368 goto fail;
369 }
370
371 if (context->timer_slack_ns_set)
372 if (prctl(PR_SET_TIMERSLACK, context->timer_slack_ns_set) < 0) {
373 r = EXIT_TIMERSLACK;
374 goto fail;
375 }
376
377 if (close_fds(fds, n_fds) < 0 ||
378 shift_fds(fds, n_fds) < 0 ||
379 flags_fds(fds, n_fds) < 0) {
380 r = EXIT_FDS;
381 goto fail;
382 }
383
384 for (i = 0; i < RLIMIT_NLIMITS; i++) {
385 if (!context->rlimit[i])
386 continue;
387
388 if (setrlimit(i, context->rlimit[i]) < 0) {
389 r = EXIT_LIMITS;
390 goto fail;
391 }
392 }
393
394 if (context->secure_bits) {
395 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
396 r = EXIT_SECUREBITS;
397 goto fail;
398 }
399 }
400
401 if (n_fds > 0) {
402 char a[64], b[64];
403 char *listen_env[3] = {
404 a,
405 b,
406 NULL
407 };
408
409 snprintf(a, sizeof(a), "LISTEN_PID=%llu", (unsigned long long) getpid());
410 snprintf(b, sizeof(b), "LISTEN_FDS=%u", n_fds);
411
412 a[sizeof(a)-1] = 0;
413 b[sizeof(b)-1] = 0;
414
415 if (context->environment) {
416 if (!(f = strv_merge(listen_env, context->environment))) {
417 r = EXIT_MEMORY;
418 goto fail;
419 }
420 e = f;
421 } else
422 e = listen_env;
423
424 } else
425 e = context->environment;
426
427 execve(command->path, command->argv, e);
428 r = EXIT_EXEC;
429
430 fail:
431 strv_free(f);
432 _exit(r);
433 }
434
435
436 log_debug("executed %s as %llu", command->path, (unsigned long long) pid);
437
438 *ret = pid;
439 return 0;
440 }
441
442 void exec_context_init(ExecContext *c) {
443 assert(c);
444
445 c->umask = 0002;
446 c->oom_adjust = 0;
447 c->oom_adjust_set = false;
448 c->nice = 0;
449 c->nice_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;
457
458 c->input = 0;
459 c->output = 0;
460 c->syslog_priority = LOG_DAEMON|LOG_INFO;
461
462 c->secure_bits = 0;
463 c->capability_bounding_set_drop = 0;
464 }
465
466 void exec_context_done(ExecContext *c) {
467 unsigned l;
468
469 assert(c);
470
471 strv_free(c->environment);
472 c->environment = NULL;
473
474 for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
475 free(c->rlimit[l]);
476 c->rlimit[l] = NULL;
477 }
478
479 free(c->working_directory);
480 c->working_directory = NULL;
481 free(c->root_directory);
482 c->root_directory = NULL;
483
484 free(c->syslog_identifier);
485 c->syslog_identifier = NULL;
486
487 free(c->user);
488 c->user = NULL;
489
490 free(c->group);
491 c->group = NULL;
492
493 strv_free(c->supplementary_groups);
494 c->supplementary_groups = NULL;
495
496 if (c->capabilities) {
497 cap_free(c->capabilities);
498 c->capabilities = NULL;
499 }
500 }
501
502 void exec_command_free_list(ExecCommand *c) {
503 ExecCommand *i;
504
505 while ((i = c)) {
506 LIST_REMOVE(ExecCommand, command, c, i);
507
508 free(i->path);
509 strv_free(i->argv);
510 free(i);
511 }
512 }
513
514 void exec_command_free_array(ExecCommand **c, unsigned n) {
515 unsigned i;
516
517 for (i = 0; i < n; i++) {
518 exec_command_free_list(c[i]);
519 c[i] = NULL;
520 }
521 }
522
523 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
524 char ** e;
525 unsigned i;
526
527 assert(c);
528 assert(f);
529
530 if (!prefix)
531 prefix = "";
532
533 fprintf(f,
534 "%sUMask: %04o\n"
535 "%sWorkingDirectory: %s\n"
536 "%sRootDirectory: %s\n",
537 prefix, c->umask,
538 prefix, c->working_directory ? c->working_directory : "/",
539 prefix, c->root_directory ? c->root_directory : "/");
540
541 if (c->environment)
542 for (e = c->environment; *e; e++)
543 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
544
545 if (c->nice_set)
546 fprintf(f,
547 "%sNice: %i\n",
548 prefix, c->nice);
549
550 if (c->oom_adjust_set)
551 fprintf(f,
552 "%sOOMAdjust: %i\n",
553 prefix, c->oom_adjust);
554
555 for (i = 0; i < RLIM_NLIMITS; i++)
556 if (c->rlimit[i])
557 fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
558
559 if (c->ioprio_set)
560 fprintf(f,
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));
565
566 if (c->cpu_sched_set)
567 fprintf(f,
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));
574
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);
580 fputs("\n", f);
581 }
582
583 if (c->timer_slack_ns_set)
584 fprintf(f, "%sTimerSlackNS: %lu\n", prefix, c->timer_slack_ns);
585
586 fprintf(f,
587 "%sInput: %s\n"
588 "%sOutput: %s\n",
589 prefix, exec_input_to_string(c->input),
590 prefix, exec_output_to_string(c->output));
591
592 if (c->output == EXEC_OUTPUT_SYSLOG || c->output == EXEC_OUTPUT_KERNEL)
593 fprintf(f,
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)));
598
599 if (c->capabilities) {
600 char *t;
601 if ((t = cap_to_text(c->capabilities, NULL))) {
602 fprintf(f, "%sCapabilities: %s\n",
603 prefix, t);
604 cap_free(t);
605 }
606 }
607
608 if (c->secure_bits)
609 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
610 prefix,
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" : "");
617
618 if (c->capability_bounding_set_drop) {
619 fprintf(f, "%sCapabilityBoundingSetDrop:", prefix);
620
621 for (i = 0; i <= CAP_LAST_CAP; i++)
622 if (c->capability_bounding_set_drop & (1 << i)) {
623 char *t;
624
625 if ((t = cap_to_name(i))) {
626 fprintf(f, " %s", t);
627 free(t);
628 }
629 }
630
631 fputs("\n", f);
632 }
633
634 if (c->user)
635 fprintf(f, "%sUser: %s", prefix, c->user);
636 if (c->group)
637 fprintf(f, "%sGroup: %s", prefix, c->group);
638
639 if (c->supplementary_groups) {
640 char **g;
641
642 fprintf(f, "%sSupplementaryGroups:", prefix);
643
644 STRV_FOREACH(g, c->supplementary_groups)
645 fprintf(f, " %s", *g);
646
647 fputs("\n", f);
648 }
649 }
650
651 void exec_status_fill(ExecStatus *s, pid_t pid, int code, int status) {
652 assert(s);
653
654 s->pid = pid;
655 s->code = code;
656 s->status = status;
657 s->timestamp = now(CLOCK_REALTIME);
658 }
659
660 char *exec_command_line(ExecCommand *c) {
661 size_t k;
662 char *n, *p, **a;
663 bool first = true;
664
665 assert(c);
666 assert(c->argv);
667
668 k = 1;
669 STRV_FOREACH(a, c->argv)
670 k += strlen(*a)+3;
671
672 if (!(n = new(char, k)))
673 return NULL;
674
675 p = n;
676 STRV_FOREACH(a, c->argv) {
677
678 if (!first)
679 *(p++) = ' ';
680 else
681 first = false;
682
683 if (strpbrk(*a, WHITESPACE)) {
684 *(p++) = '\'';
685 p = stpcpy(p, *a);
686 *(p++) = '\'';
687 } else
688 p = stpcpy(p, *a);
689
690 }
691
692 *p = 0;
693
694 /* FIXME: this doesn't really handle arguments that have
695 * spaces and ticks in them */
696
697 return n;
698 }
699
700 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
701 char *cmd;
702
703 assert(c);
704 assert(f);
705
706 if (!prefix)
707 prefix = "";
708
709 cmd = exec_command_line(c);
710
711 fprintf(f,
712 "%sCommand Line: %s\n",
713 prefix, cmd ? cmd : strerror(ENOMEM));
714
715 free(cmd);
716 }
717
718 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
719 assert(f);
720
721 if (!prefix)
722 prefix = "";
723
724 LIST_FOREACH(command, c, c)
725 exec_command_dump(c, f, prefix);
726 }
727
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"
733 };
734
735 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
736
737 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
738 [EXEC_INPUT_NULL] = "null",
739 [EXEC_INPUT_CONSOLE] = "console"
740 };
741
742 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);