]> git.ipfire.org Git - people/ms/systemd.git/blob - execute.c
execute: allow configuration of O_NONBLOCK flag from .service files
[people/ms/systemd.git] / execute.c
1 /*-*- Mode: C; c-basic-offset: 8 -*-*/
2
3 /***
4 This file is part of systemd.
5
6 Copyright 2010 Lennart Poettering
7
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.
12
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.
17
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/>.
20 ***/
21
22 #include <assert.h>
23 #include <dirent.h>
24 #include <errno.h>
25 #include <fcntl.h>
26 #include <unistd.h>
27 #include <string.h>
28 #include <signal.h>
29 #include <sys/socket.h>
30 #include <sys/un.h>
31 #include <sys/prctl.h>
32 #include <linux/sched.h>
33 #include <sys/types.h>
34 #include <sys/stat.h>
35
36 #include "execute.h"
37 #include "strv.h"
38 #include "macro.h"
39 #include "util.h"
40 #include "log.h"
41 #include "ioprio.h"
42 #include "securebits.h"
43
44 static int close_fds(int except[], unsigned n_except) {
45 DIR *d;
46 struct dirent *de;
47 int r = 0;
48
49 /* Modifies the fds array! (sorts it) */
50
51 if (!(d = opendir("/proc/self/fd")))
52 return -errno;
53
54 while ((de = readdir(d))) {
55 int fd;
56
57 if (de->d_name[0] == '.')
58 continue;
59
60 if ((r = safe_atoi(de->d_name, &fd)) < 0)
61 goto finish;
62
63 if (fd < 3)
64 continue;
65
66 if (fd == dirfd(d))
67 continue;
68
69 if (except) {
70 bool found;
71 unsigned i;
72
73 found = false;
74 for (i = 0; i < n_except; i++)
75 if (except[i] == fd) {
76 found = true;
77 break;
78 }
79
80 if (found)
81 continue;
82 }
83
84 if ((r = close_nointr(fd)) < 0)
85 goto finish;
86 }
87
88 finish:
89 closedir(d);
90 return r;
91 }
92
93 static int shift_fds(int fds[], unsigned n_fds) {
94 int start, restart_from;
95
96 if (n_fds <= 0)
97 return 0;
98
99 assert(fds);
100
101 start = 0;
102 for (;;) {
103 int i;
104
105 restart_from = -1;
106
107 for (i = start; i < (int) n_fds; i++) {
108 int nfd;
109
110 /* Already at right index? */
111 if (fds[i] == i+3)
112 continue;
113
114 if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
115 return -errno;
116
117 assert_se(close_nointr(fds[i]) == 0);
118 fds[i] = nfd;
119
120 /* Hmm, the fd we wanted isn't free? Then
121 * let's remember that and try again from here*/
122 if (nfd != i+3 && restart_from < 0)
123 restart_from = i;
124 }
125
126 if (restart_from < 0)
127 break;
128
129 start = restart_from;
130 }
131
132 return 0;
133 }
134
135 static int flags_fds(int fds[], unsigned n_fds, bool nonblock) {
136 unsigned i;
137
138 if (n_fds <= 0)
139 return 0;
140
141 assert(fds);
142
143 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
144
145 for (i = 0; i < n_fds; i++) {
146 int flags;
147
148 if ((flags = fcntl(fds[i], F_GETFL, 0)) < 0)
149 return -errno;
150
151 if (nonblock)
152 flags |= O_NONBLOCK;
153 else
154 flags &= ~O_NONBLOCK;
155
156 if (fcntl(fds[i], F_SETFL, flags) < 0)
157 return -errno;
158
159 /* We unconditionally drop FD_CLOEXEC from the fds,
160 * since after all we want to pass these fds to our
161 * children */
162 if ((flags = fcntl(fds[i], F_GETFD, 0)) < 0)
163 return -errno;
164
165 if (fcntl(fds[i], F_SETFD, flags &~FD_CLOEXEC) < 0)
166 return -errno;
167 }
168
169 return 0;
170 }
171
172 static int replace_null_fd(int fd, int flags) {
173 int nfd;
174 assert(fd >= 0);
175
176 close_nointr(fd);
177
178 if ((nfd = open("/dev/null", flags|O_NOCTTY)) < 0)
179 return -errno;
180
181 if (nfd != fd) {
182 close_nointr_nofail(nfd);
183 return -EIO;
184 }
185
186 return 0;
187 }
188
189 static int setup_output(const ExecContext *context, const char *ident) {
190 int r;
191
192 assert(context);
193
194 switch (context->output) {
195
196 case EXEC_OUTPUT_CONSOLE:
197 return 0;
198
199 case EXEC_OUTPUT_NULL:
200
201 if ((r = replace_null_fd(STDOUT_FILENO, O_WRONLY)) < 0 ||
202 (r = replace_null_fd(STDERR_FILENO, O_WRONLY)) < 0)
203 return r;
204
205 return 0;
206
207 case EXEC_OUTPUT_KERNEL:
208 case EXEC_OUTPUT_SYSLOG: {
209
210 int fd;
211 union {
212 struct sockaddr sa;
213 struct sockaddr_un un;
214 } sa;
215
216 close_nointr(STDOUT_FILENO);
217 close_nointr(STDERR_FILENO);
218
219 if ((fd = socket(AF_UNIX, SOCK_STREAM, 0)) < 0)
220 return -errno;
221
222 if (fd != STDOUT_FILENO) {
223 close_nointr_nofail(fd);
224 return -EIO;
225 }
226
227 zero(sa);
228 sa.sa.sa_family = AF_UNIX;
229 strncpy(sa.un.sun_path+1, LOGGER_SOCKET, sizeof(sa.un.sun_path)-1);
230
231 if (connect(fd, &sa.sa, sizeof(sa)) < 0) {
232 close_nointr_nofail(fd);
233 return -errno;
234 }
235
236 if (shutdown(fd, SHUT_RD) < 0) {
237 close_nointr_nofail(fd);
238 return -errno;
239 }
240
241 if ((fd = dup(fd)) < 0) {
242 close_nointr_nofail(fd);
243 return -errno;
244 }
245
246 if (fd != STDERR_FILENO) {
247 close_nointr_nofail(fd);
248 return -EIO;
249 }
250
251 /* We speak a very simple protocol between log server
252 * and client: one line for the log destination (kmsg
253 * or syslog), followed by the priority field,
254 * followed by the process name. Since we replaced
255 * stdin/stderr we simple use stdio to write to
256 * it. Note that we use stderr, to minimize buffer
257 * flushing issues. */
258
259 fprintf(stderr,
260 "%s\n"
261 "%i\n"
262 "%s\n",
263 context->output == EXEC_OUTPUT_KERNEL ? "kmsg" : "syslog",
264 context->syslog_priority,
265 context->syslog_identifier ? context->syslog_identifier : ident);
266
267 return 0;
268 }
269
270 default:
271 assert_not_reached("Unknown output type");
272 }
273 }
274
275 static int setup_input(const ExecContext *context) {
276 int r;
277
278 assert(context);
279
280 switch (context->input) {
281
282 case EXEC_INPUT_CONSOLE:
283 return 0;
284
285 case EXEC_INPUT_NULL:
286 if ((r = replace_null_fd(STDIN_FILENO, O_RDONLY)) < 0)
287 return r;
288
289 return 0;
290
291 default:
292 assert_not_reached("Unknown input type");
293 }
294 }
295
296 int exec_spawn(const ExecCommand *command, const ExecContext *context, int *fds, unsigned n_fds, pid_t *ret) {
297 pid_t pid;
298
299 assert(command);
300 assert(context);
301 assert(ret);
302 assert(fds || n_fds <= 0);
303
304 log_debug("about to execute %s", command->path);
305
306 if ((pid = fork()) < 0)
307 return -errno;
308
309 if (pid == 0) {
310 char **e, **f = NULL;
311 int i, r;
312 sigset_t ss;
313
314 /* child */
315
316 if (sigemptyset(&ss) < 0 ||
317 sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
318 r = EXIT_SIGNAL_MASK;
319 goto fail;
320 }
321
322 if (setpgid(0, 0) < 0) {
323 r = EXIT_PGID;
324 goto fail;
325 }
326
327 umask(context->umask);
328
329 if (setup_input(context) < 0) {
330 r = EXIT_INPUT;
331 goto fail;
332 }
333
334 if (setup_output(context, file_name_from_path(command->path)) < 0) {
335 r = EXIT_OUTPUT;
336 goto fail;
337 }
338
339 if (context->oom_adjust_set) {
340 char t[16];
341
342 snprintf(t, sizeof(t), "%i", context->oom_adjust);
343 char_array_0(t);
344
345 if (write_one_line_file("/proc/self/oom_adj", t) < 0) {
346 r = EXIT_OOM_ADJUST;
347 goto fail;
348 }
349 }
350
351 if (context->root_directory)
352 if (chroot(context->root_directory) < 0) {
353 r = EXIT_CHROOT;
354 goto fail;
355 }
356
357 if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
358 r = EXIT_CHDIR;
359 goto fail;
360 }
361
362 if (context->nice_set)
363 if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
364 r = EXIT_NICE;
365 goto fail;
366 }
367
368 if (context->cpu_sched_set) {
369 struct sched_param param;
370
371 zero(param);
372 param.sched_priority = context->cpu_sched_priority;
373
374 if (sched_setscheduler(0, context->cpu_sched_policy |
375 (context->cpu_sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0), &param) < 0) {
376 r = EXIT_SETSCHEDULER;
377 goto fail;
378 }
379 }
380
381 if (context->cpu_affinity_set)
382 if (sched_setaffinity(0, sizeof(context->cpu_affinity), &context->cpu_affinity) < 0) {
383 r = EXIT_CPUAFFINITY;
384 goto fail;
385 }
386
387 if (context->ioprio_set)
388 if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
389 r = EXIT_IOPRIO;
390 goto fail;
391 }
392
393 if (context->timer_slack_ns_set)
394 if (prctl(PR_SET_TIMERSLACK, context->timer_slack_ns_set) < 0) {
395 r = EXIT_TIMERSLACK;
396 goto fail;
397 }
398
399 if (close_fds(fds, n_fds) < 0 ||
400 shift_fds(fds, n_fds) < 0 ||
401 flags_fds(fds, n_fds, context->non_blocking) < 0) {
402 r = EXIT_FDS;
403 goto fail;
404 }
405
406 for (i = 0; i < RLIMIT_NLIMITS; i++) {
407 if (!context->rlimit[i])
408 continue;
409
410 if (setrlimit(i, context->rlimit[i]) < 0) {
411 r = EXIT_LIMITS;
412 goto fail;
413 }
414 }
415
416 if (context->secure_bits) {
417 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
418 r = EXIT_SECUREBITS;
419 goto fail;
420 }
421 }
422
423 if (n_fds > 0) {
424 char a[64], b[64];
425 char *listen_env[3] = {
426 a,
427 b,
428 NULL
429 };
430
431 snprintf(a, sizeof(a), "LISTEN_PID=%llu", (unsigned long long) getpid());
432 snprintf(b, sizeof(b), "LISTEN_FDS=%u", n_fds);
433
434 a[sizeof(a)-1] = 0;
435 b[sizeof(b)-1] = 0;
436
437 if (context->environment) {
438 if (!(f = strv_merge(listen_env, context->environment))) {
439 r = EXIT_MEMORY;
440 goto fail;
441 }
442 e = f;
443 } else
444 e = listen_env;
445
446 } else
447 e = context->environment;
448
449 execve(command->path, command->argv, e);
450 r = EXIT_EXEC;
451
452 fail:
453 strv_free(f);
454 _exit(r);
455 }
456
457
458 log_debug("executed %s as %llu", command->path, (unsigned long long) pid);
459
460 *ret = pid;
461 return 0;
462 }
463
464 void exec_context_init(ExecContext *c) {
465 assert(c);
466
467 c->umask = 0002;
468 c->oom_adjust = 0;
469 c->oom_adjust_set = false;
470 c->nice = 0;
471 c->nice_set = false;
472 c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
473 c->ioprio_set = false;
474 c->cpu_sched_policy = SCHED_OTHER;
475 c->cpu_sched_priority = 0;
476 c->cpu_sched_set = false;
477 CPU_ZERO(&c->cpu_affinity);
478 c->cpu_affinity_set = false;
479
480 c->input = 0;
481 c->output = 0;
482 c->syslog_priority = LOG_DAEMON|LOG_INFO;
483
484 c->secure_bits = 0;
485 c->capability_bounding_set_drop = 0;
486 }
487
488 void exec_context_done(ExecContext *c) {
489 unsigned l;
490
491 assert(c);
492
493 strv_free(c->environment);
494 c->environment = NULL;
495
496 for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
497 free(c->rlimit[l]);
498 c->rlimit[l] = NULL;
499 }
500
501 free(c->working_directory);
502 c->working_directory = NULL;
503 free(c->root_directory);
504 c->root_directory = NULL;
505
506 free(c->syslog_identifier);
507 c->syslog_identifier = NULL;
508
509 free(c->user);
510 c->user = NULL;
511
512 free(c->group);
513 c->group = NULL;
514
515 strv_free(c->supplementary_groups);
516 c->supplementary_groups = NULL;
517
518 if (c->capabilities) {
519 cap_free(c->capabilities);
520 c->capabilities = NULL;
521 }
522 }
523
524 void exec_command_free_list(ExecCommand *c) {
525 ExecCommand *i;
526
527 while ((i = c)) {
528 LIST_REMOVE(ExecCommand, command, c, i);
529
530 free(i->path);
531 strv_free(i->argv);
532 free(i);
533 }
534 }
535
536 void exec_command_free_array(ExecCommand **c, unsigned n) {
537 unsigned i;
538
539 for (i = 0; i < n; i++) {
540 exec_command_free_list(c[i]);
541 c[i] = NULL;
542 }
543 }
544
545 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
546 char ** e;
547 unsigned i;
548
549 assert(c);
550 assert(f);
551
552 if (!prefix)
553 prefix = "";
554
555 fprintf(f,
556 "%sUMask: %04o\n"
557 "%sWorkingDirectory: %s\n"
558 "%sRootDirectory: %s\n"
559 "%sNonBlocking: %s\n",
560 prefix, c->umask,
561 prefix, c->working_directory ? c->working_directory : "/",
562 prefix, c->root_directory ? c->root_directory : "/",
563 prefix, yes_no(c->non_blocking));
564
565 if (c->environment)
566 for (e = c->environment; *e; e++)
567 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
568
569 if (c->nice_set)
570 fprintf(f,
571 "%sNice: %i\n",
572 prefix, c->nice);
573
574 if (c->oom_adjust_set)
575 fprintf(f,
576 "%sOOMAdjust: %i\n",
577 prefix, c->oom_adjust);
578
579 for (i = 0; i < RLIM_NLIMITS; i++)
580 if (c->rlimit[i])
581 fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
582
583 if (c->ioprio_set)
584 fprintf(f,
585 "%sIOSchedulingClass: %s\n"
586 "%sIOPriority: %i\n",
587 prefix, ioprio_class_to_string(IOPRIO_PRIO_CLASS(c->ioprio)),
588 prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
589
590 if (c->cpu_sched_set)
591 fprintf(f,
592 "%sCPUSchedulingPolicy: %s\n"
593 "%sCPUSchedulingPriority: %i\n"
594 "%sCPUSchedulingResetOnFork: %s\n",
595 prefix, sched_policy_to_string(c->cpu_sched_policy),
596 prefix, c->cpu_sched_priority,
597 prefix, yes_no(c->cpu_sched_reset_on_fork));
598
599 if (c->cpu_affinity_set) {
600 fprintf(f, "%sCPUAffinity:", prefix);
601 for (i = 0; i < CPU_SETSIZE; i++)
602 if (CPU_ISSET(i, &c->cpu_affinity))
603 fprintf(f, " %i", i);
604 fputs("\n", f);
605 }
606
607 if (c->timer_slack_ns_set)
608 fprintf(f, "%sTimerSlackNS: %lu\n", prefix, c->timer_slack_ns);
609
610 fprintf(f,
611 "%sInput: %s\n"
612 "%sOutput: %s\n",
613 prefix, exec_input_to_string(c->input),
614 prefix, exec_output_to_string(c->output));
615
616 if (c->output == EXEC_OUTPUT_SYSLOG || c->output == EXEC_OUTPUT_KERNEL)
617 fprintf(f,
618 "%sSyslogFacility: %s\n"
619 "%sSyslogLevel: %s\n",
620 prefix, log_facility_to_string(LOG_FAC(c->syslog_priority)),
621 prefix, log_level_to_string(LOG_PRI(c->syslog_priority)));
622
623 if (c->capabilities) {
624 char *t;
625 if ((t = cap_to_text(c->capabilities, NULL))) {
626 fprintf(f, "%sCapabilities: %s\n",
627 prefix, t);
628 cap_free(t);
629 }
630 }
631
632 if (c->secure_bits)
633 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
634 prefix,
635 (c->secure_bits & SECURE_KEEP_CAPS) ? " keep-caps" : "",
636 (c->secure_bits & SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
637 (c->secure_bits & SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
638 (c->secure_bits & SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
639 (c->secure_bits & SECURE_NOROOT) ? " noroot" : "",
640 (c->secure_bits & SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
641
642 if (c->capability_bounding_set_drop) {
643 fprintf(f, "%sCapabilityBoundingSetDrop:", prefix);
644
645 for (i = 0; i <= CAP_LAST_CAP; i++)
646 if (c->capability_bounding_set_drop & (1 << i)) {
647 char *t;
648
649 if ((t = cap_to_name(i))) {
650 fprintf(f, " %s", t);
651 free(t);
652 }
653 }
654
655 fputs("\n", f);
656 }
657
658 if (c->user)
659 fprintf(f, "%sUser: %s", prefix, c->user);
660 if (c->group)
661 fprintf(f, "%sGroup: %s", prefix, c->group);
662
663 if (c->supplementary_groups) {
664 char **g;
665
666 fprintf(f, "%sSupplementaryGroups:", prefix);
667
668 STRV_FOREACH(g, c->supplementary_groups)
669 fprintf(f, " %s", *g);
670
671 fputs("\n", f);
672 }
673 }
674
675 void exec_status_fill(ExecStatus *s, pid_t pid, int code, int status) {
676 assert(s);
677
678 s->pid = pid;
679 s->code = code;
680 s->status = status;
681 s->timestamp = now(CLOCK_REALTIME);
682 }
683
684 char *exec_command_line(ExecCommand *c) {
685 size_t k;
686 char *n, *p, **a;
687 bool first = true;
688
689 assert(c);
690 assert(c->argv);
691
692 k = 1;
693 STRV_FOREACH(a, c->argv)
694 k += strlen(*a)+3;
695
696 if (!(n = new(char, k)))
697 return NULL;
698
699 p = n;
700 STRV_FOREACH(a, c->argv) {
701
702 if (!first)
703 *(p++) = ' ';
704 else
705 first = false;
706
707 if (strpbrk(*a, WHITESPACE)) {
708 *(p++) = '\'';
709 p = stpcpy(p, *a);
710 *(p++) = '\'';
711 } else
712 p = stpcpy(p, *a);
713
714 }
715
716 *p = 0;
717
718 /* FIXME: this doesn't really handle arguments that have
719 * spaces and ticks in them */
720
721 return n;
722 }
723
724 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
725 char *cmd;
726
727 assert(c);
728 assert(f);
729
730 if (!prefix)
731 prefix = "";
732
733 cmd = exec_command_line(c);
734
735 fprintf(f,
736 "%sCommand Line: %s\n",
737 prefix, cmd ? cmd : strerror(ENOMEM));
738
739 free(cmd);
740 }
741
742 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
743 assert(f);
744
745 if (!prefix)
746 prefix = "";
747
748 LIST_FOREACH(command, c, c)
749 exec_command_dump(c, f, prefix);
750 }
751
752 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
753 [EXEC_OUTPUT_CONSOLE] = "console",
754 [EXEC_OUTPUT_NULL] = "null",
755 [EXEC_OUTPUT_SYSLOG] = "syslog",
756 [EXEC_OUTPUT_KERNEL] = "kernel"
757 };
758
759 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
760
761 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
762 [EXEC_INPUT_NULL] = "null",
763 [EXEC_INPUT_CONSOLE] = "console"
764 };
765
766 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);