]> git.ipfire.org Git - people/ms/systemd.git/blob - execute.c
Merge remote branch 'kay/master'
[people/ms/systemd.git] / execute.c
1 /*-*- Mode: C; c-basic-offset: 8 -*-*/
2
3 #include <assert.h>
4 #include <dirent.h>
5 #include <errno.h>
6 #include <fcntl.h>
7 #include <unistd.h>
8 #include <string.h>
9 #include <signal.h>
10 #include <sys/socket.h>
11 #include <sys/un.h>
12 #include <sys/prctl.h>
13 #include <sched.h>
14
15 #include "execute.h"
16 #include "strv.h"
17 #include "macro.h"
18 #include "util.h"
19 #include "log.h"
20 #include "ioprio.h"
21 #include "securebits.h"
22
23 static int close_fds(int except[], unsigned n_except) {
24 DIR *d;
25 struct dirent *de;
26 int r = 0;
27
28 /* Modifies the fds array! (sorts it) */
29
30 if (!(d = opendir("/proc/self/fd")))
31 return -errno;
32
33 while ((de = readdir(d))) {
34 int fd;
35
36 if (de->d_name[0] == '.')
37 continue;
38
39 if ((r = safe_atoi(de->d_name, &fd)) < 0)
40 goto finish;
41
42 if (fd < 3)
43 continue;
44
45 if (fd == dirfd(d))
46 continue;
47
48 if (except) {
49 bool found;
50 unsigned i;
51
52 found = false;
53 for (i = 0; i < n_except; i++)
54 if (except[i] == fd) {
55 found = true;
56 break;
57 }
58
59 if (found)
60 continue;
61 }
62
63 if ((r = close_nointr(fd)) < 0)
64 goto finish;
65 }
66
67 finish:
68 closedir(d);
69 return r;
70 }
71
72 static int shift_fds(int fds[], unsigned n_fds) {
73 int start, restart_from;
74
75 if (n_fds <= 0)
76 return 0;
77
78 assert(fds);
79
80 start = 0;
81 for (;;) {
82 int i;
83
84 restart_from = -1;
85
86 for (i = start; i < (int) n_fds; i++) {
87 int nfd;
88
89 /* Already at right index? */
90 if (fds[i] == i+3)
91 continue;
92
93 if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
94 return -errno;
95
96 assert_se(close_nointr(fds[i]) == 0);
97 fds[i] = nfd;
98
99 /* Hmm, the fd we wanted isn't free? Then
100 * let's remember that and try again from here*/
101 if (nfd != i+3 && restart_from < 0)
102 restart_from = i;
103 }
104
105 if (restart_from < 0)
106 break;
107
108 start = restart_from;
109 }
110
111 return 0;
112 }
113
114 static int flags_fds(int fds[], unsigned n_fds) {
115 unsigned i;
116
117 if (n_fds <= 0)
118 return 0;
119
120 assert(fds);
121
122 /* Drops O_NONBLOCK and FD_CLOEXEC from the file flags */
123
124 for (i = 0; i < n_fds; i++) {
125 int flags;
126
127 if ((flags = fcntl(fds[i], F_GETFL, 0)) < 0)
128 return -errno;
129
130 /* Since we are at it, let's make sure that nobody
131 * forgot setting O_NONBLOCK for all our fds */
132
133 if (fcntl(fds[i], F_SETFL, flags &~O_NONBLOCK) < 0)
134 return -errno;
135
136 if ((flags = fcntl(fds[i], F_GETFD, 0)) < 0)
137 return -errno;
138
139 /* Also make sure nobody forgot O_CLOEXEC for all our
140 * fds */
141 if (fcntl(fds[i], F_SETFD, flags &~FD_CLOEXEC) < 0)
142 return -errno;
143 }
144
145 return 0;
146 }
147
148 static int replace_null_fd(int fd, int flags) {
149 int nfd;
150 assert(fd >= 0);
151
152 close_nointr(fd);
153
154 if ((nfd = open("/dev/null", flags|O_NOCTTY)) < 0)
155 return -errno;
156
157 if (nfd != fd) {
158 close_nointr_nofail(nfd);
159 return -EIO;
160 }
161
162 return 0;
163 }
164
165 static int setup_output(const ExecContext *context, const char *ident) {
166 int r;
167
168 assert(context);
169
170 switch (context->output) {
171
172 case EXEC_OUTPUT_CONSOLE:
173 return 0;
174
175 case EXEC_OUTPUT_NULL:
176
177 if ((r = replace_null_fd(STDOUT_FILENO, O_WRONLY)) < 0 ||
178 (r = replace_null_fd(STDERR_FILENO, O_WRONLY)) < 0)
179 return r;
180
181 return 0;
182
183 case EXEC_OUTPUT_KERNEL:
184 case EXEC_OUTPUT_SYSLOG: {
185
186 int fd;
187 union {
188 struct sockaddr sa;
189 struct sockaddr_un un;
190 } sa;
191
192 close_nointr(STDOUT_FILENO);
193 close_nointr(STDERR_FILENO);
194
195 if ((fd = socket(AF_UNIX, SOCK_STREAM, 0)) < 0)
196 return -errno;
197
198 if (fd != STDOUT_FILENO) {
199 close_nointr_nofail(fd);
200 return -EIO;
201 }
202
203 zero(sa);
204 sa.sa.sa_family = AF_UNIX;
205 strncpy(sa.un.sun_path+1, LOGGER_SOCKET, sizeof(sa.un.sun_path)-1);
206
207 if (connect(fd, &sa.sa, sizeof(sa)) < 0) {
208 close_nointr_nofail(fd);
209 return -errno;
210 }
211
212 if (shutdown(fd, SHUT_RD) < 0) {
213 close_nointr_nofail(fd);
214 return -errno;
215 }
216
217 if ((fd = dup(fd)) < 0) {
218 close_nointr_nofail(fd);
219 return -errno;
220 }
221
222 if (fd != STDERR_FILENO) {
223 close_nointr_nofail(fd);
224 return -EIO;
225 }
226
227 /* We speak a very simple protocol between log server
228 * and client: one line for the log destination (kmsg
229 * or syslog), followed by the priority field,
230 * followed by the process name. Since we replaced
231 * stdin/stderr we simple use stdio to write to
232 * it. Note that we use stderr, to minimize buffer
233 * flushing issues. */
234
235 fprintf(stderr,
236 "%s\n"
237 "%i\n"
238 "%s\n",
239 context->output == EXEC_OUTPUT_KERNEL ? "kmsg" : "syslog",
240 context->syslog_priority,
241 context->syslog_identifier ? context->syslog_identifier : ident);
242
243 return 0;
244 }
245
246 default:
247 assert_not_reached("Unknown output type");
248 }
249 }
250
251 int setup_input(const ExecContext *context) {
252 int r;
253
254 assert(context);
255
256 switch (context->input) {
257
258 case EXEC_INPUT_CONSOLE:
259 return 0;
260
261 case EXEC_INPUT_NULL:
262 if ((r = replace_null_fd(STDIN_FILENO, O_RDONLY)) < 0)
263 return r;
264
265 return 0;
266
267 default:
268 assert_not_reached("Unknown input type");
269 }
270 }
271
272 int exec_spawn(const ExecCommand *command, const ExecContext *context, int *fds, unsigned n_fds, pid_t *ret) {
273 pid_t pid;
274
275 assert(command);
276 assert(context);
277 assert(ret);
278 assert(fds || n_fds <= 0);
279
280 log_debug("about to execute %s", command->path);
281
282 if ((pid = fork()) < 0)
283 return -errno;
284
285 if (pid == 0) {
286 char **e, **f = NULL;
287 int i, r;
288 sigset_t ss;
289
290 /* child */
291
292 if (sigemptyset(&ss) < 0 ||
293 sigprocmask(SIG_SETMASK, &ss, NULL) < 0) {
294 r = EXIT_SIGNAL_MASK;
295 goto fail;
296 }
297
298 if (setpgid(0, 0) < 0) {
299 r = EXIT_PGID;
300 goto fail;
301 }
302
303 umask(context->umask);
304
305 if (setup_input(context) < 0) {
306 r = EXIT_INPUT;
307 goto fail;
308 }
309
310 if (setup_output(context, file_name_from_path(command->path)) < 0) {
311 r = EXIT_OUTPUT;
312 goto fail;
313 }
314
315 if (context->oom_adjust_set) {
316 char t[16];
317
318 snprintf(t, sizeof(t), "%i", context->oom_adjust);
319 char_array_0(t);
320
321 if (write_one_line_file("/proc/self/oom_adj", t) < 0) {
322 r = EXIT_OOM_ADJUST;
323 goto fail;
324 }
325 }
326
327 if (context->root_directory)
328 if (chroot(context->root_directory) < 0) {
329 r = EXIT_CHROOT;
330 goto fail;
331 }
332
333 if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
334 r = EXIT_CHDIR;
335 goto fail;
336 }
337
338 if (context->nice_set)
339 if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
340 r = EXIT_NICE;
341 goto fail;
342 }
343
344 if (context->cpu_sched_set) {
345 struct sched_param param;
346
347 zero(param);
348 param.sched_priority = context->cpu_sched_priority;
349
350 if (sched_setscheduler(0, context->cpu_sched_policy |
351 (context->cpu_sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0), &param) < 0) {
352 r = EXIT_SETSCHEDULER;
353 goto fail;
354 }
355 }
356
357 if (context->cpu_affinity_set)
358 if (sched_setaffinity(0, sizeof(context->cpu_affinity), &context->cpu_affinity) < 0) {
359 r = EXIT_CPUAFFINITY;
360 goto fail;
361 }
362
363 if (context->ioprio_set)
364 if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
365 r = EXIT_IOPRIO;
366 goto fail;
367 }
368
369 if (context->timer_slack_ns_set)
370 if (prctl(PR_SET_TIMERSLACK, context->timer_slack_ns_set) < 0) {
371 r = EXIT_TIMERSLACK;
372 goto fail;
373 }
374
375 if (close_fds(fds, n_fds) < 0 ||
376 shift_fds(fds, n_fds) < 0 ||
377 flags_fds(fds, n_fds) < 0) {
378 r = EXIT_FDS;
379 goto fail;
380 }
381
382 for (i = 0; i < RLIMIT_NLIMITS; i++) {
383 if (!context->rlimit[i])
384 continue;
385
386 if (setrlimit(i, context->rlimit[i]) < 0) {
387 r = EXIT_LIMITS;
388 goto fail;
389 }
390 }
391
392 if (context->secure_bits) {
393 if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
394 r = EXIT_SECUREBITS;
395 goto fail;
396 }
397 }
398
399 if (n_fds > 0) {
400 char a[64], b[64];
401 char *listen_env[3] = {
402 a,
403 b,
404 NULL
405 };
406
407 snprintf(a, sizeof(a), "LISTEN_PID=%llu", (unsigned long long) getpid());
408 snprintf(b, sizeof(b), "LISTEN_FDS=%u", n_fds);
409
410 a[sizeof(a)-1] = 0;
411 b[sizeof(b)-1] = 0;
412
413 if (context->environment) {
414 if (!(f = strv_merge(listen_env, context->environment))) {
415 r = EXIT_MEMORY;
416 goto fail;
417 }
418 e = f;
419 } else
420 e = listen_env;
421
422 } else
423 e = context->environment;
424
425 execve(command->path, command->argv, e);
426 r = EXIT_EXEC;
427
428 fail:
429 strv_free(f);
430 _exit(r);
431 }
432
433
434 log_debug("executed %s as %llu", command->path, (unsigned long long) pid);
435
436 *ret = pid;
437 return 0;
438 }
439
440 void exec_context_init(ExecContext *c) {
441 assert(c);
442
443 c->umask = 0002;
444 c->oom_adjust = 0;
445 c->oom_adjust_set = false;
446 c->nice = 0;
447 c->nice_set = false;
448 c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
449 c->ioprio_set = false;
450 c->cpu_sched_policy = SCHED_OTHER;
451 c->cpu_sched_priority = 0;
452 c->cpu_sched_set = false;
453 CPU_ZERO(&c->cpu_affinity);
454 c->cpu_affinity_set = false;
455
456 c->input = 0;
457 c->output = 0;
458 c->syslog_priority = LOG_DAEMON|LOG_INFO;
459
460 c->secure_bits = 0;
461 c->capability_bounding_set_drop = 0;
462 }
463
464 void exec_context_done(ExecContext *c) {
465 unsigned l;
466
467 assert(c);
468
469 strv_free(c->environment);
470 c->environment = NULL;
471
472 for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
473 free(c->rlimit[l]);
474 c->rlimit[l] = NULL;
475 }
476
477 free(c->working_directory);
478 c->working_directory = NULL;
479 free(c->root_directory);
480 c->root_directory = NULL;
481
482 free(c->syslog_identifier);
483 c->syslog_identifier = NULL;
484
485 free(c->user);
486 c->user = NULL;
487
488 free(c->group);
489 c->group = NULL;
490
491 strv_free(c->supplementary_groups);
492 c->supplementary_groups = NULL;
493
494 if (c->capabilities) {
495 cap_free(c->capabilities);
496 c->capabilities = NULL;
497 }
498 }
499
500 void exec_command_free_list(ExecCommand *c) {
501 ExecCommand *i;
502
503 while ((i = c)) {
504 LIST_REMOVE(ExecCommand, command, c, i);
505
506 free(i->path);
507 strv_free(i->argv);
508 free(i);
509 }
510 }
511
512 void exec_command_free_array(ExecCommand **c, unsigned n) {
513 unsigned i;
514
515 for (i = 0; i < n; i++) {
516 exec_command_free_list(c[i]);
517 c[i] = NULL;
518 }
519 }
520
521 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
522 char ** e;
523 unsigned i;
524
525 assert(c);
526 assert(f);
527
528 if (!prefix)
529 prefix = "";
530
531 fprintf(f,
532 "%sUMask: %04o\n"
533 "%sWorkingDirectory: %s\n"
534 "%sRootDirectory: %s\n",
535 prefix, c->umask,
536 prefix, c->working_directory ? c->working_directory : "/",
537 prefix, c->root_directory ? c->root_directory : "/");
538
539 if (c->environment)
540 for (e = c->environment; *e; e++)
541 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
542
543 if (c->nice_set)
544 fprintf(f,
545 "%sNice: %i\n",
546 prefix, c->nice);
547
548 if (c->oom_adjust_set)
549 fprintf(f,
550 "%sOOMAdjust: %i\n",
551 prefix, c->oom_adjust);
552
553 for (i = 0; i < RLIM_NLIMITS; i++)
554 if (c->rlimit[i])
555 fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
556
557 if (c->ioprio_set)
558 fprintf(f,
559 "%sIOSchedulingClass: %s\n"
560 "%sIOPriority: %i\n",
561 prefix, ioprio_class_to_string(IOPRIO_PRIO_CLASS(c->ioprio)),
562 prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
563
564 if (c->cpu_sched_set)
565 fprintf(f,
566 "%sCPUSchedulingPolicy: %s\n"
567 "%sCPUSchedulingPriority: %i\n"
568 "%sCPUSchedulingResetOnFork: %s\n",
569 prefix, sched_policy_to_string(c->cpu_sched_policy),
570 prefix, c->cpu_sched_priority,
571 prefix, yes_no(c->cpu_sched_reset_on_fork));
572
573 if (c->cpu_affinity_set) {
574 fprintf(f, "%sCPUAffinity:", prefix);
575 for (i = 0; i < CPU_SETSIZE; i++)
576 if (CPU_ISSET(i, &c->cpu_affinity))
577 fprintf(f, " %i", i);
578 fputs("\n", f);
579 }
580
581 if (c->timer_slack_ns_set)
582 fprintf(f, "%sTimerSlackNS: %lu\n", prefix, c->timer_slack_ns);
583
584 fprintf(f,
585 "%sInput: %s\n"
586 "%sOutput: %s\n",
587 prefix, exec_input_to_string(c->input),
588 prefix, exec_output_to_string(c->output));
589
590 if (c->output == EXEC_OUTPUT_SYSLOG || c->output == EXEC_OUTPUT_KERNEL)
591 fprintf(f,
592 "%sSyslogFacility: %s\n"
593 "%sSyslogLevel: %s\n",
594 prefix, log_facility_to_string(LOG_FAC(c->syslog_priority)),
595 prefix, log_level_to_string(LOG_PRI(c->syslog_priority)));
596
597 if (c->capabilities) {
598 char *t;
599 if ((t = cap_to_text(c->capabilities, NULL))) {
600 fprintf(f, "%sCapabilities: %s\n",
601 prefix, t);
602 cap_free(t);
603 }
604 }
605
606 if (c->secure_bits)
607 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
608 prefix,
609 (c->secure_bits & SECURE_KEEP_CAPS) ? " keep-caps" : "",
610 (c->secure_bits & SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
611 (c->secure_bits & SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
612 (c->secure_bits & SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
613 (c->secure_bits & SECURE_NOROOT) ? " noroot" : "",
614 (c->secure_bits & SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
615
616 if (c->capability_bounding_set_drop) {
617 fprintf(f, "%sCapabilityBoundingSetDrop:", prefix);
618
619 for (i = 0; i <= CAP_LAST_CAP; i++)
620 if (c->capability_bounding_set_drop & (1 << i)) {
621 char *t;
622
623 if ((t = cap_to_name(i))) {
624 fprintf(f, " %s", t);
625 free(t);
626 }
627 }
628
629 fputs("\n", f);
630 }
631
632 if (c->user)
633 fprintf(f, "%sUser: %s", prefix, c->user);
634 if (c->group)
635 fprintf(f, "%sGroup: %s", prefix, c->group);
636
637 if (c->supplementary_groups) {
638 char **g;
639
640 fprintf(f, "%sSupplementaryGroups:", prefix);
641
642 STRV_FOREACH(g, c->supplementary_groups)
643 fprintf(f, " %s", *g);
644
645 fputs("\n", f);
646 }
647 }
648
649 void exec_status_fill(ExecStatus *s, pid_t pid, int code, int status) {
650 assert(s);
651
652 s->pid = pid;
653 s->code = code;
654 s->status = status;
655 s->timestamp = now(CLOCK_REALTIME);
656 }
657
658 char *exec_command_line(ExecCommand *c) {
659 size_t k;
660 char *n, *p, **a;
661 bool first = true;
662
663 assert(c);
664 assert(c->argv);
665
666 k = 1;
667 STRV_FOREACH(a, c->argv)
668 k += strlen(*a)+3;
669
670 if (!(n = new(char, k)))
671 return NULL;
672
673 p = n;
674 STRV_FOREACH(a, c->argv) {
675
676 if (!first)
677 *(p++) = ' ';
678 else
679 first = false;
680
681 if (strpbrk(*a, WHITESPACE)) {
682 *(p++) = '\'';
683 p = stpcpy(p, *a);
684 *(p++) = '\'';
685 } else
686 p = stpcpy(p, *a);
687
688 }
689
690 *p = 0;
691
692 /* FIXME: this doesn't really handle arguments that have
693 * spaces and ticks in them */
694
695 return n;
696 }
697
698 void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
699 char *cmd;
700
701 assert(c);
702 assert(f);
703
704 if (!prefix)
705 prefix = "";
706
707 cmd = exec_command_line(c);
708
709 fprintf(f,
710 "%sCommand Line: %s\n",
711 prefix, cmd ? cmd : strerror(ENOMEM));
712
713 free(cmd);
714 }
715
716 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
717 assert(f);
718
719 if (!prefix)
720 prefix = "";
721
722 LIST_FOREACH(command, c, c)
723 exec_command_dump(c, f, prefix);
724 }
725
726 static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
727 [EXEC_OUTPUT_CONSOLE] = "console",
728 [EXEC_OUTPUT_NULL] = "null",
729 [EXEC_OUTPUT_SYSLOG] = "syslog",
730 [EXEC_OUTPUT_KERNEL] = "kernel"
731 };
732
733 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
734
735 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
736 [EXEC_INPUT_NULL] = "null",
737 [EXEC_INPUT_CONSOLE] = "console"
738 };
739
740 DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);