]>
Commit | Line | Data |
---|---|---|
53e1b683 | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
0b452006 RC |
2 | /*** |
3 | This file is part of systemd. | |
4 | ||
5 | Copyright 2010 Lennart Poettering | |
6 | ||
7 | systemd is free software; you can redistribute it and/or modify it | |
8 | under the terms of the GNU Lesser General Public License as published by | |
9 | the Free Software Foundation; either version 2.1 of the License, or | |
10 | (at your option) any later version. | |
11 | ||
12 | systemd is distributed in the hope that it will be useful, but | |
13 | WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
15 | Lesser General Public License for more details. | |
16 | ||
17 | You should have received a copy of the GNU Lesser General Public License | |
18 | along with systemd; If not, see <http://www.gnu.org/licenses/>. | |
19 | ***/ | |
20 | ||
4f5dd394 | 21 | #include <ctype.h> |
0b452006 | 22 | #include <errno.h> |
11c3a366 TA |
23 | #include <limits.h> |
24 | #include <linux/oom.h> | |
7b3e062c | 25 | #include <sched.h> |
0b452006 | 26 | #include <signal.h> |
4f5dd394 LP |
27 | #include <stdbool.h> |
28 | #include <stdio.h> | |
35bbbf85 | 29 | #include <stdio_ext.h> |
11c3a366 | 30 | #include <stdlib.h> |
4f5dd394 | 31 | #include <string.h> |
9bfaffd5 | 32 | #include <sys/mman.h> |
7b3e062c | 33 | #include <sys/personality.h> |
405f8907 | 34 | #include <sys/prctl.h> |
4f5dd394 LP |
35 | #include <sys/types.h> |
36 | #include <sys/wait.h> | |
11c3a366 | 37 | #include <syslog.h> |
4f5dd394 | 38 | #include <unistd.h> |
349cc4a5 | 39 | #if HAVE_VALGRIND_VALGRIND_H |
dcadc967 EV |
40 | #include <valgrind/valgrind.h> |
41 | #endif | |
0b452006 | 42 | |
b5efdb8a | 43 | #include "alloc-util.h" |
6e5f1b57 | 44 | #include "architecture.h" |
4f5dd394 | 45 | #include "escape.h" |
3ffd4af2 | 46 | #include "fd-util.h" |
0b452006 | 47 | #include "fileio.h" |
f4f15635 | 48 | #include "fs-util.h" |
7b3e062c | 49 | #include "ioprio.h" |
0b452006 | 50 | #include "log.h" |
11c3a366 TA |
51 | #include "macro.h" |
52 | #include "missing.h" | |
93cc7779 | 53 | #include "process-util.h" |
8869a0b4 | 54 | #include "raw-clone.h" |
93cc7779 | 55 | #include "signal-util.h" |
1359fffa | 56 | #include "stat-util.h" |
7b3e062c | 57 | #include "string-table.h" |
07630cea | 58 | #include "string-util.h" |
4c253ed1 | 59 | #include "terminal-util.h" |
b1d4f8e1 | 60 | #include "user-util.h" |
4f5dd394 | 61 | #include "util.h" |
0b452006 RC |
62 | |
63 | int get_process_state(pid_t pid) { | |
64 | const char *p; | |
65 | char state; | |
66 | int r; | |
67 | _cleanup_free_ char *line = NULL; | |
68 | ||
69 | assert(pid >= 0); | |
70 | ||
71 | p = procfs_file_alloca(pid, "stat"); | |
a644184a | 72 | |
0b452006 | 73 | r = read_one_line_file(p, &line); |
a644184a LP |
74 | if (r == -ENOENT) |
75 | return -ESRCH; | |
0b452006 RC |
76 | if (r < 0) |
77 | return r; | |
78 | ||
79 | p = strrchr(line, ')'); | |
80 | if (!p) | |
81 | return -EIO; | |
82 | ||
83 | p++; | |
84 | ||
85 | if (sscanf(p, " %c", &state) != 1) | |
86 | return -EIO; | |
87 | ||
88 | return (unsigned char) state; | |
89 | } | |
90 | ||
91 | int get_process_comm(pid_t pid, char **name) { | |
92 | const char *p; | |
93 | int r; | |
94 | ||
95 | assert(name); | |
96 | assert(pid >= 0); | |
97 | ||
98 | p = procfs_file_alloca(pid, "comm"); | |
99 | ||
100 | r = read_one_line_file(p, name); | |
101 | if (r == -ENOENT) | |
102 | return -ESRCH; | |
103 | ||
104 | return r; | |
105 | } | |
106 | ||
107 | int get_process_cmdline(pid_t pid, size_t max_length, bool comm_fallback, char **line) { | |
108 | _cleanup_fclose_ FILE *f = NULL; | |
ba4cd7e2 | 109 | bool space = false; |
c0534780 | 110 | char *k, *ans = NULL; |
0b452006 RC |
111 | const char *p; |
112 | int c; | |
113 | ||
114 | assert(line); | |
115 | assert(pid >= 0); | |
116 | ||
69281c49 LP |
117 | /* Retrieves a process' command line. Replaces unprintable characters while doing so by whitespace (coalescing |
118 | * multiple sequential ones into one). If max_length is != 0 will return a string of the specified size at most | |
119 | * (the trailing NUL byte does count towards the length here!), abbreviated with a "..." ellipsis. If | |
120 | * comm_fallback is true and the process has no command line set (the case for kernel threads), or has a | |
121 | * command line that resolves to the empty string will return the "comm" name of the process instead. | |
122 | * | |
123 | * Returns -ESRCH if the process doesn't exist, and -ENOENT if the process has no command line (and | |
c0534780 | 124 | * comm_fallback is false). Returns 0 and sets *line otherwise. */ |
69281c49 | 125 | |
0b452006 RC |
126 | p = procfs_file_alloca(pid, "cmdline"); |
127 | ||
128 | f = fopen(p, "re"); | |
a644184a LP |
129 | if (!f) { |
130 | if (errno == ENOENT) | |
131 | return -ESRCH; | |
0b452006 | 132 | return -errno; |
a644184a | 133 | } |
0b452006 | 134 | |
35bbbf85 LP |
135 | (void) __fsetlocking(f, FSETLOCKING_BYCALLER); |
136 | ||
69281c49 LP |
137 | if (max_length == 1) { |
138 | ||
139 | /* If there's only room for one byte, return the empty string */ | |
c0534780 ZJS |
140 | ans = new0(char, 1); |
141 | if (!ans) | |
69281c49 LP |
142 | return -ENOMEM; |
143 | ||
c0534780 | 144 | *line = ans; |
69281c49 LP |
145 | return 0; |
146 | ||
147 | } else if (max_length == 0) { | |
0b452006 RC |
148 | size_t len = 0, allocated = 0; |
149 | ||
150 | while ((c = getc(f)) != EOF) { | |
151 | ||
c0534780 ZJS |
152 | if (!GREEDY_REALLOC(ans, allocated, len+3)) { |
153 | free(ans); | |
0b452006 RC |
154 | return -ENOMEM; |
155 | } | |
156 | ||
ba4cd7e2 MP |
157 | if (isprint(c)) { |
158 | if (space) { | |
c0534780 | 159 | ans[len++] = ' '; |
ba4cd7e2 MP |
160 | space = false; |
161 | } | |
162 | ||
c0534780 | 163 | ans[len++] = c; |
69281c49 | 164 | } else if (len > 0) |
ba4cd7e2 MP |
165 | space = true; |
166 | } | |
0b452006 RC |
167 | |
168 | if (len > 0) | |
c0534780 | 169 | ans[len] = '\0'; |
b09df4e2 | 170 | else |
c0534780 | 171 | ans = mfree(ans); |
0b452006 RC |
172 | |
173 | } else { | |
69281c49 | 174 | bool dotdotdot = false; |
0b452006 RC |
175 | size_t left; |
176 | ||
c0534780 ZJS |
177 | ans = new(char, max_length); |
178 | if (!ans) | |
0b452006 RC |
179 | return -ENOMEM; |
180 | ||
c0534780 | 181 | k = ans; |
0b452006 RC |
182 | left = max_length; |
183 | while ((c = getc(f)) != EOF) { | |
184 | ||
185 | if (isprint(c)) { | |
69281c49 | 186 | |
0b452006 | 187 | if (space) { |
69281c49 LP |
188 | if (left <= 2) { |
189 | dotdotdot = true; | |
0b452006 | 190 | break; |
69281c49 | 191 | } |
0b452006 RC |
192 | |
193 | *(k++) = ' '; | |
194 | left--; | |
195 | space = false; | |
196 | } | |
197 | ||
69281c49 LP |
198 | if (left <= 1) { |
199 | dotdotdot = true; | |
0b452006 | 200 | break; |
69281c49 | 201 | } |
0b452006 RC |
202 | |
203 | *(k++) = (char) c; | |
204 | left--; | |
c0534780 | 205 | } else if (k > ans) |
0b452006 RC |
206 | space = true; |
207 | } | |
208 | ||
69281c49 LP |
209 | if (dotdotdot) { |
210 | if (max_length <= 4) { | |
c0534780 | 211 | k = ans; |
69281c49 LP |
212 | left = max_length; |
213 | } else { | |
c0534780 | 214 | k = ans + max_length - 4; |
69281c49 LP |
215 | left = 4; |
216 | ||
217 | /* Eat up final spaces */ | |
c0534780 | 218 | while (k > ans && isspace(k[-1])) { |
69281c49 LP |
219 | k--; |
220 | left++; | |
221 | } | |
222 | } | |
223 | ||
224 | strncpy(k, "...", left-1); | |
b09df4e2 | 225 | k[left-1] = 0; |
0b452006 RC |
226 | } else |
227 | *k = 0; | |
228 | } | |
229 | ||
230 | /* Kernel threads have no argv[] */ | |
c0534780 | 231 | if (isempty(ans)) { |
0b452006 RC |
232 | _cleanup_free_ char *t = NULL; |
233 | int h; | |
234 | ||
c0534780 | 235 | free(ans); |
0b452006 RC |
236 | |
237 | if (!comm_fallback) | |
238 | return -ENOENT; | |
239 | ||
240 | h = get_process_comm(pid, &t); | |
241 | if (h < 0) | |
242 | return h; | |
243 | ||
69281c49 | 244 | if (max_length == 0) |
c0534780 | 245 | ans = strjoin("[", t, "]"); |
69281c49 LP |
246 | else { |
247 | size_t l; | |
248 | ||
249 | l = strlen(t); | |
250 | ||
251 | if (l + 3 <= max_length) | |
c0534780 | 252 | ans = strjoin("[", t, "]"); |
69281c49 LP |
253 | else if (max_length <= 6) { |
254 | ||
c0534780 ZJS |
255 | ans = new(char, max_length); |
256 | if (!ans) | |
69281c49 LP |
257 | return -ENOMEM; |
258 | ||
c0534780 ZJS |
259 | memcpy(ans, "[...]", max_length-1); |
260 | ans[max_length-1] = 0; | |
69281c49 LP |
261 | } else { |
262 | char *e; | |
263 | ||
264 | t[max_length - 6] = 0; | |
265 | ||
266 | /* Chop off final spaces */ | |
267 | e = strchr(t, 0); | |
268 | while (e > t && isspace(e[-1])) | |
269 | e--; | |
270 | *e = 0; | |
271 | ||
c0534780 | 272 | ans = strjoin("[", t, "...]"); |
69281c49 LP |
273 | } |
274 | } | |
c0534780 | 275 | if (!ans) |
0b452006 RC |
276 | return -ENOMEM; |
277 | } | |
278 | ||
c0534780 | 279 | *line = ans; |
0b452006 RC |
280 | return 0; |
281 | } | |
282 | ||
9bfaffd5 LP |
283 | int rename_process(const char name[]) { |
284 | static size_t mm_size = 0; | |
285 | static char *mm = NULL; | |
286 | bool truncated = false; | |
287 | size_t l; | |
288 | ||
289 | /* This is a like a poor man's setproctitle(). It changes the comm field, argv[0], and also the glibc's | |
290 | * internally used name of the process. For the first one a limit of 16 chars applies; to the second one in | |
291 | * many cases one of 10 (i.e. length of "/sbin/init") — however if we have CAP_SYS_RESOURCES it is unbounded; | |
292 | * to the third one 7 (i.e. the length of "systemd". If you pass a longer string it will likely be | |
293 | * truncated. | |
294 | * | |
295 | * Returns 0 if a name was set but truncated, > 0 if it was set but not truncated. */ | |
296 | ||
297 | if (isempty(name)) | |
298 | return -EINVAL; /* let's not confuse users unnecessarily with an empty name */ | |
405f8907 | 299 | |
1096bb8a LP |
300 | if (!is_main_thread()) |
301 | return -EPERM; /* Let's not allow setting the process name from other threads than the main one, as we | |
302 | * cache things without locking, and we make assumptions that PR_SET_NAME sets the | |
303 | * process name that isn't correct on any other threads */ | |
304 | ||
9bfaffd5 | 305 | l = strlen(name); |
405f8907 | 306 | |
5a8af747 LP |
307 | /* First step, change the comm field. The main thread's comm is identical to the process comm. This means we |
308 | * can use PR_SET_NAME, which sets the thread name for the calling thread. */ | |
309 | if (prctl(PR_SET_NAME, name) < 0) | |
310 | log_debug_errno(errno, "PR_SET_NAME failed: %m"); | |
9bfaffd5 LP |
311 | if (l > 15) /* Linux process names can be 15 chars at max */ |
312 | truncated = true; | |
313 | ||
314 | /* Second step, change glibc's ID of the process name. */ | |
315 | if (program_invocation_name) { | |
316 | size_t k; | |
317 | ||
318 | k = strlen(program_invocation_name); | |
319 | strncpy(program_invocation_name, name, k); | |
320 | if (l > k) | |
321 | truncated = true; | |
322 | } | |
323 | ||
324 | /* Third step, completely replace the argv[] array the kernel maintains for us. This requires privileges, but | |
325 | * has the advantage that the argv[] array is exactly what we want it to be, and not filled up with zeros at | |
13e785f7 | 326 | * the end. This is the best option for changing /proc/self/cmdline. */ |
01f989c6 JW |
327 | |
328 | /* Let's not bother with this if we don't have euid == 0. Strictly speaking we should check for the | |
329 | * CAP_SYS_RESOURCE capability which is independent of the euid. In our own code the capability generally is | |
330 | * present only for euid == 0, hence let's use this as quick bypass check, to avoid calling mmap() if | |
331 | * PR_SET_MM_ARG_{START,END} fails with EPERM later on anyway. After all geteuid() is dead cheap to call, but | |
332 | * mmap() is not. */ | |
333 | if (geteuid() != 0) | |
334 | log_debug("Skipping PR_SET_MM, as we don't have privileges."); | |
335 | else if (mm_size < l+1) { | |
9bfaffd5 LP |
336 | size_t nn_size; |
337 | char *nn; | |
338 | ||
9bfaffd5 LP |
339 | nn_size = PAGE_ALIGN(l+1); |
340 | nn = mmap(NULL, nn_size, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); | |
341 | if (nn == MAP_FAILED) { | |
342 | log_debug_errno(errno, "mmap() failed: %m"); | |
343 | goto use_saved_argv; | |
344 | } | |
405f8907 | 345 | |
9bfaffd5 LP |
346 | strncpy(nn, name, nn_size); |
347 | ||
348 | /* Now, let's tell the kernel about this new memory */ | |
349 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_START, (unsigned long) nn, 0, 0) < 0) { | |
350 | log_debug_errno(errno, "PR_SET_MM_ARG_START failed, proceeding without: %m"); | |
351 | (void) munmap(nn, nn_size); | |
352 | goto use_saved_argv; | |
353 | } | |
354 | ||
355 | /* And update the end pointer to the new end, too. If this fails, we don't really know what to do, it's | |
356 | * pretty unlikely that we can rollback, hence we'll just accept the failure, and continue. */ | |
357 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) nn + l + 1, 0, 0) < 0) | |
358 | log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m"); | |
359 | ||
360 | if (mm) | |
361 | (void) munmap(mm, mm_size); | |
362 | ||
363 | mm = nn; | |
364 | mm_size = nn_size; | |
01f989c6 | 365 | } else { |
9bfaffd5 LP |
366 | strncpy(mm, name, mm_size); |
367 | ||
01f989c6 JW |
368 | /* Update the end pointer, continuing regardless of any failure. */ |
369 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) mm + l + 1, 0, 0) < 0) | |
370 | log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m"); | |
371 | } | |
372 | ||
9bfaffd5 LP |
373 | use_saved_argv: |
374 | /* Fourth step: in all cases we'll also update the original argv[], so that our own code gets it right too if | |
375 | * it still looks here */ | |
405f8907 LP |
376 | |
377 | if (saved_argc > 0) { | |
378 | int i; | |
379 | ||
9bfaffd5 LP |
380 | if (saved_argv[0]) { |
381 | size_t k; | |
382 | ||
383 | k = strlen(saved_argv[0]); | |
384 | strncpy(saved_argv[0], name, k); | |
385 | if (l > k) | |
386 | truncated = true; | |
387 | } | |
405f8907 LP |
388 | |
389 | for (i = 1; i < saved_argc; i++) { | |
390 | if (!saved_argv[i]) | |
391 | break; | |
392 | ||
393 | memzero(saved_argv[i], strlen(saved_argv[i])); | |
394 | } | |
395 | } | |
9bfaffd5 LP |
396 | |
397 | return !truncated; | |
405f8907 LP |
398 | } |
399 | ||
0b452006 RC |
400 | int is_kernel_thread(pid_t pid) { |
401 | const char *p; | |
402 | size_t count; | |
403 | char c; | |
404 | bool eof; | |
405 | FILE *f; | |
406 | ||
4c701096 | 407 | if (IN_SET(pid, 0, 1) || pid == getpid_cached()) /* pid 1, and we ourselves certainly aren't a kernel thread */ |
0b452006 RC |
408 | return 0; |
409 | ||
a6149b93 | 410 | assert(pid > 1); |
0b452006 RC |
411 | |
412 | p = procfs_file_alloca(pid, "cmdline"); | |
413 | f = fopen(p, "re"); | |
a644184a LP |
414 | if (!f) { |
415 | if (errno == ENOENT) | |
416 | return -ESRCH; | |
0b452006 | 417 | return -errno; |
a644184a | 418 | } |
0b452006 | 419 | |
35bbbf85 LP |
420 | (void) __fsetlocking(f, FSETLOCKING_BYCALLER); |
421 | ||
0b452006 RC |
422 | count = fread(&c, 1, 1, f); |
423 | eof = feof(f); | |
424 | fclose(f); | |
425 | ||
426 | /* Kernel threads have an empty cmdline */ | |
427 | ||
428 | if (count <= 0) | |
429 | return eof ? 1 : -errno; | |
430 | ||
431 | return 0; | |
432 | } | |
433 | ||
434 | int get_process_capeff(pid_t pid, char **capeff) { | |
435 | const char *p; | |
a644184a | 436 | int r; |
0b452006 RC |
437 | |
438 | assert(capeff); | |
439 | assert(pid >= 0); | |
440 | ||
441 | p = procfs_file_alloca(pid, "status"); | |
442 | ||
c4cd1d4d | 443 | r = get_proc_field(p, "CapEff", WHITESPACE, capeff); |
a644184a LP |
444 | if (r == -ENOENT) |
445 | return -ESRCH; | |
446 | ||
447 | return r; | |
0b452006 RC |
448 | } |
449 | ||
450 | static int get_process_link_contents(const char *proc_file, char **name) { | |
451 | int r; | |
452 | ||
453 | assert(proc_file); | |
454 | assert(name); | |
455 | ||
456 | r = readlink_malloc(proc_file, name); | |
a644184a LP |
457 | if (r == -ENOENT) |
458 | return -ESRCH; | |
0b452006 | 459 | if (r < 0) |
a644184a | 460 | return r; |
0b452006 RC |
461 | |
462 | return 0; | |
463 | } | |
464 | ||
465 | int get_process_exe(pid_t pid, char **name) { | |
466 | const char *p; | |
467 | char *d; | |
468 | int r; | |
469 | ||
470 | assert(pid >= 0); | |
471 | ||
472 | p = procfs_file_alloca(pid, "exe"); | |
473 | r = get_process_link_contents(p, name); | |
474 | if (r < 0) | |
475 | return r; | |
476 | ||
477 | d = endswith(*name, " (deleted)"); | |
478 | if (d) | |
479 | *d = '\0'; | |
480 | ||
481 | return 0; | |
482 | } | |
483 | ||
484 | static int get_process_id(pid_t pid, const char *field, uid_t *uid) { | |
485 | _cleanup_fclose_ FILE *f = NULL; | |
486 | char line[LINE_MAX]; | |
487 | const char *p; | |
488 | ||
489 | assert(field); | |
490 | assert(uid); | |
491 | ||
07b38ba5 | 492 | if (pid < 0) |
6f8cbcdb LP |
493 | return -EINVAL; |
494 | ||
0b452006 RC |
495 | p = procfs_file_alloca(pid, "status"); |
496 | f = fopen(p, "re"); | |
a644184a LP |
497 | if (!f) { |
498 | if (errno == ENOENT) | |
499 | return -ESRCH; | |
0b452006 | 500 | return -errno; |
a644184a | 501 | } |
0b452006 | 502 | |
35bbbf85 LP |
503 | (void) __fsetlocking(f, FSETLOCKING_BYCALLER); |
504 | ||
0b452006 RC |
505 | FOREACH_LINE(line, f, return -errno) { |
506 | char *l; | |
507 | ||
508 | l = strstrip(line); | |
509 | ||
510 | if (startswith(l, field)) { | |
511 | l += strlen(field); | |
512 | l += strspn(l, WHITESPACE); | |
513 | ||
514 | l[strcspn(l, WHITESPACE)] = 0; | |
515 | ||
516 | return parse_uid(l, uid); | |
517 | } | |
518 | } | |
519 | ||
520 | return -EIO; | |
521 | } | |
522 | ||
523 | int get_process_uid(pid_t pid, uid_t *uid) { | |
6f8cbcdb LP |
524 | |
525 | if (pid == 0 || pid == getpid_cached()) { | |
526 | *uid = getuid(); | |
527 | return 0; | |
528 | } | |
529 | ||
0b452006 RC |
530 | return get_process_id(pid, "Uid:", uid); |
531 | } | |
532 | ||
533 | int get_process_gid(pid_t pid, gid_t *gid) { | |
6f8cbcdb LP |
534 | |
535 | if (pid == 0 || pid == getpid_cached()) { | |
536 | *gid = getgid(); | |
537 | return 0; | |
538 | } | |
539 | ||
0b452006 RC |
540 | assert_cc(sizeof(uid_t) == sizeof(gid_t)); |
541 | return get_process_id(pid, "Gid:", gid); | |
542 | } | |
543 | ||
544 | int get_process_cwd(pid_t pid, char **cwd) { | |
545 | const char *p; | |
546 | ||
547 | assert(pid >= 0); | |
548 | ||
549 | p = procfs_file_alloca(pid, "cwd"); | |
550 | ||
551 | return get_process_link_contents(p, cwd); | |
552 | } | |
553 | ||
554 | int get_process_root(pid_t pid, char **root) { | |
555 | const char *p; | |
556 | ||
557 | assert(pid >= 0); | |
558 | ||
559 | p = procfs_file_alloca(pid, "root"); | |
560 | ||
561 | return get_process_link_contents(p, root); | |
562 | } | |
563 | ||
564 | int get_process_environ(pid_t pid, char **env) { | |
565 | _cleanup_fclose_ FILE *f = NULL; | |
566 | _cleanup_free_ char *outcome = NULL; | |
567 | int c; | |
568 | const char *p; | |
569 | size_t allocated = 0, sz = 0; | |
570 | ||
571 | assert(pid >= 0); | |
572 | assert(env); | |
573 | ||
574 | p = procfs_file_alloca(pid, "environ"); | |
575 | ||
576 | f = fopen(p, "re"); | |
a644184a LP |
577 | if (!f) { |
578 | if (errno == ENOENT) | |
579 | return -ESRCH; | |
0b452006 | 580 | return -errno; |
a644184a | 581 | } |
0b452006 | 582 | |
35bbbf85 LP |
583 | (void) __fsetlocking(f, FSETLOCKING_BYCALLER); |
584 | ||
0b452006 RC |
585 | while ((c = fgetc(f)) != EOF) { |
586 | if (!GREEDY_REALLOC(outcome, allocated, sz + 5)) | |
587 | return -ENOMEM; | |
588 | ||
589 | if (c == '\0') | |
590 | outcome[sz++] = '\n'; | |
591 | else | |
592 | sz += cescape_char(c, outcome + sz); | |
593 | } | |
594 | ||
03c55bc0 LP |
595 | if (!outcome) { |
596 | outcome = strdup(""); | |
597 | if (!outcome) | |
598 | return -ENOMEM; | |
599 | } else | |
600 | outcome[sz] = '\0'; | |
de8763b6 | 601 | |
0b452006 RC |
602 | *env = outcome; |
603 | outcome = NULL; | |
604 | ||
605 | return 0; | |
606 | } | |
607 | ||
6bc73acb | 608 | int get_process_ppid(pid_t pid, pid_t *_ppid) { |
0b452006 RC |
609 | int r; |
610 | _cleanup_free_ char *line = NULL; | |
611 | long unsigned ppid; | |
612 | const char *p; | |
613 | ||
614 | assert(pid >= 0); | |
615 | assert(_ppid); | |
616 | ||
6f8cbcdb | 617 | if (pid == 0 || pid == getpid_cached()) { |
0b452006 RC |
618 | *_ppid = getppid(); |
619 | return 0; | |
620 | } | |
621 | ||
622 | p = procfs_file_alloca(pid, "stat"); | |
623 | r = read_one_line_file(p, &line); | |
a644184a LP |
624 | if (r == -ENOENT) |
625 | return -ESRCH; | |
0b452006 RC |
626 | if (r < 0) |
627 | return r; | |
628 | ||
629 | /* Let's skip the pid and comm fields. The latter is enclosed | |
630 | * in () but does not escape any () in its value, so let's | |
631 | * skip over it manually */ | |
632 | ||
633 | p = strrchr(line, ')'); | |
634 | if (!p) | |
635 | return -EIO; | |
636 | ||
637 | p++; | |
638 | ||
639 | if (sscanf(p, " " | |
640 | "%*c " /* state */ | |
641 | "%lu ", /* ppid */ | |
642 | &ppid) != 1) | |
643 | return -EIO; | |
644 | ||
645 | if ((long unsigned) (pid_t) ppid != ppid) | |
646 | return -ERANGE; | |
647 | ||
648 | *_ppid = (pid_t) ppid; | |
649 | ||
650 | return 0; | |
651 | } | |
652 | ||
653 | int wait_for_terminate(pid_t pid, siginfo_t *status) { | |
654 | siginfo_t dummy; | |
655 | ||
656 | assert(pid >= 1); | |
657 | ||
658 | if (!status) | |
659 | status = &dummy; | |
660 | ||
661 | for (;;) { | |
662 | zero(*status); | |
663 | ||
664 | if (waitid(P_PID, pid, status, WEXITED) < 0) { | |
665 | ||
666 | if (errno == EINTR) | |
667 | continue; | |
668 | ||
3f0083a2 | 669 | return negative_errno(); |
0b452006 RC |
670 | } |
671 | ||
672 | return 0; | |
673 | } | |
674 | } | |
675 | ||
676 | /* | |
677 | * Return values: | |
678 | * < 0 : wait_for_terminate() failed to get the state of the | |
679 | * process, the process was terminated by a signal, or | |
680 | * failed for an unknown reason. | |
681 | * >=0 : The process terminated normally, and its exit code is | |
682 | * returned. | |
683 | * | |
684 | * That is, success is indicated by a return value of zero, and an | |
685 | * error is indicated by a non-zero value. | |
686 | * | |
687 | * A warning is emitted if the process terminates abnormally, | |
688 | * and also if it returns non-zero unless check_exit_code is true. | |
689 | */ | |
7d4904fe LP |
690 | int wait_for_terminate_and_check(const char *name, pid_t pid, WaitFlags flags) { |
691 | _cleanup_free_ char *buffer = NULL; | |
0b452006 | 692 | siginfo_t status; |
7d4904fe | 693 | int r, prio; |
0b452006 | 694 | |
0b452006 RC |
695 | assert(pid > 1); |
696 | ||
7d4904fe LP |
697 | if (!name) { |
698 | r = get_process_comm(pid, &buffer); | |
699 | if (r < 0) | |
700 | log_debug_errno(r, "Failed to acquire process name of " PID_FMT ", ignoring: %m", pid); | |
701 | else | |
702 | name = buffer; | |
703 | } | |
704 | ||
705 | prio = flags & WAIT_LOG_ABNORMAL ? LOG_ERR : LOG_DEBUG; | |
706 | ||
0b452006 RC |
707 | r = wait_for_terminate(pid, &status); |
708 | if (r < 0) | |
7d4904fe | 709 | return log_full_errno(prio, r, "Failed to wait for %s: %m", strna(name)); |
0b452006 RC |
710 | |
711 | if (status.si_code == CLD_EXITED) { | |
7d4904fe LP |
712 | if (status.si_status != EXIT_SUCCESS) |
713 | log_full(flags & WAIT_LOG_NON_ZERO_EXIT_STATUS ? LOG_ERR : LOG_DEBUG, | |
714 | "%s failed with exit status %i.", strna(name), status.si_status); | |
0b452006 RC |
715 | else |
716 | log_debug("%s succeeded.", name); | |
717 | ||
718 | return status.si_status; | |
7d4904fe | 719 | |
3742095b | 720 | } else if (IN_SET(status.si_code, CLD_KILLED, CLD_DUMPED)) { |
0b452006 | 721 | |
7d4904fe | 722 | log_full(prio, "%s terminated by signal %s.", strna(name), signal_to_string(status.si_status)); |
0b452006 RC |
723 | return -EPROTO; |
724 | } | |
725 | ||
7d4904fe | 726 | log_full(prio, "%s failed due to unknown reason.", strna(name)); |
0b452006 RC |
727 | return -EPROTO; |
728 | } | |
729 | ||
d5641e0d KW |
730 | /* |
731 | * Return values: | |
732 | * < 0 : wait_for_terminate_with_timeout() failed to get the state of the | |
733 | * process, the process timed out, the process was terminated by a | |
734 | * signal, or failed for an unknown reason. | |
735 | * >=0 : The process terminated normally with no failures. | |
736 | * | |
737 | * Success is indicated by a return value of zero, a timeout is indicated | |
738 | * by ETIMEDOUT, and all other child failure states are indicated by error | |
739 | * is indicated by a non-zero value. | |
740 | */ | |
741 | int wait_for_terminate_with_timeout(pid_t pid, usec_t timeout) { | |
742 | sigset_t mask; | |
743 | int r; | |
744 | usec_t until; | |
745 | ||
746 | assert_se(sigemptyset(&mask) == 0); | |
747 | assert_se(sigaddset(&mask, SIGCHLD) == 0); | |
748 | ||
749 | /* Drop into a sigtimewait-based timeout. Waiting for the | |
750 | * pid to exit. */ | |
751 | until = now(CLOCK_MONOTONIC) + timeout; | |
752 | for (;;) { | |
753 | usec_t n; | |
754 | siginfo_t status = {}; | |
755 | struct timespec ts; | |
756 | ||
757 | n = now(CLOCK_MONOTONIC); | |
758 | if (n >= until) | |
759 | break; | |
760 | ||
761 | r = sigtimedwait(&mask, NULL, timespec_store(&ts, until - n)) < 0 ? -errno : 0; | |
762 | /* Assuming we woke due to the child exiting. */ | |
763 | if (waitid(P_PID, pid, &status, WEXITED|WNOHANG) == 0) { | |
764 | if (status.si_pid == pid) { | |
765 | /* This is the correct child.*/ | |
766 | if (status.si_code == CLD_EXITED) | |
767 | return (status.si_status == 0) ? 0 : -EPROTO; | |
768 | else | |
769 | return -EPROTO; | |
770 | } | |
771 | } | |
772 | /* Not the child, check for errors and proceed appropriately */ | |
773 | if (r < 0) { | |
774 | switch (r) { | |
775 | case -EAGAIN: | |
776 | /* Timed out, child is likely hung. */ | |
777 | return -ETIMEDOUT; | |
778 | case -EINTR: | |
779 | /* Received a different signal and should retry */ | |
780 | continue; | |
781 | default: | |
782 | /* Return any unexpected errors */ | |
783 | return r; | |
784 | } | |
785 | } | |
786 | } | |
787 | ||
788 | return -EPROTO; | |
789 | } | |
790 | ||
89c9030d LP |
791 | void sigkill_wait(pid_t pid) { |
792 | assert(pid > 1); | |
793 | ||
794 | if (kill(pid, SIGKILL) > 0) | |
795 | (void) wait_for_terminate(pid, NULL); | |
796 | } | |
797 | ||
798 | void sigkill_waitp(pid_t *pid) { | |
dfd14786 LP |
799 | PROTECT_ERRNO; |
800 | ||
4d0d3d41 LP |
801 | if (!pid) |
802 | return; | |
803 | if (*pid <= 1) | |
804 | return; | |
805 | ||
89c9030d | 806 | sigkill_wait(*pid); |
4d0d3d41 LP |
807 | } |
808 | ||
0b452006 RC |
809 | int kill_and_sigcont(pid_t pid, int sig) { |
810 | int r; | |
811 | ||
812 | r = kill(pid, sig) < 0 ? -errno : 0; | |
813 | ||
26f417d3 LP |
814 | /* If this worked, also send SIGCONT, unless we already just sent a SIGCONT, or SIGKILL was sent which isn't |
815 | * affected by a process being suspended anyway. */ | |
a3d8d68c | 816 | if (r >= 0 && !IN_SET(sig, SIGCONT, SIGKILL)) |
26f417d3 | 817 | (void) kill(pid, SIGCONT); |
0b452006 RC |
818 | |
819 | return r; | |
820 | } | |
821 | ||
822 | int getenv_for_pid(pid_t pid, const char *field, char **_value) { | |
823 | _cleanup_fclose_ FILE *f = NULL; | |
824 | char *value = NULL; | |
825 | int r; | |
826 | bool done = false; | |
827 | size_t l; | |
828 | const char *path; | |
829 | ||
830 | assert(pid >= 0); | |
831 | assert(field); | |
832 | assert(_value); | |
833 | ||
834 | path = procfs_file_alloca(pid, "environ"); | |
835 | ||
836 | f = fopen(path, "re"); | |
a644184a LP |
837 | if (!f) { |
838 | if (errno == ENOENT) | |
839 | return -ESRCH; | |
0b452006 | 840 | return -errno; |
a644184a | 841 | } |
0b452006 | 842 | |
35bbbf85 LP |
843 | (void) __fsetlocking(f, FSETLOCKING_BYCALLER); |
844 | ||
0b452006 RC |
845 | l = strlen(field); |
846 | r = 0; | |
847 | ||
848 | do { | |
849 | char line[LINE_MAX]; | |
850 | unsigned i; | |
851 | ||
852 | for (i = 0; i < sizeof(line)-1; i++) { | |
853 | int c; | |
854 | ||
855 | c = getc(f); | |
856 | if (_unlikely_(c == EOF)) { | |
857 | done = true; | |
858 | break; | |
859 | } else if (c == 0) | |
860 | break; | |
861 | ||
862 | line[i] = c; | |
863 | } | |
864 | line[i] = 0; | |
865 | ||
041b5ae1 | 866 | if (strneq(line, field, l) && line[l] == '=') { |
0b452006 RC |
867 | value = strdup(line + l + 1); |
868 | if (!value) | |
869 | return -ENOMEM; | |
870 | ||
871 | r = 1; | |
872 | break; | |
873 | } | |
874 | ||
875 | } while (!done); | |
876 | ||
877 | *_value = value; | |
878 | return r; | |
879 | } | |
880 | ||
881 | bool pid_is_unwaited(pid_t pid) { | |
882 | /* Checks whether a PID is still valid at all, including a zombie */ | |
883 | ||
07b38ba5 | 884 | if (pid < 0) |
0b452006 RC |
885 | return false; |
886 | ||
5fd9b2c5 LP |
887 | if (pid <= 1) /* If we or PID 1 would be dead and have been waited for, this code would not be running */ |
888 | return true; | |
889 | ||
6f8cbcdb LP |
890 | if (pid == getpid_cached()) |
891 | return true; | |
892 | ||
0b452006 RC |
893 | if (kill(pid, 0) >= 0) |
894 | return true; | |
895 | ||
896 | return errno != ESRCH; | |
897 | } | |
898 | ||
899 | bool pid_is_alive(pid_t pid) { | |
900 | int r; | |
901 | ||
902 | /* Checks whether a PID is still valid and not a zombie */ | |
903 | ||
07b38ba5 | 904 | if (pid < 0) |
0b452006 RC |
905 | return false; |
906 | ||
5fd9b2c5 LP |
907 | if (pid <= 1) /* If we or PID 1 would be a zombie, this code would not be running */ |
908 | return true; | |
909 | ||
6f8cbcdb LP |
910 | if (pid == getpid_cached()) |
911 | return true; | |
912 | ||
0b452006 | 913 | r = get_process_state(pid); |
4c701096 | 914 | if (IN_SET(r, -ESRCH, 'Z')) |
0b452006 RC |
915 | return false; |
916 | ||
917 | return true; | |
918 | } | |
d4510856 | 919 | |
1359fffa MS |
920 | int pid_from_same_root_fs(pid_t pid) { |
921 | const char *root; | |
922 | ||
07b38ba5 | 923 | if (pid < 0) |
6f8cbcdb LP |
924 | return false; |
925 | ||
926 | if (pid == 0 || pid == getpid_cached()) | |
927 | return true; | |
1359fffa MS |
928 | |
929 | root = procfs_file_alloca(pid, "root"); | |
930 | ||
e3f791a2 | 931 | return files_same(root, "/proc/1/root", 0); |
1359fffa MS |
932 | } |
933 | ||
d4510856 LP |
934 | bool is_main_thread(void) { |
935 | static thread_local int cached = 0; | |
936 | ||
937 | if (_unlikely_(cached == 0)) | |
df0ff127 | 938 | cached = getpid_cached() == gettid() ? 1 : -1; |
d4510856 LP |
939 | |
940 | return cached > 0; | |
941 | } | |
7b3e062c LP |
942 | |
943 | noreturn void freeze(void) { | |
944 | ||
3da48d7a EV |
945 | log_close(); |
946 | ||
7b3e062c LP |
947 | /* Make sure nobody waits for us on a socket anymore */ |
948 | close_all_fds(NULL, 0); | |
949 | ||
950 | sync(); | |
951 | ||
952 | for (;;) | |
953 | pause(); | |
954 | } | |
955 | ||
956 | bool oom_score_adjust_is_valid(int oa) { | |
957 | return oa >= OOM_SCORE_ADJ_MIN && oa <= OOM_SCORE_ADJ_MAX; | |
958 | } | |
959 | ||
960 | unsigned long personality_from_string(const char *p) { | |
6e5f1b57 | 961 | int architecture; |
7b3e062c | 962 | |
0c0fea07 LP |
963 | if (!p) |
964 | return PERSONALITY_INVALID; | |
965 | ||
6e5f1b57 LP |
966 | /* Parse a personality specifier. We use our own identifiers that indicate specific ABIs, rather than just |
967 | * hints regarding the register size, since we want to keep things open for multiple locally supported ABIs for | |
968 | * the same register size. */ | |
969 | ||
970 | architecture = architecture_from_string(p); | |
971 | if (architecture < 0) | |
972 | return PERSONALITY_INVALID; | |
7b3e062c | 973 | |
0c0fea07 | 974 | if (architecture == native_architecture()) |
7b3e062c | 975 | return PER_LINUX; |
0c0fea07 LP |
976 | #ifdef SECONDARY_ARCHITECTURE |
977 | if (architecture == SECONDARY_ARCHITECTURE) | |
f2d1736c | 978 | return PER_LINUX32; |
7b3e062c LP |
979 | #endif |
980 | ||
981 | return PERSONALITY_INVALID; | |
982 | } | |
983 | ||
984 | const char* personality_to_string(unsigned long p) { | |
6e5f1b57 | 985 | int architecture = _ARCHITECTURE_INVALID; |
7b3e062c | 986 | |
7b3e062c | 987 | if (p == PER_LINUX) |
0c0fea07 LP |
988 | architecture = native_architecture(); |
989 | #ifdef SECONDARY_ARCHITECTURE | |
6e5f1b57 | 990 | else if (p == PER_LINUX32) |
0c0fea07 | 991 | architecture = SECONDARY_ARCHITECTURE; |
7b3e062c LP |
992 | #endif |
993 | ||
6e5f1b57 LP |
994 | if (architecture < 0) |
995 | return NULL; | |
996 | ||
997 | return architecture_to_string(architecture); | |
7b3e062c LP |
998 | } |
999 | ||
21022b9d LP |
1000 | int safe_personality(unsigned long p) { |
1001 | int ret; | |
1002 | ||
1003 | /* So here's the deal, personality() is weirdly defined by glibc. In some cases it returns a failure via errno, | |
1004 | * and in others as negative return value containing an errno-like value. Let's work around this: this is a | |
1005 | * wrapper that uses errno if it is set, and uses the return value otherwise. And then it sets both errno and | |
1006 | * the return value indicating the same issue, so that we are definitely on the safe side. | |
1007 | * | |
1008 | * See https://github.com/systemd/systemd/issues/6737 */ | |
1009 | ||
1010 | errno = 0; | |
1011 | ret = personality(p); | |
1012 | if (ret < 0) { | |
1013 | if (errno != 0) | |
1014 | return -errno; | |
1015 | ||
1016 | errno = -ret; | |
1017 | } | |
1018 | ||
1019 | return ret; | |
1020 | } | |
1021 | ||
e8132d63 LP |
1022 | int opinionated_personality(unsigned long *ret) { |
1023 | int current; | |
1024 | ||
1025 | /* Returns the current personality, or PERSONALITY_INVALID if we can't determine it. This function is a bit | |
1026 | * opinionated though, and ignores all the finer-grained bits and exotic personalities, only distinguishing the | |
1027 | * two most relevant personalities: PER_LINUX and PER_LINUX32. */ | |
1028 | ||
21022b9d | 1029 | current = safe_personality(PERSONALITY_INVALID); |
e8132d63 | 1030 | if (current < 0) |
21022b9d | 1031 | return current; |
e8132d63 LP |
1032 | |
1033 | if (((unsigned long) current & 0xffff) == PER_LINUX32) | |
1034 | *ret = PER_LINUX32; | |
1035 | else | |
1036 | *ret = PER_LINUX; | |
1037 | ||
1038 | return 0; | |
1039 | } | |
1040 | ||
dcadc967 | 1041 | void valgrind_summary_hack(void) { |
349cc4a5 | 1042 | #if HAVE_VALGRIND_VALGRIND_H |
df0ff127 | 1043 | if (getpid_cached() == 1 && RUNNING_ON_VALGRIND) { |
dcadc967 | 1044 | pid_t pid; |
8869a0b4 | 1045 | pid = raw_clone(SIGCHLD); |
dcadc967 EV |
1046 | if (pid < 0) |
1047 | log_emergency_errno(errno, "Failed to fork off valgrind helper: %m"); | |
1048 | else if (pid == 0) | |
1049 | exit(EXIT_SUCCESS); | |
1050 | else { | |
1051 | log_info("Spawned valgrind helper as PID "PID_FMT".", pid); | |
1052 | (void) wait_for_terminate(pid, NULL); | |
1053 | } | |
1054 | } | |
1055 | #endif | |
1056 | } | |
1057 | ||
291d565a LP |
1058 | int pid_compare_func(const void *a, const void *b) { |
1059 | const pid_t *p = a, *q = b; | |
1060 | ||
1061 | /* Suitable for usage in qsort() */ | |
1062 | ||
1063 | if (*p < *q) | |
1064 | return -1; | |
1065 | if (*p > *q) | |
1066 | return 1; | |
1067 | return 0; | |
1068 | } | |
1069 | ||
7f452159 LP |
1070 | int ioprio_parse_priority(const char *s, int *ret) { |
1071 | int i, r; | |
1072 | ||
1073 | assert(s); | |
1074 | assert(ret); | |
1075 | ||
1076 | r = safe_atoi(s, &i); | |
1077 | if (r < 0) | |
1078 | return r; | |
1079 | ||
1080 | if (!ioprio_priority_is_valid(i)) | |
1081 | return -EINVAL; | |
1082 | ||
1083 | *ret = i; | |
1084 | return 0; | |
1085 | } | |
1086 | ||
5c30a6d2 LP |
1087 | /* The cached PID, possible values: |
1088 | * | |
1089 | * == UNSET [0] → cache not initialized yet | |
1090 | * == BUSY [-1] → some thread is initializing it at the moment | |
1091 | * any other → the cached PID | |
1092 | */ | |
1093 | ||
1094 | #define CACHED_PID_UNSET ((pid_t) 0) | |
1095 | #define CACHED_PID_BUSY ((pid_t) -1) | |
1096 | ||
1097 | static pid_t cached_pid = CACHED_PID_UNSET; | |
1098 | ||
799a960d | 1099 | void reset_cached_pid(void) { |
5c30a6d2 LP |
1100 | /* Invoked in the child after a fork(), i.e. at the first moment the PID changed */ |
1101 | cached_pid = CACHED_PID_UNSET; | |
1102 | } | |
1103 | ||
1104 | /* We use glibc __register_atfork() + __dso_handle directly here, as they are not included in the glibc | |
1105 | * headers. __register_atfork() is mostly equivalent to pthread_atfork(), but doesn't require us to link against | |
1106 | * libpthread, as it is part of glibc anyway. */ | |
1107 | extern int __register_atfork(void (*prepare) (void), void (*parent) (void), void (*child) (void), void * __dso_handle); | |
1108 | extern void* __dso_handle __attribute__ ((__weak__)); | |
1109 | ||
1110 | pid_t getpid_cached(void) { | |
1111 | pid_t current_value; | |
1112 | ||
1113 | /* getpid_cached() is much like getpid(), but caches the value in local memory, to avoid having to invoke a | |
1114 | * system call each time. This restores glibc behaviour from before 2.24, when getpid() was unconditionally | |
1115 | * cached. Starting with 2.24 getpid() started to become prohibitively expensive when used for detecting when | |
1116 | * objects were used across fork()s. With this caching the old behaviour is somewhat restored. | |
1117 | * | |
1118 | * https://bugzilla.redhat.com/show_bug.cgi?id=1443976 | |
a4041e4f | 1119 | * https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=c579f48edba88380635ab98cb612030e3ed8691e |
5c30a6d2 LP |
1120 | */ |
1121 | ||
1122 | current_value = __sync_val_compare_and_swap(&cached_pid, CACHED_PID_UNSET, CACHED_PID_BUSY); | |
1123 | ||
1124 | switch (current_value) { | |
1125 | ||
1126 | case CACHED_PID_UNSET: { /* Not initialized yet, then do so now */ | |
1127 | pid_t new_pid; | |
1128 | ||
1129 | new_pid = getpid(); | |
1130 | ||
1131 | if (__register_atfork(NULL, NULL, reset_cached_pid, __dso_handle) != 0) { | |
1132 | /* OOM? Let's try again later */ | |
1133 | cached_pid = CACHED_PID_UNSET; | |
1134 | return new_pid; | |
1135 | } | |
1136 | ||
1137 | cached_pid = new_pid; | |
1138 | return new_pid; | |
1139 | } | |
1140 | ||
1141 | case CACHED_PID_BUSY: /* Somebody else is currently initializing */ | |
1142 | return getpid(); | |
1143 | ||
1144 | default: /* Properly initialized */ | |
1145 | return current_value; | |
1146 | } | |
1147 | } | |
1148 | ||
fba868fa LP |
1149 | int must_be_root(void) { |
1150 | ||
1151 | if (geteuid() == 0) | |
1152 | return 0; | |
1153 | ||
1154 | log_error("Need to be root."); | |
1155 | return -EPERM; | |
1156 | } | |
1157 | ||
4c253ed1 LP |
1158 | int safe_fork_full( |
1159 | const char *name, | |
1160 | const int except_fds[], | |
1161 | size_t n_except_fds, | |
1162 | ForkFlags flags, | |
1163 | pid_t *ret_pid) { | |
1164 | ||
1165 | pid_t original_pid, pid; | |
1f5d1e02 | 1166 | sigset_t saved_ss, ss; |
a41f6217 | 1167 | bool block_signals = false; |
b6e1fff1 | 1168 | int prio, r; |
4c253ed1 LP |
1169 | |
1170 | /* A wrapper around fork(), that does a couple of important initializations in addition to mere forking. Always | |
1171 | * returns the child's PID in *ret_pid. Returns == 0 in the child, and > 0 in the parent. */ | |
1172 | ||
b6e1fff1 LP |
1173 | prio = flags & FORK_LOG ? LOG_ERR : LOG_DEBUG; |
1174 | ||
4c253ed1 LP |
1175 | original_pid = getpid_cached(); |
1176 | ||
1f5d1e02 | 1177 | if (flags & (FORK_RESET_SIGNALS|FORK_DEATHSIG)) { |
4c253ed1 | 1178 | |
1f5d1e02 LP |
1179 | /* We temporarily block all signals, so that the new child has them blocked initially. This way, we can |
1180 | * be sure that SIGTERMs are not lost we might send to the child. */ | |
4c253ed1 | 1181 | |
4c253ed1 | 1182 | if (sigfillset(&ss) < 0) |
b6e1fff1 | 1183 | return log_full_errno(prio, errno, "Failed to reset signal set: %m"); |
4c253ed1 | 1184 | |
1f5d1e02 LP |
1185 | block_signals = true; |
1186 | ||
1187 | } else if (flags & FORK_WAIT) { | |
1188 | ||
1189 | /* Let's block SIGCHLD at least, so that we can safely watch for the child process */ | |
1190 | ||
1191 | if (sigemptyset(&ss) < 0) | |
1192 | return log_full_errno(prio, errno, "Failed to clear signal set: %m"); | |
1193 | ||
1194 | if (sigaddset(&ss, SIGCHLD) < 0) | |
1195 | return log_full_errno(prio, errno, "Failed to add SIGCHLD to signal set: %m"); | |
1196 | ||
1197 | block_signals = true; | |
4c253ed1 LP |
1198 | } |
1199 | ||
1f5d1e02 LP |
1200 | if (block_signals) |
1201 | if (sigprocmask(SIG_SETMASK, &ss, &saved_ss) < 0) | |
1202 | return log_full_errno(prio, errno, "Failed to set signal mask: %m"); | |
1203 | ||
be39f6ee LP |
1204 | if (flags & FORK_NEW_MOUNTNS) |
1205 | pid = raw_clone(SIGCHLD|CLONE_NEWNS); | |
1206 | else | |
1207 | pid = fork(); | |
4c253ed1 LP |
1208 | if (pid < 0) { |
1209 | r = -errno; | |
1210 | ||
1211 | if (block_signals) /* undo what we did above */ | |
1212 | (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL); | |
1213 | ||
b6e1fff1 | 1214 | return log_full_errno(prio, r, "Failed to fork: %m"); |
4c253ed1 LP |
1215 | } |
1216 | if (pid > 0) { | |
1217 | /* We are in the parent process */ | |
1218 | ||
1f5d1e02 LP |
1219 | log_debug("Successfully forked off '%s' as PID " PID_FMT ".", strna(name), pid); |
1220 | ||
1221 | if (flags & FORK_WAIT) { | |
1222 | r = wait_for_terminate_and_check(name, pid, (flags & FORK_LOG ? WAIT_LOG : 0)); | |
1223 | if (r < 0) | |
1224 | return r; | |
1225 | if (r != EXIT_SUCCESS) /* exit status > 0 should be treated as failure, too */ | |
1226 | return -EPROTO; | |
1227 | } | |
1228 | ||
4c253ed1 LP |
1229 | if (block_signals) /* undo what we did above */ |
1230 | (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL); | |
1231 | ||
4c253ed1 LP |
1232 | if (ret_pid) |
1233 | *ret_pid = pid; | |
1234 | ||
1235 | return 1; | |
1236 | } | |
1237 | ||
1238 | /* We are in the child process */ | |
1239 | ||
1240 | if (flags & FORK_REOPEN_LOG) { | |
1241 | /* Close the logs if requested, before we log anything. And make sure we reopen it if needed. */ | |
1242 | log_close(); | |
1243 | log_set_open_when_needed(true); | |
1244 | } | |
1245 | ||
1246 | if (name) { | |
1247 | r = rename_process(name); | |
1248 | if (r < 0) | |
b6e1fff1 LP |
1249 | log_full_errno(flags & FORK_LOG ? LOG_WARNING : LOG_DEBUG, |
1250 | r, "Failed to rename process, ignoring: %m"); | |
4c253ed1 LP |
1251 | } |
1252 | ||
1253 | if (flags & FORK_DEATHSIG) | |
1254 | if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0) { | |
b6e1fff1 | 1255 | log_full_errno(prio, errno, "Failed to set death signal: %m"); |
4c253ed1 LP |
1256 | _exit(EXIT_FAILURE); |
1257 | } | |
1258 | ||
1259 | if (flags & FORK_RESET_SIGNALS) { | |
1260 | r = reset_all_signal_handlers(); | |
1261 | if (r < 0) { | |
b6e1fff1 | 1262 | log_full_errno(prio, r, "Failed to reset signal handlers: %m"); |
4c253ed1 LP |
1263 | _exit(EXIT_FAILURE); |
1264 | } | |
1265 | ||
1266 | /* This implicitly undoes the signal mask stuff we did before the fork()ing above */ | |
1267 | r = reset_signal_mask(); | |
1268 | if (r < 0) { | |
b6e1fff1 | 1269 | log_full_errno(prio, r, "Failed to reset signal mask: %m"); |
4c253ed1 LP |
1270 | _exit(EXIT_FAILURE); |
1271 | } | |
1272 | } else if (block_signals) { /* undo what we did above */ | |
1273 | if (sigprocmask(SIG_SETMASK, &saved_ss, NULL) < 0) { | |
b6e1fff1 | 1274 | log_full_errno(prio, errno, "Failed to restore signal mask: %m"); |
4c253ed1 LP |
1275 | _exit(EXIT_FAILURE); |
1276 | } | |
1277 | } | |
1278 | ||
1279 | if (flags & FORK_DEATHSIG) { | |
7ddc2dc5 | 1280 | pid_t ppid; |
4c253ed1 LP |
1281 | /* Let's see if the parent PID is still the one we started from? If not, then the parent |
1282 | * already died by the time we set PR_SET_PDEATHSIG, hence let's emulate the effect */ | |
1283 | ||
7ddc2dc5 SL |
1284 | ppid = getppid(); |
1285 | if (ppid == 0) | |
1286 | /* Parent is in a differn't PID namespace. */; | |
1287 | else if (ppid != original_pid) { | |
4c253ed1 LP |
1288 | log_debug("Parent died early, raising SIGTERM."); |
1289 | (void) raise(SIGTERM); | |
1290 | _exit(EXIT_FAILURE); | |
1291 | } | |
1292 | } | |
1293 | ||
1294 | if (flags & FORK_CLOSE_ALL_FDS) { | |
1295 | /* Close the logs here in case it got reopened above, as close_all_fds() would close them for us */ | |
1296 | log_close(); | |
1297 | ||
1298 | r = close_all_fds(except_fds, n_except_fds); | |
1299 | if (r < 0) { | |
b6e1fff1 | 1300 | log_full_errno(prio, r, "Failed to close all file descriptors: %m"); |
4c253ed1 LP |
1301 | _exit(EXIT_FAILURE); |
1302 | } | |
1303 | } | |
1304 | ||
1305 | /* When we were asked to reopen the logs, do so again now */ | |
1306 | if (flags & FORK_REOPEN_LOG) { | |
1307 | log_open(); | |
1308 | log_set_open_when_needed(false); | |
1309 | } | |
1310 | ||
1311 | if (flags & FORK_NULL_STDIO) { | |
1312 | r = make_null_stdio(); | |
1313 | if (r < 0) { | |
b6e1fff1 | 1314 | log_full_errno(prio, r, "Failed to connect stdin/stdout to /dev/null: %m"); |
4c253ed1 LP |
1315 | _exit(EXIT_FAILURE); |
1316 | } | |
1317 | } | |
1318 | ||
1319 | if (ret_pid) | |
1320 | *ret_pid = getpid_cached(); | |
1321 | ||
1322 | return 0; | |
1323 | } | |
1324 | ||
78752f2e LP |
1325 | int fork_agent(const char *name, const int except[], unsigned n_except, pid_t *ret_pid, const char *path, ...) { |
1326 | bool stdout_is_tty, stderr_is_tty; | |
1327 | unsigned n, i; | |
1328 | va_list ap; | |
1329 | char **l; | |
1330 | int r; | |
1331 | ||
1332 | assert(path); | |
1333 | ||
1334 | /* Spawns a temporary TTY agent, making sure it goes away when we go away */ | |
1335 | ||
1336 | r = safe_fork_full(name, except, n_except, FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_CLOSE_ALL_FDS, ret_pid); | |
1337 | if (r < 0) | |
1338 | return r; | |
1339 | if (r > 0) | |
1340 | return 0; | |
1341 | ||
1342 | /* In the child: */ | |
1343 | ||
1344 | stdout_is_tty = isatty(STDOUT_FILENO); | |
1345 | stderr_is_tty = isatty(STDERR_FILENO); | |
1346 | ||
1347 | if (!stdout_is_tty || !stderr_is_tty) { | |
1348 | int fd; | |
1349 | ||
1350 | /* Detach from stdout/stderr. and reopen | |
1351 | * /dev/tty for them. This is important to | |
1352 | * ensure that when systemctl is started via | |
1353 | * popen() or a similar call that expects to | |
1354 | * read EOF we actually do generate EOF and | |
1355 | * not delay this indefinitely by because we | |
1356 | * keep an unused copy of stdin around. */ | |
1357 | fd = open("/dev/tty", O_WRONLY); | |
1358 | if (fd < 0) { | |
1359 | log_error_errno(errno, "Failed to open /dev/tty: %m"); | |
1360 | _exit(EXIT_FAILURE); | |
1361 | } | |
1362 | ||
1363 | if (!stdout_is_tty && dup2(fd, STDOUT_FILENO) < 0) { | |
1364 | log_error_errno(errno, "Failed to dup2 /dev/tty: %m"); | |
1365 | _exit(EXIT_FAILURE); | |
1366 | } | |
1367 | ||
1368 | if (!stderr_is_tty && dup2(fd, STDERR_FILENO) < 0) { | |
1369 | log_error_errno(errno, "Failed to dup2 /dev/tty: %m"); | |
1370 | _exit(EXIT_FAILURE); | |
1371 | } | |
1372 | ||
1373 | if (fd > STDERR_FILENO) | |
1374 | close(fd); | |
1375 | } | |
1376 | ||
1377 | /* Count arguments */ | |
1378 | va_start(ap, path); | |
1379 | for (n = 0; va_arg(ap, char*); n++) | |
1380 | ; | |
1381 | va_end(ap); | |
1382 | ||
1383 | /* Allocate strv */ | |
1384 | l = alloca(sizeof(char *) * (n + 1)); | |
1385 | ||
1386 | /* Fill in arguments */ | |
1387 | va_start(ap, path); | |
1388 | for (i = 0; i <= n; i++) | |
1389 | l[i] = va_arg(ap, char*); | |
1390 | va_end(ap); | |
1391 | ||
1392 | execv(path, l); | |
1393 | _exit(EXIT_FAILURE); | |
1394 | } | |
1395 | ||
7b3e062c LP |
1396 | static const char *const ioprio_class_table[] = { |
1397 | [IOPRIO_CLASS_NONE] = "none", | |
1398 | [IOPRIO_CLASS_RT] = "realtime", | |
1399 | [IOPRIO_CLASS_BE] = "best-effort", | |
1400 | [IOPRIO_CLASS_IDLE] = "idle" | |
1401 | }; | |
1402 | ||
1403 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ioprio_class, int, INT_MAX); | |
1404 | ||
1405 | static const char *const sigchld_code_table[] = { | |
1406 | [CLD_EXITED] = "exited", | |
1407 | [CLD_KILLED] = "killed", | |
1408 | [CLD_DUMPED] = "dumped", | |
1409 | [CLD_TRAPPED] = "trapped", | |
1410 | [CLD_STOPPED] = "stopped", | |
1411 | [CLD_CONTINUED] = "continued", | |
1412 | }; | |
1413 | ||
1414 | DEFINE_STRING_TABLE_LOOKUP(sigchld_code, int); | |
1415 | ||
1416 | static const char* const sched_policy_table[] = { | |
1417 | [SCHED_OTHER] = "other", | |
1418 | [SCHED_BATCH] = "batch", | |
1419 | [SCHED_IDLE] = "idle", | |
1420 | [SCHED_FIFO] = "fifo", | |
1421 | [SCHED_RR] = "rr" | |
1422 | }; | |
1423 | ||
1424 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(sched_policy, int, INT_MAX); |