]>
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> | |
11c3a366 | 29 | #include <stdlib.h> |
4f5dd394 | 30 | #include <string.h> |
9bfaffd5 | 31 | #include <sys/mman.h> |
7b3e062c | 32 | #include <sys/personality.h> |
405f8907 | 33 | #include <sys/prctl.h> |
4f5dd394 LP |
34 | #include <sys/types.h> |
35 | #include <sys/wait.h> | |
11c3a366 | 36 | #include <syslog.h> |
4f5dd394 | 37 | #include <unistd.h> |
349cc4a5 | 38 | #if HAVE_VALGRIND_VALGRIND_H |
dcadc967 EV |
39 | #include <valgrind/valgrind.h> |
40 | #endif | |
0b452006 | 41 | |
b5efdb8a | 42 | #include "alloc-util.h" |
6e5f1b57 | 43 | #include "architecture.h" |
4f5dd394 | 44 | #include "escape.h" |
3ffd4af2 | 45 | #include "fd-util.h" |
0b452006 | 46 | #include "fileio.h" |
f4f15635 | 47 | #include "fs-util.h" |
7b3e062c | 48 | #include "ioprio.h" |
0b452006 | 49 | #include "log.h" |
11c3a366 TA |
50 | #include "macro.h" |
51 | #include "missing.h" | |
93cc7779 | 52 | #include "process-util.h" |
8869a0b4 | 53 | #include "raw-clone.h" |
93cc7779 | 54 | #include "signal-util.h" |
1359fffa | 55 | #include "stat-util.h" |
7b3e062c | 56 | #include "string-table.h" |
07630cea | 57 | #include "string-util.h" |
b1d4f8e1 | 58 | #include "user-util.h" |
4f5dd394 | 59 | #include "util.h" |
0b452006 RC |
60 | |
61 | int get_process_state(pid_t pid) { | |
62 | const char *p; | |
63 | char state; | |
64 | int r; | |
65 | _cleanup_free_ char *line = NULL; | |
66 | ||
67 | assert(pid >= 0); | |
68 | ||
69 | p = procfs_file_alloca(pid, "stat"); | |
a644184a | 70 | |
0b452006 | 71 | r = read_one_line_file(p, &line); |
a644184a LP |
72 | if (r == -ENOENT) |
73 | return -ESRCH; | |
0b452006 RC |
74 | if (r < 0) |
75 | return r; | |
76 | ||
77 | p = strrchr(line, ')'); | |
78 | if (!p) | |
79 | return -EIO; | |
80 | ||
81 | p++; | |
82 | ||
83 | if (sscanf(p, " %c", &state) != 1) | |
84 | return -EIO; | |
85 | ||
86 | return (unsigned char) state; | |
87 | } | |
88 | ||
89 | int get_process_comm(pid_t pid, char **name) { | |
90 | const char *p; | |
91 | int r; | |
92 | ||
93 | assert(name); | |
94 | assert(pid >= 0); | |
95 | ||
96 | p = procfs_file_alloca(pid, "comm"); | |
97 | ||
98 | r = read_one_line_file(p, name); | |
99 | if (r == -ENOENT) | |
100 | return -ESRCH; | |
101 | ||
102 | return r; | |
103 | } | |
104 | ||
105 | int get_process_cmdline(pid_t pid, size_t max_length, bool comm_fallback, char **line) { | |
106 | _cleanup_fclose_ FILE *f = NULL; | |
ba4cd7e2 | 107 | bool space = false; |
c0534780 | 108 | char *k, *ans = NULL; |
0b452006 RC |
109 | const char *p; |
110 | int c; | |
111 | ||
112 | assert(line); | |
113 | assert(pid >= 0); | |
114 | ||
69281c49 LP |
115 | /* Retrieves a process' command line. Replaces unprintable characters while doing so by whitespace (coalescing |
116 | * multiple sequential ones into one). If max_length is != 0 will return a string of the specified size at most | |
117 | * (the trailing NUL byte does count towards the length here!), abbreviated with a "..." ellipsis. If | |
118 | * comm_fallback is true and the process has no command line set (the case for kernel threads), or has a | |
119 | * command line that resolves to the empty string will return the "comm" name of the process instead. | |
120 | * | |
121 | * Returns -ESRCH if the process doesn't exist, and -ENOENT if the process has no command line (and | |
c0534780 | 122 | * comm_fallback is false). Returns 0 and sets *line otherwise. */ |
69281c49 | 123 | |
0b452006 RC |
124 | p = procfs_file_alloca(pid, "cmdline"); |
125 | ||
126 | f = fopen(p, "re"); | |
a644184a LP |
127 | if (!f) { |
128 | if (errno == ENOENT) | |
129 | return -ESRCH; | |
0b452006 | 130 | return -errno; |
a644184a | 131 | } |
0b452006 | 132 | |
69281c49 LP |
133 | if (max_length == 1) { |
134 | ||
135 | /* If there's only room for one byte, return the empty string */ | |
c0534780 ZJS |
136 | ans = new0(char, 1); |
137 | if (!ans) | |
69281c49 LP |
138 | return -ENOMEM; |
139 | ||
c0534780 | 140 | *line = ans; |
69281c49 LP |
141 | return 0; |
142 | ||
143 | } else if (max_length == 0) { | |
0b452006 RC |
144 | size_t len = 0, allocated = 0; |
145 | ||
146 | while ((c = getc(f)) != EOF) { | |
147 | ||
c0534780 ZJS |
148 | if (!GREEDY_REALLOC(ans, allocated, len+3)) { |
149 | free(ans); | |
0b452006 RC |
150 | return -ENOMEM; |
151 | } | |
152 | ||
ba4cd7e2 MP |
153 | if (isprint(c)) { |
154 | if (space) { | |
c0534780 | 155 | ans[len++] = ' '; |
ba4cd7e2 MP |
156 | space = false; |
157 | } | |
158 | ||
c0534780 | 159 | ans[len++] = c; |
69281c49 | 160 | } else if (len > 0) |
ba4cd7e2 MP |
161 | space = true; |
162 | } | |
0b452006 RC |
163 | |
164 | if (len > 0) | |
c0534780 | 165 | ans[len] = '\0'; |
b09df4e2 | 166 | else |
c0534780 | 167 | ans = mfree(ans); |
0b452006 RC |
168 | |
169 | } else { | |
69281c49 | 170 | bool dotdotdot = false; |
0b452006 RC |
171 | size_t left; |
172 | ||
c0534780 ZJS |
173 | ans = new(char, max_length); |
174 | if (!ans) | |
0b452006 RC |
175 | return -ENOMEM; |
176 | ||
c0534780 | 177 | k = ans; |
0b452006 RC |
178 | left = max_length; |
179 | while ((c = getc(f)) != EOF) { | |
180 | ||
181 | if (isprint(c)) { | |
69281c49 | 182 | |
0b452006 | 183 | if (space) { |
69281c49 LP |
184 | if (left <= 2) { |
185 | dotdotdot = true; | |
0b452006 | 186 | break; |
69281c49 | 187 | } |
0b452006 RC |
188 | |
189 | *(k++) = ' '; | |
190 | left--; | |
191 | space = false; | |
192 | } | |
193 | ||
69281c49 LP |
194 | if (left <= 1) { |
195 | dotdotdot = true; | |
0b452006 | 196 | break; |
69281c49 | 197 | } |
0b452006 RC |
198 | |
199 | *(k++) = (char) c; | |
200 | left--; | |
c0534780 | 201 | } else if (k > ans) |
0b452006 RC |
202 | space = true; |
203 | } | |
204 | ||
69281c49 LP |
205 | if (dotdotdot) { |
206 | if (max_length <= 4) { | |
c0534780 | 207 | k = ans; |
69281c49 LP |
208 | left = max_length; |
209 | } else { | |
c0534780 | 210 | k = ans + max_length - 4; |
69281c49 LP |
211 | left = 4; |
212 | ||
213 | /* Eat up final spaces */ | |
c0534780 | 214 | while (k > ans && isspace(k[-1])) { |
69281c49 LP |
215 | k--; |
216 | left++; | |
217 | } | |
218 | } | |
219 | ||
220 | strncpy(k, "...", left-1); | |
b09df4e2 | 221 | k[left-1] = 0; |
0b452006 RC |
222 | } else |
223 | *k = 0; | |
224 | } | |
225 | ||
226 | /* Kernel threads have no argv[] */ | |
c0534780 | 227 | if (isempty(ans)) { |
0b452006 RC |
228 | _cleanup_free_ char *t = NULL; |
229 | int h; | |
230 | ||
c0534780 | 231 | free(ans); |
0b452006 RC |
232 | |
233 | if (!comm_fallback) | |
234 | return -ENOENT; | |
235 | ||
236 | h = get_process_comm(pid, &t); | |
237 | if (h < 0) | |
238 | return h; | |
239 | ||
69281c49 | 240 | if (max_length == 0) |
c0534780 | 241 | ans = strjoin("[", t, "]"); |
69281c49 LP |
242 | else { |
243 | size_t l; | |
244 | ||
245 | l = strlen(t); | |
246 | ||
247 | if (l + 3 <= max_length) | |
c0534780 | 248 | ans = strjoin("[", t, "]"); |
69281c49 LP |
249 | else if (max_length <= 6) { |
250 | ||
c0534780 ZJS |
251 | ans = new(char, max_length); |
252 | if (!ans) | |
69281c49 LP |
253 | return -ENOMEM; |
254 | ||
c0534780 ZJS |
255 | memcpy(ans, "[...]", max_length-1); |
256 | ans[max_length-1] = 0; | |
69281c49 LP |
257 | } else { |
258 | char *e; | |
259 | ||
260 | t[max_length - 6] = 0; | |
261 | ||
262 | /* Chop off final spaces */ | |
263 | e = strchr(t, 0); | |
264 | while (e > t && isspace(e[-1])) | |
265 | e--; | |
266 | *e = 0; | |
267 | ||
c0534780 | 268 | ans = strjoin("[", t, "...]"); |
69281c49 LP |
269 | } |
270 | } | |
c0534780 | 271 | if (!ans) |
0b452006 RC |
272 | return -ENOMEM; |
273 | } | |
274 | ||
c0534780 | 275 | *line = ans; |
0b452006 RC |
276 | return 0; |
277 | } | |
278 | ||
9bfaffd5 LP |
279 | int rename_process(const char name[]) { |
280 | static size_t mm_size = 0; | |
281 | static char *mm = NULL; | |
282 | bool truncated = false; | |
283 | size_t l; | |
284 | ||
285 | /* This is a like a poor man's setproctitle(). It changes the comm field, argv[0], and also the glibc's | |
286 | * internally used name of the process. For the first one a limit of 16 chars applies; to the second one in | |
287 | * many cases one of 10 (i.e. length of "/sbin/init") — however if we have CAP_SYS_RESOURCES it is unbounded; | |
288 | * to the third one 7 (i.e. the length of "systemd". If you pass a longer string it will likely be | |
289 | * truncated. | |
290 | * | |
291 | * Returns 0 if a name was set but truncated, > 0 if it was set but not truncated. */ | |
292 | ||
293 | if (isempty(name)) | |
294 | return -EINVAL; /* let's not confuse users unnecessarily with an empty name */ | |
405f8907 | 295 | |
9bfaffd5 | 296 | l = strlen(name); |
405f8907 | 297 | |
9bfaffd5 | 298 | /* First step, change the comm field. */ |
79d62972 | 299 | (void) prctl(PR_SET_NAME, name); |
9bfaffd5 LP |
300 | if (l > 15) /* Linux process names can be 15 chars at max */ |
301 | truncated = true; | |
302 | ||
303 | /* Second step, change glibc's ID of the process name. */ | |
304 | if (program_invocation_name) { | |
305 | size_t k; | |
306 | ||
307 | k = strlen(program_invocation_name); | |
308 | strncpy(program_invocation_name, name, k); | |
309 | if (l > k) | |
310 | truncated = true; | |
311 | } | |
312 | ||
313 | /* Third step, completely replace the argv[] array the kernel maintains for us. This requires privileges, but | |
314 | * has the advantage that the argv[] array is exactly what we want it to be, and not filled up with zeros at | |
13e785f7 | 315 | * the end. This is the best option for changing /proc/self/cmdline. */ |
01f989c6 JW |
316 | |
317 | /* Let's not bother with this if we don't have euid == 0. Strictly speaking we should check for the | |
318 | * CAP_SYS_RESOURCE capability which is independent of the euid. In our own code the capability generally is | |
319 | * present only for euid == 0, hence let's use this as quick bypass check, to avoid calling mmap() if | |
320 | * PR_SET_MM_ARG_{START,END} fails with EPERM later on anyway. After all geteuid() is dead cheap to call, but | |
321 | * mmap() is not. */ | |
322 | if (geteuid() != 0) | |
323 | log_debug("Skipping PR_SET_MM, as we don't have privileges."); | |
324 | else if (mm_size < l+1) { | |
9bfaffd5 LP |
325 | size_t nn_size; |
326 | char *nn; | |
327 | ||
9bfaffd5 LP |
328 | nn_size = PAGE_ALIGN(l+1); |
329 | nn = mmap(NULL, nn_size, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); | |
330 | if (nn == MAP_FAILED) { | |
331 | log_debug_errno(errno, "mmap() failed: %m"); | |
332 | goto use_saved_argv; | |
333 | } | |
405f8907 | 334 | |
9bfaffd5 LP |
335 | strncpy(nn, name, nn_size); |
336 | ||
337 | /* Now, let's tell the kernel about this new memory */ | |
338 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_START, (unsigned long) nn, 0, 0) < 0) { | |
339 | log_debug_errno(errno, "PR_SET_MM_ARG_START failed, proceeding without: %m"); | |
340 | (void) munmap(nn, nn_size); | |
341 | goto use_saved_argv; | |
342 | } | |
343 | ||
344 | /* And update the end pointer to the new end, too. If this fails, we don't really know what to do, it's | |
345 | * pretty unlikely that we can rollback, hence we'll just accept the failure, and continue. */ | |
346 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) nn + l + 1, 0, 0) < 0) | |
347 | log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m"); | |
348 | ||
349 | if (mm) | |
350 | (void) munmap(mm, mm_size); | |
351 | ||
352 | mm = nn; | |
353 | mm_size = nn_size; | |
01f989c6 | 354 | } else { |
9bfaffd5 LP |
355 | strncpy(mm, name, mm_size); |
356 | ||
01f989c6 JW |
357 | /* Update the end pointer, continuing regardless of any failure. */ |
358 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) mm + l + 1, 0, 0) < 0) | |
359 | log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m"); | |
360 | } | |
361 | ||
9bfaffd5 LP |
362 | use_saved_argv: |
363 | /* Fourth step: in all cases we'll also update the original argv[], so that our own code gets it right too if | |
364 | * it still looks here */ | |
405f8907 LP |
365 | |
366 | if (saved_argc > 0) { | |
367 | int i; | |
368 | ||
9bfaffd5 LP |
369 | if (saved_argv[0]) { |
370 | size_t k; | |
371 | ||
372 | k = strlen(saved_argv[0]); | |
373 | strncpy(saved_argv[0], name, k); | |
374 | if (l > k) | |
375 | truncated = true; | |
376 | } | |
405f8907 LP |
377 | |
378 | for (i = 1; i < saved_argc; i++) { | |
379 | if (!saved_argv[i]) | |
380 | break; | |
381 | ||
382 | memzero(saved_argv[i], strlen(saved_argv[i])); | |
383 | } | |
384 | } | |
9bfaffd5 LP |
385 | |
386 | return !truncated; | |
405f8907 LP |
387 | } |
388 | ||
0b452006 RC |
389 | int is_kernel_thread(pid_t pid) { |
390 | const char *p; | |
391 | size_t count; | |
392 | char c; | |
393 | bool eof; | |
394 | FILE *f; | |
395 | ||
4c701096 | 396 | if (IN_SET(pid, 0, 1) || pid == getpid_cached()) /* pid 1, and we ourselves certainly aren't a kernel thread */ |
0b452006 RC |
397 | return 0; |
398 | ||
a6149b93 | 399 | assert(pid > 1); |
0b452006 RC |
400 | |
401 | p = procfs_file_alloca(pid, "cmdline"); | |
402 | f = fopen(p, "re"); | |
a644184a LP |
403 | if (!f) { |
404 | if (errno == ENOENT) | |
405 | return -ESRCH; | |
0b452006 | 406 | return -errno; |
a644184a | 407 | } |
0b452006 RC |
408 | |
409 | count = fread(&c, 1, 1, f); | |
410 | eof = feof(f); | |
411 | fclose(f); | |
412 | ||
413 | /* Kernel threads have an empty cmdline */ | |
414 | ||
415 | if (count <= 0) | |
416 | return eof ? 1 : -errno; | |
417 | ||
418 | return 0; | |
419 | } | |
420 | ||
421 | int get_process_capeff(pid_t pid, char **capeff) { | |
422 | const char *p; | |
a644184a | 423 | int r; |
0b452006 RC |
424 | |
425 | assert(capeff); | |
426 | assert(pid >= 0); | |
427 | ||
428 | p = procfs_file_alloca(pid, "status"); | |
429 | ||
c4cd1d4d | 430 | r = get_proc_field(p, "CapEff", WHITESPACE, capeff); |
a644184a LP |
431 | if (r == -ENOENT) |
432 | return -ESRCH; | |
433 | ||
434 | return r; | |
0b452006 RC |
435 | } |
436 | ||
437 | static int get_process_link_contents(const char *proc_file, char **name) { | |
438 | int r; | |
439 | ||
440 | assert(proc_file); | |
441 | assert(name); | |
442 | ||
443 | r = readlink_malloc(proc_file, name); | |
a644184a LP |
444 | if (r == -ENOENT) |
445 | return -ESRCH; | |
0b452006 | 446 | if (r < 0) |
a644184a | 447 | return r; |
0b452006 RC |
448 | |
449 | return 0; | |
450 | } | |
451 | ||
452 | int get_process_exe(pid_t pid, char **name) { | |
453 | const char *p; | |
454 | char *d; | |
455 | int r; | |
456 | ||
457 | assert(pid >= 0); | |
458 | ||
459 | p = procfs_file_alloca(pid, "exe"); | |
460 | r = get_process_link_contents(p, name); | |
461 | if (r < 0) | |
462 | return r; | |
463 | ||
464 | d = endswith(*name, " (deleted)"); | |
465 | if (d) | |
466 | *d = '\0'; | |
467 | ||
468 | return 0; | |
469 | } | |
470 | ||
471 | static int get_process_id(pid_t pid, const char *field, uid_t *uid) { | |
472 | _cleanup_fclose_ FILE *f = NULL; | |
473 | char line[LINE_MAX]; | |
474 | const char *p; | |
475 | ||
476 | assert(field); | |
477 | assert(uid); | |
478 | ||
07b38ba5 | 479 | if (pid < 0) |
6f8cbcdb LP |
480 | return -EINVAL; |
481 | ||
0b452006 RC |
482 | p = procfs_file_alloca(pid, "status"); |
483 | f = fopen(p, "re"); | |
a644184a LP |
484 | if (!f) { |
485 | if (errno == ENOENT) | |
486 | return -ESRCH; | |
0b452006 | 487 | return -errno; |
a644184a | 488 | } |
0b452006 RC |
489 | |
490 | FOREACH_LINE(line, f, return -errno) { | |
491 | char *l; | |
492 | ||
493 | l = strstrip(line); | |
494 | ||
495 | if (startswith(l, field)) { | |
496 | l += strlen(field); | |
497 | l += strspn(l, WHITESPACE); | |
498 | ||
499 | l[strcspn(l, WHITESPACE)] = 0; | |
500 | ||
501 | return parse_uid(l, uid); | |
502 | } | |
503 | } | |
504 | ||
505 | return -EIO; | |
506 | } | |
507 | ||
508 | int get_process_uid(pid_t pid, uid_t *uid) { | |
6f8cbcdb LP |
509 | |
510 | if (pid == 0 || pid == getpid_cached()) { | |
511 | *uid = getuid(); | |
512 | return 0; | |
513 | } | |
514 | ||
0b452006 RC |
515 | return get_process_id(pid, "Uid:", uid); |
516 | } | |
517 | ||
518 | int get_process_gid(pid_t pid, gid_t *gid) { | |
6f8cbcdb LP |
519 | |
520 | if (pid == 0 || pid == getpid_cached()) { | |
521 | *gid = getgid(); | |
522 | return 0; | |
523 | } | |
524 | ||
0b452006 RC |
525 | assert_cc(sizeof(uid_t) == sizeof(gid_t)); |
526 | return get_process_id(pid, "Gid:", gid); | |
527 | } | |
528 | ||
529 | int get_process_cwd(pid_t pid, char **cwd) { | |
530 | const char *p; | |
531 | ||
532 | assert(pid >= 0); | |
533 | ||
534 | p = procfs_file_alloca(pid, "cwd"); | |
535 | ||
536 | return get_process_link_contents(p, cwd); | |
537 | } | |
538 | ||
539 | int get_process_root(pid_t pid, char **root) { | |
540 | const char *p; | |
541 | ||
542 | assert(pid >= 0); | |
543 | ||
544 | p = procfs_file_alloca(pid, "root"); | |
545 | ||
546 | return get_process_link_contents(p, root); | |
547 | } | |
548 | ||
549 | int get_process_environ(pid_t pid, char **env) { | |
550 | _cleanup_fclose_ FILE *f = NULL; | |
551 | _cleanup_free_ char *outcome = NULL; | |
552 | int c; | |
553 | const char *p; | |
554 | size_t allocated = 0, sz = 0; | |
555 | ||
556 | assert(pid >= 0); | |
557 | assert(env); | |
558 | ||
559 | p = procfs_file_alloca(pid, "environ"); | |
560 | ||
561 | f = fopen(p, "re"); | |
a644184a LP |
562 | if (!f) { |
563 | if (errno == ENOENT) | |
564 | return -ESRCH; | |
0b452006 | 565 | return -errno; |
a644184a | 566 | } |
0b452006 RC |
567 | |
568 | while ((c = fgetc(f)) != EOF) { | |
569 | if (!GREEDY_REALLOC(outcome, allocated, sz + 5)) | |
570 | return -ENOMEM; | |
571 | ||
572 | if (c == '\0') | |
573 | outcome[sz++] = '\n'; | |
574 | else | |
575 | sz += cescape_char(c, outcome + sz); | |
576 | } | |
577 | ||
03c55bc0 LP |
578 | if (!outcome) { |
579 | outcome = strdup(""); | |
580 | if (!outcome) | |
581 | return -ENOMEM; | |
582 | } else | |
583 | outcome[sz] = '\0'; | |
de8763b6 | 584 | |
0b452006 RC |
585 | *env = outcome; |
586 | outcome = NULL; | |
587 | ||
588 | return 0; | |
589 | } | |
590 | ||
6bc73acb | 591 | int get_process_ppid(pid_t pid, pid_t *_ppid) { |
0b452006 RC |
592 | int r; |
593 | _cleanup_free_ char *line = NULL; | |
594 | long unsigned ppid; | |
595 | const char *p; | |
596 | ||
597 | assert(pid >= 0); | |
598 | assert(_ppid); | |
599 | ||
6f8cbcdb | 600 | if (pid == 0 || pid == getpid_cached()) { |
0b452006 RC |
601 | *_ppid = getppid(); |
602 | return 0; | |
603 | } | |
604 | ||
605 | p = procfs_file_alloca(pid, "stat"); | |
606 | r = read_one_line_file(p, &line); | |
a644184a LP |
607 | if (r == -ENOENT) |
608 | return -ESRCH; | |
0b452006 RC |
609 | if (r < 0) |
610 | return r; | |
611 | ||
612 | /* Let's skip the pid and comm fields. The latter is enclosed | |
613 | * in () but does not escape any () in its value, so let's | |
614 | * skip over it manually */ | |
615 | ||
616 | p = strrchr(line, ')'); | |
617 | if (!p) | |
618 | return -EIO; | |
619 | ||
620 | p++; | |
621 | ||
622 | if (sscanf(p, " " | |
623 | "%*c " /* state */ | |
624 | "%lu ", /* ppid */ | |
625 | &ppid) != 1) | |
626 | return -EIO; | |
627 | ||
628 | if ((long unsigned) (pid_t) ppid != ppid) | |
629 | return -ERANGE; | |
630 | ||
631 | *_ppid = (pid_t) ppid; | |
632 | ||
633 | return 0; | |
634 | } | |
635 | ||
636 | int wait_for_terminate(pid_t pid, siginfo_t *status) { | |
637 | siginfo_t dummy; | |
638 | ||
639 | assert(pid >= 1); | |
640 | ||
641 | if (!status) | |
642 | status = &dummy; | |
643 | ||
644 | for (;;) { | |
645 | zero(*status); | |
646 | ||
647 | if (waitid(P_PID, pid, status, WEXITED) < 0) { | |
648 | ||
649 | if (errno == EINTR) | |
650 | continue; | |
651 | ||
3f0083a2 | 652 | return negative_errno(); |
0b452006 RC |
653 | } |
654 | ||
655 | return 0; | |
656 | } | |
657 | } | |
658 | ||
659 | /* | |
660 | * Return values: | |
661 | * < 0 : wait_for_terminate() failed to get the state of the | |
662 | * process, the process was terminated by a signal, or | |
663 | * failed for an unknown reason. | |
664 | * >=0 : The process terminated normally, and its exit code is | |
665 | * returned. | |
666 | * | |
667 | * That is, success is indicated by a return value of zero, and an | |
668 | * error is indicated by a non-zero value. | |
669 | * | |
670 | * A warning is emitted if the process terminates abnormally, | |
671 | * and also if it returns non-zero unless check_exit_code is true. | |
672 | */ | |
673 | int wait_for_terminate_and_warn(const char *name, pid_t pid, bool check_exit_code) { | |
674 | int r; | |
675 | siginfo_t status; | |
676 | ||
677 | assert(name); | |
678 | assert(pid > 1); | |
679 | ||
680 | r = wait_for_terminate(pid, &status); | |
681 | if (r < 0) | |
682 | return log_warning_errno(r, "Failed to wait for %s: %m", name); | |
683 | ||
684 | if (status.si_code == CLD_EXITED) { | |
685 | if (status.si_status != 0) | |
686 | log_full(check_exit_code ? LOG_WARNING : LOG_DEBUG, | |
687 | "%s failed with error code %i.", name, status.si_status); | |
688 | else | |
689 | log_debug("%s succeeded.", name); | |
690 | ||
691 | return status.si_status; | |
3742095b | 692 | } else if (IN_SET(status.si_code, CLD_KILLED, CLD_DUMPED)) { |
0b452006 RC |
693 | |
694 | log_warning("%s terminated by signal %s.", name, signal_to_string(status.si_status)); | |
695 | return -EPROTO; | |
696 | } | |
697 | ||
698 | log_warning("%s failed due to unknown reason.", name); | |
699 | return -EPROTO; | |
700 | } | |
701 | ||
89c9030d LP |
702 | void sigkill_wait(pid_t pid) { |
703 | assert(pid > 1); | |
704 | ||
705 | if (kill(pid, SIGKILL) > 0) | |
706 | (void) wait_for_terminate(pid, NULL); | |
707 | } | |
708 | ||
709 | void sigkill_waitp(pid_t *pid) { | |
4d0d3d41 LP |
710 | if (!pid) |
711 | return; | |
712 | if (*pid <= 1) | |
713 | return; | |
714 | ||
89c9030d | 715 | sigkill_wait(*pid); |
4d0d3d41 LP |
716 | } |
717 | ||
0b452006 RC |
718 | int kill_and_sigcont(pid_t pid, int sig) { |
719 | int r; | |
720 | ||
721 | r = kill(pid, sig) < 0 ? -errno : 0; | |
722 | ||
26f417d3 LP |
723 | /* If this worked, also send SIGCONT, unless we already just sent a SIGCONT, or SIGKILL was sent which isn't |
724 | * affected by a process being suspended anyway. */ | |
a3d8d68c | 725 | if (r >= 0 && !IN_SET(sig, SIGCONT, SIGKILL)) |
26f417d3 | 726 | (void) kill(pid, SIGCONT); |
0b452006 RC |
727 | |
728 | return r; | |
729 | } | |
730 | ||
731 | int getenv_for_pid(pid_t pid, const char *field, char **_value) { | |
732 | _cleanup_fclose_ FILE *f = NULL; | |
733 | char *value = NULL; | |
734 | int r; | |
735 | bool done = false; | |
736 | size_t l; | |
737 | const char *path; | |
738 | ||
739 | assert(pid >= 0); | |
740 | assert(field); | |
741 | assert(_value); | |
742 | ||
743 | path = procfs_file_alloca(pid, "environ"); | |
744 | ||
745 | f = fopen(path, "re"); | |
a644184a LP |
746 | if (!f) { |
747 | if (errno == ENOENT) | |
748 | return -ESRCH; | |
0b452006 | 749 | return -errno; |
a644184a | 750 | } |
0b452006 RC |
751 | |
752 | l = strlen(field); | |
753 | r = 0; | |
754 | ||
755 | do { | |
756 | char line[LINE_MAX]; | |
757 | unsigned i; | |
758 | ||
759 | for (i = 0; i < sizeof(line)-1; i++) { | |
760 | int c; | |
761 | ||
762 | c = getc(f); | |
763 | if (_unlikely_(c == EOF)) { | |
764 | done = true; | |
765 | break; | |
766 | } else if (c == 0) | |
767 | break; | |
768 | ||
769 | line[i] = c; | |
770 | } | |
771 | line[i] = 0; | |
772 | ||
041b5ae1 | 773 | if (strneq(line, field, l) && line[l] == '=') { |
0b452006 RC |
774 | value = strdup(line + l + 1); |
775 | if (!value) | |
776 | return -ENOMEM; | |
777 | ||
778 | r = 1; | |
779 | break; | |
780 | } | |
781 | ||
782 | } while (!done); | |
783 | ||
784 | *_value = value; | |
785 | return r; | |
786 | } | |
787 | ||
788 | bool pid_is_unwaited(pid_t pid) { | |
789 | /* Checks whether a PID is still valid at all, including a zombie */ | |
790 | ||
07b38ba5 | 791 | if (pid < 0) |
0b452006 RC |
792 | return false; |
793 | ||
5fd9b2c5 LP |
794 | if (pid <= 1) /* If we or PID 1 would be dead and have been waited for, this code would not be running */ |
795 | return true; | |
796 | ||
6f8cbcdb LP |
797 | if (pid == getpid_cached()) |
798 | return true; | |
799 | ||
0b452006 RC |
800 | if (kill(pid, 0) >= 0) |
801 | return true; | |
802 | ||
803 | return errno != ESRCH; | |
804 | } | |
805 | ||
806 | bool pid_is_alive(pid_t pid) { | |
807 | int r; | |
808 | ||
809 | /* Checks whether a PID is still valid and not a zombie */ | |
810 | ||
07b38ba5 | 811 | if (pid < 0) |
0b452006 RC |
812 | return false; |
813 | ||
5fd9b2c5 LP |
814 | if (pid <= 1) /* If we or PID 1 would be a zombie, this code would not be running */ |
815 | return true; | |
816 | ||
6f8cbcdb LP |
817 | if (pid == getpid_cached()) |
818 | return true; | |
819 | ||
0b452006 | 820 | r = get_process_state(pid); |
4c701096 | 821 | if (IN_SET(r, -ESRCH, 'Z')) |
0b452006 RC |
822 | return false; |
823 | ||
824 | return true; | |
825 | } | |
d4510856 | 826 | |
1359fffa MS |
827 | int pid_from_same_root_fs(pid_t pid) { |
828 | const char *root; | |
829 | ||
07b38ba5 | 830 | if (pid < 0) |
6f8cbcdb LP |
831 | return false; |
832 | ||
833 | if (pid == 0 || pid == getpid_cached()) | |
834 | return true; | |
1359fffa MS |
835 | |
836 | root = procfs_file_alloca(pid, "root"); | |
837 | ||
e3f791a2 | 838 | return files_same(root, "/proc/1/root", 0); |
1359fffa MS |
839 | } |
840 | ||
d4510856 LP |
841 | bool is_main_thread(void) { |
842 | static thread_local int cached = 0; | |
843 | ||
844 | if (_unlikely_(cached == 0)) | |
df0ff127 | 845 | cached = getpid_cached() == gettid() ? 1 : -1; |
d4510856 LP |
846 | |
847 | return cached > 0; | |
848 | } | |
7b3e062c LP |
849 | |
850 | noreturn void freeze(void) { | |
851 | ||
3da48d7a EV |
852 | log_close(); |
853 | ||
7b3e062c LP |
854 | /* Make sure nobody waits for us on a socket anymore */ |
855 | close_all_fds(NULL, 0); | |
856 | ||
857 | sync(); | |
858 | ||
859 | for (;;) | |
860 | pause(); | |
861 | } | |
862 | ||
863 | bool oom_score_adjust_is_valid(int oa) { | |
864 | return oa >= OOM_SCORE_ADJ_MIN && oa <= OOM_SCORE_ADJ_MAX; | |
865 | } | |
866 | ||
867 | unsigned long personality_from_string(const char *p) { | |
6e5f1b57 | 868 | int architecture; |
7b3e062c | 869 | |
0c0fea07 LP |
870 | if (!p) |
871 | return PERSONALITY_INVALID; | |
872 | ||
6e5f1b57 LP |
873 | /* Parse a personality specifier. We use our own identifiers that indicate specific ABIs, rather than just |
874 | * hints regarding the register size, since we want to keep things open for multiple locally supported ABIs for | |
875 | * the same register size. */ | |
876 | ||
877 | architecture = architecture_from_string(p); | |
878 | if (architecture < 0) | |
879 | return PERSONALITY_INVALID; | |
7b3e062c | 880 | |
0c0fea07 | 881 | if (architecture == native_architecture()) |
7b3e062c | 882 | return PER_LINUX; |
0c0fea07 LP |
883 | #ifdef SECONDARY_ARCHITECTURE |
884 | if (architecture == SECONDARY_ARCHITECTURE) | |
f2d1736c | 885 | return PER_LINUX32; |
7b3e062c LP |
886 | #endif |
887 | ||
888 | return PERSONALITY_INVALID; | |
889 | } | |
890 | ||
891 | const char* personality_to_string(unsigned long p) { | |
6e5f1b57 | 892 | int architecture = _ARCHITECTURE_INVALID; |
7b3e062c | 893 | |
7b3e062c | 894 | if (p == PER_LINUX) |
0c0fea07 LP |
895 | architecture = native_architecture(); |
896 | #ifdef SECONDARY_ARCHITECTURE | |
6e5f1b57 | 897 | else if (p == PER_LINUX32) |
0c0fea07 | 898 | architecture = SECONDARY_ARCHITECTURE; |
7b3e062c LP |
899 | #endif |
900 | ||
6e5f1b57 LP |
901 | if (architecture < 0) |
902 | return NULL; | |
903 | ||
904 | return architecture_to_string(architecture); | |
7b3e062c LP |
905 | } |
906 | ||
21022b9d LP |
907 | int safe_personality(unsigned long p) { |
908 | int ret; | |
909 | ||
910 | /* So here's the deal, personality() is weirdly defined by glibc. In some cases it returns a failure via errno, | |
911 | * and in others as negative return value containing an errno-like value. Let's work around this: this is a | |
912 | * wrapper that uses errno if it is set, and uses the return value otherwise. And then it sets both errno and | |
913 | * the return value indicating the same issue, so that we are definitely on the safe side. | |
914 | * | |
915 | * See https://github.com/systemd/systemd/issues/6737 */ | |
916 | ||
917 | errno = 0; | |
918 | ret = personality(p); | |
919 | if (ret < 0) { | |
920 | if (errno != 0) | |
921 | return -errno; | |
922 | ||
923 | errno = -ret; | |
924 | } | |
925 | ||
926 | return ret; | |
927 | } | |
928 | ||
e8132d63 LP |
929 | int opinionated_personality(unsigned long *ret) { |
930 | int current; | |
931 | ||
932 | /* Returns the current personality, or PERSONALITY_INVALID if we can't determine it. This function is a bit | |
933 | * opinionated though, and ignores all the finer-grained bits and exotic personalities, only distinguishing the | |
934 | * two most relevant personalities: PER_LINUX and PER_LINUX32. */ | |
935 | ||
21022b9d | 936 | current = safe_personality(PERSONALITY_INVALID); |
e8132d63 | 937 | if (current < 0) |
21022b9d | 938 | return current; |
e8132d63 LP |
939 | |
940 | if (((unsigned long) current & 0xffff) == PER_LINUX32) | |
941 | *ret = PER_LINUX32; | |
942 | else | |
943 | *ret = PER_LINUX; | |
944 | ||
945 | return 0; | |
946 | } | |
947 | ||
dcadc967 | 948 | void valgrind_summary_hack(void) { |
349cc4a5 | 949 | #if HAVE_VALGRIND_VALGRIND_H |
df0ff127 | 950 | if (getpid_cached() == 1 && RUNNING_ON_VALGRIND) { |
dcadc967 | 951 | pid_t pid; |
8869a0b4 | 952 | pid = raw_clone(SIGCHLD); |
dcadc967 EV |
953 | if (pid < 0) |
954 | log_emergency_errno(errno, "Failed to fork off valgrind helper: %m"); | |
955 | else if (pid == 0) | |
956 | exit(EXIT_SUCCESS); | |
957 | else { | |
958 | log_info("Spawned valgrind helper as PID "PID_FMT".", pid); | |
959 | (void) wait_for_terminate(pid, NULL); | |
960 | } | |
961 | } | |
962 | #endif | |
963 | } | |
964 | ||
291d565a LP |
965 | int pid_compare_func(const void *a, const void *b) { |
966 | const pid_t *p = a, *q = b; | |
967 | ||
968 | /* Suitable for usage in qsort() */ | |
969 | ||
970 | if (*p < *q) | |
971 | return -1; | |
972 | if (*p > *q) | |
973 | return 1; | |
974 | return 0; | |
975 | } | |
976 | ||
7f452159 LP |
977 | int ioprio_parse_priority(const char *s, int *ret) { |
978 | int i, r; | |
979 | ||
980 | assert(s); | |
981 | assert(ret); | |
982 | ||
983 | r = safe_atoi(s, &i); | |
984 | if (r < 0) | |
985 | return r; | |
986 | ||
987 | if (!ioprio_priority_is_valid(i)) | |
988 | return -EINVAL; | |
989 | ||
990 | *ret = i; | |
991 | return 0; | |
992 | } | |
993 | ||
5c30a6d2 LP |
994 | /* The cached PID, possible values: |
995 | * | |
996 | * == UNSET [0] → cache not initialized yet | |
997 | * == BUSY [-1] → some thread is initializing it at the moment | |
998 | * any other → the cached PID | |
999 | */ | |
1000 | ||
1001 | #define CACHED_PID_UNSET ((pid_t) 0) | |
1002 | #define CACHED_PID_BUSY ((pid_t) -1) | |
1003 | ||
1004 | static pid_t cached_pid = CACHED_PID_UNSET; | |
1005 | ||
1006 | static void reset_cached_pid(void) { | |
1007 | /* Invoked in the child after a fork(), i.e. at the first moment the PID changed */ | |
1008 | cached_pid = CACHED_PID_UNSET; | |
1009 | } | |
1010 | ||
1011 | /* We use glibc __register_atfork() + __dso_handle directly here, as they are not included in the glibc | |
1012 | * headers. __register_atfork() is mostly equivalent to pthread_atfork(), but doesn't require us to link against | |
1013 | * libpthread, as it is part of glibc anyway. */ | |
1014 | extern int __register_atfork(void (*prepare) (void), void (*parent) (void), void (*child) (void), void * __dso_handle); | |
1015 | extern void* __dso_handle __attribute__ ((__weak__)); | |
1016 | ||
1017 | pid_t getpid_cached(void) { | |
1018 | pid_t current_value; | |
1019 | ||
1020 | /* getpid_cached() is much like getpid(), but caches the value in local memory, to avoid having to invoke a | |
1021 | * system call each time. This restores glibc behaviour from before 2.24, when getpid() was unconditionally | |
1022 | * cached. Starting with 2.24 getpid() started to become prohibitively expensive when used for detecting when | |
1023 | * objects were used across fork()s. With this caching the old behaviour is somewhat restored. | |
1024 | * | |
1025 | * https://bugzilla.redhat.com/show_bug.cgi?id=1443976 | |
a4041e4f | 1026 | * https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=c579f48edba88380635ab98cb612030e3ed8691e |
5c30a6d2 LP |
1027 | */ |
1028 | ||
1029 | current_value = __sync_val_compare_and_swap(&cached_pid, CACHED_PID_UNSET, CACHED_PID_BUSY); | |
1030 | ||
1031 | switch (current_value) { | |
1032 | ||
1033 | case CACHED_PID_UNSET: { /* Not initialized yet, then do so now */ | |
1034 | pid_t new_pid; | |
1035 | ||
1036 | new_pid = getpid(); | |
1037 | ||
1038 | if (__register_atfork(NULL, NULL, reset_cached_pid, __dso_handle) != 0) { | |
1039 | /* OOM? Let's try again later */ | |
1040 | cached_pid = CACHED_PID_UNSET; | |
1041 | return new_pid; | |
1042 | } | |
1043 | ||
1044 | cached_pid = new_pid; | |
1045 | return new_pid; | |
1046 | } | |
1047 | ||
1048 | case CACHED_PID_BUSY: /* Somebody else is currently initializing */ | |
1049 | return getpid(); | |
1050 | ||
1051 | default: /* Properly initialized */ | |
1052 | return current_value; | |
1053 | } | |
1054 | } | |
1055 | ||
7b3e062c LP |
1056 | static const char *const ioprio_class_table[] = { |
1057 | [IOPRIO_CLASS_NONE] = "none", | |
1058 | [IOPRIO_CLASS_RT] = "realtime", | |
1059 | [IOPRIO_CLASS_BE] = "best-effort", | |
1060 | [IOPRIO_CLASS_IDLE] = "idle" | |
1061 | }; | |
1062 | ||
1063 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ioprio_class, int, INT_MAX); | |
1064 | ||
1065 | static const char *const sigchld_code_table[] = { | |
1066 | [CLD_EXITED] = "exited", | |
1067 | [CLD_KILLED] = "killed", | |
1068 | [CLD_DUMPED] = "dumped", | |
1069 | [CLD_TRAPPED] = "trapped", | |
1070 | [CLD_STOPPED] = "stopped", | |
1071 | [CLD_CONTINUED] = "continued", | |
1072 | }; | |
1073 | ||
1074 | DEFINE_STRING_TABLE_LOOKUP(sigchld_code, int); | |
1075 | ||
1076 | static const char* const sched_policy_table[] = { | |
1077 | [SCHED_OTHER] = "other", | |
1078 | [SCHED_BATCH] = "batch", | |
1079 | [SCHED_IDLE] = "idle", | |
1080 | [SCHED_FIFO] = "fifo", | |
1081 | [SCHED_RR] = "rr" | |
1082 | }; | |
1083 | ||
1084 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(sched_policy, int, INT_MAX); |