]>
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 | |
0b452006 RC |
6 | ***/ |
7 | ||
4f5dd394 | 8 | #include <ctype.h> |
0b452006 | 9 | #include <errno.h> |
11c3a366 TA |
10 | #include <limits.h> |
11 | #include <linux/oom.h> | |
7b3e062c | 12 | #include <sched.h> |
0b452006 | 13 | #include <signal.h> |
4f5dd394 LP |
14 | #include <stdbool.h> |
15 | #include <stdio.h> | |
35bbbf85 | 16 | #include <stdio_ext.h> |
11c3a366 | 17 | #include <stdlib.h> |
4f5dd394 | 18 | #include <string.h> |
9bfaffd5 | 19 | #include <sys/mman.h> |
7b3e062c | 20 | #include <sys/personality.h> |
405f8907 | 21 | #include <sys/prctl.h> |
4f5dd394 LP |
22 | #include <sys/types.h> |
23 | #include <sys/wait.h> | |
11c3a366 | 24 | #include <syslog.h> |
4f5dd394 | 25 | #include <unistd.h> |
349cc4a5 | 26 | #if HAVE_VALGRIND_VALGRIND_H |
dcadc967 EV |
27 | #include <valgrind/valgrind.h> |
28 | #endif | |
0b452006 | 29 | |
b5efdb8a | 30 | #include "alloc-util.h" |
6e5f1b57 | 31 | #include "architecture.h" |
4f5dd394 | 32 | #include "escape.h" |
3ffd4af2 | 33 | #include "fd-util.h" |
0b452006 | 34 | #include "fileio.h" |
f4f15635 | 35 | #include "fs-util.h" |
7b3e062c | 36 | #include "ioprio.h" |
0b452006 | 37 | #include "log.h" |
11c3a366 TA |
38 | #include "macro.h" |
39 | #include "missing.h" | |
93cc7779 | 40 | #include "process-util.h" |
8869a0b4 | 41 | #include "raw-clone.h" |
93cc7779 | 42 | #include "signal-util.h" |
1359fffa | 43 | #include "stat-util.h" |
7b3e062c | 44 | #include "string-table.h" |
07630cea | 45 | #include "string-util.h" |
4c253ed1 | 46 | #include "terminal-util.h" |
b1d4f8e1 | 47 | #include "user-util.h" |
4f5dd394 | 48 | #include "util.h" |
0b452006 RC |
49 | |
50 | int get_process_state(pid_t pid) { | |
51 | const char *p; | |
52 | char state; | |
53 | int r; | |
54 | _cleanup_free_ char *line = NULL; | |
55 | ||
56 | assert(pid >= 0); | |
57 | ||
58 | p = procfs_file_alloca(pid, "stat"); | |
a644184a | 59 | |
0b452006 | 60 | r = read_one_line_file(p, &line); |
a644184a LP |
61 | if (r == -ENOENT) |
62 | return -ESRCH; | |
0b452006 RC |
63 | if (r < 0) |
64 | return r; | |
65 | ||
66 | p = strrchr(line, ')'); | |
67 | if (!p) | |
68 | return -EIO; | |
69 | ||
70 | p++; | |
71 | ||
72 | if (sscanf(p, " %c", &state) != 1) | |
73 | return -EIO; | |
74 | ||
75 | return (unsigned char) state; | |
76 | } | |
77 | ||
78 | int get_process_comm(pid_t pid, char **name) { | |
79 | const char *p; | |
80 | int r; | |
81 | ||
82 | assert(name); | |
83 | assert(pid >= 0); | |
84 | ||
85 | p = procfs_file_alloca(pid, "comm"); | |
86 | ||
87 | r = read_one_line_file(p, name); | |
88 | if (r == -ENOENT) | |
89 | return -ESRCH; | |
90 | ||
91 | return r; | |
92 | } | |
93 | ||
94 | int get_process_cmdline(pid_t pid, size_t max_length, bool comm_fallback, char **line) { | |
95 | _cleanup_fclose_ FILE *f = NULL; | |
ba4cd7e2 | 96 | bool space = false; |
c0534780 | 97 | char *k, *ans = NULL; |
0b452006 RC |
98 | const char *p; |
99 | int c; | |
100 | ||
101 | assert(line); | |
102 | assert(pid >= 0); | |
103 | ||
69281c49 LP |
104 | /* Retrieves a process' command line. Replaces unprintable characters while doing so by whitespace (coalescing |
105 | * multiple sequential ones into one). If max_length is != 0 will return a string of the specified size at most | |
106 | * (the trailing NUL byte does count towards the length here!), abbreviated with a "..." ellipsis. If | |
107 | * comm_fallback is true and the process has no command line set (the case for kernel threads), or has a | |
108 | * command line that resolves to the empty string will return the "comm" name of the process instead. | |
109 | * | |
110 | * Returns -ESRCH if the process doesn't exist, and -ENOENT if the process has no command line (and | |
c0534780 | 111 | * comm_fallback is false). Returns 0 and sets *line otherwise. */ |
69281c49 | 112 | |
0b452006 RC |
113 | p = procfs_file_alloca(pid, "cmdline"); |
114 | ||
115 | f = fopen(p, "re"); | |
a644184a LP |
116 | if (!f) { |
117 | if (errno == ENOENT) | |
118 | return -ESRCH; | |
0b452006 | 119 | return -errno; |
a644184a | 120 | } |
0b452006 | 121 | |
35bbbf85 LP |
122 | (void) __fsetlocking(f, FSETLOCKING_BYCALLER); |
123 | ||
69281c49 LP |
124 | if (max_length == 1) { |
125 | ||
126 | /* If there's only room for one byte, return the empty string */ | |
c0534780 ZJS |
127 | ans = new0(char, 1); |
128 | if (!ans) | |
69281c49 LP |
129 | return -ENOMEM; |
130 | ||
c0534780 | 131 | *line = ans; |
69281c49 LP |
132 | return 0; |
133 | ||
134 | } else if (max_length == 0) { | |
0b452006 RC |
135 | size_t len = 0, allocated = 0; |
136 | ||
137 | while ((c = getc(f)) != EOF) { | |
138 | ||
c0534780 ZJS |
139 | if (!GREEDY_REALLOC(ans, allocated, len+3)) { |
140 | free(ans); | |
0b452006 RC |
141 | return -ENOMEM; |
142 | } | |
143 | ||
ba4cd7e2 MP |
144 | if (isprint(c)) { |
145 | if (space) { | |
c0534780 | 146 | ans[len++] = ' '; |
ba4cd7e2 MP |
147 | space = false; |
148 | } | |
149 | ||
c0534780 | 150 | ans[len++] = c; |
69281c49 | 151 | } else if (len > 0) |
ba4cd7e2 MP |
152 | space = true; |
153 | } | |
0b452006 RC |
154 | |
155 | if (len > 0) | |
c0534780 | 156 | ans[len] = '\0'; |
b09df4e2 | 157 | else |
c0534780 | 158 | ans = mfree(ans); |
0b452006 RC |
159 | |
160 | } else { | |
69281c49 | 161 | bool dotdotdot = false; |
0b452006 RC |
162 | size_t left; |
163 | ||
c0534780 ZJS |
164 | ans = new(char, max_length); |
165 | if (!ans) | |
0b452006 RC |
166 | return -ENOMEM; |
167 | ||
c0534780 | 168 | k = ans; |
0b452006 RC |
169 | left = max_length; |
170 | while ((c = getc(f)) != EOF) { | |
171 | ||
172 | if (isprint(c)) { | |
69281c49 | 173 | |
0b452006 | 174 | if (space) { |
69281c49 LP |
175 | if (left <= 2) { |
176 | dotdotdot = true; | |
0b452006 | 177 | break; |
69281c49 | 178 | } |
0b452006 RC |
179 | |
180 | *(k++) = ' '; | |
181 | left--; | |
182 | space = false; | |
183 | } | |
184 | ||
69281c49 LP |
185 | if (left <= 1) { |
186 | dotdotdot = true; | |
0b452006 | 187 | break; |
69281c49 | 188 | } |
0b452006 RC |
189 | |
190 | *(k++) = (char) c; | |
191 | left--; | |
c0534780 | 192 | } else if (k > ans) |
0b452006 RC |
193 | space = true; |
194 | } | |
195 | ||
69281c49 LP |
196 | if (dotdotdot) { |
197 | if (max_length <= 4) { | |
c0534780 | 198 | k = ans; |
69281c49 LP |
199 | left = max_length; |
200 | } else { | |
c0534780 | 201 | k = ans + max_length - 4; |
69281c49 LP |
202 | left = 4; |
203 | ||
204 | /* Eat up final spaces */ | |
c0534780 | 205 | while (k > ans && isspace(k[-1])) { |
69281c49 LP |
206 | k--; |
207 | left++; | |
208 | } | |
209 | } | |
210 | ||
211 | strncpy(k, "...", left-1); | |
b09df4e2 | 212 | k[left-1] = 0; |
0b452006 RC |
213 | } else |
214 | *k = 0; | |
215 | } | |
216 | ||
217 | /* Kernel threads have no argv[] */ | |
c0534780 | 218 | if (isempty(ans)) { |
0b452006 RC |
219 | _cleanup_free_ char *t = NULL; |
220 | int h; | |
221 | ||
c0534780 | 222 | free(ans); |
0b452006 RC |
223 | |
224 | if (!comm_fallback) | |
225 | return -ENOENT; | |
226 | ||
227 | h = get_process_comm(pid, &t); | |
228 | if (h < 0) | |
229 | return h; | |
230 | ||
69281c49 | 231 | if (max_length == 0) |
c0534780 | 232 | ans = strjoin("[", t, "]"); |
69281c49 LP |
233 | else { |
234 | size_t l; | |
235 | ||
236 | l = strlen(t); | |
237 | ||
238 | if (l + 3 <= max_length) | |
c0534780 | 239 | ans = strjoin("[", t, "]"); |
69281c49 LP |
240 | else if (max_length <= 6) { |
241 | ||
c0534780 ZJS |
242 | ans = new(char, max_length); |
243 | if (!ans) | |
69281c49 LP |
244 | return -ENOMEM; |
245 | ||
c0534780 ZJS |
246 | memcpy(ans, "[...]", max_length-1); |
247 | ans[max_length-1] = 0; | |
69281c49 LP |
248 | } else { |
249 | char *e; | |
250 | ||
251 | t[max_length - 6] = 0; | |
252 | ||
253 | /* Chop off final spaces */ | |
254 | e = strchr(t, 0); | |
255 | while (e > t && isspace(e[-1])) | |
256 | e--; | |
257 | *e = 0; | |
258 | ||
c0534780 | 259 | ans = strjoin("[", t, "...]"); |
69281c49 LP |
260 | } |
261 | } | |
c0534780 | 262 | if (!ans) |
0b452006 RC |
263 | return -ENOMEM; |
264 | } | |
265 | ||
c0534780 | 266 | *line = ans; |
0b452006 RC |
267 | return 0; |
268 | } | |
269 | ||
9bfaffd5 LP |
270 | int rename_process(const char name[]) { |
271 | static size_t mm_size = 0; | |
272 | static char *mm = NULL; | |
273 | bool truncated = false; | |
274 | size_t l; | |
275 | ||
276 | /* This is a like a poor man's setproctitle(). It changes the comm field, argv[0], and also the glibc's | |
277 | * internally used name of the process. For the first one a limit of 16 chars applies; to the second one in | |
278 | * many cases one of 10 (i.e. length of "/sbin/init") — however if we have CAP_SYS_RESOURCES it is unbounded; | |
279 | * to the third one 7 (i.e. the length of "systemd". If you pass a longer string it will likely be | |
280 | * truncated. | |
281 | * | |
282 | * Returns 0 if a name was set but truncated, > 0 if it was set but not truncated. */ | |
283 | ||
284 | if (isempty(name)) | |
285 | return -EINVAL; /* let's not confuse users unnecessarily with an empty name */ | |
405f8907 | 286 | |
1096bb8a LP |
287 | if (!is_main_thread()) |
288 | return -EPERM; /* Let's not allow setting the process name from other threads than the main one, as we | |
289 | * cache things without locking, and we make assumptions that PR_SET_NAME sets the | |
290 | * process name that isn't correct on any other threads */ | |
291 | ||
9bfaffd5 | 292 | l = strlen(name); |
405f8907 | 293 | |
5a8af747 LP |
294 | /* First step, change the comm field. The main thread's comm is identical to the process comm. This means we |
295 | * can use PR_SET_NAME, which sets the thread name for the calling thread. */ | |
296 | if (prctl(PR_SET_NAME, name) < 0) | |
297 | log_debug_errno(errno, "PR_SET_NAME failed: %m"); | |
9bfaffd5 LP |
298 | if (l > 15) /* Linux process names can be 15 chars at max */ |
299 | truncated = true; | |
300 | ||
301 | /* Second step, change glibc's ID of the process name. */ | |
302 | if (program_invocation_name) { | |
303 | size_t k; | |
304 | ||
305 | k = strlen(program_invocation_name); | |
306 | strncpy(program_invocation_name, name, k); | |
307 | if (l > k) | |
308 | truncated = true; | |
309 | } | |
310 | ||
311 | /* Third step, completely replace the argv[] array the kernel maintains for us. This requires privileges, but | |
312 | * has the advantage that the argv[] array is exactly what we want it to be, and not filled up with zeros at | |
13e785f7 | 313 | * the end. This is the best option for changing /proc/self/cmdline. */ |
01f989c6 JW |
314 | |
315 | /* Let's not bother with this if we don't have euid == 0. Strictly speaking we should check for the | |
316 | * CAP_SYS_RESOURCE capability which is independent of the euid. In our own code the capability generally is | |
317 | * present only for euid == 0, hence let's use this as quick bypass check, to avoid calling mmap() if | |
318 | * PR_SET_MM_ARG_{START,END} fails with EPERM later on anyway. After all geteuid() is dead cheap to call, but | |
319 | * mmap() is not. */ | |
320 | if (geteuid() != 0) | |
321 | log_debug("Skipping PR_SET_MM, as we don't have privileges."); | |
322 | else if (mm_size < l+1) { | |
9bfaffd5 LP |
323 | size_t nn_size; |
324 | char *nn; | |
325 | ||
9bfaffd5 LP |
326 | nn_size = PAGE_ALIGN(l+1); |
327 | nn = mmap(NULL, nn_size, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); | |
328 | if (nn == MAP_FAILED) { | |
329 | log_debug_errno(errno, "mmap() failed: %m"); | |
330 | goto use_saved_argv; | |
331 | } | |
405f8907 | 332 | |
9bfaffd5 LP |
333 | strncpy(nn, name, nn_size); |
334 | ||
335 | /* Now, let's tell the kernel about this new memory */ | |
336 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_START, (unsigned long) nn, 0, 0) < 0) { | |
337 | log_debug_errno(errno, "PR_SET_MM_ARG_START failed, proceeding without: %m"); | |
338 | (void) munmap(nn, nn_size); | |
339 | goto use_saved_argv; | |
340 | } | |
341 | ||
342 | /* And update the end pointer to the new end, too. If this fails, we don't really know what to do, it's | |
343 | * pretty unlikely that we can rollback, hence we'll just accept the failure, and continue. */ | |
344 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) nn + l + 1, 0, 0) < 0) | |
345 | log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m"); | |
346 | ||
347 | if (mm) | |
348 | (void) munmap(mm, mm_size); | |
349 | ||
350 | mm = nn; | |
351 | mm_size = nn_size; | |
01f989c6 | 352 | } else { |
9bfaffd5 LP |
353 | strncpy(mm, name, mm_size); |
354 | ||
01f989c6 JW |
355 | /* Update the end pointer, continuing regardless of any failure. */ |
356 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) mm + l + 1, 0, 0) < 0) | |
357 | log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m"); | |
358 | } | |
359 | ||
9bfaffd5 LP |
360 | use_saved_argv: |
361 | /* Fourth step: in all cases we'll also update the original argv[], so that our own code gets it right too if | |
362 | * it still looks here */ | |
405f8907 LP |
363 | |
364 | if (saved_argc > 0) { | |
365 | int i; | |
366 | ||
9bfaffd5 LP |
367 | if (saved_argv[0]) { |
368 | size_t k; | |
369 | ||
370 | k = strlen(saved_argv[0]); | |
371 | strncpy(saved_argv[0], name, k); | |
372 | if (l > k) | |
373 | truncated = true; | |
374 | } | |
405f8907 LP |
375 | |
376 | for (i = 1; i < saved_argc; i++) { | |
377 | if (!saved_argv[i]) | |
378 | break; | |
379 | ||
380 | memzero(saved_argv[i], strlen(saved_argv[i])); | |
381 | } | |
382 | } | |
9bfaffd5 LP |
383 | |
384 | return !truncated; | |
405f8907 LP |
385 | } |
386 | ||
0b452006 | 387 | int is_kernel_thread(pid_t pid) { |
36b5119a LP |
388 | _cleanup_free_ char *line = NULL; |
389 | unsigned long long flags; | |
390 | size_t l, i; | |
0b452006 | 391 | const char *p; |
36b5119a LP |
392 | char *q; |
393 | int r; | |
0b452006 | 394 | |
4c701096 | 395 | if (IN_SET(pid, 0, 1) || pid == getpid_cached()) /* pid 1, and we ourselves certainly aren't a kernel thread */ |
0b452006 | 396 | return 0; |
36b5119a LP |
397 | if (!pid_is_valid(pid)) |
398 | return -EINVAL; | |
0b452006 | 399 | |
36b5119a LP |
400 | p = procfs_file_alloca(pid, "stat"); |
401 | r = read_one_line_file(p, &line); | |
402 | if (r == -ENOENT) | |
403 | return -ESRCH; | |
404 | if (r < 0) | |
405 | return r; | |
0b452006 | 406 | |
36b5119a LP |
407 | /* Skip past the comm field */ |
408 | q = strrchr(line, ')'); | |
409 | if (!q) | |
410 | return -EINVAL; | |
411 | q++; | |
412 | ||
413 | /* Skip 6 fields to reach the flags field */ | |
414 | for (i = 0; i < 6; i++) { | |
415 | l = strspn(q, WHITESPACE); | |
416 | if (l < 1) | |
417 | return -EINVAL; | |
418 | q += l; | |
419 | ||
420 | l = strcspn(q, WHITESPACE); | |
421 | if (l < 1) | |
422 | return -EINVAL; | |
423 | q += l; | |
a644184a | 424 | } |
0b452006 | 425 | |
36b5119a LP |
426 | /* Skip preceeding whitespace */ |
427 | l = strspn(q, WHITESPACE); | |
428 | if (l < 1) | |
429 | return -EINVAL; | |
430 | q += l; | |
35bbbf85 | 431 | |
36b5119a LP |
432 | /* Truncate the rest */ |
433 | l = strcspn(q, WHITESPACE); | |
434 | if (l < 1) | |
435 | return -EINVAL; | |
436 | q[l] = 0; | |
0b452006 | 437 | |
36b5119a LP |
438 | r = safe_atollu(q, &flags); |
439 | if (r < 0) | |
440 | return r; | |
0b452006 | 441 | |
36b5119a | 442 | return !!(flags & PF_KTHREAD); |
0b452006 RC |
443 | } |
444 | ||
445 | int get_process_capeff(pid_t pid, char **capeff) { | |
446 | const char *p; | |
a644184a | 447 | int r; |
0b452006 RC |
448 | |
449 | assert(capeff); | |
450 | assert(pid >= 0); | |
451 | ||
452 | p = procfs_file_alloca(pid, "status"); | |
453 | ||
c4cd1d4d | 454 | r = get_proc_field(p, "CapEff", WHITESPACE, capeff); |
a644184a LP |
455 | if (r == -ENOENT) |
456 | return -ESRCH; | |
457 | ||
458 | return r; | |
0b452006 RC |
459 | } |
460 | ||
461 | static int get_process_link_contents(const char *proc_file, char **name) { | |
462 | int r; | |
463 | ||
464 | assert(proc_file); | |
465 | assert(name); | |
466 | ||
467 | r = readlink_malloc(proc_file, name); | |
a644184a LP |
468 | if (r == -ENOENT) |
469 | return -ESRCH; | |
0b452006 | 470 | if (r < 0) |
a644184a | 471 | return r; |
0b452006 RC |
472 | |
473 | return 0; | |
474 | } | |
475 | ||
476 | int get_process_exe(pid_t pid, char **name) { | |
477 | const char *p; | |
478 | char *d; | |
479 | int r; | |
480 | ||
481 | assert(pid >= 0); | |
482 | ||
483 | p = procfs_file_alloca(pid, "exe"); | |
484 | r = get_process_link_contents(p, name); | |
485 | if (r < 0) | |
486 | return r; | |
487 | ||
488 | d = endswith(*name, " (deleted)"); | |
489 | if (d) | |
490 | *d = '\0'; | |
491 | ||
492 | return 0; | |
493 | } | |
494 | ||
495 | static int get_process_id(pid_t pid, const char *field, uid_t *uid) { | |
496 | _cleanup_fclose_ FILE *f = NULL; | |
497 | char line[LINE_MAX]; | |
498 | const char *p; | |
499 | ||
500 | assert(field); | |
501 | assert(uid); | |
502 | ||
07b38ba5 | 503 | if (pid < 0) |
6f8cbcdb LP |
504 | return -EINVAL; |
505 | ||
0b452006 RC |
506 | p = procfs_file_alloca(pid, "status"); |
507 | f = fopen(p, "re"); | |
a644184a LP |
508 | if (!f) { |
509 | if (errno == ENOENT) | |
510 | return -ESRCH; | |
0b452006 | 511 | return -errno; |
a644184a | 512 | } |
0b452006 | 513 | |
35bbbf85 LP |
514 | (void) __fsetlocking(f, FSETLOCKING_BYCALLER); |
515 | ||
0b452006 RC |
516 | FOREACH_LINE(line, f, return -errno) { |
517 | char *l; | |
518 | ||
519 | l = strstrip(line); | |
520 | ||
521 | if (startswith(l, field)) { | |
522 | l += strlen(field); | |
523 | l += strspn(l, WHITESPACE); | |
524 | ||
525 | l[strcspn(l, WHITESPACE)] = 0; | |
526 | ||
527 | return parse_uid(l, uid); | |
528 | } | |
529 | } | |
530 | ||
531 | return -EIO; | |
532 | } | |
533 | ||
534 | int get_process_uid(pid_t pid, uid_t *uid) { | |
6f8cbcdb LP |
535 | |
536 | if (pid == 0 || pid == getpid_cached()) { | |
537 | *uid = getuid(); | |
538 | return 0; | |
539 | } | |
540 | ||
0b452006 RC |
541 | return get_process_id(pid, "Uid:", uid); |
542 | } | |
543 | ||
544 | int get_process_gid(pid_t pid, gid_t *gid) { | |
6f8cbcdb LP |
545 | |
546 | if (pid == 0 || pid == getpid_cached()) { | |
547 | *gid = getgid(); | |
548 | return 0; | |
549 | } | |
550 | ||
0b452006 RC |
551 | assert_cc(sizeof(uid_t) == sizeof(gid_t)); |
552 | return get_process_id(pid, "Gid:", gid); | |
553 | } | |
554 | ||
555 | int get_process_cwd(pid_t pid, char **cwd) { | |
556 | const char *p; | |
557 | ||
558 | assert(pid >= 0); | |
559 | ||
560 | p = procfs_file_alloca(pid, "cwd"); | |
561 | ||
562 | return get_process_link_contents(p, cwd); | |
563 | } | |
564 | ||
565 | int get_process_root(pid_t pid, char **root) { | |
566 | const char *p; | |
567 | ||
568 | assert(pid >= 0); | |
569 | ||
570 | p = procfs_file_alloca(pid, "root"); | |
571 | ||
572 | return get_process_link_contents(p, root); | |
573 | } | |
574 | ||
575 | int get_process_environ(pid_t pid, char **env) { | |
576 | _cleanup_fclose_ FILE *f = NULL; | |
577 | _cleanup_free_ char *outcome = NULL; | |
578 | int c; | |
579 | const char *p; | |
580 | size_t allocated = 0, sz = 0; | |
581 | ||
582 | assert(pid >= 0); | |
583 | assert(env); | |
584 | ||
585 | p = procfs_file_alloca(pid, "environ"); | |
586 | ||
587 | f = fopen(p, "re"); | |
a644184a LP |
588 | if (!f) { |
589 | if (errno == ENOENT) | |
590 | return -ESRCH; | |
0b452006 | 591 | return -errno; |
a644184a | 592 | } |
0b452006 | 593 | |
35bbbf85 LP |
594 | (void) __fsetlocking(f, FSETLOCKING_BYCALLER); |
595 | ||
0b452006 RC |
596 | while ((c = fgetc(f)) != EOF) { |
597 | if (!GREEDY_REALLOC(outcome, allocated, sz + 5)) | |
598 | return -ENOMEM; | |
599 | ||
600 | if (c == '\0') | |
601 | outcome[sz++] = '\n'; | |
602 | else | |
603 | sz += cescape_char(c, outcome + sz); | |
604 | } | |
605 | ||
03c55bc0 LP |
606 | if (!outcome) { |
607 | outcome = strdup(""); | |
608 | if (!outcome) | |
609 | return -ENOMEM; | |
610 | } else | |
611 | outcome[sz] = '\0'; | |
de8763b6 | 612 | |
ae2a15bc | 613 | *env = TAKE_PTR(outcome); |
0b452006 RC |
614 | |
615 | return 0; | |
616 | } | |
617 | ||
6bc73acb | 618 | int get_process_ppid(pid_t pid, pid_t *_ppid) { |
0b452006 RC |
619 | int r; |
620 | _cleanup_free_ char *line = NULL; | |
621 | long unsigned ppid; | |
622 | const char *p; | |
623 | ||
624 | assert(pid >= 0); | |
625 | assert(_ppid); | |
626 | ||
6f8cbcdb | 627 | if (pid == 0 || pid == getpid_cached()) { |
0b452006 RC |
628 | *_ppid = getppid(); |
629 | return 0; | |
630 | } | |
631 | ||
632 | p = procfs_file_alloca(pid, "stat"); | |
633 | r = read_one_line_file(p, &line); | |
a644184a LP |
634 | if (r == -ENOENT) |
635 | return -ESRCH; | |
0b452006 RC |
636 | if (r < 0) |
637 | return r; | |
638 | ||
639 | /* Let's skip the pid and comm fields. The latter is enclosed | |
640 | * in () but does not escape any () in its value, so let's | |
641 | * skip over it manually */ | |
642 | ||
643 | p = strrchr(line, ')'); | |
644 | if (!p) | |
645 | return -EIO; | |
646 | ||
647 | p++; | |
648 | ||
649 | if (sscanf(p, " " | |
650 | "%*c " /* state */ | |
651 | "%lu ", /* ppid */ | |
652 | &ppid) != 1) | |
653 | return -EIO; | |
654 | ||
655 | if ((long unsigned) (pid_t) ppid != ppid) | |
656 | return -ERANGE; | |
657 | ||
658 | *_ppid = (pid_t) ppid; | |
659 | ||
660 | return 0; | |
661 | } | |
662 | ||
663 | int wait_for_terminate(pid_t pid, siginfo_t *status) { | |
664 | siginfo_t dummy; | |
665 | ||
666 | assert(pid >= 1); | |
667 | ||
668 | if (!status) | |
669 | status = &dummy; | |
670 | ||
671 | for (;;) { | |
672 | zero(*status); | |
673 | ||
674 | if (waitid(P_PID, pid, status, WEXITED) < 0) { | |
675 | ||
676 | if (errno == EINTR) | |
677 | continue; | |
678 | ||
3f0083a2 | 679 | return negative_errno(); |
0b452006 RC |
680 | } |
681 | ||
682 | return 0; | |
683 | } | |
684 | } | |
685 | ||
686 | /* | |
687 | * Return values: | |
688 | * < 0 : wait_for_terminate() failed to get the state of the | |
689 | * process, the process was terminated by a signal, or | |
690 | * failed for an unknown reason. | |
691 | * >=0 : The process terminated normally, and its exit code is | |
692 | * returned. | |
693 | * | |
694 | * That is, success is indicated by a return value of zero, and an | |
695 | * error is indicated by a non-zero value. | |
696 | * | |
697 | * A warning is emitted if the process terminates abnormally, | |
698 | * and also if it returns non-zero unless check_exit_code is true. | |
699 | */ | |
7d4904fe LP |
700 | int wait_for_terminate_and_check(const char *name, pid_t pid, WaitFlags flags) { |
701 | _cleanup_free_ char *buffer = NULL; | |
0b452006 | 702 | siginfo_t status; |
7d4904fe | 703 | int r, prio; |
0b452006 | 704 | |
0b452006 RC |
705 | assert(pid > 1); |
706 | ||
7d4904fe LP |
707 | if (!name) { |
708 | r = get_process_comm(pid, &buffer); | |
709 | if (r < 0) | |
710 | log_debug_errno(r, "Failed to acquire process name of " PID_FMT ", ignoring: %m", pid); | |
711 | else | |
712 | name = buffer; | |
713 | } | |
714 | ||
715 | prio = flags & WAIT_LOG_ABNORMAL ? LOG_ERR : LOG_DEBUG; | |
716 | ||
0b452006 RC |
717 | r = wait_for_terminate(pid, &status); |
718 | if (r < 0) | |
7d4904fe | 719 | return log_full_errno(prio, r, "Failed to wait for %s: %m", strna(name)); |
0b452006 RC |
720 | |
721 | if (status.si_code == CLD_EXITED) { | |
7d4904fe LP |
722 | if (status.si_status != EXIT_SUCCESS) |
723 | log_full(flags & WAIT_LOG_NON_ZERO_EXIT_STATUS ? LOG_ERR : LOG_DEBUG, | |
724 | "%s failed with exit status %i.", strna(name), status.si_status); | |
0b452006 RC |
725 | else |
726 | log_debug("%s succeeded.", name); | |
727 | ||
728 | return status.si_status; | |
7d4904fe | 729 | |
3742095b | 730 | } else if (IN_SET(status.si_code, CLD_KILLED, CLD_DUMPED)) { |
0b452006 | 731 | |
7d4904fe | 732 | log_full(prio, "%s terminated by signal %s.", strna(name), signal_to_string(status.si_status)); |
0b452006 RC |
733 | return -EPROTO; |
734 | } | |
735 | ||
7d4904fe | 736 | log_full(prio, "%s failed due to unknown reason.", strna(name)); |
0b452006 RC |
737 | return -EPROTO; |
738 | } | |
739 | ||
d5641e0d KW |
740 | /* |
741 | * Return values: | |
742 | * < 0 : wait_for_terminate_with_timeout() failed to get the state of the | |
743 | * process, the process timed out, the process was terminated by a | |
744 | * signal, or failed for an unknown reason. | |
745 | * >=0 : The process terminated normally with no failures. | |
746 | * | |
747 | * Success is indicated by a return value of zero, a timeout is indicated | |
748 | * by ETIMEDOUT, and all other child failure states are indicated by error | |
749 | * is indicated by a non-zero value. | |
750 | */ | |
751 | int wait_for_terminate_with_timeout(pid_t pid, usec_t timeout) { | |
752 | sigset_t mask; | |
753 | int r; | |
754 | usec_t until; | |
755 | ||
756 | assert_se(sigemptyset(&mask) == 0); | |
757 | assert_se(sigaddset(&mask, SIGCHLD) == 0); | |
758 | ||
759 | /* Drop into a sigtimewait-based timeout. Waiting for the | |
760 | * pid to exit. */ | |
761 | until = now(CLOCK_MONOTONIC) + timeout; | |
762 | for (;;) { | |
763 | usec_t n; | |
764 | siginfo_t status = {}; | |
765 | struct timespec ts; | |
766 | ||
767 | n = now(CLOCK_MONOTONIC); | |
768 | if (n >= until) | |
769 | break; | |
770 | ||
771 | r = sigtimedwait(&mask, NULL, timespec_store(&ts, until - n)) < 0 ? -errno : 0; | |
772 | /* Assuming we woke due to the child exiting. */ | |
773 | if (waitid(P_PID, pid, &status, WEXITED|WNOHANG) == 0) { | |
774 | if (status.si_pid == pid) { | |
775 | /* This is the correct child.*/ | |
776 | if (status.si_code == CLD_EXITED) | |
777 | return (status.si_status == 0) ? 0 : -EPROTO; | |
778 | else | |
779 | return -EPROTO; | |
780 | } | |
781 | } | |
782 | /* Not the child, check for errors and proceed appropriately */ | |
783 | if (r < 0) { | |
784 | switch (r) { | |
785 | case -EAGAIN: | |
786 | /* Timed out, child is likely hung. */ | |
787 | return -ETIMEDOUT; | |
788 | case -EINTR: | |
789 | /* Received a different signal and should retry */ | |
790 | continue; | |
791 | default: | |
792 | /* Return any unexpected errors */ | |
793 | return r; | |
794 | } | |
795 | } | |
796 | } | |
797 | ||
798 | return -EPROTO; | |
799 | } | |
800 | ||
89c9030d LP |
801 | void sigkill_wait(pid_t pid) { |
802 | assert(pid > 1); | |
803 | ||
804 | if (kill(pid, SIGKILL) > 0) | |
805 | (void) wait_for_terminate(pid, NULL); | |
806 | } | |
807 | ||
808 | void sigkill_waitp(pid_t *pid) { | |
dfd14786 LP |
809 | PROTECT_ERRNO; |
810 | ||
4d0d3d41 LP |
811 | if (!pid) |
812 | return; | |
813 | if (*pid <= 1) | |
814 | return; | |
815 | ||
89c9030d | 816 | sigkill_wait(*pid); |
4d0d3d41 LP |
817 | } |
818 | ||
392cf1d0 SL |
819 | void sigterm_wait(pid_t pid) { |
820 | assert(pid > 1); | |
821 | ||
822 | if (kill_and_sigcont(pid, SIGTERM) > 0) | |
823 | (void) wait_for_terminate(pid, NULL); | |
824 | } | |
825 | ||
0b452006 RC |
826 | int kill_and_sigcont(pid_t pid, int sig) { |
827 | int r; | |
828 | ||
829 | r = kill(pid, sig) < 0 ? -errno : 0; | |
830 | ||
26f417d3 LP |
831 | /* If this worked, also send SIGCONT, unless we already just sent a SIGCONT, or SIGKILL was sent which isn't |
832 | * affected by a process being suspended anyway. */ | |
a3d8d68c | 833 | if (r >= 0 && !IN_SET(sig, SIGCONT, SIGKILL)) |
26f417d3 | 834 | (void) kill(pid, SIGCONT); |
0b452006 RC |
835 | |
836 | return r; | |
837 | } | |
838 | ||
e70f4453 | 839 | int getenv_for_pid(pid_t pid, const char *field, char **ret) { |
0b452006 RC |
840 | _cleanup_fclose_ FILE *f = NULL; |
841 | char *value = NULL; | |
0b452006 | 842 | bool done = false; |
0b452006 | 843 | const char *path; |
e70f4453 | 844 | size_t l; |
0b452006 RC |
845 | |
846 | assert(pid >= 0); | |
847 | assert(field); | |
e70f4453 LP |
848 | assert(ret); |
849 | ||
850 | if (pid == 0 || pid == getpid_cached()) { | |
851 | const char *e; | |
852 | ||
853 | e = getenv(field); | |
854 | if (!e) { | |
855 | *ret = NULL; | |
856 | return 0; | |
857 | } | |
858 | ||
859 | value = strdup(e); | |
860 | if (!value) | |
861 | return -ENOMEM; | |
862 | ||
863 | *ret = value; | |
864 | return 1; | |
865 | } | |
0b452006 RC |
866 | |
867 | path = procfs_file_alloca(pid, "environ"); | |
868 | ||
869 | f = fopen(path, "re"); | |
a644184a LP |
870 | if (!f) { |
871 | if (errno == ENOENT) | |
872 | return -ESRCH; | |
e70f4453 | 873 | |
0b452006 | 874 | return -errno; |
a644184a | 875 | } |
0b452006 | 876 | |
35bbbf85 LP |
877 | (void) __fsetlocking(f, FSETLOCKING_BYCALLER); |
878 | ||
0b452006 | 879 | l = strlen(field); |
0b452006 RC |
880 | |
881 | do { | |
882 | char line[LINE_MAX]; | |
da6053d0 | 883 | size_t i; |
0b452006 RC |
884 | |
885 | for (i = 0; i < sizeof(line)-1; i++) { | |
886 | int c; | |
887 | ||
888 | c = getc(f); | |
889 | if (_unlikely_(c == EOF)) { | |
890 | done = true; | |
891 | break; | |
892 | } else if (c == 0) | |
893 | break; | |
894 | ||
895 | line[i] = c; | |
896 | } | |
897 | line[i] = 0; | |
898 | ||
041b5ae1 | 899 | if (strneq(line, field, l) && line[l] == '=') { |
0b452006 RC |
900 | value = strdup(line + l + 1); |
901 | if (!value) | |
902 | return -ENOMEM; | |
903 | ||
e70f4453 LP |
904 | *ret = value; |
905 | return 1; | |
0b452006 RC |
906 | } |
907 | ||
908 | } while (!done); | |
909 | ||
e70f4453 LP |
910 | *ret = NULL; |
911 | return 0; | |
0b452006 RC |
912 | } |
913 | ||
914 | bool pid_is_unwaited(pid_t pid) { | |
915 | /* Checks whether a PID is still valid at all, including a zombie */ | |
916 | ||
07b38ba5 | 917 | if (pid < 0) |
0b452006 RC |
918 | return false; |
919 | ||
5fd9b2c5 LP |
920 | if (pid <= 1) /* If we or PID 1 would be dead and have been waited for, this code would not be running */ |
921 | return true; | |
922 | ||
6f8cbcdb LP |
923 | if (pid == getpid_cached()) |
924 | return true; | |
925 | ||
0b452006 RC |
926 | if (kill(pid, 0) >= 0) |
927 | return true; | |
928 | ||
929 | return errno != ESRCH; | |
930 | } | |
931 | ||
932 | bool pid_is_alive(pid_t pid) { | |
933 | int r; | |
934 | ||
935 | /* Checks whether a PID is still valid and not a zombie */ | |
936 | ||
07b38ba5 | 937 | if (pid < 0) |
0b452006 RC |
938 | return false; |
939 | ||
5fd9b2c5 LP |
940 | if (pid <= 1) /* If we or PID 1 would be a zombie, this code would not be running */ |
941 | return true; | |
942 | ||
6f8cbcdb LP |
943 | if (pid == getpid_cached()) |
944 | return true; | |
945 | ||
0b452006 | 946 | r = get_process_state(pid); |
4c701096 | 947 | if (IN_SET(r, -ESRCH, 'Z')) |
0b452006 RC |
948 | return false; |
949 | ||
950 | return true; | |
951 | } | |
d4510856 | 952 | |
1359fffa MS |
953 | int pid_from_same_root_fs(pid_t pid) { |
954 | const char *root; | |
955 | ||
07b38ba5 | 956 | if (pid < 0) |
6f8cbcdb LP |
957 | return false; |
958 | ||
959 | if (pid == 0 || pid == getpid_cached()) | |
960 | return true; | |
1359fffa MS |
961 | |
962 | root = procfs_file_alloca(pid, "root"); | |
963 | ||
e3f791a2 | 964 | return files_same(root, "/proc/1/root", 0); |
1359fffa MS |
965 | } |
966 | ||
d4510856 LP |
967 | bool is_main_thread(void) { |
968 | static thread_local int cached = 0; | |
969 | ||
970 | if (_unlikely_(cached == 0)) | |
df0ff127 | 971 | cached = getpid_cached() == gettid() ? 1 : -1; |
d4510856 LP |
972 | |
973 | return cached > 0; | |
974 | } | |
7b3e062c | 975 | |
848e863a | 976 | _noreturn_ void freeze(void) { |
7b3e062c | 977 | |
3da48d7a EV |
978 | log_close(); |
979 | ||
7b3e062c LP |
980 | /* Make sure nobody waits for us on a socket anymore */ |
981 | close_all_fds(NULL, 0); | |
982 | ||
983 | sync(); | |
984 | ||
8647283e MS |
985 | /* Let's not freeze right away, but keep reaping zombies. */ |
986 | for (;;) { | |
987 | int r; | |
988 | siginfo_t si = {}; | |
989 | ||
990 | r = waitid(P_ALL, 0, &si, WEXITED); | |
991 | if (r < 0 && errno != EINTR) | |
992 | break; | |
993 | } | |
994 | ||
995 | /* waitid() failed with an unexpected error, things are really borked. Freeze now! */ | |
7b3e062c LP |
996 | for (;;) |
997 | pause(); | |
998 | } | |
999 | ||
1000 | bool oom_score_adjust_is_valid(int oa) { | |
1001 | return oa >= OOM_SCORE_ADJ_MIN && oa <= OOM_SCORE_ADJ_MAX; | |
1002 | } | |
1003 | ||
1004 | unsigned long personality_from_string(const char *p) { | |
6e5f1b57 | 1005 | int architecture; |
7b3e062c | 1006 | |
0c0fea07 LP |
1007 | if (!p) |
1008 | return PERSONALITY_INVALID; | |
1009 | ||
6e5f1b57 LP |
1010 | /* Parse a personality specifier. We use our own identifiers that indicate specific ABIs, rather than just |
1011 | * hints regarding the register size, since we want to keep things open for multiple locally supported ABIs for | |
1012 | * the same register size. */ | |
1013 | ||
1014 | architecture = architecture_from_string(p); | |
1015 | if (architecture < 0) | |
1016 | return PERSONALITY_INVALID; | |
7b3e062c | 1017 | |
0c0fea07 | 1018 | if (architecture == native_architecture()) |
7b3e062c | 1019 | return PER_LINUX; |
0c0fea07 LP |
1020 | #ifdef SECONDARY_ARCHITECTURE |
1021 | if (architecture == SECONDARY_ARCHITECTURE) | |
f2d1736c | 1022 | return PER_LINUX32; |
7b3e062c LP |
1023 | #endif |
1024 | ||
1025 | return PERSONALITY_INVALID; | |
1026 | } | |
1027 | ||
1028 | const char* personality_to_string(unsigned long p) { | |
6e5f1b57 | 1029 | int architecture = _ARCHITECTURE_INVALID; |
7b3e062c | 1030 | |
7b3e062c | 1031 | if (p == PER_LINUX) |
0c0fea07 LP |
1032 | architecture = native_architecture(); |
1033 | #ifdef SECONDARY_ARCHITECTURE | |
6e5f1b57 | 1034 | else if (p == PER_LINUX32) |
0c0fea07 | 1035 | architecture = SECONDARY_ARCHITECTURE; |
7b3e062c LP |
1036 | #endif |
1037 | ||
6e5f1b57 LP |
1038 | if (architecture < 0) |
1039 | return NULL; | |
1040 | ||
1041 | return architecture_to_string(architecture); | |
7b3e062c LP |
1042 | } |
1043 | ||
21022b9d LP |
1044 | int safe_personality(unsigned long p) { |
1045 | int ret; | |
1046 | ||
1047 | /* So here's the deal, personality() is weirdly defined by glibc. In some cases it returns a failure via errno, | |
1048 | * and in others as negative return value containing an errno-like value. Let's work around this: this is a | |
1049 | * wrapper that uses errno if it is set, and uses the return value otherwise. And then it sets both errno and | |
1050 | * the return value indicating the same issue, so that we are definitely on the safe side. | |
1051 | * | |
1052 | * See https://github.com/systemd/systemd/issues/6737 */ | |
1053 | ||
1054 | errno = 0; | |
1055 | ret = personality(p); | |
1056 | if (ret < 0) { | |
1057 | if (errno != 0) | |
1058 | return -errno; | |
1059 | ||
1060 | errno = -ret; | |
1061 | } | |
1062 | ||
1063 | return ret; | |
1064 | } | |
1065 | ||
e8132d63 LP |
1066 | int opinionated_personality(unsigned long *ret) { |
1067 | int current; | |
1068 | ||
1069 | /* Returns the current personality, or PERSONALITY_INVALID if we can't determine it. This function is a bit | |
1070 | * opinionated though, and ignores all the finer-grained bits and exotic personalities, only distinguishing the | |
1071 | * two most relevant personalities: PER_LINUX and PER_LINUX32. */ | |
1072 | ||
21022b9d | 1073 | current = safe_personality(PERSONALITY_INVALID); |
e8132d63 | 1074 | if (current < 0) |
21022b9d | 1075 | return current; |
e8132d63 LP |
1076 | |
1077 | if (((unsigned long) current & 0xffff) == PER_LINUX32) | |
1078 | *ret = PER_LINUX32; | |
1079 | else | |
1080 | *ret = PER_LINUX; | |
1081 | ||
1082 | return 0; | |
1083 | } | |
1084 | ||
dcadc967 | 1085 | void valgrind_summary_hack(void) { |
349cc4a5 | 1086 | #if HAVE_VALGRIND_VALGRIND_H |
df0ff127 | 1087 | if (getpid_cached() == 1 && RUNNING_ON_VALGRIND) { |
dcadc967 | 1088 | pid_t pid; |
8869a0b4 | 1089 | pid = raw_clone(SIGCHLD); |
dcadc967 EV |
1090 | if (pid < 0) |
1091 | log_emergency_errno(errno, "Failed to fork off valgrind helper: %m"); | |
1092 | else if (pid == 0) | |
1093 | exit(EXIT_SUCCESS); | |
1094 | else { | |
1095 | log_info("Spawned valgrind helper as PID "PID_FMT".", pid); | |
1096 | (void) wait_for_terminate(pid, NULL); | |
1097 | } | |
1098 | } | |
1099 | #endif | |
1100 | } | |
1101 | ||
291d565a LP |
1102 | int pid_compare_func(const void *a, const void *b) { |
1103 | const pid_t *p = a, *q = b; | |
1104 | ||
1105 | /* Suitable for usage in qsort() */ | |
1106 | ||
1107 | if (*p < *q) | |
1108 | return -1; | |
1109 | if (*p > *q) | |
1110 | return 1; | |
1111 | return 0; | |
1112 | } | |
1113 | ||
7f452159 LP |
1114 | int ioprio_parse_priority(const char *s, int *ret) { |
1115 | int i, r; | |
1116 | ||
1117 | assert(s); | |
1118 | assert(ret); | |
1119 | ||
1120 | r = safe_atoi(s, &i); | |
1121 | if (r < 0) | |
1122 | return r; | |
1123 | ||
1124 | if (!ioprio_priority_is_valid(i)) | |
1125 | return -EINVAL; | |
1126 | ||
1127 | *ret = i; | |
1128 | return 0; | |
1129 | } | |
1130 | ||
5c30a6d2 LP |
1131 | /* The cached PID, possible values: |
1132 | * | |
1133 | * == UNSET [0] → cache not initialized yet | |
1134 | * == BUSY [-1] → some thread is initializing it at the moment | |
1135 | * any other → the cached PID | |
1136 | */ | |
1137 | ||
1138 | #define CACHED_PID_UNSET ((pid_t) 0) | |
1139 | #define CACHED_PID_BUSY ((pid_t) -1) | |
1140 | ||
1141 | static pid_t cached_pid = CACHED_PID_UNSET; | |
1142 | ||
799a960d | 1143 | void reset_cached_pid(void) { |
5c30a6d2 LP |
1144 | /* Invoked in the child after a fork(), i.e. at the first moment the PID changed */ |
1145 | cached_pid = CACHED_PID_UNSET; | |
1146 | } | |
1147 | ||
1148 | /* We use glibc __register_atfork() + __dso_handle directly here, as they are not included in the glibc | |
1149 | * headers. __register_atfork() is mostly equivalent to pthread_atfork(), but doesn't require us to link against | |
1150 | * libpthread, as it is part of glibc anyway. */ | |
1151 | extern int __register_atfork(void (*prepare) (void), void (*parent) (void), void (*child) (void), void * __dso_handle); | |
1152 | extern void* __dso_handle __attribute__ ((__weak__)); | |
1153 | ||
1154 | pid_t getpid_cached(void) { | |
5d71bac3 | 1155 | static bool installed = false; |
5c30a6d2 LP |
1156 | pid_t current_value; |
1157 | ||
1158 | /* getpid_cached() is much like getpid(), but caches the value in local memory, to avoid having to invoke a | |
1159 | * system call each time. This restores glibc behaviour from before 2.24, when getpid() was unconditionally | |
1160 | * cached. Starting with 2.24 getpid() started to become prohibitively expensive when used for detecting when | |
1161 | * objects were used across fork()s. With this caching the old behaviour is somewhat restored. | |
1162 | * | |
1163 | * https://bugzilla.redhat.com/show_bug.cgi?id=1443976 | |
a4041e4f | 1164 | * https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=c579f48edba88380635ab98cb612030e3ed8691e |
5c30a6d2 LP |
1165 | */ |
1166 | ||
1167 | current_value = __sync_val_compare_and_swap(&cached_pid, CACHED_PID_UNSET, CACHED_PID_BUSY); | |
1168 | ||
1169 | switch (current_value) { | |
1170 | ||
1171 | case CACHED_PID_UNSET: { /* Not initialized yet, then do so now */ | |
1172 | pid_t new_pid; | |
1173 | ||
996def17 | 1174 | new_pid = raw_getpid(); |
5c30a6d2 | 1175 | |
5d71bac3 LP |
1176 | if (!installed) { |
1177 | /* __register_atfork() either returns 0 or -ENOMEM, in its glibc implementation. Since it's | |
1178 | * only half-documented (glibc doesn't document it but LSB does — though only superficially) | |
1179 | * we'll check for errors only in the most generic fashion possible. */ | |
1180 | ||
1181 | if (__register_atfork(NULL, NULL, reset_cached_pid, __dso_handle) != 0) { | |
1182 | /* OOM? Let's try again later */ | |
1183 | cached_pid = CACHED_PID_UNSET; | |
1184 | return new_pid; | |
1185 | } | |
1186 | ||
1187 | installed = true; | |
5c30a6d2 LP |
1188 | } |
1189 | ||
1190 | cached_pid = new_pid; | |
1191 | return new_pid; | |
1192 | } | |
1193 | ||
1194 | case CACHED_PID_BUSY: /* Somebody else is currently initializing */ | |
996def17 | 1195 | return raw_getpid(); |
5c30a6d2 LP |
1196 | |
1197 | default: /* Properly initialized */ | |
1198 | return current_value; | |
1199 | } | |
1200 | } | |
1201 | ||
fba868fa LP |
1202 | int must_be_root(void) { |
1203 | ||
1204 | if (geteuid() == 0) | |
1205 | return 0; | |
1206 | ||
1207 | log_error("Need to be root."); | |
1208 | return -EPERM; | |
1209 | } | |
1210 | ||
4c253ed1 LP |
1211 | int safe_fork_full( |
1212 | const char *name, | |
1213 | const int except_fds[], | |
1214 | size_t n_except_fds, | |
1215 | ForkFlags flags, | |
1216 | pid_t *ret_pid) { | |
1217 | ||
1218 | pid_t original_pid, pid; | |
1f5d1e02 | 1219 | sigset_t saved_ss, ss; |
a41f6217 | 1220 | bool block_signals = false; |
b6e1fff1 | 1221 | int prio, r; |
4c253ed1 LP |
1222 | |
1223 | /* A wrapper around fork(), that does a couple of important initializations in addition to mere forking. Always | |
1224 | * returns the child's PID in *ret_pid. Returns == 0 in the child, and > 0 in the parent. */ | |
1225 | ||
b6e1fff1 LP |
1226 | prio = flags & FORK_LOG ? LOG_ERR : LOG_DEBUG; |
1227 | ||
4c253ed1 LP |
1228 | original_pid = getpid_cached(); |
1229 | ||
1f5d1e02 | 1230 | if (flags & (FORK_RESET_SIGNALS|FORK_DEATHSIG)) { |
4c253ed1 | 1231 | |
1f5d1e02 LP |
1232 | /* We temporarily block all signals, so that the new child has them blocked initially. This way, we can |
1233 | * be sure that SIGTERMs are not lost we might send to the child. */ | |
4c253ed1 | 1234 | |
4c253ed1 | 1235 | if (sigfillset(&ss) < 0) |
b6e1fff1 | 1236 | return log_full_errno(prio, errno, "Failed to reset signal set: %m"); |
4c253ed1 | 1237 | |
1f5d1e02 LP |
1238 | block_signals = true; |
1239 | ||
1240 | } else if (flags & FORK_WAIT) { | |
1241 | ||
1242 | /* Let's block SIGCHLD at least, so that we can safely watch for the child process */ | |
1243 | ||
1244 | if (sigemptyset(&ss) < 0) | |
1245 | return log_full_errno(prio, errno, "Failed to clear signal set: %m"); | |
1246 | ||
1247 | if (sigaddset(&ss, SIGCHLD) < 0) | |
1248 | return log_full_errno(prio, errno, "Failed to add SIGCHLD to signal set: %m"); | |
1249 | ||
1250 | block_signals = true; | |
4c253ed1 LP |
1251 | } |
1252 | ||
1f5d1e02 LP |
1253 | if (block_signals) |
1254 | if (sigprocmask(SIG_SETMASK, &ss, &saved_ss) < 0) | |
1255 | return log_full_errno(prio, errno, "Failed to set signal mask: %m"); | |
1256 | ||
be39f6ee LP |
1257 | if (flags & FORK_NEW_MOUNTNS) |
1258 | pid = raw_clone(SIGCHLD|CLONE_NEWNS); | |
1259 | else | |
1260 | pid = fork(); | |
4c253ed1 LP |
1261 | if (pid < 0) { |
1262 | r = -errno; | |
1263 | ||
1264 | if (block_signals) /* undo what we did above */ | |
1265 | (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL); | |
1266 | ||
b6e1fff1 | 1267 | return log_full_errno(prio, r, "Failed to fork: %m"); |
4c253ed1 LP |
1268 | } |
1269 | if (pid > 0) { | |
1270 | /* We are in the parent process */ | |
1271 | ||
1f5d1e02 LP |
1272 | log_debug("Successfully forked off '%s' as PID " PID_FMT ".", strna(name), pid); |
1273 | ||
1274 | if (flags & FORK_WAIT) { | |
1275 | r = wait_for_terminate_and_check(name, pid, (flags & FORK_LOG ? WAIT_LOG : 0)); | |
1276 | if (r < 0) | |
1277 | return r; | |
1278 | if (r != EXIT_SUCCESS) /* exit status > 0 should be treated as failure, too */ | |
1279 | return -EPROTO; | |
1280 | } | |
1281 | ||
4c253ed1 LP |
1282 | if (block_signals) /* undo what we did above */ |
1283 | (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL); | |
1284 | ||
4c253ed1 LP |
1285 | if (ret_pid) |
1286 | *ret_pid = pid; | |
1287 | ||
1288 | return 1; | |
1289 | } | |
1290 | ||
1291 | /* We are in the child process */ | |
1292 | ||
1293 | if (flags & FORK_REOPEN_LOG) { | |
1294 | /* Close the logs if requested, before we log anything. And make sure we reopen it if needed. */ | |
1295 | log_close(); | |
1296 | log_set_open_when_needed(true); | |
1297 | } | |
1298 | ||
1299 | if (name) { | |
1300 | r = rename_process(name); | |
1301 | if (r < 0) | |
b6e1fff1 LP |
1302 | log_full_errno(flags & FORK_LOG ? LOG_WARNING : LOG_DEBUG, |
1303 | r, "Failed to rename process, ignoring: %m"); | |
4c253ed1 LP |
1304 | } |
1305 | ||
1306 | if (flags & FORK_DEATHSIG) | |
1307 | if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0) { | |
b6e1fff1 | 1308 | log_full_errno(prio, errno, "Failed to set death signal: %m"); |
4c253ed1 LP |
1309 | _exit(EXIT_FAILURE); |
1310 | } | |
1311 | ||
1312 | if (flags & FORK_RESET_SIGNALS) { | |
1313 | r = reset_all_signal_handlers(); | |
1314 | if (r < 0) { | |
b6e1fff1 | 1315 | log_full_errno(prio, r, "Failed to reset signal handlers: %m"); |
4c253ed1 LP |
1316 | _exit(EXIT_FAILURE); |
1317 | } | |
1318 | ||
1319 | /* This implicitly undoes the signal mask stuff we did before the fork()ing above */ | |
1320 | r = reset_signal_mask(); | |
1321 | if (r < 0) { | |
b6e1fff1 | 1322 | log_full_errno(prio, r, "Failed to reset signal mask: %m"); |
4c253ed1 LP |
1323 | _exit(EXIT_FAILURE); |
1324 | } | |
1325 | } else if (block_signals) { /* undo what we did above */ | |
1326 | if (sigprocmask(SIG_SETMASK, &saved_ss, NULL) < 0) { | |
b6e1fff1 | 1327 | log_full_errno(prio, errno, "Failed to restore signal mask: %m"); |
4c253ed1 LP |
1328 | _exit(EXIT_FAILURE); |
1329 | } | |
1330 | } | |
1331 | ||
1332 | if (flags & FORK_DEATHSIG) { | |
7ddc2dc5 | 1333 | pid_t ppid; |
4c253ed1 LP |
1334 | /* Let's see if the parent PID is still the one we started from? If not, then the parent |
1335 | * already died by the time we set PR_SET_PDEATHSIG, hence let's emulate the effect */ | |
1336 | ||
7ddc2dc5 SL |
1337 | ppid = getppid(); |
1338 | if (ppid == 0) | |
1339 | /* Parent is in a differn't PID namespace. */; | |
1340 | else if (ppid != original_pid) { | |
4c253ed1 LP |
1341 | log_debug("Parent died early, raising SIGTERM."); |
1342 | (void) raise(SIGTERM); | |
1343 | _exit(EXIT_FAILURE); | |
1344 | } | |
1345 | } | |
1346 | ||
1347 | if (flags & FORK_CLOSE_ALL_FDS) { | |
1348 | /* Close the logs here in case it got reopened above, as close_all_fds() would close them for us */ | |
1349 | log_close(); | |
1350 | ||
1351 | r = close_all_fds(except_fds, n_except_fds); | |
1352 | if (r < 0) { | |
b6e1fff1 | 1353 | log_full_errno(prio, r, "Failed to close all file descriptors: %m"); |
4c253ed1 LP |
1354 | _exit(EXIT_FAILURE); |
1355 | } | |
1356 | } | |
1357 | ||
1358 | /* When we were asked to reopen the logs, do so again now */ | |
1359 | if (flags & FORK_REOPEN_LOG) { | |
1360 | log_open(); | |
1361 | log_set_open_when_needed(false); | |
1362 | } | |
1363 | ||
1364 | if (flags & FORK_NULL_STDIO) { | |
1365 | r = make_null_stdio(); | |
1366 | if (r < 0) { | |
b6e1fff1 | 1367 | log_full_errno(prio, r, "Failed to connect stdin/stdout to /dev/null: %m"); |
4c253ed1 LP |
1368 | _exit(EXIT_FAILURE); |
1369 | } | |
1370 | } | |
1371 | ||
1372 | if (ret_pid) | |
1373 | *ret_pid = getpid_cached(); | |
1374 | ||
1375 | return 0; | |
1376 | } | |
1377 | ||
da6053d0 | 1378 | int fork_agent(const char *name, const int except[], size_t n_except, pid_t *ret_pid, const char *path, ...) { |
78752f2e | 1379 | bool stdout_is_tty, stderr_is_tty; |
da6053d0 | 1380 | size_t n, i; |
78752f2e LP |
1381 | va_list ap; |
1382 | char **l; | |
1383 | int r; | |
1384 | ||
1385 | assert(path); | |
1386 | ||
1387 | /* Spawns a temporary TTY agent, making sure it goes away when we go away */ | |
1388 | ||
1389 | r = safe_fork_full(name, except, n_except, FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_CLOSE_ALL_FDS, ret_pid); | |
1390 | if (r < 0) | |
1391 | return r; | |
1392 | if (r > 0) | |
1393 | return 0; | |
1394 | ||
1395 | /* In the child: */ | |
1396 | ||
1397 | stdout_is_tty = isatty(STDOUT_FILENO); | |
1398 | stderr_is_tty = isatty(STDERR_FILENO); | |
1399 | ||
1400 | if (!stdout_is_tty || !stderr_is_tty) { | |
1401 | int fd; | |
1402 | ||
1403 | /* Detach from stdout/stderr. and reopen | |
1404 | * /dev/tty for them. This is important to | |
1405 | * ensure that when systemctl is started via | |
1406 | * popen() or a similar call that expects to | |
1407 | * read EOF we actually do generate EOF and | |
1408 | * not delay this indefinitely by because we | |
1409 | * keep an unused copy of stdin around. */ | |
1410 | fd = open("/dev/tty", O_WRONLY); | |
1411 | if (fd < 0) { | |
1412 | log_error_errno(errno, "Failed to open /dev/tty: %m"); | |
1413 | _exit(EXIT_FAILURE); | |
1414 | } | |
1415 | ||
1416 | if (!stdout_is_tty && dup2(fd, STDOUT_FILENO) < 0) { | |
1417 | log_error_errno(errno, "Failed to dup2 /dev/tty: %m"); | |
1418 | _exit(EXIT_FAILURE); | |
1419 | } | |
1420 | ||
1421 | if (!stderr_is_tty && dup2(fd, STDERR_FILENO) < 0) { | |
1422 | log_error_errno(errno, "Failed to dup2 /dev/tty: %m"); | |
1423 | _exit(EXIT_FAILURE); | |
1424 | } | |
1425 | ||
e7685a77 | 1426 | safe_close_above_stdio(fd); |
78752f2e LP |
1427 | } |
1428 | ||
1429 | /* Count arguments */ | |
1430 | va_start(ap, path); | |
1431 | for (n = 0; va_arg(ap, char*); n++) | |
1432 | ; | |
1433 | va_end(ap); | |
1434 | ||
1435 | /* Allocate strv */ | |
cf409d15 | 1436 | l = newa(char*, n + 1); |
78752f2e LP |
1437 | |
1438 | /* Fill in arguments */ | |
1439 | va_start(ap, path); | |
1440 | for (i = 0; i <= n; i++) | |
1441 | l[i] = va_arg(ap, char*); | |
1442 | va_end(ap); | |
1443 | ||
1444 | execv(path, l); | |
1445 | _exit(EXIT_FAILURE); | |
1446 | } | |
1447 | ||
7b3e062c LP |
1448 | static const char *const ioprio_class_table[] = { |
1449 | [IOPRIO_CLASS_NONE] = "none", | |
1450 | [IOPRIO_CLASS_RT] = "realtime", | |
1451 | [IOPRIO_CLASS_BE] = "best-effort", | |
1452 | [IOPRIO_CLASS_IDLE] = "idle" | |
1453 | }; | |
1454 | ||
10062bbc | 1455 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ioprio_class, int, IOPRIO_N_CLASSES); |
7b3e062c LP |
1456 | |
1457 | static const char *const sigchld_code_table[] = { | |
1458 | [CLD_EXITED] = "exited", | |
1459 | [CLD_KILLED] = "killed", | |
1460 | [CLD_DUMPED] = "dumped", | |
1461 | [CLD_TRAPPED] = "trapped", | |
1462 | [CLD_STOPPED] = "stopped", | |
1463 | [CLD_CONTINUED] = "continued", | |
1464 | }; | |
1465 | ||
1466 | DEFINE_STRING_TABLE_LOOKUP(sigchld_code, int); | |
1467 | ||
1468 | static const char* const sched_policy_table[] = { | |
1469 | [SCHED_OTHER] = "other", | |
1470 | [SCHED_BATCH] = "batch", | |
1471 | [SCHED_IDLE] = "idle", | |
1472 | [SCHED_FIFO] = "fifo", | |
1473 | [SCHED_RR] = "rr" | |
1474 | }; | |
1475 | ||
1476 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(sched_policy, int, INT_MAX); |