]>
Commit | Line | Data |
---|---|---|
53e1b683 | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
0b452006 | 2 | |
4f5dd394 | 3 | #include <ctype.h> |
0b452006 | 4 | #include <errno.h> |
11c3a366 TA |
5 | #include <limits.h> |
6 | #include <linux/oom.h> | |
7b3e062c | 7 | #include <sched.h> |
0b452006 | 8 | #include <signal.h> |
4f5dd394 LP |
9 | #include <stdbool.h> |
10 | #include <stdio.h> | |
11c3a366 | 11 | #include <stdlib.h> |
4f5dd394 | 12 | #include <string.h> |
9bfaffd5 | 13 | #include <sys/mman.h> |
e2047ba9 | 14 | #include <sys/mount.h> |
7b3e062c | 15 | #include <sys/personality.h> |
405f8907 | 16 | #include <sys/prctl.h> |
4f5dd394 LP |
17 | #include <sys/types.h> |
18 | #include <sys/wait.h> | |
11c3a366 | 19 | #include <syslog.h> |
4f5dd394 | 20 | #include <unistd.h> |
349cc4a5 | 21 | #if HAVE_VALGRIND_VALGRIND_H |
dcadc967 EV |
22 | #include <valgrind/valgrind.h> |
23 | #endif | |
0b452006 | 24 | |
b5efdb8a | 25 | #include "alloc-util.h" |
6e5f1b57 | 26 | #include "architecture.h" |
4f5dd394 | 27 | #include "escape.h" |
21c491e1 | 28 | #include "env-util.h" |
3ffd4af2 | 29 | #include "fd-util.h" |
0b452006 | 30 | #include "fileio.h" |
f4f15635 | 31 | #include "fs-util.h" |
7b3e062c | 32 | #include "ioprio.h" |
0b452006 | 33 | #include "log.h" |
11c3a366 | 34 | #include "macro.h" |
0a970718 | 35 | #include "memory-util.h" |
11c3a366 | 36 | #include "missing.h" |
0cb8e3d1 | 37 | #include "namespace-util.h" |
93cc7779 | 38 | #include "process-util.h" |
8869a0b4 | 39 | #include "raw-clone.h" |
909106eb | 40 | #include "rlimit-util.h" |
93cc7779 | 41 | #include "signal-util.h" |
1359fffa | 42 | #include "stat-util.h" |
7b3e062c | 43 | #include "string-table.h" |
07630cea | 44 | #include "string-util.h" |
4c253ed1 | 45 | #include "terminal-util.h" |
b1d4f8e1 | 46 | #include "user-util.h" |
0b452006 | 47 | |
0a51b45c | 48 | static int get_process_state(pid_t pid) { |
0b452006 RC |
49 | const char *p; |
50 | char state; | |
51 | int r; | |
52 | _cleanup_free_ char *line = NULL; | |
53 | ||
54 | assert(pid >= 0); | |
55 | ||
56 | p = procfs_file_alloca(pid, "stat"); | |
a644184a | 57 | |
0b452006 | 58 | r = read_one_line_file(p, &line); |
a644184a LP |
59 | if (r == -ENOENT) |
60 | return -ESRCH; | |
0b452006 RC |
61 | if (r < 0) |
62 | return r; | |
63 | ||
64 | p = strrchr(line, ')'); | |
65 | if (!p) | |
66 | return -EIO; | |
67 | ||
68 | p++; | |
69 | ||
70 | if (sscanf(p, " %c", &state) != 1) | |
71 | return -EIO; | |
72 | ||
73 | return (unsigned char) state; | |
74 | } | |
75 | ||
ce268825 LP |
76 | int get_process_comm(pid_t pid, char **ret) { |
77 | _cleanup_free_ char *escaped = NULL, *comm = NULL; | |
0b452006 RC |
78 | const char *p; |
79 | int r; | |
80 | ||
ce268825 | 81 | assert(ret); |
0b452006 RC |
82 | assert(pid >= 0); |
83 | ||
ce268825 LP |
84 | escaped = new(char, TASK_COMM_LEN); |
85 | if (!escaped) | |
86 | return -ENOMEM; | |
87 | ||
0b452006 RC |
88 | p = procfs_file_alloca(pid, "comm"); |
89 | ||
ce268825 | 90 | r = read_one_line_file(p, &comm); |
0b452006 RC |
91 | if (r == -ENOENT) |
92 | return -ESRCH; | |
ce268825 LP |
93 | if (r < 0) |
94 | return r; | |
95 | ||
96 | /* Escape unprintable characters, just in case, but don't grow the string beyond the underlying size */ | |
97 | cellescape(escaped, TASK_COMM_LEN, comm); | |
0b452006 | 98 | |
ce268825 LP |
99 | *ret = TAKE_PTR(escaped); |
100 | return 0; | |
0b452006 RC |
101 | } |
102 | ||
103 | int get_process_cmdline(pid_t pid, size_t max_length, bool comm_fallback, char **line) { | |
104 | _cleanup_fclose_ FILE *f = NULL; | |
ba4cd7e2 | 105 | bool space = false; |
eb1ec489 MS |
106 | char *k; |
107 | _cleanup_free_ char *ans = NULL; | |
0b452006 | 108 | const char *p; |
fdeea3f4 | 109 | int c, r; |
0b452006 RC |
110 | |
111 | assert(line); | |
112 | assert(pid >= 0); | |
113 | ||
69281c49 LP |
114 | /* Retrieves a process' command line. Replaces unprintable characters while doing so by whitespace (coalescing |
115 | * multiple sequential ones into one). If max_length is != 0 will return a string of the specified size at most | |
116 | * (the trailing NUL byte does count towards the length here!), abbreviated with a "..." ellipsis. If | |
117 | * comm_fallback is true and the process has no command line set (the case for kernel threads), or has a | |
118 | * command line that resolves to the empty string will return the "comm" name of the process instead. | |
119 | * | |
120 | * Returns -ESRCH if the process doesn't exist, and -ENOENT if the process has no command line (and | |
c0534780 | 121 | * comm_fallback is false). Returns 0 and sets *line otherwise. */ |
69281c49 | 122 | |
0b452006 | 123 | p = procfs_file_alloca(pid, "cmdline"); |
fdeea3f4 ZJS |
124 | r = fopen_unlocked(p, "re", &f); |
125 | if (r == -ENOENT) | |
126 | return -ESRCH; | |
127 | if (r < 0) | |
128 | return r; | |
35bbbf85 | 129 | |
21c491e1 | 130 | if (max_length == 0) |
2d5d2e0c | 131 | /* This is supposed to be a safety guard against runaway command lines. */ |
21c491e1 | 132 | max_length = sc_arg_max(); |
2d5d2e0c | 133 | |
69281c49 LP |
134 | if (max_length == 1) { |
135 | ||
136 | /* If there's only room for one byte, return the empty string */ | |
c0534780 ZJS |
137 | ans = new0(char, 1); |
138 | if (!ans) | |
69281c49 LP |
139 | return -ENOMEM; |
140 | ||
eb1ec489 | 141 | *line = TAKE_PTR(ans); |
69281c49 LP |
142 | return 0; |
143 | ||
0b452006 | 144 | } else { |
69281c49 | 145 | bool dotdotdot = false; |
0b452006 RC |
146 | size_t left; |
147 | ||
c0534780 ZJS |
148 | ans = new(char, max_length); |
149 | if (!ans) | |
0b452006 RC |
150 | return -ENOMEM; |
151 | ||
c0534780 | 152 | k = ans; |
0b452006 RC |
153 | left = max_length; |
154 | while ((c = getc(f)) != EOF) { | |
155 | ||
156 | if (isprint(c)) { | |
69281c49 | 157 | |
0b452006 | 158 | if (space) { |
69281c49 LP |
159 | if (left <= 2) { |
160 | dotdotdot = true; | |
0b452006 | 161 | break; |
69281c49 | 162 | } |
0b452006 RC |
163 | |
164 | *(k++) = ' '; | |
165 | left--; | |
166 | space = false; | |
167 | } | |
168 | ||
69281c49 LP |
169 | if (left <= 1) { |
170 | dotdotdot = true; | |
0b452006 | 171 | break; |
69281c49 | 172 | } |
0b452006 RC |
173 | |
174 | *(k++) = (char) c; | |
175 | left--; | |
c0534780 | 176 | } else if (k > ans) |
0b452006 RC |
177 | space = true; |
178 | } | |
179 | ||
69281c49 LP |
180 | if (dotdotdot) { |
181 | if (max_length <= 4) { | |
c0534780 | 182 | k = ans; |
69281c49 LP |
183 | left = max_length; |
184 | } else { | |
c0534780 | 185 | k = ans + max_length - 4; |
69281c49 LP |
186 | left = 4; |
187 | ||
188 | /* Eat up final spaces */ | |
c0534780 | 189 | while (k > ans && isspace(k[-1])) { |
69281c49 LP |
190 | k--; |
191 | left++; | |
192 | } | |
193 | } | |
194 | ||
195 | strncpy(k, "...", left-1); | |
b09df4e2 | 196 | k[left-1] = 0; |
0b452006 RC |
197 | } else |
198 | *k = 0; | |
199 | } | |
200 | ||
201 | /* Kernel threads have no argv[] */ | |
c0534780 | 202 | if (isempty(ans)) { |
0b452006 RC |
203 | _cleanup_free_ char *t = NULL; |
204 | int h; | |
205 | ||
eb1ec489 | 206 | ans = mfree(ans); |
0b452006 RC |
207 | |
208 | if (!comm_fallback) | |
209 | return -ENOENT; | |
210 | ||
211 | h = get_process_comm(pid, &t); | |
212 | if (h < 0) | |
213 | return h; | |
214 | ||
2d5d2e0c | 215 | size_t l = strlen(t); |
69281c49 | 216 | |
2d5d2e0c ZJS |
217 | if (l + 3 <= max_length) { |
218 | ans = strjoin("[", t, "]"); | |
219 | if (!ans) | |
220 | return -ENOMEM; | |
69281c49 | 221 | |
2d5d2e0c ZJS |
222 | } else if (max_length <= 6) { |
223 | ans = new(char, max_length); | |
224 | if (!ans) | |
225 | return -ENOMEM; | |
69281c49 | 226 | |
2d5d2e0c ZJS |
227 | memcpy(ans, "[...]", max_length-1); |
228 | ans[max_length-1] = 0; | |
229 | } else { | |
230 | t[max_length - 6] = 0; | |
69281c49 | 231 | |
2d5d2e0c ZJS |
232 | /* Chop off final spaces */ |
233 | delete_trailing_chars(t, WHITESPACE); | |
69281c49 | 234 | |
2d5d2e0c ZJS |
235 | ans = strjoin("[", t, "...]"); |
236 | if (!ans) | |
237 | return -ENOMEM; | |
69281c49 | 238 | } |
eb1ec489 MS |
239 | |
240 | *line = TAKE_PTR(ans); | |
241 | return 0; | |
0b452006 RC |
242 | } |
243 | ||
eb1ec489 MS |
244 | k = realloc(ans, strlen(ans) + 1); |
245 | if (!k) | |
246 | return -ENOMEM; | |
247 | ||
248 | ans = NULL; | |
249 | *line = k; | |
250 | ||
0b452006 RC |
251 | return 0; |
252 | } | |
253 | ||
9bfaffd5 LP |
254 | int rename_process(const char name[]) { |
255 | static size_t mm_size = 0; | |
256 | static char *mm = NULL; | |
257 | bool truncated = false; | |
258 | size_t l; | |
259 | ||
260 | /* This is a like a poor man's setproctitle(). It changes the comm field, argv[0], and also the glibc's | |
261 | * internally used name of the process. For the first one a limit of 16 chars applies; to the second one in | |
262 | * many cases one of 10 (i.e. length of "/sbin/init") — however if we have CAP_SYS_RESOURCES it is unbounded; | |
263 | * to the third one 7 (i.e. the length of "systemd". If you pass a longer string it will likely be | |
264 | * truncated. | |
265 | * | |
266 | * Returns 0 if a name was set but truncated, > 0 if it was set but not truncated. */ | |
267 | ||
268 | if (isempty(name)) | |
269 | return -EINVAL; /* let's not confuse users unnecessarily with an empty name */ | |
405f8907 | 270 | |
1096bb8a LP |
271 | if (!is_main_thread()) |
272 | return -EPERM; /* Let's not allow setting the process name from other threads than the main one, as we | |
273 | * cache things without locking, and we make assumptions that PR_SET_NAME sets the | |
274 | * process name that isn't correct on any other threads */ | |
275 | ||
9bfaffd5 | 276 | l = strlen(name); |
405f8907 | 277 | |
5a8af747 LP |
278 | /* First step, change the comm field. The main thread's comm is identical to the process comm. This means we |
279 | * can use PR_SET_NAME, which sets the thread name for the calling thread. */ | |
280 | if (prctl(PR_SET_NAME, name) < 0) | |
281 | log_debug_errno(errno, "PR_SET_NAME failed: %m"); | |
92f14395 | 282 | if (l >= TASK_COMM_LEN) /* Linux process names can be 15 chars at max */ |
9bfaffd5 LP |
283 | truncated = true; |
284 | ||
285 | /* Second step, change glibc's ID of the process name. */ | |
286 | if (program_invocation_name) { | |
287 | size_t k; | |
288 | ||
289 | k = strlen(program_invocation_name); | |
290 | strncpy(program_invocation_name, name, k); | |
291 | if (l > k) | |
292 | truncated = true; | |
293 | } | |
294 | ||
295 | /* Third step, completely replace the argv[] array the kernel maintains for us. This requires privileges, but | |
296 | * has the advantage that the argv[] array is exactly what we want it to be, and not filled up with zeros at | |
13e785f7 | 297 | * the end. This is the best option for changing /proc/self/cmdline. */ |
01f989c6 JW |
298 | |
299 | /* Let's not bother with this if we don't have euid == 0. Strictly speaking we should check for the | |
300 | * CAP_SYS_RESOURCE capability which is independent of the euid. In our own code the capability generally is | |
301 | * present only for euid == 0, hence let's use this as quick bypass check, to avoid calling mmap() if | |
302 | * PR_SET_MM_ARG_{START,END} fails with EPERM later on anyway. After all geteuid() is dead cheap to call, but | |
303 | * mmap() is not. */ | |
304 | if (geteuid() != 0) | |
305 | log_debug("Skipping PR_SET_MM, as we don't have privileges."); | |
306 | else if (mm_size < l+1) { | |
9bfaffd5 LP |
307 | size_t nn_size; |
308 | char *nn; | |
309 | ||
9bfaffd5 LP |
310 | nn_size = PAGE_ALIGN(l+1); |
311 | nn = mmap(NULL, nn_size, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); | |
312 | if (nn == MAP_FAILED) { | |
313 | log_debug_errno(errno, "mmap() failed: %m"); | |
314 | goto use_saved_argv; | |
315 | } | |
405f8907 | 316 | |
9bfaffd5 LP |
317 | strncpy(nn, name, nn_size); |
318 | ||
319 | /* Now, let's tell the kernel about this new memory */ | |
320 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_START, (unsigned long) nn, 0, 0) < 0) { | |
14ee72b7 FS |
321 | /* HACK: prctl() API is kind of dumb on this point. The existing end address may already be |
322 | * below the desired start address, in which case the kernel may have kicked this back due | |
323 | * to a range-check failure (see linux/kernel/sys.c:validate_prctl_map() to see this in | |
324 | * action). The proper solution would be to have a prctl() API that could set both start+end | |
325 | * simultaneously, or at least let us query the existing address to anticipate this condition | |
326 | * and respond accordingly. For now, we can only guess at the cause of this failure and try | |
327 | * a workaround--which will briefly expand the arg space to something potentially huge before | |
328 | * resizing it to what we want. */ | |
329 | log_debug_errno(errno, "PR_SET_MM_ARG_START failed, attempting PR_SET_MM_ARG_END hack: %m"); | |
330 | ||
331 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) nn + l + 1, 0, 0) < 0) { | |
332 | log_debug_errno(errno, "PR_SET_MM_ARG_END hack failed, proceeding without: %m"); | |
333 | (void) munmap(nn, nn_size); | |
334 | goto use_saved_argv; | |
335 | } | |
9bfaffd5 | 336 | |
14ee72b7 FS |
337 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_START, (unsigned long) nn, 0, 0) < 0) { |
338 | log_debug_errno(errno, "PR_SET_MM_ARG_START still failed, proceeding without: %m"); | |
339 | goto use_saved_argv; | |
340 | } | |
341 | } else { | |
342 | /* And update the end pointer to the new end, too. If this fails, we don't really know what | |
343 | * to do, it's pretty unlikely that we can rollback, hence we'll just accept the failure, | |
344 | * and continue. */ | |
345 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) nn + l + 1, 0, 0) < 0) | |
346 | log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m"); | |
347 | } | |
9bfaffd5 LP |
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 | 389 | int is_kernel_thread(pid_t pid) { |
36b5119a LP |
390 | _cleanup_free_ char *line = NULL; |
391 | unsigned long long flags; | |
392 | size_t l, i; | |
0b452006 | 393 | const char *p; |
36b5119a LP |
394 | char *q; |
395 | int r; | |
0b452006 | 396 | |
4c701096 | 397 | if (IN_SET(pid, 0, 1) || pid == getpid_cached()) /* pid 1, and we ourselves certainly aren't a kernel thread */ |
0b452006 | 398 | return 0; |
36b5119a LP |
399 | if (!pid_is_valid(pid)) |
400 | return -EINVAL; | |
0b452006 | 401 | |
36b5119a LP |
402 | p = procfs_file_alloca(pid, "stat"); |
403 | r = read_one_line_file(p, &line); | |
404 | if (r == -ENOENT) | |
405 | return -ESRCH; | |
406 | if (r < 0) | |
407 | return r; | |
0b452006 | 408 | |
36b5119a LP |
409 | /* Skip past the comm field */ |
410 | q = strrchr(line, ')'); | |
411 | if (!q) | |
412 | return -EINVAL; | |
413 | q++; | |
414 | ||
415 | /* Skip 6 fields to reach the flags field */ | |
416 | for (i = 0; i < 6; i++) { | |
417 | l = strspn(q, WHITESPACE); | |
418 | if (l < 1) | |
419 | return -EINVAL; | |
420 | q += l; | |
421 | ||
422 | l = strcspn(q, WHITESPACE); | |
423 | if (l < 1) | |
424 | return -EINVAL; | |
425 | q += l; | |
a644184a | 426 | } |
0b452006 | 427 | |
f21f31b2 | 428 | /* Skip preceding whitespace */ |
36b5119a LP |
429 | l = strspn(q, WHITESPACE); |
430 | if (l < 1) | |
431 | return -EINVAL; | |
432 | q += l; | |
35bbbf85 | 433 | |
36b5119a LP |
434 | /* Truncate the rest */ |
435 | l = strcspn(q, WHITESPACE); | |
436 | if (l < 1) | |
437 | return -EINVAL; | |
438 | q[l] = 0; | |
0b452006 | 439 | |
36b5119a LP |
440 | r = safe_atollu(q, &flags); |
441 | if (r < 0) | |
442 | return r; | |
0b452006 | 443 | |
36b5119a | 444 | return !!(flags & PF_KTHREAD); |
0b452006 RC |
445 | } |
446 | ||
447 | int get_process_capeff(pid_t pid, char **capeff) { | |
448 | const char *p; | |
a644184a | 449 | int r; |
0b452006 RC |
450 | |
451 | assert(capeff); | |
452 | assert(pid >= 0); | |
453 | ||
454 | p = procfs_file_alloca(pid, "status"); | |
455 | ||
c4cd1d4d | 456 | r = get_proc_field(p, "CapEff", WHITESPACE, capeff); |
a644184a LP |
457 | if (r == -ENOENT) |
458 | return -ESRCH; | |
459 | ||
460 | return r; | |
0b452006 RC |
461 | } |
462 | ||
463 | static int get_process_link_contents(const char *proc_file, char **name) { | |
464 | int r; | |
465 | ||
466 | assert(proc_file); | |
467 | assert(name); | |
468 | ||
469 | r = readlink_malloc(proc_file, name); | |
a644184a LP |
470 | if (r == -ENOENT) |
471 | return -ESRCH; | |
0b452006 | 472 | if (r < 0) |
a644184a | 473 | return r; |
0b452006 RC |
474 | |
475 | return 0; | |
476 | } | |
477 | ||
478 | int get_process_exe(pid_t pid, char **name) { | |
479 | const char *p; | |
480 | char *d; | |
481 | int r; | |
482 | ||
483 | assert(pid >= 0); | |
484 | ||
485 | p = procfs_file_alloca(pid, "exe"); | |
486 | r = get_process_link_contents(p, name); | |
487 | if (r < 0) | |
488 | return r; | |
489 | ||
490 | d = endswith(*name, " (deleted)"); | |
491 | if (d) | |
492 | *d = '\0'; | |
493 | ||
494 | return 0; | |
495 | } | |
496 | ||
497 | static int get_process_id(pid_t pid, const char *field, uid_t *uid) { | |
498 | _cleanup_fclose_ FILE *f = NULL; | |
0b452006 | 499 | const char *p; |
7e7a16a0 | 500 | int r; |
0b452006 RC |
501 | |
502 | assert(field); | |
503 | assert(uid); | |
504 | ||
07b38ba5 | 505 | if (pid < 0) |
6f8cbcdb LP |
506 | return -EINVAL; |
507 | ||
0b452006 | 508 | p = procfs_file_alloca(pid, "status"); |
fdeea3f4 ZJS |
509 | r = fopen_unlocked(p, "re", &f); |
510 | if (r == -ENOENT) | |
511 | return -ESRCH; | |
512 | if (r < 0) | |
513 | return r; | |
35bbbf85 | 514 | |
7e7a16a0 LP |
515 | for (;;) { |
516 | _cleanup_free_ char *line = NULL; | |
0b452006 RC |
517 | char *l; |
518 | ||
7e7a16a0 LP |
519 | r = read_line(f, LONG_LINE_MAX, &line); |
520 | if (r < 0) | |
521 | return r; | |
522 | if (r == 0) | |
523 | break; | |
524 | ||
0b452006 RC |
525 | l = strstrip(line); |
526 | ||
527 | if (startswith(l, field)) { | |
528 | l += strlen(field); | |
529 | l += strspn(l, WHITESPACE); | |
530 | ||
531 | l[strcspn(l, WHITESPACE)] = 0; | |
532 | ||
533 | return parse_uid(l, uid); | |
534 | } | |
535 | } | |
536 | ||
537 | return -EIO; | |
538 | } | |
539 | ||
540 | int get_process_uid(pid_t pid, uid_t *uid) { | |
6f8cbcdb LP |
541 | |
542 | if (pid == 0 || pid == getpid_cached()) { | |
543 | *uid = getuid(); | |
544 | return 0; | |
545 | } | |
546 | ||
0b452006 RC |
547 | return get_process_id(pid, "Uid:", uid); |
548 | } | |
549 | ||
550 | int get_process_gid(pid_t pid, gid_t *gid) { | |
6f8cbcdb LP |
551 | |
552 | if (pid == 0 || pid == getpid_cached()) { | |
553 | *gid = getgid(); | |
554 | return 0; | |
555 | } | |
556 | ||
0b452006 RC |
557 | assert_cc(sizeof(uid_t) == sizeof(gid_t)); |
558 | return get_process_id(pid, "Gid:", gid); | |
559 | } | |
560 | ||
561 | int get_process_cwd(pid_t pid, char **cwd) { | |
562 | const char *p; | |
563 | ||
564 | assert(pid >= 0); | |
565 | ||
566 | p = procfs_file_alloca(pid, "cwd"); | |
567 | ||
568 | return get_process_link_contents(p, cwd); | |
569 | } | |
570 | ||
571 | int get_process_root(pid_t pid, char **root) { | |
572 | const char *p; | |
573 | ||
574 | assert(pid >= 0); | |
575 | ||
576 | p = procfs_file_alloca(pid, "root"); | |
577 | ||
578 | return get_process_link_contents(p, root); | |
579 | } | |
580 | ||
2a7797e9 LP |
581 | #define ENVIRONMENT_BLOCK_MAX (5U*1024U*1024U) |
582 | ||
0b452006 RC |
583 | int get_process_environ(pid_t pid, char **env) { |
584 | _cleanup_fclose_ FILE *f = NULL; | |
585 | _cleanup_free_ char *outcome = NULL; | |
0b452006 | 586 | size_t allocated = 0, sz = 0; |
2a7797e9 LP |
587 | const char *p; |
588 | int r; | |
0b452006 RC |
589 | |
590 | assert(pid >= 0); | |
591 | assert(env); | |
592 | ||
593 | p = procfs_file_alloca(pid, "environ"); | |
594 | ||
fdeea3f4 ZJS |
595 | r = fopen_unlocked(p, "re", &f); |
596 | if (r == -ENOENT) | |
597 | return -ESRCH; | |
598 | if (r < 0) | |
599 | return r; | |
35bbbf85 | 600 | |
2a7797e9 LP |
601 | for (;;) { |
602 | char c; | |
603 | ||
604 | if (sz >= ENVIRONMENT_BLOCK_MAX) | |
605 | return -ENOBUFS; | |
606 | ||
0b452006 RC |
607 | if (!GREEDY_REALLOC(outcome, allocated, sz + 5)) |
608 | return -ENOMEM; | |
609 | ||
2a7797e9 LP |
610 | r = safe_fgetc(f, &c); |
611 | if (r < 0) | |
612 | return r; | |
613 | if (r == 0) | |
614 | break; | |
615 | ||
0b452006 RC |
616 | if (c == '\0') |
617 | outcome[sz++] = '\n'; | |
618 | else | |
619 | sz += cescape_char(c, outcome + sz); | |
620 | } | |
621 | ||
2a7797e9 | 622 | outcome[sz] = '\0'; |
ae2a15bc | 623 | *env = TAKE_PTR(outcome); |
0b452006 RC |
624 | |
625 | return 0; | |
626 | } | |
627 | ||
6bc73acb | 628 | int get_process_ppid(pid_t pid, pid_t *_ppid) { |
0b452006 RC |
629 | int r; |
630 | _cleanup_free_ char *line = NULL; | |
631 | long unsigned ppid; | |
632 | const char *p; | |
633 | ||
634 | assert(pid >= 0); | |
635 | assert(_ppid); | |
636 | ||
6f8cbcdb | 637 | if (pid == 0 || pid == getpid_cached()) { |
0b452006 RC |
638 | *_ppid = getppid(); |
639 | return 0; | |
640 | } | |
641 | ||
642 | p = procfs_file_alloca(pid, "stat"); | |
643 | r = read_one_line_file(p, &line); | |
a644184a LP |
644 | if (r == -ENOENT) |
645 | return -ESRCH; | |
0b452006 RC |
646 | if (r < 0) |
647 | return r; | |
648 | ||
649 | /* Let's skip the pid and comm fields. The latter is enclosed | |
650 | * in () but does not escape any () in its value, so let's | |
651 | * skip over it manually */ | |
652 | ||
653 | p = strrchr(line, ')'); | |
654 | if (!p) | |
655 | return -EIO; | |
656 | ||
657 | p++; | |
658 | ||
659 | if (sscanf(p, " " | |
660 | "%*c " /* state */ | |
661 | "%lu ", /* ppid */ | |
662 | &ppid) != 1) | |
663 | return -EIO; | |
664 | ||
665 | if ((long unsigned) (pid_t) ppid != ppid) | |
666 | return -ERANGE; | |
667 | ||
668 | *_ppid = (pid_t) ppid; | |
669 | ||
670 | return 0; | |
671 | } | |
672 | ||
673 | int wait_for_terminate(pid_t pid, siginfo_t *status) { | |
674 | siginfo_t dummy; | |
675 | ||
676 | assert(pid >= 1); | |
677 | ||
678 | if (!status) | |
679 | status = &dummy; | |
680 | ||
681 | for (;;) { | |
682 | zero(*status); | |
683 | ||
684 | if (waitid(P_PID, pid, status, WEXITED) < 0) { | |
685 | ||
686 | if (errno == EINTR) | |
687 | continue; | |
688 | ||
3f0083a2 | 689 | return negative_errno(); |
0b452006 RC |
690 | } |
691 | ||
692 | return 0; | |
693 | } | |
694 | } | |
695 | ||
696 | /* | |
697 | * Return values: | |
698 | * < 0 : wait_for_terminate() failed to get the state of the | |
699 | * process, the process was terminated by a signal, or | |
700 | * failed for an unknown reason. | |
701 | * >=0 : The process terminated normally, and its exit code is | |
702 | * returned. | |
703 | * | |
704 | * That is, success is indicated by a return value of zero, and an | |
705 | * error is indicated by a non-zero value. | |
706 | * | |
707 | * A warning is emitted if the process terminates abnormally, | |
708 | * and also if it returns non-zero unless check_exit_code is true. | |
709 | */ | |
7d4904fe LP |
710 | int wait_for_terminate_and_check(const char *name, pid_t pid, WaitFlags flags) { |
711 | _cleanup_free_ char *buffer = NULL; | |
0b452006 | 712 | siginfo_t status; |
7d4904fe | 713 | int r, prio; |
0b452006 | 714 | |
0b452006 RC |
715 | assert(pid > 1); |
716 | ||
7d4904fe LP |
717 | if (!name) { |
718 | r = get_process_comm(pid, &buffer); | |
719 | if (r < 0) | |
720 | log_debug_errno(r, "Failed to acquire process name of " PID_FMT ", ignoring: %m", pid); | |
721 | else | |
722 | name = buffer; | |
723 | } | |
724 | ||
725 | prio = flags & WAIT_LOG_ABNORMAL ? LOG_ERR : LOG_DEBUG; | |
726 | ||
0b452006 RC |
727 | r = wait_for_terminate(pid, &status); |
728 | if (r < 0) | |
7d4904fe | 729 | return log_full_errno(prio, r, "Failed to wait for %s: %m", strna(name)); |
0b452006 RC |
730 | |
731 | if (status.si_code == CLD_EXITED) { | |
7d4904fe LP |
732 | if (status.si_status != EXIT_SUCCESS) |
733 | log_full(flags & WAIT_LOG_NON_ZERO_EXIT_STATUS ? LOG_ERR : LOG_DEBUG, | |
734 | "%s failed with exit status %i.", strna(name), status.si_status); | |
0b452006 RC |
735 | else |
736 | log_debug("%s succeeded.", name); | |
737 | ||
738 | return status.si_status; | |
7d4904fe | 739 | |
3742095b | 740 | } else if (IN_SET(status.si_code, CLD_KILLED, CLD_DUMPED)) { |
0b452006 | 741 | |
7d4904fe | 742 | log_full(prio, "%s terminated by signal %s.", strna(name), signal_to_string(status.si_status)); |
0b452006 RC |
743 | return -EPROTO; |
744 | } | |
745 | ||
7d4904fe | 746 | log_full(prio, "%s failed due to unknown reason.", strna(name)); |
0b452006 RC |
747 | return -EPROTO; |
748 | } | |
749 | ||
d5641e0d KW |
750 | /* |
751 | * Return values: | |
e225e5c3 LP |
752 | * |
753 | * < 0 : wait_for_terminate_with_timeout() failed to get the state of the process, the process timed out, the process | |
754 | * was terminated by a signal, or failed for an unknown reason. | |
755 | * | |
d5641e0d KW |
756 | * >=0 : The process terminated normally with no failures. |
757 | * | |
e225e5c3 LP |
758 | * Success is indicated by a return value of zero, a timeout is indicated by ETIMEDOUT, and all other child failure |
759 | * states are indicated by error is indicated by a non-zero value. | |
760 | * | |
761 | * This call assumes SIGCHLD has been blocked already, in particular before the child to wait for has been forked off | |
762 | * to remain entirely race-free. | |
d5641e0d KW |
763 | */ |
764 | int wait_for_terminate_with_timeout(pid_t pid, usec_t timeout) { | |
765 | sigset_t mask; | |
766 | int r; | |
767 | usec_t until; | |
768 | ||
769 | assert_se(sigemptyset(&mask) == 0); | |
770 | assert_se(sigaddset(&mask, SIGCHLD) == 0); | |
771 | ||
772 | /* Drop into a sigtimewait-based timeout. Waiting for the | |
773 | * pid to exit. */ | |
774 | until = now(CLOCK_MONOTONIC) + timeout; | |
775 | for (;;) { | |
776 | usec_t n; | |
777 | siginfo_t status = {}; | |
778 | struct timespec ts; | |
779 | ||
780 | n = now(CLOCK_MONOTONIC); | |
781 | if (n >= until) | |
782 | break; | |
783 | ||
784 | r = sigtimedwait(&mask, NULL, timespec_store(&ts, until - n)) < 0 ? -errno : 0; | |
785 | /* Assuming we woke due to the child exiting. */ | |
786 | if (waitid(P_PID, pid, &status, WEXITED|WNOHANG) == 0) { | |
787 | if (status.si_pid == pid) { | |
788 | /* This is the correct child.*/ | |
789 | if (status.si_code == CLD_EXITED) | |
790 | return (status.si_status == 0) ? 0 : -EPROTO; | |
791 | else | |
792 | return -EPROTO; | |
793 | } | |
794 | } | |
795 | /* Not the child, check for errors and proceed appropriately */ | |
796 | if (r < 0) { | |
797 | switch (r) { | |
798 | case -EAGAIN: | |
799 | /* Timed out, child is likely hung. */ | |
800 | return -ETIMEDOUT; | |
801 | case -EINTR: | |
802 | /* Received a different signal and should retry */ | |
803 | continue; | |
804 | default: | |
805 | /* Return any unexpected errors */ | |
806 | return r; | |
807 | } | |
808 | } | |
809 | } | |
810 | ||
811 | return -EPROTO; | |
812 | } | |
813 | ||
89c9030d LP |
814 | void sigkill_wait(pid_t pid) { |
815 | assert(pid > 1); | |
816 | ||
53640e6f | 817 | if (kill(pid, SIGKILL) >= 0) |
89c9030d LP |
818 | (void) wait_for_terminate(pid, NULL); |
819 | } | |
820 | ||
821 | void sigkill_waitp(pid_t *pid) { | |
dfd14786 LP |
822 | PROTECT_ERRNO; |
823 | ||
4d0d3d41 LP |
824 | if (!pid) |
825 | return; | |
826 | if (*pid <= 1) | |
827 | return; | |
828 | ||
89c9030d | 829 | sigkill_wait(*pid); |
4d0d3d41 LP |
830 | } |
831 | ||
392cf1d0 SL |
832 | void sigterm_wait(pid_t pid) { |
833 | assert(pid > 1); | |
834 | ||
53640e6f | 835 | if (kill_and_sigcont(pid, SIGTERM) >= 0) |
392cf1d0 SL |
836 | (void) wait_for_terminate(pid, NULL); |
837 | } | |
838 | ||
0b452006 RC |
839 | int kill_and_sigcont(pid_t pid, int sig) { |
840 | int r; | |
841 | ||
842 | r = kill(pid, sig) < 0 ? -errno : 0; | |
843 | ||
26f417d3 LP |
844 | /* If this worked, also send SIGCONT, unless we already just sent a SIGCONT, or SIGKILL was sent which isn't |
845 | * affected by a process being suspended anyway. */ | |
a3d8d68c | 846 | if (r >= 0 && !IN_SET(sig, SIGCONT, SIGKILL)) |
26f417d3 | 847 | (void) kill(pid, SIGCONT); |
0b452006 RC |
848 | |
849 | return r; | |
850 | } | |
851 | ||
e70f4453 | 852 | int getenv_for_pid(pid_t pid, const char *field, char **ret) { |
0b452006 RC |
853 | _cleanup_fclose_ FILE *f = NULL; |
854 | char *value = NULL; | |
0b452006 | 855 | const char *path; |
0d90bd92 LP |
856 | size_t l, sum = 0; |
857 | int r; | |
0b452006 RC |
858 | |
859 | assert(pid >= 0); | |
860 | assert(field); | |
e70f4453 LP |
861 | assert(ret); |
862 | ||
863 | if (pid == 0 || pid == getpid_cached()) { | |
864 | const char *e; | |
865 | ||
866 | e = getenv(field); | |
867 | if (!e) { | |
868 | *ret = NULL; | |
869 | return 0; | |
870 | } | |
871 | ||
872 | value = strdup(e); | |
873 | if (!value) | |
874 | return -ENOMEM; | |
875 | ||
876 | *ret = value; | |
877 | return 1; | |
878 | } | |
0b452006 | 879 | |
0d90bd92 LP |
880 | if (!pid_is_valid(pid)) |
881 | return -EINVAL; | |
882 | ||
0b452006 RC |
883 | path = procfs_file_alloca(pid, "environ"); |
884 | ||
fdeea3f4 ZJS |
885 | r = fopen_unlocked(path, "re", &f); |
886 | if (r == -ENOENT) | |
887 | return -ESRCH; | |
888 | if (r < 0) | |
889 | return r; | |
35bbbf85 | 890 | |
0b452006 | 891 | l = strlen(field); |
0d90bd92 LP |
892 | for (;;) { |
893 | _cleanup_free_ char *line = NULL; | |
0b452006 | 894 | |
0d90bd92 LP |
895 | if (sum > ENVIRONMENT_BLOCK_MAX) /* Give up searching eventually */ |
896 | return -ENOBUFS; | |
0b452006 | 897 | |
0d90bd92 LP |
898 | r = read_nul_string(f, LONG_LINE_MAX, &line); |
899 | if (r < 0) | |
900 | return r; | |
901 | if (r == 0) /* EOF */ | |
902 | break; | |
0b452006 | 903 | |
0d90bd92 | 904 | sum += r; |
0b452006 | 905 | |
041b5ae1 | 906 | if (strneq(line, field, l) && line[l] == '=') { |
0b452006 RC |
907 | value = strdup(line + l + 1); |
908 | if (!value) | |
909 | return -ENOMEM; | |
910 | ||
e70f4453 LP |
911 | *ret = value; |
912 | return 1; | |
0b452006 | 913 | } |
0d90bd92 | 914 | } |
0b452006 | 915 | |
e70f4453 LP |
916 | *ret = NULL; |
917 | return 0; | |
0b452006 RC |
918 | } |
919 | ||
4d051546 FB |
920 | int pid_is_my_child(pid_t pid) { |
921 | pid_t ppid; | |
922 | int r; | |
923 | ||
924 | if (pid <= 1) | |
925 | return false; | |
926 | ||
927 | r = get_process_ppid(pid, &ppid); | |
928 | if (r < 0) | |
929 | return r; | |
930 | ||
931 | return ppid == getpid_cached(); | |
932 | } | |
933 | ||
0b452006 RC |
934 | bool pid_is_unwaited(pid_t pid) { |
935 | /* Checks whether a PID is still valid at all, including 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 dead and have been waited for, this code would not be running */ |
941 | return true; | |
942 | ||
6f8cbcdb LP |
943 | if (pid == getpid_cached()) |
944 | return true; | |
945 | ||
0b452006 RC |
946 | if (kill(pid, 0) >= 0) |
947 | return true; | |
948 | ||
949 | return errno != ESRCH; | |
950 | } | |
951 | ||
952 | bool pid_is_alive(pid_t pid) { | |
953 | int r; | |
954 | ||
955 | /* Checks whether a PID is still valid and not a zombie */ | |
956 | ||
07b38ba5 | 957 | if (pid < 0) |
0b452006 RC |
958 | return false; |
959 | ||
5fd9b2c5 LP |
960 | if (pid <= 1) /* If we or PID 1 would be a zombie, this code would not be running */ |
961 | return true; | |
962 | ||
6f8cbcdb LP |
963 | if (pid == getpid_cached()) |
964 | return true; | |
965 | ||
0b452006 | 966 | r = get_process_state(pid); |
4c701096 | 967 | if (IN_SET(r, -ESRCH, 'Z')) |
0b452006 RC |
968 | return false; |
969 | ||
970 | return true; | |
971 | } | |
d4510856 | 972 | |
1359fffa MS |
973 | int pid_from_same_root_fs(pid_t pid) { |
974 | const char *root; | |
975 | ||
07b38ba5 | 976 | if (pid < 0) |
6f8cbcdb LP |
977 | return false; |
978 | ||
979 | if (pid == 0 || pid == getpid_cached()) | |
980 | return true; | |
1359fffa MS |
981 | |
982 | root = procfs_file_alloca(pid, "root"); | |
983 | ||
e3f791a2 | 984 | return files_same(root, "/proc/1/root", 0); |
1359fffa MS |
985 | } |
986 | ||
d4510856 LP |
987 | bool is_main_thread(void) { |
988 | static thread_local int cached = 0; | |
989 | ||
990 | if (_unlikely_(cached == 0)) | |
df0ff127 | 991 | cached = getpid_cached() == gettid() ? 1 : -1; |
d4510856 LP |
992 | |
993 | return cached > 0; | |
994 | } | |
7b3e062c | 995 | |
848e863a | 996 | _noreturn_ void freeze(void) { |
7b3e062c | 997 | |
3da48d7a EV |
998 | log_close(); |
999 | ||
7b3e062c | 1000 | /* Make sure nobody waits for us on a socket anymore */ |
7acf581a | 1001 | (void) close_all_fds(NULL, 0); |
7b3e062c LP |
1002 | |
1003 | sync(); | |
1004 | ||
8647283e MS |
1005 | /* Let's not freeze right away, but keep reaping zombies. */ |
1006 | for (;;) { | |
1007 | int r; | |
1008 | siginfo_t si = {}; | |
1009 | ||
1010 | r = waitid(P_ALL, 0, &si, WEXITED); | |
1011 | if (r < 0 && errno != EINTR) | |
1012 | break; | |
1013 | } | |
1014 | ||
1015 | /* waitid() failed with an unexpected error, things are really borked. Freeze now! */ | |
7b3e062c LP |
1016 | for (;;) |
1017 | pause(); | |
1018 | } | |
1019 | ||
1020 | bool oom_score_adjust_is_valid(int oa) { | |
1021 | return oa >= OOM_SCORE_ADJ_MIN && oa <= OOM_SCORE_ADJ_MAX; | |
1022 | } | |
1023 | ||
1024 | unsigned long personality_from_string(const char *p) { | |
6e5f1b57 | 1025 | int architecture; |
7b3e062c | 1026 | |
0c0fea07 LP |
1027 | if (!p) |
1028 | return PERSONALITY_INVALID; | |
1029 | ||
6e5f1b57 LP |
1030 | /* Parse a personality specifier. We use our own identifiers that indicate specific ABIs, rather than just |
1031 | * hints regarding the register size, since we want to keep things open for multiple locally supported ABIs for | |
1032 | * the same register size. */ | |
1033 | ||
1034 | architecture = architecture_from_string(p); | |
1035 | if (architecture < 0) | |
1036 | return PERSONALITY_INVALID; | |
7b3e062c | 1037 | |
0c0fea07 | 1038 | if (architecture == native_architecture()) |
7b3e062c | 1039 | return PER_LINUX; |
0c0fea07 LP |
1040 | #ifdef SECONDARY_ARCHITECTURE |
1041 | if (architecture == SECONDARY_ARCHITECTURE) | |
f2d1736c | 1042 | return PER_LINUX32; |
7b3e062c LP |
1043 | #endif |
1044 | ||
1045 | return PERSONALITY_INVALID; | |
1046 | } | |
1047 | ||
1048 | const char* personality_to_string(unsigned long p) { | |
6e5f1b57 | 1049 | int architecture = _ARCHITECTURE_INVALID; |
7b3e062c | 1050 | |
7b3e062c | 1051 | if (p == PER_LINUX) |
0c0fea07 LP |
1052 | architecture = native_architecture(); |
1053 | #ifdef SECONDARY_ARCHITECTURE | |
6e5f1b57 | 1054 | else if (p == PER_LINUX32) |
0c0fea07 | 1055 | architecture = SECONDARY_ARCHITECTURE; |
7b3e062c LP |
1056 | #endif |
1057 | ||
6e5f1b57 LP |
1058 | if (architecture < 0) |
1059 | return NULL; | |
1060 | ||
1061 | return architecture_to_string(architecture); | |
7b3e062c LP |
1062 | } |
1063 | ||
21022b9d LP |
1064 | int safe_personality(unsigned long p) { |
1065 | int ret; | |
1066 | ||
1067 | /* So here's the deal, personality() is weirdly defined by glibc. In some cases it returns a failure via errno, | |
1068 | * and in others as negative return value containing an errno-like value. Let's work around this: this is a | |
1069 | * wrapper that uses errno if it is set, and uses the return value otherwise. And then it sets both errno and | |
1070 | * the return value indicating the same issue, so that we are definitely on the safe side. | |
1071 | * | |
1072 | * See https://github.com/systemd/systemd/issues/6737 */ | |
1073 | ||
1074 | errno = 0; | |
1075 | ret = personality(p); | |
1076 | if (ret < 0) { | |
1077 | if (errno != 0) | |
1078 | return -errno; | |
1079 | ||
1080 | errno = -ret; | |
1081 | } | |
1082 | ||
1083 | return ret; | |
1084 | } | |
1085 | ||
e8132d63 LP |
1086 | int opinionated_personality(unsigned long *ret) { |
1087 | int current; | |
1088 | ||
1089 | /* Returns the current personality, or PERSONALITY_INVALID if we can't determine it. This function is a bit | |
1090 | * opinionated though, and ignores all the finer-grained bits and exotic personalities, only distinguishing the | |
1091 | * two most relevant personalities: PER_LINUX and PER_LINUX32. */ | |
1092 | ||
21022b9d | 1093 | current = safe_personality(PERSONALITY_INVALID); |
e8132d63 | 1094 | if (current < 0) |
21022b9d | 1095 | return current; |
e8132d63 LP |
1096 | |
1097 | if (((unsigned long) current & 0xffff) == PER_LINUX32) | |
1098 | *ret = PER_LINUX32; | |
1099 | else | |
1100 | *ret = PER_LINUX; | |
1101 | ||
1102 | return 0; | |
1103 | } | |
1104 | ||
dcadc967 | 1105 | void valgrind_summary_hack(void) { |
349cc4a5 | 1106 | #if HAVE_VALGRIND_VALGRIND_H |
df0ff127 | 1107 | if (getpid_cached() == 1 && RUNNING_ON_VALGRIND) { |
dcadc967 | 1108 | pid_t pid; |
8869a0b4 | 1109 | pid = raw_clone(SIGCHLD); |
dcadc967 EV |
1110 | if (pid < 0) |
1111 | log_emergency_errno(errno, "Failed to fork off valgrind helper: %m"); | |
1112 | else if (pid == 0) | |
1113 | exit(EXIT_SUCCESS); | |
1114 | else { | |
1115 | log_info("Spawned valgrind helper as PID "PID_FMT".", pid); | |
1116 | (void) wait_for_terminate(pid, NULL); | |
1117 | } | |
1118 | } | |
1119 | #endif | |
1120 | } | |
1121 | ||
93bab288 | 1122 | int pid_compare_func(const pid_t *a, const pid_t *b) { |
291d565a | 1123 | /* Suitable for usage in qsort() */ |
93bab288 | 1124 | return CMP(*a, *b); |
291d565a LP |
1125 | } |
1126 | ||
7f452159 LP |
1127 | int ioprio_parse_priority(const char *s, int *ret) { |
1128 | int i, r; | |
1129 | ||
1130 | assert(s); | |
1131 | assert(ret); | |
1132 | ||
1133 | r = safe_atoi(s, &i); | |
1134 | if (r < 0) | |
1135 | return r; | |
1136 | ||
1137 | if (!ioprio_priority_is_valid(i)) | |
1138 | return -EINVAL; | |
1139 | ||
1140 | *ret = i; | |
1141 | return 0; | |
1142 | } | |
1143 | ||
5c30a6d2 LP |
1144 | /* The cached PID, possible values: |
1145 | * | |
1146 | * == UNSET [0] → cache not initialized yet | |
1147 | * == BUSY [-1] → some thread is initializing it at the moment | |
1148 | * any other → the cached PID | |
1149 | */ | |
1150 | ||
1151 | #define CACHED_PID_UNSET ((pid_t) 0) | |
1152 | #define CACHED_PID_BUSY ((pid_t) -1) | |
1153 | ||
1154 | static pid_t cached_pid = CACHED_PID_UNSET; | |
1155 | ||
799a960d | 1156 | void reset_cached_pid(void) { |
5c30a6d2 LP |
1157 | /* Invoked in the child after a fork(), i.e. at the first moment the PID changed */ |
1158 | cached_pid = CACHED_PID_UNSET; | |
1159 | } | |
1160 | ||
1161 | /* We use glibc __register_atfork() + __dso_handle directly here, as they are not included in the glibc | |
1162 | * headers. __register_atfork() is mostly equivalent to pthread_atfork(), but doesn't require us to link against | |
1163 | * libpthread, as it is part of glibc anyway. */ | |
2bb8d8d9 | 1164 | extern int __register_atfork(void (*prepare) (void), void (*parent) (void), void (*child) (void), void *dso_handle); |
d34dae18 | 1165 | extern void* __dso_handle _weak_; |
5c30a6d2 LP |
1166 | |
1167 | pid_t getpid_cached(void) { | |
5d71bac3 | 1168 | static bool installed = false; |
5c30a6d2 LP |
1169 | pid_t current_value; |
1170 | ||
1171 | /* getpid_cached() is much like getpid(), but caches the value in local memory, to avoid having to invoke a | |
1172 | * system call each time. This restores glibc behaviour from before 2.24, when getpid() was unconditionally | |
1173 | * cached. Starting with 2.24 getpid() started to become prohibitively expensive when used for detecting when | |
1174 | * objects were used across fork()s. With this caching the old behaviour is somewhat restored. | |
1175 | * | |
1176 | * https://bugzilla.redhat.com/show_bug.cgi?id=1443976 | |
a4041e4f | 1177 | * https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=c579f48edba88380635ab98cb612030e3ed8691e |
5c30a6d2 LP |
1178 | */ |
1179 | ||
1180 | current_value = __sync_val_compare_and_swap(&cached_pid, CACHED_PID_UNSET, CACHED_PID_BUSY); | |
1181 | ||
1182 | switch (current_value) { | |
1183 | ||
1184 | case CACHED_PID_UNSET: { /* Not initialized yet, then do so now */ | |
1185 | pid_t new_pid; | |
1186 | ||
996def17 | 1187 | new_pid = raw_getpid(); |
5c30a6d2 | 1188 | |
5d71bac3 LP |
1189 | if (!installed) { |
1190 | /* __register_atfork() either returns 0 or -ENOMEM, in its glibc implementation. Since it's | |
1191 | * only half-documented (glibc doesn't document it but LSB does — though only superficially) | |
1192 | * we'll check for errors only in the most generic fashion possible. */ | |
1193 | ||
1194 | if (__register_atfork(NULL, NULL, reset_cached_pid, __dso_handle) != 0) { | |
1195 | /* OOM? Let's try again later */ | |
1196 | cached_pid = CACHED_PID_UNSET; | |
1197 | return new_pid; | |
1198 | } | |
1199 | ||
1200 | installed = true; | |
5c30a6d2 LP |
1201 | } |
1202 | ||
1203 | cached_pid = new_pid; | |
1204 | return new_pid; | |
1205 | } | |
1206 | ||
1207 | case CACHED_PID_BUSY: /* Somebody else is currently initializing */ | |
996def17 | 1208 | return raw_getpid(); |
5c30a6d2 LP |
1209 | |
1210 | default: /* Properly initialized */ | |
1211 | return current_value; | |
1212 | } | |
1213 | } | |
1214 | ||
fba868fa LP |
1215 | int must_be_root(void) { |
1216 | ||
1217 | if (geteuid() == 0) | |
1218 | return 0; | |
1219 | ||
baaa35ad | 1220 | return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Need to be root."); |
fba868fa LP |
1221 | } |
1222 | ||
4c253ed1 LP |
1223 | int safe_fork_full( |
1224 | const char *name, | |
1225 | const int except_fds[], | |
1226 | size_t n_except_fds, | |
1227 | ForkFlags flags, | |
1228 | pid_t *ret_pid) { | |
1229 | ||
1230 | pid_t original_pid, pid; | |
1f5d1e02 | 1231 | sigset_t saved_ss, ss; |
a41f6217 | 1232 | bool block_signals = false; |
b6e1fff1 | 1233 | int prio, r; |
4c253ed1 LP |
1234 | |
1235 | /* A wrapper around fork(), that does a couple of important initializations in addition to mere forking. Always | |
1236 | * returns the child's PID in *ret_pid. Returns == 0 in the child, and > 0 in the parent. */ | |
1237 | ||
b6e1fff1 LP |
1238 | prio = flags & FORK_LOG ? LOG_ERR : LOG_DEBUG; |
1239 | ||
4c253ed1 LP |
1240 | original_pid = getpid_cached(); |
1241 | ||
1f5d1e02 | 1242 | if (flags & (FORK_RESET_SIGNALS|FORK_DEATHSIG)) { |
1f5d1e02 LP |
1243 | /* We temporarily block all signals, so that the new child has them blocked initially. This way, we can |
1244 | * be sure that SIGTERMs are not lost we might send to the child. */ | |
4c253ed1 | 1245 | |
cd2a429e | 1246 | assert_se(sigfillset(&ss) >= 0); |
1f5d1e02 LP |
1247 | block_signals = true; |
1248 | ||
1249 | } else if (flags & FORK_WAIT) { | |
1f5d1e02 LP |
1250 | /* Let's block SIGCHLD at least, so that we can safely watch for the child process */ |
1251 | ||
cd2a429e ZJS |
1252 | assert_se(sigemptyset(&ss) >= 0); |
1253 | assert_se(sigaddset(&ss, SIGCHLD) >= 0); | |
1f5d1e02 | 1254 | block_signals = true; |
4c253ed1 LP |
1255 | } |
1256 | ||
1f5d1e02 LP |
1257 | if (block_signals) |
1258 | if (sigprocmask(SIG_SETMASK, &ss, &saved_ss) < 0) | |
1259 | return log_full_errno(prio, errno, "Failed to set signal mask: %m"); | |
1260 | ||
be39f6ee LP |
1261 | if (flags & FORK_NEW_MOUNTNS) |
1262 | pid = raw_clone(SIGCHLD|CLONE_NEWNS); | |
1263 | else | |
1264 | pid = fork(); | |
4c253ed1 LP |
1265 | if (pid < 0) { |
1266 | r = -errno; | |
1267 | ||
1268 | if (block_signals) /* undo what we did above */ | |
1269 | (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL); | |
1270 | ||
b6e1fff1 | 1271 | return log_full_errno(prio, r, "Failed to fork: %m"); |
4c253ed1 LP |
1272 | } |
1273 | if (pid > 0) { | |
1274 | /* We are in the parent process */ | |
1275 | ||
1f5d1e02 LP |
1276 | log_debug("Successfully forked off '%s' as PID " PID_FMT ".", strna(name), pid); |
1277 | ||
1278 | if (flags & FORK_WAIT) { | |
1279 | r = wait_for_terminate_and_check(name, pid, (flags & FORK_LOG ? WAIT_LOG : 0)); | |
1280 | if (r < 0) | |
1281 | return r; | |
1282 | if (r != EXIT_SUCCESS) /* exit status > 0 should be treated as failure, too */ | |
1283 | return -EPROTO; | |
1284 | } | |
1285 | ||
4c253ed1 LP |
1286 | if (block_signals) /* undo what we did above */ |
1287 | (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL); | |
1288 | ||
4c253ed1 LP |
1289 | if (ret_pid) |
1290 | *ret_pid = pid; | |
1291 | ||
1292 | return 1; | |
1293 | } | |
1294 | ||
1295 | /* We are in the child process */ | |
1296 | ||
1297 | if (flags & FORK_REOPEN_LOG) { | |
1298 | /* Close the logs if requested, before we log anything. And make sure we reopen it if needed. */ | |
1299 | log_close(); | |
1300 | log_set_open_when_needed(true); | |
1301 | } | |
1302 | ||
1303 | if (name) { | |
1304 | r = rename_process(name); | |
1305 | if (r < 0) | |
b6e1fff1 LP |
1306 | log_full_errno(flags & FORK_LOG ? LOG_WARNING : LOG_DEBUG, |
1307 | r, "Failed to rename process, ignoring: %m"); | |
4c253ed1 LP |
1308 | } |
1309 | ||
1310 | if (flags & FORK_DEATHSIG) | |
1311 | if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0) { | |
b6e1fff1 | 1312 | log_full_errno(prio, errno, "Failed to set death signal: %m"); |
4c253ed1 LP |
1313 | _exit(EXIT_FAILURE); |
1314 | } | |
1315 | ||
1316 | if (flags & FORK_RESET_SIGNALS) { | |
1317 | r = reset_all_signal_handlers(); | |
1318 | if (r < 0) { | |
b6e1fff1 | 1319 | log_full_errno(prio, r, "Failed to reset signal handlers: %m"); |
4c253ed1 LP |
1320 | _exit(EXIT_FAILURE); |
1321 | } | |
1322 | ||
1323 | /* This implicitly undoes the signal mask stuff we did before the fork()ing above */ | |
1324 | r = reset_signal_mask(); | |
1325 | if (r < 0) { | |
b6e1fff1 | 1326 | log_full_errno(prio, r, "Failed to reset signal mask: %m"); |
4c253ed1 LP |
1327 | _exit(EXIT_FAILURE); |
1328 | } | |
1329 | } else if (block_signals) { /* undo what we did above */ | |
1330 | if (sigprocmask(SIG_SETMASK, &saved_ss, NULL) < 0) { | |
b6e1fff1 | 1331 | log_full_errno(prio, errno, "Failed to restore signal mask: %m"); |
4c253ed1 LP |
1332 | _exit(EXIT_FAILURE); |
1333 | } | |
1334 | } | |
1335 | ||
1336 | if (flags & FORK_DEATHSIG) { | |
7ddc2dc5 | 1337 | pid_t ppid; |
4c253ed1 LP |
1338 | /* Let's see if the parent PID is still the one we started from? If not, then the parent |
1339 | * already died by the time we set PR_SET_PDEATHSIG, hence let's emulate the effect */ | |
1340 | ||
7ddc2dc5 SL |
1341 | ppid = getppid(); |
1342 | if (ppid == 0) | |
1343 | /* Parent is in a differn't PID namespace. */; | |
1344 | else if (ppid != original_pid) { | |
4c253ed1 LP |
1345 | log_debug("Parent died early, raising SIGTERM."); |
1346 | (void) raise(SIGTERM); | |
1347 | _exit(EXIT_FAILURE); | |
1348 | } | |
1349 | } | |
1350 | ||
d94a24ca | 1351 | if (FLAGS_SET(flags, FORK_NEW_MOUNTNS | FORK_MOUNTNS_SLAVE)) { |
e2047ba9 LP |
1352 | |
1353 | /* Optionally, make sure we never propagate mounts to the host. */ | |
1354 | ||
1355 | if (mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL) < 0) { | |
1356 | log_full_errno(prio, errno, "Failed to remount root directory as MS_SLAVE: %m"); | |
1357 | _exit(EXIT_FAILURE); | |
1358 | } | |
1359 | } | |
1360 | ||
4c253ed1 LP |
1361 | if (flags & FORK_CLOSE_ALL_FDS) { |
1362 | /* Close the logs here in case it got reopened above, as close_all_fds() would close them for us */ | |
1363 | log_close(); | |
1364 | ||
1365 | r = close_all_fds(except_fds, n_except_fds); | |
1366 | if (r < 0) { | |
b6e1fff1 | 1367 | log_full_errno(prio, r, "Failed to close all file descriptors: %m"); |
4c253ed1 LP |
1368 | _exit(EXIT_FAILURE); |
1369 | } | |
1370 | } | |
1371 | ||
1372 | /* When we were asked to reopen the logs, do so again now */ | |
1373 | if (flags & FORK_REOPEN_LOG) { | |
1374 | log_open(); | |
1375 | log_set_open_when_needed(false); | |
1376 | } | |
1377 | ||
1378 | if (flags & FORK_NULL_STDIO) { | |
1379 | r = make_null_stdio(); | |
1380 | if (r < 0) { | |
b6e1fff1 | 1381 | log_full_errno(prio, r, "Failed to connect stdin/stdout to /dev/null: %m"); |
4c253ed1 LP |
1382 | _exit(EXIT_FAILURE); |
1383 | } | |
1384 | } | |
1385 | ||
909106eb LP |
1386 | if (flags & FORK_RLIMIT_NOFILE_SAFE) { |
1387 | r = rlimit_nofile_safe(); | |
1388 | if (r < 0) { | |
1389 | log_full_errno(prio, r, "Failed to lower RLIMIT_NOFILE's soft limit to 1K: %m"); | |
1390 | _exit(EXIT_FAILURE); | |
1391 | } | |
1392 | } | |
1393 | ||
4c253ed1 LP |
1394 | if (ret_pid) |
1395 | *ret_pid = getpid_cached(); | |
1396 | ||
1397 | return 0; | |
1398 | } | |
1399 | ||
27096982 LP |
1400 | int namespace_fork( |
1401 | const char *outer_name, | |
1402 | const char *inner_name, | |
1403 | const int except_fds[], | |
1404 | size_t n_except_fds, | |
1405 | ForkFlags flags, | |
1406 | int pidns_fd, | |
1407 | int mntns_fd, | |
1408 | int netns_fd, | |
1409 | int userns_fd, | |
1410 | int root_fd, | |
1411 | pid_t *ret_pid) { | |
1412 | ||
1413 | int r; | |
1414 | ||
1415 | /* This is much like safe_fork(), but forks twice, and joins the specified namespaces in the middle | |
1416 | * process. This ensures that we are fully a member of the destination namespace, with pidns an all, so that | |
1417 | * /proc/self/fd works correctly. */ | |
1418 | ||
1419 | r = safe_fork_full(outer_name, except_fds, n_except_fds, (flags|FORK_DEATHSIG) & ~(FORK_REOPEN_LOG|FORK_NEW_MOUNTNS|FORK_MOUNTNS_SLAVE), ret_pid); | |
1420 | if (r < 0) | |
1421 | return r; | |
1422 | if (r == 0) { | |
1423 | pid_t pid; | |
1424 | ||
1425 | /* Child */ | |
1426 | ||
1427 | r = namespace_enter(pidns_fd, mntns_fd, netns_fd, userns_fd, root_fd); | |
1428 | if (r < 0) { | |
1429 | log_full_errno(FLAGS_SET(flags, FORK_LOG) ? LOG_ERR : LOG_DEBUG, r, "Failed to join namespace: %m"); | |
1430 | _exit(EXIT_FAILURE); | |
1431 | } | |
1432 | ||
1433 | /* We mask a few flags here that either make no sense for the grandchild, or that we don't have to do again */ | |
1434 | r = safe_fork_full(inner_name, except_fds, n_except_fds, flags & ~(FORK_WAIT|FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_NULL_STDIO), &pid); | |
1435 | if (r < 0) | |
1436 | _exit(EXIT_FAILURE); | |
1437 | if (r == 0) { | |
1438 | /* Child */ | |
1439 | if (ret_pid) | |
1440 | *ret_pid = pid; | |
1441 | return 0; | |
1442 | } | |
1443 | ||
1444 | r = wait_for_terminate_and_check(inner_name, pid, FLAGS_SET(flags, FORK_LOG) ? WAIT_LOG : 0); | |
1445 | if (r < 0) | |
1446 | _exit(EXIT_FAILURE); | |
1447 | ||
1448 | _exit(r); | |
1449 | } | |
1450 | ||
1451 | return 1; | |
1452 | } | |
1453 | ||
da6053d0 | 1454 | int fork_agent(const char *name, const int except[], size_t n_except, pid_t *ret_pid, const char *path, ...) { |
78752f2e | 1455 | bool stdout_is_tty, stderr_is_tty; |
da6053d0 | 1456 | size_t n, i; |
78752f2e LP |
1457 | va_list ap; |
1458 | char **l; | |
1459 | int r; | |
1460 | ||
1461 | assert(path); | |
1462 | ||
1463 | /* Spawns a temporary TTY agent, making sure it goes away when we go away */ | |
1464 | ||
1465 | r = safe_fork_full(name, except, n_except, FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_CLOSE_ALL_FDS, ret_pid); | |
1466 | if (r < 0) | |
1467 | return r; | |
1468 | if (r > 0) | |
1469 | return 0; | |
1470 | ||
1471 | /* In the child: */ | |
1472 | ||
1473 | stdout_is_tty = isatty(STDOUT_FILENO); | |
1474 | stderr_is_tty = isatty(STDERR_FILENO); | |
1475 | ||
1476 | if (!stdout_is_tty || !stderr_is_tty) { | |
1477 | int fd; | |
1478 | ||
1479 | /* Detach from stdout/stderr. and reopen | |
1480 | * /dev/tty for them. This is important to | |
1481 | * ensure that when systemctl is started via | |
1482 | * popen() or a similar call that expects to | |
1483 | * read EOF we actually do generate EOF and | |
1484 | * not delay this indefinitely by because we | |
1485 | * keep an unused copy of stdin around. */ | |
1486 | fd = open("/dev/tty", O_WRONLY); | |
1487 | if (fd < 0) { | |
1488 | log_error_errno(errno, "Failed to open /dev/tty: %m"); | |
1489 | _exit(EXIT_FAILURE); | |
1490 | } | |
1491 | ||
1492 | if (!stdout_is_tty && dup2(fd, STDOUT_FILENO) < 0) { | |
1493 | log_error_errno(errno, "Failed to dup2 /dev/tty: %m"); | |
1494 | _exit(EXIT_FAILURE); | |
1495 | } | |
1496 | ||
1497 | if (!stderr_is_tty && dup2(fd, STDERR_FILENO) < 0) { | |
1498 | log_error_errno(errno, "Failed to dup2 /dev/tty: %m"); | |
1499 | _exit(EXIT_FAILURE); | |
1500 | } | |
1501 | ||
e7685a77 | 1502 | safe_close_above_stdio(fd); |
78752f2e LP |
1503 | } |
1504 | ||
595225af LP |
1505 | (void) rlimit_nofile_safe(); |
1506 | ||
78752f2e LP |
1507 | /* Count arguments */ |
1508 | va_start(ap, path); | |
1509 | for (n = 0; va_arg(ap, char*); n++) | |
1510 | ; | |
1511 | va_end(ap); | |
1512 | ||
1513 | /* Allocate strv */ | |
cf409d15 | 1514 | l = newa(char*, n + 1); |
78752f2e LP |
1515 | |
1516 | /* Fill in arguments */ | |
1517 | va_start(ap, path); | |
1518 | for (i = 0; i <= n; i++) | |
1519 | l[i] = va_arg(ap, char*); | |
1520 | va_end(ap); | |
1521 | ||
1522 | execv(path, l); | |
1523 | _exit(EXIT_FAILURE); | |
1524 | } | |
1525 | ||
9f8168eb LP |
1526 | int set_oom_score_adjust(int value) { |
1527 | char t[DECIMAL_STR_MAX(int)]; | |
1528 | ||
1529 | sprintf(t, "%i", value); | |
1530 | ||
1531 | return write_string_file("/proc/self/oom_score_adj", t, | |
1532 | WRITE_STRING_FILE_VERIFY_ON_FAILURE|WRITE_STRING_FILE_DISABLE_BUFFER); | |
1533 | } | |
1534 | ||
20ee849d LP |
1535 | int cpus_in_affinity_mask(void) { |
1536 | size_t n = 16; | |
1537 | int r; | |
1538 | ||
1539 | for (;;) { | |
1540 | cpu_set_t *c; | |
1541 | ||
1542 | c = CPU_ALLOC(n); | |
1543 | if (!c) | |
1544 | return -ENOMEM; | |
1545 | ||
1546 | if (sched_getaffinity(0, CPU_ALLOC_SIZE(n), c) >= 0) { | |
1547 | int k; | |
1548 | ||
1549 | k = CPU_COUNT_S(CPU_ALLOC_SIZE(n), c); | |
1550 | CPU_FREE(c); | |
1551 | ||
1552 | if (k <= 0) | |
1553 | return -EINVAL; | |
1554 | ||
1555 | return k; | |
1556 | } | |
1557 | ||
1558 | r = -errno; | |
1559 | CPU_FREE(c); | |
1560 | ||
1561 | if (r != -EINVAL) | |
1562 | return r; | |
1563 | if (n > SIZE_MAX/2) | |
1564 | return -ENOMEM; | |
1565 | n *= 2; | |
1566 | } | |
1567 | } | |
1568 | ||
7b3e062c LP |
1569 | static const char *const ioprio_class_table[] = { |
1570 | [IOPRIO_CLASS_NONE] = "none", | |
1571 | [IOPRIO_CLASS_RT] = "realtime", | |
1572 | [IOPRIO_CLASS_BE] = "best-effort", | |
1573 | [IOPRIO_CLASS_IDLE] = "idle" | |
1574 | }; | |
1575 | ||
10062bbc | 1576 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ioprio_class, int, IOPRIO_N_CLASSES); |
7b3e062c LP |
1577 | |
1578 | static const char *const sigchld_code_table[] = { | |
1579 | [CLD_EXITED] = "exited", | |
1580 | [CLD_KILLED] = "killed", | |
1581 | [CLD_DUMPED] = "dumped", | |
1582 | [CLD_TRAPPED] = "trapped", | |
1583 | [CLD_STOPPED] = "stopped", | |
1584 | [CLD_CONTINUED] = "continued", | |
1585 | }; | |
1586 | ||
1587 | DEFINE_STRING_TABLE_LOOKUP(sigchld_code, int); | |
1588 | ||
1589 | static const char* const sched_policy_table[] = { | |
1590 | [SCHED_OTHER] = "other", | |
1591 | [SCHED_BATCH] = "batch", | |
1592 | [SCHED_IDLE] = "idle", | |
1593 | [SCHED_FIFO] = "fifo", | |
1594 | [SCHED_RR] = "rr" | |
1595 | }; | |
1596 | ||
1597 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(sched_policy, int, INT_MAX); |