]>
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> | |
35bbbf85 | 11 | #include <stdio_ext.h> |
11c3a366 | 12 | #include <stdlib.h> |
4f5dd394 | 13 | #include <string.h> |
9bfaffd5 | 14 | #include <sys/mman.h> |
e2047ba9 | 15 | #include <sys/mount.h> |
7b3e062c | 16 | #include <sys/personality.h> |
405f8907 | 17 | #include <sys/prctl.h> |
4f5dd394 LP |
18 | #include <sys/types.h> |
19 | #include <sys/wait.h> | |
11c3a366 | 20 | #include <syslog.h> |
4f5dd394 | 21 | #include <unistd.h> |
349cc4a5 | 22 | #if HAVE_VALGRIND_VALGRIND_H |
dcadc967 EV |
23 | #include <valgrind/valgrind.h> |
24 | #endif | |
0b452006 | 25 | |
b5efdb8a | 26 | #include "alloc-util.h" |
6e5f1b57 | 27 | #include "architecture.h" |
4f5dd394 | 28 | #include "escape.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 TA |
34 | #include "macro.h" |
35 | #include "missing.h" | |
93cc7779 | 36 | #include "process-util.h" |
8869a0b4 | 37 | #include "raw-clone.h" |
909106eb | 38 | #include "rlimit-util.h" |
93cc7779 | 39 | #include "signal-util.h" |
1359fffa | 40 | #include "stat-util.h" |
7b3e062c | 41 | #include "string-table.h" |
07630cea | 42 | #include "string-util.h" |
4c253ed1 | 43 | #include "terminal-util.h" |
b1d4f8e1 | 44 | #include "user-util.h" |
4f5dd394 | 45 | #include "util.h" |
0b452006 RC |
46 | |
47 | int get_process_state(pid_t pid) { | |
48 | const char *p; | |
49 | char state; | |
50 | int r; | |
51 | _cleanup_free_ char *line = NULL; | |
52 | ||
53 | assert(pid >= 0); | |
54 | ||
55 | p = procfs_file_alloca(pid, "stat"); | |
a644184a | 56 | |
0b452006 | 57 | r = read_one_line_file(p, &line); |
a644184a LP |
58 | if (r == -ENOENT) |
59 | return -ESRCH; | |
0b452006 RC |
60 | if (r < 0) |
61 | return r; | |
62 | ||
63 | p = strrchr(line, ')'); | |
64 | if (!p) | |
65 | return -EIO; | |
66 | ||
67 | p++; | |
68 | ||
69 | if (sscanf(p, " %c", &state) != 1) | |
70 | return -EIO; | |
71 | ||
72 | return (unsigned char) state; | |
73 | } | |
74 | ||
ce268825 LP |
75 | int get_process_comm(pid_t pid, char **ret) { |
76 | _cleanup_free_ char *escaped = NULL, *comm = NULL; | |
0b452006 RC |
77 | const char *p; |
78 | int r; | |
79 | ||
ce268825 | 80 | assert(ret); |
0b452006 RC |
81 | assert(pid >= 0); |
82 | ||
ce268825 LP |
83 | escaped = new(char, TASK_COMM_LEN); |
84 | if (!escaped) | |
85 | return -ENOMEM; | |
86 | ||
0b452006 RC |
87 | p = procfs_file_alloca(pid, "comm"); |
88 | ||
ce268825 | 89 | r = read_one_line_file(p, &comm); |
0b452006 RC |
90 | if (r == -ENOENT) |
91 | return -ESRCH; | |
ce268825 LP |
92 | if (r < 0) |
93 | return r; | |
94 | ||
95 | /* Escape unprintable characters, just in case, but don't grow the string beyond the underlying size */ | |
96 | cellescape(escaped, TASK_COMM_LEN, comm); | |
0b452006 | 97 | |
ce268825 LP |
98 | *ret = TAKE_PTR(escaped); |
99 | return 0; | |
0b452006 RC |
100 | } |
101 | ||
102 | int get_process_cmdline(pid_t pid, size_t max_length, bool comm_fallback, char **line) { | |
103 | _cleanup_fclose_ FILE *f = NULL; | |
ba4cd7e2 | 104 | bool space = false; |
eb1ec489 MS |
105 | char *k; |
106 | _cleanup_free_ char *ans = NULL; | |
0b452006 RC |
107 | const char *p; |
108 | int c; | |
109 | ||
110 | assert(line); | |
111 | assert(pid >= 0); | |
112 | ||
69281c49 LP |
113 | /* Retrieves a process' command line. Replaces unprintable characters while doing so by whitespace (coalescing |
114 | * multiple sequential ones into one). If max_length is != 0 will return a string of the specified size at most | |
115 | * (the trailing NUL byte does count towards the length here!), abbreviated with a "..." ellipsis. If | |
116 | * comm_fallback is true and the process has no command line set (the case for kernel threads), or has a | |
117 | * command line that resolves to the empty string will return the "comm" name of the process instead. | |
118 | * | |
119 | * Returns -ESRCH if the process doesn't exist, and -ENOENT if the process has no command line (and | |
c0534780 | 120 | * comm_fallback is false). Returns 0 and sets *line otherwise. */ |
69281c49 | 121 | |
0b452006 RC |
122 | p = procfs_file_alloca(pid, "cmdline"); |
123 | ||
124 | f = fopen(p, "re"); | |
a644184a LP |
125 | if (!f) { |
126 | if (errno == ENOENT) | |
127 | return -ESRCH; | |
0b452006 | 128 | return -errno; |
a644184a | 129 | } |
0b452006 | 130 | |
35bbbf85 LP |
131 | (void) __fsetlocking(f, FSETLOCKING_BYCALLER); |
132 | ||
2d5d2e0c ZJS |
133 | if (max_length == 0) { |
134 | /* This is supposed to be a safety guard against runaway command lines. */ | |
135 | long l = sysconf(_SC_ARG_MAX); | |
136 | assert(l > 0); | |
137 | max_length = l; | |
138 | } | |
139 | ||
69281c49 LP |
140 | if (max_length == 1) { |
141 | ||
142 | /* If there's only room for one byte, return the empty string */ | |
c0534780 ZJS |
143 | ans = new0(char, 1); |
144 | if (!ans) | |
69281c49 LP |
145 | return -ENOMEM; |
146 | ||
eb1ec489 | 147 | *line = TAKE_PTR(ans); |
69281c49 LP |
148 | return 0; |
149 | ||
0b452006 | 150 | } else { |
69281c49 | 151 | bool dotdotdot = false; |
0b452006 RC |
152 | size_t left; |
153 | ||
c0534780 ZJS |
154 | ans = new(char, max_length); |
155 | if (!ans) | |
0b452006 RC |
156 | return -ENOMEM; |
157 | ||
c0534780 | 158 | k = ans; |
0b452006 RC |
159 | left = max_length; |
160 | while ((c = getc(f)) != EOF) { | |
161 | ||
162 | if (isprint(c)) { | |
69281c49 | 163 | |
0b452006 | 164 | if (space) { |
69281c49 LP |
165 | if (left <= 2) { |
166 | dotdotdot = true; | |
0b452006 | 167 | break; |
69281c49 | 168 | } |
0b452006 RC |
169 | |
170 | *(k++) = ' '; | |
171 | left--; | |
172 | space = false; | |
173 | } | |
174 | ||
69281c49 LP |
175 | if (left <= 1) { |
176 | dotdotdot = true; | |
0b452006 | 177 | break; |
69281c49 | 178 | } |
0b452006 RC |
179 | |
180 | *(k++) = (char) c; | |
181 | left--; | |
c0534780 | 182 | } else if (k > ans) |
0b452006 RC |
183 | space = true; |
184 | } | |
185 | ||
69281c49 LP |
186 | if (dotdotdot) { |
187 | if (max_length <= 4) { | |
c0534780 | 188 | k = ans; |
69281c49 LP |
189 | left = max_length; |
190 | } else { | |
c0534780 | 191 | k = ans + max_length - 4; |
69281c49 LP |
192 | left = 4; |
193 | ||
194 | /* Eat up final spaces */ | |
c0534780 | 195 | while (k > ans && isspace(k[-1])) { |
69281c49 LP |
196 | k--; |
197 | left++; | |
198 | } | |
199 | } | |
200 | ||
201 | strncpy(k, "...", left-1); | |
b09df4e2 | 202 | k[left-1] = 0; |
0b452006 RC |
203 | } else |
204 | *k = 0; | |
205 | } | |
206 | ||
207 | /* Kernel threads have no argv[] */ | |
c0534780 | 208 | if (isempty(ans)) { |
0b452006 RC |
209 | _cleanup_free_ char *t = NULL; |
210 | int h; | |
211 | ||
eb1ec489 | 212 | ans = mfree(ans); |
0b452006 RC |
213 | |
214 | if (!comm_fallback) | |
215 | return -ENOENT; | |
216 | ||
217 | h = get_process_comm(pid, &t); | |
218 | if (h < 0) | |
219 | return h; | |
220 | ||
2d5d2e0c | 221 | size_t l = strlen(t); |
69281c49 | 222 | |
2d5d2e0c ZJS |
223 | if (l + 3 <= max_length) { |
224 | ans = strjoin("[", t, "]"); | |
225 | if (!ans) | |
226 | return -ENOMEM; | |
69281c49 | 227 | |
2d5d2e0c ZJS |
228 | } else if (max_length <= 6) { |
229 | ans = new(char, max_length); | |
230 | if (!ans) | |
231 | return -ENOMEM; | |
69281c49 | 232 | |
2d5d2e0c ZJS |
233 | memcpy(ans, "[...]", max_length-1); |
234 | ans[max_length-1] = 0; | |
235 | } else { | |
236 | t[max_length - 6] = 0; | |
69281c49 | 237 | |
2d5d2e0c ZJS |
238 | /* Chop off final spaces */ |
239 | delete_trailing_chars(t, WHITESPACE); | |
69281c49 | 240 | |
2d5d2e0c ZJS |
241 | ans = strjoin("[", t, "...]"); |
242 | if (!ans) | |
243 | return -ENOMEM; | |
69281c49 | 244 | } |
eb1ec489 MS |
245 | |
246 | *line = TAKE_PTR(ans); | |
247 | return 0; | |
0b452006 RC |
248 | } |
249 | ||
eb1ec489 MS |
250 | k = realloc(ans, strlen(ans) + 1); |
251 | if (!k) | |
252 | return -ENOMEM; | |
253 | ||
254 | ans = NULL; | |
255 | *line = k; | |
256 | ||
0b452006 RC |
257 | return 0; |
258 | } | |
259 | ||
9bfaffd5 LP |
260 | int rename_process(const char name[]) { |
261 | static size_t mm_size = 0; | |
262 | static char *mm = NULL; | |
263 | bool truncated = false; | |
264 | size_t l; | |
265 | ||
266 | /* This is a like a poor man's setproctitle(). It changes the comm field, argv[0], and also the glibc's | |
267 | * internally used name of the process. For the first one a limit of 16 chars applies; to the second one in | |
268 | * many cases one of 10 (i.e. length of "/sbin/init") — however if we have CAP_SYS_RESOURCES it is unbounded; | |
269 | * to the third one 7 (i.e. the length of "systemd". If you pass a longer string it will likely be | |
270 | * truncated. | |
271 | * | |
272 | * Returns 0 if a name was set but truncated, > 0 if it was set but not truncated. */ | |
273 | ||
274 | if (isempty(name)) | |
275 | return -EINVAL; /* let's not confuse users unnecessarily with an empty name */ | |
405f8907 | 276 | |
1096bb8a LP |
277 | if (!is_main_thread()) |
278 | return -EPERM; /* Let's not allow setting the process name from other threads than the main one, as we | |
279 | * cache things without locking, and we make assumptions that PR_SET_NAME sets the | |
280 | * process name that isn't correct on any other threads */ | |
281 | ||
9bfaffd5 | 282 | l = strlen(name); |
405f8907 | 283 | |
5a8af747 LP |
284 | /* First step, change the comm field. The main thread's comm is identical to the process comm. This means we |
285 | * can use PR_SET_NAME, which sets the thread name for the calling thread. */ | |
286 | if (prctl(PR_SET_NAME, name) < 0) | |
287 | log_debug_errno(errno, "PR_SET_NAME failed: %m"); | |
92f14395 | 288 | if (l >= TASK_COMM_LEN) /* Linux process names can be 15 chars at max */ |
9bfaffd5 LP |
289 | truncated = true; |
290 | ||
291 | /* Second step, change glibc's ID of the process name. */ | |
292 | if (program_invocation_name) { | |
293 | size_t k; | |
294 | ||
295 | k = strlen(program_invocation_name); | |
296 | strncpy(program_invocation_name, name, k); | |
297 | if (l > k) | |
298 | truncated = true; | |
299 | } | |
300 | ||
301 | /* Third step, completely replace the argv[] array the kernel maintains for us. This requires privileges, but | |
302 | * has the advantage that the argv[] array is exactly what we want it to be, and not filled up with zeros at | |
13e785f7 | 303 | * the end. This is the best option for changing /proc/self/cmdline. */ |
01f989c6 JW |
304 | |
305 | /* Let's not bother with this if we don't have euid == 0. Strictly speaking we should check for the | |
306 | * CAP_SYS_RESOURCE capability which is independent of the euid. In our own code the capability generally is | |
307 | * present only for euid == 0, hence let's use this as quick bypass check, to avoid calling mmap() if | |
308 | * PR_SET_MM_ARG_{START,END} fails with EPERM later on anyway. After all geteuid() is dead cheap to call, but | |
309 | * mmap() is not. */ | |
310 | if (geteuid() != 0) | |
311 | log_debug("Skipping PR_SET_MM, as we don't have privileges."); | |
312 | else if (mm_size < l+1) { | |
9bfaffd5 LP |
313 | size_t nn_size; |
314 | char *nn; | |
315 | ||
9bfaffd5 LP |
316 | nn_size = PAGE_ALIGN(l+1); |
317 | nn = mmap(NULL, nn_size, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); | |
318 | if (nn == MAP_FAILED) { | |
319 | log_debug_errno(errno, "mmap() failed: %m"); | |
320 | goto use_saved_argv; | |
321 | } | |
405f8907 | 322 | |
9bfaffd5 LP |
323 | strncpy(nn, name, nn_size); |
324 | ||
325 | /* Now, let's tell the kernel about this new memory */ | |
326 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_START, (unsigned long) nn, 0, 0) < 0) { | |
14ee72b7 FS |
327 | /* HACK: prctl() API is kind of dumb on this point. The existing end address may already be |
328 | * below the desired start address, in which case the kernel may have kicked this back due | |
329 | * to a range-check failure (see linux/kernel/sys.c:validate_prctl_map() to see this in | |
330 | * action). The proper solution would be to have a prctl() API that could set both start+end | |
331 | * simultaneously, or at least let us query the existing address to anticipate this condition | |
332 | * and respond accordingly. For now, we can only guess at the cause of this failure and try | |
333 | * a workaround--which will briefly expand the arg space to something potentially huge before | |
334 | * resizing it to what we want. */ | |
335 | log_debug_errno(errno, "PR_SET_MM_ARG_START failed, attempting PR_SET_MM_ARG_END hack: %m"); | |
336 | ||
337 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) nn + l + 1, 0, 0) < 0) { | |
338 | log_debug_errno(errno, "PR_SET_MM_ARG_END hack failed, proceeding without: %m"); | |
339 | (void) munmap(nn, nn_size); | |
340 | goto use_saved_argv; | |
341 | } | |
9bfaffd5 | 342 | |
14ee72b7 FS |
343 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_START, (unsigned long) nn, 0, 0) < 0) { |
344 | log_debug_errno(errno, "PR_SET_MM_ARG_START still failed, proceeding without: %m"); | |
345 | goto use_saved_argv; | |
346 | } | |
347 | } else { | |
348 | /* And update the end pointer to the new end, too. If this fails, we don't really know what | |
349 | * to do, it's pretty unlikely that we can rollback, hence we'll just accept the failure, | |
350 | * and continue. */ | |
351 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) nn + l + 1, 0, 0) < 0) | |
352 | log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m"); | |
353 | } | |
9bfaffd5 LP |
354 | |
355 | if (mm) | |
356 | (void) munmap(mm, mm_size); | |
357 | ||
358 | mm = nn; | |
359 | mm_size = nn_size; | |
01f989c6 | 360 | } else { |
9bfaffd5 LP |
361 | strncpy(mm, name, mm_size); |
362 | ||
01f989c6 JW |
363 | /* Update the end pointer, continuing regardless of any failure. */ |
364 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) mm + l + 1, 0, 0) < 0) | |
365 | log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m"); | |
366 | } | |
367 | ||
9bfaffd5 LP |
368 | use_saved_argv: |
369 | /* Fourth step: in all cases we'll also update the original argv[], so that our own code gets it right too if | |
370 | * it still looks here */ | |
405f8907 LP |
371 | |
372 | if (saved_argc > 0) { | |
373 | int i; | |
374 | ||
9bfaffd5 LP |
375 | if (saved_argv[0]) { |
376 | size_t k; | |
377 | ||
378 | k = strlen(saved_argv[0]); | |
379 | strncpy(saved_argv[0], name, k); | |
380 | if (l > k) | |
381 | truncated = true; | |
382 | } | |
405f8907 LP |
383 | |
384 | for (i = 1; i < saved_argc; i++) { | |
385 | if (!saved_argv[i]) | |
386 | break; | |
387 | ||
388 | memzero(saved_argv[i], strlen(saved_argv[i])); | |
389 | } | |
390 | } | |
9bfaffd5 LP |
391 | |
392 | return !truncated; | |
405f8907 LP |
393 | } |
394 | ||
0b452006 | 395 | int is_kernel_thread(pid_t pid) { |
36b5119a LP |
396 | _cleanup_free_ char *line = NULL; |
397 | unsigned long long flags; | |
398 | size_t l, i; | |
0b452006 | 399 | const char *p; |
36b5119a LP |
400 | char *q; |
401 | int r; | |
0b452006 | 402 | |
4c701096 | 403 | if (IN_SET(pid, 0, 1) || pid == getpid_cached()) /* pid 1, and we ourselves certainly aren't a kernel thread */ |
0b452006 | 404 | return 0; |
36b5119a LP |
405 | if (!pid_is_valid(pid)) |
406 | return -EINVAL; | |
0b452006 | 407 | |
36b5119a LP |
408 | p = procfs_file_alloca(pid, "stat"); |
409 | r = read_one_line_file(p, &line); | |
410 | if (r == -ENOENT) | |
411 | return -ESRCH; | |
412 | if (r < 0) | |
413 | return r; | |
0b452006 | 414 | |
36b5119a LP |
415 | /* Skip past the comm field */ |
416 | q = strrchr(line, ')'); | |
417 | if (!q) | |
418 | return -EINVAL; | |
419 | q++; | |
420 | ||
421 | /* Skip 6 fields to reach the flags field */ | |
422 | for (i = 0; i < 6; i++) { | |
423 | l = strspn(q, WHITESPACE); | |
424 | if (l < 1) | |
425 | return -EINVAL; | |
426 | q += l; | |
427 | ||
428 | l = strcspn(q, WHITESPACE); | |
429 | if (l < 1) | |
430 | return -EINVAL; | |
431 | q += l; | |
a644184a | 432 | } |
0b452006 | 433 | |
f21f31b2 | 434 | /* Skip preceding whitespace */ |
36b5119a LP |
435 | l = strspn(q, WHITESPACE); |
436 | if (l < 1) | |
437 | return -EINVAL; | |
438 | q += l; | |
35bbbf85 | 439 | |
36b5119a LP |
440 | /* Truncate the rest */ |
441 | l = strcspn(q, WHITESPACE); | |
442 | if (l < 1) | |
443 | return -EINVAL; | |
444 | q[l] = 0; | |
0b452006 | 445 | |
36b5119a LP |
446 | r = safe_atollu(q, &flags); |
447 | if (r < 0) | |
448 | return r; | |
0b452006 | 449 | |
36b5119a | 450 | return !!(flags & PF_KTHREAD); |
0b452006 RC |
451 | } |
452 | ||
453 | int get_process_capeff(pid_t pid, char **capeff) { | |
454 | const char *p; | |
a644184a | 455 | int r; |
0b452006 RC |
456 | |
457 | assert(capeff); | |
458 | assert(pid >= 0); | |
459 | ||
460 | p = procfs_file_alloca(pid, "status"); | |
461 | ||
c4cd1d4d | 462 | r = get_proc_field(p, "CapEff", WHITESPACE, capeff); |
a644184a LP |
463 | if (r == -ENOENT) |
464 | return -ESRCH; | |
465 | ||
466 | return r; | |
0b452006 RC |
467 | } |
468 | ||
469 | static int get_process_link_contents(const char *proc_file, char **name) { | |
470 | int r; | |
471 | ||
472 | assert(proc_file); | |
473 | assert(name); | |
474 | ||
475 | r = readlink_malloc(proc_file, name); | |
a644184a LP |
476 | if (r == -ENOENT) |
477 | return -ESRCH; | |
0b452006 | 478 | if (r < 0) |
a644184a | 479 | return r; |
0b452006 RC |
480 | |
481 | return 0; | |
482 | } | |
483 | ||
484 | int get_process_exe(pid_t pid, char **name) { | |
485 | const char *p; | |
486 | char *d; | |
487 | int r; | |
488 | ||
489 | assert(pid >= 0); | |
490 | ||
491 | p = procfs_file_alloca(pid, "exe"); | |
492 | r = get_process_link_contents(p, name); | |
493 | if (r < 0) | |
494 | return r; | |
495 | ||
496 | d = endswith(*name, " (deleted)"); | |
497 | if (d) | |
498 | *d = '\0'; | |
499 | ||
500 | return 0; | |
501 | } | |
502 | ||
503 | static int get_process_id(pid_t pid, const char *field, uid_t *uid) { | |
504 | _cleanup_fclose_ FILE *f = NULL; | |
0b452006 | 505 | const char *p; |
7e7a16a0 | 506 | int r; |
0b452006 RC |
507 | |
508 | assert(field); | |
509 | assert(uid); | |
510 | ||
07b38ba5 | 511 | if (pid < 0) |
6f8cbcdb LP |
512 | return -EINVAL; |
513 | ||
0b452006 RC |
514 | p = procfs_file_alloca(pid, "status"); |
515 | f = fopen(p, "re"); | |
a644184a LP |
516 | if (!f) { |
517 | if (errno == ENOENT) | |
518 | return -ESRCH; | |
0b452006 | 519 | return -errno; |
a644184a | 520 | } |
0b452006 | 521 | |
35bbbf85 LP |
522 | (void) __fsetlocking(f, FSETLOCKING_BYCALLER); |
523 | ||
7e7a16a0 LP |
524 | for (;;) { |
525 | _cleanup_free_ char *line = NULL; | |
0b452006 RC |
526 | char *l; |
527 | ||
7e7a16a0 LP |
528 | r = read_line(f, LONG_LINE_MAX, &line); |
529 | if (r < 0) | |
530 | return r; | |
531 | if (r == 0) | |
532 | break; | |
533 | ||
0b452006 RC |
534 | l = strstrip(line); |
535 | ||
536 | if (startswith(l, field)) { | |
537 | l += strlen(field); | |
538 | l += strspn(l, WHITESPACE); | |
539 | ||
540 | l[strcspn(l, WHITESPACE)] = 0; | |
541 | ||
542 | return parse_uid(l, uid); | |
543 | } | |
544 | } | |
545 | ||
546 | return -EIO; | |
547 | } | |
548 | ||
549 | int get_process_uid(pid_t pid, uid_t *uid) { | |
6f8cbcdb LP |
550 | |
551 | if (pid == 0 || pid == getpid_cached()) { | |
552 | *uid = getuid(); | |
553 | return 0; | |
554 | } | |
555 | ||
0b452006 RC |
556 | return get_process_id(pid, "Uid:", uid); |
557 | } | |
558 | ||
559 | int get_process_gid(pid_t pid, gid_t *gid) { | |
6f8cbcdb LP |
560 | |
561 | if (pid == 0 || pid == getpid_cached()) { | |
562 | *gid = getgid(); | |
563 | return 0; | |
564 | } | |
565 | ||
0b452006 RC |
566 | assert_cc(sizeof(uid_t) == sizeof(gid_t)); |
567 | return get_process_id(pid, "Gid:", gid); | |
568 | } | |
569 | ||
570 | int get_process_cwd(pid_t pid, char **cwd) { | |
571 | const char *p; | |
572 | ||
573 | assert(pid >= 0); | |
574 | ||
575 | p = procfs_file_alloca(pid, "cwd"); | |
576 | ||
577 | return get_process_link_contents(p, cwd); | |
578 | } | |
579 | ||
580 | int get_process_root(pid_t pid, char **root) { | |
581 | const char *p; | |
582 | ||
583 | assert(pid >= 0); | |
584 | ||
585 | p = procfs_file_alloca(pid, "root"); | |
586 | ||
587 | return get_process_link_contents(p, root); | |
588 | } | |
589 | ||
2a7797e9 LP |
590 | #define ENVIRONMENT_BLOCK_MAX (5U*1024U*1024U) |
591 | ||
0b452006 RC |
592 | int get_process_environ(pid_t pid, char **env) { |
593 | _cleanup_fclose_ FILE *f = NULL; | |
594 | _cleanup_free_ char *outcome = NULL; | |
0b452006 | 595 | size_t allocated = 0, sz = 0; |
2a7797e9 LP |
596 | const char *p; |
597 | int r; | |
0b452006 RC |
598 | |
599 | assert(pid >= 0); | |
600 | assert(env); | |
601 | ||
602 | p = procfs_file_alloca(pid, "environ"); | |
603 | ||
604 | f = fopen(p, "re"); | |
a644184a LP |
605 | if (!f) { |
606 | if (errno == ENOENT) | |
607 | return -ESRCH; | |
0b452006 | 608 | return -errno; |
a644184a | 609 | } |
0b452006 | 610 | |
35bbbf85 LP |
611 | (void) __fsetlocking(f, FSETLOCKING_BYCALLER); |
612 | ||
2a7797e9 LP |
613 | for (;;) { |
614 | char c; | |
615 | ||
616 | if (sz >= ENVIRONMENT_BLOCK_MAX) | |
617 | return -ENOBUFS; | |
618 | ||
0b452006 RC |
619 | if (!GREEDY_REALLOC(outcome, allocated, sz + 5)) |
620 | return -ENOMEM; | |
621 | ||
2a7797e9 LP |
622 | r = safe_fgetc(f, &c); |
623 | if (r < 0) | |
624 | return r; | |
625 | if (r == 0) | |
626 | break; | |
627 | ||
0b452006 RC |
628 | if (c == '\0') |
629 | outcome[sz++] = '\n'; | |
630 | else | |
631 | sz += cescape_char(c, outcome + sz); | |
632 | } | |
633 | ||
2a7797e9 | 634 | outcome[sz] = '\0'; |
ae2a15bc | 635 | *env = TAKE_PTR(outcome); |
0b452006 RC |
636 | |
637 | return 0; | |
638 | } | |
639 | ||
6bc73acb | 640 | int get_process_ppid(pid_t pid, pid_t *_ppid) { |
0b452006 RC |
641 | int r; |
642 | _cleanup_free_ char *line = NULL; | |
643 | long unsigned ppid; | |
644 | const char *p; | |
645 | ||
646 | assert(pid >= 0); | |
647 | assert(_ppid); | |
648 | ||
6f8cbcdb | 649 | if (pid == 0 || pid == getpid_cached()) { |
0b452006 RC |
650 | *_ppid = getppid(); |
651 | return 0; | |
652 | } | |
653 | ||
654 | p = procfs_file_alloca(pid, "stat"); | |
655 | r = read_one_line_file(p, &line); | |
a644184a LP |
656 | if (r == -ENOENT) |
657 | return -ESRCH; | |
0b452006 RC |
658 | if (r < 0) |
659 | return r; | |
660 | ||
661 | /* Let's skip the pid and comm fields. The latter is enclosed | |
662 | * in () but does not escape any () in its value, so let's | |
663 | * skip over it manually */ | |
664 | ||
665 | p = strrchr(line, ')'); | |
666 | if (!p) | |
667 | return -EIO; | |
668 | ||
669 | p++; | |
670 | ||
671 | if (sscanf(p, " " | |
672 | "%*c " /* state */ | |
673 | "%lu ", /* ppid */ | |
674 | &ppid) != 1) | |
675 | return -EIO; | |
676 | ||
677 | if ((long unsigned) (pid_t) ppid != ppid) | |
678 | return -ERANGE; | |
679 | ||
680 | *_ppid = (pid_t) ppid; | |
681 | ||
682 | return 0; | |
683 | } | |
684 | ||
685 | int wait_for_terminate(pid_t pid, siginfo_t *status) { | |
686 | siginfo_t dummy; | |
687 | ||
688 | assert(pid >= 1); | |
689 | ||
690 | if (!status) | |
691 | status = &dummy; | |
692 | ||
693 | for (;;) { | |
694 | zero(*status); | |
695 | ||
696 | if (waitid(P_PID, pid, status, WEXITED) < 0) { | |
697 | ||
698 | if (errno == EINTR) | |
699 | continue; | |
700 | ||
3f0083a2 | 701 | return negative_errno(); |
0b452006 RC |
702 | } |
703 | ||
704 | return 0; | |
705 | } | |
706 | } | |
707 | ||
708 | /* | |
709 | * Return values: | |
710 | * < 0 : wait_for_terminate() failed to get the state of the | |
711 | * process, the process was terminated by a signal, or | |
712 | * failed for an unknown reason. | |
713 | * >=0 : The process terminated normally, and its exit code is | |
714 | * returned. | |
715 | * | |
716 | * That is, success is indicated by a return value of zero, and an | |
717 | * error is indicated by a non-zero value. | |
718 | * | |
719 | * A warning is emitted if the process terminates abnormally, | |
720 | * and also if it returns non-zero unless check_exit_code is true. | |
721 | */ | |
7d4904fe LP |
722 | int wait_for_terminate_and_check(const char *name, pid_t pid, WaitFlags flags) { |
723 | _cleanup_free_ char *buffer = NULL; | |
0b452006 | 724 | siginfo_t status; |
7d4904fe | 725 | int r, prio; |
0b452006 | 726 | |
0b452006 RC |
727 | assert(pid > 1); |
728 | ||
7d4904fe LP |
729 | if (!name) { |
730 | r = get_process_comm(pid, &buffer); | |
731 | if (r < 0) | |
732 | log_debug_errno(r, "Failed to acquire process name of " PID_FMT ", ignoring: %m", pid); | |
733 | else | |
734 | name = buffer; | |
735 | } | |
736 | ||
737 | prio = flags & WAIT_LOG_ABNORMAL ? LOG_ERR : LOG_DEBUG; | |
738 | ||
0b452006 RC |
739 | r = wait_for_terminate(pid, &status); |
740 | if (r < 0) | |
7d4904fe | 741 | return log_full_errno(prio, r, "Failed to wait for %s: %m", strna(name)); |
0b452006 RC |
742 | |
743 | if (status.si_code == CLD_EXITED) { | |
7d4904fe LP |
744 | if (status.si_status != EXIT_SUCCESS) |
745 | log_full(flags & WAIT_LOG_NON_ZERO_EXIT_STATUS ? LOG_ERR : LOG_DEBUG, | |
746 | "%s failed with exit status %i.", strna(name), status.si_status); | |
0b452006 RC |
747 | else |
748 | log_debug("%s succeeded.", name); | |
749 | ||
750 | return status.si_status; | |
7d4904fe | 751 | |
3742095b | 752 | } else if (IN_SET(status.si_code, CLD_KILLED, CLD_DUMPED)) { |
0b452006 | 753 | |
7d4904fe | 754 | log_full(prio, "%s terminated by signal %s.", strna(name), signal_to_string(status.si_status)); |
0b452006 RC |
755 | return -EPROTO; |
756 | } | |
757 | ||
7d4904fe | 758 | log_full(prio, "%s failed due to unknown reason.", strna(name)); |
0b452006 RC |
759 | return -EPROTO; |
760 | } | |
761 | ||
d5641e0d KW |
762 | /* |
763 | * Return values: | |
e225e5c3 LP |
764 | * |
765 | * < 0 : wait_for_terminate_with_timeout() failed to get the state of the process, the process timed out, the process | |
766 | * was terminated by a signal, or failed for an unknown reason. | |
767 | * | |
d5641e0d KW |
768 | * >=0 : The process terminated normally with no failures. |
769 | * | |
e225e5c3 LP |
770 | * Success is indicated by a return value of zero, a timeout is indicated by ETIMEDOUT, and all other child failure |
771 | * states are indicated by error is indicated by a non-zero value. | |
772 | * | |
773 | * This call assumes SIGCHLD has been blocked already, in particular before the child to wait for has been forked off | |
774 | * to remain entirely race-free. | |
d5641e0d KW |
775 | */ |
776 | int wait_for_terminate_with_timeout(pid_t pid, usec_t timeout) { | |
777 | sigset_t mask; | |
778 | int r; | |
779 | usec_t until; | |
780 | ||
781 | assert_se(sigemptyset(&mask) == 0); | |
782 | assert_se(sigaddset(&mask, SIGCHLD) == 0); | |
783 | ||
784 | /* Drop into a sigtimewait-based timeout. Waiting for the | |
785 | * pid to exit. */ | |
786 | until = now(CLOCK_MONOTONIC) + timeout; | |
787 | for (;;) { | |
788 | usec_t n; | |
789 | siginfo_t status = {}; | |
790 | struct timespec ts; | |
791 | ||
792 | n = now(CLOCK_MONOTONIC); | |
793 | if (n >= until) | |
794 | break; | |
795 | ||
796 | r = sigtimedwait(&mask, NULL, timespec_store(&ts, until - n)) < 0 ? -errno : 0; | |
797 | /* Assuming we woke due to the child exiting. */ | |
798 | if (waitid(P_PID, pid, &status, WEXITED|WNOHANG) == 0) { | |
799 | if (status.si_pid == pid) { | |
800 | /* This is the correct child.*/ | |
801 | if (status.si_code == CLD_EXITED) | |
802 | return (status.si_status == 0) ? 0 : -EPROTO; | |
803 | else | |
804 | return -EPROTO; | |
805 | } | |
806 | } | |
807 | /* Not the child, check for errors and proceed appropriately */ | |
808 | if (r < 0) { | |
809 | switch (r) { | |
810 | case -EAGAIN: | |
811 | /* Timed out, child is likely hung. */ | |
812 | return -ETIMEDOUT; | |
813 | case -EINTR: | |
814 | /* Received a different signal and should retry */ | |
815 | continue; | |
816 | default: | |
817 | /* Return any unexpected errors */ | |
818 | return r; | |
819 | } | |
820 | } | |
821 | } | |
822 | ||
823 | return -EPROTO; | |
824 | } | |
825 | ||
89c9030d LP |
826 | void sigkill_wait(pid_t pid) { |
827 | assert(pid > 1); | |
828 | ||
53640e6f | 829 | if (kill(pid, SIGKILL) >= 0) |
89c9030d LP |
830 | (void) wait_for_terminate(pid, NULL); |
831 | } | |
832 | ||
833 | void sigkill_waitp(pid_t *pid) { | |
dfd14786 LP |
834 | PROTECT_ERRNO; |
835 | ||
4d0d3d41 LP |
836 | if (!pid) |
837 | return; | |
838 | if (*pid <= 1) | |
839 | return; | |
840 | ||
89c9030d | 841 | sigkill_wait(*pid); |
4d0d3d41 LP |
842 | } |
843 | ||
392cf1d0 SL |
844 | void sigterm_wait(pid_t pid) { |
845 | assert(pid > 1); | |
846 | ||
53640e6f | 847 | if (kill_and_sigcont(pid, SIGTERM) >= 0) |
392cf1d0 SL |
848 | (void) wait_for_terminate(pid, NULL); |
849 | } | |
850 | ||
0b452006 RC |
851 | int kill_and_sigcont(pid_t pid, int sig) { |
852 | int r; | |
853 | ||
854 | r = kill(pid, sig) < 0 ? -errno : 0; | |
855 | ||
26f417d3 LP |
856 | /* If this worked, also send SIGCONT, unless we already just sent a SIGCONT, or SIGKILL was sent which isn't |
857 | * affected by a process being suspended anyway. */ | |
a3d8d68c | 858 | if (r >= 0 && !IN_SET(sig, SIGCONT, SIGKILL)) |
26f417d3 | 859 | (void) kill(pid, SIGCONT); |
0b452006 RC |
860 | |
861 | return r; | |
862 | } | |
863 | ||
e70f4453 | 864 | int getenv_for_pid(pid_t pid, const char *field, char **ret) { |
0b452006 RC |
865 | _cleanup_fclose_ FILE *f = NULL; |
866 | char *value = NULL; | |
0b452006 | 867 | const char *path; |
0d90bd92 LP |
868 | size_t l, sum = 0; |
869 | int r; | |
0b452006 RC |
870 | |
871 | assert(pid >= 0); | |
872 | assert(field); | |
e70f4453 LP |
873 | assert(ret); |
874 | ||
875 | if (pid == 0 || pid == getpid_cached()) { | |
876 | const char *e; | |
877 | ||
878 | e = getenv(field); | |
879 | if (!e) { | |
880 | *ret = NULL; | |
881 | return 0; | |
882 | } | |
883 | ||
884 | value = strdup(e); | |
885 | if (!value) | |
886 | return -ENOMEM; | |
887 | ||
888 | *ret = value; | |
889 | return 1; | |
890 | } | |
0b452006 | 891 | |
0d90bd92 LP |
892 | if (!pid_is_valid(pid)) |
893 | return -EINVAL; | |
894 | ||
0b452006 RC |
895 | path = procfs_file_alloca(pid, "environ"); |
896 | ||
897 | f = fopen(path, "re"); | |
a644184a LP |
898 | if (!f) { |
899 | if (errno == ENOENT) | |
900 | return -ESRCH; | |
e70f4453 | 901 | |
0b452006 | 902 | return -errno; |
a644184a | 903 | } |
0b452006 | 904 | |
35bbbf85 LP |
905 | (void) __fsetlocking(f, FSETLOCKING_BYCALLER); |
906 | ||
0b452006 | 907 | l = strlen(field); |
0d90bd92 LP |
908 | for (;;) { |
909 | _cleanup_free_ char *line = NULL; | |
0b452006 | 910 | |
0d90bd92 LP |
911 | if (sum > ENVIRONMENT_BLOCK_MAX) /* Give up searching eventually */ |
912 | return -ENOBUFS; | |
0b452006 | 913 | |
0d90bd92 LP |
914 | r = read_nul_string(f, LONG_LINE_MAX, &line); |
915 | if (r < 0) | |
916 | return r; | |
917 | if (r == 0) /* EOF */ | |
918 | break; | |
0b452006 | 919 | |
0d90bd92 | 920 | sum += r; |
0b452006 | 921 | |
041b5ae1 | 922 | if (strneq(line, field, l) && line[l] == '=') { |
0b452006 RC |
923 | value = strdup(line + l + 1); |
924 | if (!value) | |
925 | return -ENOMEM; | |
926 | ||
e70f4453 LP |
927 | *ret = value; |
928 | return 1; | |
0b452006 | 929 | } |
0d90bd92 | 930 | } |
0b452006 | 931 | |
e70f4453 LP |
932 | *ret = NULL; |
933 | return 0; | |
0b452006 RC |
934 | } |
935 | ||
936 | bool pid_is_unwaited(pid_t pid) { | |
937 | /* Checks whether a PID is still valid at all, including a zombie */ | |
938 | ||
07b38ba5 | 939 | if (pid < 0) |
0b452006 RC |
940 | return false; |
941 | ||
5fd9b2c5 LP |
942 | if (pid <= 1) /* If we or PID 1 would be dead and have been waited for, this code would not be running */ |
943 | return true; | |
944 | ||
6f8cbcdb LP |
945 | if (pid == getpid_cached()) |
946 | return true; | |
947 | ||
0b452006 RC |
948 | if (kill(pid, 0) >= 0) |
949 | return true; | |
950 | ||
951 | return errno != ESRCH; | |
952 | } | |
953 | ||
954 | bool pid_is_alive(pid_t pid) { | |
955 | int r; | |
956 | ||
957 | /* Checks whether a PID is still valid and not a zombie */ | |
958 | ||
07b38ba5 | 959 | if (pid < 0) |
0b452006 RC |
960 | return false; |
961 | ||
5fd9b2c5 LP |
962 | if (pid <= 1) /* If we or PID 1 would be a zombie, this code would not be running */ |
963 | return true; | |
964 | ||
6f8cbcdb LP |
965 | if (pid == getpid_cached()) |
966 | return true; | |
967 | ||
0b452006 | 968 | r = get_process_state(pid); |
4c701096 | 969 | if (IN_SET(r, -ESRCH, 'Z')) |
0b452006 RC |
970 | return false; |
971 | ||
972 | return true; | |
973 | } | |
d4510856 | 974 | |
1359fffa MS |
975 | int pid_from_same_root_fs(pid_t pid) { |
976 | const char *root; | |
977 | ||
07b38ba5 | 978 | if (pid < 0) |
6f8cbcdb LP |
979 | return false; |
980 | ||
981 | if (pid == 0 || pid == getpid_cached()) | |
982 | return true; | |
1359fffa MS |
983 | |
984 | root = procfs_file_alloca(pid, "root"); | |
985 | ||
e3f791a2 | 986 | return files_same(root, "/proc/1/root", 0); |
1359fffa MS |
987 | } |
988 | ||
d4510856 LP |
989 | bool is_main_thread(void) { |
990 | static thread_local int cached = 0; | |
991 | ||
992 | if (_unlikely_(cached == 0)) | |
df0ff127 | 993 | cached = getpid_cached() == gettid() ? 1 : -1; |
d4510856 LP |
994 | |
995 | return cached > 0; | |
996 | } | |
7b3e062c | 997 | |
848e863a | 998 | _noreturn_ void freeze(void) { |
7b3e062c | 999 | |
3da48d7a EV |
1000 | log_close(); |
1001 | ||
7b3e062c LP |
1002 | /* Make sure nobody waits for us on a socket anymore */ |
1003 | close_all_fds(NULL, 0); | |
1004 | ||
1005 | sync(); | |
1006 | ||
8647283e MS |
1007 | /* Let's not freeze right away, but keep reaping zombies. */ |
1008 | for (;;) { | |
1009 | int r; | |
1010 | siginfo_t si = {}; | |
1011 | ||
1012 | r = waitid(P_ALL, 0, &si, WEXITED); | |
1013 | if (r < 0 && errno != EINTR) | |
1014 | break; | |
1015 | } | |
1016 | ||
1017 | /* waitid() failed with an unexpected error, things are really borked. Freeze now! */ | |
7b3e062c LP |
1018 | for (;;) |
1019 | pause(); | |
1020 | } | |
1021 | ||
1022 | bool oom_score_adjust_is_valid(int oa) { | |
1023 | return oa >= OOM_SCORE_ADJ_MIN && oa <= OOM_SCORE_ADJ_MAX; | |
1024 | } | |
1025 | ||
1026 | unsigned long personality_from_string(const char *p) { | |
6e5f1b57 | 1027 | int architecture; |
7b3e062c | 1028 | |
0c0fea07 LP |
1029 | if (!p) |
1030 | return PERSONALITY_INVALID; | |
1031 | ||
6e5f1b57 LP |
1032 | /* Parse a personality specifier. We use our own identifiers that indicate specific ABIs, rather than just |
1033 | * hints regarding the register size, since we want to keep things open for multiple locally supported ABIs for | |
1034 | * the same register size. */ | |
1035 | ||
1036 | architecture = architecture_from_string(p); | |
1037 | if (architecture < 0) | |
1038 | return PERSONALITY_INVALID; | |
7b3e062c | 1039 | |
0c0fea07 | 1040 | if (architecture == native_architecture()) |
7b3e062c | 1041 | return PER_LINUX; |
0c0fea07 LP |
1042 | #ifdef SECONDARY_ARCHITECTURE |
1043 | if (architecture == SECONDARY_ARCHITECTURE) | |
f2d1736c | 1044 | return PER_LINUX32; |
7b3e062c LP |
1045 | #endif |
1046 | ||
1047 | return PERSONALITY_INVALID; | |
1048 | } | |
1049 | ||
1050 | const char* personality_to_string(unsigned long p) { | |
6e5f1b57 | 1051 | int architecture = _ARCHITECTURE_INVALID; |
7b3e062c | 1052 | |
7b3e062c | 1053 | if (p == PER_LINUX) |
0c0fea07 LP |
1054 | architecture = native_architecture(); |
1055 | #ifdef SECONDARY_ARCHITECTURE | |
6e5f1b57 | 1056 | else if (p == PER_LINUX32) |
0c0fea07 | 1057 | architecture = SECONDARY_ARCHITECTURE; |
7b3e062c LP |
1058 | #endif |
1059 | ||
6e5f1b57 LP |
1060 | if (architecture < 0) |
1061 | return NULL; | |
1062 | ||
1063 | return architecture_to_string(architecture); | |
7b3e062c LP |
1064 | } |
1065 | ||
21022b9d LP |
1066 | int safe_personality(unsigned long p) { |
1067 | int ret; | |
1068 | ||
1069 | /* So here's the deal, personality() is weirdly defined by glibc. In some cases it returns a failure via errno, | |
1070 | * and in others as negative return value containing an errno-like value. Let's work around this: this is a | |
1071 | * wrapper that uses errno if it is set, and uses the return value otherwise. And then it sets both errno and | |
1072 | * the return value indicating the same issue, so that we are definitely on the safe side. | |
1073 | * | |
1074 | * See https://github.com/systemd/systemd/issues/6737 */ | |
1075 | ||
1076 | errno = 0; | |
1077 | ret = personality(p); | |
1078 | if (ret < 0) { | |
1079 | if (errno != 0) | |
1080 | return -errno; | |
1081 | ||
1082 | errno = -ret; | |
1083 | } | |
1084 | ||
1085 | return ret; | |
1086 | } | |
1087 | ||
e8132d63 LP |
1088 | int opinionated_personality(unsigned long *ret) { |
1089 | int current; | |
1090 | ||
1091 | /* Returns the current personality, or PERSONALITY_INVALID if we can't determine it. This function is a bit | |
1092 | * opinionated though, and ignores all the finer-grained bits and exotic personalities, only distinguishing the | |
1093 | * two most relevant personalities: PER_LINUX and PER_LINUX32. */ | |
1094 | ||
21022b9d | 1095 | current = safe_personality(PERSONALITY_INVALID); |
e8132d63 | 1096 | if (current < 0) |
21022b9d | 1097 | return current; |
e8132d63 LP |
1098 | |
1099 | if (((unsigned long) current & 0xffff) == PER_LINUX32) | |
1100 | *ret = PER_LINUX32; | |
1101 | else | |
1102 | *ret = PER_LINUX; | |
1103 | ||
1104 | return 0; | |
1105 | } | |
1106 | ||
dcadc967 | 1107 | void valgrind_summary_hack(void) { |
349cc4a5 | 1108 | #if HAVE_VALGRIND_VALGRIND_H |
df0ff127 | 1109 | if (getpid_cached() == 1 && RUNNING_ON_VALGRIND) { |
dcadc967 | 1110 | pid_t pid; |
8869a0b4 | 1111 | pid = raw_clone(SIGCHLD); |
dcadc967 EV |
1112 | if (pid < 0) |
1113 | log_emergency_errno(errno, "Failed to fork off valgrind helper: %m"); | |
1114 | else if (pid == 0) | |
1115 | exit(EXIT_SUCCESS); | |
1116 | else { | |
1117 | log_info("Spawned valgrind helper as PID "PID_FMT".", pid); | |
1118 | (void) wait_for_terminate(pid, NULL); | |
1119 | } | |
1120 | } | |
1121 | #endif | |
1122 | } | |
1123 | ||
93bab288 | 1124 | int pid_compare_func(const pid_t *a, const pid_t *b) { |
291d565a | 1125 | /* Suitable for usage in qsort() */ |
93bab288 | 1126 | return CMP(*a, *b); |
291d565a LP |
1127 | } |
1128 | ||
7f452159 LP |
1129 | int ioprio_parse_priority(const char *s, int *ret) { |
1130 | int i, r; | |
1131 | ||
1132 | assert(s); | |
1133 | assert(ret); | |
1134 | ||
1135 | r = safe_atoi(s, &i); | |
1136 | if (r < 0) | |
1137 | return r; | |
1138 | ||
1139 | if (!ioprio_priority_is_valid(i)) | |
1140 | return -EINVAL; | |
1141 | ||
1142 | *ret = i; | |
1143 | return 0; | |
1144 | } | |
1145 | ||
5c30a6d2 LP |
1146 | /* The cached PID, possible values: |
1147 | * | |
1148 | * == UNSET [0] → cache not initialized yet | |
1149 | * == BUSY [-1] → some thread is initializing it at the moment | |
1150 | * any other → the cached PID | |
1151 | */ | |
1152 | ||
1153 | #define CACHED_PID_UNSET ((pid_t) 0) | |
1154 | #define CACHED_PID_BUSY ((pid_t) -1) | |
1155 | ||
1156 | static pid_t cached_pid = CACHED_PID_UNSET; | |
1157 | ||
799a960d | 1158 | void reset_cached_pid(void) { |
5c30a6d2 LP |
1159 | /* Invoked in the child after a fork(), i.e. at the first moment the PID changed */ |
1160 | cached_pid = CACHED_PID_UNSET; | |
1161 | } | |
1162 | ||
1163 | /* We use glibc __register_atfork() + __dso_handle directly here, as they are not included in the glibc | |
1164 | * headers. __register_atfork() is mostly equivalent to pthread_atfork(), but doesn't require us to link against | |
1165 | * libpthread, as it is part of glibc anyway. */ | |
2bb8d8d9 | 1166 | extern int __register_atfork(void (*prepare) (void), void (*parent) (void), void (*child) (void), void *dso_handle); |
d34dae18 | 1167 | extern void* __dso_handle _weak_; |
5c30a6d2 LP |
1168 | |
1169 | pid_t getpid_cached(void) { | |
5d71bac3 | 1170 | static bool installed = false; |
5c30a6d2 LP |
1171 | pid_t current_value; |
1172 | ||
1173 | /* getpid_cached() is much like getpid(), but caches the value in local memory, to avoid having to invoke a | |
1174 | * system call each time. This restores glibc behaviour from before 2.24, when getpid() was unconditionally | |
1175 | * cached. Starting with 2.24 getpid() started to become prohibitively expensive when used for detecting when | |
1176 | * objects were used across fork()s. With this caching the old behaviour is somewhat restored. | |
1177 | * | |
1178 | * https://bugzilla.redhat.com/show_bug.cgi?id=1443976 | |
a4041e4f | 1179 | * https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=c579f48edba88380635ab98cb612030e3ed8691e |
5c30a6d2 LP |
1180 | */ |
1181 | ||
1182 | current_value = __sync_val_compare_and_swap(&cached_pid, CACHED_PID_UNSET, CACHED_PID_BUSY); | |
1183 | ||
1184 | switch (current_value) { | |
1185 | ||
1186 | case CACHED_PID_UNSET: { /* Not initialized yet, then do so now */ | |
1187 | pid_t new_pid; | |
1188 | ||
996def17 | 1189 | new_pid = raw_getpid(); |
5c30a6d2 | 1190 | |
5d71bac3 LP |
1191 | if (!installed) { |
1192 | /* __register_atfork() either returns 0 or -ENOMEM, in its glibc implementation. Since it's | |
1193 | * only half-documented (glibc doesn't document it but LSB does — though only superficially) | |
1194 | * we'll check for errors only in the most generic fashion possible. */ | |
1195 | ||
1196 | if (__register_atfork(NULL, NULL, reset_cached_pid, __dso_handle) != 0) { | |
1197 | /* OOM? Let's try again later */ | |
1198 | cached_pid = CACHED_PID_UNSET; | |
1199 | return new_pid; | |
1200 | } | |
1201 | ||
1202 | installed = true; | |
5c30a6d2 LP |
1203 | } |
1204 | ||
1205 | cached_pid = new_pid; | |
1206 | return new_pid; | |
1207 | } | |
1208 | ||
1209 | case CACHED_PID_BUSY: /* Somebody else is currently initializing */ | |
996def17 | 1210 | return raw_getpid(); |
5c30a6d2 LP |
1211 | |
1212 | default: /* Properly initialized */ | |
1213 | return current_value; | |
1214 | } | |
1215 | } | |
1216 | ||
fba868fa LP |
1217 | int must_be_root(void) { |
1218 | ||
1219 | if (geteuid() == 0) | |
1220 | return 0; | |
1221 | ||
baaa35ad | 1222 | return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Need to be root."); |
fba868fa LP |
1223 | } |
1224 | ||
4c253ed1 LP |
1225 | int safe_fork_full( |
1226 | const char *name, | |
1227 | const int except_fds[], | |
1228 | size_t n_except_fds, | |
1229 | ForkFlags flags, | |
1230 | pid_t *ret_pid) { | |
1231 | ||
1232 | pid_t original_pid, pid; | |
1f5d1e02 | 1233 | sigset_t saved_ss, ss; |
a41f6217 | 1234 | bool block_signals = false; |
b6e1fff1 | 1235 | int prio, r; |
4c253ed1 LP |
1236 | |
1237 | /* A wrapper around fork(), that does a couple of important initializations in addition to mere forking. Always | |
1238 | * returns the child's PID in *ret_pid. Returns == 0 in the child, and > 0 in the parent. */ | |
1239 | ||
b6e1fff1 LP |
1240 | prio = flags & FORK_LOG ? LOG_ERR : LOG_DEBUG; |
1241 | ||
4c253ed1 LP |
1242 | original_pid = getpid_cached(); |
1243 | ||
1f5d1e02 | 1244 | if (flags & (FORK_RESET_SIGNALS|FORK_DEATHSIG)) { |
1f5d1e02 LP |
1245 | /* We temporarily block all signals, so that the new child has them blocked initially. This way, we can |
1246 | * be sure that SIGTERMs are not lost we might send to the child. */ | |
4c253ed1 | 1247 | |
cd2a429e | 1248 | assert_se(sigfillset(&ss) >= 0); |
1f5d1e02 LP |
1249 | block_signals = true; |
1250 | ||
1251 | } else if (flags & FORK_WAIT) { | |
1f5d1e02 LP |
1252 | /* Let's block SIGCHLD at least, so that we can safely watch for the child process */ |
1253 | ||
cd2a429e ZJS |
1254 | assert_se(sigemptyset(&ss) >= 0); |
1255 | assert_se(sigaddset(&ss, SIGCHLD) >= 0); | |
1f5d1e02 | 1256 | block_signals = true; |
4c253ed1 LP |
1257 | } |
1258 | ||
1f5d1e02 LP |
1259 | if (block_signals) |
1260 | if (sigprocmask(SIG_SETMASK, &ss, &saved_ss) < 0) | |
1261 | return log_full_errno(prio, errno, "Failed to set signal mask: %m"); | |
1262 | ||
be39f6ee LP |
1263 | if (flags & FORK_NEW_MOUNTNS) |
1264 | pid = raw_clone(SIGCHLD|CLONE_NEWNS); | |
1265 | else | |
1266 | pid = fork(); | |
4c253ed1 LP |
1267 | if (pid < 0) { |
1268 | r = -errno; | |
1269 | ||
1270 | if (block_signals) /* undo what we did above */ | |
1271 | (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL); | |
1272 | ||
b6e1fff1 | 1273 | return log_full_errno(prio, r, "Failed to fork: %m"); |
4c253ed1 LP |
1274 | } |
1275 | if (pid > 0) { | |
1276 | /* We are in the parent process */ | |
1277 | ||
1f5d1e02 LP |
1278 | log_debug("Successfully forked off '%s' as PID " PID_FMT ".", strna(name), pid); |
1279 | ||
1280 | if (flags & FORK_WAIT) { | |
1281 | r = wait_for_terminate_and_check(name, pid, (flags & FORK_LOG ? WAIT_LOG : 0)); | |
1282 | if (r < 0) | |
1283 | return r; | |
1284 | if (r != EXIT_SUCCESS) /* exit status > 0 should be treated as failure, too */ | |
1285 | return -EPROTO; | |
1286 | } | |
1287 | ||
4c253ed1 LP |
1288 | if (block_signals) /* undo what we did above */ |
1289 | (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL); | |
1290 | ||
4c253ed1 LP |
1291 | if (ret_pid) |
1292 | *ret_pid = pid; | |
1293 | ||
1294 | return 1; | |
1295 | } | |
1296 | ||
1297 | /* We are in the child process */ | |
1298 | ||
1299 | if (flags & FORK_REOPEN_LOG) { | |
1300 | /* Close the logs if requested, before we log anything. And make sure we reopen it if needed. */ | |
1301 | log_close(); | |
1302 | log_set_open_when_needed(true); | |
1303 | } | |
1304 | ||
1305 | if (name) { | |
1306 | r = rename_process(name); | |
1307 | if (r < 0) | |
b6e1fff1 LP |
1308 | log_full_errno(flags & FORK_LOG ? LOG_WARNING : LOG_DEBUG, |
1309 | r, "Failed to rename process, ignoring: %m"); | |
4c253ed1 LP |
1310 | } |
1311 | ||
1312 | if (flags & FORK_DEATHSIG) | |
1313 | if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0) { | |
b6e1fff1 | 1314 | log_full_errno(prio, errno, "Failed to set death signal: %m"); |
4c253ed1 LP |
1315 | _exit(EXIT_FAILURE); |
1316 | } | |
1317 | ||
1318 | if (flags & FORK_RESET_SIGNALS) { | |
1319 | r = reset_all_signal_handlers(); | |
1320 | if (r < 0) { | |
b6e1fff1 | 1321 | log_full_errno(prio, r, "Failed to reset signal handlers: %m"); |
4c253ed1 LP |
1322 | _exit(EXIT_FAILURE); |
1323 | } | |
1324 | ||
1325 | /* This implicitly undoes the signal mask stuff we did before the fork()ing above */ | |
1326 | r = reset_signal_mask(); | |
1327 | if (r < 0) { | |
b6e1fff1 | 1328 | log_full_errno(prio, r, "Failed to reset signal mask: %m"); |
4c253ed1 LP |
1329 | _exit(EXIT_FAILURE); |
1330 | } | |
1331 | } else if (block_signals) { /* undo what we did above */ | |
1332 | if (sigprocmask(SIG_SETMASK, &saved_ss, NULL) < 0) { | |
b6e1fff1 | 1333 | log_full_errno(prio, errno, "Failed to restore signal mask: %m"); |
4c253ed1 LP |
1334 | _exit(EXIT_FAILURE); |
1335 | } | |
1336 | } | |
1337 | ||
1338 | if (flags & FORK_DEATHSIG) { | |
7ddc2dc5 | 1339 | pid_t ppid; |
4c253ed1 LP |
1340 | /* Let's see if the parent PID is still the one we started from? If not, then the parent |
1341 | * already died by the time we set PR_SET_PDEATHSIG, hence let's emulate the effect */ | |
1342 | ||
7ddc2dc5 SL |
1343 | ppid = getppid(); |
1344 | if (ppid == 0) | |
1345 | /* Parent is in a differn't PID namespace. */; | |
1346 | else if (ppid != original_pid) { | |
4c253ed1 LP |
1347 | log_debug("Parent died early, raising SIGTERM."); |
1348 | (void) raise(SIGTERM); | |
1349 | _exit(EXIT_FAILURE); | |
1350 | } | |
1351 | } | |
1352 | ||
d94a24ca | 1353 | if (FLAGS_SET(flags, FORK_NEW_MOUNTNS | FORK_MOUNTNS_SLAVE)) { |
e2047ba9 LP |
1354 | |
1355 | /* Optionally, make sure we never propagate mounts to the host. */ | |
1356 | ||
1357 | if (mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL) < 0) { | |
1358 | log_full_errno(prio, errno, "Failed to remount root directory as MS_SLAVE: %m"); | |
1359 | _exit(EXIT_FAILURE); | |
1360 | } | |
1361 | } | |
1362 | ||
4c253ed1 LP |
1363 | if (flags & FORK_CLOSE_ALL_FDS) { |
1364 | /* Close the logs here in case it got reopened above, as close_all_fds() would close them for us */ | |
1365 | log_close(); | |
1366 | ||
1367 | r = close_all_fds(except_fds, n_except_fds); | |
1368 | if (r < 0) { | |
b6e1fff1 | 1369 | log_full_errno(prio, r, "Failed to close all file descriptors: %m"); |
4c253ed1 LP |
1370 | _exit(EXIT_FAILURE); |
1371 | } | |
1372 | } | |
1373 | ||
1374 | /* When we were asked to reopen the logs, do so again now */ | |
1375 | if (flags & FORK_REOPEN_LOG) { | |
1376 | log_open(); | |
1377 | log_set_open_when_needed(false); | |
1378 | } | |
1379 | ||
1380 | if (flags & FORK_NULL_STDIO) { | |
1381 | r = make_null_stdio(); | |
1382 | if (r < 0) { | |
b6e1fff1 | 1383 | log_full_errno(prio, r, "Failed to connect stdin/stdout to /dev/null: %m"); |
4c253ed1 LP |
1384 | _exit(EXIT_FAILURE); |
1385 | } | |
1386 | } | |
1387 | ||
909106eb LP |
1388 | if (flags & FORK_RLIMIT_NOFILE_SAFE) { |
1389 | r = rlimit_nofile_safe(); | |
1390 | if (r < 0) { | |
1391 | log_full_errno(prio, r, "Failed to lower RLIMIT_NOFILE's soft limit to 1K: %m"); | |
1392 | _exit(EXIT_FAILURE); | |
1393 | } | |
1394 | } | |
1395 | ||
4c253ed1 LP |
1396 | if (ret_pid) |
1397 | *ret_pid = getpid_cached(); | |
1398 | ||
1399 | return 0; | |
1400 | } | |
1401 | ||
27096982 LP |
1402 | int namespace_fork( |
1403 | const char *outer_name, | |
1404 | const char *inner_name, | |
1405 | const int except_fds[], | |
1406 | size_t n_except_fds, | |
1407 | ForkFlags flags, | |
1408 | int pidns_fd, | |
1409 | int mntns_fd, | |
1410 | int netns_fd, | |
1411 | int userns_fd, | |
1412 | int root_fd, | |
1413 | pid_t *ret_pid) { | |
1414 | ||
1415 | int r; | |
1416 | ||
1417 | /* This is much like safe_fork(), but forks twice, and joins the specified namespaces in the middle | |
1418 | * process. This ensures that we are fully a member of the destination namespace, with pidns an all, so that | |
1419 | * /proc/self/fd works correctly. */ | |
1420 | ||
1421 | 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); | |
1422 | if (r < 0) | |
1423 | return r; | |
1424 | if (r == 0) { | |
1425 | pid_t pid; | |
1426 | ||
1427 | /* Child */ | |
1428 | ||
1429 | r = namespace_enter(pidns_fd, mntns_fd, netns_fd, userns_fd, root_fd); | |
1430 | if (r < 0) { | |
1431 | log_full_errno(FLAGS_SET(flags, FORK_LOG) ? LOG_ERR : LOG_DEBUG, r, "Failed to join namespace: %m"); | |
1432 | _exit(EXIT_FAILURE); | |
1433 | } | |
1434 | ||
1435 | /* We mask a few flags here that either make no sense for the grandchild, or that we don't have to do again */ | |
1436 | 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); | |
1437 | if (r < 0) | |
1438 | _exit(EXIT_FAILURE); | |
1439 | if (r == 0) { | |
1440 | /* Child */ | |
1441 | if (ret_pid) | |
1442 | *ret_pid = pid; | |
1443 | return 0; | |
1444 | } | |
1445 | ||
1446 | r = wait_for_terminate_and_check(inner_name, pid, FLAGS_SET(flags, FORK_LOG) ? WAIT_LOG : 0); | |
1447 | if (r < 0) | |
1448 | _exit(EXIT_FAILURE); | |
1449 | ||
1450 | _exit(r); | |
1451 | } | |
1452 | ||
1453 | return 1; | |
1454 | } | |
1455 | ||
da6053d0 | 1456 | int fork_agent(const char *name, const int except[], size_t n_except, pid_t *ret_pid, const char *path, ...) { |
78752f2e | 1457 | bool stdout_is_tty, stderr_is_tty; |
da6053d0 | 1458 | size_t n, i; |
78752f2e LP |
1459 | va_list ap; |
1460 | char **l; | |
1461 | int r; | |
1462 | ||
1463 | assert(path); | |
1464 | ||
1465 | /* Spawns a temporary TTY agent, making sure it goes away when we go away */ | |
1466 | ||
1467 | r = safe_fork_full(name, except, n_except, FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_CLOSE_ALL_FDS, ret_pid); | |
1468 | if (r < 0) | |
1469 | return r; | |
1470 | if (r > 0) | |
1471 | return 0; | |
1472 | ||
1473 | /* In the child: */ | |
1474 | ||
1475 | stdout_is_tty = isatty(STDOUT_FILENO); | |
1476 | stderr_is_tty = isatty(STDERR_FILENO); | |
1477 | ||
1478 | if (!stdout_is_tty || !stderr_is_tty) { | |
1479 | int fd; | |
1480 | ||
1481 | /* Detach from stdout/stderr. and reopen | |
1482 | * /dev/tty for them. This is important to | |
1483 | * ensure that when systemctl is started via | |
1484 | * popen() or a similar call that expects to | |
1485 | * read EOF we actually do generate EOF and | |
1486 | * not delay this indefinitely by because we | |
1487 | * keep an unused copy of stdin around. */ | |
1488 | fd = open("/dev/tty", O_WRONLY); | |
1489 | if (fd < 0) { | |
1490 | log_error_errno(errno, "Failed to open /dev/tty: %m"); | |
1491 | _exit(EXIT_FAILURE); | |
1492 | } | |
1493 | ||
1494 | if (!stdout_is_tty && dup2(fd, STDOUT_FILENO) < 0) { | |
1495 | log_error_errno(errno, "Failed to dup2 /dev/tty: %m"); | |
1496 | _exit(EXIT_FAILURE); | |
1497 | } | |
1498 | ||
1499 | if (!stderr_is_tty && dup2(fd, STDERR_FILENO) < 0) { | |
1500 | log_error_errno(errno, "Failed to dup2 /dev/tty: %m"); | |
1501 | _exit(EXIT_FAILURE); | |
1502 | } | |
1503 | ||
e7685a77 | 1504 | safe_close_above_stdio(fd); |
78752f2e LP |
1505 | } |
1506 | ||
595225af LP |
1507 | (void) rlimit_nofile_safe(); |
1508 | ||
78752f2e LP |
1509 | /* Count arguments */ |
1510 | va_start(ap, path); | |
1511 | for (n = 0; va_arg(ap, char*); n++) | |
1512 | ; | |
1513 | va_end(ap); | |
1514 | ||
1515 | /* Allocate strv */ | |
cf409d15 | 1516 | l = newa(char*, n + 1); |
78752f2e LP |
1517 | |
1518 | /* Fill in arguments */ | |
1519 | va_start(ap, path); | |
1520 | for (i = 0; i <= n; i++) | |
1521 | l[i] = va_arg(ap, char*); | |
1522 | va_end(ap); | |
1523 | ||
1524 | execv(path, l); | |
1525 | _exit(EXIT_FAILURE); | |
1526 | } | |
1527 | ||
9f8168eb LP |
1528 | int set_oom_score_adjust(int value) { |
1529 | char t[DECIMAL_STR_MAX(int)]; | |
1530 | ||
1531 | sprintf(t, "%i", value); | |
1532 | ||
1533 | return write_string_file("/proc/self/oom_score_adj", t, | |
1534 | WRITE_STRING_FILE_VERIFY_ON_FAILURE|WRITE_STRING_FILE_DISABLE_BUFFER); | |
1535 | } | |
1536 | ||
7b3e062c LP |
1537 | static const char *const ioprio_class_table[] = { |
1538 | [IOPRIO_CLASS_NONE] = "none", | |
1539 | [IOPRIO_CLASS_RT] = "realtime", | |
1540 | [IOPRIO_CLASS_BE] = "best-effort", | |
1541 | [IOPRIO_CLASS_IDLE] = "idle" | |
1542 | }; | |
1543 | ||
10062bbc | 1544 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ioprio_class, int, IOPRIO_N_CLASSES); |
7b3e062c LP |
1545 | |
1546 | static const char *const sigchld_code_table[] = { | |
1547 | [CLD_EXITED] = "exited", | |
1548 | [CLD_KILLED] = "killed", | |
1549 | [CLD_DUMPED] = "dumped", | |
1550 | [CLD_TRAPPED] = "trapped", | |
1551 | [CLD_STOPPED] = "stopped", | |
1552 | [CLD_CONTINUED] = "continued", | |
1553 | }; | |
1554 | ||
1555 | DEFINE_STRING_TABLE_LOOKUP(sigchld_code, int); | |
1556 | ||
1557 | static const char* const sched_policy_table[] = { | |
1558 | [SCHED_OTHER] = "other", | |
1559 | [SCHED_BATCH] = "batch", | |
1560 | [SCHED_IDLE] = "idle", | |
1561 | [SCHED_FIFO] = "fifo", | |
1562 | [SCHED_RR] = "rr" | |
1563 | }; | |
1564 | ||
1565 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(sched_policy, int, INT_MAX); |