]>
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> | |
0b452006 | 7 | #include <signal.h> |
4f5dd394 LP |
8 | #include <stdbool.h> |
9 | #include <stdio.h> | |
11c3a366 | 10 | #include <stdlib.h> |
4f5dd394 | 11 | #include <string.h> |
9bfaffd5 | 12 | #include <sys/mman.h> |
e2047ba9 | 13 | #include <sys/mount.h> |
7b3e062c | 14 | #include <sys/personality.h> |
405f8907 | 15 | #include <sys/prctl.h> |
4f5dd394 LP |
16 | #include <sys/types.h> |
17 | #include <sys/wait.h> | |
11c3a366 | 18 | #include <syslog.h> |
4f5dd394 | 19 | #include <unistd.h> |
349cc4a5 | 20 | #if HAVE_VALGRIND_VALGRIND_H |
dcadc967 EV |
21 | #include <valgrind/valgrind.h> |
22 | #endif | |
0b452006 | 23 | |
b5efdb8a | 24 | #include "alloc-util.h" |
6e5f1b57 | 25 | #include "architecture.h" |
4f5dd394 | 26 | #include "escape.h" |
3ffd4af2 | 27 | #include "fd-util.h" |
0b452006 | 28 | #include "fileio.h" |
f4f15635 | 29 | #include "fs-util.h" |
7b3e062c | 30 | #include "ioprio.h" |
0b452006 | 31 | #include "log.h" |
11c3a366 | 32 | #include "macro.h" |
0a970718 | 33 | #include "memory-util.h" |
11c3a366 | 34 | #include "missing.h" |
0cb8e3d1 | 35 | #include "namespace-util.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" |
0b452006 RC |
45 | |
46 | int get_process_state(pid_t pid) { | |
47 | const char *p; | |
48 | char state; | |
49 | int r; | |
50 | _cleanup_free_ char *line = NULL; | |
51 | ||
52 | assert(pid >= 0); | |
53 | ||
54 | p = procfs_file_alloca(pid, "stat"); | |
a644184a | 55 | |
0b452006 | 56 | r = read_one_line_file(p, &line); |
a644184a LP |
57 | if (r == -ENOENT) |
58 | return -ESRCH; | |
0b452006 RC |
59 | if (r < 0) |
60 | return r; | |
61 | ||
62 | p = strrchr(line, ')'); | |
63 | if (!p) | |
64 | return -EIO; | |
65 | ||
66 | p++; | |
67 | ||
68 | if (sscanf(p, " %c", &state) != 1) | |
69 | return -EIO; | |
70 | ||
71 | return (unsigned char) state; | |
72 | } | |
73 | ||
ce268825 LP |
74 | int get_process_comm(pid_t pid, char **ret) { |
75 | _cleanup_free_ char *escaped = NULL, *comm = NULL; | |
0b452006 RC |
76 | const char *p; |
77 | int r; | |
78 | ||
ce268825 | 79 | assert(ret); |
0b452006 RC |
80 | assert(pid >= 0); |
81 | ||
ce268825 LP |
82 | escaped = new(char, TASK_COMM_LEN); |
83 | if (!escaped) | |
84 | return -ENOMEM; | |
85 | ||
0b452006 RC |
86 | p = procfs_file_alloca(pid, "comm"); |
87 | ||
ce268825 | 88 | r = read_one_line_file(p, &comm); |
0b452006 RC |
89 | if (r == -ENOENT) |
90 | return -ESRCH; | |
ce268825 LP |
91 | if (r < 0) |
92 | return r; | |
93 | ||
94 | /* Escape unprintable characters, just in case, but don't grow the string beyond the underlying size */ | |
95 | cellescape(escaped, TASK_COMM_LEN, comm); | |
0b452006 | 96 | |
ce268825 LP |
97 | *ret = TAKE_PTR(escaped); |
98 | return 0; | |
0b452006 RC |
99 | } |
100 | ||
101 | int get_process_cmdline(pid_t pid, size_t max_length, bool comm_fallback, char **line) { | |
102 | _cleanup_fclose_ FILE *f = NULL; | |
ba4cd7e2 | 103 | bool space = false; |
eb1ec489 MS |
104 | char *k; |
105 | _cleanup_free_ char *ans = NULL; | |
0b452006 | 106 | const char *p; |
fdeea3f4 | 107 | int c, r; |
0b452006 RC |
108 | |
109 | assert(line); | |
110 | assert(pid >= 0); | |
111 | ||
69281c49 LP |
112 | /* Retrieves a process' command line. Replaces unprintable characters while doing so by whitespace (coalescing |
113 | * multiple sequential ones into one). If max_length is != 0 will return a string of the specified size at most | |
114 | * (the trailing NUL byte does count towards the length here!), abbreviated with a "..." ellipsis. If | |
115 | * comm_fallback is true and the process has no command line set (the case for kernel threads), or has a | |
116 | * command line that resolves to the empty string will return the "comm" name of the process instead. | |
117 | * | |
118 | * Returns -ESRCH if the process doesn't exist, and -ENOENT if the process has no command line (and | |
c0534780 | 119 | * comm_fallback is false). Returns 0 and sets *line otherwise. */ |
69281c49 | 120 | |
0b452006 | 121 | p = procfs_file_alloca(pid, "cmdline"); |
fdeea3f4 ZJS |
122 | r = fopen_unlocked(p, "re", &f); |
123 | if (r == -ENOENT) | |
124 | return -ESRCH; | |
125 | if (r < 0) | |
126 | return r; | |
35bbbf85 | 127 | |
2d5d2e0c ZJS |
128 | if (max_length == 0) { |
129 | /* This is supposed to be a safety guard against runaway command lines. */ | |
130 | long l = sysconf(_SC_ARG_MAX); | |
131 | assert(l > 0); | |
132 | max_length = l; | |
133 | } | |
134 | ||
69281c49 LP |
135 | if (max_length == 1) { |
136 | ||
137 | /* If there's only room for one byte, return the empty string */ | |
c0534780 ZJS |
138 | ans = new0(char, 1); |
139 | if (!ans) | |
69281c49 LP |
140 | return -ENOMEM; |
141 | ||
eb1ec489 | 142 | *line = TAKE_PTR(ans); |
69281c49 LP |
143 | return 0; |
144 | ||
0b452006 | 145 | } else { |
69281c49 | 146 | bool dotdotdot = false; |
0b452006 RC |
147 | size_t left; |
148 | ||
c0534780 ZJS |
149 | ans = new(char, max_length); |
150 | if (!ans) | |
0b452006 RC |
151 | return -ENOMEM; |
152 | ||
c0534780 | 153 | k = ans; |
0b452006 RC |
154 | left = max_length; |
155 | while ((c = getc(f)) != EOF) { | |
156 | ||
157 | if (isprint(c)) { | |
69281c49 | 158 | |
0b452006 | 159 | if (space) { |
69281c49 LP |
160 | if (left <= 2) { |
161 | dotdotdot = true; | |
0b452006 | 162 | break; |
69281c49 | 163 | } |
0b452006 RC |
164 | |
165 | *(k++) = ' '; | |
166 | left--; | |
167 | space = false; | |
168 | } | |
169 | ||
69281c49 LP |
170 | if (left <= 1) { |
171 | dotdotdot = true; | |
0b452006 | 172 | break; |
69281c49 | 173 | } |
0b452006 RC |
174 | |
175 | *(k++) = (char) c; | |
176 | left--; | |
c0534780 | 177 | } else if (k > ans) |
0b452006 RC |
178 | space = true; |
179 | } | |
180 | ||
69281c49 LP |
181 | if (dotdotdot) { |
182 | if (max_length <= 4) { | |
c0534780 | 183 | k = ans; |
69281c49 LP |
184 | left = max_length; |
185 | } else { | |
c0534780 | 186 | k = ans + max_length - 4; |
69281c49 LP |
187 | left = 4; |
188 | ||
189 | /* Eat up final spaces */ | |
c0534780 | 190 | while (k > ans && isspace(k[-1])) { |
69281c49 LP |
191 | k--; |
192 | left++; | |
193 | } | |
194 | } | |
195 | ||
196 | strncpy(k, "...", left-1); | |
b09df4e2 | 197 | k[left-1] = 0; |
0b452006 RC |
198 | } else |
199 | *k = 0; | |
200 | } | |
201 | ||
202 | /* Kernel threads have no argv[] */ | |
c0534780 | 203 | if (isempty(ans)) { |
0b452006 RC |
204 | _cleanup_free_ char *t = NULL; |
205 | int h; | |
206 | ||
eb1ec489 | 207 | ans = mfree(ans); |
0b452006 RC |
208 | |
209 | if (!comm_fallback) | |
210 | return -ENOENT; | |
211 | ||
212 | h = get_process_comm(pid, &t); | |
213 | if (h < 0) | |
214 | return h; | |
215 | ||
2d5d2e0c | 216 | size_t l = strlen(t); |
69281c49 | 217 | |
2d5d2e0c ZJS |
218 | if (l + 3 <= max_length) { |
219 | ans = strjoin("[", t, "]"); | |
220 | if (!ans) | |
221 | return -ENOMEM; | |
69281c49 | 222 | |
2d5d2e0c ZJS |
223 | } else if (max_length <= 6) { |
224 | ans = new(char, max_length); | |
225 | if (!ans) | |
226 | return -ENOMEM; | |
69281c49 | 227 | |
2d5d2e0c ZJS |
228 | memcpy(ans, "[...]", max_length-1); |
229 | ans[max_length-1] = 0; | |
230 | } else { | |
231 | t[max_length - 6] = 0; | |
69281c49 | 232 | |
2d5d2e0c ZJS |
233 | /* Chop off final spaces */ |
234 | delete_trailing_chars(t, WHITESPACE); | |
69281c49 | 235 | |
2d5d2e0c ZJS |
236 | ans = strjoin("[", t, "...]"); |
237 | if (!ans) | |
238 | return -ENOMEM; | |
69281c49 | 239 | } |
eb1ec489 MS |
240 | |
241 | *line = TAKE_PTR(ans); | |
242 | return 0; | |
0b452006 RC |
243 | } |
244 | ||
eb1ec489 MS |
245 | k = realloc(ans, strlen(ans) + 1); |
246 | if (!k) | |
247 | return -ENOMEM; | |
248 | ||
249 | ans = NULL; | |
250 | *line = k; | |
251 | ||
0b452006 RC |
252 | return 0; |
253 | } | |
254 | ||
9bfaffd5 LP |
255 | int rename_process(const char name[]) { |
256 | static size_t mm_size = 0; | |
257 | static char *mm = NULL; | |
258 | bool truncated = false; | |
259 | size_t l; | |
260 | ||
261 | /* This is a like a poor man's setproctitle(). It changes the comm field, argv[0], and also the glibc's | |
262 | * internally used name of the process. For the first one a limit of 16 chars applies; to the second one in | |
263 | * many cases one of 10 (i.e. length of "/sbin/init") — however if we have CAP_SYS_RESOURCES it is unbounded; | |
264 | * to the third one 7 (i.e. the length of "systemd". If you pass a longer string it will likely be | |
265 | * truncated. | |
266 | * | |
267 | * Returns 0 if a name was set but truncated, > 0 if it was set but not truncated. */ | |
268 | ||
269 | if (isempty(name)) | |
270 | return -EINVAL; /* let's not confuse users unnecessarily with an empty name */ | |
405f8907 | 271 | |
1096bb8a LP |
272 | if (!is_main_thread()) |
273 | return -EPERM; /* Let's not allow setting the process name from other threads than the main one, as we | |
274 | * cache things without locking, and we make assumptions that PR_SET_NAME sets the | |
275 | * process name that isn't correct on any other threads */ | |
276 | ||
9bfaffd5 | 277 | l = strlen(name); |
405f8907 | 278 | |
5a8af747 LP |
279 | /* First step, change the comm field. The main thread's comm is identical to the process comm. This means we |
280 | * can use PR_SET_NAME, which sets the thread name for the calling thread. */ | |
281 | if (prctl(PR_SET_NAME, name) < 0) | |
282 | log_debug_errno(errno, "PR_SET_NAME failed: %m"); | |
92f14395 | 283 | if (l >= TASK_COMM_LEN) /* Linux process names can be 15 chars at max */ |
9bfaffd5 LP |
284 | truncated = true; |
285 | ||
286 | /* Second step, change glibc's ID of the process name. */ | |
287 | if (program_invocation_name) { | |
288 | size_t k; | |
289 | ||
290 | k = strlen(program_invocation_name); | |
291 | strncpy(program_invocation_name, name, k); | |
292 | if (l > k) | |
293 | truncated = true; | |
294 | } | |
295 | ||
296 | /* Third step, completely replace the argv[] array the kernel maintains for us. This requires privileges, but | |
297 | * has the advantage that the argv[] array is exactly what we want it to be, and not filled up with zeros at | |
13e785f7 | 298 | * the end. This is the best option for changing /proc/self/cmdline. */ |
01f989c6 JW |
299 | |
300 | /* Let's not bother with this if we don't have euid == 0. Strictly speaking we should check for the | |
301 | * CAP_SYS_RESOURCE capability which is independent of the euid. In our own code the capability generally is | |
302 | * present only for euid == 0, hence let's use this as quick bypass check, to avoid calling mmap() if | |
303 | * PR_SET_MM_ARG_{START,END} fails with EPERM later on anyway. After all geteuid() is dead cheap to call, but | |
304 | * mmap() is not. */ | |
305 | if (geteuid() != 0) | |
306 | log_debug("Skipping PR_SET_MM, as we don't have privileges."); | |
307 | else if (mm_size < l+1) { | |
9bfaffd5 LP |
308 | size_t nn_size; |
309 | char *nn; | |
310 | ||
9bfaffd5 LP |
311 | nn_size = PAGE_ALIGN(l+1); |
312 | nn = mmap(NULL, nn_size, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); | |
313 | if (nn == MAP_FAILED) { | |
314 | log_debug_errno(errno, "mmap() failed: %m"); | |
315 | goto use_saved_argv; | |
316 | } | |
405f8907 | 317 | |
9bfaffd5 LP |
318 | strncpy(nn, name, nn_size); |
319 | ||
320 | /* Now, let's tell the kernel about this new memory */ | |
321 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_START, (unsigned long) nn, 0, 0) < 0) { | |
14ee72b7 FS |
322 | /* HACK: prctl() API is kind of dumb on this point. The existing end address may already be |
323 | * below the desired start address, in which case the kernel may have kicked this back due | |
324 | * to a range-check failure (see linux/kernel/sys.c:validate_prctl_map() to see this in | |
325 | * action). The proper solution would be to have a prctl() API that could set both start+end | |
326 | * simultaneously, or at least let us query the existing address to anticipate this condition | |
327 | * and respond accordingly. For now, we can only guess at the cause of this failure and try | |
328 | * a workaround--which will briefly expand the arg space to something potentially huge before | |
329 | * resizing it to what we want. */ | |
330 | log_debug_errno(errno, "PR_SET_MM_ARG_START failed, attempting PR_SET_MM_ARG_END hack: %m"); | |
331 | ||
332 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) nn + l + 1, 0, 0) < 0) { | |
333 | log_debug_errno(errno, "PR_SET_MM_ARG_END hack failed, proceeding without: %m"); | |
334 | (void) munmap(nn, nn_size); | |
335 | goto use_saved_argv; | |
336 | } | |
9bfaffd5 | 337 | |
14ee72b7 FS |
338 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_START, (unsigned long) nn, 0, 0) < 0) { |
339 | log_debug_errno(errno, "PR_SET_MM_ARG_START still failed, proceeding without: %m"); | |
340 | goto use_saved_argv; | |
341 | } | |
342 | } else { | |
343 | /* And update the end pointer to the new end, too. If this fails, we don't really know what | |
344 | * to do, it's pretty unlikely that we can rollback, hence we'll just accept the failure, | |
345 | * and continue. */ | |
346 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) nn + l + 1, 0, 0) < 0) | |
347 | log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m"); | |
348 | } | |
9bfaffd5 LP |
349 | |
350 | if (mm) | |
351 | (void) munmap(mm, mm_size); | |
352 | ||
353 | mm = nn; | |
354 | mm_size = nn_size; | |
01f989c6 | 355 | } else { |
9bfaffd5 LP |
356 | strncpy(mm, name, mm_size); |
357 | ||
01f989c6 JW |
358 | /* Update the end pointer, continuing regardless of any failure. */ |
359 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) mm + l + 1, 0, 0) < 0) | |
360 | log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m"); | |
361 | } | |
362 | ||
9bfaffd5 LP |
363 | use_saved_argv: |
364 | /* Fourth step: in all cases we'll also update the original argv[], so that our own code gets it right too if | |
365 | * it still looks here */ | |
405f8907 LP |
366 | |
367 | if (saved_argc > 0) { | |
368 | int i; | |
369 | ||
9bfaffd5 LP |
370 | if (saved_argv[0]) { |
371 | size_t k; | |
372 | ||
373 | k = strlen(saved_argv[0]); | |
374 | strncpy(saved_argv[0], name, k); | |
375 | if (l > k) | |
376 | truncated = true; | |
377 | } | |
405f8907 LP |
378 | |
379 | for (i = 1; i < saved_argc; i++) { | |
380 | if (!saved_argv[i]) | |
381 | break; | |
382 | ||
383 | memzero(saved_argv[i], strlen(saved_argv[i])); | |
384 | } | |
385 | } | |
9bfaffd5 LP |
386 | |
387 | return !truncated; | |
405f8907 LP |
388 | } |
389 | ||
0b452006 | 390 | int is_kernel_thread(pid_t pid) { |
36b5119a LP |
391 | _cleanup_free_ char *line = NULL; |
392 | unsigned long long flags; | |
393 | size_t l, i; | |
0b452006 | 394 | const char *p; |
36b5119a LP |
395 | char *q; |
396 | int r; | |
0b452006 | 397 | |
4c701096 | 398 | if (IN_SET(pid, 0, 1) || pid == getpid_cached()) /* pid 1, and we ourselves certainly aren't a kernel thread */ |
0b452006 | 399 | return 0; |
36b5119a LP |
400 | if (!pid_is_valid(pid)) |
401 | return -EINVAL; | |
0b452006 | 402 | |
36b5119a LP |
403 | p = procfs_file_alloca(pid, "stat"); |
404 | r = read_one_line_file(p, &line); | |
405 | if (r == -ENOENT) | |
406 | return -ESRCH; | |
407 | if (r < 0) | |
408 | return r; | |
0b452006 | 409 | |
36b5119a LP |
410 | /* Skip past the comm field */ |
411 | q = strrchr(line, ')'); | |
412 | if (!q) | |
413 | return -EINVAL; | |
414 | q++; | |
415 | ||
416 | /* Skip 6 fields to reach the flags field */ | |
417 | for (i = 0; i < 6; i++) { | |
418 | l = strspn(q, WHITESPACE); | |
419 | if (l < 1) | |
420 | return -EINVAL; | |
421 | q += l; | |
422 | ||
423 | l = strcspn(q, WHITESPACE); | |
424 | if (l < 1) | |
425 | return -EINVAL; | |
426 | q += l; | |
a644184a | 427 | } |
0b452006 | 428 | |
f21f31b2 | 429 | /* Skip preceding whitespace */ |
36b5119a LP |
430 | l = strspn(q, WHITESPACE); |
431 | if (l < 1) | |
432 | return -EINVAL; | |
433 | q += l; | |
35bbbf85 | 434 | |
36b5119a LP |
435 | /* Truncate the rest */ |
436 | l = strcspn(q, WHITESPACE); | |
437 | if (l < 1) | |
438 | return -EINVAL; | |
439 | q[l] = 0; | |
0b452006 | 440 | |
36b5119a LP |
441 | r = safe_atollu(q, &flags); |
442 | if (r < 0) | |
443 | return r; | |
0b452006 | 444 | |
36b5119a | 445 | return !!(flags & PF_KTHREAD); |
0b452006 RC |
446 | } |
447 | ||
448 | int get_process_capeff(pid_t pid, char **capeff) { | |
449 | const char *p; | |
a644184a | 450 | int r; |
0b452006 RC |
451 | |
452 | assert(capeff); | |
453 | assert(pid >= 0); | |
454 | ||
455 | p = procfs_file_alloca(pid, "status"); | |
456 | ||
c4cd1d4d | 457 | r = get_proc_field(p, "CapEff", WHITESPACE, capeff); |
a644184a LP |
458 | if (r == -ENOENT) |
459 | return -ESRCH; | |
460 | ||
461 | return r; | |
0b452006 RC |
462 | } |
463 | ||
464 | static int get_process_link_contents(const char *proc_file, char **name) { | |
465 | int r; | |
466 | ||
467 | assert(proc_file); | |
468 | assert(name); | |
469 | ||
470 | r = readlink_malloc(proc_file, name); | |
a644184a LP |
471 | if (r == -ENOENT) |
472 | return -ESRCH; | |
0b452006 | 473 | if (r < 0) |
a644184a | 474 | return r; |
0b452006 RC |
475 | |
476 | return 0; | |
477 | } | |
478 | ||
479 | int get_process_exe(pid_t pid, char **name) { | |
480 | const char *p; | |
481 | char *d; | |
482 | int r; | |
483 | ||
484 | assert(pid >= 0); | |
485 | ||
486 | p = procfs_file_alloca(pid, "exe"); | |
487 | r = get_process_link_contents(p, name); | |
488 | if (r < 0) | |
489 | return r; | |
490 | ||
491 | d = endswith(*name, " (deleted)"); | |
492 | if (d) | |
493 | *d = '\0'; | |
494 | ||
495 | return 0; | |
496 | } | |
497 | ||
498 | static int get_process_id(pid_t pid, const char *field, uid_t *uid) { | |
499 | _cleanup_fclose_ FILE *f = NULL; | |
0b452006 | 500 | const char *p; |
7e7a16a0 | 501 | int r; |
0b452006 RC |
502 | |
503 | assert(field); | |
504 | assert(uid); | |
505 | ||
07b38ba5 | 506 | if (pid < 0) |
6f8cbcdb LP |
507 | return -EINVAL; |
508 | ||
0b452006 | 509 | p = procfs_file_alloca(pid, "status"); |
fdeea3f4 ZJS |
510 | r = fopen_unlocked(p, "re", &f); |
511 | if (r == -ENOENT) | |
512 | return -ESRCH; | |
513 | if (r < 0) | |
514 | return r; | |
35bbbf85 | 515 | |
7e7a16a0 LP |
516 | for (;;) { |
517 | _cleanup_free_ char *line = NULL; | |
0b452006 RC |
518 | char *l; |
519 | ||
7e7a16a0 LP |
520 | r = read_line(f, LONG_LINE_MAX, &line); |
521 | if (r < 0) | |
522 | return r; | |
523 | if (r == 0) | |
524 | break; | |
525 | ||
0b452006 RC |
526 | l = strstrip(line); |
527 | ||
528 | if (startswith(l, field)) { | |
529 | l += strlen(field); | |
530 | l += strspn(l, WHITESPACE); | |
531 | ||
532 | l[strcspn(l, WHITESPACE)] = 0; | |
533 | ||
534 | return parse_uid(l, uid); | |
535 | } | |
536 | } | |
537 | ||
538 | return -EIO; | |
539 | } | |
540 | ||
541 | int get_process_uid(pid_t pid, uid_t *uid) { | |
6f8cbcdb LP |
542 | |
543 | if (pid == 0 || pid == getpid_cached()) { | |
544 | *uid = getuid(); | |
545 | return 0; | |
546 | } | |
547 | ||
0b452006 RC |
548 | return get_process_id(pid, "Uid:", uid); |
549 | } | |
550 | ||
551 | int get_process_gid(pid_t pid, gid_t *gid) { | |
6f8cbcdb LP |
552 | |
553 | if (pid == 0 || pid == getpid_cached()) { | |
554 | *gid = getgid(); | |
555 | return 0; | |
556 | } | |
557 | ||
0b452006 RC |
558 | assert_cc(sizeof(uid_t) == sizeof(gid_t)); |
559 | return get_process_id(pid, "Gid:", gid); | |
560 | } | |
561 | ||
562 | int get_process_cwd(pid_t pid, char **cwd) { | |
563 | const char *p; | |
564 | ||
565 | assert(pid >= 0); | |
566 | ||
567 | p = procfs_file_alloca(pid, "cwd"); | |
568 | ||
569 | return get_process_link_contents(p, cwd); | |
570 | } | |
571 | ||
572 | int get_process_root(pid_t pid, char **root) { | |
573 | const char *p; | |
574 | ||
575 | assert(pid >= 0); | |
576 | ||
577 | p = procfs_file_alloca(pid, "root"); | |
578 | ||
579 | return get_process_link_contents(p, root); | |
580 | } | |
581 | ||
2a7797e9 LP |
582 | #define ENVIRONMENT_BLOCK_MAX (5U*1024U*1024U) |
583 | ||
0b452006 RC |
584 | int get_process_environ(pid_t pid, char **env) { |
585 | _cleanup_fclose_ FILE *f = NULL; | |
586 | _cleanup_free_ char *outcome = NULL; | |
0b452006 | 587 | size_t allocated = 0, sz = 0; |
2a7797e9 LP |
588 | const char *p; |
589 | int r; | |
0b452006 RC |
590 | |
591 | assert(pid >= 0); | |
592 | assert(env); | |
593 | ||
594 | p = procfs_file_alloca(pid, "environ"); | |
595 | ||
fdeea3f4 ZJS |
596 | r = fopen_unlocked(p, "re", &f); |
597 | if (r == -ENOENT) | |
598 | return -ESRCH; | |
599 | if (r < 0) | |
600 | return r; | |
35bbbf85 | 601 | |
2a7797e9 LP |
602 | for (;;) { |
603 | char c; | |
604 | ||
605 | if (sz >= ENVIRONMENT_BLOCK_MAX) | |
606 | return -ENOBUFS; | |
607 | ||
0b452006 RC |
608 | if (!GREEDY_REALLOC(outcome, allocated, sz + 5)) |
609 | return -ENOMEM; | |
610 | ||
2a7797e9 LP |
611 | r = safe_fgetc(f, &c); |
612 | if (r < 0) | |
613 | return r; | |
614 | if (r == 0) | |
615 | break; | |
616 | ||
0b452006 RC |
617 | if (c == '\0') |
618 | outcome[sz++] = '\n'; | |
619 | else | |
620 | sz += cescape_char(c, outcome + sz); | |
621 | } | |
622 | ||
2a7797e9 | 623 | outcome[sz] = '\0'; |
ae2a15bc | 624 | *env = TAKE_PTR(outcome); |
0b452006 RC |
625 | |
626 | return 0; | |
627 | } | |
628 | ||
6bc73acb | 629 | int get_process_ppid(pid_t pid, pid_t *_ppid) { |
0b452006 RC |
630 | int r; |
631 | _cleanup_free_ char *line = NULL; | |
632 | long unsigned ppid; | |
633 | const char *p; | |
634 | ||
635 | assert(pid >= 0); | |
636 | assert(_ppid); | |
637 | ||
6f8cbcdb | 638 | if (pid == 0 || pid == getpid_cached()) { |
0b452006 RC |
639 | *_ppid = getppid(); |
640 | return 0; | |
641 | } | |
642 | ||
643 | p = procfs_file_alloca(pid, "stat"); | |
644 | r = read_one_line_file(p, &line); | |
a644184a LP |
645 | if (r == -ENOENT) |
646 | return -ESRCH; | |
0b452006 RC |
647 | if (r < 0) |
648 | return r; | |
649 | ||
650 | /* Let's skip the pid and comm fields. The latter is enclosed | |
651 | * in () but does not escape any () in its value, so let's | |
652 | * skip over it manually */ | |
653 | ||
654 | p = strrchr(line, ')'); | |
655 | if (!p) | |
656 | return -EIO; | |
657 | ||
658 | p++; | |
659 | ||
660 | if (sscanf(p, " " | |
661 | "%*c " /* state */ | |
662 | "%lu ", /* ppid */ | |
663 | &ppid) != 1) | |
664 | return -EIO; | |
665 | ||
666 | if ((long unsigned) (pid_t) ppid != ppid) | |
667 | return -ERANGE; | |
668 | ||
669 | *_ppid = (pid_t) ppid; | |
670 | ||
671 | return 0; | |
672 | } | |
673 | ||
674 | int wait_for_terminate(pid_t pid, siginfo_t *status) { | |
675 | siginfo_t dummy; | |
676 | ||
677 | assert(pid >= 1); | |
678 | ||
679 | if (!status) | |
680 | status = &dummy; | |
681 | ||
682 | for (;;) { | |
683 | zero(*status); | |
684 | ||
685 | if (waitid(P_PID, pid, status, WEXITED) < 0) { | |
686 | ||
687 | if (errno == EINTR) | |
688 | continue; | |
689 | ||
3f0083a2 | 690 | return negative_errno(); |
0b452006 RC |
691 | } |
692 | ||
693 | return 0; | |
694 | } | |
695 | } | |
696 | ||
697 | /* | |
698 | * Return values: | |
699 | * < 0 : wait_for_terminate() failed to get the state of the | |
700 | * process, the process was terminated by a signal, or | |
701 | * failed for an unknown reason. | |
702 | * >=0 : The process terminated normally, and its exit code is | |
703 | * returned. | |
704 | * | |
705 | * That is, success is indicated by a return value of zero, and an | |
706 | * error is indicated by a non-zero value. | |
707 | * | |
708 | * A warning is emitted if the process terminates abnormally, | |
709 | * and also if it returns non-zero unless check_exit_code is true. | |
710 | */ | |
7d4904fe LP |
711 | int wait_for_terminate_and_check(const char *name, pid_t pid, WaitFlags flags) { |
712 | _cleanup_free_ char *buffer = NULL; | |
0b452006 | 713 | siginfo_t status; |
7d4904fe | 714 | int r, prio; |
0b452006 | 715 | |
0b452006 RC |
716 | assert(pid > 1); |
717 | ||
7d4904fe LP |
718 | if (!name) { |
719 | r = get_process_comm(pid, &buffer); | |
720 | if (r < 0) | |
721 | log_debug_errno(r, "Failed to acquire process name of " PID_FMT ", ignoring: %m", pid); | |
722 | else | |
723 | name = buffer; | |
724 | } | |
725 | ||
726 | prio = flags & WAIT_LOG_ABNORMAL ? LOG_ERR : LOG_DEBUG; | |
727 | ||
0b452006 RC |
728 | r = wait_for_terminate(pid, &status); |
729 | if (r < 0) | |
7d4904fe | 730 | return log_full_errno(prio, r, "Failed to wait for %s: %m", strna(name)); |
0b452006 RC |
731 | |
732 | if (status.si_code == CLD_EXITED) { | |
7d4904fe LP |
733 | if (status.si_status != EXIT_SUCCESS) |
734 | log_full(flags & WAIT_LOG_NON_ZERO_EXIT_STATUS ? LOG_ERR : LOG_DEBUG, | |
735 | "%s failed with exit status %i.", strna(name), status.si_status); | |
0b452006 RC |
736 | else |
737 | log_debug("%s succeeded.", name); | |
738 | ||
739 | return status.si_status; | |
7d4904fe | 740 | |
3742095b | 741 | } else if (IN_SET(status.si_code, CLD_KILLED, CLD_DUMPED)) { |
0b452006 | 742 | |
7d4904fe | 743 | log_full(prio, "%s terminated by signal %s.", strna(name), signal_to_string(status.si_status)); |
0b452006 RC |
744 | return -EPROTO; |
745 | } | |
746 | ||
7d4904fe | 747 | log_full(prio, "%s failed due to unknown reason.", strna(name)); |
0b452006 RC |
748 | return -EPROTO; |
749 | } | |
750 | ||
d5641e0d KW |
751 | /* |
752 | * Return values: | |
e225e5c3 LP |
753 | * |
754 | * < 0 : wait_for_terminate_with_timeout() failed to get the state of the process, the process timed out, the process | |
755 | * was terminated by a signal, or failed for an unknown reason. | |
756 | * | |
d5641e0d KW |
757 | * >=0 : The process terminated normally with no failures. |
758 | * | |
e225e5c3 LP |
759 | * Success is indicated by a return value of zero, a timeout is indicated by ETIMEDOUT, and all other child failure |
760 | * states are indicated by error is indicated by a non-zero value. | |
761 | * | |
762 | * This call assumes SIGCHLD has been blocked already, in particular before the child to wait for has been forked off | |
763 | * to remain entirely race-free. | |
d5641e0d KW |
764 | */ |
765 | int wait_for_terminate_with_timeout(pid_t pid, usec_t timeout) { | |
766 | sigset_t mask; | |
767 | int r; | |
768 | usec_t until; | |
769 | ||
770 | assert_se(sigemptyset(&mask) == 0); | |
771 | assert_se(sigaddset(&mask, SIGCHLD) == 0); | |
772 | ||
773 | /* Drop into a sigtimewait-based timeout. Waiting for the | |
774 | * pid to exit. */ | |
775 | until = now(CLOCK_MONOTONIC) + timeout; | |
776 | for (;;) { | |
777 | usec_t n; | |
778 | siginfo_t status = {}; | |
779 | struct timespec ts; | |
780 | ||
781 | n = now(CLOCK_MONOTONIC); | |
782 | if (n >= until) | |
783 | break; | |
784 | ||
785 | r = sigtimedwait(&mask, NULL, timespec_store(&ts, until - n)) < 0 ? -errno : 0; | |
786 | /* Assuming we woke due to the child exiting. */ | |
787 | if (waitid(P_PID, pid, &status, WEXITED|WNOHANG) == 0) { | |
788 | if (status.si_pid == pid) { | |
789 | /* This is the correct child.*/ | |
790 | if (status.si_code == CLD_EXITED) | |
791 | return (status.si_status == 0) ? 0 : -EPROTO; | |
792 | else | |
793 | return -EPROTO; | |
794 | } | |
795 | } | |
796 | /* Not the child, check for errors and proceed appropriately */ | |
797 | if (r < 0) { | |
798 | switch (r) { | |
799 | case -EAGAIN: | |
800 | /* Timed out, child is likely hung. */ | |
801 | return -ETIMEDOUT; | |
802 | case -EINTR: | |
803 | /* Received a different signal and should retry */ | |
804 | continue; | |
805 | default: | |
806 | /* Return any unexpected errors */ | |
807 | return r; | |
808 | } | |
809 | } | |
810 | } | |
811 | ||
812 | return -EPROTO; | |
813 | } | |
814 | ||
89c9030d LP |
815 | void sigkill_wait(pid_t pid) { |
816 | assert(pid > 1); | |
817 | ||
53640e6f | 818 | if (kill(pid, SIGKILL) >= 0) |
89c9030d LP |
819 | (void) wait_for_terminate(pid, NULL); |
820 | } | |
821 | ||
822 | void sigkill_waitp(pid_t *pid) { | |
dfd14786 LP |
823 | PROTECT_ERRNO; |
824 | ||
4d0d3d41 LP |
825 | if (!pid) |
826 | return; | |
827 | if (*pid <= 1) | |
828 | return; | |
829 | ||
89c9030d | 830 | sigkill_wait(*pid); |
4d0d3d41 LP |
831 | } |
832 | ||
392cf1d0 SL |
833 | void sigterm_wait(pid_t pid) { |
834 | assert(pid > 1); | |
835 | ||
53640e6f | 836 | if (kill_and_sigcont(pid, SIGTERM) >= 0) |
392cf1d0 SL |
837 | (void) wait_for_terminate(pid, NULL); |
838 | } | |
839 | ||
0b452006 RC |
840 | int kill_and_sigcont(pid_t pid, int sig) { |
841 | int r; | |
842 | ||
843 | r = kill(pid, sig) < 0 ? -errno : 0; | |
844 | ||
26f417d3 LP |
845 | /* If this worked, also send SIGCONT, unless we already just sent a SIGCONT, or SIGKILL was sent which isn't |
846 | * affected by a process being suspended anyway. */ | |
a3d8d68c | 847 | if (r >= 0 && !IN_SET(sig, SIGCONT, SIGKILL)) |
26f417d3 | 848 | (void) kill(pid, SIGCONT); |
0b452006 RC |
849 | |
850 | return r; | |
851 | } | |
852 | ||
e70f4453 | 853 | int getenv_for_pid(pid_t pid, const char *field, char **ret) { |
0b452006 RC |
854 | _cleanup_fclose_ FILE *f = NULL; |
855 | char *value = NULL; | |
0b452006 | 856 | const char *path; |
0d90bd92 LP |
857 | size_t l, sum = 0; |
858 | int r; | |
0b452006 RC |
859 | |
860 | assert(pid >= 0); | |
861 | assert(field); | |
e70f4453 LP |
862 | assert(ret); |
863 | ||
864 | if (pid == 0 || pid == getpid_cached()) { | |
865 | const char *e; | |
866 | ||
867 | e = getenv(field); | |
868 | if (!e) { | |
869 | *ret = NULL; | |
870 | return 0; | |
871 | } | |
872 | ||
873 | value = strdup(e); | |
874 | if (!value) | |
875 | return -ENOMEM; | |
876 | ||
877 | *ret = value; | |
878 | return 1; | |
879 | } | |
0b452006 | 880 | |
0d90bd92 LP |
881 | if (!pid_is_valid(pid)) |
882 | return -EINVAL; | |
883 | ||
0b452006 RC |
884 | path = procfs_file_alloca(pid, "environ"); |
885 | ||
fdeea3f4 ZJS |
886 | r = fopen_unlocked(path, "re", &f); |
887 | if (r == -ENOENT) | |
888 | return -ESRCH; | |
889 | if (r < 0) | |
890 | return r; | |
35bbbf85 | 891 | |
0b452006 | 892 | l = strlen(field); |
0d90bd92 LP |
893 | for (;;) { |
894 | _cleanup_free_ char *line = NULL; | |
0b452006 | 895 | |
0d90bd92 LP |
896 | if (sum > ENVIRONMENT_BLOCK_MAX) /* Give up searching eventually */ |
897 | return -ENOBUFS; | |
0b452006 | 898 | |
0d90bd92 LP |
899 | r = read_nul_string(f, LONG_LINE_MAX, &line); |
900 | if (r < 0) | |
901 | return r; | |
902 | if (r == 0) /* EOF */ | |
903 | break; | |
0b452006 | 904 | |
0d90bd92 | 905 | sum += r; |
0b452006 | 906 | |
041b5ae1 | 907 | if (strneq(line, field, l) && line[l] == '=') { |
0b452006 RC |
908 | value = strdup(line + l + 1); |
909 | if (!value) | |
910 | return -ENOMEM; | |
911 | ||
e70f4453 LP |
912 | *ret = value; |
913 | return 1; | |
0b452006 | 914 | } |
0d90bd92 | 915 | } |
0b452006 | 916 | |
e70f4453 LP |
917 | *ret = NULL; |
918 | return 0; | |
0b452006 RC |
919 | } |
920 | ||
4d051546 FB |
921 | int pid_is_my_child(pid_t pid) { |
922 | pid_t ppid; | |
923 | int r; | |
924 | ||
925 | if (pid <= 1) | |
926 | return false; | |
927 | ||
928 | r = get_process_ppid(pid, &ppid); | |
929 | if (r < 0) | |
930 | return r; | |
931 | ||
932 | return ppid == getpid_cached(); | |
933 | } | |
934 | ||
0b452006 RC |
935 | bool pid_is_unwaited(pid_t pid) { |
936 | /* Checks whether a PID is still valid at all, including a zombie */ | |
937 | ||
07b38ba5 | 938 | if (pid < 0) |
0b452006 RC |
939 | return false; |
940 | ||
5fd9b2c5 LP |
941 | if (pid <= 1) /* If we or PID 1 would be dead and have been waited for, this code would not be running */ |
942 | return true; | |
943 | ||
6f8cbcdb LP |
944 | if (pid == getpid_cached()) |
945 | return true; | |
946 | ||
0b452006 RC |
947 | if (kill(pid, 0) >= 0) |
948 | return true; | |
949 | ||
950 | return errno != ESRCH; | |
951 | } | |
952 | ||
953 | bool pid_is_alive(pid_t pid) { | |
954 | int r; | |
955 | ||
956 | /* Checks whether a PID is still valid and not a zombie */ | |
957 | ||
07b38ba5 | 958 | if (pid < 0) |
0b452006 RC |
959 | return false; |
960 | ||
5fd9b2c5 LP |
961 | if (pid <= 1) /* If we or PID 1 would be a zombie, this code would not be running */ |
962 | return true; | |
963 | ||
6f8cbcdb LP |
964 | if (pid == getpid_cached()) |
965 | return true; | |
966 | ||
0b452006 | 967 | r = get_process_state(pid); |
4c701096 | 968 | if (IN_SET(r, -ESRCH, 'Z')) |
0b452006 RC |
969 | return false; |
970 | ||
971 | return true; | |
972 | } | |
d4510856 | 973 | |
1359fffa MS |
974 | int pid_from_same_root_fs(pid_t pid) { |
975 | const char *root; | |
976 | ||
07b38ba5 | 977 | if (pid < 0) |
6f8cbcdb LP |
978 | return false; |
979 | ||
980 | if (pid == 0 || pid == getpid_cached()) | |
981 | return true; | |
1359fffa MS |
982 | |
983 | root = procfs_file_alloca(pid, "root"); | |
984 | ||
e3f791a2 | 985 | return files_same(root, "/proc/1/root", 0); |
1359fffa MS |
986 | } |
987 | ||
d4510856 LP |
988 | bool is_main_thread(void) { |
989 | static thread_local int cached = 0; | |
990 | ||
991 | if (_unlikely_(cached == 0)) | |
df0ff127 | 992 | cached = getpid_cached() == gettid() ? 1 : -1; |
d4510856 LP |
993 | |
994 | return cached > 0; | |
995 | } | |
7b3e062c | 996 | |
848e863a | 997 | _noreturn_ void freeze(void) { |
7b3e062c | 998 | |
3da48d7a EV |
999 | log_close(); |
1000 | ||
7b3e062c | 1001 | /* Make sure nobody waits for us on a socket anymore */ |
7acf581a | 1002 | (void) close_all_fds(NULL, 0); |
7b3e062c LP |
1003 | |
1004 | sync(); | |
1005 | ||
8647283e MS |
1006 | /* Let's not freeze right away, but keep reaping zombies. */ |
1007 | for (;;) { | |
1008 | int r; | |
1009 | siginfo_t si = {}; | |
1010 | ||
1011 | r = waitid(P_ALL, 0, &si, WEXITED); | |
1012 | if (r < 0 && errno != EINTR) | |
1013 | break; | |
1014 | } | |
1015 | ||
1016 | /* waitid() failed with an unexpected error, things are really borked. Freeze now! */ | |
7b3e062c LP |
1017 | for (;;) |
1018 | pause(); | |
1019 | } | |
1020 | ||
1021 | bool oom_score_adjust_is_valid(int oa) { | |
1022 | return oa >= OOM_SCORE_ADJ_MIN && oa <= OOM_SCORE_ADJ_MAX; | |
1023 | } | |
1024 | ||
1025 | unsigned long personality_from_string(const char *p) { | |
6e5f1b57 | 1026 | int architecture; |
7b3e062c | 1027 | |
0c0fea07 LP |
1028 | if (!p) |
1029 | return PERSONALITY_INVALID; | |
1030 | ||
6e5f1b57 LP |
1031 | /* Parse a personality specifier. We use our own identifiers that indicate specific ABIs, rather than just |
1032 | * hints regarding the register size, since we want to keep things open for multiple locally supported ABIs for | |
1033 | * the same register size. */ | |
1034 | ||
1035 | architecture = architecture_from_string(p); | |
1036 | if (architecture < 0) | |
1037 | return PERSONALITY_INVALID; | |
7b3e062c | 1038 | |
0c0fea07 | 1039 | if (architecture == native_architecture()) |
7b3e062c | 1040 | return PER_LINUX; |
0c0fea07 LP |
1041 | #ifdef SECONDARY_ARCHITECTURE |
1042 | if (architecture == SECONDARY_ARCHITECTURE) | |
f2d1736c | 1043 | return PER_LINUX32; |
7b3e062c LP |
1044 | #endif |
1045 | ||
1046 | return PERSONALITY_INVALID; | |
1047 | } | |
1048 | ||
1049 | const char* personality_to_string(unsigned long p) { | |
6e5f1b57 | 1050 | int architecture = _ARCHITECTURE_INVALID; |
7b3e062c | 1051 | |
7b3e062c | 1052 | if (p == PER_LINUX) |
0c0fea07 LP |
1053 | architecture = native_architecture(); |
1054 | #ifdef SECONDARY_ARCHITECTURE | |
6e5f1b57 | 1055 | else if (p == PER_LINUX32) |
0c0fea07 | 1056 | architecture = SECONDARY_ARCHITECTURE; |
7b3e062c LP |
1057 | #endif |
1058 | ||
6e5f1b57 LP |
1059 | if (architecture < 0) |
1060 | return NULL; | |
1061 | ||
1062 | return architecture_to_string(architecture); | |
7b3e062c LP |
1063 | } |
1064 | ||
21022b9d LP |
1065 | int safe_personality(unsigned long p) { |
1066 | int ret; | |
1067 | ||
1068 | /* So here's the deal, personality() is weirdly defined by glibc. In some cases it returns a failure via errno, | |
1069 | * and in others as negative return value containing an errno-like value. Let's work around this: this is a | |
1070 | * wrapper that uses errno if it is set, and uses the return value otherwise. And then it sets both errno and | |
1071 | * the return value indicating the same issue, so that we are definitely on the safe side. | |
1072 | * | |
1073 | * See https://github.com/systemd/systemd/issues/6737 */ | |
1074 | ||
1075 | errno = 0; | |
1076 | ret = personality(p); | |
1077 | if (ret < 0) { | |
1078 | if (errno != 0) | |
1079 | return -errno; | |
1080 | ||
1081 | errno = -ret; | |
1082 | } | |
1083 | ||
1084 | return ret; | |
1085 | } | |
1086 | ||
e8132d63 LP |
1087 | int opinionated_personality(unsigned long *ret) { |
1088 | int current; | |
1089 | ||
1090 | /* Returns the current personality, or PERSONALITY_INVALID if we can't determine it. This function is a bit | |
1091 | * opinionated though, and ignores all the finer-grained bits and exotic personalities, only distinguishing the | |
1092 | * two most relevant personalities: PER_LINUX and PER_LINUX32. */ | |
1093 | ||
21022b9d | 1094 | current = safe_personality(PERSONALITY_INVALID); |
e8132d63 | 1095 | if (current < 0) |
21022b9d | 1096 | return current; |
e8132d63 LP |
1097 | |
1098 | if (((unsigned long) current & 0xffff) == PER_LINUX32) | |
1099 | *ret = PER_LINUX32; | |
1100 | else | |
1101 | *ret = PER_LINUX; | |
1102 | ||
1103 | return 0; | |
1104 | } | |
1105 | ||
dcadc967 | 1106 | void valgrind_summary_hack(void) { |
349cc4a5 | 1107 | #if HAVE_VALGRIND_VALGRIND_H |
df0ff127 | 1108 | if (getpid_cached() == 1 && RUNNING_ON_VALGRIND) { |
dcadc967 | 1109 | pid_t pid; |
8869a0b4 | 1110 | pid = raw_clone(SIGCHLD); |
dcadc967 EV |
1111 | if (pid < 0) |
1112 | log_emergency_errno(errno, "Failed to fork off valgrind helper: %m"); | |
1113 | else if (pid == 0) | |
1114 | exit(EXIT_SUCCESS); | |
1115 | else { | |
1116 | log_info("Spawned valgrind helper as PID "PID_FMT".", pid); | |
1117 | (void) wait_for_terminate(pid, NULL); | |
1118 | } | |
1119 | } | |
1120 | #endif | |
1121 | } | |
1122 | ||
93bab288 | 1123 | int pid_compare_func(const pid_t *a, const pid_t *b) { |
291d565a | 1124 | /* Suitable for usage in qsort() */ |
93bab288 | 1125 | return CMP(*a, *b); |
291d565a LP |
1126 | } |
1127 | ||
7f452159 LP |
1128 | int ioprio_parse_priority(const char *s, int *ret) { |
1129 | int i, r; | |
1130 | ||
1131 | assert(s); | |
1132 | assert(ret); | |
1133 | ||
1134 | r = safe_atoi(s, &i); | |
1135 | if (r < 0) | |
1136 | return r; | |
1137 | ||
1138 | if (!ioprio_priority_is_valid(i)) | |
1139 | return -EINVAL; | |
1140 | ||
1141 | *ret = i; | |
1142 | return 0; | |
1143 | } | |
1144 | ||
5c30a6d2 LP |
1145 | /* The cached PID, possible values: |
1146 | * | |
1147 | * == UNSET [0] → cache not initialized yet | |
1148 | * == BUSY [-1] → some thread is initializing it at the moment | |
1149 | * any other → the cached PID | |
1150 | */ | |
1151 | ||
1152 | #define CACHED_PID_UNSET ((pid_t) 0) | |
1153 | #define CACHED_PID_BUSY ((pid_t) -1) | |
1154 | ||
1155 | static pid_t cached_pid = CACHED_PID_UNSET; | |
1156 | ||
799a960d | 1157 | void reset_cached_pid(void) { |
5c30a6d2 LP |
1158 | /* Invoked in the child after a fork(), i.e. at the first moment the PID changed */ |
1159 | cached_pid = CACHED_PID_UNSET; | |
1160 | } | |
1161 | ||
1162 | /* We use glibc __register_atfork() + __dso_handle directly here, as they are not included in the glibc | |
1163 | * headers. __register_atfork() is mostly equivalent to pthread_atfork(), but doesn't require us to link against | |
1164 | * libpthread, as it is part of glibc anyway. */ | |
2bb8d8d9 | 1165 | extern int __register_atfork(void (*prepare) (void), void (*parent) (void), void (*child) (void), void *dso_handle); |
d34dae18 | 1166 | extern void* __dso_handle _weak_; |
5c30a6d2 LP |
1167 | |
1168 | pid_t getpid_cached(void) { | |
5d71bac3 | 1169 | static bool installed = false; |
5c30a6d2 LP |
1170 | pid_t current_value; |
1171 | ||
1172 | /* getpid_cached() is much like getpid(), but caches the value in local memory, to avoid having to invoke a | |
1173 | * system call each time. This restores glibc behaviour from before 2.24, when getpid() was unconditionally | |
1174 | * cached. Starting with 2.24 getpid() started to become prohibitively expensive when used for detecting when | |
1175 | * objects were used across fork()s. With this caching the old behaviour is somewhat restored. | |
1176 | * | |
1177 | * https://bugzilla.redhat.com/show_bug.cgi?id=1443976 | |
a4041e4f | 1178 | * https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=c579f48edba88380635ab98cb612030e3ed8691e |
5c30a6d2 LP |
1179 | */ |
1180 | ||
1181 | current_value = __sync_val_compare_and_swap(&cached_pid, CACHED_PID_UNSET, CACHED_PID_BUSY); | |
1182 | ||
1183 | switch (current_value) { | |
1184 | ||
1185 | case CACHED_PID_UNSET: { /* Not initialized yet, then do so now */ | |
1186 | pid_t new_pid; | |
1187 | ||
996def17 | 1188 | new_pid = raw_getpid(); |
5c30a6d2 | 1189 | |
5d71bac3 LP |
1190 | if (!installed) { |
1191 | /* __register_atfork() either returns 0 or -ENOMEM, in its glibc implementation. Since it's | |
1192 | * only half-documented (glibc doesn't document it but LSB does — though only superficially) | |
1193 | * we'll check for errors only in the most generic fashion possible. */ | |
1194 | ||
1195 | if (__register_atfork(NULL, NULL, reset_cached_pid, __dso_handle) != 0) { | |
1196 | /* OOM? Let's try again later */ | |
1197 | cached_pid = CACHED_PID_UNSET; | |
1198 | return new_pid; | |
1199 | } | |
1200 | ||
1201 | installed = true; | |
5c30a6d2 LP |
1202 | } |
1203 | ||
1204 | cached_pid = new_pid; | |
1205 | return new_pid; | |
1206 | } | |
1207 | ||
1208 | case CACHED_PID_BUSY: /* Somebody else is currently initializing */ | |
996def17 | 1209 | return raw_getpid(); |
5c30a6d2 LP |
1210 | |
1211 | default: /* Properly initialized */ | |
1212 | return current_value; | |
1213 | } | |
1214 | } | |
1215 | ||
fba868fa LP |
1216 | int must_be_root(void) { |
1217 | ||
1218 | if (geteuid() == 0) | |
1219 | return 0; | |
1220 | ||
baaa35ad | 1221 | return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Need to be root."); |
fba868fa LP |
1222 | } |
1223 | ||
4c253ed1 LP |
1224 | int safe_fork_full( |
1225 | const char *name, | |
1226 | const int except_fds[], | |
1227 | size_t n_except_fds, | |
1228 | ForkFlags flags, | |
1229 | pid_t *ret_pid) { | |
1230 | ||
1231 | pid_t original_pid, pid; | |
1f5d1e02 | 1232 | sigset_t saved_ss, ss; |
a41f6217 | 1233 | bool block_signals = false; |
b6e1fff1 | 1234 | int prio, r; |
4c253ed1 LP |
1235 | |
1236 | /* A wrapper around fork(), that does a couple of important initializations in addition to mere forking. Always | |
1237 | * returns the child's PID in *ret_pid. Returns == 0 in the child, and > 0 in the parent. */ | |
1238 | ||
b6e1fff1 LP |
1239 | prio = flags & FORK_LOG ? LOG_ERR : LOG_DEBUG; |
1240 | ||
4c253ed1 LP |
1241 | original_pid = getpid_cached(); |
1242 | ||
1f5d1e02 | 1243 | if (flags & (FORK_RESET_SIGNALS|FORK_DEATHSIG)) { |
1f5d1e02 LP |
1244 | /* We temporarily block all signals, so that the new child has them blocked initially. This way, we can |
1245 | * be sure that SIGTERMs are not lost we might send to the child. */ | |
4c253ed1 | 1246 | |
cd2a429e | 1247 | assert_se(sigfillset(&ss) >= 0); |
1f5d1e02 LP |
1248 | block_signals = true; |
1249 | ||
1250 | } else if (flags & FORK_WAIT) { | |
1f5d1e02 LP |
1251 | /* Let's block SIGCHLD at least, so that we can safely watch for the child process */ |
1252 | ||
cd2a429e ZJS |
1253 | assert_se(sigemptyset(&ss) >= 0); |
1254 | assert_se(sigaddset(&ss, SIGCHLD) >= 0); | |
1f5d1e02 | 1255 | block_signals = true; |
4c253ed1 LP |
1256 | } |
1257 | ||
1f5d1e02 LP |
1258 | if (block_signals) |
1259 | if (sigprocmask(SIG_SETMASK, &ss, &saved_ss) < 0) | |
1260 | return log_full_errno(prio, errno, "Failed to set signal mask: %m"); | |
1261 | ||
be39f6ee LP |
1262 | if (flags & FORK_NEW_MOUNTNS) |
1263 | pid = raw_clone(SIGCHLD|CLONE_NEWNS); | |
1264 | else | |
1265 | pid = fork(); | |
4c253ed1 LP |
1266 | if (pid < 0) { |
1267 | r = -errno; | |
1268 | ||
1269 | if (block_signals) /* undo what we did above */ | |
1270 | (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL); | |
1271 | ||
b6e1fff1 | 1272 | return log_full_errno(prio, r, "Failed to fork: %m"); |
4c253ed1 LP |
1273 | } |
1274 | if (pid > 0) { | |
1275 | /* We are in the parent process */ | |
1276 | ||
1f5d1e02 LP |
1277 | log_debug("Successfully forked off '%s' as PID " PID_FMT ".", strna(name), pid); |
1278 | ||
1279 | if (flags & FORK_WAIT) { | |
1280 | r = wait_for_terminate_and_check(name, pid, (flags & FORK_LOG ? WAIT_LOG : 0)); | |
1281 | if (r < 0) | |
1282 | return r; | |
1283 | if (r != EXIT_SUCCESS) /* exit status > 0 should be treated as failure, too */ | |
1284 | return -EPROTO; | |
1285 | } | |
1286 | ||
4c253ed1 LP |
1287 | if (block_signals) /* undo what we did above */ |
1288 | (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL); | |
1289 | ||
4c253ed1 LP |
1290 | if (ret_pid) |
1291 | *ret_pid = pid; | |
1292 | ||
1293 | return 1; | |
1294 | } | |
1295 | ||
1296 | /* We are in the child process */ | |
1297 | ||
1298 | if (flags & FORK_REOPEN_LOG) { | |
1299 | /* Close the logs if requested, before we log anything. And make sure we reopen it if needed. */ | |
1300 | log_close(); | |
1301 | log_set_open_when_needed(true); | |
1302 | } | |
1303 | ||
1304 | if (name) { | |
1305 | r = rename_process(name); | |
1306 | if (r < 0) | |
b6e1fff1 LP |
1307 | log_full_errno(flags & FORK_LOG ? LOG_WARNING : LOG_DEBUG, |
1308 | r, "Failed to rename process, ignoring: %m"); | |
4c253ed1 LP |
1309 | } |
1310 | ||
1311 | if (flags & FORK_DEATHSIG) | |
1312 | if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0) { | |
b6e1fff1 | 1313 | log_full_errno(prio, errno, "Failed to set death signal: %m"); |
4c253ed1 LP |
1314 | _exit(EXIT_FAILURE); |
1315 | } | |
1316 | ||
1317 | if (flags & FORK_RESET_SIGNALS) { | |
1318 | r = reset_all_signal_handlers(); | |
1319 | if (r < 0) { | |
b6e1fff1 | 1320 | log_full_errno(prio, r, "Failed to reset signal handlers: %m"); |
4c253ed1 LP |
1321 | _exit(EXIT_FAILURE); |
1322 | } | |
1323 | ||
1324 | /* This implicitly undoes the signal mask stuff we did before the fork()ing above */ | |
1325 | r = reset_signal_mask(); | |
1326 | if (r < 0) { | |
b6e1fff1 | 1327 | log_full_errno(prio, r, "Failed to reset signal mask: %m"); |
4c253ed1 LP |
1328 | _exit(EXIT_FAILURE); |
1329 | } | |
1330 | } else if (block_signals) { /* undo what we did above */ | |
1331 | if (sigprocmask(SIG_SETMASK, &saved_ss, NULL) < 0) { | |
b6e1fff1 | 1332 | log_full_errno(prio, errno, "Failed to restore signal mask: %m"); |
4c253ed1 LP |
1333 | _exit(EXIT_FAILURE); |
1334 | } | |
1335 | } | |
1336 | ||
1337 | if (flags & FORK_DEATHSIG) { | |
7ddc2dc5 | 1338 | pid_t ppid; |
4c253ed1 LP |
1339 | /* Let's see if the parent PID is still the one we started from? If not, then the parent |
1340 | * already died by the time we set PR_SET_PDEATHSIG, hence let's emulate the effect */ | |
1341 | ||
7ddc2dc5 SL |
1342 | ppid = getppid(); |
1343 | if (ppid == 0) | |
1344 | /* Parent is in a differn't PID namespace. */; | |
1345 | else if (ppid != original_pid) { | |
4c253ed1 LP |
1346 | log_debug("Parent died early, raising SIGTERM."); |
1347 | (void) raise(SIGTERM); | |
1348 | _exit(EXIT_FAILURE); | |
1349 | } | |
1350 | } | |
1351 | ||
d94a24ca | 1352 | if (FLAGS_SET(flags, FORK_NEW_MOUNTNS | FORK_MOUNTNS_SLAVE)) { |
e2047ba9 LP |
1353 | |
1354 | /* Optionally, make sure we never propagate mounts to the host. */ | |
1355 | ||
1356 | if (mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL) < 0) { | |
1357 | log_full_errno(prio, errno, "Failed to remount root directory as MS_SLAVE: %m"); | |
1358 | _exit(EXIT_FAILURE); | |
1359 | } | |
1360 | } | |
1361 | ||
4c253ed1 LP |
1362 | if (flags & FORK_CLOSE_ALL_FDS) { |
1363 | /* Close the logs here in case it got reopened above, as close_all_fds() would close them for us */ | |
1364 | log_close(); | |
1365 | ||
1366 | r = close_all_fds(except_fds, n_except_fds); | |
1367 | if (r < 0) { | |
b6e1fff1 | 1368 | log_full_errno(prio, r, "Failed to close all file descriptors: %m"); |
4c253ed1 LP |
1369 | _exit(EXIT_FAILURE); |
1370 | } | |
1371 | } | |
1372 | ||
1373 | /* When we were asked to reopen the logs, do so again now */ | |
1374 | if (flags & FORK_REOPEN_LOG) { | |
1375 | log_open(); | |
1376 | log_set_open_when_needed(false); | |
1377 | } | |
1378 | ||
1379 | if (flags & FORK_NULL_STDIO) { | |
1380 | r = make_null_stdio(); | |
1381 | if (r < 0) { | |
b6e1fff1 | 1382 | log_full_errno(prio, r, "Failed to connect stdin/stdout to /dev/null: %m"); |
4c253ed1 LP |
1383 | _exit(EXIT_FAILURE); |
1384 | } | |
1385 | } | |
1386 | ||
909106eb LP |
1387 | if (flags & FORK_RLIMIT_NOFILE_SAFE) { |
1388 | r = rlimit_nofile_safe(); | |
1389 | if (r < 0) { | |
1390 | log_full_errno(prio, r, "Failed to lower RLIMIT_NOFILE's soft limit to 1K: %m"); | |
1391 | _exit(EXIT_FAILURE); | |
1392 | } | |
1393 | } | |
1394 | ||
4c253ed1 LP |
1395 | if (ret_pid) |
1396 | *ret_pid = getpid_cached(); | |
1397 | ||
1398 | return 0; | |
1399 | } | |
1400 | ||
27096982 LP |
1401 | int namespace_fork( |
1402 | const char *outer_name, | |
1403 | const char *inner_name, | |
1404 | const int except_fds[], | |
1405 | size_t n_except_fds, | |
1406 | ForkFlags flags, | |
1407 | int pidns_fd, | |
1408 | int mntns_fd, | |
1409 | int netns_fd, | |
1410 | int userns_fd, | |
1411 | int root_fd, | |
1412 | pid_t *ret_pid) { | |
1413 | ||
1414 | int r; | |
1415 | ||
1416 | /* This is much like safe_fork(), but forks twice, and joins the specified namespaces in the middle | |
1417 | * process. This ensures that we are fully a member of the destination namespace, with pidns an all, so that | |
1418 | * /proc/self/fd works correctly. */ | |
1419 | ||
1420 | 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); | |
1421 | if (r < 0) | |
1422 | return r; | |
1423 | if (r == 0) { | |
1424 | pid_t pid; | |
1425 | ||
1426 | /* Child */ | |
1427 | ||
1428 | r = namespace_enter(pidns_fd, mntns_fd, netns_fd, userns_fd, root_fd); | |
1429 | if (r < 0) { | |
1430 | log_full_errno(FLAGS_SET(flags, FORK_LOG) ? LOG_ERR : LOG_DEBUG, r, "Failed to join namespace: %m"); | |
1431 | _exit(EXIT_FAILURE); | |
1432 | } | |
1433 | ||
1434 | /* We mask a few flags here that either make no sense for the grandchild, or that we don't have to do again */ | |
1435 | 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); | |
1436 | if (r < 0) | |
1437 | _exit(EXIT_FAILURE); | |
1438 | if (r == 0) { | |
1439 | /* Child */ | |
1440 | if (ret_pid) | |
1441 | *ret_pid = pid; | |
1442 | return 0; | |
1443 | } | |
1444 | ||
1445 | r = wait_for_terminate_and_check(inner_name, pid, FLAGS_SET(flags, FORK_LOG) ? WAIT_LOG : 0); | |
1446 | if (r < 0) | |
1447 | _exit(EXIT_FAILURE); | |
1448 | ||
1449 | _exit(r); | |
1450 | } | |
1451 | ||
1452 | return 1; | |
1453 | } | |
1454 | ||
da6053d0 | 1455 | int fork_agent(const char *name, const int except[], size_t n_except, pid_t *ret_pid, const char *path, ...) { |
78752f2e | 1456 | bool stdout_is_tty, stderr_is_tty; |
da6053d0 | 1457 | size_t n, i; |
78752f2e LP |
1458 | va_list ap; |
1459 | char **l; | |
1460 | int r; | |
1461 | ||
1462 | assert(path); | |
1463 | ||
1464 | /* Spawns a temporary TTY agent, making sure it goes away when we go away */ | |
1465 | ||
1466 | r = safe_fork_full(name, except, n_except, FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_CLOSE_ALL_FDS, ret_pid); | |
1467 | if (r < 0) | |
1468 | return r; | |
1469 | if (r > 0) | |
1470 | return 0; | |
1471 | ||
1472 | /* In the child: */ | |
1473 | ||
1474 | stdout_is_tty = isatty(STDOUT_FILENO); | |
1475 | stderr_is_tty = isatty(STDERR_FILENO); | |
1476 | ||
1477 | if (!stdout_is_tty || !stderr_is_tty) { | |
1478 | int fd; | |
1479 | ||
1480 | /* Detach from stdout/stderr. and reopen | |
1481 | * /dev/tty for them. This is important to | |
1482 | * ensure that when systemctl is started via | |
1483 | * popen() or a similar call that expects to | |
1484 | * read EOF we actually do generate EOF and | |
1485 | * not delay this indefinitely by because we | |
1486 | * keep an unused copy of stdin around. */ | |
1487 | fd = open("/dev/tty", O_WRONLY); | |
1488 | if (fd < 0) { | |
1489 | log_error_errno(errno, "Failed to open /dev/tty: %m"); | |
1490 | _exit(EXIT_FAILURE); | |
1491 | } | |
1492 | ||
1493 | if (!stdout_is_tty && dup2(fd, STDOUT_FILENO) < 0) { | |
1494 | log_error_errno(errno, "Failed to dup2 /dev/tty: %m"); | |
1495 | _exit(EXIT_FAILURE); | |
1496 | } | |
1497 | ||
1498 | if (!stderr_is_tty && dup2(fd, STDERR_FILENO) < 0) { | |
1499 | log_error_errno(errno, "Failed to dup2 /dev/tty: %m"); | |
1500 | _exit(EXIT_FAILURE); | |
1501 | } | |
1502 | ||
e7685a77 | 1503 | safe_close_above_stdio(fd); |
78752f2e LP |
1504 | } |
1505 | ||
595225af LP |
1506 | (void) rlimit_nofile_safe(); |
1507 | ||
78752f2e LP |
1508 | /* Count arguments */ |
1509 | va_start(ap, path); | |
1510 | for (n = 0; va_arg(ap, char*); n++) | |
1511 | ; | |
1512 | va_end(ap); | |
1513 | ||
1514 | /* Allocate strv */ | |
cf409d15 | 1515 | l = newa(char*, n + 1); |
78752f2e LP |
1516 | |
1517 | /* Fill in arguments */ | |
1518 | va_start(ap, path); | |
1519 | for (i = 0; i <= n; i++) | |
1520 | l[i] = va_arg(ap, char*); | |
1521 | va_end(ap); | |
1522 | ||
1523 | execv(path, l); | |
1524 | _exit(EXIT_FAILURE); | |
1525 | } | |
1526 | ||
9f8168eb LP |
1527 | int set_oom_score_adjust(int value) { |
1528 | char t[DECIMAL_STR_MAX(int)]; | |
1529 | ||
1530 | sprintf(t, "%i", value); | |
1531 | ||
1532 | return write_string_file("/proc/self/oom_score_adj", t, | |
1533 | WRITE_STRING_FILE_VERIFY_ON_FAILURE|WRITE_STRING_FILE_DISABLE_BUFFER); | |
1534 | } | |
1535 | ||
7b3e062c LP |
1536 | static const char *const ioprio_class_table[] = { |
1537 | [IOPRIO_CLASS_NONE] = "none", | |
1538 | [IOPRIO_CLASS_RT] = "realtime", | |
1539 | [IOPRIO_CLASS_BE] = "best-effort", | |
f44b3035 | 1540 | [IOPRIO_CLASS_IDLE] = "idle", |
7b3e062c LP |
1541 | }; |
1542 | ||
10062bbc | 1543 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ioprio_class, int, IOPRIO_N_CLASSES); |
7b3e062c LP |
1544 | |
1545 | static const char *const sigchld_code_table[] = { | |
1546 | [CLD_EXITED] = "exited", | |
1547 | [CLD_KILLED] = "killed", | |
1548 | [CLD_DUMPED] = "dumped", | |
1549 | [CLD_TRAPPED] = "trapped", | |
1550 | [CLD_STOPPED] = "stopped", | |
1551 | [CLD_CONTINUED] = "continued", | |
1552 | }; | |
1553 | ||
1554 | DEFINE_STRING_TABLE_LOOKUP(sigchld_code, int); | |
1555 | ||
1556 | static const char* const sched_policy_table[] = { | |
1557 | [SCHED_OTHER] = "other", | |
1558 | [SCHED_BATCH] = "batch", | |
1559 | [SCHED_IDLE] = "idle", | |
1560 | [SCHED_FIFO] = "fifo", | |
f44b3035 | 1561 | [SCHED_RR] = "rr", |
7b3e062c LP |
1562 | }; |
1563 | ||
1564 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(sched_policy, int, INT_MAX); |