]>
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" |
93cc7779 | 38 | #include "signal-util.h" |
1359fffa | 39 | #include "stat-util.h" |
7b3e062c | 40 | #include "string-table.h" |
07630cea | 41 | #include "string-util.h" |
4c253ed1 | 42 | #include "terminal-util.h" |
b1d4f8e1 | 43 | #include "user-util.h" |
4f5dd394 | 44 | #include "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; |
c0534780 | 104 | char *k, *ans = NULL; |
0b452006 RC |
105 | const char *p; |
106 | int c; | |
107 | ||
108 | assert(line); | |
109 | assert(pid >= 0); | |
110 | ||
69281c49 LP |
111 | /* Retrieves a process' command line. Replaces unprintable characters while doing so by whitespace (coalescing |
112 | * multiple sequential ones into one). If max_length is != 0 will return a string of the specified size at most | |
113 | * (the trailing NUL byte does count towards the length here!), abbreviated with a "..." ellipsis. If | |
114 | * comm_fallback is true and the process has no command line set (the case for kernel threads), or has a | |
115 | * command line that resolves to the empty string will return the "comm" name of the process instead. | |
116 | * | |
117 | * Returns -ESRCH if the process doesn't exist, and -ENOENT if the process has no command line (and | |
c0534780 | 118 | * comm_fallback is false). Returns 0 and sets *line otherwise. */ |
69281c49 | 119 | |
0b452006 RC |
120 | p = procfs_file_alloca(pid, "cmdline"); |
121 | ||
122 | f = fopen(p, "re"); | |
a644184a LP |
123 | if (!f) { |
124 | if (errno == ENOENT) | |
125 | return -ESRCH; | |
0b452006 | 126 | return -errno; |
a644184a | 127 | } |
0b452006 | 128 | |
35bbbf85 LP |
129 | (void) __fsetlocking(f, FSETLOCKING_BYCALLER); |
130 | ||
69281c49 LP |
131 | if (max_length == 1) { |
132 | ||
133 | /* If there's only room for one byte, return the empty string */ | |
c0534780 ZJS |
134 | ans = new0(char, 1); |
135 | if (!ans) | |
69281c49 LP |
136 | return -ENOMEM; |
137 | ||
c0534780 | 138 | *line = ans; |
69281c49 LP |
139 | return 0; |
140 | ||
141 | } else if (max_length == 0) { | |
0b452006 RC |
142 | size_t len = 0, allocated = 0; |
143 | ||
144 | while ((c = getc(f)) != EOF) { | |
145 | ||
c0534780 ZJS |
146 | if (!GREEDY_REALLOC(ans, allocated, len+3)) { |
147 | free(ans); | |
0b452006 RC |
148 | return -ENOMEM; |
149 | } | |
150 | ||
ba4cd7e2 MP |
151 | if (isprint(c)) { |
152 | if (space) { | |
c0534780 | 153 | ans[len++] = ' '; |
ba4cd7e2 MP |
154 | space = false; |
155 | } | |
156 | ||
c0534780 | 157 | ans[len++] = c; |
69281c49 | 158 | } else if (len > 0) |
ba4cd7e2 MP |
159 | space = true; |
160 | } | |
0b452006 RC |
161 | |
162 | if (len > 0) | |
c0534780 | 163 | ans[len] = '\0'; |
b09df4e2 | 164 | else |
c0534780 | 165 | ans = mfree(ans); |
0b452006 RC |
166 | |
167 | } else { | |
69281c49 | 168 | bool dotdotdot = false; |
0b452006 RC |
169 | size_t left; |
170 | ||
c0534780 ZJS |
171 | ans = new(char, max_length); |
172 | if (!ans) | |
0b452006 RC |
173 | return -ENOMEM; |
174 | ||
c0534780 | 175 | k = ans; |
0b452006 RC |
176 | left = max_length; |
177 | while ((c = getc(f)) != EOF) { | |
178 | ||
179 | if (isprint(c)) { | |
69281c49 | 180 | |
0b452006 | 181 | if (space) { |
69281c49 LP |
182 | if (left <= 2) { |
183 | dotdotdot = true; | |
0b452006 | 184 | break; |
69281c49 | 185 | } |
0b452006 RC |
186 | |
187 | *(k++) = ' '; | |
188 | left--; | |
189 | space = false; | |
190 | } | |
191 | ||
69281c49 LP |
192 | if (left <= 1) { |
193 | dotdotdot = true; | |
0b452006 | 194 | break; |
69281c49 | 195 | } |
0b452006 RC |
196 | |
197 | *(k++) = (char) c; | |
198 | left--; | |
c0534780 | 199 | } else if (k > ans) |
0b452006 RC |
200 | space = true; |
201 | } | |
202 | ||
69281c49 LP |
203 | if (dotdotdot) { |
204 | if (max_length <= 4) { | |
c0534780 | 205 | k = ans; |
69281c49 LP |
206 | left = max_length; |
207 | } else { | |
c0534780 | 208 | k = ans + max_length - 4; |
69281c49 LP |
209 | left = 4; |
210 | ||
211 | /* Eat up final spaces */ | |
c0534780 | 212 | while (k > ans && isspace(k[-1])) { |
69281c49 LP |
213 | k--; |
214 | left++; | |
215 | } | |
216 | } | |
217 | ||
218 | strncpy(k, "...", left-1); | |
b09df4e2 | 219 | k[left-1] = 0; |
0b452006 RC |
220 | } else |
221 | *k = 0; | |
222 | } | |
223 | ||
224 | /* Kernel threads have no argv[] */ | |
c0534780 | 225 | if (isempty(ans)) { |
0b452006 RC |
226 | _cleanup_free_ char *t = NULL; |
227 | int h; | |
228 | ||
c0534780 | 229 | free(ans); |
0b452006 RC |
230 | |
231 | if (!comm_fallback) | |
232 | return -ENOENT; | |
233 | ||
234 | h = get_process_comm(pid, &t); | |
235 | if (h < 0) | |
236 | return h; | |
237 | ||
69281c49 | 238 | if (max_length == 0) |
c0534780 | 239 | ans = strjoin("[", t, "]"); |
69281c49 LP |
240 | else { |
241 | size_t l; | |
242 | ||
243 | l = strlen(t); | |
244 | ||
245 | if (l + 3 <= max_length) | |
c0534780 | 246 | ans = strjoin("[", t, "]"); |
69281c49 LP |
247 | else if (max_length <= 6) { |
248 | ||
c0534780 ZJS |
249 | ans = new(char, max_length); |
250 | if (!ans) | |
69281c49 LP |
251 | return -ENOMEM; |
252 | ||
c0534780 ZJS |
253 | memcpy(ans, "[...]", max_length-1); |
254 | ans[max_length-1] = 0; | |
69281c49 | 255 | } else { |
69281c49 LP |
256 | t[max_length - 6] = 0; |
257 | ||
258 | /* Chop off final spaces */ | |
af897494 | 259 | delete_trailing_chars(t, WHITESPACE); |
69281c49 | 260 | |
c0534780 | 261 | ans = strjoin("[", t, "...]"); |
69281c49 LP |
262 | } |
263 | } | |
c0534780 | 264 | if (!ans) |
0b452006 RC |
265 | return -ENOMEM; |
266 | } | |
267 | ||
c0534780 | 268 | *line = ans; |
0b452006 RC |
269 | return 0; |
270 | } | |
271 | ||
9bfaffd5 LP |
272 | int rename_process(const char name[]) { |
273 | static size_t mm_size = 0; | |
274 | static char *mm = NULL; | |
275 | bool truncated = false; | |
276 | size_t l; | |
277 | ||
278 | /* This is a like a poor man's setproctitle(). It changes the comm field, argv[0], and also the glibc's | |
279 | * internally used name of the process. For the first one a limit of 16 chars applies; to the second one in | |
280 | * many cases one of 10 (i.e. length of "/sbin/init") — however if we have CAP_SYS_RESOURCES it is unbounded; | |
281 | * to the third one 7 (i.e. the length of "systemd". If you pass a longer string it will likely be | |
282 | * truncated. | |
283 | * | |
284 | * Returns 0 if a name was set but truncated, > 0 if it was set but not truncated. */ | |
285 | ||
286 | if (isempty(name)) | |
287 | return -EINVAL; /* let's not confuse users unnecessarily with an empty name */ | |
405f8907 | 288 | |
1096bb8a LP |
289 | if (!is_main_thread()) |
290 | return -EPERM; /* Let's not allow setting the process name from other threads than the main one, as we | |
291 | * cache things without locking, and we make assumptions that PR_SET_NAME sets the | |
292 | * process name that isn't correct on any other threads */ | |
293 | ||
9bfaffd5 | 294 | l = strlen(name); |
405f8907 | 295 | |
5a8af747 LP |
296 | /* First step, change the comm field. The main thread's comm is identical to the process comm. This means we |
297 | * can use PR_SET_NAME, which sets the thread name for the calling thread. */ | |
298 | if (prctl(PR_SET_NAME, name) < 0) | |
299 | log_debug_errno(errno, "PR_SET_NAME failed: %m"); | |
92f14395 | 300 | if (l >= TASK_COMM_LEN) /* Linux process names can be 15 chars at max */ |
9bfaffd5 LP |
301 | truncated = true; |
302 | ||
303 | /* Second step, change glibc's ID of the process name. */ | |
304 | if (program_invocation_name) { | |
305 | size_t k; | |
306 | ||
307 | k = strlen(program_invocation_name); | |
308 | strncpy(program_invocation_name, name, k); | |
309 | if (l > k) | |
310 | truncated = true; | |
311 | } | |
312 | ||
313 | /* Third step, completely replace the argv[] array the kernel maintains for us. This requires privileges, but | |
314 | * has the advantage that the argv[] array is exactly what we want it to be, and not filled up with zeros at | |
13e785f7 | 315 | * the end. This is the best option for changing /proc/self/cmdline. */ |
01f989c6 JW |
316 | |
317 | /* Let's not bother with this if we don't have euid == 0. Strictly speaking we should check for the | |
318 | * CAP_SYS_RESOURCE capability which is independent of the euid. In our own code the capability generally is | |
319 | * present only for euid == 0, hence let's use this as quick bypass check, to avoid calling mmap() if | |
320 | * PR_SET_MM_ARG_{START,END} fails with EPERM later on anyway. After all geteuid() is dead cheap to call, but | |
321 | * mmap() is not. */ | |
322 | if (geteuid() != 0) | |
323 | log_debug("Skipping PR_SET_MM, as we don't have privileges."); | |
324 | else if (mm_size < l+1) { | |
9bfaffd5 LP |
325 | size_t nn_size; |
326 | char *nn; | |
327 | ||
9bfaffd5 LP |
328 | nn_size = PAGE_ALIGN(l+1); |
329 | nn = mmap(NULL, nn_size, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); | |
330 | if (nn == MAP_FAILED) { | |
331 | log_debug_errno(errno, "mmap() failed: %m"); | |
332 | goto use_saved_argv; | |
333 | } | |
405f8907 | 334 | |
9bfaffd5 LP |
335 | strncpy(nn, name, nn_size); |
336 | ||
337 | /* Now, let's tell the kernel about this new memory */ | |
338 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_START, (unsigned long) nn, 0, 0) < 0) { | |
14ee72b7 FS |
339 | /* HACK: prctl() API is kind of dumb on this point. The existing end address may already be |
340 | * below the desired start address, in which case the kernel may have kicked this back due | |
341 | * to a range-check failure (see linux/kernel/sys.c:validate_prctl_map() to see this in | |
342 | * action). The proper solution would be to have a prctl() API that could set both start+end | |
343 | * simultaneously, or at least let us query the existing address to anticipate this condition | |
344 | * and respond accordingly. For now, we can only guess at the cause of this failure and try | |
345 | * a workaround--which will briefly expand the arg space to something potentially huge before | |
346 | * resizing it to what we want. */ | |
347 | log_debug_errno(errno, "PR_SET_MM_ARG_START failed, attempting PR_SET_MM_ARG_END hack: %m"); | |
348 | ||
349 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) nn + l + 1, 0, 0) < 0) { | |
350 | log_debug_errno(errno, "PR_SET_MM_ARG_END hack failed, proceeding without: %m"); | |
351 | (void) munmap(nn, nn_size); | |
352 | goto use_saved_argv; | |
353 | } | |
9bfaffd5 | 354 | |
14ee72b7 FS |
355 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_START, (unsigned long) nn, 0, 0) < 0) { |
356 | log_debug_errno(errno, "PR_SET_MM_ARG_START still failed, proceeding without: %m"); | |
357 | goto use_saved_argv; | |
358 | } | |
359 | } else { | |
360 | /* And update the end pointer to the new end, too. If this fails, we don't really know what | |
361 | * to do, it's pretty unlikely that we can rollback, hence we'll just accept the failure, | |
362 | * and continue. */ | |
363 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) nn + l + 1, 0, 0) < 0) | |
364 | log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m"); | |
365 | } | |
9bfaffd5 LP |
366 | |
367 | if (mm) | |
368 | (void) munmap(mm, mm_size); | |
369 | ||
370 | mm = nn; | |
371 | mm_size = nn_size; | |
01f989c6 | 372 | } else { |
9bfaffd5 LP |
373 | strncpy(mm, name, mm_size); |
374 | ||
01f989c6 JW |
375 | /* Update the end pointer, continuing regardless of any failure. */ |
376 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) mm + l + 1, 0, 0) < 0) | |
377 | log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m"); | |
378 | } | |
379 | ||
9bfaffd5 LP |
380 | use_saved_argv: |
381 | /* Fourth step: in all cases we'll also update the original argv[], so that our own code gets it right too if | |
382 | * it still looks here */ | |
405f8907 LP |
383 | |
384 | if (saved_argc > 0) { | |
385 | int i; | |
386 | ||
9bfaffd5 LP |
387 | if (saved_argv[0]) { |
388 | size_t k; | |
389 | ||
390 | k = strlen(saved_argv[0]); | |
391 | strncpy(saved_argv[0], name, k); | |
392 | if (l > k) | |
393 | truncated = true; | |
394 | } | |
405f8907 LP |
395 | |
396 | for (i = 1; i < saved_argc; i++) { | |
397 | if (!saved_argv[i]) | |
398 | break; | |
399 | ||
400 | memzero(saved_argv[i], strlen(saved_argv[i])); | |
401 | } | |
402 | } | |
9bfaffd5 LP |
403 | |
404 | return !truncated; | |
405f8907 LP |
405 | } |
406 | ||
0b452006 | 407 | int is_kernel_thread(pid_t pid) { |
36b5119a LP |
408 | _cleanup_free_ char *line = NULL; |
409 | unsigned long long flags; | |
410 | size_t l, i; | |
0b452006 | 411 | const char *p; |
36b5119a LP |
412 | char *q; |
413 | int r; | |
0b452006 | 414 | |
4c701096 | 415 | if (IN_SET(pid, 0, 1) || pid == getpid_cached()) /* pid 1, and we ourselves certainly aren't a kernel thread */ |
0b452006 | 416 | return 0; |
36b5119a LP |
417 | if (!pid_is_valid(pid)) |
418 | return -EINVAL; | |
0b452006 | 419 | |
36b5119a LP |
420 | p = procfs_file_alloca(pid, "stat"); |
421 | r = read_one_line_file(p, &line); | |
422 | if (r == -ENOENT) | |
423 | return -ESRCH; | |
424 | if (r < 0) | |
425 | return r; | |
0b452006 | 426 | |
36b5119a LP |
427 | /* Skip past the comm field */ |
428 | q = strrchr(line, ')'); | |
429 | if (!q) | |
430 | return -EINVAL; | |
431 | q++; | |
432 | ||
433 | /* Skip 6 fields to reach the flags field */ | |
434 | for (i = 0; i < 6; i++) { | |
435 | l = strspn(q, WHITESPACE); | |
436 | if (l < 1) | |
437 | return -EINVAL; | |
438 | q += l; | |
439 | ||
440 | l = strcspn(q, WHITESPACE); | |
441 | if (l < 1) | |
442 | return -EINVAL; | |
443 | q += l; | |
a644184a | 444 | } |
0b452006 | 445 | |
f21f31b2 | 446 | /* Skip preceding whitespace */ |
36b5119a LP |
447 | l = strspn(q, WHITESPACE); |
448 | if (l < 1) | |
449 | return -EINVAL; | |
450 | q += l; | |
35bbbf85 | 451 | |
36b5119a LP |
452 | /* Truncate the rest */ |
453 | l = strcspn(q, WHITESPACE); | |
454 | if (l < 1) | |
455 | return -EINVAL; | |
456 | q[l] = 0; | |
0b452006 | 457 | |
36b5119a LP |
458 | r = safe_atollu(q, &flags); |
459 | if (r < 0) | |
460 | return r; | |
0b452006 | 461 | |
36b5119a | 462 | return !!(flags & PF_KTHREAD); |
0b452006 RC |
463 | } |
464 | ||
465 | int get_process_capeff(pid_t pid, char **capeff) { | |
466 | const char *p; | |
a644184a | 467 | int r; |
0b452006 RC |
468 | |
469 | assert(capeff); | |
470 | assert(pid >= 0); | |
471 | ||
472 | p = procfs_file_alloca(pid, "status"); | |
473 | ||
c4cd1d4d | 474 | r = get_proc_field(p, "CapEff", WHITESPACE, capeff); |
a644184a LP |
475 | if (r == -ENOENT) |
476 | return -ESRCH; | |
477 | ||
478 | return r; | |
0b452006 RC |
479 | } |
480 | ||
481 | static int get_process_link_contents(const char *proc_file, char **name) { | |
482 | int r; | |
483 | ||
484 | assert(proc_file); | |
485 | assert(name); | |
486 | ||
487 | r = readlink_malloc(proc_file, name); | |
a644184a LP |
488 | if (r == -ENOENT) |
489 | return -ESRCH; | |
0b452006 | 490 | if (r < 0) |
a644184a | 491 | return r; |
0b452006 RC |
492 | |
493 | return 0; | |
494 | } | |
495 | ||
496 | int get_process_exe(pid_t pid, char **name) { | |
497 | const char *p; | |
498 | char *d; | |
499 | int r; | |
500 | ||
501 | assert(pid >= 0); | |
502 | ||
503 | p = procfs_file_alloca(pid, "exe"); | |
504 | r = get_process_link_contents(p, name); | |
505 | if (r < 0) | |
506 | return r; | |
507 | ||
508 | d = endswith(*name, " (deleted)"); | |
509 | if (d) | |
510 | *d = '\0'; | |
511 | ||
512 | return 0; | |
513 | } | |
514 | ||
515 | static int get_process_id(pid_t pid, const char *field, uid_t *uid) { | |
516 | _cleanup_fclose_ FILE *f = NULL; | |
0b452006 | 517 | const char *p; |
7e7a16a0 | 518 | int r; |
0b452006 RC |
519 | |
520 | assert(field); | |
521 | assert(uid); | |
522 | ||
07b38ba5 | 523 | if (pid < 0) |
6f8cbcdb LP |
524 | return -EINVAL; |
525 | ||
0b452006 RC |
526 | p = procfs_file_alloca(pid, "status"); |
527 | f = fopen(p, "re"); | |
a644184a LP |
528 | if (!f) { |
529 | if (errno == ENOENT) | |
530 | return -ESRCH; | |
0b452006 | 531 | return -errno; |
a644184a | 532 | } |
0b452006 | 533 | |
35bbbf85 LP |
534 | (void) __fsetlocking(f, FSETLOCKING_BYCALLER); |
535 | ||
7e7a16a0 LP |
536 | for (;;) { |
537 | _cleanup_free_ char *line = NULL; | |
0b452006 RC |
538 | char *l; |
539 | ||
7e7a16a0 LP |
540 | r = read_line(f, LONG_LINE_MAX, &line); |
541 | if (r < 0) | |
542 | return r; | |
543 | if (r == 0) | |
544 | break; | |
545 | ||
0b452006 RC |
546 | l = strstrip(line); |
547 | ||
548 | if (startswith(l, field)) { | |
549 | l += strlen(field); | |
550 | l += strspn(l, WHITESPACE); | |
551 | ||
552 | l[strcspn(l, WHITESPACE)] = 0; | |
553 | ||
554 | return parse_uid(l, uid); | |
555 | } | |
556 | } | |
557 | ||
558 | return -EIO; | |
559 | } | |
560 | ||
561 | int get_process_uid(pid_t pid, uid_t *uid) { | |
6f8cbcdb LP |
562 | |
563 | if (pid == 0 || pid == getpid_cached()) { | |
564 | *uid = getuid(); | |
565 | return 0; | |
566 | } | |
567 | ||
0b452006 RC |
568 | return get_process_id(pid, "Uid:", uid); |
569 | } | |
570 | ||
571 | int get_process_gid(pid_t pid, gid_t *gid) { | |
6f8cbcdb LP |
572 | |
573 | if (pid == 0 || pid == getpid_cached()) { | |
574 | *gid = getgid(); | |
575 | return 0; | |
576 | } | |
577 | ||
0b452006 RC |
578 | assert_cc(sizeof(uid_t) == sizeof(gid_t)); |
579 | return get_process_id(pid, "Gid:", gid); | |
580 | } | |
581 | ||
582 | int get_process_cwd(pid_t pid, char **cwd) { | |
583 | const char *p; | |
584 | ||
585 | assert(pid >= 0); | |
586 | ||
587 | p = procfs_file_alloca(pid, "cwd"); | |
588 | ||
589 | return get_process_link_contents(p, cwd); | |
590 | } | |
591 | ||
592 | int get_process_root(pid_t pid, char **root) { | |
593 | const char *p; | |
594 | ||
595 | assert(pid >= 0); | |
596 | ||
597 | p = procfs_file_alloca(pid, "root"); | |
598 | ||
599 | return get_process_link_contents(p, root); | |
600 | } | |
601 | ||
602 | int get_process_environ(pid_t pid, char **env) { | |
603 | _cleanup_fclose_ FILE *f = NULL; | |
604 | _cleanup_free_ char *outcome = NULL; | |
605 | int c; | |
606 | const char *p; | |
607 | size_t allocated = 0, sz = 0; | |
608 | ||
609 | assert(pid >= 0); | |
610 | assert(env); | |
611 | ||
612 | p = procfs_file_alloca(pid, "environ"); | |
613 | ||
614 | f = fopen(p, "re"); | |
a644184a LP |
615 | if (!f) { |
616 | if (errno == ENOENT) | |
617 | return -ESRCH; | |
0b452006 | 618 | return -errno; |
a644184a | 619 | } |
0b452006 | 620 | |
35bbbf85 LP |
621 | (void) __fsetlocking(f, FSETLOCKING_BYCALLER); |
622 | ||
0b452006 RC |
623 | while ((c = fgetc(f)) != EOF) { |
624 | if (!GREEDY_REALLOC(outcome, allocated, sz + 5)) | |
625 | return -ENOMEM; | |
626 | ||
627 | if (c == '\0') | |
628 | outcome[sz++] = '\n'; | |
629 | else | |
630 | sz += cescape_char(c, outcome + sz); | |
631 | } | |
632 | ||
03c55bc0 LP |
633 | if (!outcome) { |
634 | outcome = strdup(""); | |
635 | if (!outcome) | |
636 | return -ENOMEM; | |
637 | } else | |
638 | outcome[sz] = '\0'; | |
de8763b6 | 639 | |
ae2a15bc | 640 | *env = TAKE_PTR(outcome); |
0b452006 RC |
641 | |
642 | return 0; | |
643 | } | |
644 | ||
6bc73acb | 645 | int get_process_ppid(pid_t pid, pid_t *_ppid) { |
0b452006 RC |
646 | int r; |
647 | _cleanup_free_ char *line = NULL; | |
648 | long unsigned ppid; | |
649 | const char *p; | |
650 | ||
651 | assert(pid >= 0); | |
652 | assert(_ppid); | |
653 | ||
6f8cbcdb | 654 | if (pid == 0 || pid == getpid_cached()) { |
0b452006 RC |
655 | *_ppid = getppid(); |
656 | return 0; | |
657 | } | |
658 | ||
659 | p = procfs_file_alloca(pid, "stat"); | |
660 | r = read_one_line_file(p, &line); | |
a644184a LP |
661 | if (r == -ENOENT) |
662 | return -ESRCH; | |
0b452006 RC |
663 | if (r < 0) |
664 | return r; | |
665 | ||
666 | /* Let's skip the pid and comm fields. The latter is enclosed | |
667 | * in () but does not escape any () in its value, so let's | |
668 | * skip over it manually */ | |
669 | ||
670 | p = strrchr(line, ')'); | |
671 | if (!p) | |
672 | return -EIO; | |
673 | ||
674 | p++; | |
675 | ||
676 | if (sscanf(p, " " | |
677 | "%*c " /* state */ | |
678 | "%lu ", /* ppid */ | |
679 | &ppid) != 1) | |
680 | return -EIO; | |
681 | ||
682 | if ((long unsigned) (pid_t) ppid != ppid) | |
683 | return -ERANGE; | |
684 | ||
685 | *_ppid = (pid_t) ppid; | |
686 | ||
687 | return 0; | |
688 | } | |
689 | ||
690 | int wait_for_terminate(pid_t pid, siginfo_t *status) { | |
691 | siginfo_t dummy; | |
692 | ||
693 | assert(pid >= 1); | |
694 | ||
695 | if (!status) | |
696 | status = &dummy; | |
697 | ||
698 | for (;;) { | |
699 | zero(*status); | |
700 | ||
701 | if (waitid(P_PID, pid, status, WEXITED) < 0) { | |
702 | ||
703 | if (errno == EINTR) | |
704 | continue; | |
705 | ||
3f0083a2 | 706 | return negative_errno(); |
0b452006 RC |
707 | } |
708 | ||
709 | return 0; | |
710 | } | |
711 | } | |
712 | ||
713 | /* | |
714 | * Return values: | |
715 | * < 0 : wait_for_terminate() failed to get the state of the | |
716 | * process, the process was terminated by a signal, or | |
717 | * failed for an unknown reason. | |
718 | * >=0 : The process terminated normally, and its exit code is | |
719 | * returned. | |
720 | * | |
721 | * That is, success is indicated by a return value of zero, and an | |
722 | * error is indicated by a non-zero value. | |
723 | * | |
724 | * A warning is emitted if the process terminates abnormally, | |
725 | * and also if it returns non-zero unless check_exit_code is true. | |
726 | */ | |
7d4904fe LP |
727 | int wait_for_terminate_and_check(const char *name, pid_t pid, WaitFlags flags) { |
728 | _cleanup_free_ char *buffer = NULL; | |
0b452006 | 729 | siginfo_t status; |
7d4904fe | 730 | int r, prio; |
0b452006 | 731 | |
0b452006 RC |
732 | assert(pid > 1); |
733 | ||
7d4904fe LP |
734 | if (!name) { |
735 | r = get_process_comm(pid, &buffer); | |
736 | if (r < 0) | |
737 | log_debug_errno(r, "Failed to acquire process name of " PID_FMT ", ignoring: %m", pid); | |
738 | else | |
739 | name = buffer; | |
740 | } | |
741 | ||
742 | prio = flags & WAIT_LOG_ABNORMAL ? LOG_ERR : LOG_DEBUG; | |
743 | ||
0b452006 RC |
744 | r = wait_for_terminate(pid, &status); |
745 | if (r < 0) | |
7d4904fe | 746 | return log_full_errno(prio, r, "Failed to wait for %s: %m", strna(name)); |
0b452006 RC |
747 | |
748 | if (status.si_code == CLD_EXITED) { | |
7d4904fe LP |
749 | if (status.si_status != EXIT_SUCCESS) |
750 | log_full(flags & WAIT_LOG_NON_ZERO_EXIT_STATUS ? LOG_ERR : LOG_DEBUG, | |
751 | "%s failed with exit status %i.", strna(name), status.si_status); | |
0b452006 RC |
752 | else |
753 | log_debug("%s succeeded.", name); | |
754 | ||
755 | return status.si_status; | |
7d4904fe | 756 | |
3742095b | 757 | } else if (IN_SET(status.si_code, CLD_KILLED, CLD_DUMPED)) { |
0b452006 | 758 | |
7d4904fe | 759 | log_full(prio, "%s terminated by signal %s.", strna(name), signal_to_string(status.si_status)); |
0b452006 RC |
760 | return -EPROTO; |
761 | } | |
762 | ||
7d4904fe | 763 | log_full(prio, "%s failed due to unknown reason.", strna(name)); |
0b452006 RC |
764 | return -EPROTO; |
765 | } | |
766 | ||
d5641e0d KW |
767 | /* |
768 | * Return values: | |
e225e5c3 LP |
769 | * |
770 | * < 0 : wait_for_terminate_with_timeout() failed to get the state of the process, the process timed out, the process | |
771 | * was terminated by a signal, or failed for an unknown reason. | |
772 | * | |
d5641e0d KW |
773 | * >=0 : The process terminated normally with no failures. |
774 | * | |
e225e5c3 LP |
775 | * Success is indicated by a return value of zero, a timeout is indicated by ETIMEDOUT, and all other child failure |
776 | * states are indicated by error is indicated by a non-zero value. | |
777 | * | |
778 | * This call assumes SIGCHLD has been blocked already, in particular before the child to wait for has been forked off | |
779 | * to remain entirely race-free. | |
d5641e0d KW |
780 | */ |
781 | int wait_for_terminate_with_timeout(pid_t pid, usec_t timeout) { | |
782 | sigset_t mask; | |
783 | int r; | |
784 | usec_t until; | |
785 | ||
786 | assert_se(sigemptyset(&mask) == 0); | |
787 | assert_se(sigaddset(&mask, SIGCHLD) == 0); | |
788 | ||
789 | /* Drop into a sigtimewait-based timeout. Waiting for the | |
790 | * pid to exit. */ | |
791 | until = now(CLOCK_MONOTONIC) + timeout; | |
792 | for (;;) { | |
793 | usec_t n; | |
794 | siginfo_t status = {}; | |
795 | struct timespec ts; | |
796 | ||
797 | n = now(CLOCK_MONOTONIC); | |
798 | if (n >= until) | |
799 | break; | |
800 | ||
801 | r = sigtimedwait(&mask, NULL, timespec_store(&ts, until - n)) < 0 ? -errno : 0; | |
802 | /* Assuming we woke due to the child exiting. */ | |
803 | if (waitid(P_PID, pid, &status, WEXITED|WNOHANG) == 0) { | |
804 | if (status.si_pid == pid) { | |
805 | /* This is the correct child.*/ | |
806 | if (status.si_code == CLD_EXITED) | |
807 | return (status.si_status == 0) ? 0 : -EPROTO; | |
808 | else | |
809 | return -EPROTO; | |
810 | } | |
811 | } | |
812 | /* Not the child, check for errors and proceed appropriately */ | |
813 | if (r < 0) { | |
814 | switch (r) { | |
815 | case -EAGAIN: | |
816 | /* Timed out, child is likely hung. */ | |
817 | return -ETIMEDOUT; | |
818 | case -EINTR: | |
819 | /* Received a different signal and should retry */ | |
820 | continue; | |
821 | default: | |
822 | /* Return any unexpected errors */ | |
823 | return r; | |
824 | } | |
825 | } | |
826 | } | |
827 | ||
828 | return -EPROTO; | |
829 | } | |
830 | ||
89c9030d LP |
831 | void sigkill_wait(pid_t pid) { |
832 | assert(pid > 1); | |
833 | ||
53640e6f | 834 | if (kill(pid, SIGKILL) >= 0) |
89c9030d LP |
835 | (void) wait_for_terminate(pid, NULL); |
836 | } | |
837 | ||
838 | void sigkill_waitp(pid_t *pid) { | |
dfd14786 LP |
839 | PROTECT_ERRNO; |
840 | ||
4d0d3d41 LP |
841 | if (!pid) |
842 | return; | |
843 | if (*pid <= 1) | |
844 | return; | |
845 | ||
89c9030d | 846 | sigkill_wait(*pid); |
4d0d3d41 LP |
847 | } |
848 | ||
392cf1d0 SL |
849 | void sigterm_wait(pid_t pid) { |
850 | assert(pid > 1); | |
851 | ||
53640e6f | 852 | if (kill_and_sigcont(pid, SIGTERM) >= 0) |
392cf1d0 SL |
853 | (void) wait_for_terminate(pid, NULL); |
854 | } | |
855 | ||
0b452006 RC |
856 | int kill_and_sigcont(pid_t pid, int sig) { |
857 | int r; | |
858 | ||
859 | r = kill(pid, sig) < 0 ? -errno : 0; | |
860 | ||
26f417d3 LP |
861 | /* If this worked, also send SIGCONT, unless we already just sent a SIGCONT, or SIGKILL was sent which isn't |
862 | * affected by a process being suspended anyway. */ | |
a3d8d68c | 863 | if (r >= 0 && !IN_SET(sig, SIGCONT, SIGKILL)) |
26f417d3 | 864 | (void) kill(pid, SIGCONT); |
0b452006 RC |
865 | |
866 | return r; | |
867 | } | |
868 | ||
e70f4453 | 869 | int getenv_for_pid(pid_t pid, const char *field, char **ret) { |
0b452006 RC |
870 | _cleanup_fclose_ FILE *f = NULL; |
871 | char *value = NULL; | |
0b452006 | 872 | bool done = false; |
0b452006 | 873 | const char *path; |
e70f4453 | 874 | size_t l; |
0b452006 RC |
875 | |
876 | assert(pid >= 0); | |
877 | assert(field); | |
e70f4453 LP |
878 | assert(ret); |
879 | ||
880 | if (pid == 0 || pid == getpid_cached()) { | |
881 | const char *e; | |
882 | ||
883 | e = getenv(field); | |
884 | if (!e) { | |
885 | *ret = NULL; | |
886 | return 0; | |
887 | } | |
888 | ||
889 | value = strdup(e); | |
890 | if (!value) | |
891 | return -ENOMEM; | |
892 | ||
893 | *ret = value; | |
894 | return 1; | |
895 | } | |
0b452006 RC |
896 | |
897 | path = procfs_file_alloca(pid, "environ"); | |
898 | ||
899 | f = fopen(path, "re"); | |
a644184a LP |
900 | if (!f) { |
901 | if (errno == ENOENT) | |
902 | return -ESRCH; | |
e70f4453 | 903 | |
0b452006 | 904 | return -errno; |
a644184a | 905 | } |
0b452006 | 906 | |
35bbbf85 LP |
907 | (void) __fsetlocking(f, FSETLOCKING_BYCALLER); |
908 | ||
0b452006 | 909 | l = strlen(field); |
0b452006 RC |
910 | |
911 | do { | |
912 | char line[LINE_MAX]; | |
da6053d0 | 913 | size_t i; |
0b452006 RC |
914 | |
915 | for (i = 0; i < sizeof(line)-1; i++) { | |
916 | int c; | |
917 | ||
918 | c = getc(f); | |
919 | if (_unlikely_(c == EOF)) { | |
920 | done = true; | |
921 | break; | |
922 | } else if (c == 0) | |
923 | break; | |
924 | ||
925 | line[i] = c; | |
926 | } | |
927 | line[i] = 0; | |
928 | ||
041b5ae1 | 929 | if (strneq(line, field, l) && line[l] == '=') { |
0b452006 RC |
930 | value = strdup(line + l + 1); |
931 | if (!value) | |
932 | return -ENOMEM; | |
933 | ||
e70f4453 LP |
934 | *ret = value; |
935 | return 1; | |
0b452006 RC |
936 | } |
937 | ||
938 | } while (!done); | |
939 | ||
e70f4453 LP |
940 | *ret = NULL; |
941 | return 0; | |
0b452006 RC |
942 | } |
943 | ||
944 | bool pid_is_unwaited(pid_t pid) { | |
945 | /* Checks whether a PID is still valid at all, including a zombie */ | |
946 | ||
07b38ba5 | 947 | if (pid < 0) |
0b452006 RC |
948 | return false; |
949 | ||
5fd9b2c5 LP |
950 | if (pid <= 1) /* If we or PID 1 would be dead and have been waited for, this code would not be running */ |
951 | return true; | |
952 | ||
6f8cbcdb LP |
953 | if (pid == getpid_cached()) |
954 | return true; | |
955 | ||
0b452006 RC |
956 | if (kill(pid, 0) >= 0) |
957 | return true; | |
958 | ||
959 | return errno != ESRCH; | |
960 | } | |
961 | ||
962 | bool pid_is_alive(pid_t pid) { | |
963 | int r; | |
964 | ||
965 | /* Checks whether a PID is still valid and not a zombie */ | |
966 | ||
07b38ba5 | 967 | if (pid < 0) |
0b452006 RC |
968 | return false; |
969 | ||
5fd9b2c5 LP |
970 | if (pid <= 1) /* If we or PID 1 would be a zombie, this code would not be running */ |
971 | return true; | |
972 | ||
6f8cbcdb LP |
973 | if (pid == getpid_cached()) |
974 | return true; | |
975 | ||
0b452006 | 976 | r = get_process_state(pid); |
4c701096 | 977 | if (IN_SET(r, -ESRCH, 'Z')) |
0b452006 RC |
978 | return false; |
979 | ||
980 | return true; | |
981 | } | |
d4510856 | 982 | |
1359fffa MS |
983 | int pid_from_same_root_fs(pid_t pid) { |
984 | const char *root; | |
985 | ||
07b38ba5 | 986 | if (pid < 0) |
6f8cbcdb LP |
987 | return false; |
988 | ||
989 | if (pid == 0 || pid == getpid_cached()) | |
990 | return true; | |
1359fffa MS |
991 | |
992 | root = procfs_file_alloca(pid, "root"); | |
993 | ||
e3f791a2 | 994 | return files_same(root, "/proc/1/root", 0); |
1359fffa MS |
995 | } |
996 | ||
d4510856 LP |
997 | bool is_main_thread(void) { |
998 | static thread_local int cached = 0; | |
999 | ||
1000 | if (_unlikely_(cached == 0)) | |
df0ff127 | 1001 | cached = getpid_cached() == gettid() ? 1 : -1; |
d4510856 LP |
1002 | |
1003 | return cached > 0; | |
1004 | } | |
7b3e062c | 1005 | |
848e863a | 1006 | _noreturn_ void freeze(void) { |
7b3e062c | 1007 | |
3da48d7a EV |
1008 | log_close(); |
1009 | ||
7b3e062c LP |
1010 | /* Make sure nobody waits for us on a socket anymore */ |
1011 | close_all_fds(NULL, 0); | |
1012 | ||
1013 | sync(); | |
1014 | ||
8647283e MS |
1015 | /* Let's not freeze right away, but keep reaping zombies. */ |
1016 | for (;;) { | |
1017 | int r; | |
1018 | siginfo_t si = {}; | |
1019 | ||
1020 | r = waitid(P_ALL, 0, &si, WEXITED); | |
1021 | if (r < 0 && errno != EINTR) | |
1022 | break; | |
1023 | } | |
1024 | ||
1025 | /* waitid() failed with an unexpected error, things are really borked. Freeze now! */ | |
7b3e062c LP |
1026 | for (;;) |
1027 | pause(); | |
1028 | } | |
1029 | ||
1030 | bool oom_score_adjust_is_valid(int oa) { | |
1031 | return oa >= OOM_SCORE_ADJ_MIN && oa <= OOM_SCORE_ADJ_MAX; | |
1032 | } | |
1033 | ||
1034 | unsigned long personality_from_string(const char *p) { | |
6e5f1b57 | 1035 | int architecture; |
7b3e062c | 1036 | |
0c0fea07 LP |
1037 | if (!p) |
1038 | return PERSONALITY_INVALID; | |
1039 | ||
6e5f1b57 LP |
1040 | /* Parse a personality specifier. We use our own identifiers that indicate specific ABIs, rather than just |
1041 | * hints regarding the register size, since we want to keep things open for multiple locally supported ABIs for | |
1042 | * the same register size. */ | |
1043 | ||
1044 | architecture = architecture_from_string(p); | |
1045 | if (architecture < 0) | |
1046 | return PERSONALITY_INVALID; | |
7b3e062c | 1047 | |
0c0fea07 | 1048 | if (architecture == native_architecture()) |
7b3e062c | 1049 | return PER_LINUX; |
0c0fea07 LP |
1050 | #ifdef SECONDARY_ARCHITECTURE |
1051 | if (architecture == SECONDARY_ARCHITECTURE) | |
f2d1736c | 1052 | return PER_LINUX32; |
7b3e062c LP |
1053 | #endif |
1054 | ||
1055 | return PERSONALITY_INVALID; | |
1056 | } | |
1057 | ||
1058 | const char* personality_to_string(unsigned long p) { | |
6e5f1b57 | 1059 | int architecture = _ARCHITECTURE_INVALID; |
7b3e062c | 1060 | |
7b3e062c | 1061 | if (p == PER_LINUX) |
0c0fea07 LP |
1062 | architecture = native_architecture(); |
1063 | #ifdef SECONDARY_ARCHITECTURE | |
6e5f1b57 | 1064 | else if (p == PER_LINUX32) |
0c0fea07 | 1065 | architecture = SECONDARY_ARCHITECTURE; |
7b3e062c LP |
1066 | #endif |
1067 | ||
6e5f1b57 LP |
1068 | if (architecture < 0) |
1069 | return NULL; | |
1070 | ||
1071 | return architecture_to_string(architecture); | |
7b3e062c LP |
1072 | } |
1073 | ||
21022b9d LP |
1074 | int safe_personality(unsigned long p) { |
1075 | int ret; | |
1076 | ||
1077 | /* So here's the deal, personality() is weirdly defined by glibc. In some cases it returns a failure via errno, | |
1078 | * and in others as negative return value containing an errno-like value. Let's work around this: this is a | |
1079 | * wrapper that uses errno if it is set, and uses the return value otherwise. And then it sets both errno and | |
1080 | * the return value indicating the same issue, so that we are definitely on the safe side. | |
1081 | * | |
1082 | * See https://github.com/systemd/systemd/issues/6737 */ | |
1083 | ||
1084 | errno = 0; | |
1085 | ret = personality(p); | |
1086 | if (ret < 0) { | |
1087 | if (errno != 0) | |
1088 | return -errno; | |
1089 | ||
1090 | errno = -ret; | |
1091 | } | |
1092 | ||
1093 | return ret; | |
1094 | } | |
1095 | ||
e8132d63 LP |
1096 | int opinionated_personality(unsigned long *ret) { |
1097 | int current; | |
1098 | ||
1099 | /* Returns the current personality, or PERSONALITY_INVALID if we can't determine it. This function is a bit | |
1100 | * opinionated though, and ignores all the finer-grained bits and exotic personalities, only distinguishing the | |
1101 | * two most relevant personalities: PER_LINUX and PER_LINUX32. */ | |
1102 | ||
21022b9d | 1103 | current = safe_personality(PERSONALITY_INVALID); |
e8132d63 | 1104 | if (current < 0) |
21022b9d | 1105 | return current; |
e8132d63 LP |
1106 | |
1107 | if (((unsigned long) current & 0xffff) == PER_LINUX32) | |
1108 | *ret = PER_LINUX32; | |
1109 | else | |
1110 | *ret = PER_LINUX; | |
1111 | ||
1112 | return 0; | |
1113 | } | |
1114 | ||
dcadc967 | 1115 | void valgrind_summary_hack(void) { |
349cc4a5 | 1116 | #if HAVE_VALGRIND_VALGRIND_H |
df0ff127 | 1117 | if (getpid_cached() == 1 && RUNNING_ON_VALGRIND) { |
dcadc967 | 1118 | pid_t pid; |
8869a0b4 | 1119 | pid = raw_clone(SIGCHLD); |
dcadc967 EV |
1120 | if (pid < 0) |
1121 | log_emergency_errno(errno, "Failed to fork off valgrind helper: %m"); | |
1122 | else if (pid == 0) | |
1123 | exit(EXIT_SUCCESS); | |
1124 | else { | |
1125 | log_info("Spawned valgrind helper as PID "PID_FMT".", pid); | |
1126 | (void) wait_for_terminate(pid, NULL); | |
1127 | } | |
1128 | } | |
1129 | #endif | |
1130 | } | |
1131 | ||
93bab288 | 1132 | int pid_compare_func(const pid_t *a, const pid_t *b) { |
291d565a | 1133 | /* Suitable for usage in qsort() */ |
93bab288 | 1134 | return CMP(*a, *b); |
291d565a LP |
1135 | } |
1136 | ||
7f452159 LP |
1137 | int ioprio_parse_priority(const char *s, int *ret) { |
1138 | int i, r; | |
1139 | ||
1140 | assert(s); | |
1141 | assert(ret); | |
1142 | ||
1143 | r = safe_atoi(s, &i); | |
1144 | if (r < 0) | |
1145 | return r; | |
1146 | ||
1147 | if (!ioprio_priority_is_valid(i)) | |
1148 | return -EINVAL; | |
1149 | ||
1150 | *ret = i; | |
1151 | return 0; | |
1152 | } | |
1153 | ||
5c30a6d2 LP |
1154 | /* The cached PID, possible values: |
1155 | * | |
1156 | * == UNSET [0] → cache not initialized yet | |
1157 | * == BUSY [-1] → some thread is initializing it at the moment | |
1158 | * any other → the cached PID | |
1159 | */ | |
1160 | ||
1161 | #define CACHED_PID_UNSET ((pid_t) 0) | |
1162 | #define CACHED_PID_BUSY ((pid_t) -1) | |
1163 | ||
1164 | static pid_t cached_pid = CACHED_PID_UNSET; | |
1165 | ||
799a960d | 1166 | void reset_cached_pid(void) { |
5c30a6d2 LP |
1167 | /* Invoked in the child after a fork(), i.e. at the first moment the PID changed */ |
1168 | cached_pid = CACHED_PID_UNSET; | |
1169 | } | |
1170 | ||
1171 | /* We use glibc __register_atfork() + __dso_handle directly here, as they are not included in the glibc | |
1172 | * headers. __register_atfork() is mostly equivalent to pthread_atfork(), but doesn't require us to link against | |
1173 | * libpthread, as it is part of glibc anyway. */ | |
2bb8d8d9 | 1174 | extern int __register_atfork(void (*prepare) (void), void (*parent) (void), void (*child) (void), void *dso_handle); |
5c30a6d2 LP |
1175 | extern void* __dso_handle __attribute__ ((__weak__)); |
1176 | ||
1177 | pid_t getpid_cached(void) { | |
5d71bac3 | 1178 | static bool installed = false; |
5c30a6d2 LP |
1179 | pid_t current_value; |
1180 | ||
1181 | /* getpid_cached() is much like getpid(), but caches the value in local memory, to avoid having to invoke a | |
1182 | * system call each time. This restores glibc behaviour from before 2.24, when getpid() was unconditionally | |
1183 | * cached. Starting with 2.24 getpid() started to become prohibitively expensive when used for detecting when | |
1184 | * objects were used across fork()s. With this caching the old behaviour is somewhat restored. | |
1185 | * | |
1186 | * https://bugzilla.redhat.com/show_bug.cgi?id=1443976 | |
a4041e4f | 1187 | * https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=c579f48edba88380635ab98cb612030e3ed8691e |
5c30a6d2 LP |
1188 | */ |
1189 | ||
1190 | current_value = __sync_val_compare_and_swap(&cached_pid, CACHED_PID_UNSET, CACHED_PID_BUSY); | |
1191 | ||
1192 | switch (current_value) { | |
1193 | ||
1194 | case CACHED_PID_UNSET: { /* Not initialized yet, then do so now */ | |
1195 | pid_t new_pid; | |
1196 | ||
996def17 | 1197 | new_pid = raw_getpid(); |
5c30a6d2 | 1198 | |
5d71bac3 LP |
1199 | if (!installed) { |
1200 | /* __register_atfork() either returns 0 or -ENOMEM, in its glibc implementation. Since it's | |
1201 | * only half-documented (glibc doesn't document it but LSB does — though only superficially) | |
1202 | * we'll check for errors only in the most generic fashion possible. */ | |
1203 | ||
1204 | if (__register_atfork(NULL, NULL, reset_cached_pid, __dso_handle) != 0) { | |
1205 | /* OOM? Let's try again later */ | |
1206 | cached_pid = CACHED_PID_UNSET; | |
1207 | return new_pid; | |
1208 | } | |
1209 | ||
1210 | installed = true; | |
5c30a6d2 LP |
1211 | } |
1212 | ||
1213 | cached_pid = new_pid; | |
1214 | return new_pid; | |
1215 | } | |
1216 | ||
1217 | case CACHED_PID_BUSY: /* Somebody else is currently initializing */ | |
996def17 | 1218 | return raw_getpid(); |
5c30a6d2 LP |
1219 | |
1220 | default: /* Properly initialized */ | |
1221 | return current_value; | |
1222 | } | |
1223 | } | |
1224 | ||
fba868fa LP |
1225 | int must_be_root(void) { |
1226 | ||
1227 | if (geteuid() == 0) | |
1228 | return 0; | |
1229 | ||
baaa35ad | 1230 | return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Need to be root."); |
fba868fa LP |
1231 | } |
1232 | ||
4c253ed1 LP |
1233 | int safe_fork_full( |
1234 | const char *name, | |
1235 | const int except_fds[], | |
1236 | size_t n_except_fds, | |
1237 | ForkFlags flags, | |
1238 | pid_t *ret_pid) { | |
1239 | ||
1240 | pid_t original_pid, pid; | |
1f5d1e02 | 1241 | sigset_t saved_ss, ss; |
a41f6217 | 1242 | bool block_signals = false; |
b6e1fff1 | 1243 | int prio, r; |
4c253ed1 LP |
1244 | |
1245 | /* A wrapper around fork(), that does a couple of important initializations in addition to mere forking. Always | |
1246 | * returns the child's PID in *ret_pid. Returns == 0 in the child, and > 0 in the parent. */ | |
1247 | ||
b6e1fff1 LP |
1248 | prio = flags & FORK_LOG ? LOG_ERR : LOG_DEBUG; |
1249 | ||
4c253ed1 LP |
1250 | original_pid = getpid_cached(); |
1251 | ||
1f5d1e02 | 1252 | if (flags & (FORK_RESET_SIGNALS|FORK_DEATHSIG)) { |
4c253ed1 | 1253 | |
1f5d1e02 LP |
1254 | /* We temporarily block all signals, so that the new child has them blocked initially. This way, we can |
1255 | * be sure that SIGTERMs are not lost we might send to the child. */ | |
4c253ed1 | 1256 | |
4c253ed1 | 1257 | if (sigfillset(&ss) < 0) |
b6e1fff1 | 1258 | return log_full_errno(prio, errno, "Failed to reset signal set: %m"); |
4c253ed1 | 1259 | |
1f5d1e02 LP |
1260 | block_signals = true; |
1261 | ||
1262 | } else if (flags & FORK_WAIT) { | |
1263 | ||
1264 | /* Let's block SIGCHLD at least, so that we can safely watch for the child process */ | |
1265 | ||
1266 | if (sigemptyset(&ss) < 0) | |
1267 | return log_full_errno(prio, errno, "Failed to clear signal set: %m"); | |
1268 | ||
1269 | if (sigaddset(&ss, SIGCHLD) < 0) | |
1270 | return log_full_errno(prio, errno, "Failed to add SIGCHLD to signal set: %m"); | |
1271 | ||
1272 | block_signals = true; | |
4c253ed1 LP |
1273 | } |
1274 | ||
1f5d1e02 LP |
1275 | if (block_signals) |
1276 | if (sigprocmask(SIG_SETMASK, &ss, &saved_ss) < 0) | |
1277 | return log_full_errno(prio, errno, "Failed to set signal mask: %m"); | |
1278 | ||
be39f6ee LP |
1279 | if (flags & FORK_NEW_MOUNTNS) |
1280 | pid = raw_clone(SIGCHLD|CLONE_NEWNS); | |
1281 | else | |
1282 | pid = fork(); | |
4c253ed1 LP |
1283 | if (pid < 0) { |
1284 | r = -errno; | |
1285 | ||
1286 | if (block_signals) /* undo what we did above */ | |
1287 | (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL); | |
1288 | ||
b6e1fff1 | 1289 | return log_full_errno(prio, r, "Failed to fork: %m"); |
4c253ed1 LP |
1290 | } |
1291 | if (pid > 0) { | |
1292 | /* We are in the parent process */ | |
1293 | ||
1f5d1e02 LP |
1294 | log_debug("Successfully forked off '%s' as PID " PID_FMT ".", strna(name), pid); |
1295 | ||
1296 | if (flags & FORK_WAIT) { | |
1297 | r = wait_for_terminate_and_check(name, pid, (flags & FORK_LOG ? WAIT_LOG : 0)); | |
1298 | if (r < 0) | |
1299 | return r; | |
1300 | if (r != EXIT_SUCCESS) /* exit status > 0 should be treated as failure, too */ | |
1301 | return -EPROTO; | |
1302 | } | |
1303 | ||
4c253ed1 LP |
1304 | if (block_signals) /* undo what we did above */ |
1305 | (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL); | |
1306 | ||
4c253ed1 LP |
1307 | if (ret_pid) |
1308 | *ret_pid = pid; | |
1309 | ||
1310 | return 1; | |
1311 | } | |
1312 | ||
1313 | /* We are in the child process */ | |
1314 | ||
1315 | if (flags & FORK_REOPEN_LOG) { | |
1316 | /* Close the logs if requested, before we log anything. And make sure we reopen it if needed. */ | |
1317 | log_close(); | |
1318 | log_set_open_when_needed(true); | |
1319 | } | |
1320 | ||
1321 | if (name) { | |
1322 | r = rename_process(name); | |
1323 | if (r < 0) | |
b6e1fff1 LP |
1324 | log_full_errno(flags & FORK_LOG ? LOG_WARNING : LOG_DEBUG, |
1325 | r, "Failed to rename process, ignoring: %m"); | |
4c253ed1 LP |
1326 | } |
1327 | ||
1328 | if (flags & FORK_DEATHSIG) | |
1329 | if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0) { | |
b6e1fff1 | 1330 | log_full_errno(prio, errno, "Failed to set death signal: %m"); |
4c253ed1 LP |
1331 | _exit(EXIT_FAILURE); |
1332 | } | |
1333 | ||
1334 | if (flags & FORK_RESET_SIGNALS) { | |
1335 | r = reset_all_signal_handlers(); | |
1336 | if (r < 0) { | |
b6e1fff1 | 1337 | log_full_errno(prio, r, "Failed to reset signal handlers: %m"); |
4c253ed1 LP |
1338 | _exit(EXIT_FAILURE); |
1339 | } | |
1340 | ||
1341 | /* This implicitly undoes the signal mask stuff we did before the fork()ing above */ | |
1342 | r = reset_signal_mask(); | |
1343 | if (r < 0) { | |
b6e1fff1 | 1344 | log_full_errno(prio, r, "Failed to reset signal mask: %m"); |
4c253ed1 LP |
1345 | _exit(EXIT_FAILURE); |
1346 | } | |
1347 | } else if (block_signals) { /* undo what we did above */ | |
1348 | if (sigprocmask(SIG_SETMASK, &saved_ss, NULL) < 0) { | |
b6e1fff1 | 1349 | log_full_errno(prio, errno, "Failed to restore signal mask: %m"); |
4c253ed1 LP |
1350 | _exit(EXIT_FAILURE); |
1351 | } | |
1352 | } | |
1353 | ||
1354 | if (flags & FORK_DEATHSIG) { | |
7ddc2dc5 | 1355 | pid_t ppid; |
4c253ed1 LP |
1356 | /* Let's see if the parent PID is still the one we started from? If not, then the parent |
1357 | * already died by the time we set PR_SET_PDEATHSIG, hence let's emulate the effect */ | |
1358 | ||
7ddc2dc5 SL |
1359 | ppid = getppid(); |
1360 | if (ppid == 0) | |
1361 | /* Parent is in a differn't PID namespace. */; | |
1362 | else if (ppid != original_pid) { | |
4c253ed1 LP |
1363 | log_debug("Parent died early, raising SIGTERM."); |
1364 | (void) raise(SIGTERM); | |
1365 | _exit(EXIT_FAILURE); | |
1366 | } | |
1367 | } | |
1368 | ||
d94a24ca | 1369 | if (FLAGS_SET(flags, FORK_NEW_MOUNTNS | FORK_MOUNTNS_SLAVE)) { |
e2047ba9 LP |
1370 | |
1371 | /* Optionally, make sure we never propagate mounts to the host. */ | |
1372 | ||
1373 | if (mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL) < 0) { | |
1374 | log_full_errno(prio, errno, "Failed to remount root directory as MS_SLAVE: %m"); | |
1375 | _exit(EXIT_FAILURE); | |
1376 | } | |
1377 | } | |
1378 | ||
4c253ed1 LP |
1379 | if (flags & FORK_CLOSE_ALL_FDS) { |
1380 | /* Close the logs here in case it got reopened above, as close_all_fds() would close them for us */ | |
1381 | log_close(); | |
1382 | ||
1383 | r = close_all_fds(except_fds, n_except_fds); | |
1384 | if (r < 0) { | |
b6e1fff1 | 1385 | log_full_errno(prio, r, "Failed to close all file descriptors: %m"); |
4c253ed1 LP |
1386 | _exit(EXIT_FAILURE); |
1387 | } | |
1388 | } | |
1389 | ||
1390 | /* When we were asked to reopen the logs, do so again now */ | |
1391 | if (flags & FORK_REOPEN_LOG) { | |
1392 | log_open(); | |
1393 | log_set_open_when_needed(false); | |
1394 | } | |
1395 | ||
1396 | if (flags & FORK_NULL_STDIO) { | |
1397 | r = make_null_stdio(); | |
1398 | if (r < 0) { | |
b6e1fff1 | 1399 | log_full_errno(prio, r, "Failed to connect stdin/stdout to /dev/null: %m"); |
4c253ed1 LP |
1400 | _exit(EXIT_FAILURE); |
1401 | } | |
1402 | } | |
1403 | ||
1404 | if (ret_pid) | |
1405 | *ret_pid = getpid_cached(); | |
1406 | ||
1407 | return 0; | |
1408 | } | |
1409 | ||
27096982 LP |
1410 | int namespace_fork( |
1411 | const char *outer_name, | |
1412 | const char *inner_name, | |
1413 | const int except_fds[], | |
1414 | size_t n_except_fds, | |
1415 | ForkFlags flags, | |
1416 | int pidns_fd, | |
1417 | int mntns_fd, | |
1418 | int netns_fd, | |
1419 | int userns_fd, | |
1420 | int root_fd, | |
1421 | pid_t *ret_pid) { | |
1422 | ||
1423 | int r; | |
1424 | ||
1425 | /* This is much like safe_fork(), but forks twice, and joins the specified namespaces in the middle | |
1426 | * process. This ensures that we are fully a member of the destination namespace, with pidns an all, so that | |
1427 | * /proc/self/fd works correctly. */ | |
1428 | ||
1429 | 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); | |
1430 | if (r < 0) | |
1431 | return r; | |
1432 | if (r == 0) { | |
1433 | pid_t pid; | |
1434 | ||
1435 | /* Child */ | |
1436 | ||
1437 | r = namespace_enter(pidns_fd, mntns_fd, netns_fd, userns_fd, root_fd); | |
1438 | if (r < 0) { | |
1439 | log_full_errno(FLAGS_SET(flags, FORK_LOG) ? LOG_ERR : LOG_DEBUG, r, "Failed to join namespace: %m"); | |
1440 | _exit(EXIT_FAILURE); | |
1441 | } | |
1442 | ||
1443 | /* We mask a few flags here that either make no sense for the grandchild, or that we don't have to do again */ | |
1444 | 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); | |
1445 | if (r < 0) | |
1446 | _exit(EXIT_FAILURE); | |
1447 | if (r == 0) { | |
1448 | /* Child */ | |
1449 | if (ret_pid) | |
1450 | *ret_pid = pid; | |
1451 | return 0; | |
1452 | } | |
1453 | ||
1454 | r = wait_for_terminate_and_check(inner_name, pid, FLAGS_SET(flags, FORK_LOG) ? WAIT_LOG : 0); | |
1455 | if (r < 0) | |
1456 | _exit(EXIT_FAILURE); | |
1457 | ||
1458 | _exit(r); | |
1459 | } | |
1460 | ||
1461 | return 1; | |
1462 | } | |
1463 | ||
da6053d0 | 1464 | int fork_agent(const char *name, const int except[], size_t n_except, pid_t *ret_pid, const char *path, ...) { |
78752f2e | 1465 | bool stdout_is_tty, stderr_is_tty; |
da6053d0 | 1466 | size_t n, i; |
78752f2e LP |
1467 | va_list ap; |
1468 | char **l; | |
1469 | int r; | |
1470 | ||
1471 | assert(path); | |
1472 | ||
1473 | /* Spawns a temporary TTY agent, making sure it goes away when we go away */ | |
1474 | ||
1475 | r = safe_fork_full(name, except, n_except, FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_CLOSE_ALL_FDS, ret_pid); | |
1476 | if (r < 0) | |
1477 | return r; | |
1478 | if (r > 0) | |
1479 | return 0; | |
1480 | ||
1481 | /* In the child: */ | |
1482 | ||
1483 | stdout_is_tty = isatty(STDOUT_FILENO); | |
1484 | stderr_is_tty = isatty(STDERR_FILENO); | |
1485 | ||
1486 | if (!stdout_is_tty || !stderr_is_tty) { | |
1487 | int fd; | |
1488 | ||
1489 | /* Detach from stdout/stderr. and reopen | |
1490 | * /dev/tty for them. This is important to | |
1491 | * ensure that when systemctl is started via | |
1492 | * popen() or a similar call that expects to | |
1493 | * read EOF we actually do generate EOF and | |
1494 | * not delay this indefinitely by because we | |
1495 | * keep an unused copy of stdin around. */ | |
1496 | fd = open("/dev/tty", O_WRONLY); | |
1497 | if (fd < 0) { | |
1498 | log_error_errno(errno, "Failed to open /dev/tty: %m"); | |
1499 | _exit(EXIT_FAILURE); | |
1500 | } | |
1501 | ||
1502 | if (!stdout_is_tty && dup2(fd, STDOUT_FILENO) < 0) { | |
1503 | log_error_errno(errno, "Failed to dup2 /dev/tty: %m"); | |
1504 | _exit(EXIT_FAILURE); | |
1505 | } | |
1506 | ||
1507 | if (!stderr_is_tty && dup2(fd, STDERR_FILENO) < 0) { | |
1508 | log_error_errno(errno, "Failed to dup2 /dev/tty: %m"); | |
1509 | _exit(EXIT_FAILURE); | |
1510 | } | |
1511 | ||
e7685a77 | 1512 | safe_close_above_stdio(fd); |
78752f2e LP |
1513 | } |
1514 | ||
1515 | /* Count arguments */ | |
1516 | va_start(ap, path); | |
1517 | for (n = 0; va_arg(ap, char*); n++) | |
1518 | ; | |
1519 | va_end(ap); | |
1520 | ||
1521 | /* Allocate strv */ | |
cf409d15 | 1522 | l = newa(char*, n + 1); |
78752f2e LP |
1523 | |
1524 | /* Fill in arguments */ | |
1525 | va_start(ap, path); | |
1526 | for (i = 0; i <= n; i++) | |
1527 | l[i] = va_arg(ap, char*); | |
1528 | va_end(ap); | |
1529 | ||
1530 | execv(path, l); | |
1531 | _exit(EXIT_FAILURE); | |
1532 | } | |
1533 | ||
9f8168eb LP |
1534 | int set_oom_score_adjust(int value) { |
1535 | char t[DECIMAL_STR_MAX(int)]; | |
1536 | ||
1537 | sprintf(t, "%i", value); | |
1538 | ||
1539 | return write_string_file("/proc/self/oom_score_adj", t, | |
1540 | WRITE_STRING_FILE_VERIFY_ON_FAILURE|WRITE_STRING_FILE_DISABLE_BUFFER); | |
1541 | } | |
1542 | ||
7b3e062c LP |
1543 | static const char *const ioprio_class_table[] = { |
1544 | [IOPRIO_CLASS_NONE] = "none", | |
1545 | [IOPRIO_CLASS_RT] = "realtime", | |
1546 | [IOPRIO_CLASS_BE] = "best-effort", | |
1547 | [IOPRIO_CLASS_IDLE] = "idle" | |
1548 | }; | |
1549 | ||
10062bbc | 1550 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ioprio_class, int, IOPRIO_N_CLASSES); |
7b3e062c LP |
1551 | |
1552 | static const char *const sigchld_code_table[] = { | |
1553 | [CLD_EXITED] = "exited", | |
1554 | [CLD_KILLED] = "killed", | |
1555 | [CLD_DUMPED] = "dumped", | |
1556 | [CLD_TRAPPED] = "trapped", | |
1557 | [CLD_STOPPED] = "stopped", | |
1558 | [CLD_CONTINUED] = "continued", | |
1559 | }; | |
1560 | ||
1561 | DEFINE_STRING_TABLE_LOOKUP(sigchld_code, int); | |
1562 | ||
1563 | static const char* const sched_policy_table[] = { | |
1564 | [SCHED_OTHER] = "other", | |
1565 | [SCHED_BATCH] = "batch", | |
1566 | [SCHED_IDLE] = "idle", | |
1567 | [SCHED_FIFO] = "fifo", | |
1568 | [SCHED_RR] = "rr" | |
1569 | }; | |
1570 | ||
1571 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(sched_policy, int, INT_MAX); |