]>
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> | |
4f5dd394 LP |
7 | #include <stdbool.h> |
8 | #include <stdio.h> | |
11c3a366 | 9 | #include <stdlib.h> |
9bfaffd5 | 10 | #include <sys/mman.h> |
e2047ba9 | 11 | #include <sys/mount.h> |
7b3e062c | 12 | #include <sys/personality.h> |
405f8907 | 13 | #include <sys/prctl.h> |
4f5dd394 LP |
14 | #include <sys/types.h> |
15 | #include <sys/wait.h> | |
11c3a366 | 16 | #include <syslog.h> |
4f5dd394 | 17 | #include <unistd.h> |
349cc4a5 | 18 | #if HAVE_VALGRIND_VALGRIND_H |
dcadc967 EV |
19 | #include <valgrind/valgrind.h> |
20 | #endif | |
0b452006 | 21 | |
b5efdb8a | 22 | #include "alloc-util.h" |
6e5f1b57 | 23 | #include "architecture.h" |
21c491e1 | 24 | #include "env-util.h" |
39090201 | 25 | #include "errno-util.h" |
aad3a64d | 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" |
e3b4efd2 | 31 | #include "locale-util.h" |
0b452006 | 32 | #include "log.h" |
11c3a366 | 33 | #include "macro.h" |
0a970718 | 34 | #include "memory-util.h" |
f5947a5e YW |
35 | #include "missing_sched.h" |
36 | #include "missing_syscall.h" | |
0cb8e3d1 | 37 | #include "namespace-util.h" |
aad3a64d | 38 | #include "path-util.h" |
93cc7779 | 39 | #include "process-util.h" |
8869a0b4 | 40 | #include "raw-clone.h" |
909106eb | 41 | #include "rlimit-util.h" |
93cc7779 | 42 | #include "signal-util.h" |
1359fffa | 43 | #include "stat-util.h" |
298f466f | 44 | #include "stdio-util.h" |
7b3e062c | 45 | #include "string-table.h" |
07630cea | 46 | #include "string-util.h" |
4c253ed1 | 47 | #include "terminal-util.h" |
b1d4f8e1 | 48 | #include "user-util.h" |
bc28751e | 49 | #include "utf8.h" |
0b452006 | 50 | |
0e85cbcf ZJS |
51 | /* The kernel limits userspace processes to TASK_COMM_LEN (16 bytes), but allows higher values for its own |
52 | * workers, e.g. "kworker/u9:3-kcryptd/253:0". Let's pick a fixed smallish limit that will work for the kernel. | |
53 | */ | |
54 | #define COMM_MAX_LEN 128 | |
55 | ||
0a51b45c | 56 | static int get_process_state(pid_t pid) { |
5c7b9974 | 57 | _cleanup_free_ char *line = NULL; |
0b452006 RC |
58 | const char *p; |
59 | char state; | |
60 | int r; | |
0b452006 RC |
61 | |
62 | assert(pid >= 0); | |
63 | ||
5c7b9974 LP |
64 | /* Shortcut: if we are enquired about our own state, we are obviously running */ |
65 | if (pid == 0 || pid == getpid_cached()) | |
66 | return (unsigned char) 'R'; | |
67 | ||
0b452006 | 68 | p = procfs_file_alloca(pid, "stat"); |
a644184a | 69 | |
0b452006 | 70 | r = read_one_line_file(p, &line); |
a644184a LP |
71 | if (r == -ENOENT) |
72 | return -ESRCH; | |
0b452006 RC |
73 | if (r < 0) |
74 | return r; | |
75 | ||
76 | p = strrchr(line, ')'); | |
77 | if (!p) | |
78 | return -EIO; | |
79 | ||
80 | p++; | |
81 | ||
82 | if (sscanf(p, " %c", &state) != 1) | |
83 | return -EIO; | |
84 | ||
85 | return (unsigned char) state; | |
86 | } | |
87 | ||
ce268825 LP |
88 | int get_process_comm(pid_t pid, char **ret) { |
89 | _cleanup_free_ char *escaped = NULL, *comm = NULL; | |
0b452006 RC |
90 | int r; |
91 | ||
ce268825 | 92 | assert(ret); |
0b452006 RC |
93 | assert(pid >= 0); |
94 | ||
cde93ba2 LP |
95 | if (pid == 0 || pid == getpid_cached()) { |
96 | comm = new0(char, TASK_COMM_LEN + 1); /* Must fit in 16 byte according to prctl(2) */ | |
97 | if (!comm) | |
98 | return -ENOMEM; | |
99 | ||
100 | if (prctl(PR_GET_NAME, comm) < 0) | |
101 | return -errno; | |
102 | } else { | |
103 | const char *p; | |
104 | ||
105 | p = procfs_file_alloca(pid, "comm"); | |
106 | ||
107 | /* Note that process names of kernel threads can be much longer than TASK_COMM_LEN */ | |
108 | r = read_one_line_file(p, &comm); | |
109 | if (r == -ENOENT) | |
110 | return -ESRCH; | |
111 | if (r < 0) | |
112 | return r; | |
113 | } | |
114 | ||
0e85cbcf | 115 | escaped = new(char, COMM_MAX_LEN); |
ce268825 LP |
116 | if (!escaped) |
117 | return -ENOMEM; | |
118 | ||
ce268825 | 119 | /* Escape unprintable characters, just in case, but don't grow the string beyond the underlying size */ |
0e85cbcf | 120 | cellescape(escaped, COMM_MAX_LEN, comm); |
0b452006 | 121 | |
ce268825 LP |
122 | *ret = TAKE_PTR(escaped); |
123 | return 0; | |
0b452006 RC |
124 | } |
125 | ||
09c1dcee | 126 | int get_process_cmdline(pid_t pid, size_t max_columns, ProcessCmdlineFlags flags, char **line) { |
0b452006 | 127 | _cleanup_fclose_ FILE *f = NULL; |
bc28751e | 128 | _cleanup_free_ char *t = NULL, *ans = NULL; |
0b452006 | 129 | const char *p; |
bc28751e ZJS |
130 | int r; |
131 | size_t k; | |
132 | ||
133 | /* This is supposed to be a safety guard against runaway command lines. */ | |
134 | size_t max_length = sc_arg_max(); | |
0b452006 RC |
135 | |
136 | assert(line); | |
137 | assert(pid >= 0); | |
138 | ||
bc28751e ZJS |
139 | /* Retrieves a process' command line. Replaces non-utf8 bytes by replacement character (�). If |
140 | * max_columns is != -1 will return a string of the specified console width at most, abbreviated with | |
09c1dcee ZJS |
141 | * an ellipsis. If PROCESS_CMDLINE_COMM_FALLBACK is specified in flags and the process has no command |
142 | * line set (the case for kernel threads), or has a command line that resolves to the empty string | |
143 | * will return the "comm" name of the process instead. This will use at most _SC_ARG_MAX bytes of | |
144 | * input data. | |
69281c49 LP |
145 | * |
146 | * Returns -ESRCH if the process doesn't exist, and -ENOENT if the process has no command line (and | |
c0534780 | 147 | * comm_fallback is false). Returns 0 and sets *line otherwise. */ |
69281c49 | 148 | |
0b452006 | 149 | p = procfs_file_alloca(pid, "cmdline"); |
fdeea3f4 ZJS |
150 | r = fopen_unlocked(p, "re", &f); |
151 | if (r == -ENOENT) | |
152 | return -ESRCH; | |
153 | if (r < 0) | |
154 | return r; | |
35bbbf85 | 155 | |
bc28751e ZJS |
156 | /* We assume that each four-byte character uses one or two columns. If we ever check for combining |
157 | * characters, this assumption will need to be adjusted. */ | |
158 | if ((size_t) 4 * max_columns + 1 < max_columns) | |
159 | max_length = MIN(max_length, (size_t) 4 * max_columns + 1); | |
69281c49 | 160 | |
bc28751e ZJS |
161 | t = new(char, max_length); |
162 | if (!t) | |
163 | return -ENOMEM; | |
69281c49 | 164 | |
bc28751e ZJS |
165 | k = fread(t, 1, max_length, f); |
166 | if (k > 0) { | |
167 | /* Arguments are separated by NULs. Let's replace those with spaces. */ | |
168 | for (size_t i = 0; i < k - 1; i++) | |
169 | if (t[i] == '\0') | |
170 | t[i] = ' '; | |
69281c49 | 171 | |
bc28751e | 172 | t[k] = '\0'; /* Normally, t[k] is already NUL, so this is just a guard in case of short read */ |
0b452006 | 173 | } else { |
bc28751e ZJS |
174 | /* We only treat getting nothing as an error. We *could* also get an error after reading some |
175 | * data, but we ignore that case, as such an error is rather unlikely and we prefer to get | |
176 | * some data rather than none. */ | |
177 | if (ferror(f)) | |
178 | return -errno; | |
0b452006 | 179 | |
09c1dcee | 180 | if (!(flags & PROCESS_CMDLINE_COMM_FALLBACK)) |
0b452006 RC |
181 | return -ENOENT; |
182 | ||
bc28751e ZJS |
183 | /* Kernel threads have no argv[] */ |
184 | _cleanup_free_ char *t2 = NULL; | |
69281c49 | 185 | |
bc28751e ZJS |
186 | r = get_process_comm(pid, &t2); |
187 | if (r < 0) | |
188 | return r; | |
69281c49 | 189 | |
bc28751e ZJS |
190 | mfree(t); |
191 | t = strjoin("[", t2, "]"); | |
192 | if (!t) | |
193 | return -ENOMEM; | |
0b452006 RC |
194 | } |
195 | ||
bc28751e | 196 | delete_trailing_chars(t, WHITESPACE); |
eb1ec489 | 197 | |
e3b4efd2 ZJS |
198 | bool eight_bit = (flags & PROCESS_CMDLINE_USE_LOCALE) && !is_locale_utf8(); |
199 | ||
200 | ans = escape_non_printable_full(t, max_columns, eight_bit); | |
bc28751e ZJS |
201 | if (!ans) |
202 | return -ENOMEM; | |
eb1ec489 | 203 | |
bc28751e ZJS |
204 | (void) str_realloc(&ans); |
205 | *line = TAKE_PTR(ans); | |
0b452006 RC |
206 | return 0; |
207 | } | |
208 | ||
9bfaffd5 LP |
209 | int rename_process(const char name[]) { |
210 | static size_t mm_size = 0; | |
211 | static char *mm = NULL; | |
212 | bool truncated = false; | |
213 | size_t l; | |
214 | ||
215 | /* This is a like a poor man's setproctitle(). It changes the comm field, argv[0], and also the glibc's | |
216 | * internally used name of the process. For the first one a limit of 16 chars applies; to the second one in | |
217 | * many cases one of 10 (i.e. length of "/sbin/init") — however if we have CAP_SYS_RESOURCES it is unbounded; | |
218 | * to the third one 7 (i.e. the length of "systemd". If you pass a longer string it will likely be | |
219 | * truncated. | |
220 | * | |
221 | * Returns 0 if a name was set but truncated, > 0 if it was set but not truncated. */ | |
222 | ||
223 | if (isempty(name)) | |
224 | return -EINVAL; /* let's not confuse users unnecessarily with an empty name */ | |
405f8907 | 225 | |
1096bb8a LP |
226 | if (!is_main_thread()) |
227 | return -EPERM; /* Let's not allow setting the process name from other threads than the main one, as we | |
228 | * cache things without locking, and we make assumptions that PR_SET_NAME sets the | |
229 | * process name that isn't correct on any other threads */ | |
230 | ||
9bfaffd5 | 231 | l = strlen(name); |
405f8907 | 232 | |
5a8af747 LP |
233 | /* First step, change the comm field. The main thread's comm is identical to the process comm. This means we |
234 | * can use PR_SET_NAME, which sets the thread name for the calling thread. */ | |
235 | if (prctl(PR_SET_NAME, name) < 0) | |
236 | log_debug_errno(errno, "PR_SET_NAME failed: %m"); | |
0e85cbcf | 237 | if (l >= TASK_COMM_LEN) /* Linux userspace process names can be 15 chars at max */ |
9bfaffd5 LP |
238 | truncated = true; |
239 | ||
240 | /* Second step, change glibc's ID of the process name. */ | |
241 | if (program_invocation_name) { | |
242 | size_t k; | |
243 | ||
244 | k = strlen(program_invocation_name); | |
245 | strncpy(program_invocation_name, name, k); | |
246 | if (l > k) | |
247 | truncated = true; | |
248 | } | |
249 | ||
250 | /* Third step, completely replace the argv[] array the kernel maintains for us. This requires privileges, but | |
251 | * has the advantage that the argv[] array is exactly what we want it to be, and not filled up with zeros at | |
13e785f7 | 252 | * the end. This is the best option for changing /proc/self/cmdline. */ |
01f989c6 JW |
253 | |
254 | /* Let's not bother with this if we don't have euid == 0. Strictly speaking we should check for the | |
255 | * CAP_SYS_RESOURCE capability which is independent of the euid. In our own code the capability generally is | |
256 | * present only for euid == 0, hence let's use this as quick bypass check, to avoid calling mmap() if | |
257 | * PR_SET_MM_ARG_{START,END} fails with EPERM later on anyway. After all geteuid() is dead cheap to call, but | |
258 | * mmap() is not. */ | |
259 | if (geteuid() != 0) | |
260 | log_debug("Skipping PR_SET_MM, as we don't have privileges."); | |
261 | else if (mm_size < l+1) { | |
9bfaffd5 LP |
262 | size_t nn_size; |
263 | char *nn; | |
264 | ||
9bfaffd5 LP |
265 | nn_size = PAGE_ALIGN(l+1); |
266 | nn = mmap(NULL, nn_size, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); | |
267 | if (nn == MAP_FAILED) { | |
268 | log_debug_errno(errno, "mmap() failed: %m"); | |
269 | goto use_saved_argv; | |
270 | } | |
405f8907 | 271 | |
9bfaffd5 LP |
272 | strncpy(nn, name, nn_size); |
273 | ||
274 | /* Now, let's tell the kernel about this new memory */ | |
275 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_START, (unsigned long) nn, 0, 0) < 0) { | |
14ee72b7 FS |
276 | /* HACK: prctl() API is kind of dumb on this point. The existing end address may already be |
277 | * below the desired start address, in which case the kernel may have kicked this back due | |
278 | * to a range-check failure (see linux/kernel/sys.c:validate_prctl_map() to see this in | |
279 | * action). The proper solution would be to have a prctl() API that could set both start+end | |
280 | * simultaneously, or at least let us query the existing address to anticipate this condition | |
281 | * and respond accordingly. For now, we can only guess at the cause of this failure and try | |
282 | * a workaround--which will briefly expand the arg space to something potentially huge before | |
283 | * resizing it to what we want. */ | |
284 | log_debug_errno(errno, "PR_SET_MM_ARG_START failed, attempting PR_SET_MM_ARG_END hack: %m"); | |
285 | ||
286 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) nn + l + 1, 0, 0) < 0) { | |
287 | log_debug_errno(errno, "PR_SET_MM_ARG_END hack failed, proceeding without: %m"); | |
288 | (void) munmap(nn, nn_size); | |
289 | goto use_saved_argv; | |
290 | } | |
9bfaffd5 | 291 | |
14ee72b7 FS |
292 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_START, (unsigned long) nn, 0, 0) < 0) { |
293 | log_debug_errno(errno, "PR_SET_MM_ARG_START still failed, proceeding without: %m"); | |
294 | goto use_saved_argv; | |
295 | } | |
296 | } else { | |
297 | /* And update the end pointer to the new end, too. If this fails, we don't really know what | |
298 | * to do, it's pretty unlikely that we can rollback, hence we'll just accept the failure, | |
299 | * and continue. */ | |
300 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) nn + l + 1, 0, 0) < 0) | |
301 | log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m"); | |
302 | } | |
9bfaffd5 LP |
303 | |
304 | if (mm) | |
305 | (void) munmap(mm, mm_size); | |
306 | ||
307 | mm = nn; | |
308 | mm_size = nn_size; | |
01f989c6 | 309 | } else { |
9bfaffd5 LP |
310 | strncpy(mm, name, mm_size); |
311 | ||
01f989c6 JW |
312 | /* Update the end pointer, continuing regardless of any failure. */ |
313 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) mm + l + 1, 0, 0) < 0) | |
314 | log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m"); | |
315 | } | |
316 | ||
9bfaffd5 LP |
317 | use_saved_argv: |
318 | /* Fourth step: in all cases we'll also update the original argv[], so that our own code gets it right too if | |
319 | * it still looks here */ | |
405f8907 LP |
320 | |
321 | if (saved_argc > 0) { | |
322 | int i; | |
323 | ||
9bfaffd5 LP |
324 | if (saved_argv[0]) { |
325 | size_t k; | |
326 | ||
327 | k = strlen(saved_argv[0]); | |
328 | strncpy(saved_argv[0], name, k); | |
329 | if (l > k) | |
330 | truncated = true; | |
331 | } | |
405f8907 LP |
332 | |
333 | for (i = 1; i < saved_argc; i++) { | |
334 | if (!saved_argv[i]) | |
335 | break; | |
336 | ||
337 | memzero(saved_argv[i], strlen(saved_argv[i])); | |
338 | } | |
339 | } | |
9bfaffd5 LP |
340 | |
341 | return !truncated; | |
405f8907 LP |
342 | } |
343 | ||
0b452006 | 344 | int is_kernel_thread(pid_t pid) { |
36b5119a LP |
345 | _cleanup_free_ char *line = NULL; |
346 | unsigned long long flags; | |
347 | size_t l, i; | |
0b452006 | 348 | const char *p; |
36b5119a LP |
349 | char *q; |
350 | int r; | |
0b452006 | 351 | |
4c701096 | 352 | if (IN_SET(pid, 0, 1) || pid == getpid_cached()) /* pid 1, and we ourselves certainly aren't a kernel thread */ |
0b452006 | 353 | return 0; |
36b5119a LP |
354 | if (!pid_is_valid(pid)) |
355 | return -EINVAL; | |
0b452006 | 356 | |
36b5119a LP |
357 | p = procfs_file_alloca(pid, "stat"); |
358 | r = read_one_line_file(p, &line); | |
359 | if (r == -ENOENT) | |
360 | return -ESRCH; | |
361 | if (r < 0) | |
362 | return r; | |
0b452006 | 363 | |
36b5119a LP |
364 | /* Skip past the comm field */ |
365 | q = strrchr(line, ')'); | |
366 | if (!q) | |
367 | return -EINVAL; | |
368 | q++; | |
369 | ||
370 | /* Skip 6 fields to reach the flags field */ | |
371 | for (i = 0; i < 6; i++) { | |
372 | l = strspn(q, WHITESPACE); | |
373 | if (l < 1) | |
374 | return -EINVAL; | |
375 | q += l; | |
376 | ||
377 | l = strcspn(q, WHITESPACE); | |
378 | if (l < 1) | |
379 | return -EINVAL; | |
380 | q += l; | |
a644184a | 381 | } |
0b452006 | 382 | |
f21f31b2 | 383 | /* Skip preceding whitespace */ |
36b5119a LP |
384 | l = strspn(q, WHITESPACE); |
385 | if (l < 1) | |
386 | return -EINVAL; | |
387 | q += l; | |
35bbbf85 | 388 | |
36b5119a LP |
389 | /* Truncate the rest */ |
390 | l = strcspn(q, WHITESPACE); | |
391 | if (l < 1) | |
392 | return -EINVAL; | |
393 | q[l] = 0; | |
0b452006 | 394 | |
36b5119a LP |
395 | r = safe_atollu(q, &flags); |
396 | if (r < 0) | |
397 | return r; | |
0b452006 | 398 | |
36b5119a | 399 | return !!(flags & PF_KTHREAD); |
0b452006 RC |
400 | } |
401 | ||
402 | int get_process_capeff(pid_t pid, char **capeff) { | |
403 | const char *p; | |
a644184a | 404 | int r; |
0b452006 RC |
405 | |
406 | assert(capeff); | |
407 | assert(pid >= 0); | |
408 | ||
409 | p = procfs_file_alloca(pid, "status"); | |
410 | ||
c4cd1d4d | 411 | r = get_proc_field(p, "CapEff", WHITESPACE, capeff); |
a644184a LP |
412 | if (r == -ENOENT) |
413 | return -ESRCH; | |
414 | ||
415 | return r; | |
0b452006 RC |
416 | } |
417 | ||
418 | static int get_process_link_contents(const char *proc_file, char **name) { | |
419 | int r; | |
420 | ||
421 | assert(proc_file); | |
422 | assert(name); | |
423 | ||
424 | r = readlink_malloc(proc_file, name); | |
a644184a LP |
425 | if (r == -ENOENT) |
426 | return -ESRCH; | |
0b452006 | 427 | if (r < 0) |
a644184a | 428 | return r; |
0b452006 RC |
429 | |
430 | return 0; | |
431 | } | |
432 | ||
433 | int get_process_exe(pid_t pid, char **name) { | |
434 | const char *p; | |
435 | char *d; | |
436 | int r; | |
437 | ||
438 | assert(pid >= 0); | |
439 | ||
440 | p = procfs_file_alloca(pid, "exe"); | |
441 | r = get_process_link_contents(p, name); | |
442 | if (r < 0) | |
443 | return r; | |
444 | ||
445 | d = endswith(*name, " (deleted)"); | |
446 | if (d) | |
447 | *d = '\0'; | |
448 | ||
449 | return 0; | |
450 | } | |
451 | ||
452 | static int get_process_id(pid_t pid, const char *field, uid_t *uid) { | |
453 | _cleanup_fclose_ FILE *f = NULL; | |
0b452006 | 454 | const char *p; |
7e7a16a0 | 455 | int r; |
0b452006 RC |
456 | |
457 | assert(field); | |
458 | assert(uid); | |
459 | ||
07b38ba5 | 460 | if (pid < 0) |
6f8cbcdb LP |
461 | return -EINVAL; |
462 | ||
0b452006 | 463 | p = procfs_file_alloca(pid, "status"); |
fdeea3f4 ZJS |
464 | r = fopen_unlocked(p, "re", &f); |
465 | if (r == -ENOENT) | |
466 | return -ESRCH; | |
467 | if (r < 0) | |
468 | return r; | |
35bbbf85 | 469 | |
7e7a16a0 LP |
470 | for (;;) { |
471 | _cleanup_free_ char *line = NULL; | |
0b452006 RC |
472 | char *l; |
473 | ||
7e7a16a0 LP |
474 | r = read_line(f, LONG_LINE_MAX, &line); |
475 | if (r < 0) | |
476 | return r; | |
477 | if (r == 0) | |
478 | break; | |
479 | ||
0b452006 RC |
480 | l = strstrip(line); |
481 | ||
482 | if (startswith(l, field)) { | |
483 | l += strlen(field); | |
484 | l += strspn(l, WHITESPACE); | |
485 | ||
486 | l[strcspn(l, WHITESPACE)] = 0; | |
487 | ||
488 | return parse_uid(l, uid); | |
489 | } | |
490 | } | |
491 | ||
492 | return -EIO; | |
493 | } | |
494 | ||
495 | int get_process_uid(pid_t pid, uid_t *uid) { | |
6f8cbcdb LP |
496 | |
497 | if (pid == 0 || pid == getpid_cached()) { | |
498 | *uid = getuid(); | |
499 | return 0; | |
500 | } | |
501 | ||
0b452006 RC |
502 | return get_process_id(pid, "Uid:", uid); |
503 | } | |
504 | ||
505 | int get_process_gid(pid_t pid, gid_t *gid) { | |
6f8cbcdb LP |
506 | |
507 | if (pid == 0 || pid == getpid_cached()) { | |
508 | *gid = getgid(); | |
509 | return 0; | |
510 | } | |
511 | ||
0b452006 RC |
512 | assert_cc(sizeof(uid_t) == sizeof(gid_t)); |
513 | return get_process_id(pid, "Gid:", gid); | |
514 | } | |
515 | ||
516 | int get_process_cwd(pid_t pid, char **cwd) { | |
517 | const char *p; | |
518 | ||
519 | assert(pid >= 0); | |
520 | ||
aad3a64d LP |
521 | if (pid == 0 || pid == getpid_cached()) |
522 | return safe_getcwd(cwd); | |
523 | ||
0b452006 RC |
524 | p = procfs_file_alloca(pid, "cwd"); |
525 | ||
526 | return get_process_link_contents(p, cwd); | |
527 | } | |
528 | ||
529 | int get_process_root(pid_t pid, char **root) { | |
530 | const char *p; | |
531 | ||
532 | assert(pid >= 0); | |
533 | ||
534 | p = procfs_file_alloca(pid, "root"); | |
535 | ||
536 | return get_process_link_contents(p, root); | |
537 | } | |
538 | ||
2a7797e9 LP |
539 | #define ENVIRONMENT_BLOCK_MAX (5U*1024U*1024U) |
540 | ||
0b452006 RC |
541 | int get_process_environ(pid_t pid, char **env) { |
542 | _cleanup_fclose_ FILE *f = NULL; | |
543 | _cleanup_free_ char *outcome = NULL; | |
0b452006 | 544 | size_t allocated = 0, sz = 0; |
2a7797e9 LP |
545 | const char *p; |
546 | int r; | |
0b452006 RC |
547 | |
548 | assert(pid >= 0); | |
549 | assert(env); | |
550 | ||
551 | p = procfs_file_alloca(pid, "environ"); | |
552 | ||
fdeea3f4 ZJS |
553 | r = fopen_unlocked(p, "re", &f); |
554 | if (r == -ENOENT) | |
555 | return -ESRCH; | |
556 | if (r < 0) | |
557 | return r; | |
35bbbf85 | 558 | |
2a7797e9 LP |
559 | for (;;) { |
560 | char c; | |
561 | ||
562 | if (sz >= ENVIRONMENT_BLOCK_MAX) | |
563 | return -ENOBUFS; | |
564 | ||
0b452006 RC |
565 | if (!GREEDY_REALLOC(outcome, allocated, sz + 5)) |
566 | return -ENOMEM; | |
567 | ||
2a7797e9 LP |
568 | r = safe_fgetc(f, &c); |
569 | if (r < 0) | |
570 | return r; | |
571 | if (r == 0) | |
572 | break; | |
573 | ||
0b452006 RC |
574 | if (c == '\0') |
575 | outcome[sz++] = '\n'; | |
576 | else | |
577 | sz += cescape_char(c, outcome + sz); | |
578 | } | |
579 | ||
2a7797e9 | 580 | outcome[sz] = '\0'; |
ae2a15bc | 581 | *env = TAKE_PTR(outcome); |
0b452006 RC |
582 | |
583 | return 0; | |
584 | } | |
585 | ||
6bc73acb | 586 | int get_process_ppid(pid_t pid, pid_t *_ppid) { |
0b452006 RC |
587 | int r; |
588 | _cleanup_free_ char *line = NULL; | |
589 | long unsigned ppid; | |
590 | const char *p; | |
591 | ||
592 | assert(pid >= 0); | |
593 | assert(_ppid); | |
594 | ||
6f8cbcdb | 595 | if (pid == 0 || pid == getpid_cached()) { |
0b452006 RC |
596 | *_ppid = getppid(); |
597 | return 0; | |
598 | } | |
599 | ||
600 | p = procfs_file_alloca(pid, "stat"); | |
601 | r = read_one_line_file(p, &line); | |
a644184a LP |
602 | if (r == -ENOENT) |
603 | return -ESRCH; | |
0b452006 RC |
604 | if (r < 0) |
605 | return r; | |
606 | ||
607 | /* Let's skip the pid and comm fields. The latter is enclosed | |
608 | * in () but does not escape any () in its value, so let's | |
609 | * skip over it manually */ | |
610 | ||
611 | p = strrchr(line, ')'); | |
612 | if (!p) | |
613 | return -EIO; | |
614 | ||
615 | p++; | |
616 | ||
617 | if (sscanf(p, " " | |
618 | "%*c " /* state */ | |
619 | "%lu ", /* ppid */ | |
620 | &ppid) != 1) | |
621 | return -EIO; | |
622 | ||
623 | if ((long unsigned) (pid_t) ppid != ppid) | |
624 | return -ERANGE; | |
625 | ||
626 | *_ppid = (pid_t) ppid; | |
627 | ||
628 | return 0; | |
629 | } | |
630 | ||
5e37d193 FB |
631 | int get_process_umask(pid_t pid, mode_t *umask) { |
632 | _cleanup_free_ char *m = NULL; | |
633 | const char *p; | |
634 | int r; | |
635 | ||
636 | assert(umask); | |
637 | assert(pid >= 0); | |
638 | ||
639 | p = procfs_file_alloca(pid, "status"); | |
640 | ||
641 | r = get_proc_field(p, "Umask", WHITESPACE, &m); | |
642 | if (r == -ENOENT) | |
643 | return -ESRCH; | |
644 | ||
645 | return parse_mode(m, umask); | |
646 | } | |
647 | ||
0b452006 RC |
648 | int wait_for_terminate(pid_t pid, siginfo_t *status) { |
649 | siginfo_t dummy; | |
650 | ||
651 | assert(pid >= 1); | |
652 | ||
653 | if (!status) | |
654 | status = &dummy; | |
655 | ||
656 | for (;;) { | |
657 | zero(*status); | |
658 | ||
659 | if (waitid(P_PID, pid, status, WEXITED) < 0) { | |
660 | ||
661 | if (errno == EINTR) | |
662 | continue; | |
663 | ||
3f0083a2 | 664 | return negative_errno(); |
0b452006 RC |
665 | } |
666 | ||
667 | return 0; | |
668 | } | |
669 | } | |
670 | ||
671 | /* | |
672 | * Return values: | |
673 | * < 0 : wait_for_terminate() failed to get the state of the | |
674 | * process, the process was terminated by a signal, or | |
675 | * failed for an unknown reason. | |
676 | * >=0 : The process terminated normally, and its exit code is | |
677 | * returned. | |
678 | * | |
679 | * That is, success is indicated by a return value of zero, and an | |
680 | * error is indicated by a non-zero value. | |
681 | * | |
682 | * A warning is emitted if the process terminates abnormally, | |
683 | * and also if it returns non-zero unless check_exit_code is true. | |
684 | */ | |
7d4904fe LP |
685 | int wait_for_terminate_and_check(const char *name, pid_t pid, WaitFlags flags) { |
686 | _cleanup_free_ char *buffer = NULL; | |
0b452006 | 687 | siginfo_t status; |
7d4904fe | 688 | int r, prio; |
0b452006 | 689 | |
0b452006 RC |
690 | assert(pid > 1); |
691 | ||
7d4904fe LP |
692 | if (!name) { |
693 | r = get_process_comm(pid, &buffer); | |
694 | if (r < 0) | |
695 | log_debug_errno(r, "Failed to acquire process name of " PID_FMT ", ignoring: %m", pid); | |
696 | else | |
697 | name = buffer; | |
698 | } | |
699 | ||
700 | prio = flags & WAIT_LOG_ABNORMAL ? LOG_ERR : LOG_DEBUG; | |
701 | ||
0b452006 RC |
702 | r = wait_for_terminate(pid, &status); |
703 | if (r < 0) | |
7d4904fe | 704 | return log_full_errno(prio, r, "Failed to wait for %s: %m", strna(name)); |
0b452006 RC |
705 | |
706 | if (status.si_code == CLD_EXITED) { | |
7d4904fe LP |
707 | if (status.si_status != EXIT_SUCCESS) |
708 | log_full(flags & WAIT_LOG_NON_ZERO_EXIT_STATUS ? LOG_ERR : LOG_DEBUG, | |
709 | "%s failed with exit status %i.", strna(name), status.si_status); | |
0b452006 RC |
710 | else |
711 | log_debug("%s succeeded.", name); | |
712 | ||
713 | return status.si_status; | |
7d4904fe | 714 | |
3742095b | 715 | } else if (IN_SET(status.si_code, CLD_KILLED, CLD_DUMPED)) { |
0b452006 | 716 | |
7d4904fe | 717 | log_full(prio, "%s terminated by signal %s.", strna(name), signal_to_string(status.si_status)); |
0b452006 RC |
718 | return -EPROTO; |
719 | } | |
720 | ||
7d4904fe | 721 | log_full(prio, "%s failed due to unknown reason.", strna(name)); |
0b452006 RC |
722 | return -EPROTO; |
723 | } | |
724 | ||
d5641e0d KW |
725 | /* |
726 | * Return values: | |
e225e5c3 LP |
727 | * |
728 | * < 0 : wait_for_terminate_with_timeout() failed to get the state of the process, the process timed out, the process | |
729 | * was terminated by a signal, or failed for an unknown reason. | |
730 | * | |
d5641e0d KW |
731 | * >=0 : The process terminated normally with no failures. |
732 | * | |
e225e5c3 LP |
733 | * Success is indicated by a return value of zero, a timeout is indicated by ETIMEDOUT, and all other child failure |
734 | * states are indicated by error is indicated by a non-zero value. | |
735 | * | |
736 | * This call assumes SIGCHLD has been blocked already, in particular before the child to wait for has been forked off | |
737 | * to remain entirely race-free. | |
d5641e0d KW |
738 | */ |
739 | int wait_for_terminate_with_timeout(pid_t pid, usec_t timeout) { | |
740 | sigset_t mask; | |
741 | int r; | |
742 | usec_t until; | |
743 | ||
744 | assert_se(sigemptyset(&mask) == 0); | |
745 | assert_se(sigaddset(&mask, SIGCHLD) == 0); | |
746 | ||
747 | /* Drop into a sigtimewait-based timeout. Waiting for the | |
748 | * pid to exit. */ | |
749 | until = now(CLOCK_MONOTONIC) + timeout; | |
750 | for (;;) { | |
751 | usec_t n; | |
752 | siginfo_t status = {}; | |
753 | struct timespec ts; | |
754 | ||
755 | n = now(CLOCK_MONOTONIC); | |
756 | if (n >= until) | |
757 | break; | |
758 | ||
759 | r = sigtimedwait(&mask, NULL, timespec_store(&ts, until - n)) < 0 ? -errno : 0; | |
760 | /* Assuming we woke due to the child exiting. */ | |
761 | if (waitid(P_PID, pid, &status, WEXITED|WNOHANG) == 0) { | |
762 | if (status.si_pid == pid) { | |
763 | /* This is the correct child.*/ | |
764 | if (status.si_code == CLD_EXITED) | |
765 | return (status.si_status == 0) ? 0 : -EPROTO; | |
766 | else | |
767 | return -EPROTO; | |
768 | } | |
769 | } | |
770 | /* Not the child, check for errors and proceed appropriately */ | |
771 | if (r < 0) { | |
772 | switch (r) { | |
773 | case -EAGAIN: | |
774 | /* Timed out, child is likely hung. */ | |
775 | return -ETIMEDOUT; | |
776 | case -EINTR: | |
777 | /* Received a different signal and should retry */ | |
778 | continue; | |
779 | default: | |
780 | /* Return any unexpected errors */ | |
781 | return r; | |
782 | } | |
783 | } | |
784 | } | |
785 | ||
786 | return -EPROTO; | |
787 | } | |
788 | ||
89c9030d LP |
789 | void sigkill_wait(pid_t pid) { |
790 | assert(pid > 1); | |
791 | ||
53640e6f | 792 | if (kill(pid, SIGKILL) >= 0) |
89c9030d LP |
793 | (void) wait_for_terminate(pid, NULL); |
794 | } | |
795 | ||
796 | void sigkill_waitp(pid_t *pid) { | |
dfd14786 LP |
797 | PROTECT_ERRNO; |
798 | ||
4d0d3d41 LP |
799 | if (!pid) |
800 | return; | |
801 | if (*pid <= 1) | |
802 | return; | |
803 | ||
89c9030d | 804 | sigkill_wait(*pid); |
4d0d3d41 LP |
805 | } |
806 | ||
392cf1d0 SL |
807 | void sigterm_wait(pid_t pid) { |
808 | assert(pid > 1); | |
809 | ||
53640e6f | 810 | if (kill_and_sigcont(pid, SIGTERM) >= 0) |
392cf1d0 SL |
811 | (void) wait_for_terminate(pid, NULL); |
812 | } | |
813 | ||
0b452006 RC |
814 | int kill_and_sigcont(pid_t pid, int sig) { |
815 | int r; | |
816 | ||
817 | r = kill(pid, sig) < 0 ? -errno : 0; | |
818 | ||
26f417d3 LP |
819 | /* If this worked, also send SIGCONT, unless we already just sent a SIGCONT, or SIGKILL was sent which isn't |
820 | * affected by a process being suspended anyway. */ | |
a3d8d68c | 821 | if (r >= 0 && !IN_SET(sig, SIGCONT, SIGKILL)) |
26f417d3 | 822 | (void) kill(pid, SIGCONT); |
0b452006 RC |
823 | |
824 | return r; | |
825 | } | |
826 | ||
e70f4453 | 827 | int getenv_for_pid(pid_t pid, const char *field, char **ret) { |
0b452006 RC |
828 | _cleanup_fclose_ FILE *f = NULL; |
829 | char *value = NULL; | |
0b452006 | 830 | const char *path; |
0d90bd92 LP |
831 | size_t l, sum = 0; |
832 | int r; | |
0b452006 RC |
833 | |
834 | assert(pid >= 0); | |
835 | assert(field); | |
e70f4453 LP |
836 | assert(ret); |
837 | ||
838 | if (pid == 0 || pid == getpid_cached()) { | |
839 | const char *e; | |
840 | ||
841 | e = getenv(field); | |
842 | if (!e) { | |
843 | *ret = NULL; | |
844 | return 0; | |
845 | } | |
846 | ||
847 | value = strdup(e); | |
848 | if (!value) | |
849 | return -ENOMEM; | |
850 | ||
851 | *ret = value; | |
852 | return 1; | |
853 | } | |
0b452006 | 854 | |
0d90bd92 LP |
855 | if (!pid_is_valid(pid)) |
856 | return -EINVAL; | |
857 | ||
0b452006 RC |
858 | path = procfs_file_alloca(pid, "environ"); |
859 | ||
fdeea3f4 ZJS |
860 | r = fopen_unlocked(path, "re", &f); |
861 | if (r == -ENOENT) | |
862 | return -ESRCH; | |
863 | if (r < 0) | |
864 | return r; | |
35bbbf85 | 865 | |
0b452006 | 866 | l = strlen(field); |
0d90bd92 LP |
867 | for (;;) { |
868 | _cleanup_free_ char *line = NULL; | |
0b452006 | 869 | |
0d90bd92 LP |
870 | if (sum > ENVIRONMENT_BLOCK_MAX) /* Give up searching eventually */ |
871 | return -ENOBUFS; | |
0b452006 | 872 | |
0d90bd92 LP |
873 | r = read_nul_string(f, LONG_LINE_MAX, &line); |
874 | if (r < 0) | |
875 | return r; | |
876 | if (r == 0) /* EOF */ | |
877 | break; | |
0b452006 | 878 | |
0d90bd92 | 879 | sum += r; |
0b452006 | 880 | |
041b5ae1 | 881 | if (strneq(line, field, l) && line[l] == '=') { |
0b452006 RC |
882 | value = strdup(line + l + 1); |
883 | if (!value) | |
884 | return -ENOMEM; | |
885 | ||
e70f4453 LP |
886 | *ret = value; |
887 | return 1; | |
0b452006 | 888 | } |
0d90bd92 | 889 | } |
0b452006 | 890 | |
e70f4453 LP |
891 | *ret = NULL; |
892 | return 0; | |
0b452006 RC |
893 | } |
894 | ||
4d051546 FB |
895 | int pid_is_my_child(pid_t pid) { |
896 | pid_t ppid; | |
897 | int r; | |
898 | ||
899 | if (pid <= 1) | |
900 | return false; | |
901 | ||
902 | r = get_process_ppid(pid, &ppid); | |
903 | if (r < 0) | |
904 | return r; | |
905 | ||
906 | return ppid == getpid_cached(); | |
907 | } | |
908 | ||
0b452006 RC |
909 | bool pid_is_unwaited(pid_t pid) { |
910 | /* Checks whether a PID is still valid at all, including a zombie */ | |
911 | ||
07b38ba5 | 912 | if (pid < 0) |
0b452006 RC |
913 | return false; |
914 | ||
5fd9b2c5 LP |
915 | if (pid <= 1) /* If we or PID 1 would be dead and have been waited for, this code would not be running */ |
916 | return true; | |
917 | ||
6f8cbcdb LP |
918 | if (pid == getpid_cached()) |
919 | return true; | |
920 | ||
0b452006 RC |
921 | if (kill(pid, 0) >= 0) |
922 | return true; | |
923 | ||
924 | return errno != ESRCH; | |
925 | } | |
926 | ||
927 | bool pid_is_alive(pid_t pid) { | |
928 | int r; | |
929 | ||
930 | /* Checks whether a PID is still valid and not a zombie */ | |
931 | ||
07b38ba5 | 932 | if (pid < 0) |
0b452006 RC |
933 | return false; |
934 | ||
5fd9b2c5 LP |
935 | if (pid <= 1) /* If we or PID 1 would be a zombie, this code would not be running */ |
936 | return true; | |
937 | ||
6f8cbcdb LP |
938 | if (pid == getpid_cached()) |
939 | return true; | |
940 | ||
0b452006 | 941 | r = get_process_state(pid); |
4c701096 | 942 | if (IN_SET(r, -ESRCH, 'Z')) |
0b452006 RC |
943 | return false; |
944 | ||
945 | return true; | |
946 | } | |
d4510856 | 947 | |
1359fffa MS |
948 | int pid_from_same_root_fs(pid_t pid) { |
949 | const char *root; | |
950 | ||
07b38ba5 | 951 | if (pid < 0) |
6f8cbcdb LP |
952 | return false; |
953 | ||
954 | if (pid == 0 || pid == getpid_cached()) | |
955 | return true; | |
1359fffa MS |
956 | |
957 | root = procfs_file_alloca(pid, "root"); | |
958 | ||
e3f791a2 | 959 | return files_same(root, "/proc/1/root", 0); |
1359fffa MS |
960 | } |
961 | ||
d4510856 LP |
962 | bool is_main_thread(void) { |
963 | static thread_local int cached = 0; | |
964 | ||
965 | if (_unlikely_(cached == 0)) | |
df0ff127 | 966 | cached = getpid_cached() == gettid() ? 1 : -1; |
d4510856 LP |
967 | |
968 | return cached > 0; | |
969 | } | |
7b3e062c | 970 | |
848e863a | 971 | _noreturn_ void freeze(void) { |
7b3e062c | 972 | |
3da48d7a EV |
973 | log_close(); |
974 | ||
7b3e062c | 975 | /* Make sure nobody waits for us on a socket anymore */ |
7acf581a | 976 | (void) close_all_fds(NULL, 0); |
7b3e062c LP |
977 | |
978 | sync(); | |
979 | ||
8647283e MS |
980 | /* Let's not freeze right away, but keep reaping zombies. */ |
981 | for (;;) { | |
982 | int r; | |
983 | siginfo_t si = {}; | |
984 | ||
985 | r = waitid(P_ALL, 0, &si, WEXITED); | |
986 | if (r < 0 && errno != EINTR) | |
987 | break; | |
988 | } | |
989 | ||
990 | /* waitid() failed with an unexpected error, things are really borked. Freeze now! */ | |
7b3e062c LP |
991 | for (;;) |
992 | pause(); | |
993 | } | |
994 | ||
995 | bool oom_score_adjust_is_valid(int oa) { | |
996 | return oa >= OOM_SCORE_ADJ_MIN && oa <= OOM_SCORE_ADJ_MAX; | |
997 | } | |
998 | ||
999 | unsigned long personality_from_string(const char *p) { | |
6e5f1b57 | 1000 | int architecture; |
7b3e062c | 1001 | |
0c0fea07 LP |
1002 | if (!p) |
1003 | return PERSONALITY_INVALID; | |
1004 | ||
6e5f1b57 LP |
1005 | /* Parse a personality specifier. We use our own identifiers that indicate specific ABIs, rather than just |
1006 | * hints regarding the register size, since we want to keep things open for multiple locally supported ABIs for | |
1007 | * the same register size. */ | |
1008 | ||
1009 | architecture = architecture_from_string(p); | |
1010 | if (architecture < 0) | |
1011 | return PERSONALITY_INVALID; | |
7b3e062c | 1012 | |
0c0fea07 | 1013 | if (architecture == native_architecture()) |
7b3e062c | 1014 | return PER_LINUX; |
0c0fea07 LP |
1015 | #ifdef SECONDARY_ARCHITECTURE |
1016 | if (architecture == SECONDARY_ARCHITECTURE) | |
f2d1736c | 1017 | return PER_LINUX32; |
7b3e062c LP |
1018 | #endif |
1019 | ||
1020 | return PERSONALITY_INVALID; | |
1021 | } | |
1022 | ||
1023 | const char* personality_to_string(unsigned long p) { | |
6e5f1b57 | 1024 | int architecture = _ARCHITECTURE_INVALID; |
7b3e062c | 1025 | |
7b3e062c | 1026 | if (p == PER_LINUX) |
0c0fea07 LP |
1027 | architecture = native_architecture(); |
1028 | #ifdef SECONDARY_ARCHITECTURE | |
6e5f1b57 | 1029 | else if (p == PER_LINUX32) |
0c0fea07 | 1030 | architecture = SECONDARY_ARCHITECTURE; |
7b3e062c LP |
1031 | #endif |
1032 | ||
6e5f1b57 LP |
1033 | if (architecture < 0) |
1034 | return NULL; | |
1035 | ||
1036 | return architecture_to_string(architecture); | |
7b3e062c LP |
1037 | } |
1038 | ||
21022b9d LP |
1039 | int safe_personality(unsigned long p) { |
1040 | int ret; | |
1041 | ||
1042 | /* So here's the deal, personality() is weirdly defined by glibc. In some cases it returns a failure via errno, | |
1043 | * and in others as negative return value containing an errno-like value. Let's work around this: this is a | |
1044 | * wrapper that uses errno if it is set, and uses the return value otherwise. And then it sets both errno and | |
1045 | * the return value indicating the same issue, so that we are definitely on the safe side. | |
1046 | * | |
1047 | * See https://github.com/systemd/systemd/issues/6737 */ | |
1048 | ||
1049 | errno = 0; | |
1050 | ret = personality(p); | |
1051 | if (ret < 0) { | |
1052 | if (errno != 0) | |
1053 | return -errno; | |
1054 | ||
1055 | errno = -ret; | |
1056 | } | |
1057 | ||
1058 | return ret; | |
1059 | } | |
1060 | ||
e8132d63 LP |
1061 | int opinionated_personality(unsigned long *ret) { |
1062 | int current; | |
1063 | ||
1064 | /* Returns the current personality, or PERSONALITY_INVALID if we can't determine it. This function is a bit | |
1065 | * opinionated though, and ignores all the finer-grained bits and exotic personalities, only distinguishing the | |
1066 | * two most relevant personalities: PER_LINUX and PER_LINUX32. */ | |
1067 | ||
21022b9d | 1068 | current = safe_personality(PERSONALITY_INVALID); |
e8132d63 | 1069 | if (current < 0) |
21022b9d | 1070 | return current; |
e8132d63 LP |
1071 | |
1072 | if (((unsigned long) current & 0xffff) == PER_LINUX32) | |
1073 | *ret = PER_LINUX32; | |
1074 | else | |
1075 | *ret = PER_LINUX; | |
1076 | ||
1077 | return 0; | |
1078 | } | |
1079 | ||
dcadc967 | 1080 | void valgrind_summary_hack(void) { |
349cc4a5 | 1081 | #if HAVE_VALGRIND_VALGRIND_H |
df0ff127 | 1082 | if (getpid_cached() == 1 && RUNNING_ON_VALGRIND) { |
dcadc967 | 1083 | pid_t pid; |
8869a0b4 | 1084 | pid = raw_clone(SIGCHLD); |
dcadc967 EV |
1085 | if (pid < 0) |
1086 | log_emergency_errno(errno, "Failed to fork off valgrind helper: %m"); | |
1087 | else if (pid == 0) | |
1088 | exit(EXIT_SUCCESS); | |
1089 | else { | |
1090 | log_info("Spawned valgrind helper as PID "PID_FMT".", pid); | |
1091 | (void) wait_for_terminate(pid, NULL); | |
1092 | } | |
1093 | } | |
1094 | #endif | |
1095 | } | |
1096 | ||
93bab288 | 1097 | int pid_compare_func(const pid_t *a, const pid_t *b) { |
291d565a | 1098 | /* Suitable for usage in qsort() */ |
93bab288 | 1099 | return CMP(*a, *b); |
291d565a LP |
1100 | } |
1101 | ||
7f452159 LP |
1102 | int ioprio_parse_priority(const char *s, int *ret) { |
1103 | int i, r; | |
1104 | ||
1105 | assert(s); | |
1106 | assert(ret); | |
1107 | ||
1108 | r = safe_atoi(s, &i); | |
1109 | if (r < 0) | |
1110 | return r; | |
1111 | ||
1112 | if (!ioprio_priority_is_valid(i)) | |
1113 | return -EINVAL; | |
1114 | ||
1115 | *ret = i; | |
1116 | return 0; | |
1117 | } | |
1118 | ||
5c30a6d2 LP |
1119 | /* The cached PID, possible values: |
1120 | * | |
1121 | * == UNSET [0] → cache not initialized yet | |
1122 | * == BUSY [-1] → some thread is initializing it at the moment | |
1123 | * any other → the cached PID | |
1124 | */ | |
1125 | ||
1126 | #define CACHED_PID_UNSET ((pid_t) 0) | |
1127 | #define CACHED_PID_BUSY ((pid_t) -1) | |
1128 | ||
1129 | static pid_t cached_pid = CACHED_PID_UNSET; | |
1130 | ||
799a960d | 1131 | void reset_cached_pid(void) { |
5c30a6d2 LP |
1132 | /* Invoked in the child after a fork(), i.e. at the first moment the PID changed */ |
1133 | cached_pid = CACHED_PID_UNSET; | |
1134 | } | |
1135 | ||
1136 | /* We use glibc __register_atfork() + __dso_handle directly here, as they are not included in the glibc | |
1137 | * headers. __register_atfork() is mostly equivalent to pthread_atfork(), but doesn't require us to link against | |
1138 | * libpthread, as it is part of glibc anyway. */ | |
2bb8d8d9 | 1139 | extern int __register_atfork(void (*prepare) (void), void (*parent) (void), void (*child) (void), void *dso_handle); |
d34dae18 | 1140 | extern void* __dso_handle _weak_; |
5c30a6d2 LP |
1141 | |
1142 | pid_t getpid_cached(void) { | |
5d71bac3 | 1143 | static bool installed = false; |
5c30a6d2 LP |
1144 | pid_t current_value; |
1145 | ||
1146 | /* getpid_cached() is much like getpid(), but caches the value in local memory, to avoid having to invoke a | |
1147 | * system call each time. This restores glibc behaviour from before 2.24, when getpid() was unconditionally | |
1148 | * cached. Starting with 2.24 getpid() started to become prohibitively expensive when used for detecting when | |
1149 | * objects were used across fork()s. With this caching the old behaviour is somewhat restored. | |
1150 | * | |
1151 | * https://bugzilla.redhat.com/show_bug.cgi?id=1443976 | |
a4041e4f | 1152 | * https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=c579f48edba88380635ab98cb612030e3ed8691e |
5c30a6d2 LP |
1153 | */ |
1154 | ||
1155 | current_value = __sync_val_compare_and_swap(&cached_pid, CACHED_PID_UNSET, CACHED_PID_BUSY); | |
1156 | ||
1157 | switch (current_value) { | |
1158 | ||
1159 | case CACHED_PID_UNSET: { /* Not initialized yet, then do so now */ | |
1160 | pid_t new_pid; | |
1161 | ||
996def17 | 1162 | new_pid = raw_getpid(); |
5c30a6d2 | 1163 | |
5d71bac3 LP |
1164 | if (!installed) { |
1165 | /* __register_atfork() either returns 0 or -ENOMEM, in its glibc implementation. Since it's | |
1166 | * only half-documented (glibc doesn't document it but LSB does — though only superficially) | |
1167 | * we'll check for errors only in the most generic fashion possible. */ | |
1168 | ||
1169 | if (__register_atfork(NULL, NULL, reset_cached_pid, __dso_handle) != 0) { | |
1170 | /* OOM? Let's try again later */ | |
1171 | cached_pid = CACHED_PID_UNSET; | |
1172 | return new_pid; | |
1173 | } | |
1174 | ||
1175 | installed = true; | |
5c30a6d2 LP |
1176 | } |
1177 | ||
1178 | cached_pid = new_pid; | |
1179 | return new_pid; | |
1180 | } | |
1181 | ||
1182 | case CACHED_PID_BUSY: /* Somebody else is currently initializing */ | |
996def17 | 1183 | return raw_getpid(); |
5c30a6d2 LP |
1184 | |
1185 | default: /* Properly initialized */ | |
1186 | return current_value; | |
1187 | } | |
1188 | } | |
1189 | ||
fba868fa LP |
1190 | int must_be_root(void) { |
1191 | ||
1192 | if (geteuid() == 0) | |
1193 | return 0; | |
1194 | ||
baaa35ad | 1195 | return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Need to be root."); |
fba868fa LP |
1196 | } |
1197 | ||
4c253ed1 LP |
1198 | int safe_fork_full( |
1199 | const char *name, | |
1200 | const int except_fds[], | |
1201 | size_t n_except_fds, | |
1202 | ForkFlags flags, | |
1203 | pid_t *ret_pid) { | |
1204 | ||
1205 | pid_t original_pid, pid; | |
1f5d1e02 | 1206 | sigset_t saved_ss, ss; |
a41f6217 | 1207 | bool block_signals = false; |
b6e1fff1 | 1208 | int prio, r; |
4c253ed1 LP |
1209 | |
1210 | /* A wrapper around fork(), that does a couple of important initializations in addition to mere forking. Always | |
1211 | * returns the child's PID in *ret_pid. Returns == 0 in the child, and > 0 in the parent. */ | |
1212 | ||
b6e1fff1 LP |
1213 | prio = flags & FORK_LOG ? LOG_ERR : LOG_DEBUG; |
1214 | ||
4c253ed1 LP |
1215 | original_pid = getpid_cached(); |
1216 | ||
1f5d1e02 | 1217 | if (flags & (FORK_RESET_SIGNALS|FORK_DEATHSIG)) { |
1f5d1e02 LP |
1218 | /* We temporarily block all signals, so that the new child has them blocked initially. This way, we can |
1219 | * be sure that SIGTERMs are not lost we might send to the child. */ | |
4c253ed1 | 1220 | |
cd2a429e | 1221 | assert_se(sigfillset(&ss) >= 0); |
1f5d1e02 LP |
1222 | block_signals = true; |
1223 | ||
1224 | } else if (flags & FORK_WAIT) { | |
1f5d1e02 LP |
1225 | /* Let's block SIGCHLD at least, so that we can safely watch for the child process */ |
1226 | ||
cd2a429e ZJS |
1227 | assert_se(sigemptyset(&ss) >= 0); |
1228 | assert_se(sigaddset(&ss, SIGCHLD) >= 0); | |
1f5d1e02 | 1229 | block_signals = true; |
4c253ed1 LP |
1230 | } |
1231 | ||
1f5d1e02 LP |
1232 | if (block_signals) |
1233 | if (sigprocmask(SIG_SETMASK, &ss, &saved_ss) < 0) | |
1234 | return log_full_errno(prio, errno, "Failed to set signal mask: %m"); | |
1235 | ||
be39f6ee LP |
1236 | if (flags & FORK_NEW_MOUNTNS) |
1237 | pid = raw_clone(SIGCHLD|CLONE_NEWNS); | |
1238 | else | |
1239 | pid = fork(); | |
4c253ed1 LP |
1240 | if (pid < 0) { |
1241 | r = -errno; | |
1242 | ||
1243 | if (block_signals) /* undo what we did above */ | |
1244 | (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL); | |
1245 | ||
b6e1fff1 | 1246 | return log_full_errno(prio, r, "Failed to fork: %m"); |
4c253ed1 LP |
1247 | } |
1248 | if (pid > 0) { | |
1249 | /* We are in the parent process */ | |
1250 | ||
1f5d1e02 LP |
1251 | log_debug("Successfully forked off '%s' as PID " PID_FMT ".", strna(name), pid); |
1252 | ||
1253 | if (flags & FORK_WAIT) { | |
1254 | r = wait_for_terminate_and_check(name, pid, (flags & FORK_LOG ? WAIT_LOG : 0)); | |
1255 | if (r < 0) | |
1256 | return r; | |
1257 | if (r != EXIT_SUCCESS) /* exit status > 0 should be treated as failure, too */ | |
1258 | return -EPROTO; | |
1259 | } | |
1260 | ||
4c253ed1 LP |
1261 | if (block_signals) /* undo what we did above */ |
1262 | (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL); | |
1263 | ||
4c253ed1 LP |
1264 | if (ret_pid) |
1265 | *ret_pid = pid; | |
1266 | ||
1267 | return 1; | |
1268 | } | |
1269 | ||
1270 | /* We are in the child process */ | |
1271 | ||
1272 | if (flags & FORK_REOPEN_LOG) { | |
1273 | /* Close the logs if requested, before we log anything. And make sure we reopen it if needed. */ | |
1274 | log_close(); | |
1275 | log_set_open_when_needed(true); | |
1276 | } | |
1277 | ||
1278 | if (name) { | |
1279 | r = rename_process(name); | |
1280 | if (r < 0) | |
b6e1fff1 LP |
1281 | log_full_errno(flags & FORK_LOG ? LOG_WARNING : LOG_DEBUG, |
1282 | r, "Failed to rename process, ignoring: %m"); | |
4c253ed1 LP |
1283 | } |
1284 | ||
1285 | if (flags & FORK_DEATHSIG) | |
1286 | if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0) { | |
b6e1fff1 | 1287 | log_full_errno(prio, errno, "Failed to set death signal: %m"); |
4c253ed1 LP |
1288 | _exit(EXIT_FAILURE); |
1289 | } | |
1290 | ||
1291 | if (flags & FORK_RESET_SIGNALS) { | |
1292 | r = reset_all_signal_handlers(); | |
1293 | if (r < 0) { | |
b6e1fff1 | 1294 | log_full_errno(prio, r, "Failed to reset signal handlers: %m"); |
4c253ed1 LP |
1295 | _exit(EXIT_FAILURE); |
1296 | } | |
1297 | ||
1298 | /* This implicitly undoes the signal mask stuff we did before the fork()ing above */ | |
1299 | r = reset_signal_mask(); | |
1300 | if (r < 0) { | |
b6e1fff1 | 1301 | log_full_errno(prio, r, "Failed to reset signal mask: %m"); |
4c253ed1 LP |
1302 | _exit(EXIT_FAILURE); |
1303 | } | |
1304 | } else if (block_signals) { /* undo what we did above */ | |
1305 | if (sigprocmask(SIG_SETMASK, &saved_ss, NULL) < 0) { | |
b6e1fff1 | 1306 | log_full_errno(prio, errno, "Failed to restore signal mask: %m"); |
4c253ed1 LP |
1307 | _exit(EXIT_FAILURE); |
1308 | } | |
1309 | } | |
1310 | ||
1311 | if (flags & FORK_DEATHSIG) { | |
7ddc2dc5 | 1312 | pid_t ppid; |
4c253ed1 LP |
1313 | /* Let's see if the parent PID is still the one we started from? If not, then the parent |
1314 | * already died by the time we set PR_SET_PDEATHSIG, hence let's emulate the effect */ | |
1315 | ||
7ddc2dc5 SL |
1316 | ppid = getppid(); |
1317 | if (ppid == 0) | |
1318 | /* Parent is in a differn't PID namespace. */; | |
1319 | else if (ppid != original_pid) { | |
4c253ed1 LP |
1320 | log_debug("Parent died early, raising SIGTERM."); |
1321 | (void) raise(SIGTERM); | |
1322 | _exit(EXIT_FAILURE); | |
1323 | } | |
1324 | } | |
1325 | ||
d94a24ca | 1326 | if (FLAGS_SET(flags, FORK_NEW_MOUNTNS | FORK_MOUNTNS_SLAVE)) { |
e2047ba9 LP |
1327 | |
1328 | /* Optionally, make sure we never propagate mounts to the host. */ | |
1329 | ||
1330 | if (mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL) < 0) { | |
1331 | log_full_errno(prio, errno, "Failed to remount root directory as MS_SLAVE: %m"); | |
1332 | _exit(EXIT_FAILURE); | |
1333 | } | |
1334 | } | |
1335 | ||
4c253ed1 LP |
1336 | if (flags & FORK_CLOSE_ALL_FDS) { |
1337 | /* Close the logs here in case it got reopened above, as close_all_fds() would close them for us */ | |
1338 | log_close(); | |
1339 | ||
1340 | r = close_all_fds(except_fds, n_except_fds); | |
1341 | if (r < 0) { | |
b6e1fff1 | 1342 | log_full_errno(prio, r, "Failed to close all file descriptors: %m"); |
4c253ed1 LP |
1343 | _exit(EXIT_FAILURE); |
1344 | } | |
1345 | } | |
1346 | ||
1347 | /* When we were asked to reopen the logs, do so again now */ | |
1348 | if (flags & FORK_REOPEN_LOG) { | |
1349 | log_open(); | |
1350 | log_set_open_when_needed(false); | |
1351 | } | |
1352 | ||
1353 | if (flags & FORK_NULL_STDIO) { | |
1354 | r = make_null_stdio(); | |
1355 | if (r < 0) { | |
b6e1fff1 | 1356 | log_full_errno(prio, r, "Failed to connect stdin/stdout to /dev/null: %m"); |
4c253ed1 LP |
1357 | _exit(EXIT_FAILURE); |
1358 | } | |
8987afc4 LP |
1359 | |
1360 | } else if (flags & FORK_STDOUT_TO_STDERR) { | |
8987afc4 | 1361 | if (dup2(STDERR_FILENO, STDOUT_FILENO) < 0) { |
bddeb54c | 1362 | log_full_errno(prio, errno, "Failed to connect stdout to stderr: %m"); |
8987afc4 LP |
1363 | _exit(EXIT_FAILURE); |
1364 | } | |
4c253ed1 LP |
1365 | } |
1366 | ||
909106eb LP |
1367 | if (flags & FORK_RLIMIT_NOFILE_SAFE) { |
1368 | r = rlimit_nofile_safe(); | |
1369 | if (r < 0) { | |
1370 | log_full_errno(prio, r, "Failed to lower RLIMIT_NOFILE's soft limit to 1K: %m"); | |
1371 | _exit(EXIT_FAILURE); | |
1372 | } | |
1373 | } | |
1374 | ||
4c253ed1 LP |
1375 | if (ret_pid) |
1376 | *ret_pid = getpid_cached(); | |
1377 | ||
1378 | return 0; | |
1379 | } | |
1380 | ||
27096982 LP |
1381 | int namespace_fork( |
1382 | const char *outer_name, | |
1383 | const char *inner_name, | |
1384 | const int except_fds[], | |
1385 | size_t n_except_fds, | |
1386 | ForkFlags flags, | |
1387 | int pidns_fd, | |
1388 | int mntns_fd, | |
1389 | int netns_fd, | |
1390 | int userns_fd, | |
1391 | int root_fd, | |
1392 | pid_t *ret_pid) { | |
1393 | ||
1394 | int r; | |
1395 | ||
1396 | /* This is much like safe_fork(), but forks twice, and joins the specified namespaces in the middle | |
1397 | * process. This ensures that we are fully a member of the destination namespace, with pidns an all, so that | |
1398 | * /proc/self/fd works correctly. */ | |
1399 | ||
1400 | 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); | |
1401 | if (r < 0) | |
1402 | return r; | |
1403 | if (r == 0) { | |
1404 | pid_t pid; | |
1405 | ||
1406 | /* Child */ | |
1407 | ||
1408 | r = namespace_enter(pidns_fd, mntns_fd, netns_fd, userns_fd, root_fd); | |
1409 | if (r < 0) { | |
1410 | log_full_errno(FLAGS_SET(flags, FORK_LOG) ? LOG_ERR : LOG_DEBUG, r, "Failed to join namespace: %m"); | |
1411 | _exit(EXIT_FAILURE); | |
1412 | } | |
1413 | ||
1414 | /* We mask a few flags here that either make no sense for the grandchild, or that we don't have to do again */ | |
1415 | 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); | |
1416 | if (r < 0) | |
1417 | _exit(EXIT_FAILURE); | |
1418 | if (r == 0) { | |
1419 | /* Child */ | |
1420 | if (ret_pid) | |
1421 | *ret_pid = pid; | |
1422 | return 0; | |
1423 | } | |
1424 | ||
1425 | r = wait_for_terminate_and_check(inner_name, pid, FLAGS_SET(flags, FORK_LOG) ? WAIT_LOG : 0); | |
1426 | if (r < 0) | |
1427 | _exit(EXIT_FAILURE); | |
1428 | ||
1429 | _exit(r); | |
1430 | } | |
1431 | ||
1432 | return 1; | |
1433 | } | |
1434 | ||
da6053d0 | 1435 | int fork_agent(const char *name, const int except[], size_t n_except, pid_t *ret_pid, const char *path, ...) { |
78752f2e | 1436 | bool stdout_is_tty, stderr_is_tty; |
da6053d0 | 1437 | size_t n, i; |
78752f2e LP |
1438 | va_list ap; |
1439 | char **l; | |
1440 | int r; | |
1441 | ||
1442 | assert(path); | |
1443 | ||
1444 | /* Spawns a temporary TTY agent, making sure it goes away when we go away */ | |
1445 | ||
1446 | r = safe_fork_full(name, except, n_except, FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_CLOSE_ALL_FDS, ret_pid); | |
1447 | if (r < 0) | |
1448 | return r; | |
1449 | if (r > 0) | |
1450 | return 0; | |
1451 | ||
1452 | /* In the child: */ | |
1453 | ||
1454 | stdout_is_tty = isatty(STDOUT_FILENO); | |
1455 | stderr_is_tty = isatty(STDERR_FILENO); | |
1456 | ||
1457 | if (!stdout_is_tty || !stderr_is_tty) { | |
1458 | int fd; | |
1459 | ||
1460 | /* Detach from stdout/stderr. and reopen | |
1461 | * /dev/tty for them. This is important to | |
1462 | * ensure that when systemctl is started via | |
1463 | * popen() or a similar call that expects to | |
1464 | * read EOF we actually do generate EOF and | |
1465 | * not delay this indefinitely by because we | |
1466 | * keep an unused copy of stdin around. */ | |
1467 | fd = open("/dev/tty", O_WRONLY); | |
1468 | if (fd < 0) { | |
1469 | log_error_errno(errno, "Failed to open /dev/tty: %m"); | |
1470 | _exit(EXIT_FAILURE); | |
1471 | } | |
1472 | ||
1473 | if (!stdout_is_tty && dup2(fd, STDOUT_FILENO) < 0) { | |
1474 | log_error_errno(errno, "Failed to dup2 /dev/tty: %m"); | |
1475 | _exit(EXIT_FAILURE); | |
1476 | } | |
1477 | ||
1478 | if (!stderr_is_tty && dup2(fd, STDERR_FILENO) < 0) { | |
1479 | log_error_errno(errno, "Failed to dup2 /dev/tty: %m"); | |
1480 | _exit(EXIT_FAILURE); | |
1481 | } | |
1482 | ||
e7685a77 | 1483 | safe_close_above_stdio(fd); |
78752f2e LP |
1484 | } |
1485 | ||
595225af LP |
1486 | (void) rlimit_nofile_safe(); |
1487 | ||
78752f2e LP |
1488 | /* Count arguments */ |
1489 | va_start(ap, path); | |
1490 | for (n = 0; va_arg(ap, char*); n++) | |
1491 | ; | |
1492 | va_end(ap); | |
1493 | ||
1494 | /* Allocate strv */ | |
cf409d15 | 1495 | l = newa(char*, n + 1); |
78752f2e LP |
1496 | |
1497 | /* Fill in arguments */ | |
1498 | va_start(ap, path); | |
1499 | for (i = 0; i <= n; i++) | |
1500 | l[i] = va_arg(ap, char*); | |
1501 | va_end(ap); | |
1502 | ||
1503 | execv(path, l); | |
1504 | _exit(EXIT_FAILURE); | |
1505 | } | |
1506 | ||
9f8168eb LP |
1507 | int set_oom_score_adjust(int value) { |
1508 | char t[DECIMAL_STR_MAX(int)]; | |
1509 | ||
1510 | sprintf(t, "%i", value); | |
1511 | ||
1512 | return write_string_file("/proc/self/oom_score_adj", t, | |
1513 | WRITE_STRING_FILE_VERIFY_ON_FAILURE|WRITE_STRING_FILE_DISABLE_BUFFER); | |
1514 | } | |
1515 | ||
298f466f LP |
1516 | int pidfd_get_pid(int fd, pid_t *ret) { |
1517 | char path[STRLEN("/proc/self/fdinfo/") + DECIMAL_STR_MAX(int)]; | |
1518 | _cleanup_free_ char *fdinfo = NULL; | |
1519 | char *p; | |
1520 | int r; | |
1521 | ||
1522 | if (fd < 0) | |
1523 | return -EBADF; | |
1524 | ||
1525 | xsprintf(path, "/proc/self/fdinfo/%i", fd); | |
1526 | ||
1527 | r = read_full_file(path, &fdinfo, NULL); | |
1528 | if (r == -ENOENT) /* if fdinfo doesn't exist we assume the process does not exist */ | |
1529 | return -ESRCH; | |
1530 | if (r < 0) | |
1531 | return r; | |
1532 | ||
1533 | p = startswith(fdinfo, "Pid:"); | |
1534 | if (!p) { | |
1535 | p = strstr(fdinfo, "\nPid:"); | |
1536 | if (!p) | |
1537 | return -ENOTTY; /* not a pidfd? */ | |
1538 | ||
1539 | p += 5; | |
1540 | } | |
1541 | ||
1542 | p += strspn(p, WHITESPACE); | |
1543 | p[strcspn(p, WHITESPACE)] = 0; | |
1544 | ||
1545 | return parse_pid(p, ret); | |
1546 | } | |
1547 | ||
39090201 DJL |
1548 | static int rlimit_to_nice(rlim_t limit) { |
1549 | if (limit <= 1) | |
1550 | return PRIO_MAX-1; /* i.e. 19 */ | |
1551 | ||
1552 | if (limit >= -PRIO_MIN + PRIO_MAX) | |
1553 | return PRIO_MIN; /* i.e. -20 */ | |
1554 | ||
1555 | return PRIO_MAX - (int) limit; | |
1556 | } | |
1557 | ||
1558 | int setpriority_closest(int priority) { | |
1559 | int current, limit, saved_errno; | |
1560 | struct rlimit highest; | |
1561 | ||
1562 | /* Try to set requested nice level */ | |
1563 | if (setpriority(PRIO_PROCESS, 0, priority) >= 0) | |
1564 | return 1; | |
1565 | ||
1566 | /* Permission failed */ | |
1567 | saved_errno = -errno; | |
1568 | if (!ERRNO_IS_PRIVILEGE(saved_errno)) | |
1569 | return saved_errno; | |
1570 | ||
1571 | errno = 0; | |
1572 | current = getpriority(PRIO_PROCESS, 0); | |
1573 | if (errno != 0) | |
1574 | return -errno; | |
1575 | ||
1576 | if (priority == current) | |
1577 | return 1; | |
1578 | ||
1579 | /* Hmm, we'd expect that raising the nice level from our status quo would always work. If it doesn't, | |
1580 | * then the whole setpriority() system call is blocked to us, hence let's propagate the error | |
1581 | * right-away */ | |
1582 | if (priority > current) | |
1583 | return saved_errno; | |
1584 | ||
1585 | if (getrlimit(RLIMIT_NICE, &highest) < 0) | |
1586 | return -errno; | |
1587 | ||
1588 | limit = rlimit_to_nice(highest.rlim_cur); | |
1589 | ||
1590 | /* We are already less nice than limit allows us */ | |
1591 | if (current < limit) { | |
1592 | log_debug("Cannot raise nice level, permissions and the resource limit do not allow it."); | |
1593 | return 0; | |
1594 | } | |
1595 | ||
1596 | /* Push to the allowed limit */ | |
1597 | if (setpriority(PRIO_PROCESS, 0, limit) < 0) | |
1598 | return -errno; | |
1599 | ||
1600 | log_debug("Cannot set requested nice level (%i), used next best (%i).", priority, limit); | |
1601 | return 0; | |
1602 | } | |
1603 | ||
7b3e062c LP |
1604 | static const char *const ioprio_class_table[] = { |
1605 | [IOPRIO_CLASS_NONE] = "none", | |
1606 | [IOPRIO_CLASS_RT] = "realtime", | |
1607 | [IOPRIO_CLASS_BE] = "best-effort", | |
f44b3035 | 1608 | [IOPRIO_CLASS_IDLE] = "idle", |
7b3e062c LP |
1609 | }; |
1610 | ||
10062bbc | 1611 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ioprio_class, int, IOPRIO_N_CLASSES); |
7b3e062c LP |
1612 | |
1613 | static const char *const sigchld_code_table[] = { | |
1614 | [CLD_EXITED] = "exited", | |
1615 | [CLD_KILLED] = "killed", | |
1616 | [CLD_DUMPED] = "dumped", | |
1617 | [CLD_TRAPPED] = "trapped", | |
1618 | [CLD_STOPPED] = "stopped", | |
1619 | [CLD_CONTINUED] = "continued", | |
1620 | }; | |
1621 | ||
1622 | DEFINE_STRING_TABLE_LOOKUP(sigchld_code, int); | |
1623 | ||
1624 | static const char* const sched_policy_table[] = { | |
1625 | [SCHED_OTHER] = "other", | |
1626 | [SCHED_BATCH] = "batch", | |
1627 | [SCHED_IDLE] = "idle", | |
1628 | [SCHED_FIFO] = "fifo", | |
f44b3035 | 1629 | [SCHED_RR] = "rr", |
7b3e062c LP |
1630 | }; |
1631 | ||
1632 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(sched_policy, int, INT_MAX); |