]>
Commit | Line | Data |
---|---|---|
db9ecf05 | 1 | /* SPDX-License-Identifier: LGPL-2.1-or-later */ |
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) { |
bc28751e | 127 | _cleanup_free_ char *t = NULL, *ans = NULL; |
0b452006 | 128 | const char *p; |
bc28751e | 129 | size_t k; |
7b7a060e | 130 | int r; |
0b452006 RC |
131 | |
132 | assert(line); | |
133 | assert(pid >= 0); | |
134 | ||
bc28751e ZJS |
135 | /* Retrieves a process' command line. Replaces non-utf8 bytes by replacement character (�). If |
136 | * max_columns is != -1 will return a string of the specified console width at most, abbreviated with | |
09c1dcee ZJS |
137 | * an ellipsis. If PROCESS_CMDLINE_COMM_FALLBACK is specified in flags and the process has no command |
138 | * line set (the case for kernel threads), or has a command line that resolves to the empty string | |
139 | * will return the "comm" name of the process instead. This will use at most _SC_ARG_MAX bytes of | |
140 | * input data. | |
69281c49 LP |
141 | * |
142 | * Returns -ESRCH if the process doesn't exist, and -ENOENT if the process has no command line (and | |
c0534780 | 143 | * comm_fallback is false). Returns 0 and sets *line otherwise. */ |
69281c49 | 144 | |
0b452006 | 145 | p = procfs_file_alloca(pid, "cmdline"); |
7b7a060e | 146 | r = read_full_virtual_file(p, &t, &k); |
fdeea3f4 ZJS |
147 | if (r == -ENOENT) |
148 | return -ESRCH; | |
149 | if (r < 0) | |
150 | return r; | |
35bbbf85 | 151 | |
bc28751e ZJS |
152 | if (k > 0) { |
153 | /* Arguments are separated by NULs. Let's replace those with spaces. */ | |
154 | for (size_t i = 0; i < k - 1; i++) | |
155 | if (t[i] == '\0') | |
156 | t[i] = ' '; | |
0b452006 | 157 | } else { |
09c1dcee | 158 | if (!(flags & PROCESS_CMDLINE_COMM_FALLBACK)) |
0b452006 RC |
159 | return -ENOENT; |
160 | ||
bc28751e ZJS |
161 | /* Kernel threads have no argv[] */ |
162 | _cleanup_free_ char *t2 = NULL; | |
69281c49 | 163 | |
bc28751e ZJS |
164 | r = get_process_comm(pid, &t2); |
165 | if (r < 0) | |
166 | return r; | |
69281c49 | 167 | |
7b7a060e | 168 | free(t); |
bc28751e ZJS |
169 | t = strjoin("[", t2, "]"); |
170 | if (!t) | |
171 | return -ENOMEM; | |
0b452006 RC |
172 | } |
173 | ||
bc28751e | 174 | delete_trailing_chars(t, WHITESPACE); |
eb1ec489 | 175 | |
e3b4efd2 ZJS |
176 | bool eight_bit = (flags & PROCESS_CMDLINE_USE_LOCALE) && !is_locale_utf8(); |
177 | ||
178 | ans = escape_non_printable_full(t, max_columns, eight_bit); | |
bc28751e ZJS |
179 | if (!ans) |
180 | return -ENOMEM; | |
eb1ec489 | 181 | |
bc28751e ZJS |
182 | (void) str_realloc(&ans); |
183 | *line = TAKE_PTR(ans); | |
0b452006 RC |
184 | return 0; |
185 | } | |
186 | ||
ad4f7f67 | 187 | static int update_argv(const char name[], size_t l) { |
c55104ce ZJS |
188 | static int can_do = -1; |
189 | ||
190 | if (can_do == 0) | |
191 | return 0; | |
192 | can_do = false; /* We'll set it to true only if the whole process works */ | |
193 | ||
01f989c6 JW |
194 | /* Let's not bother with this if we don't have euid == 0. Strictly speaking we should check for the |
195 | * CAP_SYS_RESOURCE capability which is independent of the euid. In our own code the capability generally is | |
196 | * present only for euid == 0, hence let's use this as quick bypass check, to avoid calling mmap() if | |
197 | * PR_SET_MM_ARG_{START,END} fails with EPERM later on anyway. After all geteuid() is dead cheap to call, but | |
198 | * mmap() is not. */ | |
199 | if (geteuid() != 0) | |
ad4f7f67 ZJS |
200 | return log_debug_errno(SYNTHETIC_ERRNO(EPERM), |
201 | "Skipping PR_SET_MM, as we don't have privileges."); | |
202 | ||
203 | static size_t mm_size = 0; | |
204 | static char *mm = NULL; | |
205 | int r; | |
206 | ||
207 | if (mm_size < l+1) { | |
9bfaffd5 LP |
208 | size_t nn_size; |
209 | char *nn; | |
210 | ||
9bfaffd5 LP |
211 | nn_size = PAGE_ALIGN(l+1); |
212 | nn = mmap(NULL, nn_size, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); | |
ad4f7f67 ZJS |
213 | if (nn == MAP_FAILED) |
214 | return log_debug_errno(errno, "mmap() failed: %m"); | |
405f8907 | 215 | |
9bfaffd5 LP |
216 | strncpy(nn, name, nn_size); |
217 | ||
218 | /* Now, let's tell the kernel about this new memory */ | |
219 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_START, (unsigned long) nn, 0, 0) < 0) { | |
c55104ce ZJS |
220 | if (ERRNO_IS_PRIVILEGE(errno)) |
221 | return log_debug_errno(errno, "PR_SET_MM_ARG_START failed: %m"); | |
222 | ||
14ee72b7 FS |
223 | /* HACK: prctl() API is kind of dumb on this point. The existing end address may already be |
224 | * below the desired start address, in which case the kernel may have kicked this back due | |
225 | * to a range-check failure (see linux/kernel/sys.c:validate_prctl_map() to see this in | |
226 | * action). The proper solution would be to have a prctl() API that could set both start+end | |
227 | * simultaneously, or at least let us query the existing address to anticipate this condition | |
228 | * and respond accordingly. For now, we can only guess at the cause of this failure and try | |
229 | * a workaround--which will briefly expand the arg space to something potentially huge before | |
230 | * resizing it to what we want. */ | |
231 | log_debug_errno(errno, "PR_SET_MM_ARG_START failed, attempting PR_SET_MM_ARG_END hack: %m"); | |
232 | ||
233 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) nn + l + 1, 0, 0) < 0) { | |
ad4f7f67 | 234 | r = log_debug_errno(errno, "PR_SET_MM_ARG_END hack failed, proceeding without: %m"); |
14ee72b7 | 235 | (void) munmap(nn, nn_size); |
ad4f7f67 | 236 | return r; |
14ee72b7 | 237 | } |
9bfaffd5 | 238 | |
ad4f7f67 ZJS |
239 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_START, (unsigned long) nn, 0, 0) < 0) |
240 | return log_debug_errno(errno, "PR_SET_MM_ARG_START still failed, proceeding without: %m"); | |
14ee72b7 FS |
241 | } else { |
242 | /* And update the end pointer to the new end, too. If this fails, we don't really know what | |
243 | * to do, it's pretty unlikely that we can rollback, hence we'll just accept the failure, | |
244 | * and continue. */ | |
245 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) nn + l + 1, 0, 0) < 0) | |
246 | log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m"); | |
247 | } | |
9bfaffd5 LP |
248 | |
249 | if (mm) | |
250 | (void) munmap(mm, mm_size); | |
251 | ||
252 | mm = nn; | |
253 | mm_size = nn_size; | |
01f989c6 | 254 | } else { |
9bfaffd5 LP |
255 | strncpy(mm, name, mm_size); |
256 | ||
01f989c6 JW |
257 | /* Update the end pointer, continuing regardless of any failure. */ |
258 | if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) mm + l + 1, 0, 0) < 0) | |
259 | log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m"); | |
260 | } | |
261 | ||
c55104ce | 262 | can_do = true; |
ad4f7f67 ZJS |
263 | return 0; |
264 | } | |
265 | ||
266 | int rename_process(const char name[]) { | |
267 | bool truncated = false; | |
268 | ||
269 | /* This is a like a poor man's setproctitle(). It changes the comm field, argv[0], and also the glibc's | |
270 | * internally used name of the process. For the first one a limit of 16 chars applies; to the second one in | |
271 | * many cases one of 10 (i.e. length of "/sbin/init") — however if we have CAP_SYS_RESOURCES it is unbounded; | |
272 | * to the third one 7 (i.e. the length of "systemd". If you pass a longer string it will likely be | |
273 | * truncated. | |
274 | * | |
275 | * Returns 0 if a name was set but truncated, > 0 if it was set but not truncated. */ | |
276 | ||
277 | if (isempty(name)) | |
278 | return -EINVAL; /* let's not confuse users unnecessarily with an empty name */ | |
279 | ||
280 | if (!is_main_thread()) | |
281 | return -EPERM; /* Let's not allow setting the process name from other threads than the main one, as we | |
282 | * cache things without locking, and we make assumptions that PR_SET_NAME sets the | |
283 | * process name that isn't correct on any other threads */ | |
284 | ||
285 | size_t l = strlen(name); | |
286 | ||
287 | /* First step, change the comm field. The main thread's comm is identical to the process comm. This means we | |
288 | * can use PR_SET_NAME, which sets the thread name for the calling thread. */ | |
289 | if (prctl(PR_SET_NAME, name) < 0) | |
290 | log_debug_errno(errno, "PR_SET_NAME failed: %m"); | |
291 | if (l >= TASK_COMM_LEN) /* Linux userspace process names can be 15 chars at max */ | |
292 | truncated = true; | |
293 | ||
294 | /* Second step, change glibc's ID of the process name. */ | |
295 | if (program_invocation_name) { | |
296 | size_t k; | |
297 | ||
298 | k = strlen(program_invocation_name); | |
299 | strncpy(program_invocation_name, name, k); | |
300 | if (l > k) | |
301 | truncated = true; | |
302 | } | |
303 | ||
304 | /* Third step, completely replace the argv[] array the kernel maintains for us. This requires privileges, but | |
305 | * has the advantage that the argv[] array is exactly what we want it to be, and not filled up with zeros at | |
306 | * the end. This is the best option for changing /proc/self/cmdline. */ | |
307 | (void) update_argv(name, l); | |
308 | ||
9bfaffd5 LP |
309 | /* Fourth step: in all cases we'll also update the original argv[], so that our own code gets it right too if |
310 | * it still looks here */ | |
405f8907 | 311 | if (saved_argc > 0) { |
9bfaffd5 LP |
312 | if (saved_argv[0]) { |
313 | size_t k; | |
314 | ||
315 | k = strlen(saved_argv[0]); | |
316 | strncpy(saved_argv[0], name, k); | |
317 | if (l > k) | |
318 | truncated = true; | |
319 | } | |
405f8907 | 320 | |
ad4f7f67 | 321 | for (int i = 1; i < saved_argc; i++) { |
405f8907 LP |
322 | if (!saved_argv[i]) |
323 | break; | |
324 | ||
325 | memzero(saved_argv[i], strlen(saved_argv[i])); | |
326 | } | |
327 | } | |
9bfaffd5 LP |
328 | |
329 | return !truncated; | |
405f8907 LP |
330 | } |
331 | ||
0b452006 | 332 | int is_kernel_thread(pid_t pid) { |
36b5119a LP |
333 | _cleanup_free_ char *line = NULL; |
334 | unsigned long long flags; | |
335 | size_t l, i; | |
0b452006 | 336 | const char *p; |
36b5119a LP |
337 | char *q; |
338 | int r; | |
0b452006 | 339 | |
4c701096 | 340 | if (IN_SET(pid, 0, 1) || pid == getpid_cached()) /* pid 1, and we ourselves certainly aren't a kernel thread */ |
0b452006 | 341 | return 0; |
36b5119a LP |
342 | if (!pid_is_valid(pid)) |
343 | return -EINVAL; | |
0b452006 | 344 | |
36b5119a LP |
345 | p = procfs_file_alloca(pid, "stat"); |
346 | r = read_one_line_file(p, &line); | |
347 | if (r == -ENOENT) | |
348 | return -ESRCH; | |
349 | if (r < 0) | |
350 | return r; | |
0b452006 | 351 | |
36b5119a LP |
352 | /* Skip past the comm field */ |
353 | q = strrchr(line, ')'); | |
354 | if (!q) | |
355 | return -EINVAL; | |
356 | q++; | |
357 | ||
358 | /* Skip 6 fields to reach the flags field */ | |
359 | for (i = 0; i < 6; i++) { | |
360 | l = strspn(q, WHITESPACE); | |
361 | if (l < 1) | |
362 | return -EINVAL; | |
363 | q += l; | |
364 | ||
365 | l = strcspn(q, WHITESPACE); | |
366 | if (l < 1) | |
367 | return -EINVAL; | |
368 | q += l; | |
a644184a | 369 | } |
0b452006 | 370 | |
f21f31b2 | 371 | /* Skip preceding whitespace */ |
36b5119a LP |
372 | l = strspn(q, WHITESPACE); |
373 | if (l < 1) | |
374 | return -EINVAL; | |
375 | q += l; | |
35bbbf85 | 376 | |
36b5119a LP |
377 | /* Truncate the rest */ |
378 | l = strcspn(q, WHITESPACE); | |
379 | if (l < 1) | |
380 | return -EINVAL; | |
381 | q[l] = 0; | |
0b452006 | 382 | |
36b5119a LP |
383 | r = safe_atollu(q, &flags); |
384 | if (r < 0) | |
385 | return r; | |
0b452006 | 386 | |
36b5119a | 387 | return !!(flags & PF_KTHREAD); |
0b452006 RC |
388 | } |
389 | ||
390 | int get_process_capeff(pid_t pid, char **capeff) { | |
391 | const char *p; | |
a644184a | 392 | int r; |
0b452006 RC |
393 | |
394 | assert(capeff); | |
395 | assert(pid >= 0); | |
396 | ||
397 | p = procfs_file_alloca(pid, "status"); | |
398 | ||
c4cd1d4d | 399 | r = get_proc_field(p, "CapEff", WHITESPACE, capeff); |
a644184a LP |
400 | if (r == -ENOENT) |
401 | return -ESRCH; | |
402 | ||
403 | return r; | |
0b452006 RC |
404 | } |
405 | ||
406 | static int get_process_link_contents(const char *proc_file, char **name) { | |
407 | int r; | |
408 | ||
409 | assert(proc_file); | |
410 | assert(name); | |
411 | ||
412 | r = readlink_malloc(proc_file, name); | |
a644184a LP |
413 | if (r == -ENOENT) |
414 | return -ESRCH; | |
0b452006 | 415 | if (r < 0) |
a644184a | 416 | return r; |
0b452006 RC |
417 | |
418 | return 0; | |
419 | } | |
420 | ||
421 | int get_process_exe(pid_t pid, char **name) { | |
422 | const char *p; | |
423 | char *d; | |
424 | int r; | |
425 | ||
426 | assert(pid >= 0); | |
427 | ||
428 | p = procfs_file_alloca(pid, "exe"); | |
429 | r = get_process_link_contents(p, name); | |
430 | if (r < 0) | |
431 | return r; | |
432 | ||
433 | d = endswith(*name, " (deleted)"); | |
434 | if (d) | |
435 | *d = '\0'; | |
436 | ||
437 | return 0; | |
438 | } | |
439 | ||
440 | static int get_process_id(pid_t pid, const char *field, uid_t *uid) { | |
441 | _cleanup_fclose_ FILE *f = NULL; | |
0b452006 | 442 | const char *p; |
7e7a16a0 | 443 | int r; |
0b452006 RC |
444 | |
445 | assert(field); | |
446 | assert(uid); | |
447 | ||
07b38ba5 | 448 | if (pid < 0) |
6f8cbcdb LP |
449 | return -EINVAL; |
450 | ||
0b452006 | 451 | p = procfs_file_alloca(pid, "status"); |
fdeea3f4 ZJS |
452 | r = fopen_unlocked(p, "re", &f); |
453 | if (r == -ENOENT) | |
454 | return -ESRCH; | |
455 | if (r < 0) | |
456 | return r; | |
35bbbf85 | 457 | |
7e7a16a0 LP |
458 | for (;;) { |
459 | _cleanup_free_ char *line = NULL; | |
0b452006 RC |
460 | char *l; |
461 | ||
7e7a16a0 LP |
462 | r = read_line(f, LONG_LINE_MAX, &line); |
463 | if (r < 0) | |
464 | return r; | |
465 | if (r == 0) | |
466 | break; | |
467 | ||
0b452006 RC |
468 | l = strstrip(line); |
469 | ||
470 | if (startswith(l, field)) { | |
471 | l += strlen(field); | |
472 | l += strspn(l, WHITESPACE); | |
473 | ||
474 | l[strcspn(l, WHITESPACE)] = 0; | |
475 | ||
476 | return parse_uid(l, uid); | |
477 | } | |
478 | } | |
479 | ||
480 | return -EIO; | |
481 | } | |
482 | ||
483 | int get_process_uid(pid_t pid, uid_t *uid) { | |
6f8cbcdb LP |
484 | |
485 | if (pid == 0 || pid == getpid_cached()) { | |
486 | *uid = getuid(); | |
487 | return 0; | |
488 | } | |
489 | ||
0b452006 RC |
490 | return get_process_id(pid, "Uid:", uid); |
491 | } | |
492 | ||
493 | int get_process_gid(pid_t pid, gid_t *gid) { | |
6f8cbcdb LP |
494 | |
495 | if (pid == 0 || pid == getpid_cached()) { | |
496 | *gid = getgid(); | |
497 | return 0; | |
498 | } | |
499 | ||
0b452006 RC |
500 | assert_cc(sizeof(uid_t) == sizeof(gid_t)); |
501 | return get_process_id(pid, "Gid:", gid); | |
502 | } | |
503 | ||
504 | int get_process_cwd(pid_t pid, char **cwd) { | |
505 | const char *p; | |
506 | ||
507 | assert(pid >= 0); | |
508 | ||
aad3a64d LP |
509 | if (pid == 0 || pid == getpid_cached()) |
510 | return safe_getcwd(cwd); | |
511 | ||
0b452006 RC |
512 | p = procfs_file_alloca(pid, "cwd"); |
513 | ||
514 | return get_process_link_contents(p, cwd); | |
515 | } | |
516 | ||
517 | int get_process_root(pid_t pid, char **root) { | |
518 | const char *p; | |
519 | ||
520 | assert(pid >= 0); | |
521 | ||
522 | p = procfs_file_alloca(pid, "root"); | |
523 | ||
524 | return get_process_link_contents(p, root); | |
525 | } | |
526 | ||
2a7797e9 LP |
527 | #define ENVIRONMENT_BLOCK_MAX (5U*1024U*1024U) |
528 | ||
0b452006 RC |
529 | int get_process_environ(pid_t pid, char **env) { |
530 | _cleanup_fclose_ FILE *f = NULL; | |
531 | _cleanup_free_ char *outcome = NULL; | |
0b452006 | 532 | size_t allocated = 0, sz = 0; |
2a7797e9 LP |
533 | const char *p; |
534 | int r; | |
0b452006 RC |
535 | |
536 | assert(pid >= 0); | |
537 | assert(env); | |
538 | ||
539 | p = procfs_file_alloca(pid, "environ"); | |
540 | ||
fdeea3f4 ZJS |
541 | r = fopen_unlocked(p, "re", &f); |
542 | if (r == -ENOENT) | |
543 | return -ESRCH; | |
544 | if (r < 0) | |
545 | return r; | |
35bbbf85 | 546 | |
2a7797e9 LP |
547 | for (;;) { |
548 | char c; | |
549 | ||
550 | if (sz >= ENVIRONMENT_BLOCK_MAX) | |
551 | return -ENOBUFS; | |
552 | ||
0b452006 RC |
553 | if (!GREEDY_REALLOC(outcome, allocated, sz + 5)) |
554 | return -ENOMEM; | |
555 | ||
2a7797e9 LP |
556 | r = safe_fgetc(f, &c); |
557 | if (r < 0) | |
558 | return r; | |
559 | if (r == 0) | |
560 | break; | |
561 | ||
0b452006 RC |
562 | if (c == '\0') |
563 | outcome[sz++] = '\n'; | |
564 | else | |
565 | sz += cescape_char(c, outcome + sz); | |
566 | } | |
567 | ||
2a7797e9 | 568 | outcome[sz] = '\0'; |
ae2a15bc | 569 | *env = TAKE_PTR(outcome); |
0b452006 RC |
570 | |
571 | return 0; | |
572 | } | |
573 | ||
6bc73acb | 574 | int get_process_ppid(pid_t pid, pid_t *_ppid) { |
0b452006 RC |
575 | int r; |
576 | _cleanup_free_ char *line = NULL; | |
577 | long unsigned ppid; | |
578 | const char *p; | |
579 | ||
580 | assert(pid >= 0); | |
581 | assert(_ppid); | |
582 | ||
6f8cbcdb | 583 | if (pid == 0 || pid == getpid_cached()) { |
0b452006 RC |
584 | *_ppid = getppid(); |
585 | return 0; | |
586 | } | |
587 | ||
588 | p = procfs_file_alloca(pid, "stat"); | |
589 | r = read_one_line_file(p, &line); | |
a644184a LP |
590 | if (r == -ENOENT) |
591 | return -ESRCH; | |
0b452006 RC |
592 | if (r < 0) |
593 | return r; | |
594 | ||
595 | /* Let's skip the pid and comm fields. The latter is enclosed | |
596 | * in () but does not escape any () in its value, so let's | |
597 | * skip over it manually */ | |
598 | ||
599 | p = strrchr(line, ')'); | |
600 | if (!p) | |
601 | return -EIO; | |
602 | ||
603 | p++; | |
604 | ||
605 | if (sscanf(p, " " | |
606 | "%*c " /* state */ | |
607 | "%lu ", /* ppid */ | |
608 | &ppid) != 1) | |
609 | return -EIO; | |
610 | ||
611 | if ((long unsigned) (pid_t) ppid != ppid) | |
612 | return -ERANGE; | |
613 | ||
614 | *_ppid = (pid_t) ppid; | |
615 | ||
616 | return 0; | |
617 | } | |
618 | ||
5e37d193 FB |
619 | int get_process_umask(pid_t pid, mode_t *umask) { |
620 | _cleanup_free_ char *m = NULL; | |
621 | const char *p; | |
622 | int r; | |
623 | ||
624 | assert(umask); | |
625 | assert(pid >= 0); | |
626 | ||
627 | p = procfs_file_alloca(pid, "status"); | |
628 | ||
629 | r = get_proc_field(p, "Umask", WHITESPACE, &m); | |
630 | if (r == -ENOENT) | |
631 | return -ESRCH; | |
632 | ||
633 | return parse_mode(m, umask); | |
634 | } | |
635 | ||
0b452006 RC |
636 | int wait_for_terminate(pid_t pid, siginfo_t *status) { |
637 | siginfo_t dummy; | |
638 | ||
639 | assert(pid >= 1); | |
640 | ||
641 | if (!status) | |
642 | status = &dummy; | |
643 | ||
644 | for (;;) { | |
645 | zero(*status); | |
646 | ||
647 | if (waitid(P_PID, pid, status, WEXITED) < 0) { | |
648 | ||
649 | if (errno == EINTR) | |
650 | continue; | |
651 | ||
3f0083a2 | 652 | return negative_errno(); |
0b452006 RC |
653 | } |
654 | ||
655 | return 0; | |
656 | } | |
657 | } | |
658 | ||
659 | /* | |
660 | * Return values: | |
661 | * < 0 : wait_for_terminate() failed to get the state of the | |
662 | * process, the process was terminated by a signal, or | |
663 | * failed for an unknown reason. | |
664 | * >=0 : The process terminated normally, and its exit code is | |
665 | * returned. | |
666 | * | |
667 | * That is, success is indicated by a return value of zero, and an | |
668 | * error is indicated by a non-zero value. | |
669 | * | |
670 | * A warning is emitted if the process terminates abnormally, | |
671 | * and also if it returns non-zero unless check_exit_code is true. | |
672 | */ | |
7d4904fe LP |
673 | int wait_for_terminate_and_check(const char *name, pid_t pid, WaitFlags flags) { |
674 | _cleanup_free_ char *buffer = NULL; | |
0b452006 | 675 | siginfo_t status; |
7d4904fe | 676 | int r, prio; |
0b452006 | 677 | |
0b452006 RC |
678 | assert(pid > 1); |
679 | ||
7d4904fe LP |
680 | if (!name) { |
681 | r = get_process_comm(pid, &buffer); | |
682 | if (r < 0) | |
683 | log_debug_errno(r, "Failed to acquire process name of " PID_FMT ", ignoring: %m", pid); | |
684 | else | |
685 | name = buffer; | |
686 | } | |
687 | ||
688 | prio = flags & WAIT_LOG_ABNORMAL ? LOG_ERR : LOG_DEBUG; | |
689 | ||
0b452006 RC |
690 | r = wait_for_terminate(pid, &status); |
691 | if (r < 0) | |
7d4904fe | 692 | return log_full_errno(prio, r, "Failed to wait for %s: %m", strna(name)); |
0b452006 RC |
693 | |
694 | if (status.si_code == CLD_EXITED) { | |
7d4904fe LP |
695 | if (status.si_status != EXIT_SUCCESS) |
696 | log_full(flags & WAIT_LOG_NON_ZERO_EXIT_STATUS ? LOG_ERR : LOG_DEBUG, | |
697 | "%s failed with exit status %i.", strna(name), status.si_status); | |
0b452006 RC |
698 | else |
699 | log_debug("%s succeeded.", name); | |
700 | ||
701 | return status.si_status; | |
7d4904fe | 702 | |
3742095b | 703 | } else if (IN_SET(status.si_code, CLD_KILLED, CLD_DUMPED)) { |
0b452006 | 704 | |
7d4904fe | 705 | log_full(prio, "%s terminated by signal %s.", strna(name), signal_to_string(status.si_status)); |
0b452006 RC |
706 | return -EPROTO; |
707 | } | |
708 | ||
7d4904fe | 709 | log_full(prio, "%s failed due to unknown reason.", strna(name)); |
0b452006 RC |
710 | return -EPROTO; |
711 | } | |
712 | ||
d5641e0d KW |
713 | /* |
714 | * Return values: | |
e225e5c3 LP |
715 | * |
716 | * < 0 : wait_for_terminate_with_timeout() failed to get the state of the process, the process timed out, the process | |
717 | * was terminated by a signal, or failed for an unknown reason. | |
718 | * | |
d5641e0d KW |
719 | * >=0 : The process terminated normally with no failures. |
720 | * | |
e225e5c3 LP |
721 | * Success is indicated by a return value of zero, a timeout is indicated by ETIMEDOUT, and all other child failure |
722 | * states are indicated by error is indicated by a non-zero value. | |
723 | * | |
724 | * This call assumes SIGCHLD has been blocked already, in particular before the child to wait for has been forked off | |
725 | * to remain entirely race-free. | |
d5641e0d KW |
726 | */ |
727 | int wait_for_terminate_with_timeout(pid_t pid, usec_t timeout) { | |
728 | sigset_t mask; | |
729 | int r; | |
730 | usec_t until; | |
731 | ||
732 | assert_se(sigemptyset(&mask) == 0); | |
733 | assert_se(sigaddset(&mask, SIGCHLD) == 0); | |
734 | ||
735 | /* Drop into a sigtimewait-based timeout. Waiting for the | |
736 | * pid to exit. */ | |
496db330 | 737 | until = usec_add(now(CLOCK_MONOTONIC), timeout); |
d5641e0d KW |
738 | for (;;) { |
739 | usec_t n; | |
740 | siginfo_t status = {}; | |
741 | struct timespec ts; | |
742 | ||
743 | n = now(CLOCK_MONOTONIC); | |
744 | if (n >= until) | |
745 | break; | |
746 | ||
747 | r = sigtimedwait(&mask, NULL, timespec_store(&ts, until - n)) < 0 ? -errno : 0; | |
748 | /* Assuming we woke due to the child exiting. */ | |
749 | if (waitid(P_PID, pid, &status, WEXITED|WNOHANG) == 0) { | |
750 | if (status.si_pid == pid) { | |
751 | /* This is the correct child.*/ | |
752 | if (status.si_code == CLD_EXITED) | |
753 | return (status.si_status == 0) ? 0 : -EPROTO; | |
754 | else | |
755 | return -EPROTO; | |
756 | } | |
757 | } | |
758 | /* Not the child, check for errors and proceed appropriately */ | |
759 | if (r < 0) { | |
760 | switch (r) { | |
761 | case -EAGAIN: | |
762 | /* Timed out, child is likely hung. */ | |
763 | return -ETIMEDOUT; | |
764 | case -EINTR: | |
765 | /* Received a different signal and should retry */ | |
766 | continue; | |
767 | default: | |
768 | /* Return any unexpected errors */ | |
769 | return r; | |
770 | } | |
771 | } | |
772 | } | |
773 | ||
774 | return -EPROTO; | |
775 | } | |
776 | ||
89c9030d LP |
777 | void sigkill_wait(pid_t pid) { |
778 | assert(pid > 1); | |
779 | ||
53640e6f | 780 | if (kill(pid, SIGKILL) >= 0) |
89c9030d LP |
781 | (void) wait_for_terminate(pid, NULL); |
782 | } | |
783 | ||
784 | void sigkill_waitp(pid_t *pid) { | |
dfd14786 LP |
785 | PROTECT_ERRNO; |
786 | ||
4d0d3d41 LP |
787 | if (!pid) |
788 | return; | |
789 | if (*pid <= 1) | |
790 | return; | |
791 | ||
89c9030d | 792 | sigkill_wait(*pid); |
4d0d3d41 LP |
793 | } |
794 | ||
392cf1d0 SL |
795 | void sigterm_wait(pid_t pid) { |
796 | assert(pid > 1); | |
797 | ||
53640e6f | 798 | if (kill_and_sigcont(pid, SIGTERM) >= 0) |
392cf1d0 SL |
799 | (void) wait_for_terminate(pid, NULL); |
800 | } | |
801 | ||
0b452006 RC |
802 | int kill_and_sigcont(pid_t pid, int sig) { |
803 | int r; | |
804 | ||
805 | r = kill(pid, sig) < 0 ? -errno : 0; | |
806 | ||
26f417d3 LP |
807 | /* If this worked, also send SIGCONT, unless we already just sent a SIGCONT, or SIGKILL was sent which isn't |
808 | * affected by a process being suspended anyway. */ | |
a3d8d68c | 809 | if (r >= 0 && !IN_SET(sig, SIGCONT, SIGKILL)) |
26f417d3 | 810 | (void) kill(pid, SIGCONT); |
0b452006 RC |
811 | |
812 | return r; | |
813 | } | |
814 | ||
e70f4453 | 815 | int getenv_for_pid(pid_t pid, const char *field, char **ret) { |
0b452006 RC |
816 | _cleanup_fclose_ FILE *f = NULL; |
817 | char *value = NULL; | |
0b452006 | 818 | const char *path; |
0d90bd92 LP |
819 | size_t l, sum = 0; |
820 | int r; | |
0b452006 RC |
821 | |
822 | assert(pid >= 0); | |
823 | assert(field); | |
e70f4453 LP |
824 | assert(ret); |
825 | ||
826 | if (pid == 0 || pid == getpid_cached()) { | |
827 | const char *e; | |
828 | ||
829 | e = getenv(field); | |
830 | if (!e) { | |
831 | *ret = NULL; | |
832 | return 0; | |
833 | } | |
834 | ||
835 | value = strdup(e); | |
836 | if (!value) | |
837 | return -ENOMEM; | |
838 | ||
839 | *ret = value; | |
840 | return 1; | |
841 | } | |
0b452006 | 842 | |
0d90bd92 LP |
843 | if (!pid_is_valid(pid)) |
844 | return -EINVAL; | |
845 | ||
0b452006 RC |
846 | path = procfs_file_alloca(pid, "environ"); |
847 | ||
fdeea3f4 ZJS |
848 | r = fopen_unlocked(path, "re", &f); |
849 | if (r == -ENOENT) | |
850 | return -ESRCH; | |
851 | if (r < 0) | |
852 | return r; | |
35bbbf85 | 853 | |
0b452006 | 854 | l = strlen(field); |
0d90bd92 LP |
855 | for (;;) { |
856 | _cleanup_free_ char *line = NULL; | |
0b452006 | 857 | |
0d90bd92 LP |
858 | if (sum > ENVIRONMENT_BLOCK_MAX) /* Give up searching eventually */ |
859 | return -ENOBUFS; | |
0b452006 | 860 | |
0d90bd92 LP |
861 | r = read_nul_string(f, LONG_LINE_MAX, &line); |
862 | if (r < 0) | |
863 | return r; | |
864 | if (r == 0) /* EOF */ | |
865 | break; | |
0b452006 | 866 | |
0d90bd92 | 867 | sum += r; |
0b452006 | 868 | |
041b5ae1 | 869 | if (strneq(line, field, l) && line[l] == '=') { |
0b452006 RC |
870 | value = strdup(line + l + 1); |
871 | if (!value) | |
872 | return -ENOMEM; | |
873 | ||
e70f4453 LP |
874 | *ret = value; |
875 | return 1; | |
0b452006 | 876 | } |
0d90bd92 | 877 | } |
0b452006 | 878 | |
e70f4453 LP |
879 | *ret = NULL; |
880 | return 0; | |
0b452006 RC |
881 | } |
882 | ||
4d051546 FB |
883 | int pid_is_my_child(pid_t pid) { |
884 | pid_t ppid; | |
885 | int r; | |
886 | ||
887 | if (pid <= 1) | |
888 | return false; | |
889 | ||
890 | r = get_process_ppid(pid, &ppid); | |
891 | if (r < 0) | |
892 | return r; | |
893 | ||
894 | return ppid == getpid_cached(); | |
895 | } | |
896 | ||
0b452006 RC |
897 | bool pid_is_unwaited(pid_t pid) { |
898 | /* Checks whether a PID is still valid at all, including a zombie */ | |
899 | ||
07b38ba5 | 900 | if (pid < 0) |
0b452006 RC |
901 | return false; |
902 | ||
5fd9b2c5 LP |
903 | if (pid <= 1) /* If we or PID 1 would be dead and have been waited for, this code would not be running */ |
904 | return true; | |
905 | ||
6f8cbcdb LP |
906 | if (pid == getpid_cached()) |
907 | return true; | |
908 | ||
0b452006 RC |
909 | if (kill(pid, 0) >= 0) |
910 | return true; | |
911 | ||
912 | return errno != ESRCH; | |
913 | } | |
914 | ||
915 | bool pid_is_alive(pid_t pid) { | |
916 | int r; | |
917 | ||
918 | /* Checks whether a PID is still valid and not a zombie */ | |
919 | ||
07b38ba5 | 920 | if (pid < 0) |
0b452006 RC |
921 | return false; |
922 | ||
5fd9b2c5 LP |
923 | if (pid <= 1) /* If we or PID 1 would be a zombie, this code would not be running */ |
924 | return true; | |
925 | ||
6f8cbcdb LP |
926 | if (pid == getpid_cached()) |
927 | return true; | |
928 | ||
0b452006 | 929 | r = get_process_state(pid); |
4c701096 | 930 | if (IN_SET(r, -ESRCH, 'Z')) |
0b452006 RC |
931 | return false; |
932 | ||
933 | return true; | |
934 | } | |
d4510856 | 935 | |
1359fffa MS |
936 | int pid_from_same_root_fs(pid_t pid) { |
937 | const char *root; | |
938 | ||
07b38ba5 | 939 | if (pid < 0) |
6f8cbcdb LP |
940 | return false; |
941 | ||
942 | if (pid == 0 || pid == getpid_cached()) | |
943 | return true; | |
1359fffa MS |
944 | |
945 | root = procfs_file_alloca(pid, "root"); | |
946 | ||
e3f791a2 | 947 | return files_same(root, "/proc/1/root", 0); |
1359fffa MS |
948 | } |
949 | ||
d4510856 LP |
950 | bool is_main_thread(void) { |
951 | static thread_local int cached = 0; | |
952 | ||
953 | if (_unlikely_(cached == 0)) | |
df0ff127 | 954 | cached = getpid_cached() == gettid() ? 1 : -1; |
d4510856 LP |
955 | |
956 | return cached > 0; | |
957 | } | |
7b3e062c | 958 | |
848e863a | 959 | _noreturn_ void freeze(void) { |
7b3e062c | 960 | |
3da48d7a EV |
961 | log_close(); |
962 | ||
7b3e062c | 963 | /* Make sure nobody waits for us on a socket anymore */ |
7acf581a | 964 | (void) close_all_fds(NULL, 0); |
7b3e062c LP |
965 | |
966 | sync(); | |
967 | ||
8647283e MS |
968 | /* Let's not freeze right away, but keep reaping zombies. */ |
969 | for (;;) { | |
970 | int r; | |
971 | siginfo_t si = {}; | |
972 | ||
973 | r = waitid(P_ALL, 0, &si, WEXITED); | |
974 | if (r < 0 && errno != EINTR) | |
975 | break; | |
976 | } | |
977 | ||
978 | /* waitid() failed with an unexpected error, things are really borked. Freeze now! */ | |
7b3e062c LP |
979 | for (;;) |
980 | pause(); | |
981 | } | |
982 | ||
983 | bool oom_score_adjust_is_valid(int oa) { | |
984 | return oa >= OOM_SCORE_ADJ_MIN && oa <= OOM_SCORE_ADJ_MAX; | |
985 | } | |
986 | ||
987 | unsigned long personality_from_string(const char *p) { | |
6e5f1b57 | 988 | int architecture; |
7b3e062c | 989 | |
0c0fea07 LP |
990 | if (!p) |
991 | return PERSONALITY_INVALID; | |
992 | ||
6e5f1b57 LP |
993 | /* Parse a personality specifier. We use our own identifiers that indicate specific ABIs, rather than just |
994 | * hints regarding the register size, since we want to keep things open for multiple locally supported ABIs for | |
995 | * the same register size. */ | |
996 | ||
997 | architecture = architecture_from_string(p); | |
998 | if (architecture < 0) | |
999 | return PERSONALITY_INVALID; | |
7b3e062c | 1000 | |
0c0fea07 | 1001 | if (architecture == native_architecture()) |
7b3e062c | 1002 | return PER_LINUX; |
0c0fea07 LP |
1003 | #ifdef SECONDARY_ARCHITECTURE |
1004 | if (architecture == SECONDARY_ARCHITECTURE) | |
f2d1736c | 1005 | return PER_LINUX32; |
7b3e062c LP |
1006 | #endif |
1007 | ||
1008 | return PERSONALITY_INVALID; | |
1009 | } | |
1010 | ||
1011 | const char* personality_to_string(unsigned long p) { | |
6e5f1b57 | 1012 | int architecture = _ARCHITECTURE_INVALID; |
7b3e062c | 1013 | |
7b3e062c | 1014 | if (p == PER_LINUX) |
0c0fea07 LP |
1015 | architecture = native_architecture(); |
1016 | #ifdef SECONDARY_ARCHITECTURE | |
6e5f1b57 | 1017 | else if (p == PER_LINUX32) |
0c0fea07 | 1018 | architecture = SECONDARY_ARCHITECTURE; |
7b3e062c LP |
1019 | #endif |
1020 | ||
6e5f1b57 LP |
1021 | if (architecture < 0) |
1022 | return NULL; | |
1023 | ||
1024 | return architecture_to_string(architecture); | |
7b3e062c LP |
1025 | } |
1026 | ||
21022b9d LP |
1027 | int safe_personality(unsigned long p) { |
1028 | int ret; | |
1029 | ||
1030 | /* So here's the deal, personality() is weirdly defined by glibc. In some cases it returns a failure via errno, | |
1031 | * and in others as negative return value containing an errno-like value. Let's work around this: this is a | |
1032 | * wrapper that uses errno if it is set, and uses the return value otherwise. And then it sets both errno and | |
1033 | * the return value indicating the same issue, so that we are definitely on the safe side. | |
1034 | * | |
1035 | * See https://github.com/systemd/systemd/issues/6737 */ | |
1036 | ||
1037 | errno = 0; | |
1038 | ret = personality(p); | |
1039 | if (ret < 0) { | |
1040 | if (errno != 0) | |
1041 | return -errno; | |
1042 | ||
1043 | errno = -ret; | |
1044 | } | |
1045 | ||
1046 | return ret; | |
1047 | } | |
1048 | ||
e8132d63 LP |
1049 | int opinionated_personality(unsigned long *ret) { |
1050 | int current; | |
1051 | ||
1052 | /* Returns the current personality, or PERSONALITY_INVALID if we can't determine it. This function is a bit | |
1053 | * opinionated though, and ignores all the finer-grained bits and exotic personalities, only distinguishing the | |
1054 | * two most relevant personalities: PER_LINUX and PER_LINUX32. */ | |
1055 | ||
21022b9d | 1056 | current = safe_personality(PERSONALITY_INVALID); |
e8132d63 | 1057 | if (current < 0) |
21022b9d | 1058 | return current; |
e8132d63 LP |
1059 | |
1060 | if (((unsigned long) current & 0xffff) == PER_LINUX32) | |
1061 | *ret = PER_LINUX32; | |
1062 | else | |
1063 | *ret = PER_LINUX; | |
1064 | ||
1065 | return 0; | |
1066 | } | |
1067 | ||
dcadc967 | 1068 | void valgrind_summary_hack(void) { |
349cc4a5 | 1069 | #if HAVE_VALGRIND_VALGRIND_H |
df0ff127 | 1070 | if (getpid_cached() == 1 && RUNNING_ON_VALGRIND) { |
dcadc967 | 1071 | pid_t pid; |
8869a0b4 | 1072 | pid = raw_clone(SIGCHLD); |
dcadc967 EV |
1073 | if (pid < 0) |
1074 | log_emergency_errno(errno, "Failed to fork off valgrind helper: %m"); | |
1075 | else if (pid == 0) | |
1076 | exit(EXIT_SUCCESS); | |
1077 | else { | |
1078 | log_info("Spawned valgrind helper as PID "PID_FMT".", pid); | |
1079 | (void) wait_for_terminate(pid, NULL); | |
1080 | } | |
1081 | } | |
1082 | #endif | |
1083 | } | |
1084 | ||
93bab288 | 1085 | int pid_compare_func(const pid_t *a, const pid_t *b) { |
291d565a | 1086 | /* Suitable for usage in qsort() */ |
93bab288 | 1087 | return CMP(*a, *b); |
291d565a LP |
1088 | } |
1089 | ||
7f452159 LP |
1090 | int ioprio_parse_priority(const char *s, int *ret) { |
1091 | int i, r; | |
1092 | ||
1093 | assert(s); | |
1094 | assert(ret); | |
1095 | ||
1096 | r = safe_atoi(s, &i); | |
1097 | if (r < 0) | |
1098 | return r; | |
1099 | ||
1100 | if (!ioprio_priority_is_valid(i)) | |
1101 | return -EINVAL; | |
1102 | ||
1103 | *ret = i; | |
1104 | return 0; | |
1105 | } | |
1106 | ||
5c30a6d2 LP |
1107 | /* The cached PID, possible values: |
1108 | * | |
1109 | * == UNSET [0] → cache not initialized yet | |
1110 | * == BUSY [-1] → some thread is initializing it at the moment | |
1111 | * any other → the cached PID | |
1112 | */ | |
1113 | ||
1114 | #define CACHED_PID_UNSET ((pid_t) 0) | |
1115 | #define CACHED_PID_BUSY ((pid_t) -1) | |
1116 | ||
1117 | static pid_t cached_pid = CACHED_PID_UNSET; | |
1118 | ||
799a960d | 1119 | void reset_cached_pid(void) { |
5c30a6d2 LP |
1120 | /* Invoked in the child after a fork(), i.e. at the first moment the PID changed */ |
1121 | cached_pid = CACHED_PID_UNSET; | |
1122 | } | |
1123 | ||
1124 | /* We use glibc __register_atfork() + __dso_handle directly here, as they are not included in the glibc | |
1125 | * headers. __register_atfork() is mostly equivalent to pthread_atfork(), but doesn't require us to link against | |
1126 | * libpthread, as it is part of glibc anyway. */ | |
2bb8d8d9 | 1127 | extern int __register_atfork(void (*prepare) (void), void (*parent) (void), void (*child) (void), void *dso_handle); |
d34dae18 | 1128 | extern void* __dso_handle _weak_; |
5c30a6d2 LP |
1129 | |
1130 | pid_t getpid_cached(void) { | |
5d71bac3 | 1131 | static bool installed = false; |
5c30a6d2 LP |
1132 | pid_t current_value; |
1133 | ||
1134 | /* getpid_cached() is much like getpid(), but caches the value in local memory, to avoid having to invoke a | |
1135 | * system call each time. This restores glibc behaviour from before 2.24, when getpid() was unconditionally | |
1136 | * cached. Starting with 2.24 getpid() started to become prohibitively expensive when used for detecting when | |
1137 | * objects were used across fork()s. With this caching the old behaviour is somewhat restored. | |
1138 | * | |
1139 | * https://bugzilla.redhat.com/show_bug.cgi?id=1443976 | |
a4041e4f | 1140 | * https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=c579f48edba88380635ab98cb612030e3ed8691e |
5c30a6d2 LP |
1141 | */ |
1142 | ||
1143 | current_value = __sync_val_compare_and_swap(&cached_pid, CACHED_PID_UNSET, CACHED_PID_BUSY); | |
1144 | ||
1145 | switch (current_value) { | |
1146 | ||
1147 | case CACHED_PID_UNSET: { /* Not initialized yet, then do so now */ | |
1148 | pid_t new_pid; | |
1149 | ||
996def17 | 1150 | new_pid = raw_getpid(); |
5c30a6d2 | 1151 | |
5d71bac3 LP |
1152 | if (!installed) { |
1153 | /* __register_atfork() either returns 0 or -ENOMEM, in its glibc implementation. Since it's | |
1154 | * only half-documented (glibc doesn't document it but LSB does — though only superficially) | |
1155 | * we'll check for errors only in the most generic fashion possible. */ | |
1156 | ||
1157 | if (__register_atfork(NULL, NULL, reset_cached_pid, __dso_handle) != 0) { | |
1158 | /* OOM? Let's try again later */ | |
1159 | cached_pid = CACHED_PID_UNSET; | |
1160 | return new_pid; | |
1161 | } | |
1162 | ||
1163 | installed = true; | |
5c30a6d2 LP |
1164 | } |
1165 | ||
1166 | cached_pid = new_pid; | |
1167 | return new_pid; | |
1168 | } | |
1169 | ||
1170 | case CACHED_PID_BUSY: /* Somebody else is currently initializing */ | |
996def17 | 1171 | return raw_getpid(); |
5c30a6d2 LP |
1172 | |
1173 | default: /* Properly initialized */ | |
1174 | return current_value; | |
1175 | } | |
1176 | } | |
1177 | ||
fba868fa LP |
1178 | int must_be_root(void) { |
1179 | ||
1180 | if (geteuid() == 0) | |
1181 | return 0; | |
1182 | ||
baaa35ad | 1183 | return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Need to be root."); |
fba868fa LP |
1184 | } |
1185 | ||
4b3abcd0 MG |
1186 | static void restore_sigsetp(sigset_t **ssp) { |
1187 | if (*ssp) | |
1188 | (void) sigprocmask(SIG_SETMASK, *ssp, NULL); | |
1189 | } | |
1190 | ||
4c253ed1 LP |
1191 | int safe_fork_full( |
1192 | const char *name, | |
1193 | const int except_fds[], | |
1194 | size_t n_except_fds, | |
1195 | ForkFlags flags, | |
1196 | pid_t *ret_pid) { | |
1197 | ||
1198 | pid_t original_pid, pid; | |
1f5d1e02 | 1199 | sigset_t saved_ss, ss; |
4b3abcd0 | 1200 | _cleanup_(restore_sigsetp) sigset_t *saved_ssp = NULL; |
d7e38115 | 1201 | bool block_signals = false, block_all = false; |
b6e1fff1 | 1202 | int prio, r; |
4c253ed1 LP |
1203 | |
1204 | /* A wrapper around fork(), that does a couple of important initializations in addition to mere forking. Always | |
1205 | * returns the child's PID in *ret_pid. Returns == 0 in the child, and > 0 in the parent. */ | |
1206 | ||
b6e1fff1 LP |
1207 | prio = flags & FORK_LOG ? LOG_ERR : LOG_DEBUG; |
1208 | ||
4c253ed1 LP |
1209 | original_pid = getpid_cached(); |
1210 | ||
48f813c4 LP |
1211 | if (flags & FORK_FLUSH_STDIO) { |
1212 | fflush(stdout); | |
1213 | fflush(stderr); /* This one shouldn't be necessary, stderr should be unbuffered anyway, but let's better be safe than sorry */ | |
1214 | } | |
1215 | ||
1f5d1e02 | 1216 | if (flags & (FORK_RESET_SIGNALS|FORK_DEATHSIG)) { |
1f5d1e02 LP |
1217 | /* We temporarily block all signals, so that the new child has them blocked initially. This way, we can |
1218 | * be sure that SIGTERMs are not lost we might send to the child. */ | |
4c253ed1 | 1219 | |
cd2a429e | 1220 | assert_se(sigfillset(&ss) >= 0); |
d7e38115 | 1221 | block_signals = block_all = true; |
1f5d1e02 LP |
1222 | |
1223 | } else if (flags & FORK_WAIT) { | |
1f5d1e02 LP |
1224 | /* Let's block SIGCHLD at least, so that we can safely watch for the child process */ |
1225 | ||
cd2a429e ZJS |
1226 | assert_se(sigemptyset(&ss) >= 0); |
1227 | assert_se(sigaddset(&ss, SIGCHLD) >= 0); | |
1f5d1e02 | 1228 | block_signals = true; |
4c253ed1 LP |
1229 | } |
1230 | ||
4b3abcd0 | 1231 | if (block_signals) { |
1f5d1e02 LP |
1232 | if (sigprocmask(SIG_SETMASK, &ss, &saved_ss) < 0) |
1233 | return log_full_errno(prio, errno, "Failed to set signal mask: %m"); | |
4b3abcd0 MG |
1234 | saved_ssp = &saved_ss; |
1235 | } | |
1f5d1e02 | 1236 | |
be39f6ee LP |
1237 | if (flags & FORK_NEW_MOUNTNS) |
1238 | pid = raw_clone(SIGCHLD|CLONE_NEWNS); | |
1239 | else | |
1240 | pid = fork(); | |
4b3abcd0 MG |
1241 | if (pid < 0) |
1242 | return log_full_errno(prio, errno, "Failed to fork: %m"); | |
4c253ed1 LP |
1243 | if (pid > 0) { |
1244 | /* We are in the parent process */ | |
1245 | ||
1f5d1e02 LP |
1246 | log_debug("Successfully forked off '%s' as PID " PID_FMT ".", strna(name), pid); |
1247 | ||
1248 | if (flags & FORK_WAIT) { | |
d7e38115 MG |
1249 | if (block_all) { |
1250 | /* undo everything except SIGCHLD */ | |
1251 | ss = saved_ss; | |
1252 | assert_se(sigaddset(&ss, SIGCHLD) >= 0); | |
1253 | (void) sigprocmask(SIG_SETMASK, &ss, NULL); | |
1254 | } | |
1255 | ||
1f5d1e02 LP |
1256 | r = wait_for_terminate_and_check(name, pid, (flags & FORK_LOG ? WAIT_LOG : 0)); |
1257 | if (r < 0) | |
1258 | return r; | |
1259 | if (r != EXIT_SUCCESS) /* exit status > 0 should be treated as failure, too */ | |
1260 | return -EPROTO; | |
1261 | } | |
1262 | ||
4c253ed1 LP |
1263 | if (ret_pid) |
1264 | *ret_pid = pid; | |
1265 | ||
1266 | return 1; | |
1267 | } | |
1268 | ||
1269 | /* We are in the child process */ | |
1270 | ||
4b3abcd0 MG |
1271 | /* Restore signal mask manually */ |
1272 | saved_ssp = NULL; | |
1273 | ||
4c253ed1 LP |
1274 | if (flags & FORK_REOPEN_LOG) { |
1275 | /* Close the logs if requested, before we log anything. And make sure we reopen it if needed. */ | |
1276 | log_close(); | |
1277 | log_set_open_when_needed(true); | |
1278 | } | |
1279 | ||
1280 | if (name) { | |
1281 | r = rename_process(name); | |
1282 | if (r < 0) | |
b6e1fff1 LP |
1283 | log_full_errno(flags & FORK_LOG ? LOG_WARNING : LOG_DEBUG, |
1284 | r, "Failed to rename process, ignoring: %m"); | |
4c253ed1 LP |
1285 | } |
1286 | ||
97033ba4 LP |
1287 | if (flags & (FORK_DEATHSIG|FORK_DEATHSIG_SIGINT)) |
1288 | if (prctl(PR_SET_PDEATHSIG, (flags & FORK_DEATHSIG_SIGINT) ? SIGINT : SIGTERM) < 0) { | |
b6e1fff1 | 1289 | log_full_errno(prio, errno, "Failed to set death signal: %m"); |
4c253ed1 LP |
1290 | _exit(EXIT_FAILURE); |
1291 | } | |
1292 | ||
1293 | if (flags & FORK_RESET_SIGNALS) { | |
1294 | r = reset_all_signal_handlers(); | |
1295 | if (r < 0) { | |
b6e1fff1 | 1296 | log_full_errno(prio, r, "Failed to reset signal handlers: %m"); |
4c253ed1 LP |
1297 | _exit(EXIT_FAILURE); |
1298 | } | |
1299 | ||
1300 | /* This implicitly undoes the signal mask stuff we did before the fork()ing above */ | |
1301 | r = reset_signal_mask(); | |
1302 | if (r < 0) { | |
b6e1fff1 | 1303 | log_full_errno(prio, r, "Failed to reset signal mask: %m"); |
4c253ed1 LP |
1304 | _exit(EXIT_FAILURE); |
1305 | } | |
1306 | } else if (block_signals) { /* undo what we did above */ | |
1307 | if (sigprocmask(SIG_SETMASK, &saved_ss, NULL) < 0) { | |
b6e1fff1 | 1308 | log_full_errno(prio, errno, "Failed to restore signal mask: %m"); |
4c253ed1 LP |
1309 | _exit(EXIT_FAILURE); |
1310 | } | |
1311 | } | |
1312 | ||
1313 | if (flags & FORK_DEATHSIG) { | |
7ddc2dc5 | 1314 | pid_t ppid; |
4c253ed1 LP |
1315 | /* Let's see if the parent PID is still the one we started from? If not, then the parent |
1316 | * already died by the time we set PR_SET_PDEATHSIG, hence let's emulate the effect */ | |
1317 | ||
7ddc2dc5 SL |
1318 | ppid = getppid(); |
1319 | if (ppid == 0) | |
d7b34e38 | 1320 | /* Parent is in a different PID namespace. */; |
7ddc2dc5 | 1321 | else if (ppid != original_pid) { |
4c253ed1 LP |
1322 | log_debug("Parent died early, raising SIGTERM."); |
1323 | (void) raise(SIGTERM); | |
1324 | _exit(EXIT_FAILURE); | |
1325 | } | |
1326 | } | |
1327 | ||
d94a24ca | 1328 | if (FLAGS_SET(flags, FORK_NEW_MOUNTNS | FORK_MOUNTNS_SLAVE)) { |
e2047ba9 LP |
1329 | |
1330 | /* Optionally, make sure we never propagate mounts to the host. */ | |
1331 | ||
1332 | if (mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL) < 0) { | |
1333 | log_full_errno(prio, errno, "Failed to remount root directory as MS_SLAVE: %m"); | |
1334 | _exit(EXIT_FAILURE); | |
1335 | } | |
1336 | } | |
1337 | ||
4c253ed1 LP |
1338 | if (flags & FORK_CLOSE_ALL_FDS) { |
1339 | /* Close the logs here in case it got reopened above, as close_all_fds() would close them for us */ | |
1340 | log_close(); | |
1341 | ||
1342 | r = close_all_fds(except_fds, n_except_fds); | |
1343 | if (r < 0) { | |
b6e1fff1 | 1344 | log_full_errno(prio, r, "Failed to close all file descriptors: %m"); |
4c253ed1 LP |
1345 | _exit(EXIT_FAILURE); |
1346 | } | |
1347 | } | |
1348 | ||
1349 | /* When we were asked to reopen the logs, do so again now */ | |
1350 | if (flags & FORK_REOPEN_LOG) { | |
1351 | log_open(); | |
1352 | log_set_open_when_needed(false); | |
1353 | } | |
1354 | ||
1355 | if (flags & FORK_NULL_STDIO) { | |
1356 | r = make_null_stdio(); | |
1357 | if (r < 0) { | |
b6e1fff1 | 1358 | log_full_errno(prio, r, "Failed to connect stdin/stdout to /dev/null: %m"); |
4c253ed1 LP |
1359 | _exit(EXIT_FAILURE); |
1360 | } | |
8987afc4 LP |
1361 | |
1362 | } else if (flags & FORK_STDOUT_TO_STDERR) { | |
8987afc4 | 1363 | if (dup2(STDERR_FILENO, STDOUT_FILENO) < 0) { |
bddeb54c | 1364 | log_full_errno(prio, errno, "Failed to connect stdout to stderr: %m"); |
8987afc4 LP |
1365 | _exit(EXIT_FAILURE); |
1366 | } | |
4c253ed1 LP |
1367 | } |
1368 | ||
909106eb LP |
1369 | if (flags & FORK_RLIMIT_NOFILE_SAFE) { |
1370 | r = rlimit_nofile_safe(); | |
1371 | if (r < 0) { | |
1372 | log_full_errno(prio, r, "Failed to lower RLIMIT_NOFILE's soft limit to 1K: %m"); | |
1373 | _exit(EXIT_FAILURE); | |
1374 | } | |
1375 | } | |
1376 | ||
4c253ed1 LP |
1377 | if (ret_pid) |
1378 | *ret_pid = getpid_cached(); | |
1379 | ||
1380 | return 0; | |
1381 | } | |
1382 | ||
27096982 LP |
1383 | int namespace_fork( |
1384 | const char *outer_name, | |
1385 | const char *inner_name, | |
1386 | const int except_fds[], | |
1387 | size_t n_except_fds, | |
1388 | ForkFlags flags, | |
1389 | int pidns_fd, | |
1390 | int mntns_fd, | |
1391 | int netns_fd, | |
1392 | int userns_fd, | |
1393 | int root_fd, | |
1394 | pid_t *ret_pid) { | |
1395 | ||
1396 | int r; | |
1397 | ||
1398 | /* This is much like safe_fork(), but forks twice, and joins the specified namespaces in the middle | |
1399 | * process. This ensures that we are fully a member of the destination namespace, with pidns an all, so that | |
1400 | * /proc/self/fd works correctly. */ | |
1401 | ||
1402 | 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); | |
1403 | if (r < 0) | |
1404 | return r; | |
1405 | if (r == 0) { | |
1406 | pid_t pid; | |
1407 | ||
1408 | /* Child */ | |
1409 | ||
1410 | r = namespace_enter(pidns_fd, mntns_fd, netns_fd, userns_fd, root_fd); | |
1411 | if (r < 0) { | |
1412 | log_full_errno(FLAGS_SET(flags, FORK_LOG) ? LOG_ERR : LOG_DEBUG, r, "Failed to join namespace: %m"); | |
1413 | _exit(EXIT_FAILURE); | |
1414 | } | |
1415 | ||
1416 | /* We mask a few flags here that either make no sense for the grandchild, or that we don't have to do again */ | |
1417 | 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); | |
1418 | if (r < 0) | |
1419 | _exit(EXIT_FAILURE); | |
1420 | if (r == 0) { | |
1421 | /* Child */ | |
1422 | if (ret_pid) | |
1423 | *ret_pid = pid; | |
1424 | return 0; | |
1425 | } | |
1426 | ||
1427 | r = wait_for_terminate_and_check(inner_name, pid, FLAGS_SET(flags, FORK_LOG) ? WAIT_LOG : 0); | |
1428 | if (r < 0) | |
1429 | _exit(EXIT_FAILURE); | |
1430 | ||
1431 | _exit(r); | |
1432 | } | |
1433 | ||
1434 | return 1; | |
1435 | } | |
1436 | ||
da6053d0 | 1437 | int fork_agent(const char *name, const int except[], size_t n_except, pid_t *ret_pid, const char *path, ...) { |
78752f2e | 1438 | bool stdout_is_tty, stderr_is_tty; |
da6053d0 | 1439 | size_t n, i; |
78752f2e LP |
1440 | va_list ap; |
1441 | char **l; | |
1442 | int r; | |
1443 | ||
1444 | assert(path); | |
1445 | ||
1446 | /* Spawns a temporary TTY agent, making sure it goes away when we go away */ | |
1447 | ||
7e0ed2e9 SB |
1448 | r = safe_fork_full(name, |
1449 | except, | |
1450 | n_except, | |
1451 | FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_CLOSE_ALL_FDS|FORK_REOPEN_LOG, | |
1452 | ret_pid); | |
78752f2e LP |
1453 | if (r < 0) |
1454 | return r; | |
1455 | if (r > 0) | |
1456 | return 0; | |
1457 | ||
1458 | /* In the child: */ | |
1459 | ||
1460 | stdout_is_tty = isatty(STDOUT_FILENO); | |
1461 | stderr_is_tty = isatty(STDERR_FILENO); | |
1462 | ||
1463 | if (!stdout_is_tty || !stderr_is_tty) { | |
1464 | int fd; | |
1465 | ||
1466 | /* Detach from stdout/stderr. and reopen | |
1467 | * /dev/tty for them. This is important to | |
1468 | * ensure that when systemctl is started via | |
1469 | * popen() or a similar call that expects to | |
1470 | * read EOF we actually do generate EOF and | |
1471 | * not delay this indefinitely by because we | |
1472 | * keep an unused copy of stdin around. */ | |
1473 | fd = open("/dev/tty", O_WRONLY); | |
1474 | if (fd < 0) { | |
1475 | log_error_errno(errno, "Failed to open /dev/tty: %m"); | |
1476 | _exit(EXIT_FAILURE); | |
1477 | } | |
1478 | ||
1479 | if (!stdout_is_tty && dup2(fd, STDOUT_FILENO) < 0) { | |
1480 | log_error_errno(errno, "Failed to dup2 /dev/tty: %m"); | |
1481 | _exit(EXIT_FAILURE); | |
1482 | } | |
1483 | ||
1484 | if (!stderr_is_tty && dup2(fd, STDERR_FILENO) < 0) { | |
1485 | log_error_errno(errno, "Failed to dup2 /dev/tty: %m"); | |
1486 | _exit(EXIT_FAILURE); | |
1487 | } | |
1488 | ||
e7685a77 | 1489 | safe_close_above_stdio(fd); |
78752f2e LP |
1490 | } |
1491 | ||
595225af LP |
1492 | (void) rlimit_nofile_safe(); |
1493 | ||
78752f2e LP |
1494 | /* Count arguments */ |
1495 | va_start(ap, path); | |
1496 | for (n = 0; va_arg(ap, char*); n++) | |
1497 | ; | |
1498 | va_end(ap); | |
1499 | ||
1500 | /* Allocate strv */ | |
cf409d15 | 1501 | l = newa(char*, n + 1); |
78752f2e LP |
1502 | |
1503 | /* Fill in arguments */ | |
1504 | va_start(ap, path); | |
1505 | for (i = 0; i <= n; i++) | |
1506 | l[i] = va_arg(ap, char*); | |
1507 | va_end(ap); | |
1508 | ||
1509 | execv(path, l); | |
1510 | _exit(EXIT_FAILURE); | |
1511 | } | |
1512 | ||
9f8168eb LP |
1513 | int set_oom_score_adjust(int value) { |
1514 | char t[DECIMAL_STR_MAX(int)]; | |
1515 | ||
1516 | sprintf(t, "%i", value); | |
1517 | ||
1518 | return write_string_file("/proc/self/oom_score_adj", t, | |
1519 | WRITE_STRING_FILE_VERIFY_ON_FAILURE|WRITE_STRING_FILE_DISABLE_BUFFER); | |
1520 | } | |
1521 | ||
298f466f LP |
1522 | int pidfd_get_pid(int fd, pid_t *ret) { |
1523 | char path[STRLEN("/proc/self/fdinfo/") + DECIMAL_STR_MAX(int)]; | |
1524 | _cleanup_free_ char *fdinfo = NULL; | |
1525 | char *p; | |
1526 | int r; | |
1527 | ||
1528 | if (fd < 0) | |
1529 | return -EBADF; | |
1530 | ||
1531 | xsprintf(path, "/proc/self/fdinfo/%i", fd); | |
1532 | ||
627055ce | 1533 | r = read_full_virtual_file(path, &fdinfo, NULL); |
298f466f LP |
1534 | if (r == -ENOENT) /* if fdinfo doesn't exist we assume the process does not exist */ |
1535 | return -ESRCH; | |
1536 | if (r < 0) | |
1537 | return r; | |
1538 | ||
1539 | p = startswith(fdinfo, "Pid:"); | |
1540 | if (!p) { | |
1541 | p = strstr(fdinfo, "\nPid:"); | |
1542 | if (!p) | |
1543 | return -ENOTTY; /* not a pidfd? */ | |
1544 | ||
1545 | p += 5; | |
1546 | } | |
1547 | ||
1548 | p += strspn(p, WHITESPACE); | |
1549 | p[strcspn(p, WHITESPACE)] = 0; | |
1550 | ||
1551 | return parse_pid(p, ret); | |
1552 | } | |
1553 | ||
39090201 DJL |
1554 | static int rlimit_to_nice(rlim_t limit) { |
1555 | if (limit <= 1) | |
1556 | return PRIO_MAX-1; /* i.e. 19 */ | |
1557 | ||
1558 | if (limit >= -PRIO_MIN + PRIO_MAX) | |
1559 | return PRIO_MIN; /* i.e. -20 */ | |
1560 | ||
1561 | return PRIO_MAX - (int) limit; | |
1562 | } | |
1563 | ||
1564 | int setpriority_closest(int priority) { | |
1565 | int current, limit, saved_errno; | |
1566 | struct rlimit highest; | |
1567 | ||
1568 | /* Try to set requested nice level */ | |
1569 | if (setpriority(PRIO_PROCESS, 0, priority) >= 0) | |
1570 | return 1; | |
1571 | ||
1572 | /* Permission failed */ | |
1573 | saved_errno = -errno; | |
1574 | if (!ERRNO_IS_PRIVILEGE(saved_errno)) | |
1575 | return saved_errno; | |
1576 | ||
1577 | errno = 0; | |
1578 | current = getpriority(PRIO_PROCESS, 0); | |
1579 | if (errno != 0) | |
1580 | return -errno; | |
1581 | ||
1582 | if (priority == current) | |
1583 | return 1; | |
1584 | ||
1585 | /* Hmm, we'd expect that raising the nice level from our status quo would always work. If it doesn't, | |
1586 | * then the whole setpriority() system call is blocked to us, hence let's propagate the error | |
1587 | * right-away */ | |
1588 | if (priority > current) | |
1589 | return saved_errno; | |
1590 | ||
1591 | if (getrlimit(RLIMIT_NICE, &highest) < 0) | |
1592 | return -errno; | |
1593 | ||
1594 | limit = rlimit_to_nice(highest.rlim_cur); | |
1595 | ||
1596 | /* We are already less nice than limit allows us */ | |
1597 | if (current < limit) { | |
1598 | log_debug("Cannot raise nice level, permissions and the resource limit do not allow it."); | |
1599 | return 0; | |
1600 | } | |
1601 | ||
1602 | /* Push to the allowed limit */ | |
1603 | if (setpriority(PRIO_PROCESS, 0, limit) < 0) | |
1604 | return -errno; | |
1605 | ||
1606 | log_debug("Cannot set requested nice level (%i), used next best (%i).", priority, limit); | |
1607 | return 0; | |
1608 | } | |
1609 | ||
2306d177 LP |
1610 | bool invoked_as(char *argv[], const char *token) { |
1611 | if (!argv || isempty(argv[0])) | |
1612 | return false; | |
1613 | ||
1614 | if (isempty(token)) | |
1615 | return false; | |
1616 | ||
1617 | return strstr(last_path_component(argv[0]), token); | |
1618 | } | |
1619 | ||
7b3e062c LP |
1620 | static const char *const ioprio_class_table[] = { |
1621 | [IOPRIO_CLASS_NONE] = "none", | |
1622 | [IOPRIO_CLASS_RT] = "realtime", | |
1623 | [IOPRIO_CLASS_BE] = "best-effort", | |
f44b3035 | 1624 | [IOPRIO_CLASS_IDLE] = "idle", |
7b3e062c LP |
1625 | }; |
1626 | ||
10062bbc | 1627 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ioprio_class, int, IOPRIO_N_CLASSES); |
7b3e062c LP |
1628 | |
1629 | static const char *const sigchld_code_table[] = { | |
1630 | [CLD_EXITED] = "exited", | |
1631 | [CLD_KILLED] = "killed", | |
1632 | [CLD_DUMPED] = "dumped", | |
1633 | [CLD_TRAPPED] = "trapped", | |
1634 | [CLD_STOPPED] = "stopped", | |
1635 | [CLD_CONTINUED] = "continued", | |
1636 | }; | |
1637 | ||
1638 | DEFINE_STRING_TABLE_LOOKUP(sigchld_code, int); | |
1639 | ||
1640 | static const char* const sched_policy_table[] = { | |
1641 | [SCHED_OTHER] = "other", | |
1642 | [SCHED_BATCH] = "batch", | |
1643 | [SCHED_IDLE] = "idle", | |
1644 | [SCHED_FIFO] = "fifo", | |
f44b3035 | 1645 | [SCHED_RR] = "rr", |
7b3e062c LP |
1646 | }; |
1647 | ||
1648 | DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(sched_policy, int, INT_MAX); |