]>
Commit | Line | Data |
---|---|---|
53e1b683 | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
8c6db833 | 2 | |
84ac7bea | 3 | #include <dirent.h> |
8c6db833 | 4 | #include <errno.h> |
84ac7bea | 5 | #include <ftw.h> |
11c3a366 | 6 | #include <limits.h> |
8c6db833 | 7 | #include <signal.h> |
11c3a366 | 8 | #include <stddef.h> |
35bbbf85 | 9 | #include <stdio_ext.h> |
8c6db833 | 10 | #include <stdlib.h> |
84ac7bea | 11 | #include <string.h> |
672c48cc | 12 | #include <sys/stat.h> |
11c3a366 | 13 | #include <sys/statfs.h> |
672c48cc | 14 | #include <sys/types.h> |
f98c2585 | 15 | #include <sys/utsname.h> |
4b58153d | 16 | #include <sys/xattr.h> |
84ac7bea | 17 | #include <unistd.h> |
8c6db833 | 18 | |
b5efdb8a | 19 | #include "alloc-util.h" |
3ffd4af2 | 20 | #include "cgroup-util.h" |
93cc7779 | 21 | #include "def.h" |
a0956174 | 22 | #include "dirent-util.h" |
84ac7bea | 23 | #include "extract-word.h" |
3ffd4af2 | 24 | #include "fd-util.h" |
84ac7bea | 25 | #include "fileio.h" |
f97b34a6 | 26 | #include "format-util.h" |
f4f15635 | 27 | #include "fs-util.h" |
93cc7779 | 28 | #include "log.h" |
84ac7bea LP |
29 | #include "login-util.h" |
30 | #include "macro.h" | |
93cc7779 | 31 | #include "missing.h" |
84ac7bea | 32 | #include "mkdir.h" |
6bedfcbb | 33 | #include "parse-util.h" |
9eb977db | 34 | #include "path-util.h" |
872a590e | 35 | #include "proc-cmdline.h" |
84ac7bea LP |
36 | #include "process-util.h" |
37 | #include "set.h" | |
9444b1f2 | 38 | #include "special.h" |
872a590e | 39 | #include "stat-util.h" |
d054f0a4 | 40 | #include "stdio-util.h" |
8b43440b | 41 | #include "string-table.h" |
07630cea | 42 | #include "string-util.h" |
aae7e17f | 43 | #include "strv.h" |
84ac7bea | 44 | #include "unit-name.h" |
b1d4f8e1 | 45 | #include "user-util.h" |
8c6db833 | 46 | |
c6c18be3 | 47 | int cg_enumerate_processes(const char *controller, const char *path, FILE **_f) { |
7027ff61 | 48 | _cleanup_free_ char *fs = NULL; |
c6c18be3 | 49 | FILE *f; |
7027ff61 | 50 | int r; |
c6c18be3 | 51 | |
c6c18be3 LP |
52 | assert(_f); |
53 | ||
c3175a7f LP |
54 | r = cg_get_path(controller, path, "cgroup.procs", &fs); |
55 | if (r < 0) | |
c6c18be3 LP |
56 | return r; |
57 | ||
58 | f = fopen(fs, "re"); | |
c6c18be3 LP |
59 | if (!f) |
60 | return -errno; | |
61 | ||
62 | *_f = f; | |
63 | return 0; | |
64 | } | |
65 | ||
c6c18be3 LP |
66 | int cg_read_pid(FILE *f, pid_t *_pid) { |
67 | unsigned long ul; | |
68 | ||
69 | /* Note that the cgroup.procs might contain duplicates! See | |
70 | * cgroups.txt for details. */ | |
71 | ||
7027ff61 LP |
72 | assert(f); |
73 | assert(_pid); | |
74 | ||
c6c18be3 LP |
75 | errno = 0; |
76 | if (fscanf(f, "%lu", &ul) != 1) { | |
77 | ||
78 | if (feof(f)) | |
79 | return 0; | |
80 | ||
f5e5c28f | 81 | return errno > 0 ? -errno : -EIO; |
c6c18be3 LP |
82 | } |
83 | ||
84 | if (ul <= 0) | |
85 | return -EIO; | |
86 | ||
87 | *_pid = (pid_t) ul; | |
88 | return 1; | |
89 | } | |
90 | ||
8b238b13 LP |
91 | int cg_read_event( |
92 | const char *controller, | |
93 | const char *path, | |
94 | const char *event, | |
95 | char **val) { | |
96 | ||
ab2c3861 TH |
97 | _cleanup_free_ char *events = NULL, *content = NULL; |
98 | char *p, *line; | |
99 | int r; | |
100 | ||
101 | r = cg_get_path(controller, path, "cgroup.events", &events); | |
102 | if (r < 0) | |
103 | return r; | |
104 | ||
105 | r = read_full_file(events, &content, NULL); | |
106 | if (r < 0) | |
107 | return r; | |
108 | ||
109 | p = content; | |
110 | while ((line = strsep(&p, "\n"))) { | |
111 | char *key; | |
112 | ||
113 | key = strsep(&line, " "); | |
114 | if (!key || !line) | |
115 | return -EINVAL; | |
116 | ||
117 | if (strcmp(key, event)) | |
118 | continue; | |
119 | ||
120 | *val = strdup(line); | |
121 | return 0; | |
122 | } | |
123 | ||
124 | return -ENOENT; | |
125 | } | |
126 | ||
3228995c CB |
127 | bool cg_ns_supported(void) { |
128 | static thread_local int enabled = -1; | |
129 | ||
130 | if (enabled >= 0) | |
131 | return enabled; | |
132 | ||
0887fa71 LP |
133 | if (access("/proc/self/ns/cgroup", F_OK) < 0) { |
134 | if (errno != ENOENT) | |
135 | log_debug_errno(errno, "Failed to check whether /proc/self/ns/cgroup is available, assuming not: %m"); | |
136 | enabled = false; | |
137 | } else | |
138 | enabled = true; | |
3228995c CB |
139 | |
140 | return enabled; | |
141 | } | |
142 | ||
35d2e7ec | 143 | int cg_enumerate_subgroups(const char *controller, const char *path, DIR **_d) { |
7027ff61 | 144 | _cleanup_free_ char *fs = NULL; |
35d2e7ec LP |
145 | int r; |
146 | DIR *d; | |
147 | ||
35d2e7ec LP |
148 | assert(_d); |
149 | ||
150 | /* This is not recursive! */ | |
151 | ||
c3175a7f LP |
152 | r = cg_get_path(controller, path, NULL, &fs); |
153 | if (r < 0) | |
35d2e7ec LP |
154 | return r; |
155 | ||
156 | d = opendir(fs); | |
35d2e7ec LP |
157 | if (!d) |
158 | return -errno; | |
159 | ||
160 | *_d = d; | |
161 | return 0; | |
162 | } | |
163 | ||
164 | int cg_read_subgroup(DIR *d, char **fn) { | |
165 | struct dirent *de; | |
166 | ||
167 | assert(d); | |
7027ff61 | 168 | assert(fn); |
35d2e7ec | 169 | |
f01327ad | 170 | FOREACH_DIRENT_ALL(de, d, return -errno) { |
35d2e7ec LP |
171 | char *b; |
172 | ||
173 | if (de->d_type != DT_DIR) | |
174 | continue; | |
175 | ||
49bfc877 | 176 | if (dot_or_dot_dot(de->d_name)) |
35d2e7ec LP |
177 | continue; |
178 | ||
7027ff61 LP |
179 | b = strdup(de->d_name); |
180 | if (!b) | |
35d2e7ec LP |
181 | return -ENOMEM; |
182 | ||
183 | *fn = b; | |
184 | return 1; | |
185 | } | |
186 | ||
35d2e7ec LP |
187 | return 0; |
188 | } | |
189 | ||
4ad49000 | 190 | int cg_rmdir(const char *controller, const char *path) { |
7027ff61 | 191 | _cleanup_free_ char *p = NULL; |
35d2e7ec LP |
192 | int r; |
193 | ||
ad293f5a LP |
194 | r = cg_get_path(controller, path, NULL, &p); |
195 | if (r < 0) | |
35d2e7ec LP |
196 | return r; |
197 | ||
198 | r = rmdir(p); | |
7027ff61 LP |
199 | if (r < 0 && errno != ENOENT) |
200 | return -errno; | |
35d2e7ec | 201 | |
b4cccbc1 | 202 | r = cg_hybrid_unified(); |
f20db199 | 203 | if (r <= 0) |
b4cccbc1 | 204 | return r; |
b4cccbc1 LP |
205 | |
206 | if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { | |
2977724b TH |
207 | r = cg_rmdir(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path); |
208 | if (r < 0) | |
209 | log_warning_errno(r, "Failed to remove compat systemd cgroup %s: %m", path); | |
210 | } | |
211 | ||
7027ff61 | 212 | return 0; |
35d2e7ec LP |
213 | } |
214 | ||
1d98fef1 LP |
215 | int cg_kill( |
216 | const char *controller, | |
217 | const char *path, | |
218 | int sig, | |
219 | CGroupFlags flags, | |
220 | Set *s, | |
221 | cg_kill_log_func_t log_kill, | |
222 | void *userdata) { | |
223 | ||
7027ff61 | 224 | _cleanup_set_free_ Set *allocated_set = NULL; |
35d2e7ec | 225 | bool done = false; |
c53d2d54 | 226 | int r, ret = 0, ret_log_kill = 0; |
35d2e7ec | 227 | pid_t my_pid; |
8c6db833 | 228 | |
8c6db833 LP |
229 | assert(sig >= 0); |
230 | ||
0d5b4810 LP |
231 | /* Don't send SIGCONT twice. Also, SIGKILL always works even when process is suspended, hence don't send |
232 | * SIGCONT on SIGKILL. */ | |
233 | if (IN_SET(sig, SIGCONT, SIGKILL)) | |
234 | flags &= ~CGROUP_SIGCONT; | |
235 | ||
8c6db833 LP |
236 | /* This goes through the tasks list and kills them all. This |
237 | * is repeated until no further processes are added to the | |
238 | * tasks list, to properly handle forking processes */ | |
239 | ||
7027ff61 | 240 | if (!s) { |
d5099efc | 241 | s = allocated_set = set_new(NULL); |
7027ff61 | 242 | if (!s) |
ca949c9d | 243 | return -ENOMEM; |
7027ff61 | 244 | } |
8c6db833 | 245 | |
df0ff127 | 246 | my_pid = getpid_cached(); |
8c6db833 LP |
247 | |
248 | do { | |
7027ff61 | 249 | _cleanup_fclose_ FILE *f = NULL; |
0b172489 | 250 | pid_t pid = 0; |
8c6db833 LP |
251 | done = true; |
252 | ||
7027ff61 LP |
253 | r = cg_enumerate_processes(controller, path, &f); |
254 | if (r < 0) { | |
4c633005 | 255 | if (ret >= 0 && r != -ENOENT) |
7027ff61 | 256 | return r; |
35d2e7ec | 257 | |
7027ff61 | 258 | return ret; |
35d2e7ec | 259 | } |
c6c18be3 LP |
260 | |
261 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
8c6db833 | 262 | |
1d98fef1 | 263 | if ((flags & CGROUP_IGNORE_SELF) && pid == my_pid) |
c6c18be3 | 264 | continue; |
8c6db833 | 265 | |
fea72cc0 | 266 | if (set_get(s, PID_TO_PTR(pid)) == PID_TO_PTR(pid)) |
c6c18be3 | 267 | continue; |
8c6db833 | 268 | |
1d98fef1 | 269 | if (log_kill) |
c53d2d54 | 270 | ret_log_kill = log_kill(pid, sig, userdata); |
1d98fef1 | 271 | |
8c6db833 LP |
272 | /* If we haven't killed this process yet, kill |
273 | * it */ | |
4c633005 LP |
274 | if (kill(pid, sig) < 0) { |
275 | if (ret >= 0 && errno != ESRCH) | |
8c6db833 | 276 | ret = -errno; |
6e8314c4 | 277 | } else { |
1d98fef1 | 278 | if (flags & CGROUP_SIGCONT) |
e155a0aa | 279 | (void) kill(pid, SIGCONT); |
430c18ed | 280 | |
c53d2d54 DB |
281 | if (ret == 0) { |
282 | if (log_kill) | |
283 | ret = ret_log_kill; | |
284 | else | |
285 | ret = 1; | |
286 | } | |
430c18ed | 287 | } |
8c6db833 | 288 | |
8c6db833 LP |
289 | done = false; |
290 | ||
fea72cc0 | 291 | r = set_put(s, PID_TO_PTR(pid)); |
7027ff61 | 292 | if (r < 0) { |
35d2e7ec | 293 | if (ret >= 0) |
7027ff61 | 294 | return r; |
35d2e7ec | 295 | |
7027ff61 | 296 | return ret; |
35d2e7ec LP |
297 | } |
298 | } | |
299 | ||
300 | if (r < 0) { | |
301 | if (ret >= 0) | |
7027ff61 | 302 | return r; |
35d2e7ec | 303 | |
7027ff61 | 304 | return ret; |
8c6db833 LP |
305 | } |
306 | ||
8c6db833 LP |
307 | /* To avoid racing against processes which fork |
308 | * quicker than we can kill them we repeat this until | |
309 | * no new pids need to be killed. */ | |
310 | ||
35d2e7ec | 311 | } while (!done); |
8c6db833 | 312 | |
35d2e7ec | 313 | return ret; |
8c6db833 LP |
314 | } |
315 | ||
1d98fef1 LP |
316 | int cg_kill_recursive( |
317 | const char *controller, | |
318 | const char *path, | |
319 | int sig, | |
320 | CGroupFlags flags, | |
321 | Set *s, | |
322 | cg_kill_log_func_t log_kill, | |
323 | void *userdata) { | |
324 | ||
7027ff61 LP |
325 | _cleanup_set_free_ Set *allocated_set = NULL; |
326 | _cleanup_closedir_ DIR *d = NULL; | |
e155a0aa | 327 | int r, ret; |
35d2e7ec | 328 | char *fn; |
8c6db833 LP |
329 | |
330 | assert(path); | |
8c6db833 LP |
331 | assert(sig >= 0); |
332 | ||
7027ff61 | 333 | if (!s) { |
d5099efc | 334 | s = allocated_set = set_new(NULL); |
7027ff61 | 335 | if (!s) |
ca949c9d | 336 | return -ENOMEM; |
7027ff61 | 337 | } |
ca949c9d | 338 | |
1d98fef1 | 339 | ret = cg_kill(controller, path, sig, flags, s, log_kill, userdata); |
8c6db833 | 340 | |
7027ff61 LP |
341 | r = cg_enumerate_subgroups(controller, path, &d); |
342 | if (r < 0) { | |
4c633005 | 343 | if (ret >= 0 && r != -ENOENT) |
7027ff61 | 344 | return r; |
8c6db833 | 345 | |
7027ff61 | 346 | return ret; |
35d2e7ec | 347 | } |
8c6db833 | 348 | |
35d2e7ec | 349 | while ((r = cg_read_subgroup(d, &fn)) > 0) { |
7027ff61 | 350 | _cleanup_free_ char *p = NULL; |
8c6db833 | 351 | |
605405c6 | 352 | p = strjoin(path, "/", fn); |
35d2e7ec | 353 | free(fn); |
7027ff61 LP |
354 | if (!p) |
355 | return -ENOMEM; | |
8c6db833 | 356 | |
1d98fef1 | 357 | r = cg_kill_recursive(controller, p, sig, flags, s, log_kill, userdata); |
e155a0aa | 358 | if (r != 0 && ret >= 0) |
35d2e7ec | 359 | ret = r; |
8c6db833 | 360 | } |
7027ff61 | 361 | if (ret >= 0 && r < 0) |
35d2e7ec LP |
362 | ret = r; |
363 | ||
1d98fef1 | 364 | if (flags & CGROUP_REMOVE) { |
4ad49000 | 365 | r = cg_rmdir(controller, path); |
4c701096 | 366 | if (r < 0 && ret >= 0 && !IN_SET(r, -ENOENT, -EBUSY)) |
7027ff61 LP |
367 | return r; |
368 | } | |
ca949c9d | 369 | |
8c6db833 LP |
370 | return ret; |
371 | } | |
372 | ||
1d98fef1 LP |
373 | int cg_migrate( |
374 | const char *cfrom, | |
375 | const char *pfrom, | |
376 | const char *cto, | |
377 | const char *pto, | |
378 | CGroupFlags flags) { | |
379 | ||
35d2e7ec | 380 | bool done = false; |
246aa6dd | 381 | _cleanup_set_free_ Set *s = NULL; |
8c6db833 LP |
382 | int r, ret = 0; |
383 | pid_t my_pid; | |
384 | ||
246aa6dd LP |
385 | assert(cfrom); |
386 | assert(pfrom); | |
387 | assert(cto); | |
388 | assert(pto); | |
8c6db833 | 389 | |
d5099efc | 390 | s = set_new(NULL); |
246aa6dd | 391 | if (!s) |
35d2e7ec LP |
392 | return -ENOMEM; |
393 | ||
df0ff127 | 394 | my_pid = getpid_cached(); |
8c6db833 LP |
395 | |
396 | do { | |
7027ff61 | 397 | _cleanup_fclose_ FILE *f = NULL; |
0b172489 | 398 | pid_t pid = 0; |
8c6db833 LP |
399 | done = true; |
400 | ||
b043cd0b | 401 | r = cg_enumerate_processes(cfrom, pfrom, &f); |
246aa6dd | 402 | if (r < 0) { |
4c633005 | 403 | if (ret >= 0 && r != -ENOENT) |
7027ff61 | 404 | return r; |
35d2e7ec | 405 | |
246aa6dd | 406 | return ret; |
35d2e7ec | 407 | } |
c6c18be3 LP |
408 | |
409 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
8c6db833 | 410 | |
35d2e7ec LP |
411 | /* This might do weird stuff if we aren't a |
412 | * single-threaded program. However, we | |
413 | * luckily know we are not */ | |
1d98fef1 | 414 | if ((flags & CGROUP_IGNORE_SELF) && pid == my_pid) |
c6c18be3 | 415 | continue; |
8c6db833 | 416 | |
fea72cc0 | 417 | if (set_get(s, PID_TO_PTR(pid)) == PID_TO_PTR(pid)) |
35d2e7ec LP |
418 | continue; |
419 | ||
9b84c7f9 LP |
420 | /* Ignore kernel threads. Since they can only |
421 | * exist in the root cgroup, we only check for | |
422 | * them there. */ | |
423 | if (cfrom && | |
57ea45e1 | 424 | empty_or_root(pfrom) && |
9b84c7f9 LP |
425 | is_kernel_thread(pid) > 0) |
426 | continue; | |
427 | ||
246aa6dd LP |
428 | r = cg_attach(cto, pto, pid); |
429 | if (r < 0) { | |
4c633005 | 430 | if (ret >= 0 && r != -ESRCH) |
35d2e7ec LP |
431 | ret = r; |
432 | } else if (ret == 0) | |
433 | ret = 1; | |
8c6db833 | 434 | |
8c6db833 | 435 | done = false; |
35d2e7ec | 436 | |
fea72cc0 | 437 | r = set_put(s, PID_TO_PTR(pid)); |
246aa6dd | 438 | if (r < 0) { |
35d2e7ec | 439 | if (ret >= 0) |
7027ff61 | 440 | return r; |
35d2e7ec | 441 | |
246aa6dd | 442 | return ret; |
35d2e7ec LP |
443 | } |
444 | } | |
445 | ||
446 | if (r < 0) { | |
447 | if (ret >= 0) | |
7027ff61 | 448 | return r; |
35d2e7ec | 449 | |
246aa6dd | 450 | return ret; |
8c6db833 | 451 | } |
35d2e7ec | 452 | } while (!done); |
8c6db833 | 453 | |
35d2e7ec | 454 | return ret; |
8c6db833 LP |
455 | } |
456 | ||
4ad49000 LP |
457 | int cg_migrate_recursive( |
458 | const char *cfrom, | |
459 | const char *pfrom, | |
460 | const char *cto, | |
461 | const char *pto, | |
1d98fef1 | 462 | CGroupFlags flags) { |
4ad49000 | 463 | |
246aa6dd | 464 | _cleanup_closedir_ DIR *d = NULL; |
7027ff61 | 465 | int r, ret = 0; |
35d2e7ec | 466 | char *fn; |
8c6db833 | 467 | |
246aa6dd LP |
468 | assert(cfrom); |
469 | assert(pfrom); | |
470 | assert(cto); | |
471 | assert(pto); | |
8c6db833 | 472 | |
1d98fef1 | 473 | ret = cg_migrate(cfrom, pfrom, cto, pto, flags); |
8c6db833 | 474 | |
246aa6dd LP |
475 | r = cg_enumerate_subgroups(cfrom, pfrom, &d); |
476 | if (r < 0) { | |
4c633005 | 477 | if (ret >= 0 && r != -ENOENT) |
7027ff61 LP |
478 | return r; |
479 | ||
246aa6dd | 480 | return ret; |
35d2e7ec LP |
481 | } |
482 | ||
483 | while ((r = cg_read_subgroup(d, &fn)) > 0) { | |
246aa6dd | 484 | _cleanup_free_ char *p = NULL; |
8c6db833 | 485 | |
605405c6 | 486 | p = strjoin(pfrom, "/", fn); |
35d2e7ec | 487 | free(fn); |
e155a0aa LP |
488 | if (!p) |
489 | return -ENOMEM; | |
8c6db833 | 490 | |
1d98fef1 | 491 | r = cg_migrate_recursive(cfrom, p, cto, pto, flags); |
35d2e7ec LP |
492 | if (r != 0 && ret >= 0) |
493 | ret = r; | |
8c6db833 LP |
494 | } |
495 | ||
35d2e7ec LP |
496 | if (r < 0 && ret >= 0) |
497 | ret = r; | |
498 | ||
1d98fef1 | 499 | if (flags & CGROUP_REMOVE) { |
4ad49000 | 500 | r = cg_rmdir(cfrom, pfrom); |
4c701096 | 501 | if (r < 0 && ret >= 0 && !IN_SET(r, -ENOENT, -EBUSY)) |
246aa6dd LP |
502 | return r; |
503 | } | |
8c6db833 LP |
504 | |
505 | return ret; | |
506 | } | |
507 | ||
13b84ec7 LP |
508 | int cg_migrate_recursive_fallback( |
509 | const char *cfrom, | |
510 | const char *pfrom, | |
511 | const char *cto, | |
512 | const char *pto, | |
1d98fef1 | 513 | CGroupFlags flags) { |
13b84ec7 LP |
514 | |
515 | int r; | |
516 | ||
517 | assert(cfrom); | |
518 | assert(pfrom); | |
519 | assert(cto); | |
520 | assert(pto); | |
521 | ||
1d98fef1 | 522 | r = cg_migrate_recursive(cfrom, pfrom, cto, pto, flags); |
13b84ec7 LP |
523 | if (r < 0) { |
524 | char prefix[strlen(pto) + 1]; | |
525 | ||
526 | /* This didn't work? Then let's try all prefixes of the destination */ | |
527 | ||
fecffe5d | 528 | PATH_FOREACH_PREFIX(prefix, pto) { |
e155a0aa LP |
529 | int q; |
530 | ||
1d98fef1 | 531 | q = cg_migrate_recursive(cfrom, pfrom, cto, prefix, flags); |
e155a0aa LP |
532 | if (q >= 0) |
533 | return q; | |
13b84ec7 LP |
534 | } |
535 | } | |
536 | ||
e155a0aa | 537 | return r; |
13b84ec7 LP |
538 | } |
539 | ||
efdb0237 LP |
540 | static const char *controller_to_dirname(const char *controller) { |
541 | const char *e; | |
3474ae3c | 542 | |
7027ff61 LP |
543 | assert(controller); |
544 | ||
efdb0237 LP |
545 | /* Converts a controller name to the directory name below |
546 | * /sys/fs/cgroup/ we want to mount it to. Effectively, this | |
547 | * just cuts off the name= prefixed used for named | |
548 | * hierarchies, if it is specified. */ | |
549 | ||
2977724b | 550 | if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { |
b4cccbc1 | 551 | if (cg_hybrid_unified() > 0) |
2977724b TH |
552 | controller = SYSTEMD_CGROUP_CONTROLLER_HYBRID; |
553 | else | |
554 | controller = SYSTEMD_CGROUP_CONTROLLER_LEGACY; | |
555 | } | |
b6629c4b | 556 | |
efdb0237 LP |
557 | e = startswith(controller, "name="); |
558 | if (e) | |
559 | return e; | |
560 | ||
561 | return controller; | |
3474ae3c LP |
562 | } |
563 | ||
569b19d8 LP |
564 | static int join_path_legacy(const char *controller, const char *path, const char *suffix, char **fs) { |
565 | const char *dn; | |
018ef268 | 566 | char *t = NULL; |
3474ae3c | 567 | |
efdb0237 | 568 | assert(fs); |
569b19d8 LP |
569 | assert(controller); |
570 | ||
571 | dn = controller_to_dirname(controller); | |
efdb0237 LP |
572 | |
573 | if (isempty(path) && isempty(suffix)) | |
569b19d8 | 574 | t = strappend("/sys/fs/cgroup/", dn); |
efdb0237 | 575 | else if (isempty(path)) |
605405c6 | 576 | t = strjoin("/sys/fs/cgroup/", dn, "/", suffix); |
efdb0237 | 577 | else if (isempty(suffix)) |
605405c6 | 578 | t = strjoin("/sys/fs/cgroup/", dn, "/", path); |
efdb0237 | 579 | else |
605405c6 | 580 | t = strjoin("/sys/fs/cgroup/", dn, "/", path, "/", suffix); |
efdb0237 LP |
581 | if (!t) |
582 | return -ENOMEM; | |
3474ae3c | 583 | |
efdb0237 LP |
584 | *fs = t; |
585 | return 0; | |
586 | } | |
587 | ||
588 | static int join_path_unified(const char *path, const char *suffix, char **fs) { | |
589 | char *t; | |
590 | ||
591 | assert(fs); | |
592 | ||
593 | if (isempty(path) && isempty(suffix)) | |
594 | t = strdup("/sys/fs/cgroup"); | |
595 | else if (isempty(path)) | |
596 | t = strappend("/sys/fs/cgroup/", suffix); | |
597 | else if (isempty(suffix)) | |
598 | t = strappend("/sys/fs/cgroup/", path); | |
599 | else | |
605405c6 | 600 | t = strjoin("/sys/fs/cgroup/", path, "/", suffix); |
3474ae3c LP |
601 | if (!t) |
602 | return -ENOMEM; | |
603 | ||
efdb0237 | 604 | *fs = t; |
3474ae3c LP |
605 | return 0; |
606 | } | |
607 | ||
8c6db833 | 608 | int cg_get_path(const char *controller, const char *path, const char *suffix, char **fs) { |
415fc41c | 609 | int r; |
8c6db833 | 610 | |
dbd821ac LP |
611 | assert(fs); |
612 | ||
efdb0237 LP |
613 | if (!controller) { |
614 | char *t; | |
615 | ||
569b19d8 LP |
616 | /* If no controller is specified, we return the path |
617 | * *below* the controllers, without any prefix. */ | |
efdb0237 LP |
618 | |
619 | if (!path && !suffix) | |
620 | return -EINVAL; | |
621 | ||
989189ea | 622 | if (!suffix) |
efdb0237 | 623 | t = strdup(path); |
989189ea | 624 | else if (!path) |
efdb0237 LP |
625 | t = strdup(suffix); |
626 | else | |
605405c6 | 627 | t = strjoin(path, "/", suffix); |
efdb0237 LP |
628 | if (!t) |
629 | return -ENOMEM; | |
630 | ||
858d36c1 | 631 | *fs = path_simplify(t, false); |
efdb0237 LP |
632 | return 0; |
633 | } | |
634 | ||
635 | if (!cg_controller_is_valid(controller)) | |
78edb35a LP |
636 | return -EINVAL; |
637 | ||
b4cccbc1 LP |
638 | r = cg_all_unified(); |
639 | if (r < 0) | |
640 | return r; | |
641 | if (r > 0) | |
efdb0237 | 642 | r = join_path_unified(path, suffix, fs); |
569b19d8 LP |
643 | else |
644 | r = join_path_legacy(controller, path, suffix, fs); | |
efdb0237 LP |
645 | if (r < 0) |
646 | return r; | |
7027ff61 | 647 | |
858d36c1 | 648 | path_simplify(*fs, false); |
efdb0237 | 649 | return 0; |
3474ae3c | 650 | } |
dbd821ac | 651 | |
efdb0237 | 652 | static int controller_is_accessible(const char *controller) { |
b4cccbc1 | 653 | int r; |
37099707 | 654 | |
efdb0237 | 655 | assert(controller); |
37099707 | 656 | |
efdb0237 LP |
657 | /* Checks whether a specific controller is accessible, |
658 | * i.e. its hierarchy mounted. In the unified hierarchy all | |
659 | * controllers are considered accessible, except for the named | |
660 | * hierarchies */ | |
b12afc8c | 661 | |
efdb0237 LP |
662 | if (!cg_controller_is_valid(controller)) |
663 | return -EINVAL; | |
664 | ||
b4cccbc1 LP |
665 | r = cg_all_unified(); |
666 | if (r < 0) | |
667 | return r; | |
668 | if (r > 0) { | |
efdb0237 LP |
669 | /* We don't support named hierarchies if we are using |
670 | * the unified hierarchy. */ | |
671 | ||
672 | if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) | |
673 | return 0; | |
674 | ||
675 | if (startswith(controller, "name=")) | |
676 | return -EOPNOTSUPP; | |
677 | ||
678 | } else { | |
679 | const char *cc, *dn; | |
680 | ||
681 | dn = controller_to_dirname(controller); | |
682 | cc = strjoina("/sys/fs/cgroup/", dn); | |
683 | ||
684 | if (laccess(cc, F_OK) < 0) | |
685 | return -errno; | |
686 | } | |
37099707 LP |
687 | |
688 | return 0; | |
689 | } | |
690 | ||
3474ae3c | 691 | int cg_get_path_and_check(const char *controller, const char *path, const char *suffix, char **fs) { |
37099707 | 692 | int r; |
dbd821ac | 693 | |
efdb0237 | 694 | assert(controller); |
3474ae3c | 695 | assert(fs); |
70132bd0 | 696 | |
efdb0237 LP |
697 | /* Check if the specified controller is actually accessible */ |
698 | r = controller_is_accessible(controller); | |
37099707 LP |
699 | if (r < 0) |
700 | return r; | |
3474ae3c | 701 | |
efdb0237 | 702 | return cg_get_path(controller, path, suffix, fs); |
8c6db833 LP |
703 | } |
704 | ||
e27796a0 | 705 | static int trim_cb(const char *path, const struct stat *sb, int typeflag, struct FTW *ftwbuf) { |
4ad49000 LP |
706 | assert(path); |
707 | assert(sb); | |
708 | assert(ftwbuf); | |
e27796a0 LP |
709 | |
710 | if (typeflag != FTW_DP) | |
711 | return 0; | |
712 | ||
713 | if (ftwbuf->level < 1) | |
714 | return 0; | |
715 | ||
e155a0aa | 716 | (void) rmdir(path); |
e27796a0 LP |
717 | return 0; |
718 | } | |
719 | ||
8c6db833 | 720 | int cg_trim(const char *controller, const char *path, bool delete_root) { |
7027ff61 | 721 | _cleanup_free_ char *fs = NULL; |
2977724b | 722 | int r = 0, q; |
8c6db833 | 723 | |
8c6db833 LP |
724 | assert(path); |
725 | ||
e27796a0 LP |
726 | r = cg_get_path(controller, path, NULL, &fs); |
727 | if (r < 0) | |
8c6db833 LP |
728 | return r; |
729 | ||
e27796a0 | 730 | errno = 0; |
e155a0aa LP |
731 | if (nftw(fs, trim_cb, 64, FTW_DEPTH|FTW_MOUNT|FTW_PHYS) != 0) { |
732 | if (errno == ENOENT) | |
733 | r = 0; | |
b3267152 | 734 | else if (errno > 0) |
e155a0aa LP |
735 | r = -errno; |
736 | else | |
737 | r = -EIO; | |
738 | } | |
e27796a0 LP |
739 | |
740 | if (delete_root) { | |
4ad49000 LP |
741 | if (rmdir(fs) < 0 && errno != ENOENT) |
742 | return -errno; | |
e27796a0 LP |
743 | } |
744 | ||
b4cccbc1 LP |
745 | q = cg_hybrid_unified(); |
746 | if (q < 0) | |
747 | return q; | |
748 | if (q > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { | |
2977724b TH |
749 | q = cg_trim(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, delete_root); |
750 | if (q < 0) | |
751 | log_warning_errno(q, "Failed to trim compat systemd cgroup %s: %m", path); | |
752 | } | |
753 | ||
e27796a0 | 754 | return r; |
8c6db833 LP |
755 | } |
756 | ||
65be7e06 ZJS |
757 | /* Create a cgroup in the hierarchy of controller. |
758 | * Returns 0 if the group already existed, 1 on success, negative otherwise. | |
759 | */ | |
1434ae6f LP |
760 | int cg_create(const char *controller, const char *path) { |
761 | _cleanup_free_ char *fs = NULL; | |
762 | int r; | |
763 | ||
764 | r = cg_get_path_and_check(controller, path, NULL, &fs); | |
765 | if (r < 0) | |
766 | return r; | |
767 | ||
768 | r = mkdir_parents(fs, 0755); | |
769 | if (r < 0) | |
770 | return r; | |
771 | ||
dae8b82e ZJS |
772 | r = mkdir_errno_wrapper(fs, 0755); |
773 | if (r == -EEXIST) | |
774 | return 0; | |
775 | if (r < 0) | |
776 | return r; | |
1434ae6f | 777 | |
b4cccbc1 LP |
778 | r = cg_hybrid_unified(); |
779 | if (r < 0) | |
780 | return r; | |
781 | ||
782 | if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { | |
2977724b TH |
783 | r = cg_create(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path); |
784 | if (r < 0) | |
785 | log_warning_errno(r, "Failed to create compat systemd cgroup %s: %m", path); | |
786 | } | |
787 | ||
1434ae6f LP |
788 | return 1; |
789 | } | |
790 | ||
791 | int cg_create_and_attach(const char *controller, const char *path, pid_t pid) { | |
792 | int r, q; | |
793 | ||
794 | assert(pid >= 0); | |
795 | ||
796 | r = cg_create(controller, path); | |
797 | if (r < 0) | |
798 | return r; | |
799 | ||
800 | q = cg_attach(controller, path, pid); | |
801 | if (q < 0) | |
802 | return q; | |
803 | ||
804 | /* This does not remove the cgroup on failure */ | |
805 | return r; | |
806 | } | |
807 | ||
8c6db833 | 808 | int cg_attach(const char *controller, const char *path, pid_t pid) { |
574d5f2d LP |
809 | _cleanup_free_ char *fs = NULL; |
810 | char c[DECIMAL_STR_MAX(pid_t) + 2]; | |
8c6db833 LP |
811 | int r; |
812 | ||
8c6db833 LP |
813 | assert(path); |
814 | assert(pid >= 0); | |
815 | ||
b043cd0b | 816 | r = cg_get_path_and_check(controller, path, "cgroup.procs", &fs); |
3474ae3c | 817 | if (r < 0) |
c6c18be3 | 818 | return r; |
8c6db833 LP |
819 | |
820 | if (pid == 0) | |
df0ff127 | 821 | pid = getpid_cached(); |
8c6db833 | 822 | |
d054f0a4 | 823 | xsprintf(c, PID_FMT "\n", pid); |
8c6db833 | 824 | |
604028de | 825 | r = write_string_file(fs, c, WRITE_STRING_FILE_DISABLE_BUFFER); |
2977724b TH |
826 | if (r < 0) |
827 | return r; | |
828 | ||
b4cccbc1 LP |
829 | r = cg_hybrid_unified(); |
830 | if (r < 0) | |
831 | return r; | |
832 | ||
833 | if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { | |
2977724b TH |
834 | r = cg_attach(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, pid); |
835 | if (r < 0) | |
bd68e99b | 836 | log_warning_errno(r, "Failed to attach "PID_FMT" to compat systemd cgroup %s: %m", pid, path); |
2977724b TH |
837 | } |
838 | ||
839 | return 0; | |
8c6db833 LP |
840 | } |
841 | ||
13b84ec7 LP |
842 | int cg_attach_fallback(const char *controller, const char *path, pid_t pid) { |
843 | int r; | |
844 | ||
845 | assert(controller); | |
846 | assert(path); | |
847 | assert(pid >= 0); | |
848 | ||
849 | r = cg_attach(controller, path, pid); | |
850 | if (r < 0) { | |
851 | char prefix[strlen(path) + 1]; | |
852 | ||
853 | /* This didn't work? Then let's try all prefixes of | |
854 | * the destination */ | |
855 | ||
fecffe5d | 856 | PATH_FOREACH_PREFIX(prefix, path) { |
e155a0aa LP |
857 | int q; |
858 | ||
859 | q = cg_attach(controller, prefix, pid); | |
860 | if (q >= 0) | |
861 | return q; | |
13b84ec7 LP |
862 | } |
863 | } | |
864 | ||
e155a0aa | 865 | return r; |
13b84ec7 LP |
866 | } |
867 | ||
62b9bb26 | 868 | int cg_set_access( |
2d76d14e LP |
869 | const char *controller, |
870 | const char *path, | |
2d76d14e LP |
871 | uid_t uid, |
872 | gid_t gid) { | |
873 | ||
62b9bb26 LP |
874 | struct Attribute { |
875 | const char *name; | |
876 | bool fatal; | |
877 | }; | |
878 | ||
4e1dfa45 | 879 | /* cgroup v1, aka legacy/non-unified */ |
62b9bb26 LP |
880 | static const struct Attribute legacy_attributes[] = { |
881 | { "cgroup.procs", true }, | |
882 | { "tasks", false }, | |
883 | { "cgroup.clone_children", false }, | |
884 | {}, | |
885 | }; | |
886 | ||
4e1dfa45 | 887 | /* cgroup v2, aka unified */ |
62b9bb26 LP |
888 | static const struct Attribute unified_attributes[] = { |
889 | { "cgroup.procs", true }, | |
890 | { "cgroup.subtree_control", true }, | |
891 | { "cgroup.threads", false }, | |
892 | {}, | |
893 | }; | |
894 | ||
895 | static const struct Attribute* const attributes[] = { | |
896 | [false] = legacy_attributes, | |
897 | [true] = unified_attributes, | |
898 | }; | |
974efc46 | 899 | |
40853aa5 | 900 | _cleanup_free_ char *fs = NULL; |
62b9bb26 LP |
901 | const struct Attribute *i; |
902 | int r, unified; | |
8c6db833 | 903 | |
8c6db833 LP |
904 | assert(path); |
905 | ||
62b9bb26 | 906 | if (uid == UID_INVALID && gid == GID_INVALID) |
8d53b453 LP |
907 | return 0; |
908 | ||
62b9bb26 LP |
909 | unified = cg_unified_controller(controller); |
910 | if (unified < 0) | |
911 | return unified; | |
8c6db833 | 912 | |
62b9bb26 LP |
913 | /* Configure access to the cgroup itself */ |
914 | r = cg_get_path(controller, path, NULL, &fs); | |
974efc46 LP |
915 | if (r < 0) |
916 | return r; | |
8c6db833 | 917 | |
62b9bb26 | 918 | r = chmod_and_chown(fs, 0755, uid, gid); |
b4cccbc1 LP |
919 | if (r < 0) |
920 | return r; | |
40853aa5 | 921 | |
62b9bb26 LP |
922 | /* Configure access to the cgroup's attributes */ |
923 | for (i = attributes[unified]; i->name; i++) { | |
40853aa5 | 924 | fs = mfree(fs); |
40853aa5 | 925 | |
62b9bb26 | 926 | r = cg_get_path(controller, path, i->name, &fs); |
40853aa5 LP |
927 | if (r < 0) |
928 | return r; | |
efdb0237 | 929 | |
62b9bb26 LP |
930 | r = chmod_and_chown(fs, 0644, uid, gid); |
931 | if (r < 0) { | |
932 | if (i->fatal) | |
933 | return r; | |
5beac75e | 934 | |
62b9bb26 LP |
935 | log_debug_errno(r, "Failed to set access on cgroup %s, ignoring: %m", fs); |
936 | } | |
937 | } | |
938 | ||
939 | if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { | |
940 | r = cg_hybrid_unified(); | |
2977724b | 941 | if (r < 0) |
62b9bb26 LP |
942 | return r; |
943 | if (r > 0) { | |
944 | /* Always propagate access mode from unified to legacy controller */ | |
945 | r = cg_set_access(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, uid, gid); | |
946 | if (r < 0) | |
947 | log_debug_errno(r, "Failed to set access on compatibility systemd cgroup %s, ignoring: %m", path); | |
948 | } | |
2977724b | 949 | } |
974efc46 | 950 | |
efdb0237 | 951 | return 0; |
8c6db833 LP |
952 | } |
953 | ||
4b58153d LP |
954 | int cg_set_xattr(const char *controller, const char *path, const char *name, const void *value, size_t size, int flags) { |
955 | _cleanup_free_ char *fs = NULL; | |
956 | int r; | |
957 | ||
958 | assert(path); | |
959 | assert(name); | |
960 | assert(value || size <= 0); | |
961 | ||
962 | r = cg_get_path(controller, path, NULL, &fs); | |
963 | if (r < 0) | |
964 | return r; | |
965 | ||
966 | if (setxattr(fs, name, value, size, flags) < 0) | |
967 | return -errno; | |
968 | ||
969 | return 0; | |
970 | } | |
971 | ||
972 | int cg_get_xattr(const char *controller, const char *path, const char *name, void *value, size_t size) { | |
973 | _cleanup_free_ char *fs = NULL; | |
974 | ssize_t n; | |
975 | int r; | |
976 | ||
977 | assert(path); | |
978 | assert(name); | |
979 | ||
980 | r = cg_get_path(controller, path, NULL, &fs); | |
981 | if (r < 0) | |
982 | return r; | |
983 | ||
984 | n = getxattr(fs, name, value, size); | |
985 | if (n < 0) | |
986 | return -errno; | |
987 | ||
988 | return (int) n; | |
989 | } | |
990 | ||
7027ff61 | 991 | int cg_pid_get_path(const char *controller, pid_t pid, char **path) { |
7027ff61 | 992 | _cleanup_fclose_ FILE *f = NULL; |
b6629c4b | 993 | const char *fs, *controller_str; |
d2b39cb6 | 994 | int unified, r; |
efdb0237 | 995 | size_t cs = 0; |
8c6db833 | 996 | |
8c6db833 | 997 | assert(path); |
c6c18be3 | 998 | assert(pid >= 0); |
8c6db833 | 999 | |
5da38d07 TH |
1000 | if (controller) { |
1001 | if (!cg_controller_is_valid(controller)) | |
1002 | return -EINVAL; | |
1003 | } else | |
1004 | controller = SYSTEMD_CGROUP_CONTROLLER; | |
1005 | ||
c22800e4 | 1006 | unified = cg_unified_controller(controller); |
b4cccbc1 LP |
1007 | if (unified < 0) |
1008 | return unified; | |
1009 | if (unified == 0) { | |
b6629c4b TH |
1010 | if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) |
1011 | controller_str = SYSTEMD_CGROUP_CONTROLLER_LEGACY; | |
1012 | else | |
1013 | controller_str = controller; | |
1014 | ||
1015 | cs = strlen(controller_str); | |
1016 | } | |
7027ff61 | 1017 | |
b68fa010 | 1018 | fs = procfs_file_alloca(pid, "cgroup"); |
c6c18be3 | 1019 | f = fopen(fs, "re"); |
4c633005 LP |
1020 | if (!f) |
1021 | return errno == ENOENT ? -ESRCH : -errno; | |
1022 | ||
35bbbf85 LP |
1023 | (void) __fsetlocking(f, FSETLOCKING_BYCALLER); |
1024 | ||
d2b39cb6 LP |
1025 | for (;;) { |
1026 | _cleanup_free_ char *line = NULL; | |
efdb0237 | 1027 | char *e, *p; |
c6c18be3 | 1028 | |
d2b39cb6 LP |
1029 | r = read_line(f, LONG_LINE_MAX, &line); |
1030 | if (r < 0) | |
1031 | return r; | |
1032 | if (r == 0) | |
1033 | break; | |
c6c18be3 | 1034 | |
efdb0237 LP |
1035 | if (unified) { |
1036 | e = startswith(line, "0:"); | |
1037 | if (!e) | |
1038 | continue; | |
c6c18be3 | 1039 | |
efdb0237 LP |
1040 | e = strchr(e, ':'); |
1041 | if (!e) | |
1042 | continue; | |
1043 | } else { | |
1044 | char *l; | |
1045 | size_t k; | |
1046 | const char *word, *state; | |
1047 | bool found = false; | |
1048 | ||
1049 | l = strchr(line, ':'); | |
1050 | if (!l) | |
1051 | continue; | |
8af8afd6 | 1052 | |
efdb0237 LP |
1053 | l++; |
1054 | e = strchr(l, ':'); | |
1055 | if (!e) | |
1056 | continue; | |
8af8afd6 | 1057 | |
efdb0237 | 1058 | *e = 0; |
00d4b1e6 | 1059 | FOREACH_WORD_SEPARATOR(word, k, l, ",", state) |
b6629c4b | 1060 | if (k == cs && memcmp(word, controller_str, cs) == 0) { |
efdb0237 LP |
1061 | found = true; |
1062 | break; | |
1063 | } | |
efdb0237 LP |
1064 | if (!found) |
1065 | continue; | |
8af8afd6 LP |
1066 | } |
1067 | ||
8af8afd6 | 1068 | p = strdup(e + 1); |
7027ff61 LP |
1069 | if (!p) |
1070 | return -ENOMEM; | |
c6c18be3 | 1071 | |
5e20b0a4 LP |
1072 | /* Truncate suffix indicating the process is a zombie */ |
1073 | e = endswith(p, " (deleted)"); | |
1074 | if (e) | |
1075 | *e = 0; | |
1076 | ||
c6c18be3 | 1077 | *path = p; |
7027ff61 | 1078 | return 0; |
c6c18be3 LP |
1079 | } |
1080 | ||
1c80e425 | 1081 | return -ENODATA; |
8c6db833 LP |
1082 | } |
1083 | ||
1084 | int cg_install_release_agent(const char *controller, const char *agent) { | |
7027ff61 | 1085 | _cleanup_free_ char *fs = NULL, *contents = NULL; |
efdb0237 | 1086 | const char *sc; |
415fc41c | 1087 | int r; |
8c6db833 | 1088 | |
8c6db833 LP |
1089 | assert(agent); |
1090 | ||
c22800e4 | 1091 | r = cg_unified_controller(controller); |
b4cccbc1 LP |
1092 | if (r < 0) |
1093 | return r; | |
1094 | if (r > 0) /* doesn't apply to unified hierarchy */ | |
efdb0237 LP |
1095 | return -EOPNOTSUPP; |
1096 | ||
7027ff61 LP |
1097 | r = cg_get_path(controller, NULL, "release_agent", &fs); |
1098 | if (r < 0) | |
c6c18be3 | 1099 | return r; |
8c6db833 | 1100 | |
7027ff61 LP |
1101 | r = read_one_line_file(fs, &contents); |
1102 | if (r < 0) | |
1103 | return r; | |
8c6db833 LP |
1104 | |
1105 | sc = strstrip(contents); | |
e155a0aa | 1106 | if (isempty(sc)) { |
604028de | 1107 | r = write_string_file(fs, agent, WRITE_STRING_FILE_DISABLE_BUFFER); |
574d5f2d | 1108 | if (r < 0) |
7027ff61 | 1109 | return r; |
b8725df8 | 1110 | } else if (!path_equal(sc, agent)) |
7027ff61 | 1111 | return -EEXIST; |
8c6db833 | 1112 | |
0da16248 | 1113 | fs = mfree(fs); |
7027ff61 LP |
1114 | r = cg_get_path(controller, NULL, "notify_on_release", &fs); |
1115 | if (r < 0) | |
1116 | return r; | |
8c6db833 | 1117 | |
0da16248 | 1118 | contents = mfree(contents); |
7027ff61 LP |
1119 | r = read_one_line_file(fs, &contents); |
1120 | if (r < 0) | |
1121 | return r; | |
8c6db833 LP |
1122 | |
1123 | sc = strstrip(contents); | |
8c6db833 | 1124 | if (streq(sc, "0")) { |
604028de | 1125 | r = write_string_file(fs, "1", WRITE_STRING_FILE_DISABLE_BUFFER); |
7027ff61 LP |
1126 | if (r < 0) |
1127 | return r; | |
c6c18be3 | 1128 | |
7027ff61 LP |
1129 | return 1; |
1130 | } | |
8c6db833 | 1131 | |
7027ff61 LP |
1132 | if (!streq(sc, "1")) |
1133 | return -EIO; | |
8c6db833 | 1134 | |
7027ff61 | 1135 | return 0; |
8c6db833 LP |
1136 | } |
1137 | ||
ad929bcc KS |
1138 | int cg_uninstall_release_agent(const char *controller) { |
1139 | _cleanup_free_ char *fs = NULL; | |
415fc41c | 1140 | int r; |
efdb0237 | 1141 | |
c22800e4 | 1142 | r = cg_unified_controller(controller); |
b4cccbc1 LP |
1143 | if (r < 0) |
1144 | return r; | |
1145 | if (r > 0) /* Doesn't apply to unified hierarchy */ | |
efdb0237 | 1146 | return -EOPNOTSUPP; |
ad929bcc | 1147 | |
ac9ef333 LP |
1148 | r = cg_get_path(controller, NULL, "notify_on_release", &fs); |
1149 | if (r < 0) | |
1150 | return r; | |
1151 | ||
604028de | 1152 | r = write_string_file(fs, "0", WRITE_STRING_FILE_DISABLE_BUFFER); |
ac9ef333 LP |
1153 | if (r < 0) |
1154 | return r; | |
1155 | ||
0da16248 | 1156 | fs = mfree(fs); |
ac9ef333 | 1157 | |
ad929bcc KS |
1158 | r = cg_get_path(controller, NULL, "release_agent", &fs); |
1159 | if (r < 0) | |
1160 | return r; | |
1161 | ||
604028de | 1162 | r = write_string_file(fs, "", WRITE_STRING_FILE_DISABLE_BUFFER); |
ad929bcc KS |
1163 | if (r < 0) |
1164 | return r; | |
1165 | ||
ac9ef333 | 1166 | return 0; |
ad929bcc KS |
1167 | } |
1168 | ||
6f883237 | 1169 | int cg_is_empty(const char *controller, const char *path) { |
7027ff61 | 1170 | _cleanup_fclose_ FILE *f = NULL; |
efdb0237 | 1171 | pid_t pid; |
7027ff61 | 1172 | int r; |
8c6db833 | 1173 | |
8c6db833 LP |
1174 | assert(path); |
1175 | ||
b043cd0b | 1176 | r = cg_enumerate_processes(controller, path, &f); |
6f883237 | 1177 | if (r == -ENOENT) |
1bcf3fc6 | 1178 | return true; |
c3175a7f | 1179 | if (r < 0) |
6f883237 | 1180 | return r; |
8c6db833 | 1181 | |
6f883237 | 1182 | r = cg_read_pid(f, &pid); |
c6c18be3 LP |
1183 | if (r < 0) |
1184 | return r; | |
8c6db833 | 1185 | |
6f883237 | 1186 | return r == 0; |
8c6db833 LP |
1187 | } |
1188 | ||
6f883237 | 1189 | int cg_is_empty_recursive(const char *controller, const char *path) { |
415fc41c | 1190 | int r; |
8c6db833 | 1191 | |
8c6db833 LP |
1192 | assert(path); |
1193 | ||
6fd66507 | 1194 | /* The root cgroup is always populated */ |
57ea45e1 | 1195 | if (controller && empty_or_root(path)) |
efdb0237 | 1196 | return false; |
6fd66507 | 1197 | |
c22800e4 | 1198 | r = cg_unified_controller(controller); |
b4cccbc1 LP |
1199 | if (r < 0) |
1200 | return r; | |
1201 | if (r > 0) { | |
ab2c3861 | 1202 | _cleanup_free_ char *t = NULL; |
8c6db833 | 1203 | |
efdb0237 | 1204 | /* On the unified hierarchy we can check empty state |
ab2c3861 | 1205 | * via the "populated" attribute of "cgroup.events". */ |
8c6db833 | 1206 | |
ab2c3861 | 1207 | r = cg_read_event(controller, path, "populated", &t); |
1bcf3fc6 ZJS |
1208 | if (r == -ENOENT) |
1209 | return true; | |
efdb0237 LP |
1210 | if (r < 0) |
1211 | return r; | |
1212 | ||
1213 | return streq(t, "0"); | |
1214 | } else { | |
1215 | _cleanup_closedir_ DIR *d = NULL; | |
1216 | char *fn; | |
8c6db833 | 1217 | |
efdb0237 | 1218 | r = cg_is_empty(controller, path); |
35d2e7ec | 1219 | if (r <= 0) |
7027ff61 | 1220 | return r; |
35d2e7ec | 1221 | |
efdb0237 LP |
1222 | r = cg_enumerate_subgroups(controller, path, &d); |
1223 | if (r == -ENOENT) | |
1bcf3fc6 | 1224 | return true; |
efdb0237 LP |
1225 | if (r < 0) |
1226 | return r; | |
35d2e7ec | 1227 | |
efdb0237 LP |
1228 | while ((r = cg_read_subgroup(d, &fn)) > 0) { |
1229 | _cleanup_free_ char *p = NULL; | |
1230 | ||
605405c6 | 1231 | p = strjoin(path, "/", fn); |
efdb0237 LP |
1232 | free(fn); |
1233 | if (!p) | |
1234 | return -ENOMEM; | |
1235 | ||
1236 | r = cg_is_empty_recursive(controller, p); | |
1237 | if (r <= 0) | |
1238 | return r; | |
1239 | } | |
1240 | if (r < 0) | |
1241 | return r; | |
1242 | ||
1243 | return true; | |
1244 | } | |
35d2e7ec LP |
1245 | } |
1246 | ||
1247 | int cg_split_spec(const char *spec, char **controller, char **path) { | |
35d2e7ec | 1248 | char *t = NULL, *u = NULL; |
efdb0237 | 1249 | const char *e; |
35d2e7ec LP |
1250 | |
1251 | assert(spec); | |
35d2e7ec LP |
1252 | |
1253 | if (*spec == '/') { | |
99be45a4 | 1254 | if (!path_is_normalized(spec)) |
e884315e | 1255 | return -EINVAL; |
35d2e7ec LP |
1256 | |
1257 | if (path) { | |
246aa6dd LP |
1258 | t = strdup(spec); |
1259 | if (!t) | |
35d2e7ec LP |
1260 | return -ENOMEM; |
1261 | ||
858d36c1 | 1262 | *path = path_simplify(t, false); |
8c6db833 LP |
1263 | } |
1264 | ||
35d2e7ec LP |
1265 | if (controller) |
1266 | *controller = NULL; | |
1267 | ||
1268 | return 0; | |
8c6db833 LP |
1269 | } |
1270 | ||
246aa6dd LP |
1271 | e = strchr(spec, ':'); |
1272 | if (!e) { | |
185a0874 | 1273 | if (!cg_controller_is_valid(spec)) |
35d2e7ec LP |
1274 | return -EINVAL; |
1275 | ||
1276 | if (controller) { | |
efdb0237 | 1277 | t = strdup(spec); |
246aa6dd | 1278 | if (!t) |
35d2e7ec LP |
1279 | return -ENOMEM; |
1280 | ||
1281 | *controller = t; | |
1282 | } | |
1283 | ||
1284 | if (path) | |
1285 | *path = NULL; | |
1286 | ||
1287 | return 0; | |
8c6db833 LP |
1288 | } |
1289 | ||
efdb0237 | 1290 | t = strndup(spec, e-spec); |
e884315e LP |
1291 | if (!t) |
1292 | return -ENOMEM; | |
185a0874 | 1293 | if (!cg_controller_is_valid(t)) { |
e884315e | 1294 | free(t); |
35d2e7ec | 1295 | return -EINVAL; |
246aa6dd LP |
1296 | } |
1297 | ||
efdb0237 LP |
1298 | if (isempty(e+1)) |
1299 | u = NULL; | |
1300 | else { | |
baa89da4 LP |
1301 | u = strdup(e+1); |
1302 | if (!u) { | |
1303 | free(t); | |
1304 | return -ENOMEM; | |
1305 | } | |
35d2e7ec | 1306 | |
99be45a4 | 1307 | if (!path_is_normalized(u) || |
baa89da4 LP |
1308 | !path_is_absolute(u)) { |
1309 | free(t); | |
1310 | free(u); | |
1311 | return -EINVAL; | |
1312 | } | |
1313 | ||
858d36c1 | 1314 | path_simplify(u, false); |
baa89da4 | 1315 | } |
5954c074 | 1316 | |
35d2e7ec LP |
1317 | if (controller) |
1318 | *controller = t; | |
e884315e LP |
1319 | else |
1320 | free(t); | |
35d2e7ec LP |
1321 | |
1322 | if (path) | |
1323 | *path = u; | |
e884315e LP |
1324 | else |
1325 | free(u); | |
35d2e7ec LP |
1326 | |
1327 | return 0; | |
8c6db833 | 1328 | } |
c6c18be3 | 1329 | |
7027ff61 | 1330 | int cg_mangle_path(const char *path, char **result) { |
78edb35a LP |
1331 | _cleanup_free_ char *c = NULL, *p = NULL; |
1332 | char *t; | |
35d2e7ec LP |
1333 | int r; |
1334 | ||
1335 | assert(path); | |
1336 | assert(result); | |
1337 | ||
73e231ab | 1338 | /* First, check if it already is a filesystem path */ |
7027ff61 | 1339 | if (path_startswith(path, "/sys/fs/cgroup")) { |
35d2e7ec | 1340 | |
b69d29ce LP |
1341 | t = strdup(path); |
1342 | if (!t) | |
35d2e7ec LP |
1343 | return -ENOMEM; |
1344 | ||
858d36c1 | 1345 | *result = path_simplify(t, false); |
35d2e7ec LP |
1346 | return 0; |
1347 | } | |
1348 | ||
73e231ab | 1349 | /* Otherwise, treat it as cg spec */ |
b69d29ce LP |
1350 | r = cg_split_spec(path, &c, &p); |
1351 | if (r < 0) | |
35d2e7ec LP |
1352 | return r; |
1353 | ||
efdb0237 | 1354 | return cg_get_path(c ?: SYSTEMD_CGROUP_CONTROLLER, p ?: "/", NULL, result); |
35d2e7ec | 1355 | } |
1f73f0f1 | 1356 | |
7027ff61 | 1357 | int cg_get_root_path(char **path) { |
9444b1f2 | 1358 | char *p, *e; |
7027ff61 LP |
1359 | int r; |
1360 | ||
1361 | assert(path); | |
1362 | ||
9444b1f2 | 1363 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 1, &p); |
7027ff61 LP |
1364 | if (r < 0) |
1365 | return r; | |
1366 | ||
efdb0237 LP |
1367 | e = endswith(p, "/" SPECIAL_INIT_SCOPE); |
1368 | if (!e) | |
1369 | e = endswith(p, "/" SPECIAL_SYSTEM_SLICE); /* legacy */ | |
1370 | if (!e) | |
1371 | e = endswith(p, "/system"); /* even more legacy */ | |
9444b1f2 | 1372 | if (e) |
7027ff61 LP |
1373 | *e = 0; |
1374 | ||
1f73f0f1 LP |
1375 | *path = p; |
1376 | return 0; | |
1377 | } | |
b59e2465 | 1378 | |
751bc6ac LP |
1379 | int cg_shift_path(const char *cgroup, const char *root, const char **shifted) { |
1380 | _cleanup_free_ char *rt = NULL; | |
1381 | char *p; | |
ba1261bc LP |
1382 | int r; |
1383 | ||
e9174f29 | 1384 | assert(cgroup); |
751bc6ac | 1385 | assert(shifted); |
e9174f29 LP |
1386 | |
1387 | if (!root) { | |
1388 | /* If the root was specified let's use that, otherwise | |
1389 | * let's determine it from PID 1 */ | |
1390 | ||
751bc6ac | 1391 | r = cg_get_root_path(&rt); |
e9174f29 LP |
1392 | if (r < 0) |
1393 | return r; | |
1394 | ||
751bc6ac | 1395 | root = rt; |
e9174f29 | 1396 | } |
ba1261bc | 1397 | |
751bc6ac | 1398 | p = path_startswith(cgroup, root); |
efdb0237 | 1399 | if (p && p > cgroup) |
751bc6ac LP |
1400 | *shifted = p - 1; |
1401 | else | |
1402 | *shifted = cgroup; | |
1403 | ||
1404 | return 0; | |
1405 | } | |
1406 | ||
1407 | int cg_pid_get_path_shifted(pid_t pid, const char *root, char **cgroup) { | |
1408 | _cleanup_free_ char *raw = NULL; | |
1409 | const char *c; | |
1410 | int r; | |
1411 | ||
1412 | assert(pid >= 0); | |
1413 | assert(cgroup); | |
1414 | ||
1415 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &raw); | |
7027ff61 | 1416 | if (r < 0) |
ba1261bc | 1417 | return r; |
ba1261bc | 1418 | |
751bc6ac LP |
1419 | r = cg_shift_path(raw, root, &c); |
1420 | if (r < 0) | |
1421 | return r; | |
ba1261bc | 1422 | |
ae2a15bc LP |
1423 | if (c == raw) |
1424 | *cgroup = TAKE_PTR(raw); | |
1425 | else { | |
751bc6ac | 1426 | char *n; |
ba1261bc | 1427 | |
751bc6ac LP |
1428 | n = strdup(c); |
1429 | if (!n) | |
ba1261bc | 1430 | return -ENOMEM; |
ba1261bc | 1431 | |
751bc6ac LP |
1432 | *cgroup = n; |
1433 | } | |
ba1261bc LP |
1434 | |
1435 | return 0; | |
1436 | } | |
1437 | ||
9ed794a3 | 1438 | int cg_path_decode_unit(const char *cgroup, char **unit) { |
8b0849e9 LP |
1439 | char *c, *s; |
1440 | size_t n; | |
ef1673d1 MT |
1441 | |
1442 | assert(cgroup); | |
6c03089c | 1443 | assert(unit); |
ef1673d1 | 1444 | |
8b0849e9 LP |
1445 | n = strcspn(cgroup, "/"); |
1446 | if (n < 3) | |
1447 | return -ENXIO; | |
1448 | ||
1449 | c = strndupa(cgroup, n); | |
ae018d9b | 1450 | c = cg_unescape(c); |
ef1673d1 | 1451 | |
7410616c | 1452 | if (!unit_name_is_valid(c, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) |
cfeaa44a | 1453 | return -ENXIO; |
ef1673d1 | 1454 | |
d7bd3de0 | 1455 | s = strdup(c); |
6c03089c LP |
1456 | if (!s) |
1457 | return -ENOMEM; | |
1458 | ||
1459 | *unit = s; | |
ef1673d1 MT |
1460 | return 0; |
1461 | } | |
1462 | ||
8b0849e9 LP |
1463 | static bool valid_slice_name(const char *p, size_t n) { |
1464 | ||
1465 | if (!p) | |
1466 | return false; | |
1467 | ||
fbd0b64f | 1468 | if (n < STRLEN("x.slice")) |
8b0849e9 LP |
1469 | return false; |
1470 | ||
1471 | if (memcmp(p + n - 6, ".slice", 6) == 0) { | |
1472 | char buf[n+1], *c; | |
1473 | ||
1474 | memcpy(buf, p, n); | |
1475 | buf[n] = 0; | |
1476 | ||
1477 | c = cg_unescape(buf); | |
1478 | ||
7410616c | 1479 | return unit_name_is_valid(c, UNIT_NAME_PLAIN); |
8b0849e9 LP |
1480 | } |
1481 | ||
1482 | return false; | |
1483 | } | |
1484 | ||
9444b1f2 | 1485 | static const char *skip_slices(const char *p) { |
8b0849e9 LP |
1486 | assert(p); |
1487 | ||
9444b1f2 LP |
1488 | /* Skips over all slice assignments */ |
1489 | ||
1490 | for (;;) { | |
1021b21b LP |
1491 | size_t n; |
1492 | ||
9444b1f2 LP |
1493 | p += strspn(p, "/"); |
1494 | ||
1495 | n = strcspn(p, "/"); | |
8b0849e9 | 1496 | if (!valid_slice_name(p, n)) |
9444b1f2 LP |
1497 | return p; |
1498 | ||
1499 | p += n; | |
1500 | } | |
1501 | } | |
1502 | ||
8b0849e9 | 1503 | int cg_path_get_unit(const char *path, char **ret) { |
6c03089c | 1504 | const char *e; |
8b0849e9 LP |
1505 | char *unit; |
1506 | int r; | |
6c03089c LP |
1507 | |
1508 | assert(path); | |
8b0849e9 | 1509 | assert(ret); |
6c03089c | 1510 | |
9444b1f2 | 1511 | e = skip_slices(path); |
6c03089c | 1512 | |
8b0849e9 LP |
1513 | r = cg_path_decode_unit(e, &unit); |
1514 | if (r < 0) | |
1515 | return r; | |
1516 | ||
1517 | /* We skipped over the slices, don't accept any now */ | |
1518 | if (endswith(unit, ".slice")) { | |
1519 | free(unit); | |
1520 | return -ENXIO; | |
1521 | } | |
1522 | ||
1523 | *ret = unit; | |
1524 | return 0; | |
6c03089c LP |
1525 | } |
1526 | ||
1527 | int cg_pid_get_unit(pid_t pid, char **unit) { | |
7fd1b19b | 1528 | _cleanup_free_ char *cgroup = NULL; |
ba1261bc | 1529 | int r; |
ba1261bc | 1530 | |
ef1673d1 MT |
1531 | assert(unit); |
1532 | ||
7027ff61 | 1533 | r = cg_pid_get_path_shifted(pid, NULL, &cgroup); |
ef1673d1 MT |
1534 | if (r < 0) |
1535 | return r; | |
1536 | ||
6c03089c LP |
1537 | return cg_path_get_unit(cgroup, unit); |
1538 | } | |
ef1673d1 | 1539 | |
d4fffc4b ZJS |
1540 | /** |
1541 | * Skip session-*.scope, but require it to be there. | |
1542 | */ | |
9444b1f2 LP |
1543 | static const char *skip_session(const char *p) { |
1544 | size_t n; | |
1545 | ||
8b0849e9 LP |
1546 | if (isempty(p)) |
1547 | return NULL; | |
9444b1f2 LP |
1548 | |
1549 | p += strspn(p, "/"); | |
1550 | ||
1551 | n = strcspn(p, "/"); | |
fbd0b64f | 1552 | if (n < STRLEN("session-x.scope")) |
d4fffc4b ZJS |
1553 | return NULL; |
1554 | ||
8b0849e9 LP |
1555 | if (memcmp(p, "session-", 8) == 0 && memcmp(p + n - 6, ".scope", 6) == 0) { |
1556 | char buf[n - 8 - 6 + 1]; | |
1557 | ||
1558 | memcpy(buf, p + 8, n - 8 - 6); | |
1559 | buf[n - 8 - 6] = 0; | |
d4fffc4b | 1560 | |
8b0849e9 LP |
1561 | /* Note that session scopes never need unescaping, |
1562 | * since they cannot conflict with the kernel's own | |
1563 | * names, hence we don't need to call cg_unescape() | |
1564 | * here. */ | |
1565 | ||
1566 | if (!session_id_valid(buf)) | |
1567 | return false; | |
1568 | ||
1569 | p += n; | |
1570 | p += strspn(p, "/"); | |
1571 | return p; | |
1572 | } | |
1573 | ||
1574 | return NULL; | |
d4fffc4b ZJS |
1575 | } |
1576 | ||
1577 | /** | |
1578 | * Skip user@*.service, but require it to be there. | |
1579 | */ | |
1580 | static const char *skip_user_manager(const char *p) { | |
1581 | size_t n; | |
1582 | ||
8b0849e9 LP |
1583 | if (isempty(p)) |
1584 | return NULL; | |
d4fffc4b ZJS |
1585 | |
1586 | p += strspn(p, "/"); | |
1587 | ||
1588 | n = strcspn(p, "/"); | |
fbd0b64f | 1589 | if (n < STRLEN("user@x.service")) |
6c03089c | 1590 | return NULL; |
ef1673d1 | 1591 | |
8b0849e9 LP |
1592 | if (memcmp(p, "user@", 5) == 0 && memcmp(p + n - 8, ".service", 8) == 0) { |
1593 | char buf[n - 5 - 8 + 1]; | |
9444b1f2 | 1594 | |
8b0849e9 LP |
1595 | memcpy(buf, p + 5, n - 5 - 8); |
1596 | buf[n - 5 - 8] = 0; | |
1597 | ||
1598 | /* Note that user manager services never need unescaping, | |
1599 | * since they cannot conflict with the kernel's own | |
1600 | * names, hence we don't need to call cg_unescape() | |
1601 | * here. */ | |
1602 | ||
1603 | if (parse_uid(buf, NULL) < 0) | |
1604 | return NULL; | |
1605 | ||
1606 | p += n; | |
1607 | p += strspn(p, "/"); | |
1608 | ||
1609 | return p; | |
1610 | } | |
1611 | ||
1612 | return NULL; | |
9444b1f2 LP |
1613 | } |
1614 | ||
329ac4bc | 1615 | static const char *skip_user_prefix(const char *path) { |
d4fffc4b | 1616 | const char *e, *t; |
ef1673d1 | 1617 | |
6c03089c | 1618 | assert(path); |
ba1261bc | 1619 | |
9444b1f2 LP |
1620 | /* Skip slices, if there are any */ |
1621 | e = skip_slices(path); | |
ba1261bc | 1622 | |
329ac4bc | 1623 | /* Skip the user manager, if it's in the path now... */ |
8b0849e9 | 1624 | t = skip_user_manager(e); |
329ac4bc LP |
1625 | if (t) |
1626 | return t; | |
8b0849e9 | 1627 | |
329ac4bc LP |
1628 | /* Alternatively skip the user session if it is in the path... */ |
1629 | return skip_session(e); | |
1630 | } | |
32081481 | 1631 | |
329ac4bc LP |
1632 | int cg_path_get_user_unit(const char *path, char **ret) { |
1633 | const char *t; | |
6c03089c | 1634 | |
329ac4bc LP |
1635 | assert(path); |
1636 | assert(ret); | |
8b0849e9 | 1637 | |
329ac4bc LP |
1638 | t = skip_user_prefix(path); |
1639 | if (!t) | |
8b0849e9 | 1640 | return -ENXIO; |
8b0849e9 | 1641 | |
329ac4bc LP |
1642 | /* And from here on it looks pretty much the same as for a |
1643 | * system unit, hence let's use the same parser from here | |
1644 | * on. */ | |
1645 | return cg_path_get_unit(t, ret); | |
ef1673d1 | 1646 | } |
ba1261bc | 1647 | |
ef1673d1 | 1648 | int cg_pid_get_user_unit(pid_t pid, char **unit) { |
7fd1b19b | 1649 | _cleanup_free_ char *cgroup = NULL; |
6c03089c LP |
1650 | int r; |
1651 | ||
1652 | assert(unit); | |
1653 | ||
7027ff61 | 1654 | r = cg_pid_get_path_shifted(pid, NULL, &cgroup); |
6c03089c LP |
1655 | if (r < 0) |
1656 | return r; | |
1657 | ||
1658 | return cg_path_get_user_unit(cgroup, unit); | |
ba1261bc | 1659 | } |
e884315e | 1660 | |
7027ff61 | 1661 | int cg_path_get_machine_name(const char *path, char **machine) { |
efdb0237 LP |
1662 | _cleanup_free_ char *u = NULL; |
1663 | const char *sl; | |
89f7c846 | 1664 | int r; |
374ec6ab | 1665 | |
89f7c846 LP |
1666 | r = cg_path_get_unit(path, &u); |
1667 | if (r < 0) | |
1668 | return r; | |
7027ff61 | 1669 | |
efdb0237 | 1670 | sl = strjoina("/run/systemd/machines/unit:", u); |
89f7c846 | 1671 | return readlink_malloc(sl, machine); |
7027ff61 LP |
1672 | } |
1673 | ||
1674 | int cg_pid_get_machine_name(pid_t pid, char **machine) { | |
7fd1b19b | 1675 | _cleanup_free_ char *cgroup = NULL; |
7027ff61 LP |
1676 | int r; |
1677 | ||
1678 | assert(machine); | |
1679 | ||
1680 | r = cg_pid_get_path_shifted(pid, NULL, &cgroup); | |
1681 | if (r < 0) | |
1682 | return r; | |
1683 | ||
1684 | return cg_path_get_machine_name(cgroup, machine); | |
1685 | } | |
1686 | ||
1687 | int cg_path_get_session(const char *path, char **session) { | |
8b0849e9 LP |
1688 | _cleanup_free_ char *unit = NULL; |
1689 | char *start, *end; | |
1690 | int r; | |
7027ff61 LP |
1691 | |
1692 | assert(path); | |
7027ff61 | 1693 | |
8b0849e9 LP |
1694 | r = cg_path_get_unit(path, &unit); |
1695 | if (r < 0) | |
1696 | return r; | |
7027ff61 | 1697 | |
8b0849e9 LP |
1698 | start = startswith(unit, "session-"); |
1699 | if (!start) | |
cfeaa44a | 1700 | return -ENXIO; |
8b0849e9 LP |
1701 | end = endswith(start, ".scope"); |
1702 | if (!end) | |
cfeaa44a | 1703 | return -ENXIO; |
8b0849e9 LP |
1704 | |
1705 | *end = 0; | |
1706 | if (!session_id_valid(start)) | |
cfeaa44a | 1707 | return -ENXIO; |
374ec6ab | 1708 | |
af08d2f9 | 1709 | if (session) { |
8b0849e9 | 1710 | char *rr; |
af08d2f9 | 1711 | |
8b0849e9 LP |
1712 | rr = strdup(start); |
1713 | if (!rr) | |
af08d2f9 LP |
1714 | return -ENOMEM; |
1715 | ||
8b0849e9 | 1716 | *session = rr; |
af08d2f9 | 1717 | } |
7027ff61 | 1718 | |
7027ff61 LP |
1719 | return 0; |
1720 | } | |
1721 | ||
1722 | int cg_pid_get_session(pid_t pid, char **session) { | |
7fd1b19b | 1723 | _cleanup_free_ char *cgroup = NULL; |
7027ff61 LP |
1724 | int r; |
1725 | ||
7027ff61 LP |
1726 | r = cg_pid_get_path_shifted(pid, NULL, &cgroup); |
1727 | if (r < 0) | |
1728 | return r; | |
1729 | ||
1730 | return cg_path_get_session(cgroup, session); | |
1731 | } | |
1732 | ||
ae018d9b | 1733 | int cg_path_get_owner_uid(const char *path, uid_t *uid) { |
374ec6ab | 1734 | _cleanup_free_ char *slice = NULL; |
8b0849e9 | 1735 | char *start, *end; |
374ec6ab | 1736 | int r; |
ae018d9b LP |
1737 | |
1738 | assert(path); | |
ae018d9b | 1739 | |
374ec6ab LP |
1740 | r = cg_path_get_slice(path, &slice); |
1741 | if (r < 0) | |
1742 | return r; | |
ae018d9b | 1743 | |
674eb685 LP |
1744 | start = startswith(slice, "user-"); |
1745 | if (!start) | |
cfeaa44a | 1746 | return -ENXIO; |
8b0849e9 | 1747 | end = endswith(start, ".slice"); |
674eb685 | 1748 | if (!end) |
cfeaa44a | 1749 | return -ENXIO; |
ae018d9b | 1750 | |
8b0849e9 LP |
1751 | *end = 0; |
1752 | if (parse_uid(start, uid) < 0) | |
cfeaa44a | 1753 | return -ENXIO; |
674eb685 | 1754 | |
674eb685 | 1755 | return 0; |
ae018d9b LP |
1756 | } |
1757 | ||
1758 | int cg_pid_get_owner_uid(pid_t pid, uid_t *uid) { | |
1759 | _cleanup_free_ char *cgroup = NULL; | |
1760 | int r; | |
1761 | ||
ae018d9b LP |
1762 | r = cg_pid_get_path_shifted(pid, NULL, &cgroup); |
1763 | if (r < 0) | |
1764 | return r; | |
1765 | ||
1766 | return cg_path_get_owner_uid(cgroup, uid); | |
1767 | } | |
1768 | ||
1021b21b LP |
1769 | int cg_path_get_slice(const char *p, char **slice) { |
1770 | const char *e = NULL; | |
1021b21b LP |
1771 | |
1772 | assert(p); | |
1773 | assert(slice); | |
1774 | ||
329ac4bc LP |
1775 | /* Finds the right-most slice unit from the beginning, but |
1776 | * stops before we come to the first non-slice unit. */ | |
1777 | ||
1021b21b LP |
1778 | for (;;) { |
1779 | size_t n; | |
1780 | ||
1781 | p += strspn(p, "/"); | |
1782 | ||
1783 | n = strcspn(p, "/"); | |
8b0849e9 | 1784 | if (!valid_slice_name(p, n)) { |
1021b21b | 1785 | |
8b0849e9 LP |
1786 | if (!e) { |
1787 | char *s; | |
1021b21b | 1788 | |
e5d855d3 | 1789 | s = strdup(SPECIAL_ROOT_SLICE); |
8b0849e9 LP |
1790 | if (!s) |
1791 | return -ENOMEM; | |
1021b21b | 1792 | |
8b0849e9 LP |
1793 | *slice = s; |
1794 | return 0; | |
1795 | } | |
1796 | ||
1797 | return cg_path_decode_unit(e, slice); | |
1021b21b LP |
1798 | } |
1799 | ||
1800 | e = p; | |
1021b21b LP |
1801 | p += n; |
1802 | } | |
1803 | } | |
1804 | ||
1805 | int cg_pid_get_slice(pid_t pid, char **slice) { | |
1806 | _cleanup_free_ char *cgroup = NULL; | |
1807 | int r; | |
1808 | ||
1809 | assert(slice); | |
1810 | ||
1811 | r = cg_pid_get_path_shifted(pid, NULL, &cgroup); | |
1812 | if (r < 0) | |
1813 | return r; | |
1814 | ||
1815 | return cg_path_get_slice(cgroup, slice); | |
1816 | } | |
1817 | ||
329ac4bc LP |
1818 | int cg_path_get_user_slice(const char *p, char **slice) { |
1819 | const char *t; | |
1820 | assert(p); | |
1821 | assert(slice); | |
1822 | ||
1823 | t = skip_user_prefix(p); | |
1824 | if (!t) | |
1825 | return -ENXIO; | |
1826 | ||
1827 | /* And now it looks pretty much the same as for a system | |
1828 | * slice, so let's just use the same parser from here on. */ | |
1829 | return cg_path_get_slice(t, slice); | |
1830 | } | |
1831 | ||
1832 | int cg_pid_get_user_slice(pid_t pid, char **slice) { | |
1833 | _cleanup_free_ char *cgroup = NULL; | |
1834 | int r; | |
1835 | ||
1836 | assert(slice); | |
1837 | ||
1838 | r = cg_pid_get_path_shifted(pid, NULL, &cgroup); | |
1839 | if (r < 0) | |
1840 | return r; | |
1841 | ||
1842 | return cg_path_get_user_slice(cgroup, slice); | |
1843 | } | |
1844 | ||
ae018d9b LP |
1845 | char *cg_escape(const char *p) { |
1846 | bool need_prefix = false; | |
1847 | ||
1848 | /* This implements very minimal escaping for names to be used | |
1849 | * as file names in the cgroup tree: any name which might | |
1850 | * conflict with a kernel name or is prefixed with '_' is | |
1851 | * prefixed with a '_'. That way, when reading cgroup names it | |
1852 | * is sufficient to remove a single prefixing underscore if | |
1853 | * there is one. */ | |
1854 | ||
1855 | /* The return value of this function (unlike cg_unescape()) | |
1856 | * needs free()! */ | |
1857 | ||
4c701096 | 1858 | if (IN_SET(p[0], 0, '_', '.') || |
0cbd293e | 1859 | STR_IN_SET(p, "notify_on_release", "release_agent", "tasks") || |
efdb0237 | 1860 | startswith(p, "cgroup.")) |
ae018d9b LP |
1861 | need_prefix = true; |
1862 | else { | |
1863 | const char *dot; | |
1864 | ||
1865 | dot = strrchr(p, '.'); | |
1866 | if (dot) { | |
efdb0237 LP |
1867 | CGroupController c; |
1868 | size_t l = dot - p; | |
ae018d9b | 1869 | |
efdb0237 LP |
1870 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { |
1871 | const char *n; | |
1872 | ||
1873 | n = cgroup_controller_to_string(c); | |
ae018d9b | 1874 | |
efdb0237 LP |
1875 | if (l != strlen(n)) |
1876 | continue; | |
ae018d9b | 1877 | |
efdb0237 LP |
1878 | if (memcmp(p, n, l) != 0) |
1879 | continue; | |
1880 | ||
1881 | need_prefix = true; | |
1882 | break; | |
ae018d9b LP |
1883 | } |
1884 | } | |
1885 | } | |
1886 | ||
1887 | if (need_prefix) | |
1888 | return strappend("_", p); | |
efdb0237 LP |
1889 | |
1890 | return strdup(p); | |
ae018d9b LP |
1891 | } |
1892 | ||
1893 | char *cg_unescape(const char *p) { | |
1894 | assert(p); | |
1895 | ||
1896 | /* The return value of this function (unlike cg_escape()) | |
1897 | * doesn't need free()! */ | |
1898 | ||
1899 | if (p[0] == '_') | |
1900 | return (char*) p+1; | |
1901 | ||
1902 | return (char*) p; | |
1903 | } | |
78edb35a LP |
1904 | |
1905 | #define CONTROLLER_VALID \ | |
4b549144 | 1906 | DIGITS LETTERS \ |
78edb35a LP |
1907 | "_" |
1908 | ||
185a0874 | 1909 | bool cg_controller_is_valid(const char *p) { |
78edb35a LP |
1910 | const char *t, *s; |
1911 | ||
1912 | if (!p) | |
1913 | return false; | |
1914 | ||
b6629c4b TH |
1915 | if (streq(p, SYSTEMD_CGROUP_CONTROLLER)) |
1916 | return true; | |
1917 | ||
185a0874 DJL |
1918 | s = startswith(p, "name="); |
1919 | if (s) | |
1920 | p = s; | |
78edb35a | 1921 | |
4c701096 | 1922 | if (IN_SET(*p, 0, '_')) |
78edb35a LP |
1923 | return false; |
1924 | ||
1925 | for (t = p; *t; t++) | |
1926 | if (!strchr(CONTROLLER_VALID, *t)) | |
1927 | return false; | |
1928 | ||
1929 | if (t - p > FILENAME_MAX) | |
1930 | return false; | |
1931 | ||
1932 | return true; | |
1933 | } | |
a016b922 LP |
1934 | |
1935 | int cg_slice_to_path(const char *unit, char **ret) { | |
1936 | _cleanup_free_ char *p = NULL, *s = NULL, *e = NULL; | |
1937 | const char *dash; | |
7410616c | 1938 | int r; |
a016b922 LP |
1939 | |
1940 | assert(unit); | |
1941 | assert(ret); | |
1942 | ||
e5d855d3 | 1943 | if (streq(unit, SPECIAL_ROOT_SLICE)) { |
c96cc582 LP |
1944 | char *x; |
1945 | ||
1946 | x = strdup(""); | |
1947 | if (!x) | |
1948 | return -ENOMEM; | |
1949 | *ret = x; | |
1950 | return 0; | |
1951 | } | |
1952 | ||
7410616c | 1953 | if (!unit_name_is_valid(unit, UNIT_NAME_PLAIN)) |
a016b922 LP |
1954 | return -EINVAL; |
1955 | ||
1956 | if (!endswith(unit, ".slice")) | |
1957 | return -EINVAL; | |
1958 | ||
7410616c LP |
1959 | r = unit_name_to_prefix(unit, &p); |
1960 | if (r < 0) | |
1961 | return r; | |
a016b922 LP |
1962 | |
1963 | dash = strchr(p, '-'); | |
e66e5b61 LP |
1964 | |
1965 | /* Don't allow initial dashes */ | |
1966 | if (dash == p) | |
1967 | return -EINVAL; | |
1968 | ||
a016b922 LP |
1969 | while (dash) { |
1970 | _cleanup_free_ char *escaped = NULL; | |
1971 | char n[dash - p + sizeof(".slice")]; | |
1972 | ||
989290db | 1973 | #if HAS_FEATURE_MEMORY_SANITIZER |
1c56d501 ZJS |
1974 | /* msan doesn't instrument stpncpy, so it thinks |
1975 | * n is later used unitialized: | |
1976 | * https://github.com/google/sanitizers/issues/926 | |
1977 | */ | |
1978 | zero(n); | |
1979 | #endif | |
1980 | ||
e66e5b61 | 1981 | /* Don't allow trailing or double dashes */ |
4c701096 | 1982 | if (IN_SET(dash[1], 0, '-')) |
c96cc582 | 1983 | return -EINVAL; |
a016b922 | 1984 | |
c96cc582 | 1985 | strcpy(stpncpy(n, p, dash - p), ".slice"); |
7410616c | 1986 | if (!unit_name_is_valid(n, UNIT_NAME_PLAIN)) |
a016b922 LP |
1987 | return -EINVAL; |
1988 | ||
1989 | escaped = cg_escape(n); | |
1990 | if (!escaped) | |
1991 | return -ENOMEM; | |
1992 | ||
1993 | if (!strextend(&s, escaped, "/", NULL)) | |
1994 | return -ENOMEM; | |
1995 | ||
1996 | dash = strchr(dash+1, '-'); | |
1997 | } | |
1998 | ||
1999 | e = cg_escape(unit); | |
2000 | if (!e) | |
2001 | return -ENOMEM; | |
2002 | ||
2003 | if (!strextend(&s, e, NULL)) | |
2004 | return -ENOMEM; | |
2005 | ||
ae2a15bc | 2006 | *ret = TAKE_PTR(s); |
a016b922 LP |
2007 | |
2008 | return 0; | |
2009 | } | |
4ad49000 LP |
2010 | |
2011 | int cg_set_attribute(const char *controller, const char *path, const char *attribute, const char *value) { | |
2012 | _cleanup_free_ char *p = NULL; | |
2013 | int r; | |
2014 | ||
2015 | r = cg_get_path(controller, path, attribute, &p); | |
2016 | if (r < 0) | |
2017 | return r; | |
2018 | ||
604028de | 2019 | return write_string_file(p, value, WRITE_STRING_FILE_DISABLE_BUFFER); |
4ad49000 LP |
2020 | } |
2021 | ||
934277fe LP |
2022 | int cg_get_attribute(const char *controller, const char *path, const char *attribute, char **ret) { |
2023 | _cleanup_free_ char *p = NULL; | |
2024 | int r; | |
2025 | ||
2026 | r = cg_get_path(controller, path, attribute, &p); | |
2027 | if (r < 0) | |
2028 | return r; | |
2029 | ||
2030 | return read_one_line_file(p, ret); | |
2031 | } | |
2032 | ||
b734a4ff LP |
2033 | int cg_get_keyed_attribute( |
2034 | const char *controller, | |
2035 | const char *path, | |
2036 | const char *attribute, | |
2037 | char **keys, | |
2038 | char **ret_values) { | |
66ebf6c0 | 2039 | |
b734a4ff | 2040 | _cleanup_free_ char *filename = NULL, *contents = NULL; |
b734a4ff | 2041 | const char *p; |
9177fa9f | 2042 | size_t n, i, n_done = 0; |
b734a4ff LP |
2043 | char **v; |
2044 | int r; | |
2045 | ||
4e1dfa45 | 2046 | /* Reads one or more fields of a cgroup v2 keyed attribute file. The 'keys' parameter should be an strv with |
b734a4ff LP |
2047 | * all keys to retrieve. The 'ret_values' parameter should be passed as string size with the same number of |
2048 | * entries as 'keys'. On success each entry will be set to the value of the matching key. | |
2049 | * | |
2050 | * If the attribute file doesn't exist at all returns ENOENT, if any key is not found returns ENXIO. */ | |
66ebf6c0 TH |
2051 | |
2052 | r = cg_get_path(controller, path, attribute, &filename); | |
2053 | if (r < 0) | |
2054 | return r; | |
2055 | ||
b734a4ff | 2056 | r = read_full_file(filename, &contents, NULL); |
66ebf6c0 TH |
2057 | if (r < 0) |
2058 | return r; | |
2059 | ||
b734a4ff LP |
2060 | n = strv_length(keys); |
2061 | if (n == 0) /* No keys to retrieve? That's easy, we are done then */ | |
2062 | return 0; | |
66ebf6c0 | 2063 | |
b734a4ff LP |
2064 | /* Let's build this up in a temporary array for now in order not to clobber the return parameter on failure */ |
2065 | v = newa0(char*, n); | |
66ebf6c0 | 2066 | |
b734a4ff LP |
2067 | for (p = contents; *p;) { |
2068 | const char *w = NULL; | |
b734a4ff | 2069 | |
9177fa9f ZJS |
2070 | for (i = 0; i < n; i++) |
2071 | if (!v[i]) { | |
b734a4ff LP |
2072 | w = first_word(p, keys[i]); |
2073 | if (w) | |
2074 | break; | |
66ebf6c0 | 2075 | } |
66ebf6c0 | 2076 | |
b734a4ff | 2077 | if (w) { |
b734a4ff LP |
2078 | size_t l; |
2079 | ||
2080 | l = strcspn(w, NEWLINE); | |
9177fa9f ZJS |
2081 | v[i] = strndup(w, l); |
2082 | if (!v[i]) { | |
b734a4ff LP |
2083 | r = -ENOMEM; |
2084 | goto fail; | |
66ebf6c0 | 2085 | } |
b734a4ff | 2086 | |
b734a4ff | 2087 | n_done++; |
b734a4ff LP |
2088 | if (n_done >= n) |
2089 | goto done; | |
2090 | ||
2091 | p = w + l; | |
9177fa9f | 2092 | } else |
b734a4ff | 2093 | p += strcspn(p, NEWLINE); |
b734a4ff LP |
2094 | |
2095 | p += strspn(p, NEWLINE); | |
66ebf6c0 TH |
2096 | } |
2097 | ||
b734a4ff LP |
2098 | r = -ENXIO; |
2099 | ||
2100 | fail: | |
2101 | for (i = 0; i < n; i++) | |
2102 | free(v[i]); | |
2103 | ||
2104 | return r; | |
2105 | ||
2106 | done: | |
2107 | memcpy(ret_values, v, sizeof(char*) * n); | |
66ebf6c0 | 2108 | return 0; |
b734a4ff | 2109 | |
66ebf6c0 TH |
2110 | } |
2111 | ||
efdb0237 LP |
2112 | int cg_create_everywhere(CGroupMask supported, CGroupMask mask, const char *path) { |
2113 | CGroupController c; | |
e353faa0 | 2114 | CGroupMask done; |
65be7e06 | 2115 | bool created; |
415fc41c | 2116 | int r; |
4ad49000 LP |
2117 | |
2118 | /* This one will create a cgroup in our private tree, but also | |
2119 | * duplicate it in the trees specified in mask, and remove it | |
65be7e06 ZJS |
2120 | * in all others. |
2121 | * | |
2122 | * Returns 0 if the group already existed in the systemd hierarchy, | |
2123 | * 1 on success, negative otherwise. | |
2124 | */ | |
4ad49000 LP |
2125 | |
2126 | /* First create the cgroup in our own hierarchy. */ | |
2127 | r = cg_create(SYSTEMD_CGROUP_CONTROLLER, path); | |
2128 | if (r < 0) | |
2129 | return r; | |
490c5a37 | 2130 | created = r; |
4ad49000 | 2131 | |
efdb0237 | 2132 | /* If we are in the unified hierarchy, we are done now */ |
b4cccbc1 LP |
2133 | r = cg_all_unified(); |
2134 | if (r < 0) | |
2135 | return r; | |
2136 | if (r > 0) | |
65be7e06 | 2137 | return created; |
efdb0237 | 2138 | |
e353faa0 LP |
2139 | supported &= CGROUP_MASK_V1; |
2140 | mask = CGROUP_MASK_EXTEND_JOINED(mask); | |
2141 | done = 0; | |
2142 | ||
efdb0237 LP |
2143 | /* Otherwise, do the same in the other hierarchies */ |
2144 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { | |
2145 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
2146 | const char *n; | |
2147 | ||
e353faa0 | 2148 | if (!FLAGS_SET(supported, bit)) |
ab275f23 LP |
2149 | continue; |
2150 | ||
e353faa0 LP |
2151 | if (FLAGS_SET(done, bit)) |
2152 | continue; | |
efdb0237 | 2153 | |
e353faa0 | 2154 | n = cgroup_controller_to_string(c); |
f99850a0 | 2155 | if (FLAGS_SET(mask, bit)) |
efdb0237 | 2156 | (void) cg_create(n, path); |
e353faa0 | 2157 | else |
efdb0237 | 2158 | (void) cg_trim(n, path, true); |
e353faa0 LP |
2159 | |
2160 | done |= CGROUP_MASK_EXTEND_JOINED(bit); | |
4ad49000 LP |
2161 | } |
2162 | ||
65be7e06 | 2163 | return created; |
4ad49000 LP |
2164 | } |
2165 | ||
efdb0237 LP |
2166 | int cg_attach_everywhere(CGroupMask supported, const char *path, pid_t pid, cg_migrate_callback_t path_callback, void *userdata) { |
2167 | CGroupController c; | |
e353faa0 | 2168 | CGroupMask done; |
415fc41c | 2169 | int r; |
4ad49000 LP |
2170 | |
2171 | r = cg_attach(SYSTEMD_CGROUP_CONTROLLER, path, pid); | |
13b84ec7 LP |
2172 | if (r < 0) |
2173 | return r; | |
4ad49000 | 2174 | |
b4cccbc1 LP |
2175 | r = cg_all_unified(); |
2176 | if (r < 0) | |
2177 | return r; | |
2178 | if (r > 0) | |
efdb0237 | 2179 | return 0; |
7b3fd631 | 2180 | |
e353faa0 LP |
2181 | supported &= CGROUP_MASK_V1; |
2182 | done = 0; | |
2183 | ||
efdb0237 LP |
2184 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { |
2185 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
2186 | const char *p = NULL; | |
7b3fd631 | 2187 | |
e353faa0 | 2188 | if (!FLAGS_SET(supported, bit)) |
ab275f23 LP |
2189 | continue; |
2190 | ||
e353faa0 | 2191 | if (FLAGS_SET(done, bit)) |
efdb0237 | 2192 | continue; |
7b3fd631 | 2193 | |
efdb0237 LP |
2194 | if (path_callback) |
2195 | p = path_callback(bit, userdata); | |
efdb0237 LP |
2196 | if (!p) |
2197 | p = path; | |
4ad49000 | 2198 | |
efdb0237 | 2199 | (void) cg_attach_fallback(cgroup_controller_to_string(c), p, pid); |
e353faa0 | 2200 | done |= CGROUP_MASK_EXTEND_JOINED(bit); |
4ad49000 LP |
2201 | } |
2202 | ||
13b84ec7 | 2203 | return 0; |
4ad49000 LP |
2204 | } |
2205 | ||
efdb0237 | 2206 | int cg_attach_many_everywhere(CGroupMask supported, const char *path, Set* pids, cg_migrate_callback_t path_callback, void *userdata) { |
6c12b52e LP |
2207 | Iterator i; |
2208 | void *pidp; | |
2209 | int r = 0; | |
2210 | ||
2211 | SET_FOREACH(pidp, pids, i) { | |
fea72cc0 | 2212 | pid_t pid = PTR_TO_PID(pidp); |
13b84ec7 | 2213 | int q; |
6c12b52e | 2214 | |
7b3fd631 | 2215 | q = cg_attach_everywhere(supported, path, pid, path_callback, userdata); |
efdb0237 | 2216 | if (q < 0 && r >= 0) |
13b84ec7 | 2217 | r = q; |
6c12b52e LP |
2218 | } |
2219 | ||
2220 | return r; | |
2221 | } | |
2222 | ||
efdb0237 | 2223 | int cg_migrate_everywhere(CGroupMask supported, const char *from, const char *to, cg_migrate_callback_t to_callback, void *userdata) { |
b3c5bad3 | 2224 | CGroupController c; |
e353faa0 | 2225 | CGroupMask done; |
b4cccbc1 | 2226 | int r = 0, q; |
4ad49000 | 2227 | |
13b84ec7 | 2228 | if (!path_equal(from, to)) { |
1d98fef1 | 2229 | r = cg_migrate_recursive(SYSTEMD_CGROUP_CONTROLLER, from, SYSTEMD_CGROUP_CONTROLLER, to, CGROUP_REMOVE); |
13b84ec7 LP |
2230 | if (r < 0) |
2231 | return r; | |
2232 | } | |
4ad49000 | 2233 | |
b4cccbc1 LP |
2234 | q = cg_all_unified(); |
2235 | if (q < 0) | |
2236 | return q; | |
2237 | if (q > 0) | |
efdb0237 | 2238 | return r; |
03b90d4b | 2239 | |
e353faa0 LP |
2240 | supported &= CGROUP_MASK_V1; |
2241 | done = 0; | |
2242 | ||
efdb0237 LP |
2243 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { |
2244 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
2245 | const char *p = NULL; | |
03b90d4b | 2246 | |
e353faa0 | 2247 | if (!FLAGS_SET(supported, bit)) |
ab275f23 LP |
2248 | continue; |
2249 | ||
e353faa0 | 2250 | if (FLAGS_SET(done, bit)) |
efdb0237 | 2251 | continue; |
03b90d4b | 2252 | |
efdb0237 LP |
2253 | if (to_callback) |
2254 | p = to_callback(bit, userdata); | |
efdb0237 LP |
2255 | if (!p) |
2256 | p = to; | |
2257 | ||
1d98fef1 | 2258 | (void) cg_migrate_recursive_fallback(SYSTEMD_CGROUP_CONTROLLER, to, cgroup_controller_to_string(c), p, 0); |
e353faa0 | 2259 | done |= CGROUP_MASK_EXTEND_JOINED(bit); |
4ad49000 LP |
2260 | } |
2261 | ||
e353faa0 | 2262 | return r; |
4ad49000 LP |
2263 | } |
2264 | ||
efdb0237 LP |
2265 | int cg_trim_everywhere(CGroupMask supported, const char *path, bool delete_root) { |
2266 | CGroupController c; | |
e353faa0 | 2267 | CGroupMask done; |
b4cccbc1 | 2268 | int r, q; |
4ad49000 LP |
2269 | |
2270 | r = cg_trim(SYSTEMD_CGROUP_CONTROLLER, path, delete_root); | |
2271 | if (r < 0) | |
2272 | return r; | |
2273 | ||
b4cccbc1 LP |
2274 | q = cg_all_unified(); |
2275 | if (q < 0) | |
2276 | return q; | |
2277 | if (q > 0) | |
efdb0237 LP |
2278 | return r; |
2279 | ||
e353faa0 LP |
2280 | supported &= CGROUP_MASK_V1; |
2281 | done = 0; | |
2282 | ||
efdb0237 LP |
2283 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { |
2284 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
2285 | ||
e353faa0 | 2286 | if (!FLAGS_SET(supported, bit)) |
ab275f23 LP |
2287 | continue; |
2288 | ||
e353faa0 | 2289 | if (FLAGS_SET(done, bit)) |
efdb0237 | 2290 | continue; |
4ad49000 | 2291 | |
efdb0237 | 2292 | (void) cg_trim(cgroup_controller_to_string(c), path, delete_root); |
e353faa0 | 2293 | done |= CGROUP_MASK_EXTEND_JOINED(bit); |
4ad49000 LP |
2294 | } |
2295 | ||
e353faa0 | 2296 | return r; |
4ad49000 LP |
2297 | } |
2298 | ||
aae7e17f | 2299 | int cg_mask_to_string(CGroupMask mask, char **ret) { |
ec635a2d LP |
2300 | _cleanup_free_ char *s = NULL; |
2301 | size_t n = 0, allocated = 0; | |
2302 | bool space = false; | |
aae7e17f | 2303 | CGroupController c; |
aae7e17f FB |
2304 | |
2305 | assert(ret); | |
2306 | ||
2307 | if (mask == 0) { | |
2308 | *ret = NULL; | |
2309 | return 0; | |
2310 | } | |
2311 | ||
2312 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { | |
ec635a2d LP |
2313 | const char *k; |
2314 | size_t l; | |
aae7e17f | 2315 | |
f99850a0 | 2316 | if (!FLAGS_SET(mask, CGROUP_CONTROLLER_TO_MASK(c))) |
aae7e17f FB |
2317 | continue; |
2318 | ||
ec635a2d LP |
2319 | k = cgroup_controller_to_string(c); |
2320 | l = strlen(k); | |
2321 | ||
2322 | if (!GREEDY_REALLOC(s, allocated, n + space + l + 1)) | |
2323 | return -ENOMEM; | |
2324 | ||
2325 | if (space) | |
2326 | s[n] = ' '; | |
2327 | memcpy(s + n + space, k, l); | |
2328 | n += space + l; | |
2329 | ||
2330 | space = true; | |
aae7e17f FB |
2331 | } |
2332 | ||
ec635a2d | 2333 | assert(s); |
aae7e17f | 2334 | |
ec635a2d | 2335 | s[n] = 0; |
ae2a15bc | 2336 | *ret = TAKE_PTR(s); |
ec635a2d | 2337 | |
aae7e17f FB |
2338 | return 0; |
2339 | } | |
2340 | ||
38a90d45 LP |
2341 | int cg_mask_from_string(const char *value, CGroupMask *ret) { |
2342 | CGroupMask m = 0; | |
2343 | ||
2344 | assert(ret); | |
aae7e17f FB |
2345 | assert(value); |
2346 | ||
2347 | for (;;) { | |
2348 | _cleanup_free_ char *n = NULL; | |
2349 | CGroupController v; | |
2350 | int r; | |
2351 | ||
2352 | r = extract_first_word(&value, &n, NULL, 0); | |
2353 | if (r < 0) | |
2354 | return r; | |
2355 | if (r == 0) | |
2356 | break; | |
2357 | ||
2358 | v = cgroup_controller_from_string(n); | |
2359 | if (v < 0) | |
2360 | continue; | |
2361 | ||
38a90d45 | 2362 | m |= CGROUP_CONTROLLER_TO_MASK(v); |
aae7e17f | 2363 | } |
38a90d45 LP |
2364 | |
2365 | *ret = m; | |
aae7e17f FB |
2366 | return 0; |
2367 | } | |
2368 | ||
efdb0237 | 2369 | int cg_mask_supported(CGroupMask *ret) { |
38a90d45 | 2370 | CGroupMask mask; |
415fc41c | 2371 | int r; |
efdb0237 | 2372 | |
67558d15 LP |
2373 | /* Determines the mask of supported cgroup controllers. Only includes controllers we can make sense of and that |
2374 | * are actually accessible. Only covers real controllers, i.e. not the CGROUP_CONTROLLER_BPF_xyz | |
2375 | * pseudo-controllers. */ | |
4ad49000 | 2376 | |
b4cccbc1 LP |
2377 | r = cg_all_unified(); |
2378 | if (r < 0) | |
2379 | return r; | |
2380 | if (r > 0) { | |
5f4c5fef | 2381 | _cleanup_free_ char *root = NULL, *controllers = NULL, *path = NULL; |
efdb0237 LP |
2382 | |
2383 | /* In the unified hierarchy we can read the supported | |
2384 | * and accessible controllers from a the top-level | |
2385 | * cgroup attribute */ | |
2386 | ||
5f4c5fef LP |
2387 | r = cg_get_root_path(&root); |
2388 | if (r < 0) | |
2389 | return r; | |
2390 | ||
2391 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, root, "cgroup.controllers", &path); | |
2392 | if (r < 0) | |
2393 | return r; | |
2394 | ||
2395 | r = read_one_line_file(path, &controllers); | |
efdb0237 LP |
2396 | if (r < 0) |
2397 | return r; | |
4ad49000 | 2398 | |
aae7e17f FB |
2399 | r = cg_mask_from_string(controllers, &mask); |
2400 | if (r < 0) | |
2401 | return r; | |
efdb0237 | 2402 | |
03afd780 | 2403 | /* Currently, we support the cpu, memory, io and pids controller in the unified hierarchy, mask |
03a7b521 | 2404 | * everything else off. */ |
03afd780 | 2405 | mask &= CGROUP_MASK_V2; |
efdb0237 LP |
2406 | |
2407 | } else { | |
2408 | CGroupController c; | |
2409 | ||
03afd780 | 2410 | /* In the legacy hierarchy, we check which hierarchies are mounted. */ |
efdb0237 | 2411 | |
38a90d45 | 2412 | mask = 0; |
efdb0237 | 2413 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { |
03afd780 | 2414 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); |
efdb0237 LP |
2415 | const char *n; |
2416 | ||
03afd780 LP |
2417 | if (!FLAGS_SET(CGROUP_MASK_V1, bit)) |
2418 | continue; | |
2419 | ||
efdb0237 LP |
2420 | n = cgroup_controller_to_string(c); |
2421 | if (controller_is_accessible(n) >= 0) | |
03afd780 | 2422 | mask |= bit; |
efdb0237 | 2423 | } |
4ad49000 LP |
2424 | } |
2425 | ||
efdb0237 LP |
2426 | *ret = mask; |
2427 | return 0; | |
4ad49000 | 2428 | } |
b12afc8c | 2429 | |
6925a0de LP |
2430 | int cg_kernel_controllers(Set **ret) { |
2431 | _cleanup_set_free_free_ Set *controllers = NULL; | |
b12afc8c | 2432 | _cleanup_fclose_ FILE *f = NULL; |
b12afc8c LP |
2433 | int r; |
2434 | ||
6925a0de | 2435 | assert(ret); |
b12afc8c | 2436 | |
f09e86bc LS |
2437 | /* Determines the full list of kernel-known controllers. Might include controllers we don't actually support |
2438 | * and controllers that aren't currently accessible (because not mounted). This does not include "name=" | |
2439 | * pseudo-controllers. */ | |
e155a0aa | 2440 | |
6925a0de LP |
2441 | controllers = set_new(&string_hash_ops); |
2442 | if (!controllers) | |
2443 | return -ENOMEM; | |
2444 | ||
b12afc8c LP |
2445 | f = fopen("/proc/cgroups", "re"); |
2446 | if (!f) { | |
6925a0de LP |
2447 | if (errno == ENOENT) { |
2448 | *ret = NULL; | |
b12afc8c | 2449 | return 0; |
6925a0de LP |
2450 | } |
2451 | ||
b12afc8c LP |
2452 | return -errno; |
2453 | } | |
2454 | ||
35bbbf85 LP |
2455 | (void) __fsetlocking(f, FSETLOCKING_BYCALLER); |
2456 | ||
b12afc8c | 2457 | /* Ignore the header line */ |
2351e44d | 2458 | (void) read_line(f, (size_t) -1, NULL); |
b12afc8c LP |
2459 | |
2460 | for (;;) { | |
2461 | char *controller; | |
2462 | int enabled = 0; | |
2463 | ||
2464 | errno = 0; | |
2465 | if (fscanf(f, "%ms %*i %*i %i", &controller, &enabled) != 2) { | |
2466 | ||
2467 | if (feof(f)) | |
2468 | break; | |
2469 | ||
b3267152 | 2470 | if (ferror(f) && errno > 0) |
b12afc8c LP |
2471 | return -errno; |
2472 | ||
2473 | return -EBADMSG; | |
2474 | } | |
2475 | ||
2476 | if (!enabled) { | |
2477 | free(controller); | |
2478 | continue; | |
2479 | } | |
2480 | ||
efdb0237 | 2481 | if (!cg_controller_is_valid(controller)) { |
b12afc8c LP |
2482 | free(controller); |
2483 | return -EBADMSG; | |
2484 | } | |
2485 | ||
2486 | r = set_consume(controllers, controller); | |
2487 | if (r < 0) | |
2488 | return r; | |
2489 | } | |
2490 | ||
1cc6c93a | 2491 | *ret = TAKE_PTR(controllers); |
6925a0de | 2492 | |
b12afc8c LP |
2493 | return 0; |
2494 | } | |
efdb0237 | 2495 | |
5da38d07 TH |
2496 | static thread_local CGroupUnified unified_cache = CGROUP_UNIFIED_UNKNOWN; |
2497 | ||
4e1dfa45 | 2498 | /* The hybrid mode was initially implemented in v232 and simply mounted cgroup2 on /sys/fs/cgroup/systemd. This |
c22800e4 LP |
2499 | * unfortunately broke other tools (such as docker) which expected the v1 "name=systemd" hierarchy on |
2500 | * /sys/fs/cgroup/systemd. From v233 and on, the hybrid mode mountnbs v2 on /sys/fs/cgroup/unified and maintains | |
2501 | * "name=systemd" hierarchy on /sys/fs/cgroup/systemd for compatibility with other tools. | |
f08e9287 | 2502 | * |
c22800e4 LP |
2503 | * To keep live upgrade working, we detect and support v232 layout. When v232 layout is detected, to keep cgroup v2 |
2504 | * process management but disable the compat dual layout, we return %true on | |
2505 | * cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) and %false on cg_hybrid_unified(). | |
f08e9287 TH |
2506 | */ |
2507 | static thread_local bool unified_systemd_v232; | |
2508 | ||
1fcca10e | 2509 | static int cg_unified_update(void) { |
efdb0237 | 2510 | |
efdb0237 LP |
2511 | struct statfs fs; |
2512 | ||
2513 | /* Checks if we support the unified hierarchy. Returns an | |
2514 | * error when the cgroup hierarchies aren't mounted yet or we | |
2515 | * have any other trouble determining if the unified hierarchy | |
2516 | * is supported. */ | |
2517 | ||
5da38d07 TH |
2518 | if (unified_cache >= CGROUP_UNIFIED_NONE) |
2519 | return 0; | |
efdb0237 LP |
2520 | |
2521 | if (statfs("/sys/fs/cgroup/", &fs) < 0) | |
c028bed1 | 2522 | return log_debug_errno(errno, "statfs(\"/sys/fs/cgroup/\") failed: %m"); |
efdb0237 | 2523 | |
9aa21133 ZJS |
2524 | if (F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) { |
2525 | log_debug("Found cgroup2 on /sys/fs/cgroup/, full unified hierarchy"); | |
5da38d07 | 2526 | unified_cache = CGROUP_UNIFIED_ALL; |
9aa21133 | 2527 | } else if (F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC)) { |
2977724b | 2528 | if (statfs("/sys/fs/cgroup/unified/", &fs) == 0 && |
f08e9287 | 2529 | F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) { |
9aa21133 | 2530 | log_debug("Found cgroup2 on /sys/fs/cgroup/unified, unified hierarchy for systemd controller"); |
2977724b | 2531 | unified_cache = CGROUP_UNIFIED_SYSTEMD; |
f08e9287 | 2532 | unified_systemd_v232 = false; |
f08e9287 | 2533 | } else { |
2977724b | 2534 | if (statfs("/sys/fs/cgroup/systemd/", &fs) < 0) |
9aa21133 | 2535 | return log_debug_errno(errno, "statfs(\"/sys/fs/cgroup/systemd\" failed: %m"); |
5535d8f7 EV |
2536 | |
2537 | if (F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) { | |
2538 | log_debug("Found cgroup2 on /sys/fs/cgroup/systemd, unified hierarchy for systemd controller (v232 variant)"); | |
2539 | unified_cache = CGROUP_UNIFIED_SYSTEMD; | |
2540 | unified_systemd_v232 = true; | |
2541 | } else if (F_TYPE_EQUAL(fs.f_type, CGROUP_SUPER_MAGIC)) { | |
2542 | log_debug("Found cgroup on /sys/fs/cgroup/systemd, legacy hierarchy"); | |
2543 | unified_cache = CGROUP_UNIFIED_NONE; | |
2544 | } else { | |
2545 | log_debug("Unexpected filesystem type %llx mounted on /sys/fs/cgroup/systemd, assuming legacy hierarchy", | |
9aa21133 | 2546 | (unsigned long long) fs.f_type); |
5535d8f7 | 2547 | unified_cache = CGROUP_UNIFIED_NONE; |
9aa21133 | 2548 | } |
2977724b | 2549 | } |
baaa35ad ZJS |
2550 | } else |
2551 | return log_debug_errno(SYNTHETIC_ERRNO(ENOMEDIUM), | |
2552 | "Unknown filesystem type %llx mounted on /sys/fs/cgroup.", | |
2553 | (unsigned long long)fs.f_type); | |
efdb0237 | 2554 | |
5da38d07 TH |
2555 | return 0; |
2556 | } | |
2557 | ||
c22800e4 | 2558 | int cg_unified_controller(const char *controller) { |
b4cccbc1 | 2559 | int r; |
5da38d07 | 2560 | |
1fcca10e | 2561 | r = cg_unified_update(); |
b4cccbc1 LP |
2562 | if (r < 0) |
2563 | return r; | |
5da38d07 | 2564 | |
fc9ae717 LP |
2565 | if (unified_cache == CGROUP_UNIFIED_NONE) |
2566 | return false; | |
2567 | ||
2568 | if (unified_cache >= CGROUP_UNIFIED_ALL) | |
2569 | return true; | |
2570 | ||
2571 | return streq_ptr(controller, SYSTEMD_CGROUP_CONTROLLER); | |
5da38d07 TH |
2572 | } |
2573 | ||
b4cccbc1 | 2574 | int cg_all_unified(void) { |
4bb652ac LP |
2575 | int r; |
2576 | ||
2577 | r = cg_unified_update(); | |
2578 | if (r < 0) | |
2579 | return r; | |
2580 | ||
2581 | return unified_cache >= CGROUP_UNIFIED_ALL; | |
efdb0237 LP |
2582 | } |
2583 | ||
b4cccbc1 LP |
2584 | int cg_hybrid_unified(void) { |
2585 | int r; | |
2977724b | 2586 | |
1fcca10e | 2587 | r = cg_unified_update(); |
b4cccbc1 LP |
2588 | if (r < 0) |
2589 | return r; | |
2977724b | 2590 | |
f08e9287 | 2591 | return unified_cache == CGROUP_UNIFIED_SYSTEMD && !unified_systemd_v232; |
2977724b TH |
2592 | } |
2593 | ||
415fc41c | 2594 | int cg_unified_flush(void) { |
5da38d07 | 2595 | unified_cache = CGROUP_UNIFIED_UNKNOWN; |
415fc41c | 2596 | |
1fcca10e | 2597 | return cg_unified_update(); |
efdb0237 LP |
2598 | } |
2599 | ||
27adcc97 LP |
2600 | int cg_enable_everywhere( |
2601 | CGroupMask supported, | |
2602 | CGroupMask mask, | |
2603 | const char *p, | |
2604 | CGroupMask *ret_result_mask) { | |
2605 | ||
77fa610b | 2606 | _cleanup_fclose_ FILE *f = NULL; |
efdb0237 LP |
2607 | _cleanup_free_ char *fs = NULL; |
2608 | CGroupController c; | |
27adcc97 | 2609 | CGroupMask ret = 0; |
415fc41c | 2610 | int r; |
efdb0237 LP |
2611 | |
2612 | assert(p); | |
2613 | ||
27adcc97 LP |
2614 | if (supported == 0) { |
2615 | if (ret_result_mask) | |
2616 | *ret_result_mask = 0; | |
efdb0237 | 2617 | return 0; |
27adcc97 | 2618 | } |
efdb0237 | 2619 | |
b4cccbc1 LP |
2620 | r = cg_all_unified(); |
2621 | if (r < 0) | |
2622 | return r; | |
27adcc97 LP |
2623 | if (r == 0) { |
2624 | /* On the legacy hiearchy there's no concept of "enabling" controllers in cgroups defined. Let's claim | |
2625 | * complete success right away. (If you wonder why we return the full mask here, rather than zero: the | |
2626 | * caller tends to use the returned mask later on to compare if all controllers where properly joined, | |
2627 | * and if not requeues realization. This use is the primary purpose of the return value, hence let's | |
2628 | * minimize surprises here and reduce triggers for re-realization by always saying we fully | |
2629 | * succeeded.) */ | |
2630 | if (ret_result_mask) | |
2631 | *ret_result_mask = mask & supported & CGROUP_MASK_V2; /* If you wonder why we mask this with | |
2632 | * CGROUP_MASK_V2: The 'supported' mask | |
2633 | * might contain pure-V1 or BPF | |
2634 | * controllers, and we never want to | |
2635 | * claim that we could enable those with | |
2636 | * cgroup.subtree_control */ | |
efdb0237 | 2637 | return 0; |
27adcc97 | 2638 | } |
efdb0237 LP |
2639 | |
2640 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, p, "cgroup.subtree_control", &fs); | |
2641 | if (r < 0) | |
2642 | return r; | |
2643 | ||
2644 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { | |
2645 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
2646 | const char *n; | |
2647 | ||
ab275f23 LP |
2648 | if (!FLAGS_SET(CGROUP_MASK_V2, bit)) |
2649 | continue; | |
2650 | ||
f99850a0 | 2651 | if (!FLAGS_SET(supported, bit)) |
efdb0237 LP |
2652 | continue; |
2653 | ||
2654 | n = cgroup_controller_to_string(c); | |
2655 | { | |
2656 | char s[1 + strlen(n) + 1]; | |
2657 | ||
f99850a0 | 2658 | s[0] = FLAGS_SET(mask, bit) ? '+' : '-'; |
efdb0237 LP |
2659 | strcpy(s + 1, n); |
2660 | ||
77fa610b LP |
2661 | if (!f) { |
2662 | f = fopen(fs, "we"); | |
54b5ba1d LP |
2663 | if (!f) |
2664 | return log_debug_errno(errno, "Failed to open cgroup.subtree_control file of %s: %m", p); | |
77fa610b LP |
2665 | } |
2666 | ||
604028de | 2667 | r = write_string_stream(f, s, WRITE_STRING_FILE_DISABLE_BUFFER); |
96aa6591 | 2668 | if (r < 0) { |
94f344fb LP |
2669 | log_debug_errno(r, "Failed to %s controller %s for %s (%s): %m", |
2670 | FLAGS_SET(mask, bit) ? "enable" : "disable", n, p, fs); | |
96aa6591 | 2671 | clearerr(f); |
27adcc97 LP |
2672 | |
2673 | /* If we can't turn off a controller, leave it on in the reported resulting mask. This | |
2674 | * happens for example when we attempt to turn off a controller up in the tree that is | |
2675 | * used down in the tree. */ | |
2676 | if (!FLAGS_SET(mask, bit) && r == -EBUSY) /* You might wonder why we check for EBUSY | |
2677 | * only here, and not follow the same logic | |
2678 | * for other errors such as EINVAL or | |
2679 | * EOPNOTSUPP or anything else. That's | |
2680 | * because EBUSY indicates that the | |
2681 | * controllers is currently enabled and | |
2682 | * cannot be disabled because something down | |
2683 | * the hierarchy is still using it. Any other | |
2684 | * error most likely means something like "I | |
2685 | * never heard of this controller" or | |
2686 | * similar. In the former case it's hence | |
2687 | * safe to assume the controller is still on | |
2688 | * after the failed operation, while in the | |
2689 | * latter case it's safer to assume the | |
2690 | * controller is unknown and hence certainly | |
2691 | * not enabled. */ | |
2692 | ret |= bit; | |
2693 | } else { | |
2694 | /* Otherwise, if we managed to turn on a controller, set the bit reflecting that. */ | |
2695 | if (FLAGS_SET(mask, bit)) | |
2696 | ret |= bit; | |
96aa6591 | 2697 | } |
efdb0237 LP |
2698 | } |
2699 | } | |
2700 | ||
27adcc97 LP |
2701 | /* Let's return the precise set of controllers now enabled for the cgroup. */ |
2702 | if (ret_result_mask) | |
2703 | *ret_result_mask = ret; | |
2704 | ||
efdb0237 LP |
2705 | return 0; |
2706 | } | |
2707 | ||
2708 | bool cg_is_unified_wanted(void) { | |
2709 | static thread_local int wanted = -1; | |
415fc41c | 2710 | int r; |
1d84ad94 | 2711 | bool b; |
77fab2a9 | 2712 | const bool is_default = DEFAULT_HIERARCHY == CGROUP_UNIFIED_ALL; |
5f086dc7 | 2713 | _cleanup_free_ char *c = NULL; |
efdb0237 | 2714 | |
77fab2a9 | 2715 | /* If we have a cached value, return that. */ |
efdb0237 LP |
2716 | if (wanted >= 0) |
2717 | return wanted; | |
2718 | ||
239a3d09 ZJS |
2719 | /* If the hierarchy is already mounted, then follow whatever |
2720 | * was chosen for it. */ | |
2721 | if (cg_unified_flush() >= 0) | |
b4cccbc1 | 2722 | return (wanted = unified_cache >= CGROUP_UNIFIED_ALL); |
239a3d09 | 2723 | |
5f086dc7 CD |
2724 | /* If we were explicitly passed systemd.unified_cgroup_hierarchy, |
2725 | * respect that. */ | |
1d84ad94 | 2726 | r = proc_cmdline_get_bool("systemd.unified_cgroup_hierarchy", &b); |
5f086dc7 CD |
2727 | if (r > 0) |
2728 | return (wanted = b); | |
2729 | ||
2730 | /* If we passed cgroup_no_v1=all with no other instructions, it seems | |
2731 | * highly unlikely that we want to use hybrid or legacy hierarchy. */ | |
2732 | r = proc_cmdline_get_key("cgroup_no_v1", 0, &c); | |
2733 | if (r > 0 && streq_ptr(c, "all")) | |
2734 | return (wanted = true); | |
efdb0237 | 2735 | |
5f086dc7 | 2736 | return (wanted = is_default); |
efdb0237 LP |
2737 | } |
2738 | ||
2739 | bool cg_is_legacy_wanted(void) { | |
239a3d09 ZJS |
2740 | static thread_local int wanted = -1; |
2741 | ||
2742 | /* If we have a cached value, return that. */ | |
2743 | if (wanted >= 0) | |
2744 | return wanted; | |
2745 | ||
4e1dfa45 | 2746 | /* Check if we have cgroup v2 already mounted. */ |
1b59cf04 ZJS |
2747 | if (cg_unified_flush() >= 0 && |
2748 | unified_cache == CGROUP_UNIFIED_ALL) | |
239a3d09 | 2749 | return (wanted = false); |
1b59cf04 ZJS |
2750 | |
2751 | /* Otherwise, assume that at least partial legacy is wanted, | |
4e1dfa45 | 2752 | * since cgroup v2 should already be mounted at this point. */ |
239a3d09 | 2753 | return (wanted = true); |
efdb0237 LP |
2754 | } |
2755 | ||
a4464b95 | 2756 | bool cg_is_hybrid_wanted(void) { |
5da38d07 | 2757 | static thread_local int wanted = -1; |
415fc41c | 2758 | int r; |
1d84ad94 | 2759 | bool b; |
c19739db ZJS |
2760 | const bool is_default = DEFAULT_HIERARCHY >= CGROUP_UNIFIED_SYSTEMD; |
2761 | /* We default to true if the default is "hybrid", obviously, | |
2762 | * but also when the default is "unified", because if we get | |
2763 | * called, it means that unified hierarchy was not mounted. */ | |
5da38d07 | 2764 | |
77fab2a9 | 2765 | /* If we have a cached value, return that. */ |
5da38d07 TH |
2766 | if (wanted >= 0) |
2767 | return wanted; | |
2768 | ||
239a3d09 ZJS |
2769 | /* If the hierarchy is already mounted, then follow whatever |
2770 | * was chosen for it. */ | |
2771 | if (cg_unified_flush() >= 0 && | |
2772 | unified_cache == CGROUP_UNIFIED_ALL) | |
2773 | return (wanted = false); | |
2774 | ||
77fab2a9 ZJS |
2775 | /* Otherwise, let's see what the kernel command line has to say. |
2776 | * Since checking is expensive, cache a non-error result. */ | |
1d84ad94 | 2777 | r = proc_cmdline_get_bool("systemd.legacy_systemd_cgroup_controller", &b); |
5da38d07 | 2778 | |
2dcb526d ZJS |
2779 | /* The meaning of the kernel option is reversed wrt. to the return value |
2780 | * of this function, hence the negation. */ | |
77fab2a9 | 2781 | return (wanted = r > 0 ? !b : is_default); |
5da38d07 TH |
2782 | } |
2783 | ||
13c31542 TH |
2784 | int cg_weight_parse(const char *s, uint64_t *ret) { |
2785 | uint64_t u; | |
2786 | int r; | |
2787 | ||
2788 | if (isempty(s)) { | |
2789 | *ret = CGROUP_WEIGHT_INVALID; | |
2790 | return 0; | |
2791 | } | |
2792 | ||
2793 | r = safe_atou64(s, &u); | |
2794 | if (r < 0) | |
2795 | return r; | |
2796 | ||
2797 | if (u < CGROUP_WEIGHT_MIN || u > CGROUP_WEIGHT_MAX) | |
2798 | return -ERANGE; | |
2799 | ||
2800 | *ret = u; | |
2801 | return 0; | |
2802 | } | |
2803 | ||
9be57249 TH |
2804 | const uint64_t cgroup_io_limit_defaults[_CGROUP_IO_LIMIT_TYPE_MAX] = { |
2805 | [CGROUP_IO_RBPS_MAX] = CGROUP_LIMIT_MAX, | |
2806 | [CGROUP_IO_WBPS_MAX] = CGROUP_LIMIT_MAX, | |
ac06a0cf TH |
2807 | [CGROUP_IO_RIOPS_MAX] = CGROUP_LIMIT_MAX, |
2808 | [CGROUP_IO_WIOPS_MAX] = CGROUP_LIMIT_MAX, | |
9be57249 TH |
2809 | }; |
2810 | ||
2811 | static const char* const cgroup_io_limit_type_table[_CGROUP_IO_LIMIT_TYPE_MAX] = { | |
2812 | [CGROUP_IO_RBPS_MAX] = "IOReadBandwidthMax", | |
2813 | [CGROUP_IO_WBPS_MAX] = "IOWriteBandwidthMax", | |
ac06a0cf TH |
2814 | [CGROUP_IO_RIOPS_MAX] = "IOReadIOPSMax", |
2815 | [CGROUP_IO_WIOPS_MAX] = "IOWriteIOPSMax", | |
9be57249 TH |
2816 | }; |
2817 | ||
2818 | DEFINE_STRING_TABLE_LOOKUP(cgroup_io_limit_type, CGroupIOLimitType); | |
2819 | ||
d53d9474 LP |
2820 | int cg_cpu_shares_parse(const char *s, uint64_t *ret) { |
2821 | uint64_t u; | |
2822 | int r; | |
2823 | ||
2824 | if (isempty(s)) { | |
2825 | *ret = CGROUP_CPU_SHARES_INVALID; | |
2826 | return 0; | |
2827 | } | |
2828 | ||
2829 | r = safe_atou64(s, &u); | |
2830 | if (r < 0) | |
2831 | return r; | |
2832 | ||
2833 | if (u < CGROUP_CPU_SHARES_MIN || u > CGROUP_CPU_SHARES_MAX) | |
2834 | return -ERANGE; | |
2835 | ||
2836 | *ret = u; | |
2837 | return 0; | |
2838 | } | |
2839 | ||
2840 | int cg_blkio_weight_parse(const char *s, uint64_t *ret) { | |
2841 | uint64_t u; | |
2842 | int r; | |
2843 | ||
2844 | if (isempty(s)) { | |
2845 | *ret = CGROUP_BLKIO_WEIGHT_INVALID; | |
2846 | return 0; | |
2847 | } | |
2848 | ||
2849 | r = safe_atou64(s, &u); | |
2850 | if (r < 0) | |
2851 | return r; | |
2852 | ||
2853 | if (u < CGROUP_BLKIO_WEIGHT_MIN || u > CGROUP_BLKIO_WEIGHT_MAX) | |
2854 | return -ERANGE; | |
2855 | ||
2856 | *ret = u; | |
2857 | return 0; | |
2858 | } | |
2859 | ||
f0bef277 EV |
2860 | bool is_cgroup_fs(const struct statfs *s) { |
2861 | return is_fs_type(s, CGROUP_SUPER_MAGIC) || | |
2862 | is_fs_type(s, CGROUP2_SUPER_MAGIC); | |
2863 | } | |
2864 | ||
2865 | bool fd_is_cgroup_fs(int fd) { | |
2866 | struct statfs s; | |
2867 | ||
2868 | if (fstatfs(fd, &s) < 0) | |
2869 | return -errno; | |
2870 | ||
2871 | return is_cgroup_fs(&s); | |
2872 | } | |
2873 | ||
efdb0237 LP |
2874 | static const char *cgroup_controller_table[_CGROUP_CONTROLLER_MAX] = { |
2875 | [CGROUP_CONTROLLER_CPU] = "cpu", | |
2876 | [CGROUP_CONTROLLER_CPUACCT] = "cpuacct", | |
13c31542 | 2877 | [CGROUP_CONTROLLER_IO] = "io", |
efdb0237 LP |
2878 | [CGROUP_CONTROLLER_BLKIO] = "blkio", |
2879 | [CGROUP_CONTROLLER_MEMORY] = "memory", | |
3905f127 | 2880 | [CGROUP_CONTROLLER_DEVICES] = "devices", |
03a7b521 | 2881 | [CGROUP_CONTROLLER_PIDS] = "pids", |
17f14955 | 2882 | [CGROUP_CONTROLLER_BPF_FIREWALL] = "bpf-firewall", |
084c7007 | 2883 | [CGROUP_CONTROLLER_BPF_DEVICES] = "bpf-devices", |
efdb0237 LP |
2884 | }; |
2885 | ||
2886 | DEFINE_STRING_TABLE_LOOKUP(cgroup_controller, CGroupController); | |
f98c2585 CD |
2887 | |
2888 | CGroupMask get_cpu_accounting_mask(void) { | |
2889 | static CGroupMask needed_mask = (CGroupMask) -1; | |
2890 | ||
2891 | /* On kernel ≥4.15 with unified hierarchy, cpu.stat's usage_usec is | |
2892 | * provided externally from the CPU controller, which means we don't | |
2893 | * need to enable the CPU controller just to get metrics. This is good, | |
2894 | * because enabling the CPU controller comes at a minor performance | |
2895 | * hit, especially when it's propagated deep into large hierarchies. | |
2896 | * There's also no separate CPU accounting controller available within | |
2897 | * a unified hierarchy. | |
2898 | * | |
2899 | * This combination of factors results in the desired cgroup mask to | |
2900 | * enable for CPU accounting varying as follows: | |
2901 | * | |
2902 | * ╔═════════════════════╤═════════════════════╗ | |
2903 | * ║ Linux ≥4.15 │ Linux <4.15 ║ | |
2904 | * ╔═══════════════╬═════════════════════╪═════════════════════╣ | |
2905 | * ║ Unified ║ nothing │ CGROUP_MASK_CPU ║ | |
2906 | * ╟───────────────╫─────────────────────┼─────────────────────╢ | |
2907 | * ║ Hybrid/Legacy ║ CGROUP_MASK_CPUACCT │ CGROUP_MASK_CPUACCT ║ | |
2908 | * ╚═══════════════╩═════════════════════╧═════════════════════╝ | |
2909 | * | |
2910 | * We check kernel version here instead of manually checking whether | |
2911 | * cpu.stat is present for every cgroup, as that check in itself would | |
2912 | * already be fairly expensive. | |
2913 | * | |
2914 | * Kernels where this patch has been backported will therefore have the | |
2915 | * CPU controller enabled unnecessarily. This is more expensive than | |
2916 | * necessary, but harmless. ☺️ | |
2917 | */ | |
2918 | ||
2919 | if (needed_mask == (CGroupMask) -1) { | |
2920 | if (cg_all_unified()) { | |
2921 | struct utsname u; | |
2922 | assert_se(uname(&u) >= 0); | |
2923 | ||
2924 | if (str_verscmp(u.release, "4.15") < 0) | |
2925 | needed_mask = CGROUP_MASK_CPU; | |
2926 | else | |
2927 | needed_mask = 0; | |
2928 | } else | |
2929 | needed_mask = CGROUP_MASK_CPUACCT; | |
2930 | } | |
2931 | ||
2932 | return needed_mask; | |
2933 | } | |
2934 | ||
2935 | bool cpu_accounting_is_cheap(void) { | |
2936 | return get_cpu_accounting_mask() == 0; | |
2937 | } |