]>
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> |
8c6db833 | 9 | #include <stdlib.h> |
84ac7bea | 10 | #include <string.h> |
672c48cc | 11 | #include <sys/stat.h> |
11c3a366 | 12 | #include <sys/statfs.h> |
672c48cc | 13 | #include <sys/types.h> |
f98c2585 | 14 | #include <sys/utsname.h> |
4b58153d | 15 | #include <sys/xattr.h> |
84ac7bea | 16 | #include <unistd.h> |
8c6db833 | 17 | |
b5efdb8a | 18 | #include "alloc-util.h" |
3ffd4af2 | 19 | #include "cgroup-util.h" |
93cc7779 | 20 | #include "def.h" |
a0956174 | 21 | #include "dirent-util.h" |
84ac7bea | 22 | #include "extract-word.h" |
3ffd4af2 | 23 | #include "fd-util.h" |
84ac7bea | 24 | #include "fileio.h" |
f97b34a6 | 25 | #include "format-util.h" |
f4f15635 | 26 | #include "fs-util.h" |
93cc7779 | 27 | #include "log.h" |
84ac7bea LP |
28 | #include "login-util.h" |
29 | #include "macro.h" | |
93cc7779 | 30 | #include "missing.h" |
84ac7bea | 31 | #include "mkdir.h" |
6bedfcbb | 32 | #include "parse-util.h" |
9eb977db | 33 | #include "path-util.h" |
872a590e | 34 | #include "proc-cmdline.h" |
84ac7bea LP |
35 | #include "process-util.h" |
36 | #include "set.h" | |
9444b1f2 | 37 | #include "special.h" |
872a590e | 38 | #include "stat-util.h" |
d054f0a4 | 39 | #include "stdio-util.h" |
8b43440b | 40 | #include "string-table.h" |
07630cea | 41 | #include "string-util.h" |
aae7e17f | 42 | #include "strv.h" |
84ac7bea | 43 | #include "unit-name.h" |
b1d4f8e1 | 44 | #include "user-util.h" |
8c6db833 | 45 | |
c6c18be3 | 46 | int cg_enumerate_processes(const char *controller, const char *path, FILE **_f) { |
7027ff61 | 47 | _cleanup_free_ char *fs = NULL; |
c6c18be3 | 48 | FILE *f; |
7027ff61 | 49 | int r; |
c6c18be3 | 50 | |
c6c18be3 LP |
51 | assert(_f); |
52 | ||
c3175a7f LP |
53 | r = cg_get_path(controller, path, "cgroup.procs", &fs); |
54 | if (r < 0) | |
c6c18be3 LP |
55 | return r; |
56 | ||
57 | f = fopen(fs, "re"); | |
c6c18be3 LP |
58 | if (!f) |
59 | return -errno; | |
60 | ||
61 | *_f = f; | |
62 | return 0; | |
63 | } | |
64 | ||
c6c18be3 LP |
65 | int cg_read_pid(FILE *f, pid_t *_pid) { |
66 | unsigned long ul; | |
67 | ||
68 | /* Note that the cgroup.procs might contain duplicates! See | |
69 | * cgroups.txt for details. */ | |
70 | ||
7027ff61 LP |
71 | assert(f); |
72 | assert(_pid); | |
73 | ||
c6c18be3 LP |
74 | errno = 0; |
75 | if (fscanf(f, "%lu", &ul) != 1) { | |
76 | ||
77 | if (feof(f)) | |
78 | return 0; | |
79 | ||
f5e5c28f | 80 | return errno > 0 ? -errno : -EIO; |
c6c18be3 LP |
81 | } |
82 | ||
83 | if (ul <= 0) | |
84 | return -EIO; | |
85 | ||
86 | *_pid = (pid_t) ul; | |
87 | return 1; | |
88 | } | |
89 | ||
8b238b13 LP |
90 | int cg_read_event( |
91 | const char *controller, | |
92 | const char *path, | |
93 | const char *event, | |
31a9be23 | 94 | char **ret) { |
8b238b13 | 95 | |
ab2c3861 | 96 | _cleanup_free_ char *events = NULL, *content = NULL; |
ab2c3861 TH |
97 | int r; |
98 | ||
99 | r = cg_get_path(controller, path, "cgroup.events", &events); | |
100 | if (r < 0) | |
101 | return r; | |
102 | ||
103 | r = read_full_file(events, &content, NULL); | |
104 | if (r < 0) | |
105 | return r; | |
106 | ||
31a9be23 YW |
107 | for (const char *p = content;;) { |
108 | _cleanup_free_ char *line = NULL, *key = NULL, *val = NULL; | |
109 | const char *q; | |
110 | ||
111 | r = extract_first_word(&p, &line, "\n", 0); | |
112 | if (r < 0) | |
113 | return r; | |
114 | if (r == 0) | |
115 | return -ENOENT; | |
116 | ||
117 | q = line; | |
118 | r = extract_first_word(&q, &key, " ", 0); | |
119 | if (r < 0) | |
120 | return r; | |
121 | if (r == 0) | |
ab2c3861 TH |
122 | return -EINVAL; |
123 | ||
31a9be23 | 124 | if (!streq(key, event)) |
ab2c3861 TH |
125 | continue; |
126 | ||
31a9be23 YW |
127 | val = strdup(q); |
128 | if (!val) | |
129 | return -ENOMEM; | |
130 | ||
131 | *ret = TAKE_PTR(val); | |
ab2c3861 TH |
132 | return 0; |
133 | } | |
ab2c3861 TH |
134 | } |
135 | ||
3228995c CB |
136 | bool cg_ns_supported(void) { |
137 | static thread_local int enabled = -1; | |
138 | ||
139 | if (enabled >= 0) | |
140 | return enabled; | |
141 | ||
0887fa71 LP |
142 | if (access("/proc/self/ns/cgroup", F_OK) < 0) { |
143 | if (errno != ENOENT) | |
144 | log_debug_errno(errno, "Failed to check whether /proc/self/ns/cgroup is available, assuming not: %m"); | |
145 | enabled = false; | |
146 | } else | |
147 | enabled = true; | |
3228995c CB |
148 | |
149 | return enabled; | |
150 | } | |
151 | ||
35d2e7ec | 152 | int cg_enumerate_subgroups(const char *controller, const char *path, DIR **_d) { |
7027ff61 | 153 | _cleanup_free_ char *fs = NULL; |
35d2e7ec LP |
154 | int r; |
155 | DIR *d; | |
156 | ||
35d2e7ec LP |
157 | assert(_d); |
158 | ||
159 | /* This is not recursive! */ | |
160 | ||
c3175a7f LP |
161 | r = cg_get_path(controller, path, NULL, &fs); |
162 | if (r < 0) | |
35d2e7ec LP |
163 | return r; |
164 | ||
165 | d = opendir(fs); | |
35d2e7ec LP |
166 | if (!d) |
167 | return -errno; | |
168 | ||
169 | *_d = d; | |
170 | return 0; | |
171 | } | |
172 | ||
173 | int cg_read_subgroup(DIR *d, char **fn) { | |
174 | struct dirent *de; | |
175 | ||
176 | assert(d); | |
7027ff61 | 177 | assert(fn); |
35d2e7ec | 178 | |
f01327ad | 179 | FOREACH_DIRENT_ALL(de, d, return -errno) { |
35d2e7ec LP |
180 | char *b; |
181 | ||
182 | if (de->d_type != DT_DIR) | |
183 | continue; | |
184 | ||
49bfc877 | 185 | if (dot_or_dot_dot(de->d_name)) |
35d2e7ec LP |
186 | continue; |
187 | ||
7027ff61 LP |
188 | b = strdup(de->d_name); |
189 | if (!b) | |
35d2e7ec LP |
190 | return -ENOMEM; |
191 | ||
192 | *fn = b; | |
193 | return 1; | |
194 | } | |
195 | ||
35d2e7ec LP |
196 | return 0; |
197 | } | |
198 | ||
4ad49000 | 199 | int cg_rmdir(const char *controller, const char *path) { |
7027ff61 | 200 | _cleanup_free_ char *p = NULL; |
35d2e7ec LP |
201 | int r; |
202 | ||
ad293f5a LP |
203 | r = cg_get_path(controller, path, NULL, &p); |
204 | if (r < 0) | |
35d2e7ec LP |
205 | return r; |
206 | ||
207 | r = rmdir(p); | |
7027ff61 LP |
208 | if (r < 0 && errno != ENOENT) |
209 | return -errno; | |
35d2e7ec | 210 | |
b4cccbc1 | 211 | r = cg_hybrid_unified(); |
f20db199 | 212 | if (r <= 0) |
b4cccbc1 | 213 | return r; |
b4cccbc1 LP |
214 | |
215 | if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { | |
2977724b TH |
216 | r = cg_rmdir(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path); |
217 | if (r < 0) | |
218 | log_warning_errno(r, "Failed to remove compat systemd cgroup %s: %m", path); | |
219 | } | |
220 | ||
7027ff61 | 221 | return 0; |
35d2e7ec LP |
222 | } |
223 | ||
1d98fef1 LP |
224 | int cg_kill( |
225 | const char *controller, | |
226 | const char *path, | |
227 | int sig, | |
228 | CGroupFlags flags, | |
229 | Set *s, | |
230 | cg_kill_log_func_t log_kill, | |
231 | void *userdata) { | |
232 | ||
7027ff61 | 233 | _cleanup_set_free_ Set *allocated_set = NULL; |
35d2e7ec | 234 | bool done = false; |
c53d2d54 | 235 | int r, ret = 0, ret_log_kill = 0; |
35d2e7ec | 236 | pid_t my_pid; |
8c6db833 | 237 | |
8c6db833 LP |
238 | assert(sig >= 0); |
239 | ||
0d5b4810 LP |
240 | /* Don't send SIGCONT twice. Also, SIGKILL always works even when process is suspended, hence don't send |
241 | * SIGCONT on SIGKILL. */ | |
242 | if (IN_SET(sig, SIGCONT, SIGKILL)) | |
243 | flags &= ~CGROUP_SIGCONT; | |
244 | ||
8c6db833 LP |
245 | /* This goes through the tasks list and kills them all. This |
246 | * is repeated until no further processes are added to the | |
247 | * tasks list, to properly handle forking processes */ | |
248 | ||
7027ff61 | 249 | if (!s) { |
d5099efc | 250 | s = allocated_set = set_new(NULL); |
7027ff61 | 251 | if (!s) |
ca949c9d | 252 | return -ENOMEM; |
7027ff61 | 253 | } |
8c6db833 | 254 | |
df0ff127 | 255 | my_pid = getpid_cached(); |
8c6db833 LP |
256 | |
257 | do { | |
7027ff61 | 258 | _cleanup_fclose_ FILE *f = NULL; |
0b172489 | 259 | pid_t pid = 0; |
8c6db833 LP |
260 | done = true; |
261 | ||
7027ff61 LP |
262 | r = cg_enumerate_processes(controller, path, &f); |
263 | if (r < 0) { | |
4c633005 | 264 | if (ret >= 0 && r != -ENOENT) |
7027ff61 | 265 | return r; |
35d2e7ec | 266 | |
7027ff61 | 267 | return ret; |
35d2e7ec | 268 | } |
c6c18be3 LP |
269 | |
270 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
8c6db833 | 271 | |
1d98fef1 | 272 | if ((flags & CGROUP_IGNORE_SELF) && pid == my_pid) |
c6c18be3 | 273 | continue; |
8c6db833 | 274 | |
fea72cc0 | 275 | if (set_get(s, PID_TO_PTR(pid)) == PID_TO_PTR(pid)) |
c6c18be3 | 276 | continue; |
8c6db833 | 277 | |
1d98fef1 | 278 | if (log_kill) |
c53d2d54 | 279 | ret_log_kill = log_kill(pid, sig, userdata); |
1d98fef1 | 280 | |
8c6db833 LP |
281 | /* If we haven't killed this process yet, kill |
282 | * it */ | |
4c633005 LP |
283 | if (kill(pid, sig) < 0) { |
284 | if (ret >= 0 && errno != ESRCH) | |
8c6db833 | 285 | ret = -errno; |
6e8314c4 | 286 | } else { |
1d98fef1 | 287 | if (flags & CGROUP_SIGCONT) |
e155a0aa | 288 | (void) kill(pid, SIGCONT); |
430c18ed | 289 | |
c53d2d54 DB |
290 | if (ret == 0) { |
291 | if (log_kill) | |
292 | ret = ret_log_kill; | |
293 | else | |
294 | ret = 1; | |
295 | } | |
430c18ed | 296 | } |
8c6db833 | 297 | |
8c6db833 LP |
298 | done = false; |
299 | ||
fea72cc0 | 300 | r = set_put(s, PID_TO_PTR(pid)); |
7027ff61 | 301 | if (r < 0) { |
35d2e7ec | 302 | if (ret >= 0) |
7027ff61 | 303 | return r; |
35d2e7ec | 304 | |
7027ff61 | 305 | return ret; |
35d2e7ec LP |
306 | } |
307 | } | |
308 | ||
309 | if (r < 0) { | |
310 | if (ret >= 0) | |
7027ff61 | 311 | return r; |
35d2e7ec | 312 | |
7027ff61 | 313 | return ret; |
8c6db833 LP |
314 | } |
315 | ||
8c6db833 LP |
316 | /* To avoid racing against processes which fork |
317 | * quicker than we can kill them we repeat this until | |
318 | * no new pids need to be killed. */ | |
319 | ||
35d2e7ec | 320 | } while (!done); |
8c6db833 | 321 | |
35d2e7ec | 322 | return ret; |
8c6db833 LP |
323 | } |
324 | ||
1d98fef1 LP |
325 | int cg_kill_recursive( |
326 | const char *controller, | |
327 | const char *path, | |
328 | int sig, | |
329 | CGroupFlags flags, | |
330 | Set *s, | |
331 | cg_kill_log_func_t log_kill, | |
332 | void *userdata) { | |
333 | ||
7027ff61 LP |
334 | _cleanup_set_free_ Set *allocated_set = NULL; |
335 | _cleanup_closedir_ DIR *d = NULL; | |
e155a0aa | 336 | int r, ret; |
35d2e7ec | 337 | char *fn; |
8c6db833 LP |
338 | |
339 | assert(path); | |
8c6db833 LP |
340 | assert(sig >= 0); |
341 | ||
7027ff61 | 342 | if (!s) { |
d5099efc | 343 | s = allocated_set = set_new(NULL); |
7027ff61 | 344 | if (!s) |
ca949c9d | 345 | return -ENOMEM; |
7027ff61 | 346 | } |
ca949c9d | 347 | |
1d98fef1 | 348 | ret = cg_kill(controller, path, sig, flags, s, log_kill, userdata); |
8c6db833 | 349 | |
7027ff61 LP |
350 | r = cg_enumerate_subgroups(controller, path, &d); |
351 | if (r < 0) { | |
4c633005 | 352 | if (ret >= 0 && r != -ENOENT) |
7027ff61 | 353 | return r; |
8c6db833 | 354 | |
7027ff61 | 355 | return ret; |
35d2e7ec | 356 | } |
8c6db833 | 357 | |
35d2e7ec | 358 | while ((r = cg_read_subgroup(d, &fn)) > 0) { |
7027ff61 | 359 | _cleanup_free_ char *p = NULL; |
8c6db833 | 360 | |
605405c6 | 361 | p = strjoin(path, "/", fn); |
35d2e7ec | 362 | free(fn); |
7027ff61 LP |
363 | if (!p) |
364 | return -ENOMEM; | |
8c6db833 | 365 | |
1d98fef1 | 366 | r = cg_kill_recursive(controller, p, sig, flags, s, log_kill, userdata); |
e155a0aa | 367 | if (r != 0 && ret >= 0) |
35d2e7ec | 368 | ret = r; |
8c6db833 | 369 | } |
7027ff61 | 370 | if (ret >= 0 && r < 0) |
35d2e7ec LP |
371 | ret = r; |
372 | ||
1d98fef1 | 373 | if (flags & CGROUP_REMOVE) { |
4ad49000 | 374 | r = cg_rmdir(controller, path); |
4c701096 | 375 | if (r < 0 && ret >= 0 && !IN_SET(r, -ENOENT, -EBUSY)) |
7027ff61 LP |
376 | return r; |
377 | } | |
ca949c9d | 378 | |
8c6db833 LP |
379 | return ret; |
380 | } | |
381 | ||
1d98fef1 LP |
382 | int cg_migrate( |
383 | const char *cfrom, | |
384 | const char *pfrom, | |
385 | const char *cto, | |
386 | const char *pto, | |
387 | CGroupFlags flags) { | |
388 | ||
35d2e7ec | 389 | bool done = false; |
246aa6dd | 390 | _cleanup_set_free_ Set *s = NULL; |
8c6db833 LP |
391 | int r, ret = 0; |
392 | pid_t my_pid; | |
393 | ||
246aa6dd LP |
394 | assert(cfrom); |
395 | assert(pfrom); | |
396 | assert(cto); | |
397 | assert(pto); | |
8c6db833 | 398 | |
d5099efc | 399 | s = set_new(NULL); |
246aa6dd | 400 | if (!s) |
35d2e7ec LP |
401 | return -ENOMEM; |
402 | ||
df0ff127 | 403 | my_pid = getpid_cached(); |
8c6db833 LP |
404 | |
405 | do { | |
7027ff61 | 406 | _cleanup_fclose_ FILE *f = NULL; |
0b172489 | 407 | pid_t pid = 0; |
8c6db833 LP |
408 | done = true; |
409 | ||
b043cd0b | 410 | r = cg_enumerate_processes(cfrom, pfrom, &f); |
246aa6dd | 411 | if (r < 0) { |
4c633005 | 412 | if (ret >= 0 && r != -ENOENT) |
7027ff61 | 413 | return r; |
35d2e7ec | 414 | |
246aa6dd | 415 | return ret; |
35d2e7ec | 416 | } |
c6c18be3 LP |
417 | |
418 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
8c6db833 | 419 | |
35d2e7ec LP |
420 | /* This might do weird stuff if we aren't a |
421 | * single-threaded program. However, we | |
422 | * luckily know we are not */ | |
1d98fef1 | 423 | if ((flags & CGROUP_IGNORE_SELF) && pid == my_pid) |
c6c18be3 | 424 | continue; |
8c6db833 | 425 | |
fea72cc0 | 426 | if (set_get(s, PID_TO_PTR(pid)) == PID_TO_PTR(pid)) |
35d2e7ec LP |
427 | continue; |
428 | ||
9b84c7f9 LP |
429 | /* Ignore kernel threads. Since they can only |
430 | * exist in the root cgroup, we only check for | |
431 | * them there. */ | |
432 | if (cfrom && | |
57ea45e1 | 433 | empty_or_root(pfrom) && |
9b84c7f9 LP |
434 | is_kernel_thread(pid) > 0) |
435 | continue; | |
436 | ||
246aa6dd LP |
437 | r = cg_attach(cto, pto, pid); |
438 | if (r < 0) { | |
4c633005 | 439 | if (ret >= 0 && r != -ESRCH) |
35d2e7ec LP |
440 | ret = r; |
441 | } else if (ret == 0) | |
442 | ret = 1; | |
8c6db833 | 443 | |
8c6db833 | 444 | done = false; |
35d2e7ec | 445 | |
fea72cc0 | 446 | r = set_put(s, PID_TO_PTR(pid)); |
246aa6dd | 447 | if (r < 0) { |
35d2e7ec | 448 | if (ret >= 0) |
7027ff61 | 449 | return r; |
35d2e7ec | 450 | |
246aa6dd | 451 | return ret; |
35d2e7ec LP |
452 | } |
453 | } | |
454 | ||
455 | if (r < 0) { | |
456 | if (ret >= 0) | |
7027ff61 | 457 | return r; |
35d2e7ec | 458 | |
246aa6dd | 459 | return ret; |
8c6db833 | 460 | } |
35d2e7ec | 461 | } while (!done); |
8c6db833 | 462 | |
35d2e7ec | 463 | return ret; |
8c6db833 LP |
464 | } |
465 | ||
4ad49000 LP |
466 | int cg_migrate_recursive( |
467 | const char *cfrom, | |
468 | const char *pfrom, | |
469 | const char *cto, | |
470 | const char *pto, | |
1d98fef1 | 471 | CGroupFlags flags) { |
4ad49000 | 472 | |
246aa6dd | 473 | _cleanup_closedir_ DIR *d = NULL; |
7027ff61 | 474 | int r, ret = 0; |
35d2e7ec | 475 | char *fn; |
8c6db833 | 476 | |
246aa6dd LP |
477 | assert(cfrom); |
478 | assert(pfrom); | |
479 | assert(cto); | |
480 | assert(pto); | |
8c6db833 | 481 | |
1d98fef1 | 482 | ret = cg_migrate(cfrom, pfrom, cto, pto, flags); |
8c6db833 | 483 | |
246aa6dd LP |
484 | r = cg_enumerate_subgroups(cfrom, pfrom, &d); |
485 | if (r < 0) { | |
4c633005 | 486 | if (ret >= 0 && r != -ENOENT) |
7027ff61 LP |
487 | return r; |
488 | ||
246aa6dd | 489 | return ret; |
35d2e7ec LP |
490 | } |
491 | ||
492 | while ((r = cg_read_subgroup(d, &fn)) > 0) { | |
246aa6dd | 493 | _cleanup_free_ char *p = NULL; |
8c6db833 | 494 | |
605405c6 | 495 | p = strjoin(pfrom, "/", fn); |
35d2e7ec | 496 | free(fn); |
e155a0aa LP |
497 | if (!p) |
498 | return -ENOMEM; | |
8c6db833 | 499 | |
1d98fef1 | 500 | r = cg_migrate_recursive(cfrom, p, cto, pto, flags); |
35d2e7ec LP |
501 | if (r != 0 && ret >= 0) |
502 | ret = r; | |
8c6db833 LP |
503 | } |
504 | ||
35d2e7ec LP |
505 | if (r < 0 && ret >= 0) |
506 | ret = r; | |
507 | ||
1d98fef1 | 508 | if (flags & CGROUP_REMOVE) { |
4ad49000 | 509 | r = cg_rmdir(cfrom, pfrom); |
4c701096 | 510 | if (r < 0 && ret >= 0 && !IN_SET(r, -ENOENT, -EBUSY)) |
246aa6dd LP |
511 | return r; |
512 | } | |
8c6db833 LP |
513 | |
514 | return ret; | |
515 | } | |
516 | ||
13b84ec7 LP |
517 | int cg_migrate_recursive_fallback( |
518 | const char *cfrom, | |
519 | const char *pfrom, | |
520 | const char *cto, | |
521 | const char *pto, | |
1d98fef1 | 522 | CGroupFlags flags) { |
13b84ec7 LP |
523 | |
524 | int r; | |
525 | ||
526 | assert(cfrom); | |
527 | assert(pfrom); | |
528 | assert(cto); | |
529 | assert(pto); | |
530 | ||
1d98fef1 | 531 | r = cg_migrate_recursive(cfrom, pfrom, cto, pto, flags); |
13b84ec7 LP |
532 | if (r < 0) { |
533 | char prefix[strlen(pto) + 1]; | |
534 | ||
535 | /* This didn't work? Then let's try all prefixes of the destination */ | |
536 | ||
fecffe5d | 537 | PATH_FOREACH_PREFIX(prefix, pto) { |
e155a0aa LP |
538 | int q; |
539 | ||
1d98fef1 | 540 | q = cg_migrate_recursive(cfrom, pfrom, cto, prefix, flags); |
e155a0aa LP |
541 | if (q >= 0) |
542 | return q; | |
13b84ec7 LP |
543 | } |
544 | } | |
545 | ||
e155a0aa | 546 | return r; |
13b84ec7 LP |
547 | } |
548 | ||
efdb0237 LP |
549 | static const char *controller_to_dirname(const char *controller) { |
550 | const char *e; | |
3474ae3c | 551 | |
7027ff61 LP |
552 | assert(controller); |
553 | ||
efdb0237 LP |
554 | /* Converts a controller name to the directory name below |
555 | * /sys/fs/cgroup/ we want to mount it to. Effectively, this | |
556 | * just cuts off the name= prefixed used for named | |
557 | * hierarchies, if it is specified. */ | |
558 | ||
2977724b | 559 | if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { |
b4cccbc1 | 560 | if (cg_hybrid_unified() > 0) |
2977724b TH |
561 | controller = SYSTEMD_CGROUP_CONTROLLER_HYBRID; |
562 | else | |
563 | controller = SYSTEMD_CGROUP_CONTROLLER_LEGACY; | |
564 | } | |
b6629c4b | 565 | |
efdb0237 LP |
566 | e = startswith(controller, "name="); |
567 | if (e) | |
568 | return e; | |
569 | ||
570 | return controller; | |
3474ae3c LP |
571 | } |
572 | ||
569b19d8 LP |
573 | static int join_path_legacy(const char *controller, const char *path, const char *suffix, char **fs) { |
574 | const char *dn; | |
018ef268 | 575 | char *t = NULL; |
3474ae3c | 576 | |
efdb0237 | 577 | assert(fs); |
569b19d8 LP |
578 | assert(controller); |
579 | ||
580 | dn = controller_to_dirname(controller); | |
efdb0237 LP |
581 | |
582 | if (isempty(path) && isempty(suffix)) | |
569b19d8 | 583 | t = strappend("/sys/fs/cgroup/", dn); |
efdb0237 | 584 | else if (isempty(path)) |
605405c6 | 585 | t = strjoin("/sys/fs/cgroup/", dn, "/", suffix); |
efdb0237 | 586 | else if (isempty(suffix)) |
605405c6 | 587 | t = strjoin("/sys/fs/cgroup/", dn, "/", path); |
efdb0237 | 588 | else |
605405c6 | 589 | t = strjoin("/sys/fs/cgroup/", dn, "/", path, "/", suffix); |
efdb0237 LP |
590 | if (!t) |
591 | return -ENOMEM; | |
3474ae3c | 592 | |
efdb0237 LP |
593 | *fs = t; |
594 | return 0; | |
595 | } | |
596 | ||
597 | static int join_path_unified(const char *path, const char *suffix, char **fs) { | |
598 | char *t; | |
599 | ||
600 | assert(fs); | |
601 | ||
602 | if (isempty(path) && isempty(suffix)) | |
603 | t = strdup("/sys/fs/cgroup"); | |
604 | else if (isempty(path)) | |
605 | t = strappend("/sys/fs/cgroup/", suffix); | |
606 | else if (isempty(suffix)) | |
607 | t = strappend("/sys/fs/cgroup/", path); | |
608 | else | |
605405c6 | 609 | t = strjoin("/sys/fs/cgroup/", path, "/", suffix); |
3474ae3c LP |
610 | if (!t) |
611 | return -ENOMEM; | |
612 | ||
efdb0237 | 613 | *fs = t; |
3474ae3c LP |
614 | return 0; |
615 | } | |
616 | ||
8c6db833 | 617 | int cg_get_path(const char *controller, const char *path, const char *suffix, char **fs) { |
415fc41c | 618 | int r; |
8c6db833 | 619 | |
dbd821ac LP |
620 | assert(fs); |
621 | ||
efdb0237 LP |
622 | if (!controller) { |
623 | char *t; | |
624 | ||
569b19d8 LP |
625 | /* If no controller is specified, we return the path |
626 | * *below* the controllers, without any prefix. */ | |
efdb0237 LP |
627 | |
628 | if (!path && !suffix) | |
629 | return -EINVAL; | |
630 | ||
989189ea | 631 | if (!suffix) |
efdb0237 | 632 | t = strdup(path); |
989189ea | 633 | else if (!path) |
efdb0237 LP |
634 | t = strdup(suffix); |
635 | else | |
605405c6 | 636 | t = strjoin(path, "/", suffix); |
efdb0237 LP |
637 | if (!t) |
638 | return -ENOMEM; | |
639 | ||
858d36c1 | 640 | *fs = path_simplify(t, false); |
efdb0237 LP |
641 | return 0; |
642 | } | |
643 | ||
644 | if (!cg_controller_is_valid(controller)) | |
78edb35a LP |
645 | return -EINVAL; |
646 | ||
b4cccbc1 LP |
647 | r = cg_all_unified(); |
648 | if (r < 0) | |
649 | return r; | |
650 | if (r > 0) | |
efdb0237 | 651 | r = join_path_unified(path, suffix, fs); |
569b19d8 LP |
652 | else |
653 | r = join_path_legacy(controller, path, suffix, fs); | |
efdb0237 LP |
654 | if (r < 0) |
655 | return r; | |
7027ff61 | 656 | |
858d36c1 | 657 | path_simplify(*fs, false); |
efdb0237 | 658 | return 0; |
3474ae3c | 659 | } |
dbd821ac | 660 | |
efdb0237 | 661 | static int controller_is_accessible(const char *controller) { |
b4cccbc1 | 662 | int r; |
37099707 | 663 | |
efdb0237 | 664 | assert(controller); |
37099707 | 665 | |
efdb0237 LP |
666 | /* Checks whether a specific controller is accessible, |
667 | * i.e. its hierarchy mounted. In the unified hierarchy all | |
668 | * controllers are considered accessible, except for the named | |
669 | * hierarchies */ | |
b12afc8c | 670 | |
efdb0237 LP |
671 | if (!cg_controller_is_valid(controller)) |
672 | return -EINVAL; | |
673 | ||
b4cccbc1 LP |
674 | r = cg_all_unified(); |
675 | if (r < 0) | |
676 | return r; | |
677 | if (r > 0) { | |
efdb0237 LP |
678 | /* We don't support named hierarchies if we are using |
679 | * the unified hierarchy. */ | |
680 | ||
681 | if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) | |
682 | return 0; | |
683 | ||
684 | if (startswith(controller, "name=")) | |
685 | return -EOPNOTSUPP; | |
686 | ||
687 | } else { | |
688 | const char *cc, *dn; | |
689 | ||
690 | dn = controller_to_dirname(controller); | |
691 | cc = strjoina("/sys/fs/cgroup/", dn); | |
692 | ||
693 | if (laccess(cc, F_OK) < 0) | |
694 | return -errno; | |
695 | } | |
37099707 LP |
696 | |
697 | return 0; | |
698 | } | |
699 | ||
3474ae3c | 700 | int cg_get_path_and_check(const char *controller, const char *path, const char *suffix, char **fs) { |
37099707 | 701 | int r; |
dbd821ac | 702 | |
efdb0237 | 703 | assert(controller); |
3474ae3c | 704 | assert(fs); |
70132bd0 | 705 | |
efdb0237 LP |
706 | /* Check if the specified controller is actually accessible */ |
707 | r = controller_is_accessible(controller); | |
37099707 LP |
708 | if (r < 0) |
709 | return r; | |
3474ae3c | 710 | |
efdb0237 | 711 | return cg_get_path(controller, path, suffix, fs); |
8c6db833 LP |
712 | } |
713 | ||
e27796a0 | 714 | static int trim_cb(const char *path, const struct stat *sb, int typeflag, struct FTW *ftwbuf) { |
4ad49000 LP |
715 | assert(path); |
716 | assert(sb); | |
717 | assert(ftwbuf); | |
e27796a0 LP |
718 | |
719 | if (typeflag != FTW_DP) | |
720 | return 0; | |
721 | ||
722 | if (ftwbuf->level < 1) | |
723 | return 0; | |
724 | ||
e155a0aa | 725 | (void) rmdir(path); |
e27796a0 LP |
726 | return 0; |
727 | } | |
728 | ||
8c6db833 | 729 | int cg_trim(const char *controller, const char *path, bool delete_root) { |
7027ff61 | 730 | _cleanup_free_ char *fs = NULL; |
2977724b | 731 | int r = 0, q; |
8c6db833 | 732 | |
8c6db833 LP |
733 | assert(path); |
734 | ||
e27796a0 LP |
735 | r = cg_get_path(controller, path, NULL, &fs); |
736 | if (r < 0) | |
8c6db833 LP |
737 | return r; |
738 | ||
e27796a0 | 739 | errno = 0; |
e155a0aa LP |
740 | if (nftw(fs, trim_cb, 64, FTW_DEPTH|FTW_MOUNT|FTW_PHYS) != 0) { |
741 | if (errno == ENOENT) | |
742 | r = 0; | |
b3267152 | 743 | else if (errno > 0) |
e155a0aa LP |
744 | r = -errno; |
745 | else | |
746 | r = -EIO; | |
747 | } | |
e27796a0 LP |
748 | |
749 | if (delete_root) { | |
4ad49000 LP |
750 | if (rmdir(fs) < 0 && errno != ENOENT) |
751 | return -errno; | |
e27796a0 LP |
752 | } |
753 | ||
b4cccbc1 LP |
754 | q = cg_hybrid_unified(); |
755 | if (q < 0) | |
756 | return q; | |
757 | if (q > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { | |
2977724b TH |
758 | q = cg_trim(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, delete_root); |
759 | if (q < 0) | |
760 | log_warning_errno(q, "Failed to trim compat systemd cgroup %s: %m", path); | |
761 | } | |
762 | ||
e27796a0 | 763 | return r; |
8c6db833 LP |
764 | } |
765 | ||
65be7e06 ZJS |
766 | /* Create a cgroup in the hierarchy of controller. |
767 | * Returns 0 if the group already existed, 1 on success, negative otherwise. | |
768 | */ | |
1434ae6f LP |
769 | int cg_create(const char *controller, const char *path) { |
770 | _cleanup_free_ char *fs = NULL; | |
771 | int r; | |
772 | ||
773 | r = cg_get_path_and_check(controller, path, NULL, &fs); | |
774 | if (r < 0) | |
775 | return r; | |
776 | ||
777 | r = mkdir_parents(fs, 0755); | |
778 | if (r < 0) | |
779 | return r; | |
780 | ||
dae8b82e ZJS |
781 | r = mkdir_errno_wrapper(fs, 0755); |
782 | if (r == -EEXIST) | |
783 | return 0; | |
784 | if (r < 0) | |
785 | return r; | |
1434ae6f | 786 | |
b4cccbc1 LP |
787 | r = cg_hybrid_unified(); |
788 | if (r < 0) | |
789 | return r; | |
790 | ||
791 | if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { | |
2977724b TH |
792 | r = cg_create(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path); |
793 | if (r < 0) | |
794 | log_warning_errno(r, "Failed to create compat systemd cgroup %s: %m", path); | |
795 | } | |
796 | ||
1434ae6f LP |
797 | return 1; |
798 | } | |
799 | ||
800 | int cg_create_and_attach(const char *controller, const char *path, pid_t pid) { | |
801 | int r, q; | |
802 | ||
803 | assert(pid >= 0); | |
804 | ||
805 | r = cg_create(controller, path); | |
806 | if (r < 0) | |
807 | return r; | |
808 | ||
809 | q = cg_attach(controller, path, pid); | |
810 | if (q < 0) | |
811 | return q; | |
812 | ||
813 | /* This does not remove the cgroup on failure */ | |
814 | return r; | |
815 | } | |
816 | ||
8c6db833 | 817 | int cg_attach(const char *controller, const char *path, pid_t pid) { |
574d5f2d LP |
818 | _cleanup_free_ char *fs = NULL; |
819 | char c[DECIMAL_STR_MAX(pid_t) + 2]; | |
8c6db833 LP |
820 | int r; |
821 | ||
8c6db833 LP |
822 | assert(path); |
823 | assert(pid >= 0); | |
824 | ||
b043cd0b | 825 | r = cg_get_path_and_check(controller, path, "cgroup.procs", &fs); |
3474ae3c | 826 | if (r < 0) |
c6c18be3 | 827 | return r; |
8c6db833 LP |
828 | |
829 | if (pid == 0) | |
df0ff127 | 830 | pid = getpid_cached(); |
8c6db833 | 831 | |
d054f0a4 | 832 | xsprintf(c, PID_FMT "\n", pid); |
8c6db833 | 833 | |
604028de | 834 | r = write_string_file(fs, c, WRITE_STRING_FILE_DISABLE_BUFFER); |
2977724b TH |
835 | if (r < 0) |
836 | return r; | |
837 | ||
b4cccbc1 LP |
838 | r = cg_hybrid_unified(); |
839 | if (r < 0) | |
840 | return r; | |
841 | ||
842 | if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { | |
2977724b TH |
843 | r = cg_attach(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, pid); |
844 | if (r < 0) | |
bd68e99b | 845 | log_warning_errno(r, "Failed to attach "PID_FMT" to compat systemd cgroup %s: %m", pid, path); |
2977724b TH |
846 | } |
847 | ||
848 | return 0; | |
8c6db833 LP |
849 | } |
850 | ||
13b84ec7 LP |
851 | int cg_attach_fallback(const char *controller, const char *path, pid_t pid) { |
852 | int r; | |
853 | ||
854 | assert(controller); | |
855 | assert(path); | |
856 | assert(pid >= 0); | |
857 | ||
858 | r = cg_attach(controller, path, pid); | |
859 | if (r < 0) { | |
860 | char prefix[strlen(path) + 1]; | |
861 | ||
862 | /* This didn't work? Then let's try all prefixes of | |
863 | * the destination */ | |
864 | ||
fecffe5d | 865 | PATH_FOREACH_PREFIX(prefix, path) { |
e155a0aa LP |
866 | int q; |
867 | ||
868 | q = cg_attach(controller, prefix, pid); | |
869 | if (q >= 0) | |
870 | return q; | |
13b84ec7 LP |
871 | } |
872 | } | |
873 | ||
e155a0aa | 874 | return r; |
13b84ec7 LP |
875 | } |
876 | ||
62b9bb26 | 877 | int cg_set_access( |
2d76d14e LP |
878 | const char *controller, |
879 | const char *path, | |
2d76d14e LP |
880 | uid_t uid, |
881 | gid_t gid) { | |
882 | ||
62b9bb26 LP |
883 | struct Attribute { |
884 | const char *name; | |
885 | bool fatal; | |
886 | }; | |
887 | ||
4e1dfa45 | 888 | /* cgroup v1, aka legacy/non-unified */ |
62b9bb26 LP |
889 | static const struct Attribute legacy_attributes[] = { |
890 | { "cgroup.procs", true }, | |
891 | { "tasks", false }, | |
892 | { "cgroup.clone_children", false }, | |
893 | {}, | |
894 | }; | |
895 | ||
4e1dfa45 | 896 | /* cgroup v2, aka unified */ |
62b9bb26 LP |
897 | static const struct Attribute unified_attributes[] = { |
898 | { "cgroup.procs", true }, | |
899 | { "cgroup.subtree_control", true }, | |
900 | { "cgroup.threads", false }, | |
901 | {}, | |
902 | }; | |
903 | ||
904 | static const struct Attribute* const attributes[] = { | |
905 | [false] = legacy_attributes, | |
906 | [true] = unified_attributes, | |
907 | }; | |
974efc46 | 908 | |
40853aa5 | 909 | _cleanup_free_ char *fs = NULL; |
62b9bb26 LP |
910 | const struct Attribute *i; |
911 | int r, unified; | |
8c6db833 | 912 | |
8c6db833 LP |
913 | assert(path); |
914 | ||
62b9bb26 | 915 | if (uid == UID_INVALID && gid == GID_INVALID) |
8d53b453 LP |
916 | return 0; |
917 | ||
62b9bb26 LP |
918 | unified = cg_unified_controller(controller); |
919 | if (unified < 0) | |
920 | return unified; | |
8c6db833 | 921 | |
62b9bb26 LP |
922 | /* Configure access to the cgroup itself */ |
923 | r = cg_get_path(controller, path, NULL, &fs); | |
974efc46 LP |
924 | if (r < 0) |
925 | return r; | |
8c6db833 | 926 | |
62b9bb26 | 927 | r = chmod_and_chown(fs, 0755, uid, gid); |
b4cccbc1 LP |
928 | if (r < 0) |
929 | return r; | |
40853aa5 | 930 | |
62b9bb26 LP |
931 | /* Configure access to the cgroup's attributes */ |
932 | for (i = attributes[unified]; i->name; i++) { | |
40853aa5 | 933 | fs = mfree(fs); |
40853aa5 | 934 | |
62b9bb26 | 935 | r = cg_get_path(controller, path, i->name, &fs); |
40853aa5 LP |
936 | if (r < 0) |
937 | return r; | |
efdb0237 | 938 | |
62b9bb26 LP |
939 | r = chmod_and_chown(fs, 0644, uid, gid); |
940 | if (r < 0) { | |
941 | if (i->fatal) | |
942 | return r; | |
5beac75e | 943 | |
62b9bb26 LP |
944 | log_debug_errno(r, "Failed to set access on cgroup %s, ignoring: %m", fs); |
945 | } | |
946 | } | |
947 | ||
948 | if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { | |
949 | r = cg_hybrid_unified(); | |
2977724b | 950 | if (r < 0) |
62b9bb26 LP |
951 | return r; |
952 | if (r > 0) { | |
953 | /* Always propagate access mode from unified to legacy controller */ | |
954 | r = cg_set_access(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, uid, gid); | |
955 | if (r < 0) | |
956 | log_debug_errno(r, "Failed to set access on compatibility systemd cgroup %s, ignoring: %m", path); | |
957 | } | |
2977724b | 958 | } |
974efc46 | 959 | |
efdb0237 | 960 | return 0; |
8c6db833 LP |
961 | } |
962 | ||
4b58153d LP |
963 | int cg_set_xattr(const char *controller, const char *path, const char *name, const void *value, size_t size, int flags) { |
964 | _cleanup_free_ char *fs = NULL; | |
965 | int r; | |
966 | ||
967 | assert(path); | |
968 | assert(name); | |
969 | assert(value || size <= 0); | |
970 | ||
971 | r = cg_get_path(controller, path, NULL, &fs); | |
972 | if (r < 0) | |
973 | return r; | |
974 | ||
975 | if (setxattr(fs, name, value, size, flags) < 0) | |
976 | return -errno; | |
977 | ||
978 | return 0; | |
979 | } | |
980 | ||
981 | int cg_get_xattr(const char *controller, const char *path, const char *name, void *value, size_t size) { | |
982 | _cleanup_free_ char *fs = NULL; | |
983 | ssize_t n; | |
984 | int r; | |
985 | ||
986 | assert(path); | |
987 | assert(name); | |
988 | ||
989 | r = cg_get_path(controller, path, NULL, &fs); | |
990 | if (r < 0) | |
991 | return r; | |
992 | ||
993 | n = getxattr(fs, name, value, size); | |
994 | if (n < 0) | |
995 | return -errno; | |
996 | ||
997 | return (int) n; | |
998 | } | |
999 | ||
7027ff61 | 1000 | int cg_pid_get_path(const char *controller, pid_t pid, char **path) { |
7027ff61 | 1001 | _cleanup_fclose_ FILE *f = NULL; |
b6629c4b | 1002 | const char *fs, *controller_str; |
d2b39cb6 | 1003 | int unified, r; |
efdb0237 | 1004 | size_t cs = 0; |
8c6db833 | 1005 | |
8c6db833 | 1006 | assert(path); |
c6c18be3 | 1007 | assert(pid >= 0); |
8c6db833 | 1008 | |
5da38d07 TH |
1009 | if (controller) { |
1010 | if (!cg_controller_is_valid(controller)) | |
1011 | return -EINVAL; | |
1012 | } else | |
1013 | controller = SYSTEMD_CGROUP_CONTROLLER; | |
1014 | ||
c22800e4 | 1015 | unified = cg_unified_controller(controller); |
b4cccbc1 LP |
1016 | if (unified < 0) |
1017 | return unified; | |
1018 | if (unified == 0) { | |
b6629c4b TH |
1019 | if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) |
1020 | controller_str = SYSTEMD_CGROUP_CONTROLLER_LEGACY; | |
1021 | else | |
1022 | controller_str = controller; | |
1023 | ||
1024 | cs = strlen(controller_str); | |
1025 | } | |
7027ff61 | 1026 | |
b68fa010 | 1027 | fs = procfs_file_alloca(pid, "cgroup"); |
fdeea3f4 ZJS |
1028 | r = fopen_unlocked(fs, "re", &f); |
1029 | if (r == -ENOENT) | |
1030 | return -ESRCH; | |
1031 | if (r < 0) | |
1032 | return r; | |
35bbbf85 | 1033 | |
d2b39cb6 LP |
1034 | for (;;) { |
1035 | _cleanup_free_ char *line = NULL; | |
efdb0237 | 1036 | char *e, *p; |
c6c18be3 | 1037 | |
d2b39cb6 LP |
1038 | r = read_line(f, LONG_LINE_MAX, &line); |
1039 | if (r < 0) | |
1040 | return r; | |
1041 | if (r == 0) | |
1042 | break; | |
c6c18be3 | 1043 | |
efdb0237 LP |
1044 | if (unified) { |
1045 | e = startswith(line, "0:"); | |
1046 | if (!e) | |
1047 | continue; | |
c6c18be3 | 1048 | |
efdb0237 LP |
1049 | e = strchr(e, ':'); |
1050 | if (!e) | |
1051 | continue; | |
1052 | } else { | |
1053 | char *l; | |
1054 | size_t k; | |
1055 | const char *word, *state; | |
1056 | bool found = false; | |
1057 | ||
1058 | l = strchr(line, ':'); | |
1059 | if (!l) | |
1060 | continue; | |
8af8afd6 | 1061 | |
efdb0237 LP |
1062 | l++; |
1063 | e = strchr(l, ':'); | |
1064 | if (!e) | |
1065 | continue; | |
8af8afd6 | 1066 | |
efdb0237 | 1067 | *e = 0; |
00d4b1e6 | 1068 | FOREACH_WORD_SEPARATOR(word, k, l, ",", state) |
b6629c4b | 1069 | if (k == cs && memcmp(word, controller_str, cs) == 0) { |
efdb0237 LP |
1070 | found = true; |
1071 | break; | |
1072 | } | |
efdb0237 LP |
1073 | if (!found) |
1074 | continue; | |
8af8afd6 LP |
1075 | } |
1076 | ||
8af8afd6 | 1077 | p = strdup(e + 1); |
7027ff61 LP |
1078 | if (!p) |
1079 | return -ENOMEM; | |
c6c18be3 | 1080 | |
5e20b0a4 LP |
1081 | /* Truncate suffix indicating the process is a zombie */ |
1082 | e = endswith(p, " (deleted)"); | |
1083 | if (e) | |
1084 | *e = 0; | |
1085 | ||
c6c18be3 | 1086 | *path = p; |
7027ff61 | 1087 | return 0; |
c6c18be3 LP |
1088 | } |
1089 | ||
1c80e425 | 1090 | return -ENODATA; |
8c6db833 LP |
1091 | } |
1092 | ||
1093 | int cg_install_release_agent(const char *controller, const char *agent) { | |
7027ff61 | 1094 | _cleanup_free_ char *fs = NULL, *contents = NULL; |
efdb0237 | 1095 | const char *sc; |
415fc41c | 1096 | int r; |
8c6db833 | 1097 | |
8c6db833 LP |
1098 | assert(agent); |
1099 | ||
c22800e4 | 1100 | r = cg_unified_controller(controller); |
b4cccbc1 LP |
1101 | if (r < 0) |
1102 | return r; | |
1103 | if (r > 0) /* doesn't apply to unified hierarchy */ | |
efdb0237 LP |
1104 | return -EOPNOTSUPP; |
1105 | ||
7027ff61 LP |
1106 | r = cg_get_path(controller, NULL, "release_agent", &fs); |
1107 | if (r < 0) | |
c6c18be3 | 1108 | return r; |
8c6db833 | 1109 | |
7027ff61 LP |
1110 | r = read_one_line_file(fs, &contents); |
1111 | if (r < 0) | |
1112 | return r; | |
8c6db833 LP |
1113 | |
1114 | sc = strstrip(contents); | |
e155a0aa | 1115 | if (isempty(sc)) { |
604028de | 1116 | r = write_string_file(fs, agent, WRITE_STRING_FILE_DISABLE_BUFFER); |
574d5f2d | 1117 | if (r < 0) |
7027ff61 | 1118 | return r; |
b8725df8 | 1119 | } else if (!path_equal(sc, agent)) |
7027ff61 | 1120 | return -EEXIST; |
8c6db833 | 1121 | |
0da16248 | 1122 | fs = mfree(fs); |
7027ff61 LP |
1123 | r = cg_get_path(controller, NULL, "notify_on_release", &fs); |
1124 | if (r < 0) | |
1125 | return r; | |
8c6db833 | 1126 | |
0da16248 | 1127 | contents = mfree(contents); |
7027ff61 LP |
1128 | r = read_one_line_file(fs, &contents); |
1129 | if (r < 0) | |
1130 | return r; | |
8c6db833 LP |
1131 | |
1132 | sc = strstrip(contents); | |
8c6db833 | 1133 | if (streq(sc, "0")) { |
604028de | 1134 | r = write_string_file(fs, "1", WRITE_STRING_FILE_DISABLE_BUFFER); |
7027ff61 LP |
1135 | if (r < 0) |
1136 | return r; | |
c6c18be3 | 1137 | |
7027ff61 LP |
1138 | return 1; |
1139 | } | |
8c6db833 | 1140 | |
7027ff61 LP |
1141 | if (!streq(sc, "1")) |
1142 | return -EIO; | |
8c6db833 | 1143 | |
7027ff61 | 1144 | return 0; |
8c6db833 LP |
1145 | } |
1146 | ||
ad929bcc KS |
1147 | int cg_uninstall_release_agent(const char *controller) { |
1148 | _cleanup_free_ char *fs = NULL; | |
415fc41c | 1149 | int r; |
efdb0237 | 1150 | |
c22800e4 | 1151 | r = cg_unified_controller(controller); |
b4cccbc1 LP |
1152 | if (r < 0) |
1153 | return r; | |
1154 | if (r > 0) /* Doesn't apply to unified hierarchy */ | |
efdb0237 | 1155 | return -EOPNOTSUPP; |
ad929bcc | 1156 | |
ac9ef333 LP |
1157 | r = cg_get_path(controller, NULL, "notify_on_release", &fs); |
1158 | if (r < 0) | |
1159 | return r; | |
1160 | ||
604028de | 1161 | r = write_string_file(fs, "0", WRITE_STRING_FILE_DISABLE_BUFFER); |
ac9ef333 LP |
1162 | if (r < 0) |
1163 | return r; | |
1164 | ||
0da16248 | 1165 | fs = mfree(fs); |
ac9ef333 | 1166 | |
ad929bcc KS |
1167 | r = cg_get_path(controller, NULL, "release_agent", &fs); |
1168 | if (r < 0) | |
1169 | return r; | |
1170 | ||
604028de | 1171 | r = write_string_file(fs, "", WRITE_STRING_FILE_DISABLE_BUFFER); |
ad929bcc KS |
1172 | if (r < 0) |
1173 | return r; | |
1174 | ||
ac9ef333 | 1175 | return 0; |
ad929bcc KS |
1176 | } |
1177 | ||
6f883237 | 1178 | int cg_is_empty(const char *controller, const char *path) { |
7027ff61 | 1179 | _cleanup_fclose_ FILE *f = NULL; |
efdb0237 | 1180 | pid_t pid; |
7027ff61 | 1181 | int r; |
8c6db833 | 1182 | |
8c6db833 LP |
1183 | assert(path); |
1184 | ||
b043cd0b | 1185 | r = cg_enumerate_processes(controller, path, &f); |
6f883237 | 1186 | if (r == -ENOENT) |
1bcf3fc6 | 1187 | return true; |
c3175a7f | 1188 | if (r < 0) |
6f883237 | 1189 | return r; |
8c6db833 | 1190 | |
6f883237 | 1191 | r = cg_read_pid(f, &pid); |
c6c18be3 LP |
1192 | if (r < 0) |
1193 | return r; | |
8c6db833 | 1194 | |
6f883237 | 1195 | return r == 0; |
8c6db833 LP |
1196 | } |
1197 | ||
6f883237 | 1198 | int cg_is_empty_recursive(const char *controller, const char *path) { |
415fc41c | 1199 | int r; |
8c6db833 | 1200 | |
8c6db833 LP |
1201 | assert(path); |
1202 | ||
6fd66507 | 1203 | /* The root cgroup is always populated */ |
57ea45e1 | 1204 | if (controller && empty_or_root(path)) |
efdb0237 | 1205 | return false; |
6fd66507 | 1206 | |
c22800e4 | 1207 | r = cg_unified_controller(controller); |
b4cccbc1 LP |
1208 | if (r < 0) |
1209 | return r; | |
1210 | if (r > 0) { | |
ab2c3861 | 1211 | _cleanup_free_ char *t = NULL; |
8c6db833 | 1212 | |
efdb0237 | 1213 | /* On the unified hierarchy we can check empty state |
ab2c3861 | 1214 | * via the "populated" attribute of "cgroup.events". */ |
8c6db833 | 1215 | |
ab2c3861 | 1216 | r = cg_read_event(controller, path, "populated", &t); |
1bcf3fc6 ZJS |
1217 | if (r == -ENOENT) |
1218 | return true; | |
efdb0237 LP |
1219 | if (r < 0) |
1220 | return r; | |
1221 | ||
1222 | return streq(t, "0"); | |
1223 | } else { | |
1224 | _cleanup_closedir_ DIR *d = NULL; | |
1225 | char *fn; | |
8c6db833 | 1226 | |
efdb0237 | 1227 | r = cg_is_empty(controller, path); |
35d2e7ec | 1228 | if (r <= 0) |
7027ff61 | 1229 | return r; |
35d2e7ec | 1230 | |
efdb0237 LP |
1231 | r = cg_enumerate_subgroups(controller, path, &d); |
1232 | if (r == -ENOENT) | |
1bcf3fc6 | 1233 | return true; |
efdb0237 LP |
1234 | if (r < 0) |
1235 | return r; | |
35d2e7ec | 1236 | |
efdb0237 LP |
1237 | while ((r = cg_read_subgroup(d, &fn)) > 0) { |
1238 | _cleanup_free_ char *p = NULL; | |
1239 | ||
605405c6 | 1240 | p = strjoin(path, "/", fn); |
efdb0237 LP |
1241 | free(fn); |
1242 | if (!p) | |
1243 | return -ENOMEM; | |
1244 | ||
1245 | r = cg_is_empty_recursive(controller, p); | |
1246 | if (r <= 0) | |
1247 | return r; | |
1248 | } | |
1249 | if (r < 0) | |
1250 | return r; | |
1251 | ||
1252 | return true; | |
1253 | } | |
35d2e7ec LP |
1254 | } |
1255 | ||
1256 | int cg_split_spec(const char *spec, char **controller, char **path) { | |
35d2e7ec | 1257 | char *t = NULL, *u = NULL; |
efdb0237 | 1258 | const char *e; |
35d2e7ec LP |
1259 | |
1260 | assert(spec); | |
35d2e7ec LP |
1261 | |
1262 | if (*spec == '/') { | |
99be45a4 | 1263 | if (!path_is_normalized(spec)) |
e884315e | 1264 | return -EINVAL; |
35d2e7ec LP |
1265 | |
1266 | if (path) { | |
246aa6dd LP |
1267 | t = strdup(spec); |
1268 | if (!t) | |
35d2e7ec LP |
1269 | return -ENOMEM; |
1270 | ||
858d36c1 | 1271 | *path = path_simplify(t, false); |
8c6db833 LP |
1272 | } |
1273 | ||
35d2e7ec LP |
1274 | if (controller) |
1275 | *controller = NULL; | |
1276 | ||
1277 | return 0; | |
8c6db833 LP |
1278 | } |
1279 | ||
246aa6dd LP |
1280 | e = strchr(spec, ':'); |
1281 | if (!e) { | |
185a0874 | 1282 | if (!cg_controller_is_valid(spec)) |
35d2e7ec LP |
1283 | return -EINVAL; |
1284 | ||
1285 | if (controller) { | |
efdb0237 | 1286 | t = strdup(spec); |
246aa6dd | 1287 | if (!t) |
35d2e7ec LP |
1288 | return -ENOMEM; |
1289 | ||
1290 | *controller = t; | |
1291 | } | |
1292 | ||
1293 | if (path) | |
1294 | *path = NULL; | |
1295 | ||
1296 | return 0; | |
8c6db833 LP |
1297 | } |
1298 | ||
efdb0237 | 1299 | t = strndup(spec, e-spec); |
e884315e LP |
1300 | if (!t) |
1301 | return -ENOMEM; | |
185a0874 | 1302 | if (!cg_controller_is_valid(t)) { |
e884315e | 1303 | free(t); |
35d2e7ec | 1304 | return -EINVAL; |
246aa6dd LP |
1305 | } |
1306 | ||
efdb0237 LP |
1307 | if (isempty(e+1)) |
1308 | u = NULL; | |
1309 | else { | |
baa89da4 LP |
1310 | u = strdup(e+1); |
1311 | if (!u) { | |
1312 | free(t); | |
1313 | return -ENOMEM; | |
1314 | } | |
35d2e7ec | 1315 | |
99be45a4 | 1316 | if (!path_is_normalized(u) || |
baa89da4 LP |
1317 | !path_is_absolute(u)) { |
1318 | free(t); | |
1319 | free(u); | |
1320 | return -EINVAL; | |
1321 | } | |
1322 | ||
858d36c1 | 1323 | path_simplify(u, false); |
baa89da4 | 1324 | } |
5954c074 | 1325 | |
35d2e7ec LP |
1326 | if (controller) |
1327 | *controller = t; | |
e884315e LP |
1328 | else |
1329 | free(t); | |
35d2e7ec LP |
1330 | |
1331 | if (path) | |
1332 | *path = u; | |
e884315e LP |
1333 | else |
1334 | free(u); | |
35d2e7ec LP |
1335 | |
1336 | return 0; | |
8c6db833 | 1337 | } |
c6c18be3 | 1338 | |
7027ff61 | 1339 | int cg_mangle_path(const char *path, char **result) { |
78edb35a LP |
1340 | _cleanup_free_ char *c = NULL, *p = NULL; |
1341 | char *t; | |
35d2e7ec LP |
1342 | int r; |
1343 | ||
1344 | assert(path); | |
1345 | assert(result); | |
1346 | ||
73e231ab | 1347 | /* First, check if it already is a filesystem path */ |
7027ff61 | 1348 | if (path_startswith(path, "/sys/fs/cgroup")) { |
35d2e7ec | 1349 | |
b69d29ce LP |
1350 | t = strdup(path); |
1351 | if (!t) | |
35d2e7ec LP |
1352 | return -ENOMEM; |
1353 | ||
858d36c1 | 1354 | *result = path_simplify(t, false); |
35d2e7ec LP |
1355 | return 0; |
1356 | } | |
1357 | ||
73e231ab | 1358 | /* Otherwise, treat it as cg spec */ |
b69d29ce LP |
1359 | r = cg_split_spec(path, &c, &p); |
1360 | if (r < 0) | |
35d2e7ec LP |
1361 | return r; |
1362 | ||
efdb0237 | 1363 | return cg_get_path(c ?: SYSTEMD_CGROUP_CONTROLLER, p ?: "/", NULL, result); |
35d2e7ec | 1364 | } |
1f73f0f1 | 1365 | |
7027ff61 | 1366 | int cg_get_root_path(char **path) { |
9444b1f2 | 1367 | char *p, *e; |
7027ff61 LP |
1368 | int r; |
1369 | ||
1370 | assert(path); | |
1371 | ||
9444b1f2 | 1372 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 1, &p); |
7027ff61 LP |
1373 | if (r < 0) |
1374 | return r; | |
1375 | ||
efdb0237 LP |
1376 | e = endswith(p, "/" SPECIAL_INIT_SCOPE); |
1377 | if (!e) | |
1378 | e = endswith(p, "/" SPECIAL_SYSTEM_SLICE); /* legacy */ | |
1379 | if (!e) | |
1380 | e = endswith(p, "/system"); /* even more legacy */ | |
9444b1f2 | 1381 | if (e) |
7027ff61 LP |
1382 | *e = 0; |
1383 | ||
1f73f0f1 LP |
1384 | *path = p; |
1385 | return 0; | |
1386 | } | |
b59e2465 | 1387 | |
751bc6ac LP |
1388 | int cg_shift_path(const char *cgroup, const char *root, const char **shifted) { |
1389 | _cleanup_free_ char *rt = NULL; | |
1390 | char *p; | |
ba1261bc LP |
1391 | int r; |
1392 | ||
e9174f29 | 1393 | assert(cgroup); |
751bc6ac | 1394 | assert(shifted); |
e9174f29 LP |
1395 | |
1396 | if (!root) { | |
1397 | /* If the root was specified let's use that, otherwise | |
1398 | * let's determine it from PID 1 */ | |
1399 | ||
751bc6ac | 1400 | r = cg_get_root_path(&rt); |
e9174f29 LP |
1401 | if (r < 0) |
1402 | return r; | |
1403 | ||
751bc6ac | 1404 | root = rt; |
e9174f29 | 1405 | } |
ba1261bc | 1406 | |
751bc6ac | 1407 | p = path_startswith(cgroup, root); |
efdb0237 | 1408 | if (p && p > cgroup) |
751bc6ac LP |
1409 | *shifted = p - 1; |
1410 | else | |
1411 | *shifted = cgroup; | |
1412 | ||
1413 | return 0; | |
1414 | } | |
1415 | ||
1416 | int cg_pid_get_path_shifted(pid_t pid, const char *root, char **cgroup) { | |
1417 | _cleanup_free_ char *raw = NULL; | |
1418 | const char *c; | |
1419 | int r; | |
1420 | ||
1421 | assert(pid >= 0); | |
1422 | assert(cgroup); | |
1423 | ||
1424 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &raw); | |
7027ff61 | 1425 | if (r < 0) |
ba1261bc | 1426 | return r; |
ba1261bc | 1427 | |
751bc6ac LP |
1428 | r = cg_shift_path(raw, root, &c); |
1429 | if (r < 0) | |
1430 | return r; | |
ba1261bc | 1431 | |
ae2a15bc LP |
1432 | if (c == raw) |
1433 | *cgroup = TAKE_PTR(raw); | |
1434 | else { | |
751bc6ac | 1435 | char *n; |
ba1261bc | 1436 | |
751bc6ac LP |
1437 | n = strdup(c); |
1438 | if (!n) | |
ba1261bc | 1439 | return -ENOMEM; |
ba1261bc | 1440 | |
751bc6ac LP |
1441 | *cgroup = n; |
1442 | } | |
ba1261bc LP |
1443 | |
1444 | return 0; | |
1445 | } | |
1446 | ||
9ed794a3 | 1447 | int cg_path_decode_unit(const char *cgroup, char **unit) { |
8b0849e9 LP |
1448 | char *c, *s; |
1449 | size_t n; | |
ef1673d1 MT |
1450 | |
1451 | assert(cgroup); | |
6c03089c | 1452 | assert(unit); |
ef1673d1 | 1453 | |
8b0849e9 LP |
1454 | n = strcspn(cgroup, "/"); |
1455 | if (n < 3) | |
1456 | return -ENXIO; | |
1457 | ||
1458 | c = strndupa(cgroup, n); | |
ae018d9b | 1459 | c = cg_unescape(c); |
ef1673d1 | 1460 | |
7410616c | 1461 | if (!unit_name_is_valid(c, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) |
cfeaa44a | 1462 | return -ENXIO; |
ef1673d1 | 1463 | |
d7bd3de0 | 1464 | s = strdup(c); |
6c03089c LP |
1465 | if (!s) |
1466 | return -ENOMEM; | |
1467 | ||
1468 | *unit = s; | |
ef1673d1 MT |
1469 | return 0; |
1470 | } | |
1471 | ||
8b0849e9 LP |
1472 | static bool valid_slice_name(const char *p, size_t n) { |
1473 | ||
1474 | if (!p) | |
1475 | return false; | |
1476 | ||
fbd0b64f | 1477 | if (n < STRLEN("x.slice")) |
8b0849e9 LP |
1478 | return false; |
1479 | ||
1480 | if (memcmp(p + n - 6, ".slice", 6) == 0) { | |
1481 | char buf[n+1], *c; | |
1482 | ||
1483 | memcpy(buf, p, n); | |
1484 | buf[n] = 0; | |
1485 | ||
1486 | c = cg_unescape(buf); | |
1487 | ||
7410616c | 1488 | return unit_name_is_valid(c, UNIT_NAME_PLAIN); |
8b0849e9 LP |
1489 | } |
1490 | ||
1491 | return false; | |
1492 | } | |
1493 | ||
9444b1f2 | 1494 | static const char *skip_slices(const char *p) { |
8b0849e9 LP |
1495 | assert(p); |
1496 | ||
9444b1f2 LP |
1497 | /* Skips over all slice assignments */ |
1498 | ||
1499 | for (;;) { | |
1021b21b LP |
1500 | size_t n; |
1501 | ||
9444b1f2 LP |
1502 | p += strspn(p, "/"); |
1503 | ||
1504 | n = strcspn(p, "/"); | |
8b0849e9 | 1505 | if (!valid_slice_name(p, n)) |
9444b1f2 LP |
1506 | return p; |
1507 | ||
1508 | p += n; | |
1509 | } | |
1510 | } | |
1511 | ||
8b0849e9 | 1512 | int cg_path_get_unit(const char *path, char **ret) { |
6c03089c | 1513 | const char *e; |
8b0849e9 LP |
1514 | char *unit; |
1515 | int r; | |
6c03089c LP |
1516 | |
1517 | assert(path); | |
8b0849e9 | 1518 | assert(ret); |
6c03089c | 1519 | |
9444b1f2 | 1520 | e = skip_slices(path); |
6c03089c | 1521 | |
8b0849e9 LP |
1522 | r = cg_path_decode_unit(e, &unit); |
1523 | if (r < 0) | |
1524 | return r; | |
1525 | ||
1526 | /* We skipped over the slices, don't accept any now */ | |
1527 | if (endswith(unit, ".slice")) { | |
1528 | free(unit); | |
1529 | return -ENXIO; | |
1530 | } | |
1531 | ||
1532 | *ret = unit; | |
1533 | return 0; | |
6c03089c LP |
1534 | } |
1535 | ||
1536 | int cg_pid_get_unit(pid_t pid, char **unit) { | |
7fd1b19b | 1537 | _cleanup_free_ char *cgroup = NULL; |
ba1261bc | 1538 | int r; |
ba1261bc | 1539 | |
ef1673d1 MT |
1540 | assert(unit); |
1541 | ||
7027ff61 | 1542 | r = cg_pid_get_path_shifted(pid, NULL, &cgroup); |
ef1673d1 MT |
1543 | if (r < 0) |
1544 | return r; | |
1545 | ||
6c03089c LP |
1546 | return cg_path_get_unit(cgroup, unit); |
1547 | } | |
ef1673d1 | 1548 | |
d4fffc4b ZJS |
1549 | /** |
1550 | * Skip session-*.scope, but require it to be there. | |
1551 | */ | |
9444b1f2 LP |
1552 | static const char *skip_session(const char *p) { |
1553 | size_t n; | |
1554 | ||
8b0849e9 LP |
1555 | if (isempty(p)) |
1556 | return NULL; | |
9444b1f2 LP |
1557 | |
1558 | p += strspn(p, "/"); | |
1559 | ||
1560 | n = strcspn(p, "/"); | |
fbd0b64f | 1561 | if (n < STRLEN("session-x.scope")) |
d4fffc4b ZJS |
1562 | return NULL; |
1563 | ||
8b0849e9 LP |
1564 | if (memcmp(p, "session-", 8) == 0 && memcmp(p + n - 6, ".scope", 6) == 0) { |
1565 | char buf[n - 8 - 6 + 1]; | |
1566 | ||
1567 | memcpy(buf, p + 8, n - 8 - 6); | |
1568 | buf[n - 8 - 6] = 0; | |
d4fffc4b | 1569 | |
8b0849e9 LP |
1570 | /* Note that session scopes never need unescaping, |
1571 | * since they cannot conflict with the kernel's own | |
1572 | * names, hence we don't need to call cg_unescape() | |
1573 | * here. */ | |
1574 | ||
1575 | if (!session_id_valid(buf)) | |
1576 | return false; | |
1577 | ||
1578 | p += n; | |
1579 | p += strspn(p, "/"); | |
1580 | return p; | |
1581 | } | |
1582 | ||
1583 | return NULL; | |
d4fffc4b ZJS |
1584 | } |
1585 | ||
1586 | /** | |
1587 | * Skip user@*.service, but require it to be there. | |
1588 | */ | |
1589 | static const char *skip_user_manager(const char *p) { | |
1590 | size_t n; | |
1591 | ||
8b0849e9 LP |
1592 | if (isempty(p)) |
1593 | return NULL; | |
d4fffc4b ZJS |
1594 | |
1595 | p += strspn(p, "/"); | |
1596 | ||
1597 | n = strcspn(p, "/"); | |
fbd0b64f | 1598 | if (n < STRLEN("user@x.service")) |
6c03089c | 1599 | return NULL; |
ef1673d1 | 1600 | |
8b0849e9 LP |
1601 | if (memcmp(p, "user@", 5) == 0 && memcmp(p + n - 8, ".service", 8) == 0) { |
1602 | char buf[n - 5 - 8 + 1]; | |
9444b1f2 | 1603 | |
8b0849e9 LP |
1604 | memcpy(buf, p + 5, n - 5 - 8); |
1605 | buf[n - 5 - 8] = 0; | |
1606 | ||
1607 | /* Note that user manager services never need unescaping, | |
1608 | * since they cannot conflict with the kernel's own | |
1609 | * names, hence we don't need to call cg_unescape() | |
1610 | * here. */ | |
1611 | ||
1612 | if (parse_uid(buf, NULL) < 0) | |
1613 | return NULL; | |
1614 | ||
1615 | p += n; | |
1616 | p += strspn(p, "/"); | |
1617 | ||
1618 | return p; | |
1619 | } | |
1620 | ||
1621 | return NULL; | |
9444b1f2 LP |
1622 | } |
1623 | ||
329ac4bc | 1624 | static const char *skip_user_prefix(const char *path) { |
d4fffc4b | 1625 | const char *e, *t; |
ef1673d1 | 1626 | |
6c03089c | 1627 | assert(path); |
ba1261bc | 1628 | |
9444b1f2 LP |
1629 | /* Skip slices, if there are any */ |
1630 | e = skip_slices(path); | |
ba1261bc | 1631 | |
329ac4bc | 1632 | /* Skip the user manager, if it's in the path now... */ |
8b0849e9 | 1633 | t = skip_user_manager(e); |
329ac4bc LP |
1634 | if (t) |
1635 | return t; | |
8b0849e9 | 1636 | |
329ac4bc LP |
1637 | /* Alternatively skip the user session if it is in the path... */ |
1638 | return skip_session(e); | |
1639 | } | |
32081481 | 1640 | |
329ac4bc LP |
1641 | int cg_path_get_user_unit(const char *path, char **ret) { |
1642 | const char *t; | |
6c03089c | 1643 | |
329ac4bc LP |
1644 | assert(path); |
1645 | assert(ret); | |
8b0849e9 | 1646 | |
329ac4bc LP |
1647 | t = skip_user_prefix(path); |
1648 | if (!t) | |
8b0849e9 | 1649 | return -ENXIO; |
8b0849e9 | 1650 | |
329ac4bc LP |
1651 | /* And from here on it looks pretty much the same as for a |
1652 | * system unit, hence let's use the same parser from here | |
1653 | * on. */ | |
1654 | return cg_path_get_unit(t, ret); | |
ef1673d1 | 1655 | } |
ba1261bc | 1656 | |
ef1673d1 | 1657 | int cg_pid_get_user_unit(pid_t pid, char **unit) { |
7fd1b19b | 1658 | _cleanup_free_ char *cgroup = NULL; |
6c03089c LP |
1659 | int r; |
1660 | ||
1661 | assert(unit); | |
1662 | ||
7027ff61 | 1663 | r = cg_pid_get_path_shifted(pid, NULL, &cgroup); |
6c03089c LP |
1664 | if (r < 0) |
1665 | return r; | |
1666 | ||
1667 | return cg_path_get_user_unit(cgroup, unit); | |
ba1261bc | 1668 | } |
e884315e | 1669 | |
7027ff61 | 1670 | int cg_path_get_machine_name(const char *path, char **machine) { |
efdb0237 LP |
1671 | _cleanup_free_ char *u = NULL; |
1672 | const char *sl; | |
89f7c846 | 1673 | int r; |
374ec6ab | 1674 | |
89f7c846 LP |
1675 | r = cg_path_get_unit(path, &u); |
1676 | if (r < 0) | |
1677 | return r; | |
7027ff61 | 1678 | |
efdb0237 | 1679 | sl = strjoina("/run/systemd/machines/unit:", u); |
89f7c846 | 1680 | return readlink_malloc(sl, machine); |
7027ff61 LP |
1681 | } |
1682 | ||
1683 | int cg_pid_get_machine_name(pid_t pid, char **machine) { | |
7fd1b19b | 1684 | _cleanup_free_ char *cgroup = NULL; |
7027ff61 LP |
1685 | int r; |
1686 | ||
1687 | assert(machine); | |
1688 | ||
1689 | r = cg_pid_get_path_shifted(pid, NULL, &cgroup); | |
1690 | if (r < 0) | |
1691 | return r; | |
1692 | ||
1693 | return cg_path_get_machine_name(cgroup, machine); | |
1694 | } | |
1695 | ||
1696 | int cg_path_get_session(const char *path, char **session) { | |
8b0849e9 LP |
1697 | _cleanup_free_ char *unit = NULL; |
1698 | char *start, *end; | |
1699 | int r; | |
7027ff61 LP |
1700 | |
1701 | assert(path); | |
7027ff61 | 1702 | |
8b0849e9 LP |
1703 | r = cg_path_get_unit(path, &unit); |
1704 | if (r < 0) | |
1705 | return r; | |
7027ff61 | 1706 | |
8b0849e9 LP |
1707 | start = startswith(unit, "session-"); |
1708 | if (!start) | |
cfeaa44a | 1709 | return -ENXIO; |
8b0849e9 LP |
1710 | end = endswith(start, ".scope"); |
1711 | if (!end) | |
cfeaa44a | 1712 | return -ENXIO; |
8b0849e9 LP |
1713 | |
1714 | *end = 0; | |
1715 | if (!session_id_valid(start)) | |
cfeaa44a | 1716 | return -ENXIO; |
374ec6ab | 1717 | |
af08d2f9 | 1718 | if (session) { |
8b0849e9 | 1719 | char *rr; |
af08d2f9 | 1720 | |
8b0849e9 LP |
1721 | rr = strdup(start); |
1722 | if (!rr) | |
af08d2f9 LP |
1723 | return -ENOMEM; |
1724 | ||
8b0849e9 | 1725 | *session = rr; |
af08d2f9 | 1726 | } |
7027ff61 | 1727 | |
7027ff61 LP |
1728 | return 0; |
1729 | } | |
1730 | ||
1731 | int cg_pid_get_session(pid_t pid, char **session) { | |
7fd1b19b | 1732 | _cleanup_free_ char *cgroup = NULL; |
7027ff61 LP |
1733 | int r; |
1734 | ||
7027ff61 LP |
1735 | r = cg_pid_get_path_shifted(pid, NULL, &cgroup); |
1736 | if (r < 0) | |
1737 | return r; | |
1738 | ||
1739 | return cg_path_get_session(cgroup, session); | |
1740 | } | |
1741 | ||
ae018d9b | 1742 | int cg_path_get_owner_uid(const char *path, uid_t *uid) { |
374ec6ab | 1743 | _cleanup_free_ char *slice = NULL; |
8b0849e9 | 1744 | char *start, *end; |
374ec6ab | 1745 | int r; |
ae018d9b LP |
1746 | |
1747 | assert(path); | |
ae018d9b | 1748 | |
374ec6ab LP |
1749 | r = cg_path_get_slice(path, &slice); |
1750 | if (r < 0) | |
1751 | return r; | |
ae018d9b | 1752 | |
674eb685 LP |
1753 | start = startswith(slice, "user-"); |
1754 | if (!start) | |
cfeaa44a | 1755 | return -ENXIO; |
8b0849e9 | 1756 | end = endswith(start, ".slice"); |
674eb685 | 1757 | if (!end) |
cfeaa44a | 1758 | return -ENXIO; |
ae018d9b | 1759 | |
8b0849e9 LP |
1760 | *end = 0; |
1761 | if (parse_uid(start, uid) < 0) | |
cfeaa44a | 1762 | return -ENXIO; |
674eb685 | 1763 | |
674eb685 | 1764 | return 0; |
ae018d9b LP |
1765 | } |
1766 | ||
1767 | int cg_pid_get_owner_uid(pid_t pid, uid_t *uid) { | |
1768 | _cleanup_free_ char *cgroup = NULL; | |
1769 | int r; | |
1770 | ||
ae018d9b LP |
1771 | r = cg_pid_get_path_shifted(pid, NULL, &cgroup); |
1772 | if (r < 0) | |
1773 | return r; | |
1774 | ||
1775 | return cg_path_get_owner_uid(cgroup, uid); | |
1776 | } | |
1777 | ||
1021b21b LP |
1778 | int cg_path_get_slice(const char *p, char **slice) { |
1779 | const char *e = NULL; | |
1021b21b LP |
1780 | |
1781 | assert(p); | |
1782 | assert(slice); | |
1783 | ||
329ac4bc LP |
1784 | /* Finds the right-most slice unit from the beginning, but |
1785 | * stops before we come to the first non-slice unit. */ | |
1786 | ||
1021b21b LP |
1787 | for (;;) { |
1788 | size_t n; | |
1789 | ||
1790 | p += strspn(p, "/"); | |
1791 | ||
1792 | n = strcspn(p, "/"); | |
8b0849e9 | 1793 | if (!valid_slice_name(p, n)) { |
1021b21b | 1794 | |
8b0849e9 LP |
1795 | if (!e) { |
1796 | char *s; | |
1021b21b | 1797 | |
e5d855d3 | 1798 | s = strdup(SPECIAL_ROOT_SLICE); |
8b0849e9 LP |
1799 | if (!s) |
1800 | return -ENOMEM; | |
1021b21b | 1801 | |
8b0849e9 LP |
1802 | *slice = s; |
1803 | return 0; | |
1804 | } | |
1805 | ||
1806 | return cg_path_decode_unit(e, slice); | |
1021b21b LP |
1807 | } |
1808 | ||
1809 | e = p; | |
1021b21b LP |
1810 | p += n; |
1811 | } | |
1812 | } | |
1813 | ||
1814 | int cg_pid_get_slice(pid_t pid, char **slice) { | |
1815 | _cleanup_free_ char *cgroup = NULL; | |
1816 | int r; | |
1817 | ||
1818 | assert(slice); | |
1819 | ||
1820 | r = cg_pid_get_path_shifted(pid, NULL, &cgroup); | |
1821 | if (r < 0) | |
1822 | return r; | |
1823 | ||
1824 | return cg_path_get_slice(cgroup, slice); | |
1825 | } | |
1826 | ||
329ac4bc LP |
1827 | int cg_path_get_user_slice(const char *p, char **slice) { |
1828 | const char *t; | |
1829 | assert(p); | |
1830 | assert(slice); | |
1831 | ||
1832 | t = skip_user_prefix(p); | |
1833 | if (!t) | |
1834 | return -ENXIO; | |
1835 | ||
1836 | /* And now it looks pretty much the same as for a system | |
1837 | * slice, so let's just use the same parser from here on. */ | |
1838 | return cg_path_get_slice(t, slice); | |
1839 | } | |
1840 | ||
1841 | int cg_pid_get_user_slice(pid_t pid, char **slice) { | |
1842 | _cleanup_free_ char *cgroup = NULL; | |
1843 | int r; | |
1844 | ||
1845 | assert(slice); | |
1846 | ||
1847 | r = cg_pid_get_path_shifted(pid, NULL, &cgroup); | |
1848 | if (r < 0) | |
1849 | return r; | |
1850 | ||
1851 | return cg_path_get_user_slice(cgroup, slice); | |
1852 | } | |
1853 | ||
ae018d9b LP |
1854 | char *cg_escape(const char *p) { |
1855 | bool need_prefix = false; | |
1856 | ||
1857 | /* This implements very minimal escaping for names to be used | |
1858 | * as file names in the cgroup tree: any name which might | |
1859 | * conflict with a kernel name or is prefixed with '_' is | |
1860 | * prefixed with a '_'. That way, when reading cgroup names it | |
1861 | * is sufficient to remove a single prefixing underscore if | |
1862 | * there is one. */ | |
1863 | ||
1864 | /* The return value of this function (unlike cg_unescape()) | |
1865 | * needs free()! */ | |
1866 | ||
4c701096 | 1867 | if (IN_SET(p[0], 0, '_', '.') || |
0cbd293e | 1868 | STR_IN_SET(p, "notify_on_release", "release_agent", "tasks") || |
efdb0237 | 1869 | startswith(p, "cgroup.")) |
ae018d9b LP |
1870 | need_prefix = true; |
1871 | else { | |
1872 | const char *dot; | |
1873 | ||
1874 | dot = strrchr(p, '.'); | |
1875 | if (dot) { | |
efdb0237 LP |
1876 | CGroupController c; |
1877 | size_t l = dot - p; | |
ae018d9b | 1878 | |
efdb0237 LP |
1879 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { |
1880 | const char *n; | |
1881 | ||
1882 | n = cgroup_controller_to_string(c); | |
ae018d9b | 1883 | |
efdb0237 LP |
1884 | if (l != strlen(n)) |
1885 | continue; | |
ae018d9b | 1886 | |
efdb0237 LP |
1887 | if (memcmp(p, n, l) != 0) |
1888 | continue; | |
1889 | ||
1890 | need_prefix = true; | |
1891 | break; | |
ae018d9b LP |
1892 | } |
1893 | } | |
1894 | } | |
1895 | ||
1896 | if (need_prefix) | |
1897 | return strappend("_", p); | |
efdb0237 LP |
1898 | |
1899 | return strdup(p); | |
ae018d9b LP |
1900 | } |
1901 | ||
1902 | char *cg_unescape(const char *p) { | |
1903 | assert(p); | |
1904 | ||
1905 | /* The return value of this function (unlike cg_escape()) | |
1906 | * doesn't need free()! */ | |
1907 | ||
1908 | if (p[0] == '_') | |
1909 | return (char*) p+1; | |
1910 | ||
1911 | return (char*) p; | |
1912 | } | |
78edb35a LP |
1913 | |
1914 | #define CONTROLLER_VALID \ | |
4b549144 | 1915 | DIGITS LETTERS \ |
78edb35a LP |
1916 | "_" |
1917 | ||
185a0874 | 1918 | bool cg_controller_is_valid(const char *p) { |
78edb35a LP |
1919 | const char *t, *s; |
1920 | ||
1921 | if (!p) | |
1922 | return false; | |
1923 | ||
b6629c4b TH |
1924 | if (streq(p, SYSTEMD_CGROUP_CONTROLLER)) |
1925 | return true; | |
1926 | ||
185a0874 DJL |
1927 | s = startswith(p, "name="); |
1928 | if (s) | |
1929 | p = s; | |
78edb35a | 1930 | |
4c701096 | 1931 | if (IN_SET(*p, 0, '_')) |
78edb35a LP |
1932 | return false; |
1933 | ||
1934 | for (t = p; *t; t++) | |
1935 | if (!strchr(CONTROLLER_VALID, *t)) | |
1936 | return false; | |
1937 | ||
1938 | if (t - p > FILENAME_MAX) | |
1939 | return false; | |
1940 | ||
1941 | return true; | |
1942 | } | |
a016b922 LP |
1943 | |
1944 | int cg_slice_to_path(const char *unit, char **ret) { | |
1945 | _cleanup_free_ char *p = NULL, *s = NULL, *e = NULL; | |
1946 | const char *dash; | |
7410616c | 1947 | int r; |
a016b922 LP |
1948 | |
1949 | assert(unit); | |
1950 | assert(ret); | |
1951 | ||
e5d855d3 | 1952 | if (streq(unit, SPECIAL_ROOT_SLICE)) { |
c96cc582 LP |
1953 | char *x; |
1954 | ||
1955 | x = strdup(""); | |
1956 | if (!x) | |
1957 | return -ENOMEM; | |
1958 | *ret = x; | |
1959 | return 0; | |
1960 | } | |
1961 | ||
7410616c | 1962 | if (!unit_name_is_valid(unit, UNIT_NAME_PLAIN)) |
a016b922 LP |
1963 | return -EINVAL; |
1964 | ||
1965 | if (!endswith(unit, ".slice")) | |
1966 | return -EINVAL; | |
1967 | ||
7410616c LP |
1968 | r = unit_name_to_prefix(unit, &p); |
1969 | if (r < 0) | |
1970 | return r; | |
a016b922 LP |
1971 | |
1972 | dash = strchr(p, '-'); | |
e66e5b61 LP |
1973 | |
1974 | /* Don't allow initial dashes */ | |
1975 | if (dash == p) | |
1976 | return -EINVAL; | |
1977 | ||
a016b922 LP |
1978 | while (dash) { |
1979 | _cleanup_free_ char *escaped = NULL; | |
1980 | char n[dash - p + sizeof(".slice")]; | |
1981 | ||
989290db | 1982 | #if HAS_FEATURE_MEMORY_SANITIZER |
1c56d501 | 1983 | /* msan doesn't instrument stpncpy, so it thinks |
5238e957 | 1984 | * n is later used uninitialized: |
1c56d501 ZJS |
1985 | * https://github.com/google/sanitizers/issues/926 |
1986 | */ | |
1987 | zero(n); | |
1988 | #endif | |
1989 | ||
e66e5b61 | 1990 | /* Don't allow trailing or double dashes */ |
4c701096 | 1991 | if (IN_SET(dash[1], 0, '-')) |
c96cc582 | 1992 | return -EINVAL; |
a016b922 | 1993 | |
c96cc582 | 1994 | strcpy(stpncpy(n, p, dash - p), ".slice"); |
7410616c | 1995 | if (!unit_name_is_valid(n, UNIT_NAME_PLAIN)) |
a016b922 LP |
1996 | return -EINVAL; |
1997 | ||
1998 | escaped = cg_escape(n); | |
1999 | if (!escaped) | |
2000 | return -ENOMEM; | |
2001 | ||
2002 | if (!strextend(&s, escaped, "/", NULL)) | |
2003 | return -ENOMEM; | |
2004 | ||
2005 | dash = strchr(dash+1, '-'); | |
2006 | } | |
2007 | ||
2008 | e = cg_escape(unit); | |
2009 | if (!e) | |
2010 | return -ENOMEM; | |
2011 | ||
2012 | if (!strextend(&s, e, NULL)) | |
2013 | return -ENOMEM; | |
2014 | ||
ae2a15bc | 2015 | *ret = TAKE_PTR(s); |
a016b922 LP |
2016 | |
2017 | return 0; | |
2018 | } | |
4ad49000 LP |
2019 | |
2020 | int cg_set_attribute(const char *controller, const char *path, const char *attribute, const char *value) { | |
2021 | _cleanup_free_ char *p = NULL; | |
2022 | int r; | |
2023 | ||
2024 | r = cg_get_path(controller, path, attribute, &p); | |
2025 | if (r < 0) | |
2026 | return r; | |
2027 | ||
604028de | 2028 | return write_string_file(p, value, WRITE_STRING_FILE_DISABLE_BUFFER); |
4ad49000 LP |
2029 | } |
2030 | ||
934277fe LP |
2031 | int cg_get_attribute(const char *controller, const char *path, const char *attribute, char **ret) { |
2032 | _cleanup_free_ char *p = NULL; | |
2033 | int r; | |
2034 | ||
2035 | r = cg_get_path(controller, path, attribute, &p); | |
2036 | if (r < 0) | |
2037 | return r; | |
2038 | ||
2039 | return read_one_line_file(p, ret); | |
2040 | } | |
2041 | ||
b734a4ff LP |
2042 | int cg_get_keyed_attribute( |
2043 | const char *controller, | |
2044 | const char *path, | |
2045 | const char *attribute, | |
2046 | char **keys, | |
2047 | char **ret_values) { | |
66ebf6c0 | 2048 | |
b734a4ff | 2049 | _cleanup_free_ char *filename = NULL, *contents = NULL; |
b734a4ff | 2050 | const char *p; |
9177fa9f | 2051 | size_t n, i, n_done = 0; |
b734a4ff LP |
2052 | char **v; |
2053 | int r; | |
2054 | ||
4e1dfa45 | 2055 | /* Reads one or more fields of a cgroup v2 keyed attribute file. The 'keys' parameter should be an strv with |
b734a4ff LP |
2056 | * all keys to retrieve. The 'ret_values' parameter should be passed as string size with the same number of |
2057 | * entries as 'keys'. On success each entry will be set to the value of the matching key. | |
2058 | * | |
2059 | * If the attribute file doesn't exist at all returns ENOENT, if any key is not found returns ENXIO. */ | |
66ebf6c0 TH |
2060 | |
2061 | r = cg_get_path(controller, path, attribute, &filename); | |
2062 | if (r < 0) | |
2063 | return r; | |
2064 | ||
b734a4ff | 2065 | r = read_full_file(filename, &contents, NULL); |
66ebf6c0 TH |
2066 | if (r < 0) |
2067 | return r; | |
2068 | ||
b734a4ff LP |
2069 | n = strv_length(keys); |
2070 | if (n == 0) /* No keys to retrieve? That's easy, we are done then */ | |
2071 | return 0; | |
66ebf6c0 | 2072 | |
b734a4ff LP |
2073 | /* Let's build this up in a temporary array for now in order not to clobber the return parameter on failure */ |
2074 | v = newa0(char*, n); | |
66ebf6c0 | 2075 | |
b734a4ff LP |
2076 | for (p = contents; *p;) { |
2077 | const char *w = NULL; | |
b734a4ff | 2078 | |
9177fa9f ZJS |
2079 | for (i = 0; i < n; i++) |
2080 | if (!v[i]) { | |
b734a4ff LP |
2081 | w = first_word(p, keys[i]); |
2082 | if (w) | |
2083 | break; | |
66ebf6c0 | 2084 | } |
66ebf6c0 | 2085 | |
b734a4ff | 2086 | if (w) { |
b734a4ff LP |
2087 | size_t l; |
2088 | ||
2089 | l = strcspn(w, NEWLINE); | |
9177fa9f ZJS |
2090 | v[i] = strndup(w, l); |
2091 | if (!v[i]) { | |
b734a4ff LP |
2092 | r = -ENOMEM; |
2093 | goto fail; | |
66ebf6c0 | 2094 | } |
b734a4ff | 2095 | |
b734a4ff | 2096 | n_done++; |
b734a4ff LP |
2097 | if (n_done >= n) |
2098 | goto done; | |
2099 | ||
2100 | p = w + l; | |
9177fa9f | 2101 | } else |
b734a4ff | 2102 | p += strcspn(p, NEWLINE); |
b734a4ff LP |
2103 | |
2104 | p += strspn(p, NEWLINE); | |
66ebf6c0 TH |
2105 | } |
2106 | ||
b734a4ff LP |
2107 | r = -ENXIO; |
2108 | ||
2109 | fail: | |
2110 | for (i = 0; i < n; i++) | |
2111 | free(v[i]); | |
2112 | ||
2113 | return r; | |
2114 | ||
2115 | done: | |
2116 | memcpy(ret_values, v, sizeof(char*) * n); | |
66ebf6c0 | 2117 | return 0; |
b734a4ff | 2118 | |
66ebf6c0 TH |
2119 | } |
2120 | ||
efdb0237 LP |
2121 | int cg_create_everywhere(CGroupMask supported, CGroupMask mask, const char *path) { |
2122 | CGroupController c; | |
e353faa0 | 2123 | CGroupMask done; |
65be7e06 | 2124 | bool created; |
415fc41c | 2125 | int r; |
4ad49000 LP |
2126 | |
2127 | /* This one will create a cgroup in our private tree, but also | |
2128 | * duplicate it in the trees specified in mask, and remove it | |
65be7e06 ZJS |
2129 | * in all others. |
2130 | * | |
2131 | * Returns 0 if the group already existed in the systemd hierarchy, | |
2132 | * 1 on success, negative otherwise. | |
2133 | */ | |
4ad49000 LP |
2134 | |
2135 | /* First create the cgroup in our own hierarchy. */ | |
2136 | r = cg_create(SYSTEMD_CGROUP_CONTROLLER, path); | |
2137 | if (r < 0) | |
2138 | return r; | |
490c5a37 | 2139 | created = r; |
4ad49000 | 2140 | |
efdb0237 | 2141 | /* If we are in the unified hierarchy, we are done now */ |
b4cccbc1 LP |
2142 | r = cg_all_unified(); |
2143 | if (r < 0) | |
2144 | return r; | |
2145 | if (r > 0) | |
65be7e06 | 2146 | return created; |
efdb0237 | 2147 | |
e353faa0 LP |
2148 | supported &= CGROUP_MASK_V1; |
2149 | mask = CGROUP_MASK_EXTEND_JOINED(mask); | |
2150 | done = 0; | |
2151 | ||
efdb0237 LP |
2152 | /* Otherwise, do the same in the other hierarchies */ |
2153 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { | |
2154 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
2155 | const char *n; | |
2156 | ||
e353faa0 | 2157 | if (!FLAGS_SET(supported, bit)) |
ab275f23 LP |
2158 | continue; |
2159 | ||
e353faa0 LP |
2160 | if (FLAGS_SET(done, bit)) |
2161 | continue; | |
efdb0237 | 2162 | |
e353faa0 | 2163 | n = cgroup_controller_to_string(c); |
f99850a0 | 2164 | if (FLAGS_SET(mask, bit)) |
efdb0237 | 2165 | (void) cg_create(n, path); |
e353faa0 | 2166 | else |
efdb0237 | 2167 | (void) cg_trim(n, path, true); |
e353faa0 LP |
2168 | |
2169 | done |= CGROUP_MASK_EXTEND_JOINED(bit); | |
4ad49000 LP |
2170 | } |
2171 | ||
65be7e06 | 2172 | return created; |
4ad49000 LP |
2173 | } |
2174 | ||
efdb0237 LP |
2175 | int cg_attach_everywhere(CGroupMask supported, const char *path, pid_t pid, cg_migrate_callback_t path_callback, void *userdata) { |
2176 | CGroupController c; | |
e353faa0 | 2177 | CGroupMask done; |
415fc41c | 2178 | int r; |
4ad49000 LP |
2179 | |
2180 | r = cg_attach(SYSTEMD_CGROUP_CONTROLLER, path, pid); | |
13b84ec7 LP |
2181 | if (r < 0) |
2182 | return r; | |
4ad49000 | 2183 | |
b4cccbc1 LP |
2184 | r = cg_all_unified(); |
2185 | if (r < 0) | |
2186 | return r; | |
2187 | if (r > 0) | |
efdb0237 | 2188 | return 0; |
7b3fd631 | 2189 | |
e353faa0 LP |
2190 | supported &= CGROUP_MASK_V1; |
2191 | done = 0; | |
2192 | ||
efdb0237 LP |
2193 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { |
2194 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
2195 | const char *p = NULL; | |
7b3fd631 | 2196 | |
e353faa0 | 2197 | if (!FLAGS_SET(supported, bit)) |
ab275f23 LP |
2198 | continue; |
2199 | ||
e353faa0 | 2200 | if (FLAGS_SET(done, bit)) |
efdb0237 | 2201 | continue; |
7b3fd631 | 2202 | |
efdb0237 LP |
2203 | if (path_callback) |
2204 | p = path_callback(bit, userdata); | |
efdb0237 LP |
2205 | if (!p) |
2206 | p = path; | |
4ad49000 | 2207 | |
efdb0237 | 2208 | (void) cg_attach_fallback(cgroup_controller_to_string(c), p, pid); |
e353faa0 | 2209 | done |= CGROUP_MASK_EXTEND_JOINED(bit); |
4ad49000 LP |
2210 | } |
2211 | ||
13b84ec7 | 2212 | return 0; |
4ad49000 LP |
2213 | } |
2214 | ||
efdb0237 | 2215 | int cg_attach_many_everywhere(CGroupMask supported, const char *path, Set* pids, cg_migrate_callback_t path_callback, void *userdata) { |
6c12b52e LP |
2216 | Iterator i; |
2217 | void *pidp; | |
2218 | int r = 0; | |
2219 | ||
2220 | SET_FOREACH(pidp, pids, i) { | |
fea72cc0 | 2221 | pid_t pid = PTR_TO_PID(pidp); |
13b84ec7 | 2222 | int q; |
6c12b52e | 2223 | |
7b3fd631 | 2224 | q = cg_attach_everywhere(supported, path, pid, path_callback, userdata); |
efdb0237 | 2225 | if (q < 0 && r >= 0) |
13b84ec7 | 2226 | r = q; |
6c12b52e LP |
2227 | } |
2228 | ||
2229 | return r; | |
2230 | } | |
2231 | ||
efdb0237 | 2232 | int cg_migrate_everywhere(CGroupMask supported, const char *from, const char *to, cg_migrate_callback_t to_callback, void *userdata) { |
b3c5bad3 | 2233 | CGroupController c; |
e353faa0 | 2234 | CGroupMask done; |
b4cccbc1 | 2235 | int r = 0, q; |
4ad49000 | 2236 | |
13b84ec7 | 2237 | if (!path_equal(from, to)) { |
1d98fef1 | 2238 | r = cg_migrate_recursive(SYSTEMD_CGROUP_CONTROLLER, from, SYSTEMD_CGROUP_CONTROLLER, to, CGROUP_REMOVE); |
13b84ec7 LP |
2239 | if (r < 0) |
2240 | return r; | |
2241 | } | |
4ad49000 | 2242 | |
b4cccbc1 LP |
2243 | q = cg_all_unified(); |
2244 | if (q < 0) | |
2245 | return q; | |
2246 | if (q > 0) | |
efdb0237 | 2247 | return r; |
03b90d4b | 2248 | |
e353faa0 LP |
2249 | supported &= CGROUP_MASK_V1; |
2250 | done = 0; | |
2251 | ||
efdb0237 LP |
2252 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { |
2253 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
2254 | const char *p = NULL; | |
03b90d4b | 2255 | |
e353faa0 | 2256 | if (!FLAGS_SET(supported, bit)) |
ab275f23 LP |
2257 | continue; |
2258 | ||
e353faa0 | 2259 | if (FLAGS_SET(done, bit)) |
efdb0237 | 2260 | continue; |
03b90d4b | 2261 | |
efdb0237 LP |
2262 | if (to_callback) |
2263 | p = to_callback(bit, userdata); | |
efdb0237 LP |
2264 | if (!p) |
2265 | p = to; | |
2266 | ||
1d98fef1 | 2267 | (void) cg_migrate_recursive_fallback(SYSTEMD_CGROUP_CONTROLLER, to, cgroup_controller_to_string(c), p, 0); |
e353faa0 | 2268 | done |= CGROUP_MASK_EXTEND_JOINED(bit); |
4ad49000 LP |
2269 | } |
2270 | ||
e353faa0 | 2271 | return r; |
4ad49000 LP |
2272 | } |
2273 | ||
efdb0237 LP |
2274 | int cg_trim_everywhere(CGroupMask supported, const char *path, bool delete_root) { |
2275 | CGroupController c; | |
e353faa0 | 2276 | CGroupMask done; |
b4cccbc1 | 2277 | int r, q; |
4ad49000 LP |
2278 | |
2279 | r = cg_trim(SYSTEMD_CGROUP_CONTROLLER, path, delete_root); | |
2280 | if (r < 0) | |
2281 | return r; | |
2282 | ||
b4cccbc1 LP |
2283 | q = cg_all_unified(); |
2284 | if (q < 0) | |
2285 | return q; | |
2286 | if (q > 0) | |
efdb0237 LP |
2287 | return r; |
2288 | ||
e353faa0 LP |
2289 | supported &= CGROUP_MASK_V1; |
2290 | done = 0; | |
2291 | ||
efdb0237 LP |
2292 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { |
2293 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
2294 | ||
e353faa0 | 2295 | if (!FLAGS_SET(supported, bit)) |
ab275f23 LP |
2296 | continue; |
2297 | ||
e353faa0 | 2298 | if (FLAGS_SET(done, bit)) |
efdb0237 | 2299 | continue; |
4ad49000 | 2300 | |
efdb0237 | 2301 | (void) cg_trim(cgroup_controller_to_string(c), path, delete_root); |
e353faa0 | 2302 | done |= CGROUP_MASK_EXTEND_JOINED(bit); |
4ad49000 LP |
2303 | } |
2304 | ||
e353faa0 | 2305 | return r; |
4ad49000 LP |
2306 | } |
2307 | ||
aae7e17f | 2308 | int cg_mask_to_string(CGroupMask mask, char **ret) { |
ec635a2d LP |
2309 | _cleanup_free_ char *s = NULL; |
2310 | size_t n = 0, allocated = 0; | |
2311 | bool space = false; | |
aae7e17f | 2312 | CGroupController c; |
aae7e17f FB |
2313 | |
2314 | assert(ret); | |
2315 | ||
2316 | if (mask == 0) { | |
2317 | *ret = NULL; | |
2318 | return 0; | |
2319 | } | |
2320 | ||
2321 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { | |
ec635a2d LP |
2322 | const char *k; |
2323 | size_t l; | |
aae7e17f | 2324 | |
f99850a0 | 2325 | if (!FLAGS_SET(mask, CGROUP_CONTROLLER_TO_MASK(c))) |
aae7e17f FB |
2326 | continue; |
2327 | ||
ec635a2d LP |
2328 | k = cgroup_controller_to_string(c); |
2329 | l = strlen(k); | |
2330 | ||
2331 | if (!GREEDY_REALLOC(s, allocated, n + space + l + 1)) | |
2332 | return -ENOMEM; | |
2333 | ||
2334 | if (space) | |
2335 | s[n] = ' '; | |
2336 | memcpy(s + n + space, k, l); | |
2337 | n += space + l; | |
2338 | ||
2339 | space = true; | |
aae7e17f FB |
2340 | } |
2341 | ||
ec635a2d | 2342 | assert(s); |
aae7e17f | 2343 | |
ec635a2d | 2344 | s[n] = 0; |
ae2a15bc | 2345 | *ret = TAKE_PTR(s); |
ec635a2d | 2346 | |
aae7e17f FB |
2347 | return 0; |
2348 | } | |
2349 | ||
38a90d45 LP |
2350 | int cg_mask_from_string(const char *value, CGroupMask *ret) { |
2351 | CGroupMask m = 0; | |
2352 | ||
2353 | assert(ret); | |
aae7e17f FB |
2354 | assert(value); |
2355 | ||
2356 | for (;;) { | |
2357 | _cleanup_free_ char *n = NULL; | |
2358 | CGroupController v; | |
2359 | int r; | |
2360 | ||
2361 | r = extract_first_word(&value, &n, NULL, 0); | |
2362 | if (r < 0) | |
2363 | return r; | |
2364 | if (r == 0) | |
2365 | break; | |
2366 | ||
2367 | v = cgroup_controller_from_string(n); | |
2368 | if (v < 0) | |
2369 | continue; | |
2370 | ||
38a90d45 | 2371 | m |= CGROUP_CONTROLLER_TO_MASK(v); |
aae7e17f | 2372 | } |
38a90d45 LP |
2373 | |
2374 | *ret = m; | |
aae7e17f FB |
2375 | return 0; |
2376 | } | |
2377 | ||
efdb0237 | 2378 | int cg_mask_supported(CGroupMask *ret) { |
38a90d45 | 2379 | CGroupMask mask; |
415fc41c | 2380 | int r; |
efdb0237 | 2381 | |
67558d15 LP |
2382 | /* Determines the mask of supported cgroup controllers. Only includes controllers we can make sense of and that |
2383 | * are actually accessible. Only covers real controllers, i.e. not the CGROUP_CONTROLLER_BPF_xyz | |
2384 | * pseudo-controllers. */ | |
4ad49000 | 2385 | |
b4cccbc1 LP |
2386 | r = cg_all_unified(); |
2387 | if (r < 0) | |
2388 | return r; | |
2389 | if (r > 0) { | |
5f4c5fef | 2390 | _cleanup_free_ char *root = NULL, *controllers = NULL, *path = NULL; |
efdb0237 LP |
2391 | |
2392 | /* In the unified hierarchy we can read the supported | |
2393 | * and accessible controllers from a the top-level | |
2394 | * cgroup attribute */ | |
2395 | ||
5f4c5fef LP |
2396 | r = cg_get_root_path(&root); |
2397 | if (r < 0) | |
2398 | return r; | |
2399 | ||
2400 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, root, "cgroup.controllers", &path); | |
2401 | if (r < 0) | |
2402 | return r; | |
2403 | ||
2404 | r = read_one_line_file(path, &controllers); | |
efdb0237 LP |
2405 | if (r < 0) |
2406 | return r; | |
4ad49000 | 2407 | |
aae7e17f FB |
2408 | r = cg_mask_from_string(controllers, &mask); |
2409 | if (r < 0) | |
2410 | return r; | |
efdb0237 | 2411 | |
03afd780 | 2412 | /* Currently, we support the cpu, memory, io and pids controller in the unified hierarchy, mask |
03a7b521 | 2413 | * everything else off. */ |
03afd780 | 2414 | mask &= CGROUP_MASK_V2; |
efdb0237 LP |
2415 | |
2416 | } else { | |
2417 | CGroupController c; | |
2418 | ||
03afd780 | 2419 | /* In the legacy hierarchy, we check which hierarchies are mounted. */ |
efdb0237 | 2420 | |
38a90d45 | 2421 | mask = 0; |
efdb0237 | 2422 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { |
03afd780 | 2423 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); |
efdb0237 LP |
2424 | const char *n; |
2425 | ||
03afd780 LP |
2426 | if (!FLAGS_SET(CGROUP_MASK_V1, bit)) |
2427 | continue; | |
2428 | ||
efdb0237 LP |
2429 | n = cgroup_controller_to_string(c); |
2430 | if (controller_is_accessible(n) >= 0) | |
03afd780 | 2431 | mask |= bit; |
efdb0237 | 2432 | } |
4ad49000 LP |
2433 | } |
2434 | ||
efdb0237 LP |
2435 | *ret = mask; |
2436 | return 0; | |
4ad49000 | 2437 | } |
b12afc8c | 2438 | |
6925a0de LP |
2439 | int cg_kernel_controllers(Set **ret) { |
2440 | _cleanup_set_free_free_ Set *controllers = NULL; | |
b12afc8c | 2441 | _cleanup_fclose_ FILE *f = NULL; |
b12afc8c LP |
2442 | int r; |
2443 | ||
6925a0de | 2444 | assert(ret); |
b12afc8c | 2445 | |
f09e86bc LS |
2446 | /* Determines the full list of kernel-known controllers. Might include controllers we don't actually support |
2447 | * and controllers that aren't currently accessible (because not mounted). This does not include "name=" | |
2448 | * pseudo-controllers. */ | |
e155a0aa | 2449 | |
6925a0de LP |
2450 | controllers = set_new(&string_hash_ops); |
2451 | if (!controllers) | |
2452 | return -ENOMEM; | |
2453 | ||
fdeea3f4 ZJS |
2454 | r = fopen_unlocked("/proc/cgroups", "re", &f); |
2455 | if (r == -ENOENT) { | |
2456 | *ret = NULL; | |
2457 | return 0; | |
b12afc8c | 2458 | } |
fdeea3f4 ZJS |
2459 | if (r < 0) |
2460 | return r; | |
35bbbf85 | 2461 | |
b12afc8c | 2462 | /* Ignore the header line */ |
2351e44d | 2463 | (void) read_line(f, (size_t) -1, NULL); |
b12afc8c LP |
2464 | |
2465 | for (;;) { | |
2466 | char *controller; | |
2467 | int enabled = 0; | |
2468 | ||
2469 | errno = 0; | |
2470 | if (fscanf(f, "%ms %*i %*i %i", &controller, &enabled) != 2) { | |
2471 | ||
2472 | if (feof(f)) | |
2473 | break; | |
2474 | ||
b3267152 | 2475 | if (ferror(f) && errno > 0) |
b12afc8c LP |
2476 | return -errno; |
2477 | ||
2478 | return -EBADMSG; | |
2479 | } | |
2480 | ||
2481 | if (!enabled) { | |
2482 | free(controller); | |
2483 | continue; | |
2484 | } | |
2485 | ||
efdb0237 | 2486 | if (!cg_controller_is_valid(controller)) { |
b12afc8c LP |
2487 | free(controller); |
2488 | return -EBADMSG; | |
2489 | } | |
2490 | ||
2491 | r = set_consume(controllers, controller); | |
2492 | if (r < 0) | |
2493 | return r; | |
2494 | } | |
2495 | ||
1cc6c93a | 2496 | *ret = TAKE_PTR(controllers); |
6925a0de | 2497 | |
b12afc8c LP |
2498 | return 0; |
2499 | } | |
efdb0237 | 2500 | |
5da38d07 TH |
2501 | static thread_local CGroupUnified unified_cache = CGROUP_UNIFIED_UNKNOWN; |
2502 | ||
4e1dfa45 | 2503 | /* The hybrid mode was initially implemented in v232 and simply mounted cgroup2 on /sys/fs/cgroup/systemd. This |
c22800e4 LP |
2504 | * unfortunately broke other tools (such as docker) which expected the v1 "name=systemd" hierarchy on |
2505 | * /sys/fs/cgroup/systemd. From v233 and on, the hybrid mode mountnbs v2 on /sys/fs/cgroup/unified and maintains | |
2506 | * "name=systemd" hierarchy on /sys/fs/cgroup/systemd for compatibility with other tools. | |
f08e9287 | 2507 | * |
c22800e4 LP |
2508 | * To keep live upgrade working, we detect and support v232 layout. When v232 layout is detected, to keep cgroup v2 |
2509 | * process management but disable the compat dual layout, we return %true on | |
2510 | * cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) and %false on cg_hybrid_unified(). | |
f08e9287 TH |
2511 | */ |
2512 | static thread_local bool unified_systemd_v232; | |
2513 | ||
1fcca10e | 2514 | static int cg_unified_update(void) { |
efdb0237 | 2515 | |
efdb0237 LP |
2516 | struct statfs fs; |
2517 | ||
2518 | /* Checks if we support the unified hierarchy. Returns an | |
2519 | * error when the cgroup hierarchies aren't mounted yet or we | |
2520 | * have any other trouble determining if the unified hierarchy | |
2521 | * is supported. */ | |
2522 | ||
5da38d07 TH |
2523 | if (unified_cache >= CGROUP_UNIFIED_NONE) |
2524 | return 0; | |
efdb0237 LP |
2525 | |
2526 | if (statfs("/sys/fs/cgroup/", &fs) < 0) | |
c028bed1 | 2527 | return log_debug_errno(errno, "statfs(\"/sys/fs/cgroup/\") failed: %m"); |
efdb0237 | 2528 | |
9aa21133 ZJS |
2529 | if (F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) { |
2530 | log_debug("Found cgroup2 on /sys/fs/cgroup/, full unified hierarchy"); | |
5da38d07 | 2531 | unified_cache = CGROUP_UNIFIED_ALL; |
9aa21133 | 2532 | } else if (F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC)) { |
2977724b | 2533 | if (statfs("/sys/fs/cgroup/unified/", &fs) == 0 && |
f08e9287 | 2534 | F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) { |
9aa21133 | 2535 | log_debug("Found cgroup2 on /sys/fs/cgroup/unified, unified hierarchy for systemd controller"); |
2977724b | 2536 | unified_cache = CGROUP_UNIFIED_SYSTEMD; |
f08e9287 | 2537 | unified_systemd_v232 = false; |
f08e9287 | 2538 | } else { |
2977724b | 2539 | if (statfs("/sys/fs/cgroup/systemd/", &fs) < 0) |
9aa21133 | 2540 | return log_debug_errno(errno, "statfs(\"/sys/fs/cgroup/systemd\" failed: %m"); |
5535d8f7 EV |
2541 | |
2542 | if (F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) { | |
2543 | log_debug("Found cgroup2 on /sys/fs/cgroup/systemd, unified hierarchy for systemd controller (v232 variant)"); | |
2544 | unified_cache = CGROUP_UNIFIED_SYSTEMD; | |
2545 | unified_systemd_v232 = true; | |
2546 | } else if (F_TYPE_EQUAL(fs.f_type, CGROUP_SUPER_MAGIC)) { | |
2547 | log_debug("Found cgroup on /sys/fs/cgroup/systemd, legacy hierarchy"); | |
2548 | unified_cache = CGROUP_UNIFIED_NONE; | |
2549 | } else { | |
2550 | log_debug("Unexpected filesystem type %llx mounted on /sys/fs/cgroup/systemd, assuming legacy hierarchy", | |
9aa21133 | 2551 | (unsigned long long) fs.f_type); |
5535d8f7 | 2552 | unified_cache = CGROUP_UNIFIED_NONE; |
9aa21133 | 2553 | } |
2977724b | 2554 | } |
baaa35ad ZJS |
2555 | } else |
2556 | return log_debug_errno(SYNTHETIC_ERRNO(ENOMEDIUM), | |
2557 | "Unknown filesystem type %llx mounted on /sys/fs/cgroup.", | |
2558 | (unsigned long long)fs.f_type); | |
efdb0237 | 2559 | |
5da38d07 TH |
2560 | return 0; |
2561 | } | |
2562 | ||
c22800e4 | 2563 | int cg_unified_controller(const char *controller) { |
b4cccbc1 | 2564 | int r; |
5da38d07 | 2565 | |
1fcca10e | 2566 | r = cg_unified_update(); |
b4cccbc1 LP |
2567 | if (r < 0) |
2568 | return r; | |
5da38d07 | 2569 | |
fc9ae717 LP |
2570 | if (unified_cache == CGROUP_UNIFIED_NONE) |
2571 | return false; | |
2572 | ||
2573 | if (unified_cache >= CGROUP_UNIFIED_ALL) | |
2574 | return true; | |
2575 | ||
2576 | return streq_ptr(controller, SYSTEMD_CGROUP_CONTROLLER); | |
5da38d07 TH |
2577 | } |
2578 | ||
b4cccbc1 | 2579 | int cg_all_unified(void) { |
4bb652ac LP |
2580 | int r; |
2581 | ||
2582 | r = cg_unified_update(); | |
2583 | if (r < 0) | |
2584 | return r; | |
2585 | ||
2586 | return unified_cache >= CGROUP_UNIFIED_ALL; | |
efdb0237 LP |
2587 | } |
2588 | ||
b4cccbc1 LP |
2589 | int cg_hybrid_unified(void) { |
2590 | int r; | |
2977724b | 2591 | |
1fcca10e | 2592 | r = cg_unified_update(); |
b4cccbc1 LP |
2593 | if (r < 0) |
2594 | return r; | |
2977724b | 2595 | |
f08e9287 | 2596 | return unified_cache == CGROUP_UNIFIED_SYSTEMD && !unified_systemd_v232; |
2977724b TH |
2597 | } |
2598 | ||
415fc41c | 2599 | int cg_unified_flush(void) { |
5da38d07 | 2600 | unified_cache = CGROUP_UNIFIED_UNKNOWN; |
415fc41c | 2601 | |
1fcca10e | 2602 | return cg_unified_update(); |
efdb0237 LP |
2603 | } |
2604 | ||
27adcc97 LP |
2605 | int cg_enable_everywhere( |
2606 | CGroupMask supported, | |
2607 | CGroupMask mask, | |
2608 | const char *p, | |
2609 | CGroupMask *ret_result_mask) { | |
2610 | ||
77fa610b | 2611 | _cleanup_fclose_ FILE *f = NULL; |
efdb0237 LP |
2612 | _cleanup_free_ char *fs = NULL; |
2613 | CGroupController c; | |
27adcc97 | 2614 | CGroupMask ret = 0; |
415fc41c | 2615 | int r; |
efdb0237 LP |
2616 | |
2617 | assert(p); | |
2618 | ||
27adcc97 LP |
2619 | if (supported == 0) { |
2620 | if (ret_result_mask) | |
2621 | *ret_result_mask = 0; | |
efdb0237 | 2622 | return 0; |
27adcc97 | 2623 | } |
efdb0237 | 2624 | |
b4cccbc1 LP |
2625 | r = cg_all_unified(); |
2626 | if (r < 0) | |
2627 | return r; | |
27adcc97 | 2628 | if (r == 0) { |
5238e957 | 2629 | /* On the legacy hierarchy there's no concept of "enabling" controllers in cgroups defined. Let's claim |
27adcc97 LP |
2630 | * complete success right away. (If you wonder why we return the full mask here, rather than zero: the |
2631 | * caller tends to use the returned mask later on to compare if all controllers where properly joined, | |
2632 | * and if not requeues realization. This use is the primary purpose of the return value, hence let's | |
2633 | * minimize surprises here and reduce triggers for re-realization by always saying we fully | |
2634 | * succeeded.) */ | |
2635 | if (ret_result_mask) | |
2636 | *ret_result_mask = mask & supported & CGROUP_MASK_V2; /* If you wonder why we mask this with | |
2637 | * CGROUP_MASK_V2: The 'supported' mask | |
2638 | * might contain pure-V1 or BPF | |
2639 | * controllers, and we never want to | |
2640 | * claim that we could enable those with | |
2641 | * cgroup.subtree_control */ | |
efdb0237 | 2642 | return 0; |
27adcc97 | 2643 | } |
efdb0237 LP |
2644 | |
2645 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, p, "cgroup.subtree_control", &fs); | |
2646 | if (r < 0) | |
2647 | return r; | |
2648 | ||
2649 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { | |
2650 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
2651 | const char *n; | |
2652 | ||
ab275f23 LP |
2653 | if (!FLAGS_SET(CGROUP_MASK_V2, bit)) |
2654 | continue; | |
2655 | ||
f99850a0 | 2656 | if (!FLAGS_SET(supported, bit)) |
efdb0237 LP |
2657 | continue; |
2658 | ||
2659 | n = cgroup_controller_to_string(c); | |
2660 | { | |
2661 | char s[1 + strlen(n) + 1]; | |
2662 | ||
f99850a0 | 2663 | s[0] = FLAGS_SET(mask, bit) ? '+' : '-'; |
efdb0237 LP |
2664 | strcpy(s + 1, n); |
2665 | ||
77fa610b LP |
2666 | if (!f) { |
2667 | f = fopen(fs, "we"); | |
54b5ba1d LP |
2668 | if (!f) |
2669 | return log_debug_errno(errno, "Failed to open cgroup.subtree_control file of %s: %m", p); | |
77fa610b LP |
2670 | } |
2671 | ||
604028de | 2672 | r = write_string_stream(f, s, WRITE_STRING_FILE_DISABLE_BUFFER); |
96aa6591 | 2673 | if (r < 0) { |
94f344fb LP |
2674 | log_debug_errno(r, "Failed to %s controller %s for %s (%s): %m", |
2675 | FLAGS_SET(mask, bit) ? "enable" : "disable", n, p, fs); | |
96aa6591 | 2676 | clearerr(f); |
27adcc97 LP |
2677 | |
2678 | /* If we can't turn off a controller, leave it on in the reported resulting mask. This | |
2679 | * happens for example when we attempt to turn off a controller up in the tree that is | |
2680 | * used down in the tree. */ | |
2681 | if (!FLAGS_SET(mask, bit) && r == -EBUSY) /* You might wonder why we check for EBUSY | |
2682 | * only here, and not follow the same logic | |
2683 | * for other errors such as EINVAL or | |
2684 | * EOPNOTSUPP or anything else. That's | |
2685 | * because EBUSY indicates that the | |
2686 | * controllers is currently enabled and | |
2687 | * cannot be disabled because something down | |
2688 | * the hierarchy is still using it. Any other | |
2689 | * error most likely means something like "I | |
2690 | * never heard of this controller" or | |
2691 | * similar. In the former case it's hence | |
2692 | * safe to assume the controller is still on | |
2693 | * after the failed operation, while in the | |
2694 | * latter case it's safer to assume the | |
2695 | * controller is unknown and hence certainly | |
2696 | * not enabled. */ | |
2697 | ret |= bit; | |
2698 | } else { | |
2699 | /* Otherwise, if we managed to turn on a controller, set the bit reflecting that. */ | |
2700 | if (FLAGS_SET(mask, bit)) | |
2701 | ret |= bit; | |
96aa6591 | 2702 | } |
efdb0237 LP |
2703 | } |
2704 | } | |
2705 | ||
27adcc97 LP |
2706 | /* Let's return the precise set of controllers now enabled for the cgroup. */ |
2707 | if (ret_result_mask) | |
2708 | *ret_result_mask = ret; | |
2709 | ||
efdb0237 LP |
2710 | return 0; |
2711 | } | |
2712 | ||
2713 | bool cg_is_unified_wanted(void) { | |
2714 | static thread_local int wanted = -1; | |
415fc41c | 2715 | int r; |
1d84ad94 | 2716 | bool b; |
77fab2a9 | 2717 | const bool is_default = DEFAULT_HIERARCHY == CGROUP_UNIFIED_ALL; |
5f086dc7 | 2718 | _cleanup_free_ char *c = NULL; |
efdb0237 | 2719 | |
77fab2a9 | 2720 | /* If we have a cached value, return that. */ |
efdb0237 LP |
2721 | if (wanted >= 0) |
2722 | return wanted; | |
2723 | ||
239a3d09 ZJS |
2724 | /* If the hierarchy is already mounted, then follow whatever |
2725 | * was chosen for it. */ | |
2726 | if (cg_unified_flush() >= 0) | |
b4cccbc1 | 2727 | return (wanted = unified_cache >= CGROUP_UNIFIED_ALL); |
239a3d09 | 2728 | |
5f086dc7 CD |
2729 | /* If we were explicitly passed systemd.unified_cgroup_hierarchy, |
2730 | * respect that. */ | |
1d84ad94 | 2731 | r = proc_cmdline_get_bool("systemd.unified_cgroup_hierarchy", &b); |
5f086dc7 CD |
2732 | if (r > 0) |
2733 | return (wanted = b); | |
2734 | ||
2735 | /* If we passed cgroup_no_v1=all with no other instructions, it seems | |
2736 | * highly unlikely that we want to use hybrid or legacy hierarchy. */ | |
2737 | r = proc_cmdline_get_key("cgroup_no_v1", 0, &c); | |
2738 | if (r > 0 && streq_ptr(c, "all")) | |
2739 | return (wanted = true); | |
efdb0237 | 2740 | |
5f086dc7 | 2741 | return (wanted = is_default); |
efdb0237 LP |
2742 | } |
2743 | ||
2744 | bool cg_is_legacy_wanted(void) { | |
239a3d09 ZJS |
2745 | static thread_local int wanted = -1; |
2746 | ||
2747 | /* If we have a cached value, return that. */ | |
2748 | if (wanted >= 0) | |
2749 | return wanted; | |
2750 | ||
4e1dfa45 | 2751 | /* Check if we have cgroup v2 already mounted. */ |
1b59cf04 ZJS |
2752 | if (cg_unified_flush() >= 0 && |
2753 | unified_cache == CGROUP_UNIFIED_ALL) | |
239a3d09 | 2754 | return (wanted = false); |
1b59cf04 ZJS |
2755 | |
2756 | /* Otherwise, assume that at least partial legacy is wanted, | |
4e1dfa45 | 2757 | * since cgroup v2 should already be mounted at this point. */ |
239a3d09 | 2758 | return (wanted = true); |
efdb0237 LP |
2759 | } |
2760 | ||
a4464b95 | 2761 | bool cg_is_hybrid_wanted(void) { |
5da38d07 | 2762 | static thread_local int wanted = -1; |
415fc41c | 2763 | int r; |
1d84ad94 | 2764 | bool b; |
c19739db ZJS |
2765 | const bool is_default = DEFAULT_HIERARCHY >= CGROUP_UNIFIED_SYSTEMD; |
2766 | /* We default to true if the default is "hybrid", obviously, | |
2767 | * but also when the default is "unified", because if we get | |
2768 | * called, it means that unified hierarchy was not mounted. */ | |
5da38d07 | 2769 | |
77fab2a9 | 2770 | /* If we have a cached value, return that. */ |
5da38d07 TH |
2771 | if (wanted >= 0) |
2772 | return wanted; | |
2773 | ||
239a3d09 ZJS |
2774 | /* If the hierarchy is already mounted, then follow whatever |
2775 | * was chosen for it. */ | |
2776 | if (cg_unified_flush() >= 0 && | |
2777 | unified_cache == CGROUP_UNIFIED_ALL) | |
2778 | return (wanted = false); | |
2779 | ||
77fab2a9 ZJS |
2780 | /* Otherwise, let's see what the kernel command line has to say. |
2781 | * Since checking is expensive, cache a non-error result. */ | |
1d84ad94 | 2782 | r = proc_cmdline_get_bool("systemd.legacy_systemd_cgroup_controller", &b); |
5da38d07 | 2783 | |
2dcb526d ZJS |
2784 | /* The meaning of the kernel option is reversed wrt. to the return value |
2785 | * of this function, hence the negation. */ | |
77fab2a9 | 2786 | return (wanted = r > 0 ? !b : is_default); |
5da38d07 TH |
2787 | } |
2788 | ||
13c31542 TH |
2789 | int cg_weight_parse(const char *s, uint64_t *ret) { |
2790 | uint64_t u; | |
2791 | int r; | |
2792 | ||
2793 | if (isempty(s)) { | |
2794 | *ret = CGROUP_WEIGHT_INVALID; | |
2795 | return 0; | |
2796 | } | |
2797 | ||
2798 | r = safe_atou64(s, &u); | |
2799 | if (r < 0) | |
2800 | return r; | |
2801 | ||
2802 | if (u < CGROUP_WEIGHT_MIN || u > CGROUP_WEIGHT_MAX) | |
2803 | return -ERANGE; | |
2804 | ||
2805 | *ret = u; | |
2806 | return 0; | |
2807 | } | |
2808 | ||
9be57249 TH |
2809 | const uint64_t cgroup_io_limit_defaults[_CGROUP_IO_LIMIT_TYPE_MAX] = { |
2810 | [CGROUP_IO_RBPS_MAX] = CGROUP_LIMIT_MAX, | |
2811 | [CGROUP_IO_WBPS_MAX] = CGROUP_LIMIT_MAX, | |
ac06a0cf TH |
2812 | [CGROUP_IO_RIOPS_MAX] = CGROUP_LIMIT_MAX, |
2813 | [CGROUP_IO_WIOPS_MAX] = CGROUP_LIMIT_MAX, | |
9be57249 TH |
2814 | }; |
2815 | ||
2816 | static const char* const cgroup_io_limit_type_table[_CGROUP_IO_LIMIT_TYPE_MAX] = { | |
2817 | [CGROUP_IO_RBPS_MAX] = "IOReadBandwidthMax", | |
2818 | [CGROUP_IO_WBPS_MAX] = "IOWriteBandwidthMax", | |
ac06a0cf TH |
2819 | [CGROUP_IO_RIOPS_MAX] = "IOReadIOPSMax", |
2820 | [CGROUP_IO_WIOPS_MAX] = "IOWriteIOPSMax", | |
9be57249 TH |
2821 | }; |
2822 | ||
2823 | DEFINE_STRING_TABLE_LOOKUP(cgroup_io_limit_type, CGroupIOLimitType); | |
2824 | ||
d53d9474 LP |
2825 | int cg_cpu_shares_parse(const char *s, uint64_t *ret) { |
2826 | uint64_t u; | |
2827 | int r; | |
2828 | ||
2829 | if (isempty(s)) { | |
2830 | *ret = CGROUP_CPU_SHARES_INVALID; | |
2831 | return 0; | |
2832 | } | |
2833 | ||
2834 | r = safe_atou64(s, &u); | |
2835 | if (r < 0) | |
2836 | return r; | |
2837 | ||
2838 | if (u < CGROUP_CPU_SHARES_MIN || u > CGROUP_CPU_SHARES_MAX) | |
2839 | return -ERANGE; | |
2840 | ||
2841 | *ret = u; | |
2842 | return 0; | |
2843 | } | |
2844 | ||
2845 | int cg_blkio_weight_parse(const char *s, uint64_t *ret) { | |
2846 | uint64_t u; | |
2847 | int r; | |
2848 | ||
2849 | if (isempty(s)) { | |
2850 | *ret = CGROUP_BLKIO_WEIGHT_INVALID; | |
2851 | return 0; | |
2852 | } | |
2853 | ||
2854 | r = safe_atou64(s, &u); | |
2855 | if (r < 0) | |
2856 | return r; | |
2857 | ||
2858 | if (u < CGROUP_BLKIO_WEIGHT_MIN || u > CGROUP_BLKIO_WEIGHT_MAX) | |
2859 | return -ERANGE; | |
2860 | ||
2861 | *ret = u; | |
2862 | return 0; | |
2863 | } | |
2864 | ||
f0bef277 EV |
2865 | bool is_cgroup_fs(const struct statfs *s) { |
2866 | return is_fs_type(s, CGROUP_SUPER_MAGIC) || | |
2867 | is_fs_type(s, CGROUP2_SUPER_MAGIC); | |
2868 | } | |
2869 | ||
2870 | bool fd_is_cgroup_fs(int fd) { | |
2871 | struct statfs s; | |
2872 | ||
2873 | if (fstatfs(fd, &s) < 0) | |
2874 | return -errno; | |
2875 | ||
2876 | return is_cgroup_fs(&s); | |
2877 | } | |
2878 | ||
b82f71c7 | 2879 | static const char *const cgroup_controller_table[_CGROUP_CONTROLLER_MAX] = { |
efdb0237 LP |
2880 | [CGROUP_CONTROLLER_CPU] = "cpu", |
2881 | [CGROUP_CONTROLLER_CPUACCT] = "cpuacct", | |
13c31542 | 2882 | [CGROUP_CONTROLLER_IO] = "io", |
efdb0237 LP |
2883 | [CGROUP_CONTROLLER_BLKIO] = "blkio", |
2884 | [CGROUP_CONTROLLER_MEMORY] = "memory", | |
3905f127 | 2885 | [CGROUP_CONTROLLER_DEVICES] = "devices", |
03a7b521 | 2886 | [CGROUP_CONTROLLER_PIDS] = "pids", |
17f14955 | 2887 | [CGROUP_CONTROLLER_BPF_FIREWALL] = "bpf-firewall", |
084c7007 | 2888 | [CGROUP_CONTROLLER_BPF_DEVICES] = "bpf-devices", |
efdb0237 LP |
2889 | }; |
2890 | ||
2891 | DEFINE_STRING_TABLE_LOOKUP(cgroup_controller, CGroupController); | |
f98c2585 CD |
2892 | |
2893 | CGroupMask get_cpu_accounting_mask(void) { | |
2894 | static CGroupMask needed_mask = (CGroupMask) -1; | |
2895 | ||
2896 | /* On kernel ≥4.15 with unified hierarchy, cpu.stat's usage_usec is | |
2897 | * provided externally from the CPU controller, which means we don't | |
2898 | * need to enable the CPU controller just to get metrics. This is good, | |
2899 | * because enabling the CPU controller comes at a minor performance | |
2900 | * hit, especially when it's propagated deep into large hierarchies. | |
2901 | * There's also no separate CPU accounting controller available within | |
2902 | * a unified hierarchy. | |
2903 | * | |
2904 | * This combination of factors results in the desired cgroup mask to | |
2905 | * enable for CPU accounting varying as follows: | |
2906 | * | |
2907 | * ╔═════════════════════╤═════════════════════╗ | |
2908 | * ║ Linux ≥4.15 │ Linux <4.15 ║ | |
2909 | * ╔═══════════════╬═════════════════════╪═════════════════════╣ | |
2910 | * ║ Unified ║ nothing │ CGROUP_MASK_CPU ║ | |
2911 | * ╟───────────────╫─────────────────────┼─────────────────────╢ | |
2912 | * ║ Hybrid/Legacy ║ CGROUP_MASK_CPUACCT │ CGROUP_MASK_CPUACCT ║ | |
2913 | * ╚═══════════════╩═════════════════════╧═════════════════════╝ | |
2914 | * | |
2915 | * We check kernel version here instead of manually checking whether | |
2916 | * cpu.stat is present for every cgroup, as that check in itself would | |
2917 | * already be fairly expensive. | |
2918 | * | |
2919 | * Kernels where this patch has been backported will therefore have the | |
2920 | * CPU controller enabled unnecessarily. This is more expensive than | |
2921 | * necessary, but harmless. ☺️ | |
2922 | */ | |
2923 | ||
2924 | if (needed_mask == (CGroupMask) -1) { | |
2925 | if (cg_all_unified()) { | |
2926 | struct utsname u; | |
2927 | assert_se(uname(&u) >= 0); | |
2928 | ||
2929 | if (str_verscmp(u.release, "4.15") < 0) | |
2930 | needed_mask = CGROUP_MASK_CPU; | |
2931 | else | |
2932 | needed_mask = 0; | |
2933 | } else | |
2934 | needed_mask = CGROUP_MASK_CPUACCT; | |
2935 | } | |
2936 | ||
2937 | return needed_mask; | |
2938 | } | |
2939 | ||
2940 | bool cpu_accounting_is_cheap(void) { | |
2941 | return get_cpu_accounting_mask() == 0; | |
2942 | } |