Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
0a268dbd TH |
2 | #include "cgroup-internal.h" |
3 | ||
1592c9b2 | 4 | #include <linux/ctype.h> |
0a268dbd TH |
5 | #include <linux/kmod.h> |
6 | #include <linux/sort.h> | |
1592c9b2 | 7 | #include <linux/delay.h> |
0a268dbd | 8 | #include <linux/mm.h> |
c3edc401 | 9 | #include <linux/sched/signal.h> |
56cd6973 | 10 | #include <linux/sched/task.h> |
50ff9d13 | 11 | #include <linux/magic.h> |
0a268dbd TH |
12 | #include <linux/slab.h> |
13 | #include <linux/vmalloc.h> | |
14 | #include <linux/delayacct.h> | |
15 | #include <linux/pid_namespace.h> | |
16 | #include <linux/cgroupstats.h> | |
8d2451f4 | 17 | #include <linux/fs_parser.h> |
0a268dbd TH |
18 | |
19 | #include <trace/events/cgroup.h> | |
20 | ||
06a2ae56 | 21 | #define cg_invalf(fc, fmt, ...) invalf(fc, fmt, ## __VA_ARGS__) |
8d2451f4 | 22 | |
0a268dbd TH |
23 | /* |
24 | * pidlists linger the following amount before being destroyed. The goal | |
25 | * is avoiding frequent destruction in the middle of consecutive read calls | |
26 | * Expiring in the middle is a performance problem not a correctness one. | |
27 | * 1 sec should be enough. | |
28 | */ | |
29 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
30 | ||
31 | /* Controllers blocked by the commandline in v1 */ | |
32 | static u16 cgroup_no_v1_mask; | |
33 | ||
3fc9c12d TH |
34 | /* disable named v1 mounts */ |
35 | static bool cgroup_no_v1_named; | |
36 | ||
0a268dbd TH |
37 | /* |
38 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
39 | * separate workqueue as flush domain. | |
40 | */ | |
41 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
42 | ||
43 | /* | |
44 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
45 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
46 | */ | |
1592c9b2 | 47 | static DEFINE_SPINLOCK(release_agent_path_lock); |
0a268dbd | 48 | |
d62beb7f | 49 | bool cgroup1_ssid_disabled(int ssid) |
0a268dbd TH |
50 | { |
51 | return cgroup_no_v1_mask & (1 << ssid); | |
52 | } | |
53 | ||
54 | /** | |
55 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
56 | * @from: attach to all cgroups of a given task | |
57 | * @tsk: the task to be attached | |
58 | */ | |
59 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
60 | { | |
61 | struct cgroup_root *root; | |
62 | int retval = 0; | |
63 | ||
64 | mutex_lock(&cgroup_mutex); | |
65 | percpu_down_write(&cgroup_threadgroup_rwsem); | |
66 | for_each_root(root) { | |
67 | struct cgroup *from_cgrp; | |
68 | ||
69 | if (root == &cgrp_dfl_root) | |
70 | continue; | |
71 | ||
72 | spin_lock_irq(&css_set_lock); | |
73 | from_cgrp = task_cgroup_from_root(from, root); | |
74 | spin_unlock_irq(&css_set_lock); | |
75 | ||
76 | retval = cgroup_attach_task(from_cgrp, tsk, false); | |
77 | if (retval) | |
78 | break; | |
79 | } | |
80 | percpu_up_write(&cgroup_threadgroup_rwsem); | |
81 | mutex_unlock(&cgroup_mutex); | |
82 | ||
83 | return retval; | |
84 | } | |
85 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
86 | ||
87 | /** | |
88 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another | |
89 | * @to: cgroup to which the tasks will be moved | |
90 | * @from: cgroup in which the tasks currently reside | |
91 | * | |
92 | * Locking rules between cgroup_post_fork() and the migration path | |
93 | * guarantee that, if a task is forking while being migrated, the new child | |
94 | * is guaranteed to be either visible in the source cgroup after the | |
95 | * parent's migration is complete or put into the target cgroup. No task | |
96 | * can slip out of migration through forking. | |
97 | */ | |
98 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) | |
99 | { | |
e595cd70 | 100 | DEFINE_CGROUP_MGCTX(mgctx); |
0a268dbd TH |
101 | struct cgrp_cset_link *link; |
102 | struct css_task_iter it; | |
103 | struct task_struct *task; | |
104 | int ret; | |
105 | ||
106 | if (cgroup_on_dfl(to)) | |
107 | return -EINVAL; | |
108 | ||
8cfd8147 TH |
109 | ret = cgroup_migrate_vet_dst(to); |
110 | if (ret) | |
111 | return ret; | |
0a268dbd TH |
112 | |
113 | mutex_lock(&cgroup_mutex); | |
114 | ||
115 | percpu_down_write(&cgroup_threadgroup_rwsem); | |
116 | ||
117 | /* all tasks in @from are being moved, all csets are source */ | |
118 | spin_lock_irq(&css_set_lock); | |
119 | list_for_each_entry(link, &from->cset_links, cset_link) | |
e595cd70 | 120 | cgroup_migrate_add_src(link->cset, to, &mgctx); |
0a268dbd TH |
121 | spin_unlock_irq(&css_set_lock); |
122 | ||
e595cd70 | 123 | ret = cgroup_migrate_prepare_dst(&mgctx); |
0a268dbd TH |
124 | if (ret) |
125 | goto out_err; | |
126 | ||
127 | /* | |
128 | * Migrate tasks one-by-one until @from is empty. This fails iff | |
129 | * ->can_attach() fails. | |
130 | */ | |
131 | do { | |
bc2fb7ed | 132 | css_task_iter_start(&from->self, 0, &it); |
116d2f74 PS |
133 | |
134 | do { | |
135 | task = css_task_iter_next(&it); | |
136 | } while (task && (task->flags & PF_EXITING)); | |
137 | ||
0a268dbd TH |
138 | if (task) |
139 | get_task_struct(task); | |
140 | css_task_iter_end(&it); | |
141 | ||
142 | if (task) { | |
bfc2cf6f | 143 | ret = cgroup_migrate(task, false, &mgctx); |
0a268dbd | 144 | if (!ret) |
e4f8d81c | 145 | TRACE_CGROUP_PATH(transfer_tasks, to, task, false); |
0a268dbd TH |
146 | put_task_struct(task); |
147 | } | |
148 | } while (task && !ret); | |
149 | out_err: | |
e595cd70 | 150 | cgroup_migrate_finish(&mgctx); |
0a268dbd TH |
151 | percpu_up_write(&cgroup_threadgroup_rwsem); |
152 | mutex_unlock(&cgroup_mutex); | |
153 | return ret; | |
154 | } | |
155 | ||
156 | /* | |
157 | * Stuff for reading the 'tasks'/'procs' files. | |
158 | * | |
159 | * Reading this file can return large amounts of data if a cgroup has | |
160 | * *lots* of attached tasks. So it may need several calls to read(), | |
161 | * but we cannot guarantee that the information we produce is correct | |
162 | * unless we produce it entirely atomically. | |
163 | * | |
164 | */ | |
165 | ||
166 | /* which pidlist file are we talking about? */ | |
167 | enum cgroup_filetype { | |
168 | CGROUP_FILE_PROCS, | |
169 | CGROUP_FILE_TASKS, | |
170 | }; | |
171 | ||
172 | /* | |
173 | * A pidlist is a list of pids that virtually represents the contents of one | |
174 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
175 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
176 | * to the cgroup. | |
177 | */ | |
178 | struct cgroup_pidlist { | |
179 | /* | |
180 | * used to find which pidlist is wanted. doesn't change as long as | |
181 | * this particular list stays in the list. | |
182 | */ | |
183 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
184 | /* array of xids */ | |
185 | pid_t *list; | |
186 | /* how many elements the above list has */ | |
187 | int length; | |
188 | /* each of these stored in a list by its cgroup */ | |
189 | struct list_head links; | |
190 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
191 | struct cgroup *owner; | |
192 | /* for delayed destruction */ | |
193 | struct delayed_work destroy_dwork; | |
194 | }; | |
195 | ||
0a268dbd TH |
196 | /* |
197 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
198 | * should be left afterwards. | |
199 | */ | |
d62beb7f | 200 | void cgroup1_pidlist_destroy_all(struct cgroup *cgrp) |
0a268dbd TH |
201 | { |
202 | struct cgroup_pidlist *l, *tmp_l; | |
203 | ||
204 | mutex_lock(&cgrp->pidlist_mutex); | |
205 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
206 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
207 | mutex_unlock(&cgrp->pidlist_mutex); | |
208 | ||
209 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
210 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
211 | } | |
212 | ||
213 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
214 | { | |
215 | struct delayed_work *dwork = to_delayed_work(work); | |
216 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
217 | destroy_dwork); | |
218 | struct cgroup_pidlist *tofree = NULL; | |
219 | ||
220 | mutex_lock(&l->owner->pidlist_mutex); | |
221 | ||
222 | /* | |
223 | * Destroy iff we didn't get queued again. The state won't change | |
224 | * as destroy_dwork can only be queued while locked. | |
225 | */ | |
226 | if (!delayed_work_pending(dwork)) { | |
227 | list_del(&l->links); | |
653a23ca | 228 | kvfree(l->list); |
0a268dbd TH |
229 | put_pid_ns(l->key.ns); |
230 | tofree = l; | |
231 | } | |
232 | ||
233 | mutex_unlock(&l->owner->pidlist_mutex); | |
234 | kfree(tofree); | |
235 | } | |
236 | ||
237 | /* | |
238 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries | |
239 | * Returns the number of unique elements. | |
240 | */ | |
241 | static int pidlist_uniq(pid_t *list, int length) | |
242 | { | |
243 | int src, dest = 1; | |
244 | ||
245 | /* | |
246 | * we presume the 0th element is unique, so i starts at 1. trivial | |
247 | * edge cases first; no work needs to be done for either | |
248 | */ | |
249 | if (length == 0 || length == 1) | |
250 | return length; | |
251 | /* src and dest walk down the list; dest counts unique elements */ | |
252 | for (src = 1; src < length; src++) { | |
253 | /* find next unique element */ | |
254 | while (list[src] == list[src-1]) { | |
255 | src++; | |
256 | if (src == length) | |
257 | goto after; | |
258 | } | |
259 | /* dest always points to where the next unique element goes */ | |
260 | list[dest] = list[src]; | |
261 | dest++; | |
262 | } | |
263 | after: | |
264 | return dest; | |
265 | } | |
266 | ||
267 | /* | |
268 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
269 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
270 | * different per namespace, each namespace needs differently sorted list, | |
271 | * making it impossible to use, for example, single rbtree of member tasks | |
272 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
273 | * per open file is dangerous, so cgroup had to implement shared pool of | |
274 | * pidlists keyed by cgroup and namespace. | |
275 | */ | |
276 | static int cmppid(const void *a, const void *b) | |
277 | { | |
278 | return *(pid_t *)a - *(pid_t *)b; | |
279 | } | |
280 | ||
281 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, | |
282 | enum cgroup_filetype type) | |
283 | { | |
284 | struct cgroup_pidlist *l; | |
285 | /* don't need task_nsproxy() if we're looking at ourself */ | |
286 | struct pid_namespace *ns = task_active_pid_ns(current); | |
287 | ||
288 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
289 | ||
290 | list_for_each_entry(l, &cgrp->pidlists, links) | |
291 | if (l->key.type == type && l->key.ns == ns) | |
292 | return l; | |
293 | return NULL; | |
294 | } | |
295 | ||
296 | /* | |
297 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
298 | * returns with the lock on that pidlist already held, and takes care | |
299 | * of the use count, or returns NULL with no locks held if we're out of | |
300 | * memory. | |
301 | */ | |
302 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, | |
303 | enum cgroup_filetype type) | |
304 | { | |
305 | struct cgroup_pidlist *l; | |
306 | ||
307 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
308 | ||
309 | l = cgroup_pidlist_find(cgrp, type); | |
310 | if (l) | |
311 | return l; | |
312 | ||
313 | /* entry not found; create a new one */ | |
314 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); | |
315 | if (!l) | |
316 | return l; | |
317 | ||
318 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); | |
319 | l->key.type = type; | |
320 | /* don't need task_nsproxy() if we're looking at ourself */ | |
321 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
322 | l->owner = cgrp; | |
323 | list_add(&l->links, &cgrp->pidlists); | |
324 | return l; | |
325 | } | |
326 | ||
0a268dbd TH |
327 | /* |
328 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
329 | */ | |
330 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, | |
331 | struct cgroup_pidlist **lp) | |
332 | { | |
333 | pid_t *array; | |
334 | int length; | |
335 | int pid, n = 0; /* used for populating the array */ | |
336 | struct css_task_iter it; | |
337 | struct task_struct *tsk; | |
338 | struct cgroup_pidlist *l; | |
339 | ||
340 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
341 | ||
342 | /* | |
343 | * If cgroup gets more users after we read count, we won't have | |
344 | * enough space - tough. This race is indistinguishable to the | |
345 | * caller from the case that the additional cgroup users didn't | |
346 | * show up until sometime later on. | |
347 | */ | |
348 | length = cgroup_task_count(cgrp); | |
653a23ca | 349 | array = kvmalloc_array(length, sizeof(pid_t), GFP_KERNEL); |
0a268dbd TH |
350 | if (!array) |
351 | return -ENOMEM; | |
352 | /* now, populate the array */ | |
bc2fb7ed | 353 | css_task_iter_start(&cgrp->self, 0, &it); |
0a268dbd TH |
354 | while ((tsk = css_task_iter_next(&it))) { |
355 | if (unlikely(n == length)) | |
356 | break; | |
357 | /* get tgid or pid for procs or tasks file respectively */ | |
358 | if (type == CGROUP_FILE_PROCS) | |
359 | pid = task_tgid_vnr(tsk); | |
360 | else | |
361 | pid = task_pid_vnr(tsk); | |
362 | if (pid > 0) /* make sure to only use valid results */ | |
363 | array[n++] = pid; | |
364 | } | |
365 | css_task_iter_end(&it); | |
366 | length = n; | |
367 | /* now sort & (if procs) strip out duplicates */ | |
368 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
369 | if (type == CGROUP_FILE_PROCS) | |
370 | length = pidlist_uniq(array, length); | |
371 | ||
372 | l = cgroup_pidlist_find_create(cgrp, type); | |
373 | if (!l) { | |
653a23ca | 374 | kvfree(array); |
0a268dbd TH |
375 | return -ENOMEM; |
376 | } | |
377 | ||
378 | /* store array, freeing old if necessary */ | |
653a23ca | 379 | kvfree(l->list); |
0a268dbd TH |
380 | l->list = array; |
381 | l->length = length; | |
382 | *lp = l; | |
383 | return 0; | |
384 | } | |
385 | ||
386 | /* | |
387 | * seq_file methods for the tasks/procs files. The seq_file position is the | |
388 | * next pid to display; the seq_file iterator is a pointer to the pid | |
389 | * in the cgroup->l->list array. | |
390 | */ | |
391 | ||
392 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) | |
393 | { | |
394 | /* | |
395 | * Initially we receive a position value that corresponds to | |
396 | * one more than the last pid shown (or 0 on the first call or | |
397 | * after a seek to the start). Use a binary-search to find the | |
398 | * next pid to display, if any | |
399 | */ | |
400 | struct kernfs_open_file *of = s->private; | |
401 | struct cgroup *cgrp = seq_css(s)->cgroup; | |
402 | struct cgroup_pidlist *l; | |
403 | enum cgroup_filetype type = seq_cft(s)->private; | |
404 | int index = 0, pid = *pos; | |
405 | int *iter, ret; | |
406 | ||
407 | mutex_lock(&cgrp->pidlist_mutex); | |
408 | ||
409 | /* | |
410 | * !NULL @of->priv indicates that this isn't the first start() | |
411 | * after open. If the matching pidlist is around, we can use that. | |
412 | * Look for it. Note that @of->priv can't be used directly. It | |
413 | * could already have been destroyed. | |
414 | */ | |
415 | if (of->priv) | |
416 | of->priv = cgroup_pidlist_find(cgrp, type); | |
417 | ||
418 | /* | |
419 | * Either this is the first start() after open or the matching | |
420 | * pidlist has been destroyed inbetween. Create a new one. | |
421 | */ | |
422 | if (!of->priv) { | |
423 | ret = pidlist_array_load(cgrp, type, | |
424 | (struct cgroup_pidlist **)&of->priv); | |
425 | if (ret) | |
426 | return ERR_PTR(ret); | |
427 | } | |
428 | l = of->priv; | |
429 | ||
430 | if (pid) { | |
431 | int end = l->length; | |
432 | ||
433 | while (index < end) { | |
434 | int mid = (index + end) / 2; | |
435 | if (l->list[mid] == pid) { | |
436 | index = mid; | |
437 | break; | |
438 | } else if (l->list[mid] <= pid) | |
439 | index = mid + 1; | |
440 | else | |
441 | end = mid; | |
442 | } | |
443 | } | |
444 | /* If we're off the end of the array, we're done */ | |
445 | if (index >= l->length) | |
446 | return NULL; | |
447 | /* Update the abstract position to be the actual pid that we found */ | |
448 | iter = l->list + index; | |
449 | *pos = *iter; | |
450 | return iter; | |
451 | } | |
452 | ||
453 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) | |
454 | { | |
455 | struct kernfs_open_file *of = s->private; | |
456 | struct cgroup_pidlist *l = of->priv; | |
457 | ||
458 | if (l) | |
459 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
460 | CGROUP_PIDLIST_DESTROY_DELAY); | |
461 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); | |
462 | } | |
463 | ||
464 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) | |
465 | { | |
466 | struct kernfs_open_file *of = s->private; | |
467 | struct cgroup_pidlist *l = of->priv; | |
468 | pid_t *p = v; | |
469 | pid_t *end = l->list + l->length; | |
470 | /* | |
471 | * Advance to the next pid in the array. If this goes off the | |
472 | * end, we're done | |
473 | */ | |
474 | p++; | |
475 | if (p >= end) { | |
476 | return NULL; | |
477 | } else { | |
478 | *pos = *p; | |
479 | return p; | |
480 | } | |
481 | } | |
482 | ||
483 | static int cgroup_pidlist_show(struct seq_file *s, void *v) | |
484 | { | |
485 | seq_printf(s, "%d\n", *(int *)v); | |
486 | ||
487 | return 0; | |
488 | } | |
489 | ||
715c809d TH |
490 | static ssize_t __cgroup1_procs_write(struct kernfs_open_file *of, |
491 | char *buf, size_t nbytes, loff_t off, | |
492 | bool threadgroup) | |
0a268dbd | 493 | { |
715c809d TH |
494 | struct cgroup *cgrp; |
495 | struct task_struct *task; | |
496 | const struct cred *cred, *tcred; | |
497 | ssize_t ret; | |
9a3284fa | 498 | bool locked; |
715c809d TH |
499 | |
500 | cgrp = cgroup_kn_lock_live(of->kn, false); | |
501 | if (!cgrp) | |
502 | return -ENODEV; | |
503 | ||
9a3284fa | 504 | task = cgroup_procs_write_start(buf, threadgroup, &locked); |
715c809d TH |
505 | ret = PTR_ERR_OR_ZERO(task); |
506 | if (ret) | |
507 | goto out_unlock; | |
508 | ||
509 | /* | |
510 | * Even if we're attaching all tasks in the thread group, we only | |
511 | * need to check permissions on one of them. | |
512 | */ | |
513 | cred = current_cred(); | |
514 | tcred = get_task_cred(task); | |
515 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && | |
516 | !uid_eq(cred->euid, tcred->uid) && | |
517 | !uid_eq(cred->euid, tcred->suid)) | |
518 | ret = -EACCES; | |
519 | put_cred(tcred); | |
520 | if (ret) | |
521 | goto out_finish; | |
522 | ||
523 | ret = cgroup_attach_task(cgrp, task, threadgroup); | |
524 | ||
525 | out_finish: | |
9a3284fa | 526 | cgroup_procs_write_finish(task, locked); |
715c809d TH |
527 | out_unlock: |
528 | cgroup_kn_unlock(of->kn); | |
529 | ||
530 | return ret ?: nbytes; | |
531 | } | |
532 | ||
533 | static ssize_t cgroup1_procs_write(struct kernfs_open_file *of, | |
534 | char *buf, size_t nbytes, loff_t off) | |
535 | { | |
536 | return __cgroup1_procs_write(of, buf, nbytes, off, true); | |
537 | } | |
538 | ||
539 | static ssize_t cgroup1_tasks_write(struct kernfs_open_file *of, | |
540 | char *buf, size_t nbytes, loff_t off) | |
541 | { | |
542 | return __cgroup1_procs_write(of, buf, nbytes, off, false); | |
0a268dbd TH |
543 | } |
544 | ||
545 | static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of, | |
546 | char *buf, size_t nbytes, loff_t off) | |
547 | { | |
548 | struct cgroup *cgrp; | |
549 | ||
550 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); | |
551 | ||
552 | cgrp = cgroup_kn_lock_live(of->kn, false); | |
553 | if (!cgrp) | |
554 | return -ENODEV; | |
555 | spin_lock(&release_agent_path_lock); | |
556 | strlcpy(cgrp->root->release_agent_path, strstrip(buf), | |
557 | sizeof(cgrp->root->release_agent_path)); | |
558 | spin_unlock(&release_agent_path_lock); | |
559 | cgroup_kn_unlock(of->kn); | |
560 | return nbytes; | |
561 | } | |
562 | ||
563 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) | |
564 | { | |
565 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
566 | ||
567 | spin_lock(&release_agent_path_lock); | |
568 | seq_puts(seq, cgrp->root->release_agent_path); | |
569 | spin_unlock(&release_agent_path_lock); | |
570 | seq_putc(seq, '\n'); | |
571 | return 0; | |
572 | } | |
573 | ||
574 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) | |
575 | { | |
576 | seq_puts(seq, "0\n"); | |
577 | return 0; | |
578 | } | |
579 | ||
580 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, | |
581 | struct cftype *cft) | |
582 | { | |
583 | return notify_on_release(css->cgroup); | |
584 | } | |
585 | ||
586 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, | |
587 | struct cftype *cft, u64 val) | |
588 | { | |
589 | if (val) | |
590 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); | |
591 | else | |
592 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); | |
593 | return 0; | |
594 | } | |
595 | ||
596 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, | |
597 | struct cftype *cft) | |
598 | { | |
599 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); | |
600 | } | |
601 | ||
602 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, | |
603 | struct cftype *cft, u64 val) | |
604 | { | |
605 | if (val) | |
606 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); | |
607 | else | |
608 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); | |
609 | return 0; | |
610 | } | |
611 | ||
612 | /* cgroup core interface files for the legacy hierarchies */ | |
d62beb7f | 613 | struct cftype cgroup1_base_files[] = { |
0a268dbd TH |
614 | { |
615 | .name = "cgroup.procs", | |
616 | .seq_start = cgroup_pidlist_start, | |
617 | .seq_next = cgroup_pidlist_next, | |
618 | .seq_stop = cgroup_pidlist_stop, | |
619 | .seq_show = cgroup_pidlist_show, | |
620 | .private = CGROUP_FILE_PROCS, | |
715c809d | 621 | .write = cgroup1_procs_write, |
0a268dbd TH |
622 | }, |
623 | { | |
624 | .name = "cgroup.clone_children", | |
625 | .read_u64 = cgroup_clone_children_read, | |
626 | .write_u64 = cgroup_clone_children_write, | |
627 | }, | |
628 | { | |
629 | .name = "cgroup.sane_behavior", | |
630 | .flags = CFTYPE_ONLY_ON_ROOT, | |
631 | .seq_show = cgroup_sane_behavior_show, | |
632 | }, | |
633 | { | |
634 | .name = "tasks", | |
635 | .seq_start = cgroup_pidlist_start, | |
636 | .seq_next = cgroup_pidlist_next, | |
637 | .seq_stop = cgroup_pidlist_stop, | |
638 | .seq_show = cgroup_pidlist_show, | |
639 | .private = CGROUP_FILE_TASKS, | |
715c809d | 640 | .write = cgroup1_tasks_write, |
0a268dbd TH |
641 | }, |
642 | { | |
643 | .name = "notify_on_release", | |
644 | .read_u64 = cgroup_read_notify_on_release, | |
645 | .write_u64 = cgroup_write_notify_on_release, | |
646 | }, | |
647 | { | |
648 | .name = "release_agent", | |
649 | .flags = CFTYPE_ONLY_ON_ROOT, | |
650 | .seq_show = cgroup_release_agent_show, | |
651 | .write = cgroup_release_agent_write, | |
652 | .max_write_len = PATH_MAX - 1, | |
653 | }, | |
654 | { } /* terminate */ | |
655 | }; | |
656 | ||
657 | /* Display information about each subsystem and each hierarchy */ | |
3f3942ac | 658 | int proc_cgroupstats_show(struct seq_file *m, void *v) |
0a268dbd TH |
659 | { |
660 | struct cgroup_subsys *ss; | |
661 | int i; | |
662 | ||
663 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); | |
664 | /* | |
665 | * ideally we don't want subsystems moving around while we do this. | |
666 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
667 | * subsys/hierarchy state. | |
668 | */ | |
669 | mutex_lock(&cgroup_mutex); | |
670 | ||
671 | for_each_subsys(ss, i) | |
672 | seq_printf(m, "%s\t%d\t%d\t%d\n", | |
673 | ss->legacy_name, ss->root->hierarchy_id, | |
674 | atomic_read(&ss->root->nr_cgrps), | |
675 | cgroup_ssid_enabled(i)); | |
676 | ||
677 | mutex_unlock(&cgroup_mutex); | |
678 | return 0; | |
679 | } | |
680 | ||
0a268dbd TH |
681 | /** |
682 | * cgroupstats_build - build and fill cgroupstats | |
683 | * @stats: cgroupstats to fill information into | |
684 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
685 | * been requested. | |
686 | * | |
687 | * Build and fill cgroupstats so that taskstats can export it to user | |
688 | * space. | |
689 | */ | |
690 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
691 | { | |
692 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); | |
693 | struct cgroup *cgrp; | |
694 | struct css_task_iter it; | |
695 | struct task_struct *tsk; | |
696 | ||
697 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ | |
698 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
699 | kernfs_type(kn) != KERNFS_DIR) | |
700 | return -EINVAL; | |
701 | ||
702 | mutex_lock(&cgroup_mutex); | |
703 | ||
704 | /* | |
705 | * We aren't being called from kernfs and there's no guarantee on | |
706 | * @kn->priv's validity. For this and css_tryget_online_from_dir(), | |
707 | * @kn->priv is RCU safe. Let's do the RCU dancing. | |
708 | */ | |
709 | rcu_read_lock(); | |
e0aed7c7 | 710 | cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv); |
0a268dbd TH |
711 | if (!cgrp || cgroup_is_dead(cgrp)) { |
712 | rcu_read_unlock(); | |
713 | mutex_unlock(&cgroup_mutex); | |
714 | return -ENOENT; | |
715 | } | |
716 | rcu_read_unlock(); | |
717 | ||
bc2fb7ed | 718 | css_task_iter_start(&cgrp->self, 0, &it); |
0a268dbd TH |
719 | while ((tsk = css_task_iter_next(&it))) { |
720 | switch (tsk->state) { | |
721 | case TASK_RUNNING: | |
722 | stats->nr_running++; | |
723 | break; | |
724 | case TASK_INTERRUPTIBLE: | |
725 | stats->nr_sleeping++; | |
726 | break; | |
727 | case TASK_UNINTERRUPTIBLE: | |
728 | stats->nr_uninterruptible++; | |
729 | break; | |
730 | case TASK_STOPPED: | |
731 | stats->nr_stopped++; | |
732 | break; | |
733 | default: | |
734 | if (delayacct_is_task_waiting_on_io(tsk)) | |
735 | stats->nr_io_wait++; | |
736 | break; | |
737 | } | |
738 | } | |
739 | css_task_iter_end(&it); | |
740 | ||
741 | mutex_unlock(&cgroup_mutex); | |
742 | return 0; | |
743 | } | |
744 | ||
d62beb7f | 745 | void cgroup1_check_for_release(struct cgroup *cgrp) |
0a268dbd TH |
746 | { |
747 | if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) && | |
748 | !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp)) | |
749 | schedule_work(&cgrp->release_agent_work); | |
750 | } | |
751 | ||
752 | /* | |
753 | * Notify userspace when a cgroup is released, by running the | |
754 | * configured release agent with the name of the cgroup (path | |
755 | * relative to the root of cgroup file system) as the argument. | |
756 | * | |
757 | * Most likely, this user command will try to rmdir this cgroup. | |
758 | * | |
759 | * This races with the possibility that some other task will be | |
760 | * attached to this cgroup before it is removed, or that some other | |
761 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
762 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
763 | * unused, and this cgroup will be reprieved from its death sentence, | |
764 | * to continue to serve a useful existence. Next time it's released, | |
765 | * we will get notified again, if it still has 'notify_on_release' set. | |
766 | * | |
767 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
768 | * means only wait until the task is successfully execve()'d. The | |
769 | * separate release agent task is forked by call_usermodehelper(), | |
770 | * then control in this thread returns here, without waiting for the | |
771 | * release agent task. We don't bother to wait because the caller of | |
772 | * this routine has no use for the exit status of the release agent | |
773 | * task, so no sense holding our caller up for that. | |
774 | */ | |
d62beb7f | 775 | void cgroup1_release_agent(struct work_struct *work) |
0a268dbd TH |
776 | { |
777 | struct cgroup *cgrp = | |
778 | container_of(work, struct cgroup, release_agent_work); | |
779 | char *pathbuf = NULL, *agentbuf = NULL; | |
780 | char *argv[3], *envp[3]; | |
781 | int ret; | |
782 | ||
783 | mutex_lock(&cgroup_mutex); | |
784 | ||
785 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); | |
786 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
787 | if (!pathbuf || !agentbuf) | |
788 | goto out; | |
789 | ||
790 | spin_lock_irq(&css_set_lock); | |
791 | ret = cgroup_path_ns_locked(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns); | |
792 | spin_unlock_irq(&css_set_lock); | |
793 | if (ret < 0 || ret >= PATH_MAX) | |
794 | goto out; | |
795 | ||
796 | argv[0] = agentbuf; | |
797 | argv[1] = pathbuf; | |
798 | argv[2] = NULL; | |
799 | ||
800 | /* minimal command environment */ | |
801 | envp[0] = "HOME=/"; | |
802 | envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
803 | envp[2] = NULL; | |
804 | ||
805 | mutex_unlock(&cgroup_mutex); | |
806 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
807 | goto out_free; | |
808 | out: | |
809 | mutex_unlock(&cgroup_mutex); | |
810 | out_free: | |
811 | kfree(agentbuf); | |
812 | kfree(pathbuf); | |
813 | } | |
814 | ||
815 | /* | |
816 | * cgroup_rename - Only allow simple rename of directories in place. | |
817 | */ | |
1592c9b2 TH |
818 | static int cgroup1_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
819 | const char *new_name_str) | |
0a268dbd TH |
820 | { |
821 | struct cgroup *cgrp = kn->priv; | |
822 | int ret; | |
823 | ||
824 | if (kernfs_type(kn) != KERNFS_DIR) | |
825 | return -ENOTDIR; | |
826 | if (kn->parent != new_parent) | |
827 | return -EIO; | |
828 | ||
0a268dbd TH |
829 | /* |
830 | * We're gonna grab cgroup_mutex which nests outside kernfs | |
831 | * active_ref. kernfs_rename() doesn't require active_ref | |
832 | * protection. Break them before grabbing cgroup_mutex. | |
833 | */ | |
834 | kernfs_break_active_protection(new_parent); | |
835 | kernfs_break_active_protection(kn); | |
836 | ||
837 | mutex_lock(&cgroup_mutex); | |
838 | ||
839 | ret = kernfs_rename(kn, new_parent, new_name_str); | |
840 | if (!ret) | |
e4f8d81c | 841 | TRACE_CGROUP_PATH(rename, cgrp); |
0a268dbd TH |
842 | |
843 | mutex_unlock(&cgroup_mutex); | |
844 | ||
845 | kernfs_unbreak_active_protection(kn); | |
846 | kernfs_unbreak_active_protection(new_parent); | |
847 | return ret; | |
848 | } | |
849 | ||
1592c9b2 TH |
850 | static int cgroup1_show_options(struct seq_file *seq, struct kernfs_root *kf_root) |
851 | { | |
852 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); | |
853 | struct cgroup_subsys *ss; | |
854 | int ssid; | |
855 | ||
856 | for_each_subsys(ss, ssid) | |
857 | if (root->subsys_mask & (1 << ssid)) | |
858 | seq_show_option(seq, ss->legacy_name, NULL); | |
859 | if (root->flags & CGRP_ROOT_NOPREFIX) | |
860 | seq_puts(seq, ",noprefix"); | |
861 | if (root->flags & CGRP_ROOT_XATTR) | |
862 | seq_puts(seq, ",xattr"); | |
e1cba4b8 WL |
863 | if (root->flags & CGRP_ROOT_CPUSET_V2_MODE) |
864 | seq_puts(seq, ",cpuset_v2_mode"); | |
1592c9b2 TH |
865 | |
866 | spin_lock(&release_agent_path_lock); | |
867 | if (strlen(root->release_agent_path)) | |
868 | seq_show_option(seq, "release_agent", | |
869 | root->release_agent_path); | |
870 | spin_unlock(&release_agent_path_lock); | |
871 | ||
872 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) | |
873 | seq_puts(seq, ",clone_children"); | |
874 | if (strlen(root->name)) | |
875 | seq_show_option(seq, "name", root->name); | |
876 | return 0; | |
877 | } | |
878 | ||
8d2451f4 AV |
879 | enum cgroup1_param { |
880 | Opt_all, | |
881 | Opt_clone_children, | |
882 | Opt_cpuset_v2_mode, | |
883 | Opt_name, | |
884 | Opt_none, | |
885 | Opt_noprefix, | |
886 | Opt_release_agent, | |
887 | Opt_xattr, | |
888 | }; | |
1592c9b2 | 889 | |
8d2451f4 AV |
890 | static const struct fs_parameter_spec cgroup1_param_specs[] = { |
891 | fsparam_flag ("all", Opt_all), | |
892 | fsparam_flag ("clone_children", Opt_clone_children), | |
893 | fsparam_flag ("cpuset_v2_mode", Opt_cpuset_v2_mode), | |
894 | fsparam_string("name", Opt_name), | |
895 | fsparam_flag ("none", Opt_none), | |
896 | fsparam_flag ("noprefix", Opt_noprefix), | |
897 | fsparam_string("release_agent", Opt_release_agent), | |
898 | fsparam_flag ("xattr", Opt_xattr), | |
899 | {} | |
900 | }; | |
1592c9b2 | 901 | |
8d2451f4 AV |
902 | const struct fs_parameter_description cgroup1_fs_parameters = { |
903 | .name = "cgroup1", | |
904 | .specs = cgroup1_param_specs, | |
905 | }; | |
1592c9b2 | 906 | |
8d2451f4 AV |
907 | int cgroup1_parse_param(struct fs_context *fc, struct fs_parameter *param) |
908 | { | |
909 | struct cgroup_fs_context *ctx = cgroup_fc2context(fc); | |
910 | struct cgroup_subsys *ss; | |
911 | struct fs_parse_result result; | |
912 | int opt, i; | |
913 | ||
914 | opt = fs_parse(fc, &cgroup1_fs_parameters, param, &result); | |
915 | if (opt == -ENOPARAM) { | |
916 | if (strcmp(param->key, "source") == 0) { | |
917 | fc->source = param->string; | |
918 | param->string = NULL; | |
919 | return 0; | |
920 | } | |
1592c9b2 | 921 | for_each_subsys(ss, i) { |
8d2451f4 | 922 | if (strcmp(param->key, ss->legacy_name)) |
1592c9b2 | 923 | continue; |
f5dfb531 | 924 | ctx->subsys_mask |= (1 << i); |
8d2451f4 | 925 | return 0; |
1592c9b2 | 926 | } |
8d2451f4 AV |
927 | return cg_invalf(fc, "cgroup1: Unknown subsys name '%s'", param->key); |
928 | } | |
929 | if (opt < 0) | |
930 | return opt; | |
931 | ||
932 | switch (opt) { | |
933 | case Opt_none: | |
934 | /* Explicitly have no subsystems */ | |
935 | ctx->none = true; | |
936 | break; | |
937 | case Opt_all: | |
938 | ctx->all_ss = true; | |
939 | break; | |
940 | case Opt_noprefix: | |
941 | ctx->flags |= CGRP_ROOT_NOPREFIX; | |
942 | break; | |
943 | case Opt_clone_children: | |
944 | ctx->cpuset_clone_children = true; | |
945 | break; | |
946 | case Opt_cpuset_v2_mode: | |
947 | ctx->flags |= CGRP_ROOT_CPUSET_V2_MODE; | |
948 | break; | |
949 | case Opt_xattr: | |
950 | ctx->flags |= CGRP_ROOT_XATTR; | |
951 | break; | |
952 | case Opt_release_agent: | |
953 | /* Specifying two release agents is forbidden */ | |
954 | if (ctx->release_agent) | |
955 | return cg_invalf(fc, "cgroup1: release_agent respecified"); | |
956 | ctx->release_agent = param->string; | |
957 | param->string = NULL; | |
958 | break; | |
959 | case Opt_name: | |
960 | /* blocked by boot param? */ | |
961 | if (cgroup_no_v1_named) | |
1592c9b2 | 962 | return -ENOENT; |
8d2451f4 AV |
963 | /* Can't specify an empty name */ |
964 | if (!param->size) | |
965 | return cg_invalf(fc, "cgroup1: Empty name"); | |
966 | if (param->size > MAX_CGROUP_ROOT_NAMELEN - 1) | |
967 | return cg_invalf(fc, "cgroup1: Name too long"); | |
968 | /* Must match [\w.-]+ */ | |
969 | for (i = 0; i < param->size; i++) { | |
970 | char c = param->string[i]; | |
971 | if (isalnum(c)) | |
972 | continue; | |
973 | if ((c == '.') || (c == '-') || (c == '_')) | |
974 | continue; | |
975 | return cg_invalf(fc, "cgroup1: Invalid name"); | |
976 | } | |
977 | /* Specifying two names is forbidden */ | |
978 | if (ctx->name) | |
979 | return cg_invalf(fc, "cgroup1: name respecified"); | |
980 | ctx->name = param->string; | |
981 | param->string = NULL; | |
982 | break; | |
1592c9b2 | 983 | } |
f5dfb531 AV |
984 | return 0; |
985 | } | |
986 | ||
8d2451f4 | 987 | static int check_cgroupfs_options(struct fs_context *fc) |
f5dfb531 | 988 | { |
8d2451f4 | 989 | struct cgroup_fs_context *ctx = cgroup_fc2context(fc); |
f5dfb531 AV |
990 | u16 mask = U16_MAX; |
991 | u16 enabled = 0; | |
992 | struct cgroup_subsys *ss; | |
993 | int i; | |
994 | ||
995 | #ifdef CONFIG_CPUSETS | |
996 | mask = ~((u16)1 << cpuset_cgrp_id); | |
997 | #endif | |
998 | for_each_subsys(ss, i) | |
999 | if (cgroup_ssid_enabled(i) && !cgroup1_ssid_disabled(i)) | |
1000 | enabled |= 1 << i; | |
1001 | ||
1002 | ctx->subsys_mask &= enabled; | |
1592c9b2 TH |
1003 | |
1004 | /* | |
f5dfb531 AV |
1005 | * In absense of 'none', 'name=' or subsystem name options, |
1006 | * let's default to 'all'. | |
1592c9b2 | 1007 | */ |
f5dfb531 AV |
1008 | if (!ctx->subsys_mask && !ctx->none && !ctx->name) |
1009 | ctx->all_ss = true; | |
1010 | ||
1011 | if (ctx->all_ss) { | |
1012 | /* Mutually exclusive option 'all' + subsystem name */ | |
1013 | if (ctx->subsys_mask) | |
8d2451f4 | 1014 | return cg_invalf(fc, "cgroup1: subsys name conflicts with all"); |
f5dfb531 AV |
1015 | /* 'all' => select all the subsystems */ |
1016 | ctx->subsys_mask = enabled; | |
1017 | } | |
1592c9b2 TH |
1018 | |
1019 | /* | |
1020 | * We either have to specify by name or by subsystems. (So all | |
1021 | * empty hierarchies must have a name). | |
1022 | */ | |
f5dfb531 | 1023 | if (!ctx->subsys_mask && !ctx->name) |
8d2451f4 | 1024 | return cg_invalf(fc, "cgroup1: Need name or subsystem set"); |
1592c9b2 TH |
1025 | |
1026 | /* | |
1027 | * Option noprefix was introduced just for backward compatibility | |
1028 | * with the old cpuset, so we allow noprefix only if mounting just | |
1029 | * the cpuset subsystem. | |
1030 | */ | |
f5dfb531 | 1031 | if ((ctx->flags & CGRP_ROOT_NOPREFIX) && (ctx->subsys_mask & mask)) |
8d2451f4 | 1032 | return cg_invalf(fc, "cgroup1: noprefix used incorrectly"); |
1592c9b2 TH |
1033 | |
1034 | /* Can't specify "none" and some subsystems */ | |
f5dfb531 | 1035 | if (ctx->subsys_mask && ctx->none) |
8d2451f4 | 1036 | return cg_invalf(fc, "cgroup1: none used incorrectly"); |
1592c9b2 TH |
1037 | |
1038 | return 0; | |
1039 | } | |
1040 | ||
90129625 | 1041 | int cgroup1_reconfigure(struct fs_context *fc) |
1592c9b2 | 1042 | { |
90129625 AV |
1043 | struct cgroup_fs_context *ctx = cgroup_fc2context(fc); |
1044 | struct kernfs_root *kf_root = kernfs_root_from_sb(fc->root->d_sb); | |
1592c9b2 | 1045 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
90129625 | 1046 | int ret = 0; |
1592c9b2 TH |
1047 | u16 added_mask, removed_mask; |
1048 | ||
1049 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); | |
1050 | ||
1051 | /* See what subsystems are wanted */ | |
8d2451f4 | 1052 | ret = check_cgroupfs_options(fc); |
1592c9b2 TH |
1053 | if (ret) |
1054 | goto out_unlock; | |
1055 | ||
f5dfb531 | 1056 | if (ctx->subsys_mask != root->subsys_mask || ctx->release_agent) |
1592c9b2 TH |
1057 | pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n", |
1058 | task_tgid_nr(current), current->comm); | |
1059 | ||
f5dfb531 AV |
1060 | added_mask = ctx->subsys_mask & ~root->subsys_mask; |
1061 | removed_mask = root->subsys_mask & ~ctx->subsys_mask; | |
1592c9b2 TH |
1062 | |
1063 | /* Don't allow flags or name to change at remount */ | |
f5dfb531 AV |
1064 | if ((ctx->flags ^ root->flags) || |
1065 | (ctx->name && strcmp(ctx->name, root->name))) { | |
8d2451f4 | 1066 | cg_invalf(fc, "option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"", |
f5dfb531 | 1067 | ctx->flags, ctx->name ?: "", root->flags, root->name); |
1592c9b2 TH |
1068 | ret = -EINVAL; |
1069 | goto out_unlock; | |
1070 | } | |
1071 | ||
1072 | /* remounting is not allowed for populated hierarchies */ | |
1073 | if (!list_empty(&root->cgrp.self.children)) { | |
1074 | ret = -EBUSY; | |
1075 | goto out_unlock; | |
1076 | } | |
1077 | ||
1078 | ret = rebind_subsystems(root, added_mask); | |
1079 | if (ret) | |
1080 | goto out_unlock; | |
1081 | ||
1082 | WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask)); | |
1083 | ||
f5dfb531 | 1084 | if (ctx->release_agent) { |
1592c9b2 | 1085 | spin_lock(&release_agent_path_lock); |
f5dfb531 | 1086 | strcpy(root->release_agent_path, ctx->release_agent); |
1592c9b2 TH |
1087 | spin_unlock(&release_agent_path_lock); |
1088 | } | |
1089 | ||
1090 | trace_cgroup_remount(root); | |
1091 | ||
1092 | out_unlock: | |
1592c9b2 TH |
1093 | mutex_unlock(&cgroup_mutex); |
1094 | return ret; | |
1095 | } | |
1096 | ||
1097 | struct kernfs_syscall_ops cgroup1_kf_syscall_ops = { | |
1098 | .rename = cgroup1_rename, | |
1099 | .show_options = cgroup1_show_options, | |
1592c9b2 TH |
1100 | .mkdir = cgroup_mkdir, |
1101 | .rmdir = cgroup_rmdir, | |
1102 | .show_path = cgroup_show_path, | |
1103 | }; | |
1104 | ||
6678889f AV |
1105 | /* |
1106 | * The guts of cgroup1 mount - find or create cgroup_root to use. | |
1107 | * Called with cgroup_mutex held; returns 0 on success, -E... on | |
1108 | * error and positive - in case when the candidate is busy dying. | |
1109 | * On success it stashes a reference to cgroup_root into given | |
1110 | * cgroup_fs_context; that reference is *NOT* counting towards the | |
1111 | * cgroup_root refcount. | |
1112 | */ | |
1113 | static int cgroup1_root_to_use(struct fs_context *fc) | |
1592c9b2 | 1114 | { |
7feeef58 | 1115 | struct cgroup_fs_context *ctx = cgroup_fc2context(fc); |
1592c9b2 TH |
1116 | struct cgroup_root *root; |
1117 | struct cgroup_subsys *ss; | |
1592c9b2 TH |
1118 | int i, ret; |
1119 | ||
1592c9b2 | 1120 | /* First find the desired set of subsystems */ |
8d2451f4 | 1121 | ret = check_cgroupfs_options(fc); |
1592c9b2 | 1122 | if (ret) |
6678889f | 1123 | return ret; |
1592c9b2 TH |
1124 | |
1125 | /* | |
1126 | * Destruction of cgroup root is asynchronous, so subsystems may | |
1127 | * still be dying after the previous unmount. Let's drain the | |
1128 | * dying subsystems. We just need to ensure that the ones | |
1129 | * unmounted previously finish dying and don't care about new ones | |
1130 | * starting. Testing ref liveliness is good enough. | |
1131 | */ | |
1132 | for_each_subsys(ss, i) { | |
f5dfb531 | 1133 | if (!(ctx->subsys_mask & (1 << i)) || |
1592c9b2 TH |
1134 | ss->root == &cgrp_dfl_root) |
1135 | continue; | |
1136 | ||
6678889f AV |
1137 | if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) |
1138 | return 1; /* restart */ | |
1592c9b2 TH |
1139 | cgroup_put(&ss->root->cgrp); |
1140 | } | |
1141 | ||
1142 | for_each_root(root) { | |
1143 | bool name_match = false; | |
1144 | ||
1145 | if (root == &cgrp_dfl_root) | |
1146 | continue; | |
1147 | ||
1148 | /* | |
1149 | * If we asked for a name then it must match. Also, if | |
1150 | * name matches but sybsys_mask doesn't, we should fail. | |
1151 | * Remember whether name matched. | |
1152 | */ | |
f5dfb531 AV |
1153 | if (ctx->name) { |
1154 | if (strcmp(ctx->name, root->name)) | |
1592c9b2 TH |
1155 | continue; |
1156 | name_match = true; | |
1157 | } | |
1158 | ||
1159 | /* | |
1160 | * If we asked for subsystems (or explicitly for no | |
1161 | * subsystems) then they must match. | |
1162 | */ | |
f5dfb531 AV |
1163 | if ((ctx->subsys_mask || ctx->none) && |
1164 | (ctx->subsys_mask != root->subsys_mask)) { | |
1592c9b2 TH |
1165 | if (!name_match) |
1166 | continue; | |
6678889f | 1167 | return -EBUSY; |
1592c9b2 TH |
1168 | } |
1169 | ||
f5dfb531 | 1170 | if (root->flags ^ ctx->flags) |
1592c9b2 TH |
1171 | pr_warn("new mount options do not match the existing superblock, will be ignored\n"); |
1172 | ||
cf6299b1 | 1173 | ctx->root = root; |
6678889f | 1174 | return 0; |
1592c9b2 TH |
1175 | } |
1176 | ||
1177 | /* | |
1178 | * No such thing, create a new one. name= matching without subsys | |
1179 | * specification is allowed for already existing hierarchies but we | |
1180 | * can't create new one without subsys specification. | |
1181 | */ | |
6678889f AV |
1182 | if (!ctx->subsys_mask && !ctx->none) |
1183 | return cg_invalf(fc, "cgroup1: No subsys list or none specified"); | |
1592c9b2 TH |
1184 | |
1185 | /* Hierarchies may only be created in the initial cgroup namespace. */ | |
cca8f327 | 1186 | if (ctx->ns != &init_cgroup_ns) |
6678889f | 1187 | return -EPERM; |
1592c9b2 TH |
1188 | |
1189 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
6678889f AV |
1190 | if (!root) |
1191 | return -ENOMEM; | |
1592c9b2 | 1192 | |
cf6299b1 AV |
1193 | ctx->root = root; |
1194 | init_cgroup_root(ctx); | |
1592c9b2 | 1195 | |
f5dfb531 | 1196 | ret = cgroup_setup_root(root, ctx->subsys_mask); |
1592c9b2 TH |
1197 | if (ret) |
1198 | cgroup_free_root(root); | |
6678889f AV |
1199 | return ret; |
1200 | } | |
1201 | ||
1202 | int cgroup1_get_tree(struct fs_context *fc) | |
1203 | { | |
6678889f AV |
1204 | struct cgroup_fs_context *ctx = cgroup_fc2context(fc); |
1205 | int ret; | |
1206 | ||
1207 | /* Check if the caller has permission to mount. */ | |
cca8f327 | 1208 | if (!ns_capable(ctx->ns->user_ns, CAP_SYS_ADMIN)) |
6678889f AV |
1209 | return -EPERM; |
1210 | ||
1211 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); | |
1212 | ||
1213 | ret = cgroup1_root_to_use(fc); | |
1214 | if (!ret && !percpu_ref_tryget_live(&ctx->root->cgrp.self.refcnt)) | |
1215 | ret = 1; /* restart */ | |
1592c9b2 | 1216 | |
1592c9b2 | 1217 | mutex_unlock(&cgroup_mutex); |
1592c9b2 | 1218 | |
6678889f | 1219 | if (!ret) |
cca8f327 | 1220 | ret = cgroup_do_get_tree(fc); |
6678889f AV |
1221 | |
1222 | if (!ret && percpu_ref_is_dying(&ctx->root->cgrp.self.refcnt)) { | |
71d883c3 AV |
1223 | struct super_block *sb = fc->root->d_sb; |
1224 | dput(fc->root); | |
35ac1184 | 1225 | deactivate_locked_super(sb); |
6678889f AV |
1226 | ret = 1; |
1227 | } | |
1228 | ||
1229 | if (unlikely(ret > 0)) { | |
35ac1184 | 1230 | msleep(10); |
7feeef58 | 1231 | return restart_syscall(); |
9732adc5 | 1232 | } |
71d883c3 | 1233 | return ret; |
1592c9b2 TH |
1234 | } |
1235 | ||
0a268dbd TH |
1236 | static int __init cgroup1_wq_init(void) |
1237 | { | |
1238 | /* | |
1239 | * Used to destroy pidlists and separate to serve as flush domain. | |
1240 | * Cap @max_active to 1 too. | |
1241 | */ | |
1242 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
1243 | 0, 1); | |
1244 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
1245 | return 0; | |
1246 | } | |
1247 | core_initcall(cgroup1_wq_init); | |
1248 | ||
1249 | static int __init cgroup_no_v1(char *str) | |
1250 | { | |
1251 | struct cgroup_subsys *ss; | |
1252 | char *token; | |
1253 | int i; | |
1254 | ||
1255 | while ((token = strsep(&str, ",")) != NULL) { | |
1256 | if (!*token) | |
1257 | continue; | |
1258 | ||
1259 | if (!strcmp(token, "all")) { | |
1260 | cgroup_no_v1_mask = U16_MAX; | |
3fc9c12d TH |
1261 | continue; |
1262 | } | |
1263 | ||
1264 | if (!strcmp(token, "named")) { | |
1265 | cgroup_no_v1_named = true; | |
1266 | continue; | |
0a268dbd TH |
1267 | } |
1268 | ||
1269 | for_each_subsys(ss, i) { | |
1270 | if (strcmp(token, ss->name) && | |
1271 | strcmp(token, ss->legacy_name)) | |
1272 | continue; | |
1273 | ||
1274 | cgroup_no_v1_mask |= 1 << i; | |
1275 | } | |
1276 | } | |
1277 | return 1; | |
1278 | } | |
1279 | __setup("cgroup_no_v1=", cgroup_no_v1); |