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