Loading...
Note: File does not exist in v3.5.6.
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 */
30static u16 cgroup_no_v1_mask;
31
32/* disable named v1 mounts */
33static 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 */
39static struct workqueue_struct *cgroup_pidlist_destroy_wq;
40
41/* protects cgroup_subsys->release_agent_path */
42static DEFINE_SPINLOCK(release_agent_path_lock);
43
44bool cgroup1_ssid_disabled(int ssid)
45{
46 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 */
54int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
55{
56 struct cgroup_root *root;
57 int retval = 0;
58
59 mutex_lock(&cgroup_mutex);
60 percpu_down_write(&cgroup_threadgroup_rwsem);
61 for_each_root(root) {
62 struct cgroup *from_cgrp;
63
64 if (root == &cgrp_dfl_root)
65 continue;
66
67 spin_lock_irq(&css_set_lock);
68 from_cgrp = task_cgroup_from_root(from, root);
69 spin_unlock_irq(&css_set_lock);
70
71 retval = cgroup_attach_task(from_cgrp, tsk, false);
72 if (retval)
73 break;
74 }
75 percpu_up_write(&cgroup_threadgroup_rwsem);
76 mutex_unlock(&cgroup_mutex);
77
78 return retval;
79}
80EXPORT_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 */
93int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
94{
95 DEFINE_CGROUP_MGCTX(mgctx);
96 struct cgrp_cset_link *link;
97 struct css_task_iter it;
98 struct task_struct *task;
99 int ret;
100
101 if (cgroup_on_dfl(to))
102 return -EINVAL;
103
104 ret = cgroup_migrate_vet_dst(to);
105 if (ret)
106 return ret;
107
108 mutex_lock(&cgroup_mutex);
109
110 percpu_down_write(&cgroup_threadgroup_rwsem);
111
112 /* all tasks in @from are being moved, all csets are source */
113 spin_lock_irq(&css_set_lock);
114 list_for_each_entry(link, &from->cset_links, cset_link)
115 cgroup_migrate_add_src(link->cset, to, &mgctx);
116 spin_unlock_irq(&css_set_lock);
117
118 ret = cgroup_migrate_prepare_dst(&mgctx);
119 if (ret)
120 goto out_err;
121
122 /*
123 * Migrate tasks one-by-one until @from is empty. This fails iff
124 * ->can_attach() fails.
125 */
126 do {
127 css_task_iter_start(&from->self, 0, &it);
128
129 do {
130 task = css_task_iter_next(&it);
131 } while (task && (task->flags & PF_EXITING));
132
133 if (task)
134 get_task_struct(task);
135 css_task_iter_end(&it);
136
137 if (task) {
138 ret = cgroup_migrate(task, false, &mgctx);
139 if (!ret)
140 TRACE_CGROUP_PATH(transfer_tasks, to, task, false);
141 put_task_struct(task);
142 }
143 } while (task && !ret);
144out_err:
145 cgroup_migrate_finish(&mgctx);
146 percpu_up_write(&cgroup_threadgroup_rwsem);
147 mutex_unlock(&cgroup_mutex);
148 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? */
162enum 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 */
173struct 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 */
195void cgroup1_pidlist_destroy_all(struct cgroup *cgrp)
196{
197 struct cgroup_pidlist *l, *tmp_l;
198
199 mutex_lock(&cgrp->pidlist_mutex);
200 list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links)
201 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0);
202 mutex_unlock(&cgrp->pidlist_mutex);
203
204 flush_workqueue(cgroup_pidlist_destroy_wq);
205 BUG_ON(!list_empty(&cgrp->pidlists));
206}
207
208static void cgroup_pidlist_destroy_work_fn(struct work_struct *work)
209{
210 struct delayed_work *dwork = to_delayed_work(work);
211 struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist,
212 destroy_dwork);
213 struct cgroup_pidlist *tofree = NULL;
214
215 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 if (!delayed_work_pending(dwork)) {
222 list_del(&l->links);
223 kvfree(l->list);
224 put_pid_ns(l->key.ns);
225 tofree = l;
226 }
227
228 mutex_unlock(&l->owner->pidlist_mutex);
229 kfree(tofree);
230}
231
232/*
233 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
234 * Returns the number of unique elements.
235 */
236static int pidlist_uniq(pid_t *list, int length)
237{
238 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 if (length == 0 || length == 1)
245 return length;
246 /* src and dest walk down the list; dest counts unique elements */
247 for (src = 1; src < length; src++) {
248 /* find next unique element */
249 while (list[src] == list[src-1]) {
250 src++;
251 if (src == length)
252 goto after;
253 }
254 /* dest always points to where the next unique element goes */
255 list[dest] = list[src];
256 dest++;
257 }
258after:
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 */
271static int cmppid(const void *a, const void *b)
272{
273 return *(pid_t *)a - *(pid_t *)b;
274}
275
276static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
277 enum cgroup_filetype type)
278{
279 struct cgroup_pidlist *l;
280 /* don't need task_nsproxy() if we're looking at ourself */
281 struct pid_namespace *ns = task_active_pid_ns(current);
282
283 lockdep_assert_held(&cgrp->pidlist_mutex);
284
285 list_for_each_entry(l, &cgrp->pidlists, links)
286 if (l->key.type == type && l->key.ns == ns)
287 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 */
297static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp,
298 enum cgroup_filetype type)
299{
300 struct cgroup_pidlist *l;
301
302 lockdep_assert_held(&cgrp->pidlist_mutex);
303
304 l = cgroup_pidlist_find(cgrp, type);
305 if (l)
306 return l;
307
308 /* entry not found; create a new one */
309 l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
310 if (!l)
311 return l;
312
313 INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn);
314 l->key.type = type;
315 /* don't need task_nsproxy() if we're looking at ourself */
316 l->key.ns = get_pid_ns(task_active_pid_ns(current));
317 l->owner = cgrp;
318 list_add(&l->links, &cgrp->pidlists);
319 return l;
320}
321
322/*
323 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
324 */
325static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
326 struct cgroup_pidlist **lp)
327{
328 pid_t *array;
329 int length;
330 int pid, n = 0; /* used for populating the array */
331 struct css_task_iter it;
332 struct task_struct *tsk;
333 struct cgroup_pidlist *l;
334
335 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 length = cgroup_task_count(cgrp);
344 array = kvmalloc_array(length, sizeof(pid_t), GFP_KERNEL);
345 if (!array)
346 return -ENOMEM;
347 /* now, populate the array */
348 css_task_iter_start(&cgrp->self, 0, &it);
349 while ((tsk = css_task_iter_next(&it))) {
350 if (unlikely(n == length))
351 break;
352 /* get tgid or pid for procs or tasks file respectively */
353 if (type == CGROUP_FILE_PROCS)
354 pid = task_tgid_vnr(tsk);
355 else
356 pid = task_pid_vnr(tsk);
357 if (pid > 0) /* make sure to only use valid results */
358 array[n++] = pid;
359 }
360 css_task_iter_end(&it);
361 length = n;
362 /* now sort & (if procs) strip out duplicates */
363 sort(array, length, sizeof(pid_t), cmppid, NULL);
364 if (type == CGROUP_FILE_PROCS)
365 length = pidlist_uniq(array, length);
366
367 l = cgroup_pidlist_find_create(cgrp, type);
368 if (!l) {
369 kvfree(array);
370 return -ENOMEM;
371 }
372
373 /* store array, freeing old if necessary */
374 kvfree(l->list);
375 l->list = array;
376 l->length = length;
377 *lp = l;
378 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
387static 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 struct kernfs_open_file *of = s->private;
396 struct cgroup *cgrp = seq_css(s)->cgroup;
397 struct cgroup_pidlist *l;
398 enum cgroup_filetype type = seq_cft(s)->private;
399 int index = 0, pid = *pos;
400 int *iter, ret;
401
402 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 if (of->priv)
411 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 if (!of->priv) {
418 ret = pidlist_array_load(cgrp, type,
419 (struct cgroup_pidlist **)&of->priv);
420 if (ret)
421 return ERR_PTR(ret);
422 }
423 l = of->priv;
424
425 if (pid) {
426 int end = l->length;
427
428 while (index < end) {
429 int mid = (index + end) / 2;
430 if (l->list[mid] == pid) {
431 index = mid;
432 break;
433 } else if (l->list[mid] <= pid)
434 index = mid + 1;
435 else
436 end = mid;
437 }
438 }
439 /* If we're off the end of the array, we're done */
440 if (index >= l->length)
441 return NULL;
442 /* Update the abstract position to be the actual pid that we found */
443 iter = l->list + index;
444 *pos = *iter;
445 return iter;
446}
447
448static void cgroup_pidlist_stop(struct seq_file *s, void *v)
449{
450 struct kernfs_open_file *of = s->private;
451 struct cgroup_pidlist *l = of->priv;
452
453 if (l)
454 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork,
455 CGROUP_PIDLIST_DESTROY_DELAY);
456 mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex);
457}
458
459static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
460{
461 struct kernfs_open_file *of = s->private;
462 struct cgroup_pidlist *l = of->priv;
463 pid_t *p = v;
464 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 p++;
470 if (p >= end) {
471 (*pos)++;
472 return NULL;
473 } else {
474 *pos = *p;
475 return p;
476 }
477}
478
479static int cgroup_pidlist_show(struct seq_file *s, void *v)
480{
481 seq_printf(s, "%d\n", *(int *)v);
482
483 return 0;
484}
485
486static ssize_t __cgroup1_procs_write(struct kernfs_open_file *of,
487 char *buf, size_t nbytes, loff_t off,
488 bool threadgroup)
489{
490 struct cgroup *cgrp;
491 struct task_struct *task;
492 const struct cred *cred, *tcred;
493 ssize_t ret;
494 bool locked;
495
496 cgrp = cgroup_kn_lock_live(of->kn, false);
497 if (!cgrp)
498 return -ENODEV;
499
500 task = cgroup_procs_write_start(buf, threadgroup, &locked);
501 ret = PTR_ERR_OR_ZERO(task);
502 if (ret)
503 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 cred = current_cred();
510 tcred = get_task_cred(task);
511 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
512 !uid_eq(cred->euid, tcred->uid) &&
513 !uid_eq(cred->euid, tcred->suid))
514 ret = -EACCES;
515 put_cred(tcred);
516 if (ret)
517 goto out_finish;
518
519 ret = cgroup_attach_task(cgrp, task, threadgroup);
520
521out_finish:
522 cgroup_procs_write_finish(task, locked);
523out_unlock:
524 cgroup_kn_unlock(of->kn);
525
526 return ret ?: nbytes;
527}
528
529static ssize_t cgroup1_procs_write(struct kernfs_open_file *of,
530 char *buf, size_t nbytes, loff_t off)
531{
532 return __cgroup1_procs_write(of, buf, nbytes, off, true);
533}
534
535static ssize_t cgroup1_tasks_write(struct kernfs_open_file *of,
536 char *buf, size_t nbytes, loff_t off)
537{
538 return __cgroup1_procs_write(of, buf, nbytes, off, false);
539}
540
541static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of,
542 char *buf, size_t nbytes, loff_t off)
543{
544 struct cgroup *cgrp;
545
546 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
547
548 cgrp = cgroup_kn_lock_live(of->kn, false);
549 if (!cgrp)
550 return -ENODEV;
551 spin_lock(&release_agent_path_lock);
552 strlcpy(cgrp->root->release_agent_path, strstrip(buf),
553 sizeof(cgrp->root->release_agent_path));
554 spin_unlock(&release_agent_path_lock);
555 cgroup_kn_unlock(of->kn);
556 return nbytes;
557}
558
559static int cgroup_release_agent_show(struct seq_file *seq, void *v)
560{
561 struct cgroup *cgrp = seq_css(seq)->cgroup;
562
563 spin_lock(&release_agent_path_lock);
564 seq_puts(seq, cgrp->root->release_agent_path);
565 spin_unlock(&release_agent_path_lock);
566 seq_putc(seq, '\n');
567 return 0;
568}
569
570static int cgroup_sane_behavior_show(struct seq_file *seq, void *v)
571{
572 seq_puts(seq, "0\n");
573 return 0;
574}
575
576static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css,
577 struct cftype *cft)
578{
579 return notify_on_release(css->cgroup);
580}
581
582static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css,
583 struct cftype *cft, u64 val)
584{
585 if (val)
586 set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
587 else
588 clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
589 return 0;
590}
591
592static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css,
593 struct cftype *cft)
594{
595 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
596}
597
598static int cgroup_clone_children_write(struct cgroup_subsys_state *css,
599 struct cftype *cft, u64 val)
600{
601 if (val)
602 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
603 else
604 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
605 return 0;
606}
607
608/* cgroup core interface files for the legacy hierarchies */
609struct 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 */
654int proc_cgroupstats_show(struct seq_file *m, void *v)
655{
656 struct cgroup_subsys *ss;
657 int i;
658
659 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 mutex_lock(&cgroup_mutex);
666
667 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 mutex_unlock(&cgroup_mutex);
674 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 */
686int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
687{
688 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
689 struct cgroup *cgrp;
690 struct css_task_iter it;
691 struct task_struct *tsk;
692
693 /* it should be kernfs_node belonging to cgroupfs and is a directory */
694 if (dentry->d_sb->s_type != &cgroup_fs_type || !kn ||
695 kernfs_type(kn) != KERNFS_DIR)
696 return -EINVAL;
697
698 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 rcu_read_lock();
706 cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
707 if (!cgrp || cgroup_is_dead(cgrp)) {
708 rcu_read_unlock();
709 mutex_unlock(&cgroup_mutex);
710 return -ENOENT;
711 }
712 rcu_read_unlock();
713
714 css_task_iter_start(&cgrp->self, 0, &it);
715 while ((tsk = css_task_iter_next(&it))) {
716 switch (tsk->state) {
717 case TASK_RUNNING:
718 stats->nr_running++;
719 break;
720 case TASK_INTERRUPTIBLE:
721 stats->nr_sleeping++;
722 break;
723 case TASK_UNINTERRUPTIBLE:
724 stats->nr_uninterruptible++;
725 break;
726 case TASK_STOPPED:
727 stats->nr_stopped++;
728 break;
729 default:
730 if (delayacct_is_task_waiting_on_io(tsk))
731 stats->nr_io_wait++;
732 break;
733 }
734 }
735 css_task_iter_end(&it);
736
737 mutex_unlock(&cgroup_mutex);
738 return 0;
739}
740
741void cgroup1_check_for_release(struct cgroup *cgrp)
742{
743 if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) &&
744 !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp))
745 schedule_work(&cgrp->release_agent_work);
746}
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 */
771void cgroup1_release_agent(struct work_struct *work)
772{
773 struct cgroup *cgrp =
774 container_of(work, struct cgroup, release_agent_work);
775 char *pathbuf, *agentbuf;
776 char *argv[3], *envp[3];
777 int ret;
778
779 /* snoop agent path and exit early if empty */
780 if (!cgrp->root->release_agent_path[0])
781 return;
782
783 /* prepare argument buffers */
784 pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
785 agentbuf = kmalloc(PATH_MAX, GFP_KERNEL);
786 if (!pathbuf || !agentbuf)
787 goto out_free;
788
789 spin_lock(&release_agent_path_lock);
790 strlcpy(agentbuf, cgrp->root->release_agent_path, PATH_MAX);
791 spin_unlock(&release_agent_path_lock);
792 if (!agentbuf[0])
793 goto out_free;
794
795 ret = cgroup_path_ns(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns);
796 if (ret < 0 || ret >= PATH_MAX)
797 goto out_free;
798
799 argv[0] = agentbuf;
800 argv[1] = pathbuf;
801 argv[2] = NULL;
802
803 /* minimal command environment */
804 envp[0] = "HOME=/";
805 envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
806 envp[2] = NULL;
807
808 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
809out_free:
810 kfree(agentbuf);
811 kfree(pathbuf);
812}
813
814/*
815 * cgroup_rename - Only allow simple rename of directories in place.
816 */
817static int cgroup1_rename(struct kernfs_node *kn, struct kernfs_node *new_parent,
818 const char *new_name_str)
819{
820 struct cgroup *cgrp = kn->priv;
821 int ret;
822
823 if (kernfs_type(kn) != KERNFS_DIR)
824 return -ENOTDIR;
825 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 kernfs_break_active_protection(new_parent);
834 kernfs_break_active_protection(kn);
835
836 mutex_lock(&cgroup_mutex);
837
838 ret = kernfs_rename(kn, new_parent, new_name_str);
839 if (!ret)
840 TRACE_CGROUP_PATH(rename, cgrp);
841
842 mutex_unlock(&cgroup_mutex);
843
844 kernfs_unbreak_active_protection(kn);
845 kernfs_unbreak_active_protection(new_parent);
846 return ret;
847}
848
849static int cgroup1_show_options(struct seq_file *seq, struct kernfs_root *kf_root)
850{
851 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
852 struct cgroup_subsys *ss;
853 int ssid;
854
855 for_each_subsys(ss, ssid)
856 if (root->subsys_mask & (1 << ssid))
857 seq_show_option(seq, ss->legacy_name, NULL);
858 if (root->flags & CGRP_ROOT_NOPREFIX)
859 seq_puts(seq, ",noprefix");
860 if (root->flags & CGRP_ROOT_XATTR)
861 seq_puts(seq, ",xattr");
862 if (root->flags & CGRP_ROOT_CPUSET_V2_MODE)
863 seq_puts(seq, ",cpuset_v2_mode");
864
865 spin_lock(&release_agent_path_lock);
866 if (strlen(root->release_agent_path))
867 seq_show_option(seq, "release_agent",
868 root->release_agent_path);
869 spin_unlock(&release_agent_path_lock);
870
871 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags))
872 seq_puts(seq, ",clone_children");
873 if (strlen(root->name))
874 seq_show_option(seq, "name", root->name);
875 return 0;
876}
877
878enum 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
889const 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
901int cgroup1_parse_param(struct fs_context *fc, struct fs_parameter *param)
902{
903 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
904 struct cgroup_subsys *ss;
905 struct fs_parse_result result;
906 int opt, i;
907
908 opt = fs_parse(fc, cgroup1_fs_parameters, param, &result);
909 if (opt == -ENOPARAM) {
910 if (strcmp(param->key, "source") == 0) {
911 fc->source = param->string;
912 param->string = NULL;
913 return 0;
914 }
915 for_each_subsys(ss, i) {
916 if (strcmp(param->key, ss->legacy_name))
917 continue;
918 ctx->subsys_mask |= (1 << i);
919 return 0;
920 }
921 return invalfc(fc, "Unknown subsys name '%s'", param->key);
922 }
923 if (opt < 0)
924 return opt;
925
926 switch (opt) {
927 case Opt_none:
928 /* Explicitly have no subsystems */
929 ctx->none = true;
930 break;
931 case Opt_all:
932 ctx->all_ss = true;
933 break;
934 case Opt_noprefix:
935 ctx->flags |= CGRP_ROOT_NOPREFIX;
936 break;
937 case Opt_clone_children:
938 ctx->cpuset_clone_children = true;
939 break;
940 case Opt_cpuset_v2_mode:
941 ctx->flags |= CGRP_ROOT_CPUSET_V2_MODE;
942 break;
943 case Opt_xattr:
944 ctx->flags |= CGRP_ROOT_XATTR;
945 break;
946 case Opt_release_agent:
947 /* Specifying two release agents is forbidden */
948 if (ctx->release_agent)
949 return invalfc(fc, "release_agent respecified");
950 ctx->release_agent = param->string;
951 param->string = NULL;
952 break;
953 case Opt_name:
954 /* blocked by boot param? */
955 if (cgroup_no_v1_named)
956 return -ENOENT;
957 /* Can't specify an empty name */
958 if (!param->size)
959 return invalfc(fc, "Empty name");
960 if (param->size > MAX_CGROUP_ROOT_NAMELEN - 1)
961 return invalfc(fc, "Name too long");
962 /* Must match [\w.-]+ */
963 for (i = 0; i < param->size; i++) {
964 char c = param->string[i];
965 if (isalnum(c))
966 continue;
967 if ((c == '.') || (c == '-') || (c == '_'))
968 continue;
969 return invalfc(fc, "Invalid name");
970 }
971 /* Specifying two names is forbidden */
972 if (ctx->name)
973 return invalfc(fc, "name respecified");
974 ctx->name = param->string;
975 param->string = NULL;
976 break;
977 }
978 return 0;
979}
980
981static int check_cgroupfs_options(struct fs_context *fc)
982{
983 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
984 u16 mask = U16_MAX;
985 u16 enabled = 0;
986 struct cgroup_subsys *ss;
987 int i;
988
989#ifdef CONFIG_CPUSETS
990 mask = ~((u16)1 << cpuset_cgrp_id);
991#endif
992 for_each_subsys(ss, i)
993 if (cgroup_ssid_enabled(i) && !cgroup1_ssid_disabled(i))
994 enabled |= 1 << i;
995
996 ctx->subsys_mask &= enabled;
997
998 /*
999 * In absense of 'none', 'name=' or subsystem name options,
1000 * let's default to 'all'.
1001 */
1002 if (!ctx->subsys_mask && !ctx->none && !ctx->name)
1003 ctx->all_ss = true;
1004
1005 if (ctx->all_ss) {
1006 /* Mutually exclusive option 'all' + subsystem name */
1007 if (ctx->subsys_mask)
1008 return invalfc(fc, "subsys name conflicts with all");
1009 /* 'all' => select all the subsystems */
1010 ctx->subsys_mask = enabled;
1011 }
1012
1013 /*
1014 * We either have to specify by name or by subsystems. (So all
1015 * empty hierarchies must have a name).
1016 */
1017 if (!ctx->subsys_mask && !ctx->name)
1018 return invalfc(fc, "Need name or subsystem set");
1019
1020 /*
1021 * Option noprefix was introduced just for backward compatibility
1022 * with the old cpuset, so we allow noprefix only if mounting just
1023 * the cpuset subsystem.
1024 */
1025 if ((ctx->flags & CGRP_ROOT_NOPREFIX) && (ctx->subsys_mask & mask))
1026 return invalfc(fc, "noprefix used incorrectly");
1027
1028 /* Can't specify "none" and some subsystems */
1029 if (ctx->subsys_mask && ctx->none)
1030 return invalfc(fc, "none used incorrectly");
1031
1032 return 0;
1033}
1034
1035int cgroup1_reconfigure(struct fs_context *fc)
1036{
1037 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1038 struct kernfs_root *kf_root = kernfs_root_from_sb(fc->root->d_sb);
1039 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
1040 int ret = 0;
1041 u16 added_mask, removed_mask;
1042
1043 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1044
1045 /* See what subsystems are wanted */
1046 ret = check_cgroupfs_options(fc);
1047 if (ret)
1048 goto out_unlock;
1049
1050 if (ctx->subsys_mask != root->subsys_mask || ctx->release_agent)
1051 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1052 task_tgid_nr(current), current->comm);
1053
1054 added_mask = ctx->subsys_mask & ~root->subsys_mask;
1055 removed_mask = root->subsys_mask & ~ctx->subsys_mask;
1056
1057 /* Don't allow flags or name to change at remount */
1058 if ((ctx->flags ^ root->flags) ||
1059 (ctx->name && strcmp(ctx->name, root->name))) {
1060 errorfc(fc, "option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"",
1061 ctx->flags, ctx->name ?: "", root->flags, root->name);
1062 ret = -EINVAL;
1063 goto out_unlock;
1064 }
1065
1066 /* remounting is not allowed for populated hierarchies */
1067 if (!list_empty(&root->cgrp.self.children)) {
1068 ret = -EBUSY;
1069 goto out_unlock;
1070 }
1071
1072 ret = rebind_subsystems(root, added_mask);
1073 if (ret)
1074 goto out_unlock;
1075
1076 WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask));
1077
1078 if (ctx->release_agent) {
1079 spin_lock(&release_agent_path_lock);
1080 strcpy(root->release_agent_path, ctx->release_agent);
1081 spin_unlock(&release_agent_path_lock);
1082 }
1083
1084 trace_cgroup_remount(root);
1085
1086 out_unlock:
1087 mutex_unlock(&cgroup_mutex);
1088 return ret;
1089}
1090
1091struct kernfs_syscall_ops cgroup1_kf_syscall_ops = {
1092 .rename = cgroup1_rename,
1093 .show_options = cgroup1_show_options,
1094 .mkdir = cgroup_mkdir,
1095 .rmdir = cgroup_rmdir,
1096 .show_path = cgroup_show_path,
1097};
1098
1099/*
1100 * The guts of cgroup1 mount - find or create cgroup_root to use.
1101 * Called with cgroup_mutex held; returns 0 on success, -E... on
1102 * error and positive - in case when the candidate is busy dying.
1103 * On success it stashes a reference to cgroup_root into given
1104 * cgroup_fs_context; that reference is *NOT* counting towards the
1105 * cgroup_root refcount.
1106 */
1107static int cgroup1_root_to_use(struct fs_context *fc)
1108{
1109 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1110 struct cgroup_root *root;
1111 struct cgroup_subsys *ss;
1112 int i, ret;
1113
1114 /* First find the desired set of subsystems */
1115 ret = check_cgroupfs_options(fc);
1116 if (ret)
1117 return ret;
1118
1119 /*
1120 * Destruction of cgroup root is asynchronous, so subsystems may
1121 * still be dying after the previous unmount. Let's drain the
1122 * dying subsystems. We just need to ensure that the ones
1123 * unmounted previously finish dying and don't care about new ones
1124 * starting. Testing ref liveliness is good enough.
1125 */
1126 for_each_subsys(ss, i) {
1127 if (!(ctx->subsys_mask & (1 << i)) ||
1128 ss->root == &cgrp_dfl_root)
1129 continue;
1130
1131 if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt))
1132 return 1; /* restart */
1133 cgroup_put(&ss->root->cgrp);
1134 }
1135
1136 for_each_root(root) {
1137 bool name_match = false;
1138
1139 if (root == &cgrp_dfl_root)
1140 continue;
1141
1142 /*
1143 * If we asked for a name then it must match. Also, if
1144 * name matches but sybsys_mask doesn't, we should fail.
1145 * Remember whether name matched.
1146 */
1147 if (ctx->name) {
1148 if (strcmp(ctx->name, root->name))
1149 continue;
1150 name_match = true;
1151 }
1152
1153 /*
1154 * If we asked for subsystems (or explicitly for no
1155 * subsystems) then they must match.
1156 */
1157 if ((ctx->subsys_mask || ctx->none) &&
1158 (ctx->subsys_mask != root->subsys_mask)) {
1159 if (!name_match)
1160 continue;
1161 return -EBUSY;
1162 }
1163
1164 if (root->flags ^ ctx->flags)
1165 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
1166
1167 ctx->root = root;
1168 return 0;
1169 }
1170
1171 /*
1172 * No such thing, create a new one. name= matching without subsys
1173 * specification is allowed for already existing hierarchies but we
1174 * can't create new one without subsys specification.
1175 */
1176 if (!ctx->subsys_mask && !ctx->none)
1177 return invalfc(fc, "No subsys list or none specified");
1178
1179 /* Hierarchies may only be created in the initial cgroup namespace. */
1180 if (ctx->ns != &init_cgroup_ns)
1181 return -EPERM;
1182
1183 root = kzalloc(sizeof(*root), GFP_KERNEL);
1184 if (!root)
1185 return -ENOMEM;
1186
1187 ctx->root = root;
1188 init_cgroup_root(ctx);
1189
1190 ret = cgroup_setup_root(root, ctx->subsys_mask);
1191 if (ret)
1192 cgroup_free_root(root);
1193 return ret;
1194}
1195
1196int cgroup1_get_tree(struct fs_context *fc)
1197{
1198 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1199 int ret;
1200
1201 /* Check if the caller has permission to mount. */
1202 if (!ns_capable(ctx->ns->user_ns, CAP_SYS_ADMIN))
1203 return -EPERM;
1204
1205 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1206
1207 ret = cgroup1_root_to_use(fc);
1208 if (!ret && !percpu_ref_tryget_live(&ctx->root->cgrp.self.refcnt))
1209 ret = 1; /* restart */
1210
1211 mutex_unlock(&cgroup_mutex);
1212
1213 if (!ret)
1214 ret = cgroup_do_get_tree(fc);
1215
1216 if (!ret && percpu_ref_is_dying(&ctx->root->cgrp.self.refcnt)) {
1217 struct super_block *sb = fc->root->d_sb;
1218 dput(fc->root);
1219 deactivate_locked_super(sb);
1220 ret = 1;
1221 }
1222
1223 if (unlikely(ret > 0)) {
1224 msleep(10);
1225 return restart_syscall();
1226 }
1227 return ret;
1228}
1229
1230static int __init cgroup1_wq_init(void)
1231{
1232 /*
1233 * Used to destroy pidlists and separate to serve as flush domain.
1234 * Cap @max_active to 1 too.
1235 */
1236 cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy",
1237 0, 1);
1238 BUG_ON(!cgroup_pidlist_destroy_wq);
1239 return 0;
1240}
1241core_initcall(cgroup1_wq_init);
1242
1243static int __init cgroup_no_v1(char *str)
1244{
1245 struct cgroup_subsys *ss;
1246 char *token;
1247 int i;
1248
1249 while ((token = strsep(&str, ",")) != NULL) {
1250 if (!*token)
1251 continue;
1252
1253 if (!strcmp(token, "all")) {
1254 cgroup_no_v1_mask = U16_MAX;
1255 continue;
1256 }
1257
1258 if (!strcmp(token, "named")) {
1259 cgroup_no_v1_named = true;
1260 continue;
1261 }
1262
1263 for_each_subsys(ss, i) {
1264 if (strcmp(token, ss->name) &&
1265 strcmp(token, ss->legacy_name))
1266 continue;
1267
1268 cgroup_no_v1_mask |= 1 << i;
1269 }
1270 }
1271 return 1;
1272}
1273__setup("cgroup_no_v1=", cgroup_no_v1);