Merge branch 'fixes-for-3.5' of git://git.kernel.org/pub/scm/linux/kernel/git/coolone...
[linux-2.6/luiz-linux-2.6.git] / kernel / cgroup.c
blob2097684cf19474aee676513af39de14035c4fce7
1 /*
2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
11 * Copyright notices from the original cpuset code:
12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
19 * 2003-10-10 Written by Simon Derr.
20 * 2003-10-22 Updates by Stephen Hemminger.
21 * 2004 May-July Rework by Paul Jackson.
22 * ---------------------------------------------------
24 * This file is subject to the terms and conditions of the GNU General Public
25 * License. See the file COPYING in the main directory of the Linux
26 * distribution for more details.
29 #include <linux/cgroup.h>
30 #include <linux/cred.h>
31 #include <linux/ctype.h>
32 #include <linux/errno.h>
33 #include <linux/fs.h>
34 #include <linux/init_task.h>
35 #include <linux/kernel.h>
36 #include <linux/list.h>
37 #include <linux/mm.h>
38 #include <linux/mutex.h>
39 #include <linux/mount.h>
40 #include <linux/pagemap.h>
41 #include <linux/proc_fs.h>
42 #include <linux/rcupdate.h>
43 #include <linux/sched.h>
44 #include <linux/backing-dev.h>
45 #include <linux/seq_file.h>
46 #include <linux/slab.h>
47 #include <linux/magic.h>
48 #include <linux/spinlock.h>
49 #include <linux/string.h>
50 #include <linux/sort.h>
51 #include <linux/kmod.h>
52 #include <linux/module.h>
53 #include <linux/delayacct.h>
54 #include <linux/cgroupstats.h>
55 #include <linux/hash.h>
56 #include <linux/namei.h>
57 #include <linux/pid_namespace.h>
58 #include <linux/idr.h>
59 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
60 #include <linux/eventfd.h>
61 #include <linux/poll.h>
62 #include <linux/flex_array.h> /* used in cgroup_attach_proc */
63 #include <linux/kthread.h>
65 #include <linux/atomic.h>
67 /* css deactivation bias, makes css->refcnt negative to deny new trygets */
68 #define CSS_DEACT_BIAS INT_MIN
71 * cgroup_mutex is the master lock. Any modification to cgroup or its
72 * hierarchy must be performed while holding it.
74 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
75 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
76 * release_agent_path and so on. Modifying requires both cgroup_mutex and
77 * cgroup_root_mutex. Readers can acquire either of the two. This is to
78 * break the following locking order cycle.
80 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
81 * B. namespace_sem -> cgroup_mutex
83 * B happens only through cgroup_show_options() and using cgroup_root_mutex
84 * breaks it.
86 static DEFINE_MUTEX(cgroup_mutex);
87 static DEFINE_MUTEX(cgroup_root_mutex);
90 * Generate an array of cgroup subsystem pointers. At boot time, this is
91 * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are
92 * registered after that. The mutable section of this array is protected by
93 * cgroup_mutex.
95 #define SUBSYS(_x) &_x ## _subsys,
96 static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
97 #include <linux/cgroup_subsys.h>
100 #define MAX_CGROUP_ROOT_NAMELEN 64
103 * A cgroupfs_root represents the root of a cgroup hierarchy,
104 * and may be associated with a superblock to form an active
105 * hierarchy
107 struct cgroupfs_root {
108 struct super_block *sb;
111 * The bitmask of subsystems intended to be attached to this
112 * hierarchy
114 unsigned long subsys_bits;
116 /* Unique id for this hierarchy. */
117 int hierarchy_id;
119 /* The bitmask of subsystems currently attached to this hierarchy */
120 unsigned long actual_subsys_bits;
122 /* A list running through the attached subsystems */
123 struct list_head subsys_list;
125 /* The root cgroup for this hierarchy */
126 struct cgroup top_cgroup;
128 /* Tracks how many cgroups are currently defined in hierarchy.*/
129 int number_of_cgroups;
131 /* A list running through the active hierarchies */
132 struct list_head root_list;
134 /* All cgroups on this root, cgroup_mutex protected */
135 struct list_head allcg_list;
137 /* Hierarchy-specific flags */
138 unsigned long flags;
140 /* The path to use for release notifications. */
141 char release_agent_path[PATH_MAX];
143 /* The name for this hierarchy - may be empty */
144 char name[MAX_CGROUP_ROOT_NAMELEN];
148 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
149 * subsystems that are otherwise unattached - it never has more than a
150 * single cgroup, and all tasks are part of that cgroup.
152 static struct cgroupfs_root rootnode;
155 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
157 struct cfent {
158 struct list_head node;
159 struct dentry *dentry;
160 struct cftype *type;
164 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
165 * cgroup_subsys->use_id != 0.
167 #define CSS_ID_MAX (65535)
168 struct css_id {
170 * The css to which this ID points. This pointer is set to valid value
171 * after cgroup is populated. If cgroup is removed, this will be NULL.
172 * This pointer is expected to be RCU-safe because destroy()
173 * is called after synchronize_rcu(). But for safe use, css_is_removed()
174 * css_tryget() should be used for avoiding race.
176 struct cgroup_subsys_state __rcu *css;
178 * ID of this css.
180 unsigned short id;
182 * Depth in hierarchy which this ID belongs to.
184 unsigned short depth;
186 * ID is freed by RCU. (and lookup routine is RCU safe.)
188 struct rcu_head rcu_head;
190 * Hierarchy of CSS ID belongs to.
192 unsigned short stack[0]; /* Array of Length (depth+1) */
196 * cgroup_event represents events which userspace want to receive.
198 struct cgroup_event {
200 * Cgroup which the event belongs to.
202 struct cgroup *cgrp;
204 * Control file which the event associated.
206 struct cftype *cft;
208 * eventfd to signal userspace about the event.
210 struct eventfd_ctx *eventfd;
212 * Each of these stored in a list by the cgroup.
214 struct list_head list;
216 * All fields below needed to unregister event when
217 * userspace closes eventfd.
219 poll_table pt;
220 wait_queue_head_t *wqh;
221 wait_queue_t wait;
222 struct work_struct remove;
225 /* The list of hierarchy roots */
227 static LIST_HEAD(roots);
228 static int root_count;
230 static DEFINE_IDA(hierarchy_ida);
231 static int next_hierarchy_id;
232 static DEFINE_SPINLOCK(hierarchy_id_lock);
234 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
235 #define dummytop (&rootnode.top_cgroup)
237 /* This flag indicates whether tasks in the fork and exit paths should
238 * check for fork/exit handlers to call. This avoids us having to do
239 * extra work in the fork/exit path if none of the subsystems need to
240 * be called.
242 static int need_forkexit_callback __read_mostly;
244 #ifdef CONFIG_PROVE_LOCKING
245 int cgroup_lock_is_held(void)
247 return lockdep_is_held(&cgroup_mutex);
249 #else /* #ifdef CONFIG_PROVE_LOCKING */
250 int cgroup_lock_is_held(void)
252 return mutex_is_locked(&cgroup_mutex);
254 #endif /* #else #ifdef CONFIG_PROVE_LOCKING */
256 EXPORT_SYMBOL_GPL(cgroup_lock_is_held);
258 static int css_unbias_refcnt(int refcnt)
260 return refcnt >= 0 ? refcnt : refcnt - CSS_DEACT_BIAS;
263 /* the current nr of refs, always >= 0 whether @css is deactivated or not */
264 static int css_refcnt(struct cgroup_subsys_state *css)
266 int v = atomic_read(&css->refcnt);
268 return css_unbias_refcnt(v);
271 /* convenient tests for these bits */
272 inline int cgroup_is_removed(const struct cgroup *cgrp)
274 return test_bit(CGRP_REMOVED, &cgrp->flags);
277 /* bits in struct cgroupfs_root flags field */
278 enum {
279 ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
282 static int cgroup_is_releasable(const struct cgroup *cgrp)
284 const int bits =
285 (1 << CGRP_RELEASABLE) |
286 (1 << CGRP_NOTIFY_ON_RELEASE);
287 return (cgrp->flags & bits) == bits;
290 static int notify_on_release(const struct cgroup *cgrp)
292 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
295 static int clone_children(const struct cgroup *cgrp)
297 return test_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
301 * for_each_subsys() allows you to iterate on each subsystem attached to
302 * an active hierarchy
304 #define for_each_subsys(_root, _ss) \
305 list_for_each_entry(_ss, &_root->subsys_list, sibling)
307 /* for_each_active_root() allows you to iterate across the active hierarchies */
308 #define for_each_active_root(_root) \
309 list_for_each_entry(_root, &roots, root_list)
311 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
313 return dentry->d_fsdata;
316 static inline struct cfent *__d_cfe(struct dentry *dentry)
318 return dentry->d_fsdata;
321 static inline struct cftype *__d_cft(struct dentry *dentry)
323 return __d_cfe(dentry)->type;
326 /* the list of cgroups eligible for automatic release. Protected by
327 * release_list_lock */
328 static LIST_HEAD(release_list);
329 static DEFINE_RAW_SPINLOCK(release_list_lock);
330 static void cgroup_release_agent(struct work_struct *work);
331 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
332 static void check_for_release(struct cgroup *cgrp);
334 /* Link structure for associating css_set objects with cgroups */
335 struct cg_cgroup_link {
337 * List running through cg_cgroup_links associated with a
338 * cgroup, anchored on cgroup->css_sets
340 struct list_head cgrp_link_list;
341 struct cgroup *cgrp;
343 * List running through cg_cgroup_links pointing at a
344 * single css_set object, anchored on css_set->cg_links
346 struct list_head cg_link_list;
347 struct css_set *cg;
350 /* The default css_set - used by init and its children prior to any
351 * hierarchies being mounted. It contains a pointer to the root state
352 * for each subsystem. Also used to anchor the list of css_sets. Not
353 * reference-counted, to improve performance when child cgroups
354 * haven't been created.
357 static struct css_set init_css_set;
358 static struct cg_cgroup_link init_css_set_link;
360 static int cgroup_init_idr(struct cgroup_subsys *ss,
361 struct cgroup_subsys_state *css);
363 /* css_set_lock protects the list of css_set objects, and the
364 * chain of tasks off each css_set. Nests outside task->alloc_lock
365 * due to cgroup_iter_start() */
366 static DEFINE_RWLOCK(css_set_lock);
367 static int css_set_count;
370 * hash table for cgroup groups. This improves the performance to find
371 * an existing css_set. This hash doesn't (currently) take into
372 * account cgroups in empty hierarchies.
374 #define CSS_SET_HASH_BITS 7
375 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
376 static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE];
378 static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[])
380 int i;
381 int index;
382 unsigned long tmp = 0UL;
384 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
385 tmp += (unsigned long)css[i];
386 tmp = (tmp >> 16) ^ tmp;
388 index = hash_long(tmp, CSS_SET_HASH_BITS);
390 return &css_set_table[index];
393 /* We don't maintain the lists running through each css_set to its
394 * task until after the first call to cgroup_iter_start(). This
395 * reduces the fork()/exit() overhead for people who have cgroups
396 * compiled into their kernel but not actually in use */
397 static int use_task_css_set_links __read_mostly;
399 static void __put_css_set(struct css_set *cg, int taskexit)
401 struct cg_cgroup_link *link;
402 struct cg_cgroup_link *saved_link;
404 * Ensure that the refcount doesn't hit zero while any readers
405 * can see it. Similar to atomic_dec_and_lock(), but for an
406 * rwlock
408 if (atomic_add_unless(&cg->refcount, -1, 1))
409 return;
410 write_lock(&css_set_lock);
411 if (!atomic_dec_and_test(&cg->refcount)) {
412 write_unlock(&css_set_lock);
413 return;
416 /* This css_set is dead. unlink it and release cgroup refcounts */
417 hlist_del(&cg->hlist);
418 css_set_count--;
420 list_for_each_entry_safe(link, saved_link, &cg->cg_links,
421 cg_link_list) {
422 struct cgroup *cgrp = link->cgrp;
423 list_del(&link->cg_link_list);
424 list_del(&link->cgrp_link_list);
425 if (atomic_dec_and_test(&cgrp->count) &&
426 notify_on_release(cgrp)) {
427 if (taskexit)
428 set_bit(CGRP_RELEASABLE, &cgrp->flags);
429 check_for_release(cgrp);
432 kfree(link);
435 write_unlock(&css_set_lock);
436 kfree_rcu(cg, rcu_head);
440 * refcounted get/put for css_set objects
442 static inline void get_css_set(struct css_set *cg)
444 atomic_inc(&cg->refcount);
447 static inline void put_css_set(struct css_set *cg)
449 __put_css_set(cg, 0);
452 static inline void put_css_set_taskexit(struct css_set *cg)
454 __put_css_set(cg, 1);
458 * compare_css_sets - helper function for find_existing_css_set().
459 * @cg: candidate css_set being tested
460 * @old_cg: existing css_set for a task
461 * @new_cgrp: cgroup that's being entered by the task
462 * @template: desired set of css pointers in css_set (pre-calculated)
464 * Returns true if "cg" matches "old_cg" except for the hierarchy
465 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
467 static bool compare_css_sets(struct css_set *cg,
468 struct css_set *old_cg,
469 struct cgroup *new_cgrp,
470 struct cgroup_subsys_state *template[])
472 struct list_head *l1, *l2;
474 if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
475 /* Not all subsystems matched */
476 return false;
480 * Compare cgroup pointers in order to distinguish between
481 * different cgroups in heirarchies with no subsystems. We
482 * could get by with just this check alone (and skip the
483 * memcmp above) but on most setups the memcmp check will
484 * avoid the need for this more expensive check on almost all
485 * candidates.
488 l1 = &cg->cg_links;
489 l2 = &old_cg->cg_links;
490 while (1) {
491 struct cg_cgroup_link *cgl1, *cgl2;
492 struct cgroup *cg1, *cg2;
494 l1 = l1->next;
495 l2 = l2->next;
496 /* See if we reached the end - both lists are equal length. */
497 if (l1 == &cg->cg_links) {
498 BUG_ON(l2 != &old_cg->cg_links);
499 break;
500 } else {
501 BUG_ON(l2 == &old_cg->cg_links);
503 /* Locate the cgroups associated with these links. */
504 cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
505 cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
506 cg1 = cgl1->cgrp;
507 cg2 = cgl2->cgrp;
508 /* Hierarchies should be linked in the same order. */
509 BUG_ON(cg1->root != cg2->root);
512 * If this hierarchy is the hierarchy of the cgroup
513 * that's changing, then we need to check that this
514 * css_set points to the new cgroup; if it's any other
515 * hierarchy, then this css_set should point to the
516 * same cgroup as the old css_set.
518 if (cg1->root == new_cgrp->root) {
519 if (cg1 != new_cgrp)
520 return false;
521 } else {
522 if (cg1 != cg2)
523 return false;
526 return true;
530 * find_existing_css_set() is a helper for
531 * find_css_set(), and checks to see whether an existing
532 * css_set is suitable.
534 * oldcg: the cgroup group that we're using before the cgroup
535 * transition
537 * cgrp: the cgroup that we're moving into
539 * template: location in which to build the desired set of subsystem
540 * state objects for the new cgroup group
542 static struct css_set *find_existing_css_set(
543 struct css_set *oldcg,
544 struct cgroup *cgrp,
545 struct cgroup_subsys_state *template[])
547 int i;
548 struct cgroupfs_root *root = cgrp->root;
549 struct hlist_head *hhead;
550 struct hlist_node *node;
551 struct css_set *cg;
554 * Build the set of subsystem state objects that we want to see in the
555 * new css_set. while subsystems can change globally, the entries here
556 * won't change, so no need for locking.
558 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
559 if (root->subsys_bits & (1UL << i)) {
560 /* Subsystem is in this hierarchy. So we want
561 * the subsystem state from the new
562 * cgroup */
563 template[i] = cgrp->subsys[i];
564 } else {
565 /* Subsystem is not in this hierarchy, so we
566 * don't want to change the subsystem state */
567 template[i] = oldcg->subsys[i];
571 hhead = css_set_hash(template);
572 hlist_for_each_entry(cg, node, hhead, hlist) {
573 if (!compare_css_sets(cg, oldcg, cgrp, template))
574 continue;
576 /* This css_set matches what we need */
577 return cg;
580 /* No existing cgroup group matched */
581 return NULL;
584 static void free_cg_links(struct list_head *tmp)
586 struct cg_cgroup_link *link;
587 struct cg_cgroup_link *saved_link;
589 list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
590 list_del(&link->cgrp_link_list);
591 kfree(link);
596 * allocate_cg_links() allocates "count" cg_cgroup_link structures
597 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
598 * success or a negative error
600 static int allocate_cg_links(int count, struct list_head *tmp)
602 struct cg_cgroup_link *link;
603 int i;
604 INIT_LIST_HEAD(tmp);
605 for (i = 0; i < count; i++) {
606 link = kmalloc(sizeof(*link), GFP_KERNEL);
607 if (!link) {
608 free_cg_links(tmp);
609 return -ENOMEM;
611 list_add(&link->cgrp_link_list, tmp);
613 return 0;
617 * link_css_set - a helper function to link a css_set to a cgroup
618 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
619 * @cg: the css_set to be linked
620 * @cgrp: the destination cgroup
622 static void link_css_set(struct list_head *tmp_cg_links,
623 struct css_set *cg, struct cgroup *cgrp)
625 struct cg_cgroup_link *link;
627 BUG_ON(list_empty(tmp_cg_links));
628 link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
629 cgrp_link_list);
630 link->cg = cg;
631 link->cgrp = cgrp;
632 atomic_inc(&cgrp->count);
633 list_move(&link->cgrp_link_list, &cgrp->css_sets);
635 * Always add links to the tail of the list so that the list
636 * is sorted by order of hierarchy creation
638 list_add_tail(&link->cg_link_list, &cg->cg_links);
642 * find_css_set() takes an existing cgroup group and a
643 * cgroup object, and returns a css_set object that's
644 * equivalent to the old group, but with the given cgroup
645 * substituted into the appropriate hierarchy. Must be called with
646 * cgroup_mutex held
648 static struct css_set *find_css_set(
649 struct css_set *oldcg, struct cgroup *cgrp)
651 struct css_set *res;
652 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
654 struct list_head tmp_cg_links;
656 struct hlist_head *hhead;
657 struct cg_cgroup_link *link;
659 /* First see if we already have a cgroup group that matches
660 * the desired set */
661 read_lock(&css_set_lock);
662 res = find_existing_css_set(oldcg, cgrp, template);
663 if (res)
664 get_css_set(res);
665 read_unlock(&css_set_lock);
667 if (res)
668 return res;
670 res = kmalloc(sizeof(*res), GFP_KERNEL);
671 if (!res)
672 return NULL;
674 /* Allocate all the cg_cgroup_link objects that we'll need */
675 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
676 kfree(res);
677 return NULL;
680 atomic_set(&res->refcount, 1);
681 INIT_LIST_HEAD(&res->cg_links);
682 INIT_LIST_HEAD(&res->tasks);
683 INIT_HLIST_NODE(&res->hlist);
685 /* Copy the set of subsystem state objects generated in
686 * find_existing_css_set() */
687 memcpy(res->subsys, template, sizeof(res->subsys));
689 write_lock(&css_set_lock);
690 /* Add reference counts and links from the new css_set. */
691 list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
692 struct cgroup *c = link->cgrp;
693 if (c->root == cgrp->root)
694 c = cgrp;
695 link_css_set(&tmp_cg_links, res, c);
698 BUG_ON(!list_empty(&tmp_cg_links));
700 css_set_count++;
702 /* Add this cgroup group to the hash table */
703 hhead = css_set_hash(res->subsys);
704 hlist_add_head(&res->hlist, hhead);
706 write_unlock(&css_set_lock);
708 return res;
712 * Return the cgroup for "task" from the given hierarchy. Must be
713 * called with cgroup_mutex held.
715 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
716 struct cgroupfs_root *root)
718 struct css_set *css;
719 struct cgroup *res = NULL;
721 BUG_ON(!mutex_is_locked(&cgroup_mutex));
722 read_lock(&css_set_lock);
724 * No need to lock the task - since we hold cgroup_mutex the
725 * task can't change groups, so the only thing that can happen
726 * is that it exits and its css is set back to init_css_set.
728 css = task->cgroups;
729 if (css == &init_css_set) {
730 res = &root->top_cgroup;
731 } else {
732 struct cg_cgroup_link *link;
733 list_for_each_entry(link, &css->cg_links, cg_link_list) {
734 struct cgroup *c = link->cgrp;
735 if (c->root == root) {
736 res = c;
737 break;
741 read_unlock(&css_set_lock);
742 BUG_ON(!res);
743 return res;
747 * There is one global cgroup mutex. We also require taking
748 * task_lock() when dereferencing a task's cgroup subsys pointers.
749 * See "The task_lock() exception", at the end of this comment.
751 * A task must hold cgroup_mutex to modify cgroups.
753 * Any task can increment and decrement the count field without lock.
754 * So in general, code holding cgroup_mutex can't rely on the count
755 * field not changing. However, if the count goes to zero, then only
756 * cgroup_attach_task() can increment it again. Because a count of zero
757 * means that no tasks are currently attached, therefore there is no
758 * way a task attached to that cgroup can fork (the other way to
759 * increment the count). So code holding cgroup_mutex can safely
760 * assume that if the count is zero, it will stay zero. Similarly, if
761 * a task holds cgroup_mutex on a cgroup with zero count, it
762 * knows that the cgroup won't be removed, as cgroup_rmdir()
763 * needs that mutex.
765 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
766 * (usually) take cgroup_mutex. These are the two most performance
767 * critical pieces of code here. The exception occurs on cgroup_exit(),
768 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
769 * is taken, and if the cgroup count is zero, a usermode call made
770 * to the release agent with the name of the cgroup (path relative to
771 * the root of cgroup file system) as the argument.
773 * A cgroup can only be deleted if both its 'count' of using tasks
774 * is zero, and its list of 'children' cgroups is empty. Since all
775 * tasks in the system use _some_ cgroup, and since there is always at
776 * least one task in the system (init, pid == 1), therefore, top_cgroup
777 * always has either children cgroups and/or using tasks. So we don't
778 * need a special hack to ensure that top_cgroup cannot be deleted.
780 * The task_lock() exception
782 * The need for this exception arises from the action of
783 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
784 * another. It does so using cgroup_mutex, however there are
785 * several performance critical places that need to reference
786 * task->cgroup without the expense of grabbing a system global
787 * mutex. Therefore except as noted below, when dereferencing or, as
788 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
789 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
790 * the task_struct routinely used for such matters.
792 * P.S. One more locking exception. RCU is used to guard the
793 * update of a tasks cgroup pointer by cgroup_attach_task()
797 * cgroup_lock - lock out any changes to cgroup structures
800 void cgroup_lock(void)
802 mutex_lock(&cgroup_mutex);
804 EXPORT_SYMBOL_GPL(cgroup_lock);
807 * cgroup_unlock - release lock on cgroup changes
809 * Undo the lock taken in a previous cgroup_lock() call.
811 void cgroup_unlock(void)
813 mutex_unlock(&cgroup_mutex);
815 EXPORT_SYMBOL_GPL(cgroup_unlock);
818 * A couple of forward declarations required, due to cyclic reference loop:
819 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
820 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
821 * -> cgroup_mkdir.
824 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
825 static struct dentry *cgroup_lookup(struct inode *, struct dentry *, struct nameidata *);
826 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
827 static int cgroup_populate_dir(struct cgroup *cgrp);
828 static const struct inode_operations cgroup_dir_inode_operations;
829 static const struct file_operations proc_cgroupstats_operations;
831 static struct backing_dev_info cgroup_backing_dev_info = {
832 .name = "cgroup",
833 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
836 static int alloc_css_id(struct cgroup_subsys *ss,
837 struct cgroup *parent, struct cgroup *child);
839 static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb)
841 struct inode *inode = new_inode(sb);
843 if (inode) {
844 inode->i_ino = get_next_ino();
845 inode->i_mode = mode;
846 inode->i_uid = current_fsuid();
847 inode->i_gid = current_fsgid();
848 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
849 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
851 return inode;
855 * Call subsys's pre_destroy handler.
856 * This is called before css refcnt check.
858 static int cgroup_call_pre_destroy(struct cgroup *cgrp)
860 struct cgroup_subsys *ss;
861 int ret = 0;
863 for_each_subsys(cgrp->root, ss) {
864 if (!ss->pre_destroy)
865 continue;
867 ret = ss->pre_destroy(cgrp);
868 if (ret) {
869 /* ->pre_destroy() failure is being deprecated */
870 WARN_ON_ONCE(!ss->__DEPRECATED_clear_css_refs);
871 break;
875 return ret;
878 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
880 /* is dentry a directory ? if so, kfree() associated cgroup */
881 if (S_ISDIR(inode->i_mode)) {
882 struct cgroup *cgrp = dentry->d_fsdata;
883 struct cgroup_subsys *ss;
884 BUG_ON(!(cgroup_is_removed(cgrp)));
885 /* It's possible for external users to be holding css
886 * reference counts on a cgroup; css_put() needs to
887 * be able to access the cgroup after decrementing
888 * the reference count in order to know if it needs to
889 * queue the cgroup to be handled by the release
890 * agent */
891 synchronize_rcu();
893 mutex_lock(&cgroup_mutex);
895 * Release the subsystem state objects.
897 for_each_subsys(cgrp->root, ss)
898 ss->destroy(cgrp);
900 cgrp->root->number_of_cgroups--;
901 mutex_unlock(&cgroup_mutex);
904 * We want to drop the active superblock reference from the
905 * cgroup creation after all the dentry refs are gone -
906 * kill_sb gets mighty unhappy otherwise. Mark
907 * dentry->d_fsdata with cgroup_diput() to tell
908 * cgroup_d_release() to call deactivate_super().
910 dentry->d_fsdata = cgroup_diput;
913 * if we're getting rid of the cgroup, refcount should ensure
914 * that there are no pidlists left.
916 BUG_ON(!list_empty(&cgrp->pidlists));
918 kfree_rcu(cgrp, rcu_head);
919 } else {
920 struct cfent *cfe = __d_cfe(dentry);
921 struct cgroup *cgrp = dentry->d_parent->d_fsdata;
923 WARN_ONCE(!list_empty(&cfe->node) &&
924 cgrp != &cgrp->root->top_cgroup,
925 "cfe still linked for %s\n", cfe->type->name);
926 kfree(cfe);
928 iput(inode);
931 static int cgroup_delete(const struct dentry *d)
933 return 1;
936 static void cgroup_d_release(struct dentry *dentry)
938 /* did cgroup_diput() tell me to deactivate super? */
939 if (dentry->d_fsdata == cgroup_diput)
940 deactivate_super(dentry->d_sb);
943 static void remove_dir(struct dentry *d)
945 struct dentry *parent = dget(d->d_parent);
947 d_delete(d);
948 simple_rmdir(parent->d_inode, d);
949 dput(parent);
952 static int cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
954 struct cfent *cfe;
956 lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex);
957 lockdep_assert_held(&cgroup_mutex);
959 list_for_each_entry(cfe, &cgrp->files, node) {
960 struct dentry *d = cfe->dentry;
962 if (cft && cfe->type != cft)
963 continue;
965 dget(d);
966 d_delete(d);
967 simple_unlink(d->d_inode, d);
968 list_del_init(&cfe->node);
969 dput(d);
971 return 0;
973 return -ENOENT;
976 static void cgroup_clear_directory(struct dentry *dir)
978 struct cgroup *cgrp = __d_cgrp(dir);
980 while (!list_empty(&cgrp->files))
981 cgroup_rm_file(cgrp, NULL);
985 * NOTE : the dentry must have been dget()'ed
987 static void cgroup_d_remove_dir(struct dentry *dentry)
989 struct dentry *parent;
991 cgroup_clear_directory(dentry);
993 parent = dentry->d_parent;
994 spin_lock(&parent->d_lock);
995 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
996 list_del_init(&dentry->d_u.d_child);
997 spin_unlock(&dentry->d_lock);
998 spin_unlock(&parent->d_lock);
999 remove_dir(dentry);
1003 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
1004 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
1005 * reference to css->refcnt. In general, this refcnt is expected to goes down
1006 * to zero, soon.
1008 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
1010 static DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq);
1012 static void cgroup_wakeup_rmdir_waiter(struct cgroup *cgrp)
1014 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)))
1015 wake_up_all(&cgroup_rmdir_waitq);
1018 void cgroup_exclude_rmdir(struct cgroup_subsys_state *css)
1020 css_get(css);
1023 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css)
1025 cgroup_wakeup_rmdir_waiter(css->cgroup);
1026 css_put(css);
1030 * Call with cgroup_mutex held. Drops reference counts on modules, including
1031 * any duplicate ones that parse_cgroupfs_options took. If this function
1032 * returns an error, no reference counts are touched.
1034 static int rebind_subsystems(struct cgroupfs_root *root,
1035 unsigned long final_bits)
1037 unsigned long added_bits, removed_bits;
1038 struct cgroup *cgrp = &root->top_cgroup;
1039 int i;
1041 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1042 BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
1044 removed_bits = root->actual_subsys_bits & ~final_bits;
1045 added_bits = final_bits & ~root->actual_subsys_bits;
1046 /* Check that any added subsystems are currently free */
1047 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1048 unsigned long bit = 1UL << i;
1049 struct cgroup_subsys *ss = subsys[i];
1050 if (!(bit & added_bits))
1051 continue;
1053 * Nobody should tell us to do a subsys that doesn't exist:
1054 * parse_cgroupfs_options should catch that case and refcounts
1055 * ensure that subsystems won't disappear once selected.
1057 BUG_ON(ss == NULL);
1058 if (ss->root != &rootnode) {
1059 /* Subsystem isn't free */
1060 return -EBUSY;
1064 /* Currently we don't handle adding/removing subsystems when
1065 * any child cgroups exist. This is theoretically supportable
1066 * but involves complex error handling, so it's being left until
1067 * later */
1068 if (root->number_of_cgroups > 1)
1069 return -EBUSY;
1071 /* Process each subsystem */
1072 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1073 struct cgroup_subsys *ss = subsys[i];
1074 unsigned long bit = 1UL << i;
1075 if (bit & added_bits) {
1076 /* We're binding this subsystem to this hierarchy */
1077 BUG_ON(ss == NULL);
1078 BUG_ON(cgrp->subsys[i]);
1079 BUG_ON(!dummytop->subsys[i]);
1080 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
1081 mutex_lock(&ss->hierarchy_mutex);
1082 cgrp->subsys[i] = dummytop->subsys[i];
1083 cgrp->subsys[i]->cgroup = cgrp;
1084 list_move(&ss->sibling, &root->subsys_list);
1085 ss->root = root;
1086 if (ss->bind)
1087 ss->bind(cgrp);
1088 mutex_unlock(&ss->hierarchy_mutex);
1089 /* refcount was already taken, and we're keeping it */
1090 } else if (bit & removed_bits) {
1091 /* We're removing this subsystem */
1092 BUG_ON(ss == NULL);
1093 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1094 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1095 mutex_lock(&ss->hierarchy_mutex);
1096 if (ss->bind)
1097 ss->bind(dummytop);
1098 dummytop->subsys[i]->cgroup = dummytop;
1099 cgrp->subsys[i] = NULL;
1100 subsys[i]->root = &rootnode;
1101 list_move(&ss->sibling, &rootnode.subsys_list);
1102 mutex_unlock(&ss->hierarchy_mutex);
1103 /* subsystem is now free - drop reference on module */
1104 module_put(ss->module);
1105 } else if (bit & final_bits) {
1106 /* Subsystem state should already exist */
1107 BUG_ON(ss == NULL);
1108 BUG_ON(!cgrp->subsys[i]);
1110 * a refcount was taken, but we already had one, so
1111 * drop the extra reference.
1113 module_put(ss->module);
1114 #ifdef CONFIG_MODULE_UNLOAD
1115 BUG_ON(ss->module && !module_refcount(ss->module));
1116 #endif
1117 } else {
1118 /* Subsystem state shouldn't exist */
1119 BUG_ON(cgrp->subsys[i]);
1122 root->subsys_bits = root->actual_subsys_bits = final_bits;
1123 synchronize_rcu();
1125 return 0;
1128 static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1130 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1131 struct cgroup_subsys *ss;
1133 mutex_lock(&cgroup_root_mutex);
1134 for_each_subsys(root, ss)
1135 seq_printf(seq, ",%s", ss->name);
1136 if (test_bit(ROOT_NOPREFIX, &root->flags))
1137 seq_puts(seq, ",noprefix");
1138 if (strlen(root->release_agent_path))
1139 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1140 if (clone_children(&root->top_cgroup))
1141 seq_puts(seq, ",clone_children");
1142 if (strlen(root->name))
1143 seq_printf(seq, ",name=%s", root->name);
1144 mutex_unlock(&cgroup_root_mutex);
1145 return 0;
1148 struct cgroup_sb_opts {
1149 unsigned long subsys_bits;
1150 unsigned long flags;
1151 char *release_agent;
1152 bool clone_children;
1153 char *name;
1154 /* User explicitly requested empty subsystem */
1155 bool none;
1157 struct cgroupfs_root *new_root;
1162 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1163 * with cgroup_mutex held to protect the subsys[] array. This function takes
1164 * refcounts on subsystems to be used, unless it returns error, in which case
1165 * no refcounts are taken.
1167 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1169 char *token, *o = data;
1170 bool all_ss = false, one_ss = false;
1171 unsigned long mask = (unsigned long)-1;
1172 int i;
1173 bool module_pin_failed = false;
1175 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1177 #ifdef CONFIG_CPUSETS
1178 mask = ~(1UL << cpuset_subsys_id);
1179 #endif
1181 memset(opts, 0, sizeof(*opts));
1183 while ((token = strsep(&o, ",")) != NULL) {
1184 if (!*token)
1185 return -EINVAL;
1186 if (!strcmp(token, "none")) {
1187 /* Explicitly have no subsystems */
1188 opts->none = true;
1189 continue;
1191 if (!strcmp(token, "all")) {
1192 /* Mutually exclusive option 'all' + subsystem name */
1193 if (one_ss)
1194 return -EINVAL;
1195 all_ss = true;
1196 continue;
1198 if (!strcmp(token, "noprefix")) {
1199 set_bit(ROOT_NOPREFIX, &opts->flags);
1200 continue;
1202 if (!strcmp(token, "clone_children")) {
1203 opts->clone_children = true;
1204 continue;
1206 if (!strncmp(token, "release_agent=", 14)) {
1207 /* Specifying two release agents is forbidden */
1208 if (opts->release_agent)
1209 return -EINVAL;
1210 opts->release_agent =
1211 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1212 if (!opts->release_agent)
1213 return -ENOMEM;
1214 continue;
1216 if (!strncmp(token, "name=", 5)) {
1217 const char *name = token + 5;
1218 /* Can't specify an empty name */
1219 if (!strlen(name))
1220 return -EINVAL;
1221 /* Must match [\w.-]+ */
1222 for (i = 0; i < strlen(name); i++) {
1223 char c = name[i];
1224 if (isalnum(c))
1225 continue;
1226 if ((c == '.') || (c == '-') || (c == '_'))
1227 continue;
1228 return -EINVAL;
1230 /* Specifying two names is forbidden */
1231 if (opts->name)
1232 return -EINVAL;
1233 opts->name = kstrndup(name,
1234 MAX_CGROUP_ROOT_NAMELEN - 1,
1235 GFP_KERNEL);
1236 if (!opts->name)
1237 return -ENOMEM;
1239 continue;
1242 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1243 struct cgroup_subsys *ss = subsys[i];
1244 if (ss == NULL)
1245 continue;
1246 if (strcmp(token, ss->name))
1247 continue;
1248 if (ss->disabled)
1249 continue;
1251 /* Mutually exclusive option 'all' + subsystem name */
1252 if (all_ss)
1253 return -EINVAL;
1254 set_bit(i, &opts->subsys_bits);
1255 one_ss = true;
1257 break;
1259 if (i == CGROUP_SUBSYS_COUNT)
1260 return -ENOENT;
1264 * If the 'all' option was specified select all the subsystems,
1265 * otherwise if 'none', 'name=' and a subsystem name options
1266 * were not specified, let's default to 'all'
1268 if (all_ss || (!one_ss && !opts->none && !opts->name)) {
1269 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1270 struct cgroup_subsys *ss = subsys[i];
1271 if (ss == NULL)
1272 continue;
1273 if (ss->disabled)
1274 continue;
1275 set_bit(i, &opts->subsys_bits);
1279 /* Consistency checks */
1282 * Option noprefix was introduced just for backward compatibility
1283 * with the old cpuset, so we allow noprefix only if mounting just
1284 * the cpuset subsystem.
1286 if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1287 (opts->subsys_bits & mask))
1288 return -EINVAL;
1291 /* Can't specify "none" and some subsystems */
1292 if (opts->subsys_bits && opts->none)
1293 return -EINVAL;
1296 * We either have to specify by name or by subsystems. (So all
1297 * empty hierarchies must have a name).
1299 if (!opts->subsys_bits && !opts->name)
1300 return -EINVAL;
1303 * Grab references on all the modules we'll need, so the subsystems
1304 * don't dance around before rebind_subsystems attaches them. This may
1305 * take duplicate reference counts on a subsystem that's already used,
1306 * but rebind_subsystems handles this case.
1308 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1309 unsigned long bit = 1UL << i;
1311 if (!(bit & opts->subsys_bits))
1312 continue;
1313 if (!try_module_get(subsys[i]->module)) {
1314 module_pin_failed = true;
1315 break;
1318 if (module_pin_failed) {
1320 * oops, one of the modules was going away. this means that we
1321 * raced with a module_delete call, and to the user this is
1322 * essentially a "subsystem doesn't exist" case.
1324 for (i--; i >= CGROUP_BUILTIN_SUBSYS_COUNT; i--) {
1325 /* drop refcounts only on the ones we took */
1326 unsigned long bit = 1UL << i;
1328 if (!(bit & opts->subsys_bits))
1329 continue;
1330 module_put(subsys[i]->module);
1332 return -ENOENT;
1335 return 0;
1338 static void drop_parsed_module_refcounts(unsigned long subsys_bits)
1340 int i;
1341 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1342 unsigned long bit = 1UL << i;
1344 if (!(bit & subsys_bits))
1345 continue;
1346 module_put(subsys[i]->module);
1350 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1352 int ret = 0;
1353 struct cgroupfs_root *root = sb->s_fs_info;
1354 struct cgroup *cgrp = &root->top_cgroup;
1355 struct cgroup_sb_opts opts;
1357 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1358 mutex_lock(&cgroup_mutex);
1359 mutex_lock(&cgroup_root_mutex);
1361 /* See what subsystems are wanted */
1362 ret = parse_cgroupfs_options(data, &opts);
1363 if (ret)
1364 goto out_unlock;
1366 /* See feature-removal-schedule.txt */
1367 if (opts.subsys_bits != root->actual_subsys_bits || opts.release_agent)
1368 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1369 task_tgid_nr(current), current->comm);
1371 /* Don't allow flags or name to change at remount */
1372 if (opts.flags != root->flags ||
1373 (opts.name && strcmp(opts.name, root->name))) {
1374 ret = -EINVAL;
1375 drop_parsed_module_refcounts(opts.subsys_bits);
1376 goto out_unlock;
1379 ret = rebind_subsystems(root, opts.subsys_bits);
1380 if (ret) {
1381 drop_parsed_module_refcounts(opts.subsys_bits);
1382 goto out_unlock;
1385 /* clear out any existing files and repopulate subsystem files */
1386 cgroup_clear_directory(cgrp->dentry);
1387 cgroup_populate_dir(cgrp);
1389 if (opts.release_agent)
1390 strcpy(root->release_agent_path, opts.release_agent);
1391 out_unlock:
1392 kfree(opts.release_agent);
1393 kfree(opts.name);
1394 mutex_unlock(&cgroup_root_mutex);
1395 mutex_unlock(&cgroup_mutex);
1396 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1397 return ret;
1400 static const struct super_operations cgroup_ops = {
1401 .statfs = simple_statfs,
1402 .drop_inode = generic_delete_inode,
1403 .show_options = cgroup_show_options,
1404 .remount_fs = cgroup_remount,
1407 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1409 INIT_LIST_HEAD(&cgrp->sibling);
1410 INIT_LIST_HEAD(&cgrp->children);
1411 INIT_LIST_HEAD(&cgrp->files);
1412 INIT_LIST_HEAD(&cgrp->css_sets);
1413 INIT_LIST_HEAD(&cgrp->release_list);
1414 INIT_LIST_HEAD(&cgrp->pidlists);
1415 mutex_init(&cgrp->pidlist_mutex);
1416 INIT_LIST_HEAD(&cgrp->event_list);
1417 spin_lock_init(&cgrp->event_list_lock);
1420 static void init_cgroup_root(struct cgroupfs_root *root)
1422 struct cgroup *cgrp = &root->top_cgroup;
1424 INIT_LIST_HEAD(&root->subsys_list);
1425 INIT_LIST_HEAD(&root->root_list);
1426 INIT_LIST_HEAD(&root->allcg_list);
1427 root->number_of_cgroups = 1;
1428 cgrp->root = root;
1429 cgrp->top_cgroup = cgrp;
1430 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
1431 init_cgroup_housekeeping(cgrp);
1434 static bool init_root_id(struct cgroupfs_root *root)
1436 int ret = 0;
1438 do {
1439 if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1440 return false;
1441 spin_lock(&hierarchy_id_lock);
1442 /* Try to allocate the next unused ID */
1443 ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1444 &root->hierarchy_id);
1445 if (ret == -ENOSPC)
1446 /* Try again starting from 0 */
1447 ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1448 if (!ret) {
1449 next_hierarchy_id = root->hierarchy_id + 1;
1450 } else if (ret != -EAGAIN) {
1451 /* Can only get here if the 31-bit IDR is full ... */
1452 BUG_ON(ret);
1454 spin_unlock(&hierarchy_id_lock);
1455 } while (ret);
1456 return true;
1459 static int cgroup_test_super(struct super_block *sb, void *data)
1461 struct cgroup_sb_opts *opts = data;
1462 struct cgroupfs_root *root = sb->s_fs_info;
1464 /* If we asked for a name then it must match */
1465 if (opts->name && strcmp(opts->name, root->name))
1466 return 0;
1469 * If we asked for subsystems (or explicitly for no
1470 * subsystems) then they must match
1472 if ((opts->subsys_bits || opts->none)
1473 && (opts->subsys_bits != root->subsys_bits))
1474 return 0;
1476 return 1;
1479 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1481 struct cgroupfs_root *root;
1483 if (!opts->subsys_bits && !opts->none)
1484 return NULL;
1486 root = kzalloc(sizeof(*root), GFP_KERNEL);
1487 if (!root)
1488 return ERR_PTR(-ENOMEM);
1490 if (!init_root_id(root)) {
1491 kfree(root);
1492 return ERR_PTR(-ENOMEM);
1494 init_cgroup_root(root);
1496 root->subsys_bits = opts->subsys_bits;
1497 root->flags = opts->flags;
1498 if (opts->release_agent)
1499 strcpy(root->release_agent_path, opts->release_agent);
1500 if (opts->name)
1501 strcpy(root->name, opts->name);
1502 if (opts->clone_children)
1503 set_bit(CGRP_CLONE_CHILDREN, &root->top_cgroup.flags);
1504 return root;
1507 static void cgroup_drop_root(struct cgroupfs_root *root)
1509 if (!root)
1510 return;
1512 BUG_ON(!root->hierarchy_id);
1513 spin_lock(&hierarchy_id_lock);
1514 ida_remove(&hierarchy_ida, root->hierarchy_id);
1515 spin_unlock(&hierarchy_id_lock);
1516 kfree(root);
1519 static int cgroup_set_super(struct super_block *sb, void *data)
1521 int ret;
1522 struct cgroup_sb_opts *opts = data;
1524 /* If we don't have a new root, we can't set up a new sb */
1525 if (!opts->new_root)
1526 return -EINVAL;
1528 BUG_ON(!opts->subsys_bits && !opts->none);
1530 ret = set_anon_super(sb, NULL);
1531 if (ret)
1532 return ret;
1534 sb->s_fs_info = opts->new_root;
1535 opts->new_root->sb = sb;
1537 sb->s_blocksize = PAGE_CACHE_SIZE;
1538 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1539 sb->s_magic = CGROUP_SUPER_MAGIC;
1540 sb->s_op = &cgroup_ops;
1542 return 0;
1545 static int cgroup_get_rootdir(struct super_block *sb)
1547 static const struct dentry_operations cgroup_dops = {
1548 .d_iput = cgroup_diput,
1549 .d_delete = cgroup_delete,
1550 .d_release = cgroup_d_release,
1553 struct inode *inode =
1554 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1556 if (!inode)
1557 return -ENOMEM;
1559 inode->i_fop = &simple_dir_operations;
1560 inode->i_op = &cgroup_dir_inode_operations;
1561 /* directories start off with i_nlink == 2 (for "." entry) */
1562 inc_nlink(inode);
1563 sb->s_root = d_make_root(inode);
1564 if (!sb->s_root)
1565 return -ENOMEM;
1566 /* for everything else we want ->d_op set */
1567 sb->s_d_op = &cgroup_dops;
1568 return 0;
1571 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1572 int flags, const char *unused_dev_name,
1573 void *data)
1575 struct cgroup_sb_opts opts;
1576 struct cgroupfs_root *root;
1577 int ret = 0;
1578 struct super_block *sb;
1579 struct cgroupfs_root *new_root;
1580 struct inode *inode;
1582 /* First find the desired set of subsystems */
1583 mutex_lock(&cgroup_mutex);
1584 ret = parse_cgroupfs_options(data, &opts);
1585 mutex_unlock(&cgroup_mutex);
1586 if (ret)
1587 goto out_err;
1590 * Allocate a new cgroup root. We may not need it if we're
1591 * reusing an existing hierarchy.
1593 new_root = cgroup_root_from_opts(&opts);
1594 if (IS_ERR(new_root)) {
1595 ret = PTR_ERR(new_root);
1596 goto drop_modules;
1598 opts.new_root = new_root;
1600 /* Locate an existing or new sb for this hierarchy */
1601 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, &opts);
1602 if (IS_ERR(sb)) {
1603 ret = PTR_ERR(sb);
1604 cgroup_drop_root(opts.new_root);
1605 goto drop_modules;
1608 root = sb->s_fs_info;
1609 BUG_ON(!root);
1610 if (root == opts.new_root) {
1611 /* We used the new root structure, so this is a new hierarchy */
1612 struct list_head tmp_cg_links;
1613 struct cgroup *root_cgrp = &root->top_cgroup;
1614 struct cgroupfs_root *existing_root;
1615 const struct cred *cred;
1616 int i;
1618 BUG_ON(sb->s_root != NULL);
1620 ret = cgroup_get_rootdir(sb);
1621 if (ret)
1622 goto drop_new_super;
1623 inode = sb->s_root->d_inode;
1625 mutex_lock(&inode->i_mutex);
1626 mutex_lock(&cgroup_mutex);
1627 mutex_lock(&cgroup_root_mutex);
1629 /* Check for name clashes with existing mounts */
1630 ret = -EBUSY;
1631 if (strlen(root->name))
1632 for_each_active_root(existing_root)
1633 if (!strcmp(existing_root->name, root->name))
1634 goto unlock_drop;
1637 * We're accessing css_set_count without locking
1638 * css_set_lock here, but that's OK - it can only be
1639 * increased by someone holding cgroup_lock, and
1640 * that's us. The worst that can happen is that we
1641 * have some link structures left over
1643 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1644 if (ret)
1645 goto unlock_drop;
1647 ret = rebind_subsystems(root, root->subsys_bits);
1648 if (ret == -EBUSY) {
1649 free_cg_links(&tmp_cg_links);
1650 goto unlock_drop;
1653 * There must be no failure case after here, since rebinding
1654 * takes care of subsystems' refcounts, which are explicitly
1655 * dropped in the failure exit path.
1658 /* EBUSY should be the only error here */
1659 BUG_ON(ret);
1661 list_add(&root->root_list, &roots);
1662 root_count++;
1664 sb->s_root->d_fsdata = root_cgrp;
1665 root->top_cgroup.dentry = sb->s_root;
1667 /* Link the top cgroup in this hierarchy into all
1668 * the css_set objects */
1669 write_lock(&css_set_lock);
1670 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
1671 struct hlist_head *hhead = &css_set_table[i];
1672 struct hlist_node *node;
1673 struct css_set *cg;
1675 hlist_for_each_entry(cg, node, hhead, hlist)
1676 link_css_set(&tmp_cg_links, cg, root_cgrp);
1678 write_unlock(&css_set_lock);
1680 free_cg_links(&tmp_cg_links);
1682 BUG_ON(!list_empty(&root_cgrp->sibling));
1683 BUG_ON(!list_empty(&root_cgrp->children));
1684 BUG_ON(root->number_of_cgroups != 1);
1686 cred = override_creds(&init_cred);
1687 cgroup_populate_dir(root_cgrp);
1688 revert_creds(cred);
1689 mutex_unlock(&cgroup_root_mutex);
1690 mutex_unlock(&cgroup_mutex);
1691 mutex_unlock(&inode->i_mutex);
1692 } else {
1694 * We re-used an existing hierarchy - the new root (if
1695 * any) is not needed
1697 cgroup_drop_root(opts.new_root);
1698 /* no subsys rebinding, so refcounts don't change */
1699 drop_parsed_module_refcounts(opts.subsys_bits);
1702 kfree(opts.release_agent);
1703 kfree(opts.name);
1704 return dget(sb->s_root);
1706 unlock_drop:
1707 mutex_unlock(&cgroup_root_mutex);
1708 mutex_unlock(&cgroup_mutex);
1709 mutex_unlock(&inode->i_mutex);
1710 drop_new_super:
1711 deactivate_locked_super(sb);
1712 drop_modules:
1713 drop_parsed_module_refcounts(opts.subsys_bits);
1714 out_err:
1715 kfree(opts.release_agent);
1716 kfree(opts.name);
1717 return ERR_PTR(ret);
1720 static void cgroup_kill_sb(struct super_block *sb) {
1721 struct cgroupfs_root *root = sb->s_fs_info;
1722 struct cgroup *cgrp = &root->top_cgroup;
1723 int ret;
1724 struct cg_cgroup_link *link;
1725 struct cg_cgroup_link *saved_link;
1727 BUG_ON(!root);
1729 BUG_ON(root->number_of_cgroups != 1);
1730 BUG_ON(!list_empty(&cgrp->children));
1731 BUG_ON(!list_empty(&cgrp->sibling));
1733 mutex_lock(&cgroup_mutex);
1734 mutex_lock(&cgroup_root_mutex);
1736 /* Rebind all subsystems back to the default hierarchy */
1737 ret = rebind_subsystems(root, 0);
1738 /* Shouldn't be able to fail ... */
1739 BUG_ON(ret);
1742 * Release all the links from css_sets to this hierarchy's
1743 * root cgroup
1745 write_lock(&css_set_lock);
1747 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1748 cgrp_link_list) {
1749 list_del(&link->cg_link_list);
1750 list_del(&link->cgrp_link_list);
1751 kfree(link);
1753 write_unlock(&css_set_lock);
1755 if (!list_empty(&root->root_list)) {
1756 list_del(&root->root_list);
1757 root_count--;
1760 mutex_unlock(&cgroup_root_mutex);
1761 mutex_unlock(&cgroup_mutex);
1763 kill_litter_super(sb);
1764 cgroup_drop_root(root);
1767 static struct file_system_type cgroup_fs_type = {
1768 .name = "cgroup",
1769 .mount = cgroup_mount,
1770 .kill_sb = cgroup_kill_sb,
1773 static struct kobject *cgroup_kobj;
1776 * cgroup_path - generate the path of a cgroup
1777 * @cgrp: the cgroup in question
1778 * @buf: the buffer to write the path into
1779 * @buflen: the length of the buffer
1781 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1782 * reference. Writes path of cgroup into buf. Returns 0 on success,
1783 * -errno on error.
1785 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1787 char *start;
1788 struct dentry *dentry = rcu_dereference_check(cgrp->dentry,
1789 cgroup_lock_is_held());
1791 if (!dentry || cgrp == dummytop) {
1793 * Inactive subsystems have no dentry for their root
1794 * cgroup
1796 strcpy(buf, "/");
1797 return 0;
1800 start = buf + buflen;
1802 *--start = '\0';
1803 for (;;) {
1804 int len = dentry->d_name.len;
1806 if ((start -= len) < buf)
1807 return -ENAMETOOLONG;
1808 memcpy(start, dentry->d_name.name, len);
1809 cgrp = cgrp->parent;
1810 if (!cgrp)
1811 break;
1813 dentry = rcu_dereference_check(cgrp->dentry,
1814 cgroup_lock_is_held());
1815 if (!cgrp->parent)
1816 continue;
1817 if (--start < buf)
1818 return -ENAMETOOLONG;
1819 *start = '/';
1821 memmove(buf, start, buf + buflen - start);
1822 return 0;
1824 EXPORT_SYMBOL_GPL(cgroup_path);
1827 * Control Group taskset
1829 struct task_and_cgroup {
1830 struct task_struct *task;
1831 struct cgroup *cgrp;
1832 struct css_set *cg;
1835 struct cgroup_taskset {
1836 struct task_and_cgroup single;
1837 struct flex_array *tc_array;
1838 int tc_array_len;
1839 int idx;
1840 struct cgroup *cur_cgrp;
1844 * cgroup_taskset_first - reset taskset and return the first task
1845 * @tset: taskset of interest
1847 * @tset iteration is initialized and the first task is returned.
1849 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1851 if (tset->tc_array) {
1852 tset->idx = 0;
1853 return cgroup_taskset_next(tset);
1854 } else {
1855 tset->cur_cgrp = tset->single.cgrp;
1856 return tset->single.task;
1859 EXPORT_SYMBOL_GPL(cgroup_taskset_first);
1862 * cgroup_taskset_next - iterate to the next task in taskset
1863 * @tset: taskset of interest
1865 * Return the next task in @tset. Iteration must have been initialized
1866 * with cgroup_taskset_first().
1868 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1870 struct task_and_cgroup *tc;
1872 if (!tset->tc_array || tset->idx >= tset->tc_array_len)
1873 return NULL;
1875 tc = flex_array_get(tset->tc_array, tset->idx++);
1876 tset->cur_cgrp = tc->cgrp;
1877 return tc->task;
1879 EXPORT_SYMBOL_GPL(cgroup_taskset_next);
1882 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1883 * @tset: taskset of interest
1885 * Return the cgroup for the current (last returned) task of @tset. This
1886 * function must be preceded by either cgroup_taskset_first() or
1887 * cgroup_taskset_next().
1889 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset)
1891 return tset->cur_cgrp;
1893 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup);
1896 * cgroup_taskset_size - return the number of tasks in taskset
1897 * @tset: taskset of interest
1899 int cgroup_taskset_size(struct cgroup_taskset *tset)
1901 return tset->tc_array ? tset->tc_array_len : 1;
1903 EXPORT_SYMBOL_GPL(cgroup_taskset_size);
1907 * cgroup_task_migrate - move a task from one cgroup to another.
1909 * 'guarantee' is set if the caller promises that a new css_set for the task
1910 * will already exist. If not set, this function might sleep, and can fail with
1911 * -ENOMEM. Must be called with cgroup_mutex and threadgroup locked.
1913 static void cgroup_task_migrate(struct cgroup *cgrp, struct cgroup *oldcgrp,
1914 struct task_struct *tsk, struct css_set *newcg)
1916 struct css_set *oldcg;
1919 * We are synchronized through threadgroup_lock() against PF_EXITING
1920 * setting such that we can't race against cgroup_exit() changing the
1921 * css_set to init_css_set and dropping the old one.
1923 WARN_ON_ONCE(tsk->flags & PF_EXITING);
1924 oldcg = tsk->cgroups;
1926 task_lock(tsk);
1927 rcu_assign_pointer(tsk->cgroups, newcg);
1928 task_unlock(tsk);
1930 /* Update the css_set linked lists if we're using them */
1931 write_lock(&css_set_lock);
1932 if (!list_empty(&tsk->cg_list))
1933 list_move(&tsk->cg_list, &newcg->tasks);
1934 write_unlock(&css_set_lock);
1937 * We just gained a reference on oldcg by taking it from the task. As
1938 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1939 * it here; it will be freed under RCU.
1941 put_css_set(oldcg);
1943 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1947 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1948 * @cgrp: the cgroup the task is attaching to
1949 * @tsk: the task to be attached
1951 * Call with cgroup_mutex and threadgroup locked. May take task_lock of
1952 * @tsk during call.
1954 int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1956 int retval = 0;
1957 struct cgroup_subsys *ss, *failed_ss = NULL;
1958 struct cgroup *oldcgrp;
1959 struct cgroupfs_root *root = cgrp->root;
1960 struct cgroup_taskset tset = { };
1961 struct css_set *newcg;
1963 /* @tsk either already exited or can't exit until the end */
1964 if (tsk->flags & PF_EXITING)
1965 return -ESRCH;
1967 /* Nothing to do if the task is already in that cgroup */
1968 oldcgrp = task_cgroup_from_root(tsk, root);
1969 if (cgrp == oldcgrp)
1970 return 0;
1972 tset.single.task = tsk;
1973 tset.single.cgrp = oldcgrp;
1975 for_each_subsys(root, ss) {
1976 if (ss->can_attach) {
1977 retval = ss->can_attach(cgrp, &tset);
1978 if (retval) {
1980 * Remember on which subsystem the can_attach()
1981 * failed, so that we only call cancel_attach()
1982 * against the subsystems whose can_attach()
1983 * succeeded. (See below)
1985 failed_ss = ss;
1986 goto out;
1991 newcg = find_css_set(tsk->cgroups, cgrp);
1992 if (!newcg) {
1993 retval = -ENOMEM;
1994 goto out;
1997 cgroup_task_migrate(cgrp, oldcgrp, tsk, newcg);
1999 for_each_subsys(root, ss) {
2000 if (ss->attach)
2001 ss->attach(cgrp, &tset);
2004 synchronize_rcu();
2007 * wake up rmdir() waiter. the rmdir should fail since the cgroup
2008 * is no longer empty.
2010 cgroup_wakeup_rmdir_waiter(cgrp);
2011 out:
2012 if (retval) {
2013 for_each_subsys(root, ss) {
2014 if (ss == failed_ss)
2016 * This subsystem was the one that failed the
2017 * can_attach() check earlier, so we don't need
2018 * to call cancel_attach() against it or any
2019 * remaining subsystems.
2021 break;
2022 if (ss->cancel_attach)
2023 ss->cancel_attach(cgrp, &tset);
2026 return retval;
2030 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2031 * @from: attach to all cgroups of a given task
2032 * @tsk: the task to be attached
2034 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2036 struct cgroupfs_root *root;
2037 int retval = 0;
2039 cgroup_lock();
2040 for_each_active_root(root) {
2041 struct cgroup *from_cg = task_cgroup_from_root(from, root);
2043 retval = cgroup_attach_task(from_cg, tsk);
2044 if (retval)
2045 break;
2047 cgroup_unlock();
2049 return retval;
2051 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2054 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
2055 * @cgrp: the cgroup to attach to
2056 * @leader: the threadgroup leader task_struct of the group to be attached
2058 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
2059 * task_lock of each thread in leader's threadgroup individually in turn.
2061 static int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader)
2063 int retval, i, group_size;
2064 struct cgroup_subsys *ss, *failed_ss = NULL;
2065 /* guaranteed to be initialized later, but the compiler needs this */
2066 struct cgroupfs_root *root = cgrp->root;
2067 /* threadgroup list cursor and array */
2068 struct task_struct *tsk;
2069 struct task_and_cgroup *tc;
2070 struct flex_array *group;
2071 struct cgroup_taskset tset = { };
2074 * step 0: in order to do expensive, possibly blocking operations for
2075 * every thread, we cannot iterate the thread group list, since it needs
2076 * rcu or tasklist locked. instead, build an array of all threads in the
2077 * group - group_rwsem prevents new threads from appearing, and if
2078 * threads exit, this will just be an over-estimate.
2080 group_size = get_nr_threads(leader);
2081 /* flex_array supports very large thread-groups better than kmalloc. */
2082 group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
2083 if (!group)
2084 return -ENOMEM;
2085 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2086 retval = flex_array_prealloc(group, 0, group_size - 1, GFP_KERNEL);
2087 if (retval)
2088 goto out_free_group_list;
2090 tsk = leader;
2091 i = 0;
2093 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2094 * already PF_EXITING could be freed from underneath us unless we
2095 * take an rcu_read_lock.
2097 rcu_read_lock();
2098 do {
2099 struct task_and_cgroup ent;
2101 /* @tsk either already exited or can't exit until the end */
2102 if (tsk->flags & PF_EXITING)
2103 continue;
2105 /* as per above, nr_threads may decrease, but not increase. */
2106 BUG_ON(i >= group_size);
2107 ent.task = tsk;
2108 ent.cgrp = task_cgroup_from_root(tsk, root);
2109 /* nothing to do if this task is already in the cgroup */
2110 if (ent.cgrp == cgrp)
2111 continue;
2113 * saying GFP_ATOMIC has no effect here because we did prealloc
2114 * earlier, but it's good form to communicate our expectations.
2116 retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
2117 BUG_ON(retval != 0);
2118 i++;
2119 } while_each_thread(leader, tsk);
2120 rcu_read_unlock();
2121 /* remember the number of threads in the array for later. */
2122 group_size = i;
2123 tset.tc_array = group;
2124 tset.tc_array_len = group_size;
2126 /* methods shouldn't be called if no task is actually migrating */
2127 retval = 0;
2128 if (!group_size)
2129 goto out_free_group_list;
2132 * step 1: check that we can legitimately attach to the cgroup.
2134 for_each_subsys(root, ss) {
2135 if (ss->can_attach) {
2136 retval = ss->can_attach(cgrp, &tset);
2137 if (retval) {
2138 failed_ss = ss;
2139 goto out_cancel_attach;
2145 * step 2: make sure css_sets exist for all threads to be migrated.
2146 * we use find_css_set, which allocates a new one if necessary.
2148 for (i = 0; i < group_size; i++) {
2149 tc = flex_array_get(group, i);
2150 tc->cg = find_css_set(tc->task->cgroups, cgrp);
2151 if (!tc->cg) {
2152 retval = -ENOMEM;
2153 goto out_put_css_set_refs;
2158 * step 3: now that we're guaranteed success wrt the css_sets,
2159 * proceed to move all tasks to the new cgroup. There are no
2160 * failure cases after here, so this is the commit point.
2162 for (i = 0; i < group_size; i++) {
2163 tc = flex_array_get(group, i);
2164 cgroup_task_migrate(cgrp, tc->cgrp, tc->task, tc->cg);
2166 /* nothing is sensitive to fork() after this point. */
2169 * step 4: do subsystem attach callbacks.
2171 for_each_subsys(root, ss) {
2172 if (ss->attach)
2173 ss->attach(cgrp, &tset);
2177 * step 5: success! and cleanup
2179 synchronize_rcu();
2180 cgroup_wakeup_rmdir_waiter(cgrp);
2181 retval = 0;
2182 out_put_css_set_refs:
2183 if (retval) {
2184 for (i = 0; i < group_size; i++) {
2185 tc = flex_array_get(group, i);
2186 if (!tc->cg)
2187 break;
2188 put_css_set(tc->cg);
2191 out_cancel_attach:
2192 if (retval) {
2193 for_each_subsys(root, ss) {
2194 if (ss == failed_ss)
2195 break;
2196 if (ss->cancel_attach)
2197 ss->cancel_attach(cgrp, &tset);
2200 out_free_group_list:
2201 flex_array_free(group);
2202 return retval;
2206 * Find the task_struct of the task to attach by vpid and pass it along to the
2207 * function to attach either it or all tasks in its threadgroup. Will lock
2208 * cgroup_mutex and threadgroup; may take task_lock of task.
2210 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2212 struct task_struct *tsk;
2213 const struct cred *cred = current_cred(), *tcred;
2214 int ret;
2216 if (!cgroup_lock_live_group(cgrp))
2217 return -ENODEV;
2219 retry_find_task:
2220 rcu_read_lock();
2221 if (pid) {
2222 tsk = find_task_by_vpid(pid);
2223 if (!tsk) {
2224 rcu_read_unlock();
2225 ret= -ESRCH;
2226 goto out_unlock_cgroup;
2229 * even if we're attaching all tasks in the thread group, we
2230 * only need to check permissions on one of them.
2232 tcred = __task_cred(tsk);
2233 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2234 !uid_eq(cred->euid, tcred->uid) &&
2235 !uid_eq(cred->euid, tcred->suid)) {
2236 rcu_read_unlock();
2237 ret = -EACCES;
2238 goto out_unlock_cgroup;
2240 } else
2241 tsk = current;
2243 if (threadgroup)
2244 tsk = tsk->group_leader;
2247 * Workqueue threads may acquire PF_THREAD_BOUND and become
2248 * trapped in a cpuset, or RT worker may be born in a cgroup
2249 * with no rt_runtime allocated. Just say no.
2251 if (tsk == kthreadd_task || (tsk->flags & PF_THREAD_BOUND)) {
2252 ret = -EINVAL;
2253 rcu_read_unlock();
2254 goto out_unlock_cgroup;
2257 get_task_struct(tsk);
2258 rcu_read_unlock();
2260 threadgroup_lock(tsk);
2261 if (threadgroup) {
2262 if (!thread_group_leader(tsk)) {
2264 * a race with de_thread from another thread's exec()
2265 * may strip us of our leadership, if this happens,
2266 * there is no choice but to throw this task away and
2267 * try again; this is
2268 * "double-double-toil-and-trouble-check locking".
2270 threadgroup_unlock(tsk);
2271 put_task_struct(tsk);
2272 goto retry_find_task;
2274 ret = cgroup_attach_proc(cgrp, tsk);
2275 } else
2276 ret = cgroup_attach_task(cgrp, tsk);
2277 threadgroup_unlock(tsk);
2279 put_task_struct(tsk);
2280 out_unlock_cgroup:
2281 cgroup_unlock();
2282 return ret;
2285 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2287 return attach_task_by_pid(cgrp, pid, false);
2290 static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2292 return attach_task_by_pid(cgrp, tgid, true);
2296 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
2297 * @cgrp: the cgroup to be checked for liveness
2299 * On success, returns true; the lock should be later released with
2300 * cgroup_unlock(). On failure returns false with no lock held.
2302 bool cgroup_lock_live_group(struct cgroup *cgrp)
2304 mutex_lock(&cgroup_mutex);
2305 if (cgroup_is_removed(cgrp)) {
2306 mutex_unlock(&cgroup_mutex);
2307 return false;
2309 return true;
2311 EXPORT_SYMBOL_GPL(cgroup_lock_live_group);
2313 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2314 const char *buffer)
2316 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2317 if (strlen(buffer) >= PATH_MAX)
2318 return -EINVAL;
2319 if (!cgroup_lock_live_group(cgrp))
2320 return -ENODEV;
2321 mutex_lock(&cgroup_root_mutex);
2322 strcpy(cgrp->root->release_agent_path, buffer);
2323 mutex_unlock(&cgroup_root_mutex);
2324 cgroup_unlock();
2325 return 0;
2328 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2329 struct seq_file *seq)
2331 if (!cgroup_lock_live_group(cgrp))
2332 return -ENODEV;
2333 seq_puts(seq, cgrp->root->release_agent_path);
2334 seq_putc(seq, '\n');
2335 cgroup_unlock();
2336 return 0;
2339 /* A buffer size big enough for numbers or short strings */
2340 #define CGROUP_LOCAL_BUFFER_SIZE 64
2342 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2343 struct file *file,
2344 const char __user *userbuf,
2345 size_t nbytes, loff_t *unused_ppos)
2347 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2348 int retval = 0;
2349 char *end;
2351 if (!nbytes)
2352 return -EINVAL;
2353 if (nbytes >= sizeof(buffer))
2354 return -E2BIG;
2355 if (copy_from_user(buffer, userbuf, nbytes))
2356 return -EFAULT;
2358 buffer[nbytes] = 0; /* nul-terminate */
2359 if (cft->write_u64) {
2360 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2361 if (*end)
2362 return -EINVAL;
2363 retval = cft->write_u64(cgrp, cft, val);
2364 } else {
2365 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2366 if (*end)
2367 return -EINVAL;
2368 retval = cft->write_s64(cgrp, cft, val);
2370 if (!retval)
2371 retval = nbytes;
2372 return retval;
2375 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2376 struct file *file,
2377 const char __user *userbuf,
2378 size_t nbytes, loff_t *unused_ppos)
2380 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2381 int retval = 0;
2382 size_t max_bytes = cft->max_write_len;
2383 char *buffer = local_buffer;
2385 if (!max_bytes)
2386 max_bytes = sizeof(local_buffer) - 1;
2387 if (nbytes >= max_bytes)
2388 return -E2BIG;
2389 /* Allocate a dynamic buffer if we need one */
2390 if (nbytes >= sizeof(local_buffer)) {
2391 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2392 if (buffer == NULL)
2393 return -ENOMEM;
2395 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2396 retval = -EFAULT;
2397 goto out;
2400 buffer[nbytes] = 0; /* nul-terminate */
2401 retval = cft->write_string(cgrp, cft, strstrip(buffer));
2402 if (!retval)
2403 retval = nbytes;
2404 out:
2405 if (buffer != local_buffer)
2406 kfree(buffer);
2407 return retval;
2410 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2411 size_t nbytes, loff_t *ppos)
2413 struct cftype *cft = __d_cft(file->f_dentry);
2414 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2416 if (cgroup_is_removed(cgrp))
2417 return -ENODEV;
2418 if (cft->write)
2419 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2420 if (cft->write_u64 || cft->write_s64)
2421 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2422 if (cft->write_string)
2423 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2424 if (cft->trigger) {
2425 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2426 return ret ? ret : nbytes;
2428 return -EINVAL;
2431 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2432 struct file *file,
2433 char __user *buf, size_t nbytes,
2434 loff_t *ppos)
2436 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2437 u64 val = cft->read_u64(cgrp, cft);
2438 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2440 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2443 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2444 struct file *file,
2445 char __user *buf, size_t nbytes,
2446 loff_t *ppos)
2448 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2449 s64 val = cft->read_s64(cgrp, cft);
2450 int len = sprintf(tmp, "%lld\n", (long long) val);
2452 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2455 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2456 size_t nbytes, loff_t *ppos)
2458 struct cftype *cft = __d_cft(file->f_dentry);
2459 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2461 if (cgroup_is_removed(cgrp))
2462 return -ENODEV;
2464 if (cft->read)
2465 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2466 if (cft->read_u64)
2467 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2468 if (cft->read_s64)
2469 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2470 return -EINVAL;
2474 * seqfile ops/methods for returning structured data. Currently just
2475 * supports string->u64 maps, but can be extended in future.
2478 struct cgroup_seqfile_state {
2479 struct cftype *cft;
2480 struct cgroup *cgroup;
2483 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2485 struct seq_file *sf = cb->state;
2486 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2489 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2491 struct cgroup_seqfile_state *state = m->private;
2492 struct cftype *cft = state->cft;
2493 if (cft->read_map) {
2494 struct cgroup_map_cb cb = {
2495 .fill = cgroup_map_add,
2496 .state = m,
2498 return cft->read_map(state->cgroup, cft, &cb);
2500 return cft->read_seq_string(state->cgroup, cft, m);
2503 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2505 struct seq_file *seq = file->private_data;
2506 kfree(seq->private);
2507 return single_release(inode, file);
2510 static const struct file_operations cgroup_seqfile_operations = {
2511 .read = seq_read,
2512 .write = cgroup_file_write,
2513 .llseek = seq_lseek,
2514 .release = cgroup_seqfile_release,
2517 static int cgroup_file_open(struct inode *inode, struct file *file)
2519 int err;
2520 struct cftype *cft;
2522 err = generic_file_open(inode, file);
2523 if (err)
2524 return err;
2525 cft = __d_cft(file->f_dentry);
2527 if (cft->read_map || cft->read_seq_string) {
2528 struct cgroup_seqfile_state *state =
2529 kzalloc(sizeof(*state), GFP_USER);
2530 if (!state)
2531 return -ENOMEM;
2532 state->cft = cft;
2533 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2534 file->f_op = &cgroup_seqfile_operations;
2535 err = single_open(file, cgroup_seqfile_show, state);
2536 if (err < 0)
2537 kfree(state);
2538 } else if (cft->open)
2539 err = cft->open(inode, file);
2540 else
2541 err = 0;
2543 return err;
2546 static int cgroup_file_release(struct inode *inode, struct file *file)
2548 struct cftype *cft = __d_cft(file->f_dentry);
2549 if (cft->release)
2550 return cft->release(inode, file);
2551 return 0;
2555 * cgroup_rename - Only allow simple rename of directories in place.
2557 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2558 struct inode *new_dir, struct dentry *new_dentry)
2560 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2561 return -ENOTDIR;
2562 if (new_dentry->d_inode)
2563 return -EEXIST;
2564 if (old_dir != new_dir)
2565 return -EIO;
2566 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2569 static const struct file_operations cgroup_file_operations = {
2570 .read = cgroup_file_read,
2571 .write = cgroup_file_write,
2572 .llseek = generic_file_llseek,
2573 .open = cgroup_file_open,
2574 .release = cgroup_file_release,
2577 static const struct inode_operations cgroup_dir_inode_operations = {
2578 .lookup = cgroup_lookup,
2579 .mkdir = cgroup_mkdir,
2580 .rmdir = cgroup_rmdir,
2581 .rename = cgroup_rename,
2584 static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
2586 if (dentry->d_name.len > NAME_MAX)
2587 return ERR_PTR(-ENAMETOOLONG);
2588 d_add(dentry, NULL);
2589 return NULL;
2593 * Check if a file is a control file
2595 static inline struct cftype *__file_cft(struct file *file)
2597 if (file->f_dentry->d_inode->i_fop != &cgroup_file_operations)
2598 return ERR_PTR(-EINVAL);
2599 return __d_cft(file->f_dentry);
2602 static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2603 struct super_block *sb)
2605 struct inode *inode;
2607 if (!dentry)
2608 return -ENOENT;
2609 if (dentry->d_inode)
2610 return -EEXIST;
2612 inode = cgroup_new_inode(mode, sb);
2613 if (!inode)
2614 return -ENOMEM;
2616 if (S_ISDIR(mode)) {
2617 inode->i_op = &cgroup_dir_inode_operations;
2618 inode->i_fop = &simple_dir_operations;
2620 /* start off with i_nlink == 2 (for "." entry) */
2621 inc_nlink(inode);
2623 /* start with the directory inode held, so that we can
2624 * populate it without racing with another mkdir */
2625 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2626 } else if (S_ISREG(mode)) {
2627 inode->i_size = 0;
2628 inode->i_fop = &cgroup_file_operations;
2630 d_instantiate(dentry, inode);
2631 dget(dentry); /* Extra count - pin the dentry in core */
2632 return 0;
2636 * cgroup_create_dir - create a directory for an object.
2637 * @cgrp: the cgroup we create the directory for. It must have a valid
2638 * ->parent field. And we are going to fill its ->dentry field.
2639 * @dentry: dentry of the new cgroup
2640 * @mode: mode to set on new directory.
2642 static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
2643 umode_t mode)
2645 struct dentry *parent;
2646 int error = 0;
2648 parent = cgrp->parent->dentry;
2649 error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
2650 if (!error) {
2651 dentry->d_fsdata = cgrp;
2652 inc_nlink(parent->d_inode);
2653 rcu_assign_pointer(cgrp->dentry, dentry);
2654 dget(dentry);
2656 dput(dentry);
2658 return error;
2662 * cgroup_file_mode - deduce file mode of a control file
2663 * @cft: the control file in question
2665 * returns cft->mode if ->mode is not 0
2666 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2667 * returns S_IRUGO if it has only a read handler
2668 * returns S_IWUSR if it has only a write hander
2670 static umode_t cgroup_file_mode(const struct cftype *cft)
2672 umode_t mode = 0;
2674 if (cft->mode)
2675 return cft->mode;
2677 if (cft->read || cft->read_u64 || cft->read_s64 ||
2678 cft->read_map || cft->read_seq_string)
2679 mode |= S_IRUGO;
2681 if (cft->write || cft->write_u64 || cft->write_s64 ||
2682 cft->write_string || cft->trigger)
2683 mode |= S_IWUSR;
2685 return mode;
2688 static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2689 const struct cftype *cft)
2691 struct dentry *dir = cgrp->dentry;
2692 struct cgroup *parent = __d_cgrp(dir);
2693 struct dentry *dentry;
2694 struct cfent *cfe;
2695 int error;
2696 umode_t mode;
2697 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2699 /* does @cft->flags tell us to skip creation on @cgrp? */
2700 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2701 return 0;
2702 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
2703 return 0;
2705 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
2706 strcpy(name, subsys->name);
2707 strcat(name, ".");
2709 strcat(name, cft->name);
2711 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2713 cfe = kzalloc(sizeof(*cfe), GFP_KERNEL);
2714 if (!cfe)
2715 return -ENOMEM;
2717 dentry = lookup_one_len(name, dir, strlen(name));
2718 if (IS_ERR(dentry)) {
2719 error = PTR_ERR(dentry);
2720 goto out;
2723 mode = cgroup_file_mode(cft);
2724 error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb);
2725 if (!error) {
2726 cfe->type = (void *)cft;
2727 cfe->dentry = dentry;
2728 dentry->d_fsdata = cfe;
2729 list_add_tail(&cfe->node, &parent->files);
2730 cfe = NULL;
2732 dput(dentry);
2733 out:
2734 kfree(cfe);
2735 return error;
2738 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2739 const struct cftype cfts[], bool is_add)
2741 const struct cftype *cft;
2742 int err, ret = 0;
2744 for (cft = cfts; cft->name[0] != '\0'; cft++) {
2745 if (is_add)
2746 err = cgroup_add_file(cgrp, subsys, cft);
2747 else
2748 err = cgroup_rm_file(cgrp, cft);
2749 if (err) {
2750 pr_warning("cgroup_addrm_files: failed to %s %s, err=%d\n",
2751 is_add ? "add" : "remove", cft->name, err);
2752 ret = err;
2755 return ret;
2758 static DEFINE_MUTEX(cgroup_cft_mutex);
2760 static void cgroup_cfts_prepare(void)
2761 __acquires(&cgroup_cft_mutex) __acquires(&cgroup_mutex)
2764 * Thanks to the entanglement with vfs inode locking, we can't walk
2765 * the existing cgroups under cgroup_mutex and create files.
2766 * Instead, we increment reference on all cgroups and build list of
2767 * them using @cgrp->cft_q_node. Grab cgroup_cft_mutex to ensure
2768 * exclusive access to the field.
2770 mutex_lock(&cgroup_cft_mutex);
2771 mutex_lock(&cgroup_mutex);
2774 static void cgroup_cfts_commit(struct cgroup_subsys *ss,
2775 const struct cftype *cfts, bool is_add)
2776 __releases(&cgroup_mutex) __releases(&cgroup_cft_mutex)
2778 LIST_HEAD(pending);
2779 struct cgroup *cgrp, *n;
2781 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2782 if (cfts && ss->root != &rootnode) {
2783 list_for_each_entry(cgrp, &ss->root->allcg_list, allcg_node) {
2784 dget(cgrp->dentry);
2785 list_add_tail(&cgrp->cft_q_node, &pending);
2789 mutex_unlock(&cgroup_mutex);
2792 * All new cgroups will see @cfts update on @ss->cftsets. Add/rm
2793 * files for all cgroups which were created before.
2795 list_for_each_entry_safe(cgrp, n, &pending, cft_q_node) {
2796 struct inode *inode = cgrp->dentry->d_inode;
2798 mutex_lock(&inode->i_mutex);
2799 mutex_lock(&cgroup_mutex);
2800 if (!cgroup_is_removed(cgrp))
2801 cgroup_addrm_files(cgrp, ss, cfts, is_add);
2802 mutex_unlock(&cgroup_mutex);
2803 mutex_unlock(&inode->i_mutex);
2805 list_del_init(&cgrp->cft_q_node);
2806 dput(cgrp->dentry);
2809 mutex_unlock(&cgroup_cft_mutex);
2813 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2814 * @ss: target cgroup subsystem
2815 * @cfts: zero-length name terminated array of cftypes
2817 * Register @cfts to @ss. Files described by @cfts are created for all
2818 * existing cgroups to which @ss is attached and all future cgroups will
2819 * have them too. This function can be called anytime whether @ss is
2820 * attached or not.
2822 * Returns 0 on successful registration, -errno on failure. Note that this
2823 * function currently returns 0 as long as @cfts registration is successful
2824 * even if some file creation attempts on existing cgroups fail.
2826 int cgroup_add_cftypes(struct cgroup_subsys *ss, const struct cftype *cfts)
2828 struct cftype_set *set;
2830 set = kzalloc(sizeof(*set), GFP_KERNEL);
2831 if (!set)
2832 return -ENOMEM;
2834 cgroup_cfts_prepare();
2835 set->cfts = cfts;
2836 list_add_tail(&set->node, &ss->cftsets);
2837 cgroup_cfts_commit(ss, cfts, true);
2839 return 0;
2841 EXPORT_SYMBOL_GPL(cgroup_add_cftypes);
2844 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2845 * @ss: target cgroup subsystem
2846 * @cfts: zero-length name terminated array of cftypes
2848 * Unregister @cfts from @ss. Files described by @cfts are removed from
2849 * all existing cgroups to which @ss is attached and all future cgroups
2850 * won't have them either. This function can be called anytime whether @ss
2851 * is attached or not.
2853 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2854 * registered with @ss.
2856 int cgroup_rm_cftypes(struct cgroup_subsys *ss, const struct cftype *cfts)
2858 struct cftype_set *set;
2860 cgroup_cfts_prepare();
2862 list_for_each_entry(set, &ss->cftsets, node) {
2863 if (set->cfts == cfts) {
2864 list_del_init(&set->node);
2865 cgroup_cfts_commit(ss, cfts, false);
2866 return 0;
2870 cgroup_cfts_commit(ss, NULL, false);
2871 return -ENOENT;
2875 * cgroup_task_count - count the number of tasks in a cgroup.
2876 * @cgrp: the cgroup in question
2878 * Return the number of tasks in the cgroup.
2880 int cgroup_task_count(const struct cgroup *cgrp)
2882 int count = 0;
2883 struct cg_cgroup_link *link;
2885 read_lock(&css_set_lock);
2886 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2887 count += atomic_read(&link->cg->refcount);
2889 read_unlock(&css_set_lock);
2890 return count;
2894 * Advance a list_head iterator. The iterator should be positioned at
2895 * the start of a css_set
2897 static void cgroup_advance_iter(struct cgroup *cgrp,
2898 struct cgroup_iter *it)
2900 struct list_head *l = it->cg_link;
2901 struct cg_cgroup_link *link;
2902 struct css_set *cg;
2904 /* Advance to the next non-empty css_set */
2905 do {
2906 l = l->next;
2907 if (l == &cgrp->css_sets) {
2908 it->cg_link = NULL;
2909 return;
2911 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2912 cg = link->cg;
2913 } while (list_empty(&cg->tasks));
2914 it->cg_link = l;
2915 it->task = cg->tasks.next;
2919 * To reduce the fork() overhead for systems that are not actually
2920 * using their cgroups capability, we don't maintain the lists running
2921 * through each css_set to its tasks until we see the list actually
2922 * used - in other words after the first call to cgroup_iter_start().
2924 static void cgroup_enable_task_cg_lists(void)
2926 struct task_struct *p, *g;
2927 write_lock(&css_set_lock);
2928 use_task_css_set_links = 1;
2930 * We need tasklist_lock because RCU is not safe against
2931 * while_each_thread(). Besides, a forking task that has passed
2932 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2933 * is not guaranteed to have its child immediately visible in the
2934 * tasklist if we walk through it with RCU.
2936 read_lock(&tasklist_lock);
2937 do_each_thread(g, p) {
2938 task_lock(p);
2940 * We should check if the process is exiting, otherwise
2941 * it will race with cgroup_exit() in that the list
2942 * entry won't be deleted though the process has exited.
2944 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2945 list_add(&p->cg_list, &p->cgroups->tasks);
2946 task_unlock(p);
2947 } while_each_thread(g, p);
2948 read_unlock(&tasklist_lock);
2949 write_unlock(&css_set_lock);
2952 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
2953 __acquires(css_set_lock)
2956 * The first time anyone tries to iterate across a cgroup,
2957 * we need to enable the list linking each css_set to its
2958 * tasks, and fix up all existing tasks.
2960 if (!use_task_css_set_links)
2961 cgroup_enable_task_cg_lists();
2963 read_lock(&css_set_lock);
2964 it->cg_link = &cgrp->css_sets;
2965 cgroup_advance_iter(cgrp, it);
2968 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
2969 struct cgroup_iter *it)
2971 struct task_struct *res;
2972 struct list_head *l = it->task;
2973 struct cg_cgroup_link *link;
2975 /* If the iterator cg is NULL, we have no tasks */
2976 if (!it->cg_link)
2977 return NULL;
2978 res = list_entry(l, struct task_struct, cg_list);
2979 /* Advance iterator to find next entry */
2980 l = l->next;
2981 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
2982 if (l == &link->cg->tasks) {
2983 /* We reached the end of this task list - move on to
2984 * the next cg_cgroup_link */
2985 cgroup_advance_iter(cgrp, it);
2986 } else {
2987 it->task = l;
2989 return res;
2992 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
2993 __releases(css_set_lock)
2995 read_unlock(&css_set_lock);
2998 static inline int started_after_time(struct task_struct *t1,
2999 struct timespec *time,
3000 struct task_struct *t2)
3002 int start_diff = timespec_compare(&t1->start_time, time);
3003 if (start_diff > 0) {
3004 return 1;
3005 } else if (start_diff < 0) {
3006 return 0;
3007 } else {
3009 * Arbitrarily, if two processes started at the same
3010 * time, we'll say that the lower pointer value
3011 * started first. Note that t2 may have exited by now
3012 * so this may not be a valid pointer any longer, but
3013 * that's fine - it still serves to distinguish
3014 * between two tasks started (effectively) simultaneously.
3016 return t1 > t2;
3021 * This function is a callback from heap_insert() and is used to order
3022 * the heap.
3023 * In this case we order the heap in descending task start time.
3025 static inline int started_after(void *p1, void *p2)
3027 struct task_struct *t1 = p1;
3028 struct task_struct *t2 = p2;
3029 return started_after_time(t1, &t2->start_time, t2);
3033 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3034 * @scan: struct cgroup_scanner containing arguments for the scan
3036 * Arguments include pointers to callback functions test_task() and
3037 * process_task().
3038 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3039 * and if it returns true, call process_task() for it also.
3040 * The test_task pointer may be NULL, meaning always true (select all tasks).
3041 * Effectively duplicates cgroup_iter_{start,next,end}()
3042 * but does not lock css_set_lock for the call to process_task().
3043 * The struct cgroup_scanner may be embedded in any structure of the caller's
3044 * creation.
3045 * It is guaranteed that process_task() will act on every task that
3046 * is a member of the cgroup for the duration of this call. This
3047 * function may or may not call process_task() for tasks that exit
3048 * or move to a different cgroup during the call, or are forked or
3049 * move into the cgroup during the call.
3051 * Note that test_task() may be called with locks held, and may in some
3052 * situations be called multiple times for the same task, so it should
3053 * be cheap.
3054 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3055 * pre-allocated and will be used for heap operations (and its "gt" member will
3056 * be overwritten), else a temporary heap will be used (allocation of which
3057 * may cause this function to fail).
3059 int cgroup_scan_tasks(struct cgroup_scanner *scan)
3061 int retval, i;
3062 struct cgroup_iter it;
3063 struct task_struct *p, *dropped;
3064 /* Never dereference latest_task, since it's not refcounted */
3065 struct task_struct *latest_task = NULL;
3066 struct ptr_heap tmp_heap;
3067 struct ptr_heap *heap;
3068 struct timespec latest_time = { 0, 0 };
3070 if (scan->heap) {
3071 /* The caller supplied our heap and pre-allocated its memory */
3072 heap = scan->heap;
3073 heap->gt = &started_after;
3074 } else {
3075 /* We need to allocate our own heap memory */
3076 heap = &tmp_heap;
3077 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
3078 if (retval)
3079 /* cannot allocate the heap */
3080 return retval;
3083 again:
3085 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3086 * to determine which are of interest, and using the scanner's
3087 * "process_task" callback to process any of them that need an update.
3088 * Since we don't want to hold any locks during the task updates,
3089 * gather tasks to be processed in a heap structure.
3090 * The heap is sorted by descending task start time.
3091 * If the statically-sized heap fills up, we overflow tasks that
3092 * started later, and in future iterations only consider tasks that
3093 * started after the latest task in the previous pass. This
3094 * guarantees forward progress and that we don't miss any tasks.
3096 heap->size = 0;
3097 cgroup_iter_start(scan->cg, &it);
3098 while ((p = cgroup_iter_next(scan->cg, &it))) {
3100 * Only affect tasks that qualify per the caller's callback,
3101 * if he provided one
3103 if (scan->test_task && !scan->test_task(p, scan))
3104 continue;
3106 * Only process tasks that started after the last task
3107 * we processed
3109 if (!started_after_time(p, &latest_time, latest_task))
3110 continue;
3111 dropped = heap_insert(heap, p);
3112 if (dropped == NULL) {
3114 * The new task was inserted; the heap wasn't
3115 * previously full
3117 get_task_struct(p);
3118 } else if (dropped != p) {
3120 * The new task was inserted, and pushed out a
3121 * different task
3123 get_task_struct(p);
3124 put_task_struct(dropped);
3127 * Else the new task was newer than anything already in
3128 * the heap and wasn't inserted
3131 cgroup_iter_end(scan->cg, &it);
3133 if (heap->size) {
3134 for (i = 0; i < heap->size; i++) {
3135 struct task_struct *q = heap->ptrs[i];
3136 if (i == 0) {
3137 latest_time = q->start_time;
3138 latest_task = q;
3140 /* Process the task per the caller's callback */
3141 scan->process_task(q, scan);
3142 put_task_struct(q);
3145 * If we had to process any tasks at all, scan again
3146 * in case some of them were in the middle of forking
3147 * children that didn't get processed.
3148 * Not the most efficient way to do it, but it avoids
3149 * having to take callback_mutex in the fork path
3151 goto again;
3153 if (heap == &tmp_heap)
3154 heap_free(&tmp_heap);
3155 return 0;
3159 * Stuff for reading the 'tasks'/'procs' files.
3161 * Reading this file can return large amounts of data if a cgroup has
3162 * *lots* of attached tasks. So it may need several calls to read(),
3163 * but we cannot guarantee that the information we produce is correct
3164 * unless we produce it entirely atomically.
3168 /* which pidlist file are we talking about? */
3169 enum cgroup_filetype {
3170 CGROUP_FILE_PROCS,
3171 CGROUP_FILE_TASKS,
3175 * A pidlist is a list of pids that virtually represents the contents of one
3176 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3177 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3178 * to the cgroup.
3180 struct cgroup_pidlist {
3182 * used to find which pidlist is wanted. doesn't change as long as
3183 * this particular list stays in the list.
3185 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
3186 /* array of xids */
3187 pid_t *list;
3188 /* how many elements the above list has */
3189 int length;
3190 /* how many files are using the current array */
3191 int use_count;
3192 /* each of these stored in a list by its cgroup */
3193 struct list_head links;
3194 /* pointer to the cgroup we belong to, for list removal purposes */
3195 struct cgroup *owner;
3196 /* protects the other fields */
3197 struct rw_semaphore mutex;
3201 * The following two functions "fix" the issue where there are more pids
3202 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3203 * TODO: replace with a kernel-wide solution to this problem
3205 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3206 static void *pidlist_allocate(int count)
3208 if (PIDLIST_TOO_LARGE(count))
3209 return vmalloc(count * sizeof(pid_t));
3210 else
3211 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3213 static void pidlist_free(void *p)
3215 if (is_vmalloc_addr(p))
3216 vfree(p);
3217 else
3218 kfree(p);
3220 static void *pidlist_resize(void *p, int newcount)
3222 void *newlist;
3223 /* note: if new alloc fails, old p will still be valid either way */
3224 if (is_vmalloc_addr(p)) {
3225 newlist = vmalloc(newcount * sizeof(pid_t));
3226 if (!newlist)
3227 return NULL;
3228 memcpy(newlist, p, newcount * sizeof(pid_t));
3229 vfree(p);
3230 } else {
3231 newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL);
3233 return newlist;
3237 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3238 * If the new stripped list is sufficiently smaller and there's enough memory
3239 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3240 * number of unique elements.
3242 /* is the size difference enough that we should re-allocate the array? */
3243 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3244 static int pidlist_uniq(pid_t **p, int length)
3246 int src, dest = 1;
3247 pid_t *list = *p;
3248 pid_t *newlist;
3251 * we presume the 0th element is unique, so i starts at 1. trivial
3252 * edge cases first; no work needs to be done for either
3254 if (length == 0 || length == 1)
3255 return length;
3256 /* src and dest walk down the list; dest counts unique elements */
3257 for (src = 1; src < length; src++) {
3258 /* find next unique element */
3259 while (list[src] == list[src-1]) {
3260 src++;
3261 if (src == length)
3262 goto after;
3264 /* dest always points to where the next unique element goes */
3265 list[dest] = list[src];
3266 dest++;
3268 after:
3270 * if the length difference is large enough, we want to allocate a
3271 * smaller buffer to save memory. if this fails due to out of memory,
3272 * we'll just stay with what we've got.
3274 if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) {
3275 newlist = pidlist_resize(list, dest);
3276 if (newlist)
3277 *p = newlist;
3279 return dest;
3282 static int cmppid(const void *a, const void *b)
3284 return *(pid_t *)a - *(pid_t *)b;
3288 * find the appropriate pidlist for our purpose (given procs vs tasks)
3289 * returns with the lock on that pidlist already held, and takes care
3290 * of the use count, or returns NULL with no locks held if we're out of
3291 * memory.
3293 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3294 enum cgroup_filetype type)
3296 struct cgroup_pidlist *l;
3297 /* don't need task_nsproxy() if we're looking at ourself */
3298 struct pid_namespace *ns = current->nsproxy->pid_ns;
3301 * We can't drop the pidlist_mutex before taking the l->mutex in case
3302 * the last ref-holder is trying to remove l from the list at the same
3303 * time. Holding the pidlist_mutex precludes somebody taking whichever
3304 * list we find out from under us - compare release_pid_array().
3306 mutex_lock(&cgrp->pidlist_mutex);
3307 list_for_each_entry(l, &cgrp->pidlists, links) {
3308 if (l->key.type == type && l->key.ns == ns) {
3309 /* make sure l doesn't vanish out from under us */
3310 down_write(&l->mutex);
3311 mutex_unlock(&cgrp->pidlist_mutex);
3312 return l;
3315 /* entry not found; create a new one */
3316 l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3317 if (!l) {
3318 mutex_unlock(&cgrp->pidlist_mutex);
3319 return l;
3321 init_rwsem(&l->mutex);
3322 down_write(&l->mutex);
3323 l->key.type = type;
3324 l->key.ns = get_pid_ns(ns);
3325 l->use_count = 0; /* don't increment here */
3326 l->list = NULL;
3327 l->owner = cgrp;
3328 list_add(&l->links, &cgrp->pidlists);
3329 mutex_unlock(&cgrp->pidlist_mutex);
3330 return l;
3334 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3336 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3337 struct cgroup_pidlist **lp)
3339 pid_t *array;
3340 int length;
3341 int pid, n = 0; /* used for populating the array */
3342 struct cgroup_iter it;
3343 struct task_struct *tsk;
3344 struct cgroup_pidlist *l;
3347 * If cgroup gets more users after we read count, we won't have
3348 * enough space - tough. This race is indistinguishable to the
3349 * caller from the case that the additional cgroup users didn't
3350 * show up until sometime later on.
3352 length = cgroup_task_count(cgrp);
3353 array = pidlist_allocate(length);
3354 if (!array)
3355 return -ENOMEM;
3356 /* now, populate the array */
3357 cgroup_iter_start(cgrp, &it);
3358 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3359 if (unlikely(n == length))
3360 break;
3361 /* get tgid or pid for procs or tasks file respectively */
3362 if (type == CGROUP_FILE_PROCS)
3363 pid = task_tgid_vnr(tsk);
3364 else
3365 pid = task_pid_vnr(tsk);
3366 if (pid > 0) /* make sure to only use valid results */
3367 array[n++] = pid;
3369 cgroup_iter_end(cgrp, &it);
3370 length = n;
3371 /* now sort & (if procs) strip out duplicates */
3372 sort(array, length, sizeof(pid_t), cmppid, NULL);
3373 if (type == CGROUP_FILE_PROCS)
3374 length = pidlist_uniq(&array, length);
3375 l = cgroup_pidlist_find(cgrp, type);
3376 if (!l) {
3377 pidlist_free(array);
3378 return -ENOMEM;
3380 /* store array, freeing old if necessary - lock already held */
3381 pidlist_free(l->list);
3382 l->list = array;
3383 l->length = length;
3384 l->use_count++;
3385 up_write(&l->mutex);
3386 *lp = l;
3387 return 0;
3391 * cgroupstats_build - build and fill cgroupstats
3392 * @stats: cgroupstats to fill information into
3393 * @dentry: A dentry entry belonging to the cgroup for which stats have
3394 * been requested.
3396 * Build and fill cgroupstats so that taskstats can export it to user
3397 * space.
3399 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3401 int ret = -EINVAL;
3402 struct cgroup *cgrp;
3403 struct cgroup_iter it;
3404 struct task_struct *tsk;
3407 * Validate dentry by checking the superblock operations,
3408 * and make sure it's a directory.
3410 if (dentry->d_sb->s_op != &cgroup_ops ||
3411 !S_ISDIR(dentry->d_inode->i_mode))
3412 goto err;
3414 ret = 0;
3415 cgrp = dentry->d_fsdata;
3417 cgroup_iter_start(cgrp, &it);
3418 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3419 switch (tsk->state) {
3420 case TASK_RUNNING:
3421 stats->nr_running++;
3422 break;
3423 case TASK_INTERRUPTIBLE:
3424 stats->nr_sleeping++;
3425 break;
3426 case TASK_UNINTERRUPTIBLE:
3427 stats->nr_uninterruptible++;
3428 break;
3429 case TASK_STOPPED:
3430 stats->nr_stopped++;
3431 break;
3432 default:
3433 if (delayacct_is_task_waiting_on_io(tsk))
3434 stats->nr_io_wait++;
3435 break;
3438 cgroup_iter_end(cgrp, &it);
3440 err:
3441 return ret;
3446 * seq_file methods for the tasks/procs files. The seq_file position is the
3447 * next pid to display; the seq_file iterator is a pointer to the pid
3448 * in the cgroup->l->list array.
3451 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3454 * Initially we receive a position value that corresponds to
3455 * one more than the last pid shown (or 0 on the first call or
3456 * after a seek to the start). Use a binary-search to find the
3457 * next pid to display, if any
3459 struct cgroup_pidlist *l = s->private;
3460 int index = 0, pid = *pos;
3461 int *iter;
3463 down_read(&l->mutex);
3464 if (pid) {
3465 int end = l->length;
3467 while (index < end) {
3468 int mid = (index + end) / 2;
3469 if (l->list[mid] == pid) {
3470 index = mid;
3471 break;
3472 } else if (l->list[mid] <= pid)
3473 index = mid + 1;
3474 else
3475 end = mid;
3478 /* If we're off the end of the array, we're done */
3479 if (index >= l->length)
3480 return NULL;
3481 /* Update the abstract position to be the actual pid that we found */
3482 iter = l->list + index;
3483 *pos = *iter;
3484 return iter;
3487 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3489 struct cgroup_pidlist *l = s->private;
3490 up_read(&l->mutex);
3493 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3495 struct cgroup_pidlist *l = s->private;
3496 pid_t *p = v;
3497 pid_t *end = l->list + l->length;
3499 * Advance to the next pid in the array. If this goes off the
3500 * end, we're done
3502 p++;
3503 if (p >= end) {
3504 return NULL;
3505 } else {
3506 *pos = *p;
3507 return p;
3511 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3513 return seq_printf(s, "%d\n", *(int *)v);
3517 * seq_operations functions for iterating on pidlists through seq_file -
3518 * independent of whether it's tasks or procs
3520 static const struct seq_operations cgroup_pidlist_seq_operations = {
3521 .start = cgroup_pidlist_start,
3522 .stop = cgroup_pidlist_stop,
3523 .next = cgroup_pidlist_next,
3524 .show = cgroup_pidlist_show,
3527 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3530 * the case where we're the last user of this particular pidlist will
3531 * have us remove it from the cgroup's list, which entails taking the
3532 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3533 * pidlist_mutex, we have to take pidlist_mutex first.
3535 mutex_lock(&l->owner->pidlist_mutex);
3536 down_write(&l->mutex);
3537 BUG_ON(!l->use_count);
3538 if (!--l->use_count) {
3539 /* we're the last user if refcount is 0; remove and free */
3540 list_del(&l->links);
3541 mutex_unlock(&l->owner->pidlist_mutex);
3542 pidlist_free(l->list);
3543 put_pid_ns(l->key.ns);
3544 up_write(&l->mutex);
3545 kfree(l);
3546 return;
3548 mutex_unlock(&l->owner->pidlist_mutex);
3549 up_write(&l->mutex);
3552 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3554 struct cgroup_pidlist *l;
3555 if (!(file->f_mode & FMODE_READ))
3556 return 0;
3558 * the seq_file will only be initialized if the file was opened for
3559 * reading; hence we check if it's not null only in that case.
3561 l = ((struct seq_file *)file->private_data)->private;
3562 cgroup_release_pid_array(l);
3563 return seq_release(inode, file);
3566 static const struct file_operations cgroup_pidlist_operations = {
3567 .read = seq_read,
3568 .llseek = seq_lseek,
3569 .write = cgroup_file_write,
3570 .release = cgroup_pidlist_release,
3574 * The following functions handle opens on a file that displays a pidlist
3575 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3576 * in the cgroup.
3578 /* helper function for the two below it */
3579 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3581 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3582 struct cgroup_pidlist *l;
3583 int retval;
3585 /* Nothing to do for write-only files */
3586 if (!(file->f_mode & FMODE_READ))
3587 return 0;
3589 /* have the array populated */
3590 retval = pidlist_array_load(cgrp, type, &l);
3591 if (retval)
3592 return retval;
3593 /* configure file information */
3594 file->f_op = &cgroup_pidlist_operations;
3596 retval = seq_open(file, &cgroup_pidlist_seq_operations);
3597 if (retval) {
3598 cgroup_release_pid_array(l);
3599 return retval;
3601 ((struct seq_file *)file->private_data)->private = l;
3602 return 0;
3604 static int cgroup_tasks_open(struct inode *unused, struct file *file)
3606 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3608 static int cgroup_procs_open(struct inode *unused, struct file *file)
3610 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3613 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3614 struct cftype *cft)
3616 return notify_on_release(cgrp);
3619 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3620 struct cftype *cft,
3621 u64 val)
3623 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3624 if (val)
3625 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3626 else
3627 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3628 return 0;
3632 * Unregister event and free resources.
3634 * Gets called from workqueue.
3636 static void cgroup_event_remove(struct work_struct *work)
3638 struct cgroup_event *event = container_of(work, struct cgroup_event,
3639 remove);
3640 struct cgroup *cgrp = event->cgrp;
3642 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3644 eventfd_ctx_put(event->eventfd);
3645 kfree(event);
3646 dput(cgrp->dentry);
3650 * Gets called on POLLHUP on eventfd when user closes it.
3652 * Called with wqh->lock held and interrupts disabled.
3654 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3655 int sync, void *key)
3657 struct cgroup_event *event = container_of(wait,
3658 struct cgroup_event, wait);
3659 struct cgroup *cgrp = event->cgrp;
3660 unsigned long flags = (unsigned long)key;
3662 if (flags & POLLHUP) {
3663 __remove_wait_queue(event->wqh, &event->wait);
3664 spin_lock(&cgrp->event_list_lock);
3665 list_del(&event->list);
3666 spin_unlock(&cgrp->event_list_lock);
3668 * We are in atomic context, but cgroup_event_remove() may
3669 * sleep, so we have to call it in workqueue.
3671 schedule_work(&event->remove);
3674 return 0;
3677 static void cgroup_event_ptable_queue_proc(struct file *file,
3678 wait_queue_head_t *wqh, poll_table *pt)
3680 struct cgroup_event *event = container_of(pt,
3681 struct cgroup_event, pt);
3683 event->wqh = wqh;
3684 add_wait_queue(wqh, &event->wait);
3688 * Parse input and register new cgroup event handler.
3690 * Input must be in format '<event_fd> <control_fd> <args>'.
3691 * Interpretation of args is defined by control file implementation.
3693 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3694 const char *buffer)
3696 struct cgroup_event *event = NULL;
3697 unsigned int efd, cfd;
3698 struct file *efile = NULL;
3699 struct file *cfile = NULL;
3700 char *endp;
3701 int ret;
3703 efd = simple_strtoul(buffer, &endp, 10);
3704 if (*endp != ' ')
3705 return -EINVAL;
3706 buffer = endp + 1;
3708 cfd = simple_strtoul(buffer, &endp, 10);
3709 if ((*endp != ' ') && (*endp != '\0'))
3710 return -EINVAL;
3711 buffer = endp + 1;
3713 event = kzalloc(sizeof(*event), GFP_KERNEL);
3714 if (!event)
3715 return -ENOMEM;
3716 event->cgrp = cgrp;
3717 INIT_LIST_HEAD(&event->list);
3718 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3719 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3720 INIT_WORK(&event->remove, cgroup_event_remove);
3722 efile = eventfd_fget(efd);
3723 if (IS_ERR(efile)) {
3724 ret = PTR_ERR(efile);
3725 goto fail;
3728 event->eventfd = eventfd_ctx_fileget(efile);
3729 if (IS_ERR(event->eventfd)) {
3730 ret = PTR_ERR(event->eventfd);
3731 goto fail;
3734 cfile = fget(cfd);
3735 if (!cfile) {
3736 ret = -EBADF;
3737 goto fail;
3740 /* the process need read permission on control file */
3741 /* AV: shouldn't we check that it's been opened for read instead? */
3742 ret = inode_permission(cfile->f_path.dentry->d_inode, MAY_READ);
3743 if (ret < 0)
3744 goto fail;
3746 event->cft = __file_cft(cfile);
3747 if (IS_ERR(event->cft)) {
3748 ret = PTR_ERR(event->cft);
3749 goto fail;
3752 if (!event->cft->register_event || !event->cft->unregister_event) {
3753 ret = -EINVAL;
3754 goto fail;
3757 ret = event->cft->register_event(cgrp, event->cft,
3758 event->eventfd, buffer);
3759 if (ret)
3760 goto fail;
3762 if (efile->f_op->poll(efile, &event->pt) & POLLHUP) {
3763 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3764 ret = 0;
3765 goto fail;
3769 * Events should be removed after rmdir of cgroup directory, but before
3770 * destroying subsystem state objects. Let's take reference to cgroup
3771 * directory dentry to do that.
3773 dget(cgrp->dentry);
3775 spin_lock(&cgrp->event_list_lock);
3776 list_add(&event->list, &cgrp->event_list);
3777 spin_unlock(&cgrp->event_list_lock);
3779 fput(cfile);
3780 fput(efile);
3782 return 0;
3784 fail:
3785 if (cfile)
3786 fput(cfile);
3788 if (event && event->eventfd && !IS_ERR(event->eventfd))
3789 eventfd_ctx_put(event->eventfd);
3791 if (!IS_ERR_OR_NULL(efile))
3792 fput(efile);
3794 kfree(event);
3796 return ret;
3799 static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3800 struct cftype *cft)
3802 return clone_children(cgrp);
3805 static int cgroup_clone_children_write(struct cgroup *cgrp,
3806 struct cftype *cft,
3807 u64 val)
3809 if (val)
3810 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3811 else
3812 clear_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3813 return 0;
3817 * for the common functions, 'private' gives the type of file
3819 /* for hysterical raisins, we can't put this on the older files */
3820 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3821 static struct cftype files[] = {
3823 .name = "tasks",
3824 .open = cgroup_tasks_open,
3825 .write_u64 = cgroup_tasks_write,
3826 .release = cgroup_pidlist_release,
3827 .mode = S_IRUGO | S_IWUSR,
3830 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3831 .open = cgroup_procs_open,
3832 .write_u64 = cgroup_procs_write,
3833 .release = cgroup_pidlist_release,
3834 .mode = S_IRUGO | S_IWUSR,
3837 .name = "notify_on_release",
3838 .read_u64 = cgroup_read_notify_on_release,
3839 .write_u64 = cgroup_write_notify_on_release,
3842 .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3843 .write_string = cgroup_write_event_control,
3844 .mode = S_IWUGO,
3847 .name = "cgroup.clone_children",
3848 .read_u64 = cgroup_clone_children_read,
3849 .write_u64 = cgroup_clone_children_write,
3852 .name = "release_agent",
3853 .flags = CFTYPE_ONLY_ON_ROOT,
3854 .read_seq_string = cgroup_release_agent_show,
3855 .write_string = cgroup_release_agent_write,
3856 .max_write_len = PATH_MAX,
3858 { } /* terminate */
3861 static int cgroup_populate_dir(struct cgroup *cgrp)
3863 int err;
3864 struct cgroup_subsys *ss;
3866 err = cgroup_addrm_files(cgrp, NULL, files, true);
3867 if (err < 0)
3868 return err;
3870 /* process cftsets of each subsystem */
3871 for_each_subsys(cgrp->root, ss) {
3872 struct cftype_set *set;
3874 list_for_each_entry(set, &ss->cftsets, node)
3875 cgroup_addrm_files(cgrp, ss, set->cfts, true);
3878 /* This cgroup is ready now */
3879 for_each_subsys(cgrp->root, ss) {
3880 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3882 * Update id->css pointer and make this css visible from
3883 * CSS ID functions. This pointer will be dereferened
3884 * from RCU-read-side without locks.
3886 if (css->id)
3887 rcu_assign_pointer(css->id->css, css);
3890 return 0;
3893 static void css_dput_fn(struct work_struct *work)
3895 struct cgroup_subsys_state *css =
3896 container_of(work, struct cgroup_subsys_state, dput_work);
3898 dput(css->cgroup->dentry);
3901 static void init_cgroup_css(struct cgroup_subsys_state *css,
3902 struct cgroup_subsys *ss,
3903 struct cgroup *cgrp)
3905 css->cgroup = cgrp;
3906 atomic_set(&css->refcnt, 1);
3907 css->flags = 0;
3908 css->id = NULL;
3909 if (cgrp == dummytop)
3910 set_bit(CSS_ROOT, &css->flags);
3911 BUG_ON(cgrp->subsys[ss->subsys_id]);
3912 cgrp->subsys[ss->subsys_id] = css;
3915 * If !clear_css_refs, css holds an extra ref to @cgrp->dentry
3916 * which is put on the last css_put(). dput() requires process
3917 * context, which css_put() may be called without. @css->dput_work
3918 * will be used to invoke dput() asynchronously from css_put().
3920 INIT_WORK(&css->dput_work, css_dput_fn);
3921 if (ss->__DEPRECATED_clear_css_refs)
3922 set_bit(CSS_CLEAR_CSS_REFS, &css->flags);
3925 static void cgroup_lock_hierarchy(struct cgroupfs_root *root)
3927 /* We need to take each hierarchy_mutex in a consistent order */
3928 int i;
3931 * No worry about a race with rebind_subsystems that might mess up the
3932 * locking order, since both parties are under cgroup_mutex.
3934 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3935 struct cgroup_subsys *ss = subsys[i];
3936 if (ss == NULL)
3937 continue;
3938 if (ss->root == root)
3939 mutex_lock(&ss->hierarchy_mutex);
3943 static void cgroup_unlock_hierarchy(struct cgroupfs_root *root)
3945 int i;
3947 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3948 struct cgroup_subsys *ss = subsys[i];
3949 if (ss == NULL)
3950 continue;
3951 if (ss->root == root)
3952 mutex_unlock(&ss->hierarchy_mutex);
3957 * cgroup_create - create a cgroup
3958 * @parent: cgroup that will be parent of the new cgroup
3959 * @dentry: dentry of the new cgroup
3960 * @mode: mode to set on new inode
3962 * Must be called with the mutex on the parent inode held
3964 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
3965 umode_t mode)
3967 struct cgroup *cgrp;
3968 struct cgroupfs_root *root = parent->root;
3969 int err = 0;
3970 struct cgroup_subsys *ss;
3971 struct super_block *sb = root->sb;
3973 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
3974 if (!cgrp)
3975 return -ENOMEM;
3977 /* Grab a reference on the superblock so the hierarchy doesn't
3978 * get deleted on unmount if there are child cgroups. This
3979 * can be done outside cgroup_mutex, since the sb can't
3980 * disappear while someone has an open control file on the
3981 * fs */
3982 atomic_inc(&sb->s_active);
3984 mutex_lock(&cgroup_mutex);
3986 init_cgroup_housekeeping(cgrp);
3988 cgrp->parent = parent;
3989 cgrp->root = parent->root;
3990 cgrp->top_cgroup = parent->top_cgroup;
3992 if (notify_on_release(parent))
3993 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3995 if (clone_children(parent))
3996 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3998 for_each_subsys(root, ss) {
3999 struct cgroup_subsys_state *css = ss->create(cgrp);
4001 if (IS_ERR(css)) {
4002 err = PTR_ERR(css);
4003 goto err_destroy;
4005 init_cgroup_css(css, ss, cgrp);
4006 if (ss->use_id) {
4007 err = alloc_css_id(ss, parent, cgrp);
4008 if (err)
4009 goto err_destroy;
4011 /* At error, ->destroy() callback has to free assigned ID. */
4012 if (clone_children(parent) && ss->post_clone)
4013 ss->post_clone(cgrp);
4016 cgroup_lock_hierarchy(root);
4017 list_add(&cgrp->sibling, &cgrp->parent->children);
4018 cgroup_unlock_hierarchy(root);
4019 root->number_of_cgroups++;
4021 err = cgroup_create_dir(cgrp, dentry, mode);
4022 if (err < 0)
4023 goto err_remove;
4025 /* If !clear_css_refs, each css holds a ref to the cgroup's dentry */
4026 for_each_subsys(root, ss)
4027 if (!ss->__DEPRECATED_clear_css_refs)
4028 dget(dentry);
4030 /* The cgroup directory was pre-locked for us */
4031 BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
4033 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
4035 err = cgroup_populate_dir(cgrp);
4036 /* If err < 0, we have a half-filled directory - oh well ;) */
4038 mutex_unlock(&cgroup_mutex);
4039 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
4041 return 0;
4043 err_remove:
4045 cgroup_lock_hierarchy(root);
4046 list_del(&cgrp->sibling);
4047 cgroup_unlock_hierarchy(root);
4048 root->number_of_cgroups--;
4050 err_destroy:
4052 for_each_subsys(root, ss) {
4053 if (cgrp->subsys[ss->subsys_id])
4054 ss->destroy(cgrp);
4057 mutex_unlock(&cgroup_mutex);
4059 /* Release the reference count that we took on the superblock */
4060 deactivate_super(sb);
4062 kfree(cgrp);
4063 return err;
4066 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4068 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
4070 /* the vfs holds inode->i_mutex already */
4071 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
4075 * Check the reference count on each subsystem. Since we already
4076 * established that there are no tasks in the cgroup, if the css refcount
4077 * is also 1, then there should be no outstanding references, so the
4078 * subsystem is safe to destroy. We scan across all subsystems rather than
4079 * using the per-hierarchy linked list of mounted subsystems since we can
4080 * be called via check_for_release() with no synchronization other than
4081 * RCU, and the subsystem linked list isn't RCU-safe.
4083 static int cgroup_has_css_refs(struct cgroup *cgrp)
4085 int i;
4088 * We won't need to lock the subsys array, because the subsystems
4089 * we're concerned about aren't going anywhere since our cgroup root
4090 * has a reference on them.
4092 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4093 struct cgroup_subsys *ss = subsys[i];
4094 struct cgroup_subsys_state *css;
4096 /* Skip subsystems not present or not in this hierarchy */
4097 if (ss == NULL || ss->root != cgrp->root)
4098 continue;
4100 css = cgrp->subsys[ss->subsys_id];
4102 * When called from check_for_release() it's possible
4103 * that by this point the cgroup has been removed
4104 * and the css deleted. But a false-positive doesn't
4105 * matter, since it can only happen if the cgroup
4106 * has been deleted and hence no longer needs the
4107 * release agent to be called anyway.
4109 if (css && css_refcnt(css) > 1)
4110 return 1;
4112 return 0;
4116 * Atomically mark all (or else none) of the cgroup's CSS objects as
4117 * CSS_REMOVED. Return true on success, or false if the cgroup has
4118 * busy subsystems. Call with cgroup_mutex held
4120 * Depending on whether a subsys has __DEPRECATED_clear_css_refs set or
4121 * not, cgroup removal behaves differently.
4123 * If clear is set, css refcnt for the subsystem should be zero before
4124 * cgroup removal can be committed. This is implemented by
4125 * CGRP_WAIT_ON_RMDIR and retry logic around ->pre_destroy(), which may be
4126 * called multiple times until all css refcnts reach zero and is allowed to
4127 * veto removal on any invocation. This behavior is deprecated and will be
4128 * removed as soon as the existing user (memcg) is updated.
4130 * If clear is not set, each css holds an extra reference to the cgroup's
4131 * dentry and cgroup removal proceeds regardless of css refs.
4132 * ->pre_destroy() will be called at least once and is not allowed to fail.
4133 * On the last put of each css, whenever that may be, the extra dentry ref
4134 * is put so that dentry destruction happens only after all css's are
4135 * released.
4137 static int cgroup_clear_css_refs(struct cgroup *cgrp)
4139 struct cgroup_subsys *ss;
4140 unsigned long flags;
4141 bool failed = false;
4143 local_irq_save(flags);
4146 * Block new css_tryget() by deactivating refcnt. If all refcnts
4147 * for subsystems w/ clear_css_refs set were 1 at the moment of
4148 * deactivation, we succeeded.
4150 for_each_subsys(cgrp->root, ss) {
4151 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4153 WARN_ON(atomic_read(&css->refcnt) < 0);
4154 atomic_add(CSS_DEACT_BIAS, &css->refcnt);
4156 if (ss->__DEPRECATED_clear_css_refs)
4157 failed |= css_refcnt(css) != 1;
4161 * If succeeded, set REMOVED and put all the base refs; otherwise,
4162 * restore refcnts to positive values. Either way, all in-progress
4163 * css_tryget() will be released.
4165 for_each_subsys(cgrp->root, ss) {
4166 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4168 if (!failed) {
4169 set_bit(CSS_REMOVED, &css->flags);
4170 css_put(css);
4171 } else {
4172 atomic_sub(CSS_DEACT_BIAS, &css->refcnt);
4176 local_irq_restore(flags);
4177 return !failed;
4180 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
4182 struct cgroup *cgrp = dentry->d_fsdata;
4183 struct dentry *d;
4184 struct cgroup *parent;
4185 DEFINE_WAIT(wait);
4186 struct cgroup_event *event, *tmp;
4187 int ret;
4189 /* the vfs holds both inode->i_mutex already */
4190 again:
4191 mutex_lock(&cgroup_mutex);
4192 if (atomic_read(&cgrp->count) != 0) {
4193 mutex_unlock(&cgroup_mutex);
4194 return -EBUSY;
4196 if (!list_empty(&cgrp->children)) {
4197 mutex_unlock(&cgroup_mutex);
4198 return -EBUSY;
4200 mutex_unlock(&cgroup_mutex);
4203 * In general, subsystem has no css->refcnt after pre_destroy(). But
4204 * in racy cases, subsystem may have to get css->refcnt after
4205 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
4206 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
4207 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
4208 * and subsystem's reference count handling. Please see css_get/put
4209 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
4211 set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4214 * Call pre_destroy handlers of subsys. Notify subsystems
4215 * that rmdir() request comes.
4217 ret = cgroup_call_pre_destroy(cgrp);
4218 if (ret) {
4219 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4220 return ret;
4223 mutex_lock(&cgroup_mutex);
4224 parent = cgrp->parent;
4225 if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) {
4226 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4227 mutex_unlock(&cgroup_mutex);
4228 return -EBUSY;
4230 prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE);
4231 if (!cgroup_clear_css_refs(cgrp)) {
4232 mutex_unlock(&cgroup_mutex);
4234 * Because someone may call cgroup_wakeup_rmdir_waiter() before
4235 * prepare_to_wait(), we need to check this flag.
4237 if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags))
4238 schedule();
4239 finish_wait(&cgroup_rmdir_waitq, &wait);
4240 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4241 if (signal_pending(current))
4242 return -EINTR;
4243 goto again;
4245 /* NO css_tryget() can success after here. */
4246 finish_wait(&cgroup_rmdir_waitq, &wait);
4247 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4249 raw_spin_lock(&release_list_lock);
4250 set_bit(CGRP_REMOVED, &cgrp->flags);
4251 if (!list_empty(&cgrp->release_list))
4252 list_del_init(&cgrp->release_list);
4253 raw_spin_unlock(&release_list_lock);
4255 cgroup_lock_hierarchy(cgrp->root);
4256 /* delete this cgroup from parent->children */
4257 list_del_init(&cgrp->sibling);
4258 cgroup_unlock_hierarchy(cgrp->root);
4260 list_del_init(&cgrp->allcg_node);
4262 d = dget(cgrp->dentry);
4264 cgroup_d_remove_dir(d);
4265 dput(d);
4267 set_bit(CGRP_RELEASABLE, &parent->flags);
4268 check_for_release(parent);
4271 * Unregister events and notify userspace.
4272 * Notify userspace about cgroup removing only after rmdir of cgroup
4273 * directory to avoid race between userspace and kernelspace
4275 spin_lock(&cgrp->event_list_lock);
4276 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4277 list_del(&event->list);
4278 remove_wait_queue(event->wqh, &event->wait);
4279 eventfd_signal(event->eventfd, 1);
4280 schedule_work(&event->remove);
4282 spin_unlock(&cgrp->event_list_lock);
4284 mutex_unlock(&cgroup_mutex);
4285 return 0;
4288 static void __init_or_module cgroup_init_cftsets(struct cgroup_subsys *ss)
4290 INIT_LIST_HEAD(&ss->cftsets);
4293 * base_cftset is embedded in subsys itself, no need to worry about
4294 * deregistration.
4296 if (ss->base_cftypes) {
4297 ss->base_cftset.cfts = ss->base_cftypes;
4298 list_add_tail(&ss->base_cftset.node, &ss->cftsets);
4302 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4304 struct cgroup_subsys_state *css;
4306 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4308 /* init base cftset */
4309 cgroup_init_cftsets(ss);
4311 /* Create the top cgroup state for this subsystem */
4312 list_add(&ss->sibling, &rootnode.subsys_list);
4313 ss->root = &rootnode;
4314 css = ss->create(dummytop);
4315 /* We don't handle early failures gracefully */
4316 BUG_ON(IS_ERR(css));
4317 init_cgroup_css(css, ss, dummytop);
4319 /* Update the init_css_set to contain a subsys
4320 * pointer to this state - since the subsystem is
4321 * newly registered, all tasks and hence the
4322 * init_css_set is in the subsystem's top cgroup. */
4323 init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
4325 need_forkexit_callback |= ss->fork || ss->exit;
4327 /* At system boot, before all subsystems have been
4328 * registered, no tasks have been forked, so we don't
4329 * need to invoke fork callbacks here. */
4330 BUG_ON(!list_empty(&init_task.tasks));
4332 mutex_init(&ss->hierarchy_mutex);
4333 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
4334 ss->active = 1;
4336 /* this function shouldn't be used with modular subsystems, since they
4337 * need to register a subsys_id, among other things */
4338 BUG_ON(ss->module);
4342 * cgroup_load_subsys: load and register a modular subsystem at runtime
4343 * @ss: the subsystem to load
4345 * This function should be called in a modular subsystem's initcall. If the
4346 * subsystem is built as a module, it will be assigned a new subsys_id and set
4347 * up for use. If the subsystem is built-in anyway, work is delegated to the
4348 * simpler cgroup_init_subsys.
4350 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4352 int i;
4353 struct cgroup_subsys_state *css;
4355 /* check name and function validity */
4356 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4357 ss->create == NULL || ss->destroy == NULL)
4358 return -EINVAL;
4361 * we don't support callbacks in modular subsystems. this check is
4362 * before the ss->module check for consistency; a subsystem that could
4363 * be a module should still have no callbacks even if the user isn't
4364 * compiling it as one.
4366 if (ss->fork || ss->exit)
4367 return -EINVAL;
4370 * an optionally modular subsystem is built-in: we want to do nothing,
4371 * since cgroup_init_subsys will have already taken care of it.
4373 if (ss->module == NULL) {
4374 /* a few sanity checks */
4375 BUG_ON(ss->subsys_id >= CGROUP_BUILTIN_SUBSYS_COUNT);
4376 BUG_ON(subsys[ss->subsys_id] != ss);
4377 return 0;
4380 /* init base cftset */
4381 cgroup_init_cftsets(ss);
4384 * need to register a subsys id before anything else - for example,
4385 * init_cgroup_css needs it.
4387 mutex_lock(&cgroup_mutex);
4388 /* find the first empty slot in the array */
4389 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
4390 if (subsys[i] == NULL)
4391 break;
4393 if (i == CGROUP_SUBSYS_COUNT) {
4394 /* maximum number of subsystems already registered! */
4395 mutex_unlock(&cgroup_mutex);
4396 return -EBUSY;
4398 /* assign ourselves the subsys_id */
4399 ss->subsys_id = i;
4400 subsys[i] = ss;
4403 * no ss->create seems to need anything important in the ss struct, so
4404 * this can happen first (i.e. before the rootnode attachment).
4406 css = ss->create(dummytop);
4407 if (IS_ERR(css)) {
4408 /* failure case - need to deassign the subsys[] slot. */
4409 subsys[i] = NULL;
4410 mutex_unlock(&cgroup_mutex);
4411 return PTR_ERR(css);
4414 list_add(&ss->sibling, &rootnode.subsys_list);
4415 ss->root = &rootnode;
4417 /* our new subsystem will be attached to the dummy hierarchy. */
4418 init_cgroup_css(css, ss, dummytop);
4419 /* init_idr must be after init_cgroup_css because it sets css->id. */
4420 if (ss->use_id) {
4421 int ret = cgroup_init_idr(ss, css);
4422 if (ret) {
4423 dummytop->subsys[ss->subsys_id] = NULL;
4424 ss->destroy(dummytop);
4425 subsys[i] = NULL;
4426 mutex_unlock(&cgroup_mutex);
4427 return ret;
4432 * Now we need to entangle the css into the existing css_sets. unlike
4433 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4434 * will need a new pointer to it; done by iterating the css_set_table.
4435 * furthermore, modifying the existing css_sets will corrupt the hash
4436 * table state, so each changed css_set will need its hash recomputed.
4437 * this is all done under the css_set_lock.
4439 write_lock(&css_set_lock);
4440 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
4441 struct css_set *cg;
4442 struct hlist_node *node, *tmp;
4443 struct hlist_head *bucket = &css_set_table[i], *new_bucket;
4445 hlist_for_each_entry_safe(cg, node, tmp, bucket, hlist) {
4446 /* skip entries that we already rehashed */
4447 if (cg->subsys[ss->subsys_id])
4448 continue;
4449 /* remove existing entry */
4450 hlist_del(&cg->hlist);
4451 /* set new value */
4452 cg->subsys[ss->subsys_id] = css;
4453 /* recompute hash and restore entry */
4454 new_bucket = css_set_hash(cg->subsys);
4455 hlist_add_head(&cg->hlist, new_bucket);
4458 write_unlock(&css_set_lock);
4460 mutex_init(&ss->hierarchy_mutex);
4461 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
4462 ss->active = 1;
4464 /* success! */
4465 mutex_unlock(&cgroup_mutex);
4466 return 0;
4468 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4471 * cgroup_unload_subsys: unload a modular subsystem
4472 * @ss: the subsystem to unload
4474 * This function should be called in a modular subsystem's exitcall. When this
4475 * function is invoked, the refcount on the subsystem's module will be 0, so
4476 * the subsystem will not be attached to any hierarchy.
4478 void cgroup_unload_subsys(struct cgroup_subsys *ss)
4480 struct cg_cgroup_link *link;
4481 struct hlist_head *hhead;
4483 BUG_ON(ss->module == NULL);
4486 * we shouldn't be called if the subsystem is in use, and the use of
4487 * try_module_get in parse_cgroupfs_options should ensure that it
4488 * doesn't start being used while we're killing it off.
4490 BUG_ON(ss->root != &rootnode);
4492 mutex_lock(&cgroup_mutex);
4493 /* deassign the subsys_id */
4494 BUG_ON(ss->subsys_id < CGROUP_BUILTIN_SUBSYS_COUNT);
4495 subsys[ss->subsys_id] = NULL;
4497 /* remove subsystem from rootnode's list of subsystems */
4498 list_del_init(&ss->sibling);
4501 * disentangle the css from all css_sets attached to the dummytop. as
4502 * in loading, we need to pay our respects to the hashtable gods.
4504 write_lock(&css_set_lock);
4505 list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
4506 struct css_set *cg = link->cg;
4508 hlist_del(&cg->hlist);
4509 BUG_ON(!cg->subsys[ss->subsys_id]);
4510 cg->subsys[ss->subsys_id] = NULL;
4511 hhead = css_set_hash(cg->subsys);
4512 hlist_add_head(&cg->hlist, hhead);
4514 write_unlock(&css_set_lock);
4517 * remove subsystem's css from the dummytop and free it - need to free
4518 * before marking as null because ss->destroy needs the cgrp->subsys
4519 * pointer to find their state. note that this also takes care of
4520 * freeing the css_id.
4522 ss->destroy(dummytop);
4523 dummytop->subsys[ss->subsys_id] = NULL;
4525 mutex_unlock(&cgroup_mutex);
4527 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4530 * cgroup_init_early - cgroup initialization at system boot
4532 * Initialize cgroups at system boot, and initialize any
4533 * subsystems that request early init.
4535 int __init cgroup_init_early(void)
4537 int i;
4538 atomic_set(&init_css_set.refcount, 1);
4539 INIT_LIST_HEAD(&init_css_set.cg_links);
4540 INIT_LIST_HEAD(&init_css_set.tasks);
4541 INIT_HLIST_NODE(&init_css_set.hlist);
4542 css_set_count = 1;
4543 init_cgroup_root(&rootnode);
4544 root_count = 1;
4545 init_task.cgroups = &init_css_set;
4547 init_css_set_link.cg = &init_css_set;
4548 init_css_set_link.cgrp = dummytop;
4549 list_add(&init_css_set_link.cgrp_link_list,
4550 &rootnode.top_cgroup.css_sets);
4551 list_add(&init_css_set_link.cg_link_list,
4552 &init_css_set.cg_links);
4554 for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
4555 INIT_HLIST_HEAD(&css_set_table[i]);
4557 /* at bootup time, we don't worry about modular subsystems */
4558 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4559 struct cgroup_subsys *ss = subsys[i];
4561 BUG_ON(!ss->name);
4562 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4563 BUG_ON(!ss->create);
4564 BUG_ON(!ss->destroy);
4565 if (ss->subsys_id != i) {
4566 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4567 ss->name, ss->subsys_id);
4568 BUG();
4571 if (ss->early_init)
4572 cgroup_init_subsys(ss);
4574 return 0;
4578 * cgroup_init - cgroup initialization
4580 * Register cgroup filesystem and /proc file, and initialize
4581 * any subsystems that didn't request early init.
4583 int __init cgroup_init(void)
4585 int err;
4586 int i;
4587 struct hlist_head *hhead;
4589 err = bdi_init(&cgroup_backing_dev_info);
4590 if (err)
4591 return err;
4593 /* at bootup time, we don't worry about modular subsystems */
4594 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4595 struct cgroup_subsys *ss = subsys[i];
4596 if (!ss->early_init)
4597 cgroup_init_subsys(ss);
4598 if (ss->use_id)
4599 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4602 /* Add init_css_set to the hash table */
4603 hhead = css_set_hash(init_css_set.subsys);
4604 hlist_add_head(&init_css_set.hlist, hhead);
4605 BUG_ON(!init_root_id(&rootnode));
4607 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4608 if (!cgroup_kobj) {
4609 err = -ENOMEM;
4610 goto out;
4613 err = register_filesystem(&cgroup_fs_type);
4614 if (err < 0) {
4615 kobject_put(cgroup_kobj);
4616 goto out;
4619 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4621 out:
4622 if (err)
4623 bdi_destroy(&cgroup_backing_dev_info);
4625 return err;
4629 * proc_cgroup_show()
4630 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4631 * - Used for /proc/<pid>/cgroup.
4632 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4633 * doesn't really matter if tsk->cgroup changes after we read it,
4634 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4635 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4636 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4637 * cgroup to top_cgroup.
4640 /* TODO: Use a proper seq_file iterator */
4641 static int proc_cgroup_show(struct seq_file *m, void *v)
4643 struct pid *pid;
4644 struct task_struct *tsk;
4645 char *buf;
4646 int retval;
4647 struct cgroupfs_root *root;
4649 retval = -ENOMEM;
4650 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4651 if (!buf)
4652 goto out;
4654 retval = -ESRCH;
4655 pid = m->private;
4656 tsk = get_pid_task(pid, PIDTYPE_PID);
4657 if (!tsk)
4658 goto out_free;
4660 retval = 0;
4662 mutex_lock(&cgroup_mutex);
4664 for_each_active_root(root) {
4665 struct cgroup_subsys *ss;
4666 struct cgroup *cgrp;
4667 int count = 0;
4669 seq_printf(m, "%d:", root->hierarchy_id);
4670 for_each_subsys(root, ss)
4671 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4672 if (strlen(root->name))
4673 seq_printf(m, "%sname=%s", count ? "," : "",
4674 root->name);
4675 seq_putc(m, ':');
4676 cgrp = task_cgroup_from_root(tsk, root);
4677 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4678 if (retval < 0)
4679 goto out_unlock;
4680 seq_puts(m, buf);
4681 seq_putc(m, '\n');
4684 out_unlock:
4685 mutex_unlock(&cgroup_mutex);
4686 put_task_struct(tsk);
4687 out_free:
4688 kfree(buf);
4689 out:
4690 return retval;
4693 static int cgroup_open(struct inode *inode, struct file *file)
4695 struct pid *pid = PROC_I(inode)->pid;
4696 return single_open(file, proc_cgroup_show, pid);
4699 const struct file_operations proc_cgroup_operations = {
4700 .open = cgroup_open,
4701 .read = seq_read,
4702 .llseek = seq_lseek,
4703 .release = single_release,
4706 /* Display information about each subsystem and each hierarchy */
4707 static int proc_cgroupstats_show(struct seq_file *m, void *v)
4709 int i;
4711 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4713 * ideally we don't want subsystems moving around while we do this.
4714 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4715 * subsys/hierarchy state.
4717 mutex_lock(&cgroup_mutex);
4718 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4719 struct cgroup_subsys *ss = subsys[i];
4720 if (ss == NULL)
4721 continue;
4722 seq_printf(m, "%s\t%d\t%d\t%d\n",
4723 ss->name, ss->root->hierarchy_id,
4724 ss->root->number_of_cgroups, !ss->disabled);
4726 mutex_unlock(&cgroup_mutex);
4727 return 0;
4730 static int cgroupstats_open(struct inode *inode, struct file *file)
4732 return single_open(file, proc_cgroupstats_show, NULL);
4735 static const struct file_operations proc_cgroupstats_operations = {
4736 .open = cgroupstats_open,
4737 .read = seq_read,
4738 .llseek = seq_lseek,
4739 .release = single_release,
4743 * cgroup_fork - attach newly forked task to its parents cgroup.
4744 * @child: pointer to task_struct of forking parent process.
4746 * Description: A task inherits its parent's cgroup at fork().
4748 * A pointer to the shared css_set was automatically copied in
4749 * fork.c by dup_task_struct(). However, we ignore that copy, since
4750 * it was not made under the protection of RCU, cgroup_mutex or
4751 * threadgroup_change_begin(), so it might no longer be a valid
4752 * cgroup pointer. cgroup_attach_task() might have already changed
4753 * current->cgroups, allowing the previously referenced cgroup
4754 * group to be removed and freed.
4756 * Outside the pointer validity we also need to process the css_set
4757 * inheritance between threadgoup_change_begin() and
4758 * threadgoup_change_end(), this way there is no leak in any process
4759 * wide migration performed by cgroup_attach_proc() that could otherwise
4760 * miss a thread because it is too early or too late in the fork stage.
4762 * At the point that cgroup_fork() is called, 'current' is the parent
4763 * task, and the passed argument 'child' points to the child task.
4765 void cgroup_fork(struct task_struct *child)
4768 * We don't need to task_lock() current because current->cgroups
4769 * can't be changed concurrently here. The parent obviously hasn't
4770 * exited and called cgroup_exit(), and we are synchronized against
4771 * cgroup migration through threadgroup_change_begin().
4773 child->cgroups = current->cgroups;
4774 get_css_set(child->cgroups);
4775 INIT_LIST_HEAD(&child->cg_list);
4779 * cgroup_fork_callbacks - run fork callbacks
4780 * @child: the new task
4782 * Called on a new task very soon before adding it to the
4783 * tasklist. No need to take any locks since no-one can
4784 * be operating on this task.
4786 void cgroup_fork_callbacks(struct task_struct *child)
4788 if (need_forkexit_callback) {
4789 int i;
4791 * forkexit callbacks are only supported for builtin
4792 * subsystems, and the builtin section of the subsys array is
4793 * immutable, so we don't need to lock the subsys array here.
4795 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4796 struct cgroup_subsys *ss = subsys[i];
4797 if (ss->fork)
4798 ss->fork(child);
4804 * cgroup_post_fork - called on a new task after adding it to the task list
4805 * @child: the task in question
4807 * Adds the task to the list running through its css_set if necessary.
4808 * Has to be after the task is visible on the task list in case we race
4809 * with the first call to cgroup_iter_start() - to guarantee that the
4810 * new task ends up on its list.
4812 void cgroup_post_fork(struct task_struct *child)
4815 * use_task_css_set_links is set to 1 before we walk the tasklist
4816 * under the tasklist_lock and we read it here after we added the child
4817 * to the tasklist under the tasklist_lock as well. If the child wasn't
4818 * yet in the tasklist when we walked through it from
4819 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4820 * should be visible now due to the paired locking and barriers implied
4821 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4822 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4823 * lock on fork.
4825 if (use_task_css_set_links) {
4826 write_lock(&css_set_lock);
4827 if (list_empty(&child->cg_list)) {
4829 * It's safe to use child->cgroups without task_lock()
4830 * here because we are protected through
4831 * threadgroup_change_begin() against concurrent
4832 * css_set change in cgroup_task_migrate(). Also
4833 * the task can't exit at that point until
4834 * wake_up_new_task() is called, so we are protected
4835 * against cgroup_exit() setting child->cgroup to
4836 * init_css_set.
4838 list_add(&child->cg_list, &child->cgroups->tasks);
4840 write_unlock(&css_set_lock);
4844 * cgroup_exit - detach cgroup from exiting task
4845 * @tsk: pointer to task_struct of exiting process
4846 * @run_callback: run exit callbacks?
4848 * Description: Detach cgroup from @tsk and release it.
4850 * Note that cgroups marked notify_on_release force every task in
4851 * them to take the global cgroup_mutex mutex when exiting.
4852 * This could impact scaling on very large systems. Be reluctant to
4853 * use notify_on_release cgroups where very high task exit scaling
4854 * is required on large systems.
4856 * the_top_cgroup_hack:
4858 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4860 * We call cgroup_exit() while the task is still competent to
4861 * handle notify_on_release(), then leave the task attached to the
4862 * root cgroup in each hierarchy for the remainder of its exit.
4864 * To do this properly, we would increment the reference count on
4865 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4866 * code we would add a second cgroup function call, to drop that
4867 * reference. This would just create an unnecessary hot spot on
4868 * the top_cgroup reference count, to no avail.
4870 * Normally, holding a reference to a cgroup without bumping its
4871 * count is unsafe. The cgroup could go away, or someone could
4872 * attach us to a different cgroup, decrementing the count on
4873 * the first cgroup that we never incremented. But in this case,
4874 * top_cgroup isn't going away, and either task has PF_EXITING set,
4875 * which wards off any cgroup_attach_task() attempts, or task is a failed
4876 * fork, never visible to cgroup_attach_task.
4878 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4880 struct css_set *cg;
4881 int i;
4884 * Unlink from the css_set task list if necessary.
4885 * Optimistically check cg_list before taking
4886 * css_set_lock
4888 if (!list_empty(&tsk->cg_list)) {
4889 write_lock(&css_set_lock);
4890 if (!list_empty(&tsk->cg_list))
4891 list_del_init(&tsk->cg_list);
4892 write_unlock(&css_set_lock);
4895 /* Reassign the task to the init_css_set. */
4896 task_lock(tsk);
4897 cg = tsk->cgroups;
4898 tsk->cgroups = &init_css_set;
4900 if (run_callbacks && need_forkexit_callback) {
4902 * modular subsystems can't use callbacks, so no need to lock
4903 * the subsys array
4905 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4906 struct cgroup_subsys *ss = subsys[i];
4907 if (ss->exit) {
4908 struct cgroup *old_cgrp =
4909 rcu_dereference_raw(cg->subsys[i])->cgroup;
4910 struct cgroup *cgrp = task_cgroup(tsk, i);
4911 ss->exit(cgrp, old_cgrp, tsk);
4915 task_unlock(tsk);
4917 if (cg)
4918 put_css_set_taskexit(cg);
4922 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4923 * @cgrp: the cgroup in question
4924 * @task: the task in question
4926 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4927 * hierarchy.
4929 * If we are sending in dummytop, then presumably we are creating
4930 * the top cgroup in the subsystem.
4932 * Called only by the ns (nsproxy) cgroup.
4934 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
4936 int ret;
4937 struct cgroup *target;
4939 if (cgrp == dummytop)
4940 return 1;
4942 target = task_cgroup_from_root(task, cgrp->root);
4943 while (cgrp != target && cgrp!= cgrp->top_cgroup)
4944 cgrp = cgrp->parent;
4945 ret = (cgrp == target);
4946 return ret;
4949 static void check_for_release(struct cgroup *cgrp)
4951 /* All of these checks rely on RCU to keep the cgroup
4952 * structure alive */
4953 if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
4954 && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
4955 /* Control Group is currently removeable. If it's not
4956 * already queued for a userspace notification, queue
4957 * it now */
4958 int need_schedule_work = 0;
4959 raw_spin_lock(&release_list_lock);
4960 if (!cgroup_is_removed(cgrp) &&
4961 list_empty(&cgrp->release_list)) {
4962 list_add(&cgrp->release_list, &release_list);
4963 need_schedule_work = 1;
4965 raw_spin_unlock(&release_list_lock);
4966 if (need_schedule_work)
4967 schedule_work(&release_agent_work);
4971 /* Caller must verify that the css is not for root cgroup */
4972 bool __css_tryget(struct cgroup_subsys_state *css)
4974 do {
4975 int v = css_refcnt(css);
4977 if (atomic_cmpxchg(&css->refcnt, v, v + 1) == v)
4978 return true;
4979 cpu_relax();
4980 } while (!test_bit(CSS_REMOVED, &css->flags));
4982 return false;
4984 EXPORT_SYMBOL_GPL(__css_tryget);
4986 /* Caller must verify that the css is not for root cgroup */
4987 void __css_put(struct cgroup_subsys_state *css)
4989 struct cgroup *cgrp = css->cgroup;
4990 int v;
4992 rcu_read_lock();
4993 v = css_unbias_refcnt(atomic_dec_return(&css->refcnt));
4995 switch (v) {
4996 case 1:
4997 if (notify_on_release(cgrp)) {
4998 set_bit(CGRP_RELEASABLE, &cgrp->flags);
4999 check_for_release(cgrp);
5001 cgroup_wakeup_rmdir_waiter(cgrp);
5002 break;
5003 case 0:
5004 if (!test_bit(CSS_CLEAR_CSS_REFS, &css->flags))
5005 schedule_work(&css->dput_work);
5006 break;
5008 rcu_read_unlock();
5010 EXPORT_SYMBOL_GPL(__css_put);
5013 * Notify userspace when a cgroup is released, by running the
5014 * configured release agent with the name of the cgroup (path
5015 * relative to the root of cgroup file system) as the argument.
5017 * Most likely, this user command will try to rmdir this cgroup.
5019 * This races with the possibility that some other task will be
5020 * attached to this cgroup before it is removed, or that some other
5021 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5022 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5023 * unused, and this cgroup will be reprieved from its death sentence,
5024 * to continue to serve a useful existence. Next time it's released,
5025 * we will get notified again, if it still has 'notify_on_release' set.
5027 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5028 * means only wait until the task is successfully execve()'d. The
5029 * separate release agent task is forked by call_usermodehelper(),
5030 * then control in this thread returns here, without waiting for the
5031 * release agent task. We don't bother to wait because the caller of
5032 * this routine has no use for the exit status of the release agent
5033 * task, so no sense holding our caller up for that.
5035 static void cgroup_release_agent(struct work_struct *work)
5037 BUG_ON(work != &release_agent_work);
5038 mutex_lock(&cgroup_mutex);
5039 raw_spin_lock(&release_list_lock);
5040 while (!list_empty(&release_list)) {
5041 char *argv[3], *envp[3];
5042 int i;
5043 char *pathbuf = NULL, *agentbuf = NULL;
5044 struct cgroup *cgrp = list_entry(release_list.next,
5045 struct cgroup,
5046 release_list);
5047 list_del_init(&cgrp->release_list);
5048 raw_spin_unlock(&release_list_lock);
5049 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
5050 if (!pathbuf)
5051 goto continue_free;
5052 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
5053 goto continue_free;
5054 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
5055 if (!agentbuf)
5056 goto continue_free;
5058 i = 0;
5059 argv[i++] = agentbuf;
5060 argv[i++] = pathbuf;
5061 argv[i] = NULL;
5063 i = 0;
5064 /* minimal command environment */
5065 envp[i++] = "HOME=/";
5066 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5067 envp[i] = NULL;
5069 /* Drop the lock while we invoke the usermode helper,
5070 * since the exec could involve hitting disk and hence
5071 * be a slow process */
5072 mutex_unlock(&cgroup_mutex);
5073 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
5074 mutex_lock(&cgroup_mutex);
5075 continue_free:
5076 kfree(pathbuf);
5077 kfree(agentbuf);
5078 raw_spin_lock(&release_list_lock);
5080 raw_spin_unlock(&release_list_lock);
5081 mutex_unlock(&cgroup_mutex);
5084 static int __init cgroup_disable(char *str)
5086 int i;
5087 char *token;
5089 while ((token = strsep(&str, ",")) != NULL) {
5090 if (!*token)
5091 continue;
5093 * cgroup_disable, being at boot time, can't know about module
5094 * subsystems, so we don't worry about them.
5096 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
5097 struct cgroup_subsys *ss = subsys[i];
5099 if (!strcmp(token, ss->name)) {
5100 ss->disabled = 1;
5101 printk(KERN_INFO "Disabling %s control group"
5102 " subsystem\n", ss->name);
5103 break;
5107 return 1;
5109 __setup("cgroup_disable=", cgroup_disable);
5112 * Functons for CSS ID.
5116 *To get ID other than 0, this should be called when !cgroup_is_removed().
5118 unsigned short css_id(struct cgroup_subsys_state *css)
5120 struct css_id *cssid;
5123 * This css_id() can return correct value when somone has refcnt
5124 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5125 * it's unchanged until freed.
5127 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5129 if (cssid)
5130 return cssid->id;
5131 return 0;
5133 EXPORT_SYMBOL_GPL(css_id);
5135 unsigned short css_depth(struct cgroup_subsys_state *css)
5137 struct css_id *cssid;
5139 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5141 if (cssid)
5142 return cssid->depth;
5143 return 0;
5145 EXPORT_SYMBOL_GPL(css_depth);
5148 * css_is_ancestor - test "root" css is an ancestor of "child"
5149 * @child: the css to be tested.
5150 * @root: the css supporsed to be an ancestor of the child.
5152 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5153 * this function reads css->id, the caller must hold rcu_read_lock().
5154 * But, considering usual usage, the csses should be valid objects after test.
5155 * Assuming that the caller will do some action to the child if this returns
5156 * returns true, the caller must take "child";s reference count.
5157 * If "child" is valid object and this returns true, "root" is valid, too.
5160 bool css_is_ancestor(struct cgroup_subsys_state *child,
5161 const struct cgroup_subsys_state *root)
5163 struct css_id *child_id;
5164 struct css_id *root_id;
5166 child_id = rcu_dereference(child->id);
5167 if (!child_id)
5168 return false;
5169 root_id = rcu_dereference(root->id);
5170 if (!root_id)
5171 return false;
5172 if (child_id->depth < root_id->depth)
5173 return false;
5174 if (child_id->stack[root_id->depth] != root_id->id)
5175 return false;
5176 return true;
5179 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
5181 struct css_id *id = css->id;
5182 /* When this is called before css_id initialization, id can be NULL */
5183 if (!id)
5184 return;
5186 BUG_ON(!ss->use_id);
5188 rcu_assign_pointer(id->css, NULL);
5189 rcu_assign_pointer(css->id, NULL);
5190 spin_lock(&ss->id_lock);
5191 idr_remove(&ss->idr, id->id);
5192 spin_unlock(&ss->id_lock);
5193 kfree_rcu(id, rcu_head);
5195 EXPORT_SYMBOL_GPL(free_css_id);
5198 * This is called by init or create(). Then, calls to this function are
5199 * always serialized (By cgroup_mutex() at create()).
5202 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
5204 struct css_id *newid;
5205 int myid, error, size;
5207 BUG_ON(!ss->use_id);
5209 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
5210 newid = kzalloc(size, GFP_KERNEL);
5211 if (!newid)
5212 return ERR_PTR(-ENOMEM);
5213 /* get id */
5214 if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
5215 error = -ENOMEM;
5216 goto err_out;
5218 spin_lock(&ss->id_lock);
5219 /* Don't use 0. allocates an ID of 1-65535 */
5220 error = idr_get_new_above(&ss->idr, newid, 1, &myid);
5221 spin_unlock(&ss->id_lock);
5223 /* Returns error when there are no free spaces for new ID.*/
5224 if (error) {
5225 error = -ENOSPC;
5226 goto err_out;
5228 if (myid > CSS_ID_MAX)
5229 goto remove_idr;
5231 newid->id = myid;
5232 newid->depth = depth;
5233 return newid;
5234 remove_idr:
5235 error = -ENOSPC;
5236 spin_lock(&ss->id_lock);
5237 idr_remove(&ss->idr, myid);
5238 spin_unlock(&ss->id_lock);
5239 err_out:
5240 kfree(newid);
5241 return ERR_PTR(error);
5245 static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
5246 struct cgroup_subsys_state *rootcss)
5248 struct css_id *newid;
5250 spin_lock_init(&ss->id_lock);
5251 idr_init(&ss->idr);
5253 newid = get_new_cssid(ss, 0);
5254 if (IS_ERR(newid))
5255 return PTR_ERR(newid);
5257 newid->stack[0] = newid->id;
5258 newid->css = rootcss;
5259 rootcss->id = newid;
5260 return 0;
5263 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
5264 struct cgroup *child)
5266 int subsys_id, i, depth = 0;
5267 struct cgroup_subsys_state *parent_css, *child_css;
5268 struct css_id *child_id, *parent_id;
5270 subsys_id = ss->subsys_id;
5271 parent_css = parent->subsys[subsys_id];
5272 child_css = child->subsys[subsys_id];
5273 parent_id = parent_css->id;
5274 depth = parent_id->depth + 1;
5276 child_id = get_new_cssid(ss, depth);
5277 if (IS_ERR(child_id))
5278 return PTR_ERR(child_id);
5280 for (i = 0; i < depth; i++)
5281 child_id->stack[i] = parent_id->stack[i];
5282 child_id->stack[depth] = child_id->id;
5284 * child_id->css pointer will be set after this cgroup is available
5285 * see cgroup_populate_dir()
5287 rcu_assign_pointer(child_css->id, child_id);
5289 return 0;
5293 * css_lookup - lookup css by id
5294 * @ss: cgroup subsys to be looked into.
5295 * @id: the id
5297 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5298 * NULL if not. Should be called under rcu_read_lock()
5300 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5302 struct css_id *cssid = NULL;
5304 BUG_ON(!ss->use_id);
5305 cssid = idr_find(&ss->idr, id);
5307 if (unlikely(!cssid))
5308 return NULL;
5310 return rcu_dereference(cssid->css);
5312 EXPORT_SYMBOL_GPL(css_lookup);
5315 * css_get_next - lookup next cgroup under specified hierarchy.
5316 * @ss: pointer to subsystem
5317 * @id: current position of iteration.
5318 * @root: pointer to css. search tree under this.
5319 * @foundid: position of found object.
5321 * Search next css under the specified hierarchy of rootid. Calling under
5322 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5324 struct cgroup_subsys_state *
5325 css_get_next(struct cgroup_subsys *ss, int id,
5326 struct cgroup_subsys_state *root, int *foundid)
5328 struct cgroup_subsys_state *ret = NULL;
5329 struct css_id *tmp;
5330 int tmpid;
5331 int rootid = css_id(root);
5332 int depth = css_depth(root);
5334 if (!rootid)
5335 return NULL;
5337 BUG_ON(!ss->use_id);
5338 WARN_ON_ONCE(!rcu_read_lock_held());
5340 /* fill start point for scan */
5341 tmpid = id;
5342 while (1) {
5344 * scan next entry from bitmap(tree), tmpid is updated after
5345 * idr_get_next().
5347 tmp = idr_get_next(&ss->idr, &tmpid);
5348 if (!tmp)
5349 break;
5350 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
5351 ret = rcu_dereference(tmp->css);
5352 if (ret) {
5353 *foundid = tmpid;
5354 break;
5357 /* continue to scan from next id */
5358 tmpid = tmpid + 1;
5360 return ret;
5364 * get corresponding css from file open on cgroupfs directory
5366 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5368 struct cgroup *cgrp;
5369 struct inode *inode;
5370 struct cgroup_subsys_state *css;
5372 inode = f->f_dentry->d_inode;
5373 /* check in cgroup filesystem dir */
5374 if (inode->i_op != &cgroup_dir_inode_operations)
5375 return ERR_PTR(-EBADF);
5377 if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5378 return ERR_PTR(-EINVAL);
5380 /* get cgroup */
5381 cgrp = __d_cgrp(f->f_dentry);
5382 css = cgrp->subsys[id];
5383 return css ? css : ERR_PTR(-ENOENT);
5386 #ifdef CONFIG_CGROUP_DEBUG
5387 static struct cgroup_subsys_state *debug_create(struct cgroup *cont)
5389 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5391 if (!css)
5392 return ERR_PTR(-ENOMEM);
5394 return css;
5397 static void debug_destroy(struct cgroup *cont)
5399 kfree(cont->subsys[debug_subsys_id]);
5402 static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
5404 return atomic_read(&cont->count);
5407 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
5409 return cgroup_task_count(cont);
5412 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
5414 return (u64)(unsigned long)current->cgroups;
5417 static u64 current_css_set_refcount_read(struct cgroup *cont,
5418 struct cftype *cft)
5420 u64 count;
5422 rcu_read_lock();
5423 count = atomic_read(&current->cgroups->refcount);
5424 rcu_read_unlock();
5425 return count;
5428 static int current_css_set_cg_links_read(struct cgroup *cont,
5429 struct cftype *cft,
5430 struct seq_file *seq)
5432 struct cg_cgroup_link *link;
5433 struct css_set *cg;
5435 read_lock(&css_set_lock);
5436 rcu_read_lock();
5437 cg = rcu_dereference(current->cgroups);
5438 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
5439 struct cgroup *c = link->cgrp;
5440 const char *name;
5442 if (c->dentry)
5443 name = c->dentry->d_name.name;
5444 else
5445 name = "?";
5446 seq_printf(seq, "Root %d group %s\n",
5447 c->root->hierarchy_id, name);
5449 rcu_read_unlock();
5450 read_unlock(&css_set_lock);
5451 return 0;
5454 #define MAX_TASKS_SHOWN_PER_CSS 25
5455 static int cgroup_css_links_read(struct cgroup *cont,
5456 struct cftype *cft,
5457 struct seq_file *seq)
5459 struct cg_cgroup_link *link;
5461 read_lock(&css_set_lock);
5462 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
5463 struct css_set *cg = link->cg;
5464 struct task_struct *task;
5465 int count = 0;
5466 seq_printf(seq, "css_set %p\n", cg);
5467 list_for_each_entry(task, &cg->tasks, cg_list) {
5468 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5469 seq_puts(seq, " ...\n");
5470 break;
5471 } else {
5472 seq_printf(seq, " task %d\n",
5473 task_pid_vnr(task));
5477 read_unlock(&css_set_lock);
5478 return 0;
5481 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5483 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5486 static struct cftype debug_files[] = {
5488 .name = "cgroup_refcount",
5489 .read_u64 = cgroup_refcount_read,
5492 .name = "taskcount",
5493 .read_u64 = debug_taskcount_read,
5497 .name = "current_css_set",
5498 .read_u64 = current_css_set_read,
5502 .name = "current_css_set_refcount",
5503 .read_u64 = current_css_set_refcount_read,
5507 .name = "current_css_set_cg_links",
5508 .read_seq_string = current_css_set_cg_links_read,
5512 .name = "cgroup_css_links",
5513 .read_seq_string = cgroup_css_links_read,
5517 .name = "releasable",
5518 .read_u64 = releasable_read,
5521 { } /* terminate */
5524 struct cgroup_subsys debug_subsys = {
5525 .name = "debug",
5526 .create = debug_create,
5527 .destroy = debug_destroy,
5528 .subsys_id = debug_subsys_id,
5529 .base_cftypes = debug_files,
5531 #endif /* CONFIG_CGROUP_DEBUG */