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/init_task.h>
34 #include <linux/kernel.h>
35 #include <linux/list.h>
37 #include <linux/mutex.h>
38 #include <linux/mount.h>
39 #include <linux/pagemap.h>
40 #include <linux/proc_fs.h>
41 #include <linux/rcupdate.h>
42 #include <linux/sched.h>
43 #include <linux/backing-dev.h>
44 #include <linux/seq_file.h>
45 #include <linux/slab.h>
46 #include <linux/magic.h>
47 #include <linux/spinlock.h>
48 #include <linux/string.h>
49 #include <linux/sort.h>
50 #include <linux/kmod.h>
51 #include <linux/module.h>
52 #include <linux/delayacct.h>
53 #include <linux/cgroupstats.h>
54 #include <linux/hashtable.h>
55 #include <linux/namei.h>
56 #include <linux/pid_namespace.h>
57 #include <linux/idr.h>
58 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
59 #include <linux/eventfd.h>
60 #include <linux/poll.h>
61 #include <linux/flex_array.h> /* used in cgroup_attach_task */
62 #include <linux/kthread.h>
64 #include <linux/atomic.h>
66 /* css deactivation bias, makes css->refcnt negative to deny new trygets */
67 #define CSS_DEACT_BIAS INT_MIN
70 * cgroup_mutex is the master lock. Any modification to cgroup or its
71 * hierarchy must be performed while holding it.
73 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
74 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
75 * release_agent_path and so on. Modifying requires both cgroup_mutex and
76 * cgroup_root_mutex. Readers can acquire either of the two. This is to
77 * break the following locking order cycle.
79 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
80 * B. namespace_sem -> cgroup_mutex
82 * B happens only through cgroup_show_options() and using cgroup_root_mutex
85 #ifdef CONFIG_PROVE_RCU
86 DEFINE_MUTEX(cgroup_mutex);
87 EXPORT_SYMBOL_GPL(cgroup_mutex); /* only for task_subsys_state_check() */
89 static DEFINE_MUTEX(cgroup_mutex);
92 static DEFINE_MUTEX(cgroup_root_mutex);
95 * Generate an array of cgroup subsystem pointers. At boot time, this is
96 * populated with the built in subsystems, and modular subsystems are
97 * registered after that. The mutable section of this array is protected by
100 #define SUBSYS(_x) [_x ## _subsys_id] = &_x ## _subsys,
101 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
102 static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
103 #include <linux/cgroup_subsys.h>
107 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
108 * subsystems that are otherwise unattached - it never has more than a
109 * single cgroup, and all tasks are part of that cgroup.
111 static struct cgroupfs_root rootnode;
114 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
117 struct list_head node;
118 struct dentry *dentry;
122 struct simple_xattrs xattrs;
126 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
127 * cgroup_subsys->use_id != 0.
129 #define CSS_ID_MAX (65535)
132 * The css to which this ID points. This pointer is set to valid value
133 * after cgroup is populated. If cgroup is removed, this will be NULL.
134 * This pointer is expected to be RCU-safe because destroy()
135 * is called after synchronize_rcu(). But for safe use, css_tryget()
136 * should be used for avoiding race.
138 struct cgroup_subsys_state __rcu *css;
144 * Depth in hierarchy which this ID belongs to.
146 unsigned short depth;
148 * ID is freed by RCU. (and lookup routine is RCU safe.)
150 struct rcu_head rcu_head;
152 * Hierarchy of CSS ID belongs to.
154 unsigned short stack[0]; /* Array of Length (depth+1) */
158 * cgroup_event represents events which userspace want to receive.
160 struct cgroup_event {
162 * Cgroup which the event belongs to.
166 * Control file which the event associated.
170 * eventfd to signal userspace about the event.
172 struct eventfd_ctx *eventfd;
174 * Each of these stored in a list by the cgroup.
176 struct list_head list;
178 * All fields below needed to unregister event when
179 * userspace closes eventfd.
182 wait_queue_head_t *wqh;
184 struct work_struct remove;
187 /* The list of hierarchy roots */
189 static LIST_HEAD(roots);
190 static int root_count;
192 static DEFINE_IDA(hierarchy_ida);
193 static int next_hierarchy_id;
194 static DEFINE_SPINLOCK(hierarchy_id_lock);
196 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
197 #define dummytop (&rootnode.top_cgroup)
199 static struct cgroup_name root_cgroup_name = { .name = "/" };
201 /* This flag indicates whether tasks in the fork and exit paths should
202 * check for fork/exit handlers to call. This avoids us having to do
203 * extra work in the fork/exit path if none of the subsystems need to
206 static int need_forkexit_callback __read_mostly;
208 static int cgroup_destroy_locked(struct cgroup *cgrp);
209 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
210 struct cftype cfts[], bool is_add);
212 static int css_unbias_refcnt(int refcnt)
214 return refcnt >= 0 ? refcnt : refcnt - CSS_DEACT_BIAS;
217 /* the current nr of refs, always >= 0 whether @css is deactivated or not */
218 static int css_refcnt(struct cgroup_subsys_state *css)
220 int v = atomic_read(&css->refcnt);
222 return css_unbias_refcnt(v);
225 /* convenient tests for these bits */
226 inline int cgroup_is_removed(const struct cgroup *cgrp)
228 return test_bit(CGRP_REMOVED, &cgrp->flags);
232 * cgroup_is_descendant - test ancestry
233 * @cgrp: the cgroup to be tested
234 * @ancestor: possible ancestor of @cgrp
236 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
237 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
238 * and @ancestor are accessible.
240 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor)
243 if (cgrp == ancestor)
249 EXPORT_SYMBOL_GPL(cgroup_is_descendant);
251 static int cgroup_is_releasable(const struct cgroup *cgrp)
254 (1 << CGRP_RELEASABLE) |
255 (1 << CGRP_NOTIFY_ON_RELEASE);
256 return (cgrp->flags & bits) == bits;
259 static int notify_on_release(const struct cgroup *cgrp)
261 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
265 * for_each_subsys() allows you to iterate on each subsystem attached to
266 * an active hierarchy
268 #define for_each_subsys(_root, _ss) \
269 list_for_each_entry(_ss, &_root->subsys_list, sibling)
271 /* for_each_active_root() allows you to iterate across the active hierarchies */
272 #define for_each_active_root(_root) \
273 list_for_each_entry(_root, &roots, root_list)
275 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
277 return dentry->d_fsdata;
280 static inline struct cfent *__d_cfe(struct dentry *dentry)
282 return dentry->d_fsdata;
285 static inline struct cftype *__d_cft(struct dentry *dentry)
287 return __d_cfe(dentry)->type;
291 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
292 * @cgrp: the cgroup to be checked for liveness
294 * On success, returns true; the mutex should be later unlocked. On
295 * failure returns false with no lock held.
297 static bool cgroup_lock_live_group(struct cgroup *cgrp)
299 mutex_lock(&cgroup_mutex);
300 if (cgroup_is_removed(cgrp)) {
301 mutex_unlock(&cgroup_mutex);
307 /* the list of cgroups eligible for automatic release. Protected by
308 * release_list_lock */
309 static LIST_HEAD(release_list);
310 static DEFINE_RAW_SPINLOCK(release_list_lock);
311 static void cgroup_release_agent(struct work_struct *work);
312 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
313 static void check_for_release(struct cgroup *cgrp);
315 /* Link structure for associating css_set objects with cgroups */
316 struct cg_cgroup_link {
318 * List running through cg_cgroup_links associated with a
319 * cgroup, anchored on cgroup->css_sets
321 struct list_head cgrp_link_list;
324 * List running through cg_cgroup_links pointing at a
325 * single css_set object, anchored on css_set->cg_links
327 struct list_head cg_link_list;
331 /* The default css_set - used by init and its children prior to any
332 * hierarchies being mounted. It contains a pointer to the root state
333 * for each subsystem. Also used to anchor the list of css_sets. Not
334 * reference-counted, to improve performance when child cgroups
335 * haven't been created.
338 static struct css_set init_css_set;
339 static struct cg_cgroup_link init_css_set_link;
341 static int cgroup_init_idr(struct cgroup_subsys *ss,
342 struct cgroup_subsys_state *css);
344 /* css_set_lock protects the list of css_set objects, and the
345 * chain of tasks off each css_set. Nests outside task->alloc_lock
346 * due to cgroup_iter_start() */
347 static DEFINE_RWLOCK(css_set_lock);
348 static int css_set_count;
351 * hash table for cgroup groups. This improves the performance to find
352 * an existing css_set. This hash doesn't (currently) take into
353 * account cgroups in empty hierarchies.
355 #define CSS_SET_HASH_BITS 7
356 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
358 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
361 unsigned long key = 0UL;
363 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
364 key += (unsigned long)css[i];
365 key = (key >> 16) ^ key;
370 /* We don't maintain the lists running through each css_set to its
371 * task until after the first call to cgroup_iter_start(). This
372 * reduces the fork()/exit() overhead for people who have cgroups
373 * compiled into their kernel but not actually in use */
374 static int use_task_css_set_links __read_mostly;
376 static void __put_css_set(struct css_set *cg, int taskexit)
378 struct cg_cgroup_link *link;
379 struct cg_cgroup_link *saved_link;
381 * Ensure that the refcount doesn't hit zero while any readers
382 * can see it. Similar to atomic_dec_and_lock(), but for an
385 if (atomic_add_unless(&cg->refcount, -1, 1))
387 write_lock(&css_set_lock);
388 if (!atomic_dec_and_test(&cg->refcount)) {
389 write_unlock(&css_set_lock);
393 /* This css_set is dead. unlink it and release cgroup refcounts */
394 hash_del(&cg->hlist);
397 list_for_each_entry_safe(link, saved_link, &cg->cg_links,
399 struct cgroup *cgrp = link->cgrp;
400 list_del(&link->cg_link_list);
401 list_del(&link->cgrp_link_list);
404 * We may not be holding cgroup_mutex, and if cgrp->count is
405 * dropped to 0 the cgroup can be destroyed at any time, hence
406 * rcu_read_lock is used to keep it alive.
409 if (atomic_dec_and_test(&cgrp->count) &&
410 notify_on_release(cgrp)) {
412 set_bit(CGRP_RELEASABLE, &cgrp->flags);
413 check_for_release(cgrp);
420 write_unlock(&css_set_lock);
421 kfree_rcu(cg, rcu_head);
425 * refcounted get/put for css_set objects
427 static inline void get_css_set(struct css_set *cg)
429 atomic_inc(&cg->refcount);
432 static inline void put_css_set(struct css_set *cg)
434 __put_css_set(cg, 0);
437 static inline void put_css_set_taskexit(struct css_set *cg)
439 __put_css_set(cg, 1);
443 * compare_css_sets - helper function for find_existing_css_set().
444 * @cg: candidate css_set being tested
445 * @old_cg: existing css_set for a task
446 * @new_cgrp: cgroup that's being entered by the task
447 * @template: desired set of css pointers in css_set (pre-calculated)
449 * Returns true if "cg" matches "old_cg" except for the hierarchy
450 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
452 static bool compare_css_sets(struct css_set *cg,
453 struct css_set *old_cg,
454 struct cgroup *new_cgrp,
455 struct cgroup_subsys_state *template[])
457 struct list_head *l1, *l2;
459 if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
460 /* Not all subsystems matched */
465 * Compare cgroup pointers in order to distinguish between
466 * different cgroups in heirarchies with no subsystems. We
467 * could get by with just this check alone (and skip the
468 * memcmp above) but on most setups the memcmp check will
469 * avoid the need for this more expensive check on almost all
474 l2 = &old_cg->cg_links;
476 struct cg_cgroup_link *cgl1, *cgl2;
477 struct cgroup *cg1, *cg2;
481 /* See if we reached the end - both lists are equal length. */
482 if (l1 == &cg->cg_links) {
483 BUG_ON(l2 != &old_cg->cg_links);
486 BUG_ON(l2 == &old_cg->cg_links);
488 /* Locate the cgroups associated with these links. */
489 cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
490 cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
493 /* Hierarchies should be linked in the same order. */
494 BUG_ON(cg1->root != cg2->root);
497 * If this hierarchy is the hierarchy of the cgroup
498 * that's changing, then we need to check that this
499 * css_set points to the new cgroup; if it's any other
500 * hierarchy, then this css_set should point to the
501 * same cgroup as the old css_set.
503 if (cg1->root == new_cgrp->root) {
515 * find_existing_css_set() is a helper for
516 * find_css_set(), and checks to see whether an existing
517 * css_set is suitable.
519 * oldcg: the cgroup group that we're using before the cgroup
522 * cgrp: the cgroup that we're moving into
524 * template: location in which to build the desired set of subsystem
525 * state objects for the new cgroup group
527 static struct css_set *find_existing_css_set(
528 struct css_set *oldcg,
530 struct cgroup_subsys_state *template[])
533 struct cgroupfs_root *root = cgrp->root;
538 * Build the set of subsystem state objects that we want to see in the
539 * new css_set. while subsystems can change globally, the entries here
540 * won't change, so no need for locking.
542 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
543 if (root->subsys_mask & (1UL << i)) {
544 /* Subsystem is in this hierarchy. So we want
545 * the subsystem state from the new
547 template[i] = cgrp->subsys[i];
549 /* Subsystem is not in this hierarchy, so we
550 * don't want to change the subsystem state */
551 template[i] = oldcg->subsys[i];
555 key = css_set_hash(template);
556 hash_for_each_possible(css_set_table, cg, hlist, key) {
557 if (!compare_css_sets(cg, oldcg, cgrp, template))
560 /* This css_set matches what we need */
564 /* No existing cgroup group matched */
568 static void free_cg_links(struct list_head *tmp)
570 struct cg_cgroup_link *link;
571 struct cg_cgroup_link *saved_link;
573 list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
574 list_del(&link->cgrp_link_list);
580 * allocate_cg_links() allocates "count" cg_cgroup_link structures
581 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
582 * success or a negative error
584 static int allocate_cg_links(int count, struct list_head *tmp)
586 struct cg_cgroup_link *link;
589 for (i = 0; i < count; i++) {
590 link = kmalloc(sizeof(*link), GFP_KERNEL);
595 list_add(&link->cgrp_link_list, tmp);
601 * link_css_set - a helper function to link a css_set to a cgroup
602 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
603 * @cg: the css_set to be linked
604 * @cgrp: the destination cgroup
606 static void link_css_set(struct list_head *tmp_cg_links,
607 struct css_set *cg, struct cgroup *cgrp)
609 struct cg_cgroup_link *link;
611 BUG_ON(list_empty(tmp_cg_links));
612 link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
616 atomic_inc(&cgrp->count);
617 list_move(&link->cgrp_link_list, &cgrp->css_sets);
619 * Always add links to the tail of the list so that the list
620 * is sorted by order of hierarchy creation
622 list_add_tail(&link->cg_link_list, &cg->cg_links);
626 * find_css_set() takes an existing cgroup group and a
627 * cgroup object, and returns a css_set object that's
628 * equivalent to the old group, but with the given cgroup
629 * substituted into the appropriate hierarchy. Must be called with
632 static struct css_set *find_css_set(
633 struct css_set *oldcg, struct cgroup *cgrp)
636 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
638 struct list_head tmp_cg_links;
640 struct cg_cgroup_link *link;
643 /* First see if we already have a cgroup group that matches
645 read_lock(&css_set_lock);
646 res = find_existing_css_set(oldcg, cgrp, template);
649 read_unlock(&css_set_lock);
654 res = kmalloc(sizeof(*res), GFP_KERNEL);
658 /* Allocate all the cg_cgroup_link objects that we'll need */
659 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
664 atomic_set(&res->refcount, 1);
665 INIT_LIST_HEAD(&res->cg_links);
666 INIT_LIST_HEAD(&res->tasks);
667 INIT_HLIST_NODE(&res->hlist);
669 /* Copy the set of subsystem state objects generated in
670 * find_existing_css_set() */
671 memcpy(res->subsys, template, sizeof(res->subsys));
673 write_lock(&css_set_lock);
674 /* Add reference counts and links from the new css_set. */
675 list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
676 struct cgroup *c = link->cgrp;
677 if (c->root == cgrp->root)
679 link_css_set(&tmp_cg_links, res, c);
682 BUG_ON(!list_empty(&tmp_cg_links));
686 /* Add this cgroup group to the hash table */
687 key = css_set_hash(res->subsys);
688 hash_add(css_set_table, &res->hlist, key);
690 write_unlock(&css_set_lock);
696 * Return the cgroup for "task" from the given hierarchy. Must be
697 * called with cgroup_mutex held.
699 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
700 struct cgroupfs_root *root)
703 struct cgroup *res = NULL;
705 BUG_ON(!mutex_is_locked(&cgroup_mutex));
706 read_lock(&css_set_lock);
708 * No need to lock the task - since we hold cgroup_mutex the
709 * task can't change groups, so the only thing that can happen
710 * is that it exits and its css is set back to init_css_set.
713 if (css == &init_css_set) {
714 res = &root->top_cgroup;
716 struct cg_cgroup_link *link;
717 list_for_each_entry(link, &css->cg_links, cg_link_list) {
718 struct cgroup *c = link->cgrp;
719 if (c->root == root) {
725 read_unlock(&css_set_lock);
731 * There is one global cgroup mutex. We also require taking
732 * task_lock() when dereferencing a task's cgroup subsys pointers.
733 * See "The task_lock() exception", at the end of this comment.
735 * A task must hold cgroup_mutex to modify cgroups.
737 * Any task can increment and decrement the count field without lock.
738 * So in general, code holding cgroup_mutex can't rely on the count
739 * field not changing. However, if the count goes to zero, then only
740 * cgroup_attach_task() can increment it again. Because a count of zero
741 * means that no tasks are currently attached, therefore there is no
742 * way a task attached to that cgroup can fork (the other way to
743 * increment the count). So code holding cgroup_mutex can safely
744 * assume that if the count is zero, it will stay zero. Similarly, if
745 * a task holds cgroup_mutex on a cgroup with zero count, it
746 * knows that the cgroup won't be removed, as cgroup_rmdir()
749 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
750 * (usually) take cgroup_mutex. These are the two most performance
751 * critical pieces of code here. The exception occurs on cgroup_exit(),
752 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
753 * is taken, and if the cgroup count is zero, a usermode call made
754 * to the release agent with the name of the cgroup (path relative to
755 * the root of cgroup file system) as the argument.
757 * A cgroup can only be deleted if both its 'count' of using tasks
758 * is zero, and its list of 'children' cgroups is empty. Since all
759 * tasks in the system use _some_ cgroup, and since there is always at
760 * least one task in the system (init, pid == 1), therefore, top_cgroup
761 * always has either children cgroups and/or using tasks. So we don't
762 * need a special hack to ensure that top_cgroup cannot be deleted.
764 * The task_lock() exception
766 * The need for this exception arises from the action of
767 * cgroup_attach_task(), which overwrites one task's cgroup pointer with
768 * another. It does so using cgroup_mutex, however there are
769 * several performance critical places that need to reference
770 * task->cgroup without the expense of grabbing a system global
771 * mutex. Therefore except as noted below, when dereferencing or, as
772 * in cgroup_attach_task(), modifying a task's cgroup pointer we use
773 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
774 * the task_struct routinely used for such matters.
776 * P.S. One more locking exception. RCU is used to guard the
777 * update of a tasks cgroup pointer by cgroup_attach_task()
781 * A couple of forward declarations required, due to cyclic reference loop:
782 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
783 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
787 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
788 static struct dentry *cgroup_lookup(struct inode *, struct dentry *, unsigned int);
789 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
790 static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
791 unsigned long subsys_mask);
792 static const struct inode_operations cgroup_dir_inode_operations;
793 static const struct file_operations proc_cgroupstats_operations;
795 static struct backing_dev_info cgroup_backing_dev_info = {
797 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
800 static int alloc_css_id(struct cgroup_subsys *ss,
801 struct cgroup *parent, struct cgroup *child);
803 static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb)
805 struct inode *inode = new_inode(sb);
808 inode->i_ino = get_next_ino();
809 inode->i_mode = mode;
810 inode->i_uid = current_fsuid();
811 inode->i_gid = current_fsgid();
812 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
813 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
818 static struct cgroup_name *cgroup_alloc_name(struct dentry *dentry)
820 struct cgroup_name *name;
822 name = kmalloc(sizeof(*name) + dentry->d_name.len + 1, GFP_KERNEL);
825 strcpy(name->name, dentry->d_name.name);
829 static void cgroup_free_fn(struct work_struct *work)
831 struct cgroup *cgrp = container_of(work, struct cgroup, free_work);
832 struct cgroup_subsys *ss;
834 mutex_lock(&cgroup_mutex);
836 * Release the subsystem state objects.
838 for_each_subsys(cgrp->root, ss)
841 cgrp->root->number_of_cgroups--;
842 mutex_unlock(&cgroup_mutex);
845 * We get a ref to the parent's dentry, and put the ref when
846 * this cgroup is being freed, so it's guaranteed that the
847 * parent won't be destroyed before its children.
849 dput(cgrp->parent->dentry);
851 ida_simple_remove(&cgrp->root->cgroup_ida, cgrp->id);
854 * Drop the active superblock reference that we took when we
855 * created the cgroup. This will free cgrp->root, if we are
856 * holding the last reference to @sb.
858 deactivate_super(cgrp->root->sb);
861 * if we're getting rid of the cgroup, refcount should ensure
862 * that there are no pidlists left.
864 BUG_ON(!list_empty(&cgrp->pidlists));
866 simple_xattrs_free(&cgrp->xattrs);
868 kfree(rcu_dereference_raw(cgrp->name));
872 static void cgroup_free_rcu(struct rcu_head *head)
874 struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head);
876 schedule_work(&cgrp->free_work);
879 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
881 /* is dentry a directory ? if so, kfree() associated cgroup */
882 if (S_ISDIR(inode->i_mode)) {
883 struct cgroup *cgrp = dentry->d_fsdata;
885 BUG_ON(!(cgroup_is_removed(cgrp)));
886 call_rcu(&cgrp->rcu_head, cgroup_free_rcu);
888 struct cfent *cfe = __d_cfe(dentry);
889 struct cgroup *cgrp = dentry->d_parent->d_fsdata;
891 WARN_ONCE(!list_empty(&cfe->node) &&
892 cgrp != &cgrp->root->top_cgroup,
893 "cfe still linked for %s\n", cfe->type->name);
894 simple_xattrs_free(&cfe->xattrs);
900 static int cgroup_delete(const struct dentry *d)
905 static void remove_dir(struct dentry *d)
907 struct dentry *parent = dget(d->d_parent);
910 simple_rmdir(parent->d_inode, d);
914 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
918 lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex);
919 lockdep_assert_held(&cgroup_mutex);
922 * If we're doing cleanup due to failure of cgroup_create(),
923 * the corresponding @cfe may not exist.
925 list_for_each_entry(cfe, &cgrp->files, node) {
926 struct dentry *d = cfe->dentry;
928 if (cft && cfe->type != cft)
933 simple_unlink(cgrp->dentry->d_inode, d);
934 list_del_init(&cfe->node);
942 * cgroup_clear_directory - selective removal of base and subsystem files
943 * @dir: directory containing the files
944 * @base_files: true if the base files should be removed
945 * @subsys_mask: mask of the subsystem ids whose files should be removed
947 static void cgroup_clear_directory(struct dentry *dir, bool base_files,
948 unsigned long subsys_mask)
950 struct cgroup *cgrp = __d_cgrp(dir);
951 struct cgroup_subsys *ss;
953 for_each_subsys(cgrp->root, ss) {
954 struct cftype_set *set;
955 if (!test_bit(ss->subsys_id, &subsys_mask))
957 list_for_each_entry(set, &ss->cftsets, node)
958 cgroup_addrm_files(cgrp, NULL, set->cfts, false);
961 while (!list_empty(&cgrp->files))
962 cgroup_rm_file(cgrp, NULL);
967 * NOTE : the dentry must have been dget()'ed
969 static void cgroup_d_remove_dir(struct dentry *dentry)
971 struct dentry *parent;
972 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
974 cgroup_clear_directory(dentry, true, root->subsys_mask);
976 parent = dentry->d_parent;
977 spin_lock(&parent->d_lock);
978 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
979 list_del_init(&dentry->d_u.d_child);
980 spin_unlock(&dentry->d_lock);
981 spin_unlock(&parent->d_lock);
986 * Call with cgroup_mutex held. Drops reference counts on modules, including
987 * any duplicate ones that parse_cgroupfs_options took. If this function
988 * returns an error, no reference counts are touched.
990 static int rebind_subsystems(struct cgroupfs_root *root,
991 unsigned long final_subsys_mask)
993 unsigned long added_mask, removed_mask;
994 struct cgroup *cgrp = &root->top_cgroup;
997 BUG_ON(!mutex_is_locked(&cgroup_mutex));
998 BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
1000 removed_mask = root->actual_subsys_mask & ~final_subsys_mask;
1001 added_mask = final_subsys_mask & ~root->actual_subsys_mask;
1002 /* Check that any added subsystems are currently free */
1003 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1004 unsigned long bit = 1UL << i;
1005 struct cgroup_subsys *ss = subsys[i];
1006 if (!(bit & added_mask))
1009 * Nobody should tell us to do a subsys that doesn't exist:
1010 * parse_cgroupfs_options should catch that case and refcounts
1011 * ensure that subsystems won't disappear once selected.
1014 if (ss->root != &rootnode) {
1015 /* Subsystem isn't free */
1020 /* Currently we don't handle adding/removing subsystems when
1021 * any child cgroups exist. This is theoretically supportable
1022 * but involves complex error handling, so it's being left until
1024 if (root->number_of_cgroups > 1)
1027 /* Process each subsystem */
1028 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1029 struct cgroup_subsys *ss = subsys[i];
1030 unsigned long bit = 1UL << i;
1031 if (bit & added_mask) {
1032 /* We're binding this subsystem to this hierarchy */
1034 BUG_ON(cgrp->subsys[i]);
1035 BUG_ON(!dummytop->subsys[i]);
1036 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
1037 cgrp->subsys[i] = dummytop->subsys[i];
1038 cgrp->subsys[i]->cgroup = cgrp;
1039 list_move(&ss->sibling, &root->subsys_list);
1043 /* refcount was already taken, and we're keeping it */
1044 } else if (bit & removed_mask) {
1045 /* We're removing this subsystem */
1047 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1048 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1051 dummytop->subsys[i]->cgroup = dummytop;
1052 cgrp->subsys[i] = NULL;
1053 subsys[i]->root = &rootnode;
1054 list_move(&ss->sibling, &rootnode.subsys_list);
1055 /* subsystem is now free - drop reference on module */
1056 module_put(ss->module);
1057 } else if (bit & final_subsys_mask) {
1058 /* Subsystem state should already exist */
1060 BUG_ON(!cgrp->subsys[i]);
1062 * a refcount was taken, but we already had one, so
1063 * drop the extra reference.
1065 module_put(ss->module);
1066 #ifdef CONFIG_MODULE_UNLOAD
1067 BUG_ON(ss->module && !module_refcount(ss->module));
1070 /* Subsystem state shouldn't exist */
1071 BUG_ON(cgrp->subsys[i]);
1074 root->subsys_mask = root->actual_subsys_mask = final_subsys_mask;
1079 static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1081 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1082 struct cgroup_subsys *ss;
1084 mutex_lock(&cgroup_root_mutex);
1085 for_each_subsys(root, ss)
1086 seq_printf(seq, ",%s", ss->name);
1087 if (root->flags & CGRP_ROOT_SANE_BEHAVIOR)
1088 seq_puts(seq, ",sane_behavior");
1089 if (root->flags & CGRP_ROOT_NOPREFIX)
1090 seq_puts(seq, ",noprefix");
1091 if (root->flags & CGRP_ROOT_XATTR)
1092 seq_puts(seq, ",xattr");
1093 if (strlen(root->release_agent_path))
1094 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1095 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags))
1096 seq_puts(seq, ",clone_children");
1097 if (strlen(root->name))
1098 seq_printf(seq, ",name=%s", root->name);
1099 mutex_unlock(&cgroup_root_mutex);
1103 struct cgroup_sb_opts {
1104 unsigned long subsys_mask;
1105 unsigned long flags;
1106 char *release_agent;
1107 bool cpuset_clone_children;
1109 /* User explicitly requested empty subsystem */
1112 struct cgroupfs_root *new_root;
1117 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1118 * with cgroup_mutex held to protect the subsys[] array. This function takes
1119 * refcounts on subsystems to be used, unless it returns error, in which case
1120 * no refcounts are taken.
1122 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1124 char *token, *o = data;
1125 bool all_ss = false, one_ss = false;
1126 unsigned long mask = (unsigned long)-1;
1128 bool module_pin_failed = false;
1130 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1132 #ifdef CONFIG_CPUSETS
1133 mask = ~(1UL << cpuset_subsys_id);
1136 memset(opts, 0, sizeof(*opts));
1138 while ((token = strsep(&o, ",")) != NULL) {
1141 if (!strcmp(token, "none")) {
1142 /* Explicitly have no subsystems */
1146 if (!strcmp(token, "all")) {
1147 /* Mutually exclusive option 'all' + subsystem name */
1153 if (!strcmp(token, "__DEVEL__sane_behavior")) {
1154 opts->flags |= CGRP_ROOT_SANE_BEHAVIOR;
1157 if (!strcmp(token, "noprefix")) {
1158 opts->flags |= CGRP_ROOT_NOPREFIX;
1161 if (!strcmp(token, "clone_children")) {
1162 opts->cpuset_clone_children = true;
1165 if (!strcmp(token, "xattr")) {
1166 opts->flags |= CGRP_ROOT_XATTR;
1169 if (!strncmp(token, "release_agent=", 14)) {
1170 /* Specifying two release agents is forbidden */
1171 if (opts->release_agent)
1173 opts->release_agent =
1174 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1175 if (!opts->release_agent)
1179 if (!strncmp(token, "name=", 5)) {
1180 const char *name = token + 5;
1181 /* Can't specify an empty name */
1184 /* Must match [\w.-]+ */
1185 for (i = 0; i < strlen(name); i++) {
1189 if ((c == '.') || (c == '-') || (c == '_'))
1193 /* Specifying two names is forbidden */
1196 opts->name = kstrndup(name,
1197 MAX_CGROUP_ROOT_NAMELEN - 1,
1205 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1206 struct cgroup_subsys *ss = subsys[i];
1209 if (strcmp(token, ss->name))
1214 /* Mutually exclusive option 'all' + subsystem name */
1217 set_bit(i, &opts->subsys_mask);
1222 if (i == CGROUP_SUBSYS_COUNT)
1227 * If the 'all' option was specified select all the subsystems,
1228 * otherwise if 'none', 'name=' and a subsystem name options
1229 * were not specified, let's default to 'all'
1231 if (all_ss || (!one_ss && !opts->none && !opts->name)) {
1232 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1233 struct cgroup_subsys *ss = subsys[i];
1238 set_bit(i, &opts->subsys_mask);
1242 /* Consistency checks */
1244 if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) {
1245 pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1247 if (opts->flags & CGRP_ROOT_NOPREFIX) {
1248 pr_err("cgroup: sane_behavior: noprefix is not allowed\n");
1252 if (opts->cpuset_clone_children) {
1253 pr_err("cgroup: sane_behavior: clone_children is not allowed\n");
1259 * Option noprefix was introduced just for backward compatibility
1260 * with the old cpuset, so we allow noprefix only if mounting just
1261 * the cpuset subsystem.
1263 if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask))
1267 /* Can't specify "none" and some subsystems */
1268 if (opts->subsys_mask && opts->none)
1272 * We either have to specify by name or by subsystems. (So all
1273 * empty hierarchies must have a name).
1275 if (!opts->subsys_mask && !opts->name)
1279 * Grab references on all the modules we'll need, so the subsystems
1280 * don't dance around before rebind_subsystems attaches them. This may
1281 * take duplicate reference counts on a subsystem that's already used,
1282 * but rebind_subsystems handles this case.
1284 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1285 unsigned long bit = 1UL << i;
1287 if (!(bit & opts->subsys_mask))
1289 if (!try_module_get(subsys[i]->module)) {
1290 module_pin_failed = true;
1294 if (module_pin_failed) {
1296 * oops, one of the modules was going away. this means that we
1297 * raced with a module_delete call, and to the user this is
1298 * essentially a "subsystem doesn't exist" case.
1300 for (i--; i >= 0; i--) {
1301 /* drop refcounts only on the ones we took */
1302 unsigned long bit = 1UL << i;
1304 if (!(bit & opts->subsys_mask))
1306 module_put(subsys[i]->module);
1314 static void drop_parsed_module_refcounts(unsigned long subsys_mask)
1317 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1318 unsigned long bit = 1UL << i;
1320 if (!(bit & subsys_mask))
1322 module_put(subsys[i]->module);
1326 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1329 struct cgroupfs_root *root = sb->s_fs_info;
1330 struct cgroup *cgrp = &root->top_cgroup;
1331 struct cgroup_sb_opts opts;
1332 unsigned long added_mask, removed_mask;
1334 if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) {
1335 pr_err("cgroup: sane_behavior: remount is not allowed\n");
1339 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1340 mutex_lock(&cgroup_mutex);
1341 mutex_lock(&cgroup_root_mutex);
1343 /* See what subsystems are wanted */
1344 ret = parse_cgroupfs_options(data, &opts);
1348 if (opts.subsys_mask != root->actual_subsys_mask || opts.release_agent)
1349 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1350 task_tgid_nr(current), current->comm);
1352 added_mask = opts.subsys_mask & ~root->subsys_mask;
1353 removed_mask = root->subsys_mask & ~opts.subsys_mask;
1355 /* Don't allow flags or name to change at remount */
1356 if (opts.flags != root->flags ||
1357 (opts.name && strcmp(opts.name, root->name))) {
1359 drop_parsed_module_refcounts(opts.subsys_mask);
1364 * Clear out the files of subsystems that should be removed, do
1365 * this before rebind_subsystems, since rebind_subsystems may
1366 * change this hierarchy's subsys_list.
1368 cgroup_clear_directory(cgrp->dentry, false, removed_mask);
1370 ret = rebind_subsystems(root, opts.subsys_mask);
1372 /* rebind_subsystems failed, re-populate the removed files */
1373 cgroup_populate_dir(cgrp, false, removed_mask);
1374 drop_parsed_module_refcounts(opts.subsys_mask);
1378 /* re-populate subsystem files */
1379 cgroup_populate_dir(cgrp, false, added_mask);
1381 if (opts.release_agent)
1382 strcpy(root->release_agent_path, opts.release_agent);
1384 kfree(opts.release_agent);
1386 mutex_unlock(&cgroup_root_mutex);
1387 mutex_unlock(&cgroup_mutex);
1388 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1392 static const struct super_operations cgroup_ops = {
1393 .statfs = simple_statfs,
1394 .drop_inode = generic_delete_inode,
1395 .show_options = cgroup_show_options,
1396 .remount_fs = cgroup_remount,
1399 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1401 INIT_LIST_HEAD(&cgrp->sibling);
1402 INIT_LIST_HEAD(&cgrp->children);
1403 INIT_LIST_HEAD(&cgrp->files);
1404 INIT_LIST_HEAD(&cgrp->css_sets);
1405 INIT_LIST_HEAD(&cgrp->allcg_node);
1406 INIT_LIST_HEAD(&cgrp->release_list);
1407 INIT_LIST_HEAD(&cgrp->pidlists);
1408 INIT_WORK(&cgrp->free_work, cgroup_free_fn);
1409 mutex_init(&cgrp->pidlist_mutex);
1410 INIT_LIST_HEAD(&cgrp->event_list);
1411 spin_lock_init(&cgrp->event_list_lock);
1412 simple_xattrs_init(&cgrp->xattrs);
1415 static void init_cgroup_root(struct cgroupfs_root *root)
1417 struct cgroup *cgrp = &root->top_cgroup;
1419 INIT_LIST_HEAD(&root->subsys_list);
1420 INIT_LIST_HEAD(&root->root_list);
1421 INIT_LIST_HEAD(&root->allcg_list);
1422 root->number_of_cgroups = 1;
1424 cgrp->name = &root_cgroup_name;
1425 init_cgroup_housekeeping(cgrp);
1426 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
1429 static int cgroup_init_root_id(struct cgroupfs_root *root)
1434 if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1436 spin_lock(&hierarchy_id_lock);
1437 /* Try to allocate the next unused ID */
1438 ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1439 &root->hierarchy_id);
1441 /* Try again starting from 0 */
1442 ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1444 next_hierarchy_id = root->hierarchy_id + 1;
1445 } else if (ret != -EAGAIN) {
1446 /* Can only get here if the 31-bit IDR is full ... */
1449 spin_unlock(&hierarchy_id_lock);
1454 static void cgroup_exit_root_id(struct cgroupfs_root *root)
1456 if (root->hierarchy_id) {
1457 spin_lock(&hierarchy_id_lock);
1458 ida_remove(&hierarchy_ida, root->hierarchy_id);
1459 spin_unlock(&hierarchy_id_lock);
1461 root->hierarchy_id = 0;
1465 static int cgroup_test_super(struct super_block *sb, void *data)
1467 struct cgroup_sb_opts *opts = data;
1468 struct cgroupfs_root *root = sb->s_fs_info;
1470 /* If we asked for a name then it must match */
1471 if (opts->name && strcmp(opts->name, root->name))
1475 * If we asked for subsystems (or explicitly for no
1476 * subsystems) then they must match
1478 if ((opts->subsys_mask || opts->none)
1479 && (opts->subsys_mask != root->subsys_mask))
1485 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1487 struct cgroupfs_root *root;
1489 if (!opts->subsys_mask && !opts->none)
1492 root = kzalloc(sizeof(*root), GFP_KERNEL);
1494 return ERR_PTR(-ENOMEM);
1496 init_cgroup_root(root);
1498 root->subsys_mask = opts->subsys_mask;
1499 root->flags = opts->flags;
1500 ida_init(&root->cgroup_ida);
1501 if (opts->release_agent)
1502 strcpy(root->release_agent_path, opts->release_agent);
1504 strcpy(root->name, opts->name);
1505 if (opts->cpuset_clone_children)
1506 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags);
1510 static void cgroup_free_root(struct cgroupfs_root *root)
1513 /* hierarhcy ID shoulid already have been released */
1514 WARN_ON_ONCE(root->hierarchy_id);
1516 ida_destroy(&root->cgroup_ida);
1521 static int cgroup_set_super(struct super_block *sb, void *data)
1524 struct cgroup_sb_opts *opts = data;
1526 /* If we don't have a new root, we can't set up a new sb */
1527 if (!opts->new_root)
1530 BUG_ON(!opts->subsys_mask && !opts->none);
1532 ret = set_anon_super(sb, NULL);
1536 sb->s_fs_info = opts->new_root;
1537 opts->new_root->sb = sb;
1539 sb->s_blocksize = PAGE_CACHE_SIZE;
1540 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1541 sb->s_magic = CGROUP_SUPER_MAGIC;
1542 sb->s_op = &cgroup_ops;
1547 static int cgroup_get_rootdir(struct super_block *sb)
1549 static const struct dentry_operations cgroup_dops = {
1550 .d_iput = cgroup_diput,
1551 .d_delete = cgroup_delete,
1554 struct inode *inode =
1555 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1560 inode->i_fop = &simple_dir_operations;
1561 inode->i_op = &cgroup_dir_inode_operations;
1562 /* directories start off with i_nlink == 2 (for "." entry) */
1564 sb->s_root = d_make_root(inode);
1567 /* for everything else we want ->d_op set */
1568 sb->s_d_op = &cgroup_dops;
1572 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1573 int flags, const char *unused_dev_name,
1576 struct cgroup_sb_opts opts;
1577 struct cgroupfs_root *root;
1579 struct super_block *sb;
1580 struct cgroupfs_root *new_root;
1581 struct inode *inode;
1583 /* First find the desired set of subsystems */
1584 mutex_lock(&cgroup_mutex);
1585 ret = parse_cgroupfs_options(data, &opts);
1586 mutex_unlock(&cgroup_mutex);
1591 * Allocate a new cgroup root. We may not need it if we're
1592 * reusing an existing hierarchy.
1594 new_root = cgroup_root_from_opts(&opts);
1595 if (IS_ERR(new_root)) {
1596 ret = PTR_ERR(new_root);
1599 opts.new_root = new_root;
1601 /* Locate an existing or new sb for this hierarchy */
1602 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, 0, &opts);
1605 cgroup_free_root(opts.new_root);
1609 root = sb->s_fs_info;
1611 if (root == opts.new_root) {
1612 /* We used the new root structure, so this is a new hierarchy */
1613 struct list_head tmp_cg_links;
1614 struct cgroup *root_cgrp = &root->top_cgroup;
1615 struct cgroupfs_root *existing_root;
1616 const struct cred *cred;
1620 BUG_ON(sb->s_root != NULL);
1622 ret = cgroup_get_rootdir(sb);
1624 goto drop_new_super;
1625 inode = sb->s_root->d_inode;
1627 mutex_lock(&inode->i_mutex);
1628 mutex_lock(&cgroup_mutex);
1629 mutex_lock(&cgroup_root_mutex);
1631 /* Check for name clashes with existing mounts */
1633 if (strlen(root->name))
1634 for_each_active_root(existing_root)
1635 if (!strcmp(existing_root->name, root->name))
1639 * We're accessing css_set_count without locking
1640 * css_set_lock here, but that's OK - it can only be
1641 * increased by someone holding cgroup_lock, and
1642 * that's us. The worst that can happen is that we
1643 * have some link structures left over
1645 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1649 ret = cgroup_init_root_id(root);
1653 ret = rebind_subsystems(root, root->subsys_mask);
1654 if (ret == -EBUSY) {
1655 free_cg_links(&tmp_cg_links);
1659 * There must be no failure case after here, since rebinding
1660 * takes care of subsystems' refcounts, which are explicitly
1661 * dropped in the failure exit path.
1664 /* EBUSY should be the only error here */
1667 list_add(&root->root_list, &roots);
1670 sb->s_root->d_fsdata = root_cgrp;
1671 root->top_cgroup.dentry = sb->s_root;
1673 /* Link the top cgroup in this hierarchy into all
1674 * the css_set objects */
1675 write_lock(&css_set_lock);
1676 hash_for_each(css_set_table, i, cg, hlist)
1677 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->children));
1683 BUG_ON(root->number_of_cgroups != 1);
1685 cred = override_creds(&init_cred);
1686 cgroup_populate_dir(root_cgrp, true, root->subsys_mask);
1688 mutex_unlock(&cgroup_root_mutex);
1689 mutex_unlock(&cgroup_mutex);
1690 mutex_unlock(&inode->i_mutex);
1693 * We re-used an existing hierarchy - the new root (if
1694 * any) is not needed
1696 cgroup_free_root(opts.new_root);
1698 if (((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) &&
1699 root->flags != opts.flags) {
1700 pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n");
1702 goto drop_new_super;
1705 /* no subsys rebinding, so refcounts don't change */
1706 drop_parsed_module_refcounts(opts.subsys_mask);
1709 kfree(opts.release_agent);
1711 return dget(sb->s_root);
1714 cgroup_exit_root_id(root);
1715 mutex_unlock(&cgroup_root_mutex);
1716 mutex_unlock(&cgroup_mutex);
1717 mutex_unlock(&inode->i_mutex);
1719 deactivate_locked_super(sb);
1721 drop_parsed_module_refcounts(opts.subsys_mask);
1723 kfree(opts.release_agent);
1725 return ERR_PTR(ret);
1728 static void cgroup_kill_sb(struct super_block *sb) {
1729 struct cgroupfs_root *root = sb->s_fs_info;
1730 struct cgroup *cgrp = &root->top_cgroup;
1732 struct cg_cgroup_link *link;
1733 struct cg_cgroup_link *saved_link;
1737 BUG_ON(root->number_of_cgroups != 1);
1738 BUG_ON(!list_empty(&cgrp->children));
1740 mutex_lock(&cgroup_mutex);
1741 mutex_lock(&cgroup_root_mutex);
1743 /* Rebind all subsystems back to the default hierarchy */
1744 ret = rebind_subsystems(root, 0);
1745 /* Shouldn't be able to fail ... */
1749 * Release all the links from css_sets to this hierarchy's
1752 write_lock(&css_set_lock);
1754 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1756 list_del(&link->cg_link_list);
1757 list_del(&link->cgrp_link_list);
1760 write_unlock(&css_set_lock);
1762 if (!list_empty(&root->root_list)) {
1763 list_del(&root->root_list);
1767 cgroup_exit_root_id(root);
1769 mutex_unlock(&cgroup_root_mutex);
1770 mutex_unlock(&cgroup_mutex);
1772 simple_xattrs_free(&cgrp->xattrs);
1774 kill_litter_super(sb);
1775 cgroup_free_root(root);
1778 static struct file_system_type cgroup_fs_type = {
1780 .mount = cgroup_mount,
1781 .kill_sb = cgroup_kill_sb,
1784 static struct kobject *cgroup_kobj;
1787 * cgroup_path - generate the path of a cgroup
1788 * @cgrp: the cgroup in question
1789 * @buf: the buffer to write the path into
1790 * @buflen: the length of the buffer
1792 * Writes path of cgroup into buf. Returns 0 on success, -errno on error.
1794 * We can't generate cgroup path using dentry->d_name, as accessing
1795 * dentry->name must be protected by irq-unsafe dentry->d_lock or parent
1796 * inode's i_mutex, while on the other hand cgroup_path() can be called
1797 * with some irq-safe spinlocks held.
1799 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1801 int ret = -ENAMETOOLONG;
1804 if (!cgrp->parent) {
1805 if (strlcpy(buf, "/", buflen) >= buflen)
1806 return -ENAMETOOLONG;
1810 start = buf + buflen - 1;
1815 const char *name = cgroup_name(cgrp);
1819 if ((start -= len) < buf)
1821 memcpy(start, name, len);
1827 cgrp = cgrp->parent;
1828 } while (cgrp->parent);
1830 memmove(buf, start, buf + buflen - start);
1835 EXPORT_SYMBOL_GPL(cgroup_path);
1838 * Control Group taskset
1840 struct task_and_cgroup {
1841 struct task_struct *task;
1842 struct cgroup *cgrp;
1846 struct cgroup_taskset {
1847 struct task_and_cgroup single;
1848 struct flex_array *tc_array;
1851 struct cgroup *cur_cgrp;
1855 * cgroup_taskset_first - reset taskset and return the first task
1856 * @tset: taskset of interest
1858 * @tset iteration is initialized and the first task is returned.
1860 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1862 if (tset->tc_array) {
1864 return cgroup_taskset_next(tset);
1866 tset->cur_cgrp = tset->single.cgrp;
1867 return tset->single.task;
1870 EXPORT_SYMBOL_GPL(cgroup_taskset_first);
1873 * cgroup_taskset_next - iterate to the next task in taskset
1874 * @tset: taskset of interest
1876 * Return the next task in @tset. Iteration must have been initialized
1877 * with cgroup_taskset_first().
1879 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1881 struct task_and_cgroup *tc;
1883 if (!tset->tc_array || tset->idx >= tset->tc_array_len)
1886 tc = flex_array_get(tset->tc_array, tset->idx++);
1887 tset->cur_cgrp = tc->cgrp;
1890 EXPORT_SYMBOL_GPL(cgroup_taskset_next);
1893 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1894 * @tset: taskset of interest
1896 * Return the cgroup for the current (last returned) task of @tset. This
1897 * function must be preceded by either cgroup_taskset_first() or
1898 * cgroup_taskset_next().
1900 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset)
1902 return tset->cur_cgrp;
1904 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup);
1907 * cgroup_taskset_size - return the number of tasks in taskset
1908 * @tset: taskset of interest
1910 int cgroup_taskset_size(struct cgroup_taskset *tset)
1912 return tset->tc_array ? tset->tc_array_len : 1;
1914 EXPORT_SYMBOL_GPL(cgroup_taskset_size);
1918 * cgroup_task_migrate - move a task from one cgroup to another.
1920 * Must be called with cgroup_mutex and threadgroup locked.
1922 static void cgroup_task_migrate(struct cgroup *oldcgrp,
1923 struct task_struct *tsk, struct css_set *newcg)
1925 struct css_set *oldcg;
1928 * We are synchronized through threadgroup_lock() against PF_EXITING
1929 * setting such that we can't race against cgroup_exit() changing the
1930 * css_set to init_css_set and dropping the old one.
1932 WARN_ON_ONCE(tsk->flags & PF_EXITING);
1933 oldcg = tsk->cgroups;
1936 rcu_assign_pointer(tsk->cgroups, newcg);
1939 /* Update the css_set linked lists if we're using them */
1940 write_lock(&css_set_lock);
1941 if (!list_empty(&tsk->cg_list))
1942 list_move(&tsk->cg_list, &newcg->tasks);
1943 write_unlock(&css_set_lock);
1946 * We just gained a reference on oldcg by taking it from the task. As
1947 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1948 * it here; it will be freed under RCU.
1950 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1955 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
1956 * @cgrp: the cgroup to attach to
1957 * @tsk: the task or the leader of the threadgroup to be attached
1958 * @threadgroup: attach the whole threadgroup?
1960 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
1961 * task_lock of @tsk or each thread in the threadgroup individually in turn.
1963 static int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk,
1966 int retval, i, group_size;
1967 struct cgroup_subsys *ss, *failed_ss = NULL;
1968 struct cgroupfs_root *root = cgrp->root;
1969 /* threadgroup list cursor and array */
1970 struct task_struct *leader = tsk;
1971 struct task_and_cgroup *tc;
1972 struct flex_array *group;
1973 struct cgroup_taskset tset = { };
1976 * step 0: in order to do expensive, possibly blocking operations for
1977 * every thread, we cannot iterate the thread group list, since it needs
1978 * rcu or tasklist locked. instead, build an array of all threads in the
1979 * group - group_rwsem prevents new threads from appearing, and if
1980 * threads exit, this will just be an over-estimate.
1983 group_size = get_nr_threads(tsk);
1986 /* flex_array supports very large thread-groups better than kmalloc. */
1987 group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
1990 /* pre-allocate to guarantee space while iterating in rcu read-side. */
1991 retval = flex_array_prealloc(group, 0, group_size, GFP_KERNEL);
1993 goto out_free_group_list;
1997 * Prevent freeing of tasks while we take a snapshot. Tasks that are
1998 * already PF_EXITING could be freed from underneath us unless we
1999 * take an rcu_read_lock.
2003 struct task_and_cgroup ent;
2005 /* @tsk either already exited or can't exit until the end */
2006 if (tsk->flags & PF_EXITING)
2009 /* as per above, nr_threads may decrease, but not increase. */
2010 BUG_ON(i >= group_size);
2012 ent.cgrp = task_cgroup_from_root(tsk, root);
2013 /* nothing to do if this task is already in the cgroup */
2014 if (ent.cgrp == cgrp)
2017 * saying GFP_ATOMIC has no effect here because we did prealloc
2018 * earlier, but it's good form to communicate our expectations.
2020 retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
2021 BUG_ON(retval != 0);
2026 } while_each_thread(leader, tsk);
2028 /* remember the number of threads in the array for later. */
2030 tset.tc_array = group;
2031 tset.tc_array_len = group_size;
2033 /* methods shouldn't be called if no task is actually migrating */
2036 goto out_free_group_list;
2039 * step 1: check that we can legitimately attach to the cgroup.
2041 for_each_subsys(root, ss) {
2042 if (ss->can_attach) {
2043 retval = ss->can_attach(cgrp, &tset);
2046 goto out_cancel_attach;
2052 * step 2: make sure css_sets exist for all threads to be migrated.
2053 * we use find_css_set, which allocates a new one if necessary.
2055 for (i = 0; i < group_size; i++) {
2056 tc = flex_array_get(group, i);
2057 tc->cg = find_css_set(tc->task->cgroups, cgrp);
2060 goto out_put_css_set_refs;
2065 * step 3: now that we're guaranteed success wrt the css_sets,
2066 * proceed to move all tasks to the new cgroup. There are no
2067 * failure cases after here, so this is the commit point.
2069 for (i = 0; i < group_size; i++) {
2070 tc = flex_array_get(group, i);
2071 cgroup_task_migrate(tc->cgrp, tc->task, tc->cg);
2073 /* nothing is sensitive to fork() after this point. */
2076 * step 4: do subsystem attach callbacks.
2078 for_each_subsys(root, ss) {
2080 ss->attach(cgrp, &tset);
2084 * step 5: success! and cleanup
2087 out_put_css_set_refs:
2089 for (i = 0; i < group_size; i++) {
2090 tc = flex_array_get(group, i);
2093 put_css_set(tc->cg);
2098 for_each_subsys(root, ss) {
2099 if (ss == failed_ss)
2101 if (ss->cancel_attach)
2102 ss->cancel_attach(cgrp, &tset);
2105 out_free_group_list:
2106 flex_array_free(group);
2111 * Find the task_struct of the task to attach by vpid and pass it along to the
2112 * function to attach either it or all tasks in its threadgroup. Will lock
2113 * cgroup_mutex and threadgroup; may take task_lock of task.
2115 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2117 struct task_struct *tsk;
2118 const struct cred *cred = current_cred(), *tcred;
2121 if (!cgroup_lock_live_group(cgrp))
2127 tsk = find_task_by_vpid(pid);
2131 goto out_unlock_cgroup;
2134 * even if we're attaching all tasks in the thread group, we
2135 * only need to check permissions on one of them.
2137 tcred = __task_cred(tsk);
2138 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2139 !uid_eq(cred->euid, tcred->uid) &&
2140 !uid_eq(cred->euid, tcred->suid)) {
2143 goto out_unlock_cgroup;
2149 tsk = tsk->group_leader;
2152 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2153 * trapped in a cpuset, or RT worker may be born in a cgroup
2154 * with no rt_runtime allocated. Just say no.
2156 if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) {
2159 goto out_unlock_cgroup;
2162 get_task_struct(tsk);
2165 threadgroup_lock(tsk);
2167 if (!thread_group_leader(tsk)) {
2169 * a race with de_thread from another thread's exec()
2170 * may strip us of our leadership, if this happens,
2171 * there is no choice but to throw this task away and
2172 * try again; this is
2173 * "double-double-toil-and-trouble-check locking".
2175 threadgroup_unlock(tsk);
2176 put_task_struct(tsk);
2177 goto retry_find_task;
2181 ret = cgroup_attach_task(cgrp, tsk, threadgroup);
2183 threadgroup_unlock(tsk);
2185 put_task_struct(tsk);
2187 mutex_unlock(&cgroup_mutex);
2192 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2193 * @from: attach to all cgroups of a given task
2194 * @tsk: the task to be attached
2196 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2198 struct cgroupfs_root *root;
2201 mutex_lock(&cgroup_mutex);
2202 for_each_active_root(root) {
2203 struct cgroup *from_cg = task_cgroup_from_root(from, root);
2205 retval = cgroup_attach_task(from_cg, tsk, false);
2209 mutex_unlock(&cgroup_mutex);
2213 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2215 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2217 return attach_task_by_pid(cgrp, pid, false);
2220 static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2222 return attach_task_by_pid(cgrp, tgid, true);
2225 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2228 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2229 if (strlen(buffer) >= PATH_MAX)
2231 if (!cgroup_lock_live_group(cgrp))
2233 mutex_lock(&cgroup_root_mutex);
2234 strcpy(cgrp->root->release_agent_path, buffer);
2235 mutex_unlock(&cgroup_root_mutex);
2236 mutex_unlock(&cgroup_mutex);
2240 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2241 struct seq_file *seq)
2243 if (!cgroup_lock_live_group(cgrp))
2245 seq_puts(seq, cgrp->root->release_agent_path);
2246 seq_putc(seq, '\n');
2247 mutex_unlock(&cgroup_mutex);
2251 static int cgroup_sane_behavior_show(struct cgroup *cgrp, struct cftype *cft,
2252 struct seq_file *seq)
2254 seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp));
2258 /* A buffer size big enough for numbers or short strings */
2259 #define CGROUP_LOCAL_BUFFER_SIZE 64
2261 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2263 const char __user *userbuf,
2264 size_t nbytes, loff_t *unused_ppos)
2266 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2272 if (nbytes >= sizeof(buffer))
2274 if (copy_from_user(buffer, userbuf, nbytes))
2277 buffer[nbytes] = 0; /* nul-terminate */
2278 if (cft->write_u64) {
2279 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2282 retval = cft->write_u64(cgrp, cft, val);
2284 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2287 retval = cft->write_s64(cgrp, cft, val);
2294 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2296 const char __user *userbuf,
2297 size_t nbytes, loff_t *unused_ppos)
2299 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2301 size_t max_bytes = cft->max_write_len;
2302 char *buffer = local_buffer;
2305 max_bytes = sizeof(local_buffer) - 1;
2306 if (nbytes >= max_bytes)
2308 /* Allocate a dynamic buffer if we need one */
2309 if (nbytes >= sizeof(local_buffer)) {
2310 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2314 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2319 buffer[nbytes] = 0; /* nul-terminate */
2320 retval = cft->write_string(cgrp, cft, strstrip(buffer));
2324 if (buffer != local_buffer)
2329 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2330 size_t nbytes, loff_t *ppos)
2332 struct cftype *cft = __d_cft(file->f_dentry);
2333 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2335 if (cgroup_is_removed(cgrp))
2338 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2339 if (cft->write_u64 || cft->write_s64)
2340 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2341 if (cft->write_string)
2342 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2344 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2345 return ret ? ret : nbytes;
2350 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2352 char __user *buf, size_t nbytes,
2355 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2356 u64 val = cft->read_u64(cgrp, cft);
2357 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2359 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2362 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2364 char __user *buf, size_t nbytes,
2367 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2368 s64 val = cft->read_s64(cgrp, cft);
2369 int len = sprintf(tmp, "%lld\n", (long long) val);
2371 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2374 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2375 size_t nbytes, loff_t *ppos)
2377 struct cftype *cft = __d_cft(file->f_dentry);
2378 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2380 if (cgroup_is_removed(cgrp))
2384 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2386 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2388 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2393 * seqfile ops/methods for returning structured data. Currently just
2394 * supports string->u64 maps, but can be extended in future.
2397 struct cgroup_seqfile_state {
2399 struct cgroup *cgroup;
2402 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2404 struct seq_file *sf = cb->state;
2405 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2408 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2410 struct cgroup_seqfile_state *state = m->private;
2411 struct cftype *cft = state->cft;
2412 if (cft->read_map) {
2413 struct cgroup_map_cb cb = {
2414 .fill = cgroup_map_add,
2417 return cft->read_map(state->cgroup, cft, &cb);
2419 return cft->read_seq_string(state->cgroup, cft, m);
2422 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2424 struct seq_file *seq = file->private_data;
2425 kfree(seq->private);
2426 return single_release(inode, file);
2429 static const struct file_operations cgroup_seqfile_operations = {
2431 .write = cgroup_file_write,
2432 .llseek = seq_lseek,
2433 .release = cgroup_seqfile_release,
2436 static int cgroup_file_open(struct inode *inode, struct file *file)
2441 err = generic_file_open(inode, file);
2444 cft = __d_cft(file->f_dentry);
2446 if (cft->read_map || cft->read_seq_string) {
2447 struct cgroup_seqfile_state *state =
2448 kzalloc(sizeof(*state), GFP_USER);
2452 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2453 file->f_op = &cgroup_seqfile_operations;
2454 err = single_open(file, cgroup_seqfile_show, state);
2457 } else if (cft->open)
2458 err = cft->open(inode, file);
2465 static int cgroup_file_release(struct inode *inode, struct file *file)
2467 struct cftype *cft = __d_cft(file->f_dentry);
2469 return cft->release(inode, file);
2474 * cgroup_rename - Only allow simple rename of directories in place.
2476 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2477 struct inode *new_dir, struct dentry *new_dentry)
2480 struct cgroup_name *name, *old_name;
2481 struct cgroup *cgrp;
2484 * It's convinient to use parent dir's i_mutex to protected
2487 lockdep_assert_held(&old_dir->i_mutex);
2489 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2491 if (new_dentry->d_inode)
2493 if (old_dir != new_dir)
2496 cgrp = __d_cgrp(old_dentry);
2498 name = cgroup_alloc_name(new_dentry);
2502 ret = simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2508 old_name = cgrp->name;
2509 rcu_assign_pointer(cgrp->name, name);
2511 kfree_rcu(old_name, rcu_head);
2515 static struct simple_xattrs *__d_xattrs(struct dentry *dentry)
2517 if (S_ISDIR(dentry->d_inode->i_mode))
2518 return &__d_cgrp(dentry)->xattrs;
2520 return &__d_cfe(dentry)->xattrs;
2523 static inline int xattr_enabled(struct dentry *dentry)
2525 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
2526 return root->flags & CGRP_ROOT_XATTR;
2529 static bool is_valid_xattr(const char *name)
2531 if (!strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
2532 !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
2537 static int cgroup_setxattr(struct dentry *dentry, const char *name,
2538 const void *val, size_t size, int flags)
2540 if (!xattr_enabled(dentry))
2542 if (!is_valid_xattr(name))
2544 return simple_xattr_set(__d_xattrs(dentry), name, val, size, flags);
2547 static int cgroup_removexattr(struct dentry *dentry, const char *name)
2549 if (!xattr_enabled(dentry))
2551 if (!is_valid_xattr(name))
2553 return simple_xattr_remove(__d_xattrs(dentry), name);
2556 static ssize_t cgroup_getxattr(struct dentry *dentry, const char *name,
2557 void *buf, size_t size)
2559 if (!xattr_enabled(dentry))
2561 if (!is_valid_xattr(name))
2563 return simple_xattr_get(__d_xattrs(dentry), name, buf, size);
2566 static ssize_t cgroup_listxattr(struct dentry *dentry, char *buf, size_t size)
2568 if (!xattr_enabled(dentry))
2570 return simple_xattr_list(__d_xattrs(dentry), buf, size);
2573 static const struct file_operations cgroup_file_operations = {
2574 .read = cgroup_file_read,
2575 .write = cgroup_file_write,
2576 .llseek = generic_file_llseek,
2577 .open = cgroup_file_open,
2578 .release = cgroup_file_release,
2581 static const struct inode_operations cgroup_file_inode_operations = {
2582 .setxattr = cgroup_setxattr,
2583 .getxattr = cgroup_getxattr,
2584 .listxattr = cgroup_listxattr,
2585 .removexattr = cgroup_removexattr,
2588 static const struct inode_operations cgroup_dir_inode_operations = {
2589 .lookup = cgroup_lookup,
2590 .mkdir = cgroup_mkdir,
2591 .rmdir = cgroup_rmdir,
2592 .rename = cgroup_rename,
2593 .setxattr = cgroup_setxattr,
2594 .getxattr = cgroup_getxattr,
2595 .listxattr = cgroup_listxattr,
2596 .removexattr = cgroup_removexattr,
2599 static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2601 if (dentry->d_name.len > NAME_MAX)
2602 return ERR_PTR(-ENAMETOOLONG);
2603 d_add(dentry, NULL);
2608 * Check if a file is a control file
2610 static inline struct cftype *__file_cft(struct file *file)
2612 if (file_inode(file)->i_fop != &cgroup_file_operations)
2613 return ERR_PTR(-EINVAL);
2614 return __d_cft(file->f_dentry);
2617 static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2618 struct super_block *sb)
2620 struct inode *inode;
2624 if (dentry->d_inode)
2627 inode = cgroup_new_inode(mode, sb);
2631 if (S_ISDIR(mode)) {
2632 inode->i_op = &cgroup_dir_inode_operations;
2633 inode->i_fop = &simple_dir_operations;
2635 /* start off with i_nlink == 2 (for "." entry) */
2637 inc_nlink(dentry->d_parent->d_inode);
2640 * Control reaches here with cgroup_mutex held.
2641 * @inode->i_mutex should nest outside cgroup_mutex but we
2642 * want to populate it immediately without releasing
2643 * cgroup_mutex. As @inode isn't visible to anyone else
2644 * yet, trylock will always succeed without affecting
2647 WARN_ON_ONCE(!mutex_trylock(&inode->i_mutex));
2648 } else if (S_ISREG(mode)) {
2650 inode->i_fop = &cgroup_file_operations;
2651 inode->i_op = &cgroup_file_inode_operations;
2653 d_instantiate(dentry, inode);
2654 dget(dentry); /* Extra count - pin the dentry in core */
2659 * cgroup_file_mode - deduce file mode of a control file
2660 * @cft: the control file in question
2662 * returns cft->mode if ->mode is not 0
2663 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2664 * returns S_IRUGO if it has only a read handler
2665 * returns S_IWUSR if it has only a write hander
2667 static umode_t cgroup_file_mode(const struct cftype *cft)
2674 if (cft->read || cft->read_u64 || cft->read_s64 ||
2675 cft->read_map || cft->read_seq_string)
2678 if (cft->write || cft->write_u64 || cft->write_s64 ||
2679 cft->write_string || cft->trigger)
2685 static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2688 struct dentry *dir = cgrp->dentry;
2689 struct cgroup *parent = __d_cgrp(dir);
2690 struct dentry *dentry;
2694 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2696 if (subsys && !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) {
2697 strcpy(name, subsys->name);
2700 strcat(name, cft->name);
2702 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2704 cfe = kzalloc(sizeof(*cfe), GFP_KERNEL);
2708 dentry = lookup_one_len(name, dir, strlen(name));
2709 if (IS_ERR(dentry)) {
2710 error = PTR_ERR(dentry);
2714 mode = cgroup_file_mode(cft);
2715 error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb);
2717 cfe->type = (void *)cft;
2718 cfe->dentry = dentry;
2719 dentry->d_fsdata = cfe;
2720 simple_xattrs_init(&cfe->xattrs);
2721 list_add_tail(&cfe->node, &parent->files);
2730 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2731 struct cftype cfts[], bool is_add)
2736 for (cft = cfts; cft->name[0] != '\0'; cft++) {
2737 /* does cft->flags tell us to skip this file on @cgrp? */
2738 if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp))
2740 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2742 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
2746 err = cgroup_add_file(cgrp, subsys, cft);
2748 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2752 cgroup_rm_file(cgrp, cft);
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 struct cftype *cfts, bool is_add)
2776 __releases(&cgroup_mutex) __releases(&cgroup_cft_mutex)
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) {
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);
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
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, struct cftype *cfts)
2828 struct cftype_set *set;
2830 set = kzalloc(sizeof(*set), GFP_KERNEL);
2834 cgroup_cfts_prepare();
2836 list_add_tail(&set->node, &ss->cftsets);
2837 cgroup_cfts_commit(ss, cfts, true);
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, 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);
2870 cgroup_cfts_commit(ss, NULL, false);
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)
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);
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;
2904 /* Advance to the next non-empty css_set */
2907 if (l == &cgrp->css_sets) {
2911 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2913 } while (list_empty(&cg->tasks));
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) {
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);
2947 } while_each_thread(g, p);
2948 read_unlock(&tasklist_lock);
2949 write_unlock(&css_set_lock);
2953 * cgroup_next_descendant_pre - find the next descendant for pre-order walk
2954 * @pos: the current position (%NULL to initiate traversal)
2955 * @cgroup: cgroup whose descendants to walk
2957 * To be used by cgroup_for_each_descendant_pre(). Find the next
2958 * descendant to visit for pre-order traversal of @cgroup's descendants.
2960 struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
2961 struct cgroup *cgroup)
2963 struct cgroup *next;
2965 WARN_ON_ONCE(!rcu_read_lock_held());
2967 /* if first iteration, pretend we just visited @cgroup */
2969 if (list_empty(&cgroup->children))
2974 /* visit the first child if exists */
2975 next = list_first_or_null_rcu(&pos->children, struct cgroup, sibling);
2979 /* no child, visit my or the closest ancestor's next sibling */
2981 next = list_entry_rcu(pos->sibling.next, struct cgroup,
2983 if (&next->sibling != &pos->parent->children)
2987 } while (pos != cgroup);
2991 EXPORT_SYMBOL_GPL(cgroup_next_descendant_pre);
2994 * cgroup_rightmost_descendant - return the rightmost descendant of a cgroup
2995 * @pos: cgroup of interest
2997 * Return the rightmost descendant of @pos. If there's no descendant,
2998 * @pos is returned. This can be used during pre-order traversal to skip
3001 struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos)
3003 struct cgroup *last, *tmp;
3005 WARN_ON_ONCE(!rcu_read_lock_held());
3009 /* ->prev isn't RCU safe, walk ->next till the end */
3011 list_for_each_entry_rcu(tmp, &last->children, sibling)
3017 EXPORT_SYMBOL_GPL(cgroup_rightmost_descendant);
3019 static struct cgroup *cgroup_leftmost_descendant(struct cgroup *pos)
3021 struct cgroup *last;
3025 pos = list_first_or_null_rcu(&pos->children, struct cgroup,
3033 * cgroup_next_descendant_post - find the next descendant for post-order walk
3034 * @pos: the current position (%NULL to initiate traversal)
3035 * @cgroup: cgroup whose descendants to walk
3037 * To be used by cgroup_for_each_descendant_post(). Find the next
3038 * descendant to visit for post-order traversal of @cgroup's descendants.
3040 struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
3041 struct cgroup *cgroup)
3043 struct cgroup *next;
3045 WARN_ON_ONCE(!rcu_read_lock_held());
3047 /* if first iteration, visit the leftmost descendant */
3049 next = cgroup_leftmost_descendant(cgroup);
3050 return next != cgroup ? next : NULL;
3053 /* if there's an unvisited sibling, visit its leftmost descendant */
3054 next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling);
3055 if (&next->sibling != &pos->parent->children)
3056 return cgroup_leftmost_descendant(next);
3058 /* no sibling left, visit parent */
3060 return next != cgroup ? next : NULL;
3062 EXPORT_SYMBOL_GPL(cgroup_next_descendant_post);
3064 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
3065 __acquires(css_set_lock)
3068 * The first time anyone tries to iterate across a cgroup,
3069 * we need to enable the list linking each css_set to its
3070 * tasks, and fix up all existing tasks.
3072 if (!use_task_css_set_links)
3073 cgroup_enable_task_cg_lists();
3075 read_lock(&css_set_lock);
3076 it->cg_link = &cgrp->css_sets;
3077 cgroup_advance_iter(cgrp, it);
3080 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
3081 struct cgroup_iter *it)
3083 struct task_struct *res;
3084 struct list_head *l = it->task;
3085 struct cg_cgroup_link *link;
3087 /* If the iterator cg is NULL, we have no tasks */
3090 res = list_entry(l, struct task_struct, cg_list);
3091 /* Advance iterator to find next entry */
3093 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
3094 if (l == &link->cg->tasks) {
3095 /* We reached the end of this task list - move on to
3096 * the next cg_cgroup_link */
3097 cgroup_advance_iter(cgrp, it);
3104 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
3105 __releases(css_set_lock)
3107 read_unlock(&css_set_lock);
3110 static inline int started_after_time(struct task_struct *t1,
3111 struct timespec *time,
3112 struct task_struct *t2)
3114 int start_diff = timespec_compare(&t1->start_time, time);
3115 if (start_diff > 0) {
3117 } else if (start_diff < 0) {
3121 * Arbitrarily, if two processes started at the same
3122 * time, we'll say that the lower pointer value
3123 * started first. Note that t2 may have exited by now
3124 * so this may not be a valid pointer any longer, but
3125 * that's fine - it still serves to distinguish
3126 * between two tasks started (effectively) simultaneously.
3133 * This function is a callback from heap_insert() and is used to order
3135 * In this case we order the heap in descending task start time.
3137 static inline int started_after(void *p1, void *p2)
3139 struct task_struct *t1 = p1;
3140 struct task_struct *t2 = p2;
3141 return started_after_time(t1, &t2->start_time, t2);
3145 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3146 * @scan: struct cgroup_scanner containing arguments for the scan
3148 * Arguments include pointers to callback functions test_task() and
3150 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3151 * and if it returns true, call process_task() for it also.
3152 * The test_task pointer may be NULL, meaning always true (select all tasks).
3153 * Effectively duplicates cgroup_iter_{start,next,end}()
3154 * but does not lock css_set_lock for the call to process_task().
3155 * The struct cgroup_scanner may be embedded in any structure of the caller's
3157 * It is guaranteed that process_task() will act on every task that
3158 * is a member of the cgroup for the duration of this call. This
3159 * function may or may not call process_task() for tasks that exit
3160 * or move to a different cgroup during the call, or are forked or
3161 * move into the cgroup during the call.
3163 * Note that test_task() may be called with locks held, and may in some
3164 * situations be called multiple times for the same task, so it should
3166 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3167 * pre-allocated and will be used for heap operations (and its "gt" member will
3168 * be overwritten), else a temporary heap will be used (allocation of which
3169 * may cause this function to fail).
3171 int cgroup_scan_tasks(struct cgroup_scanner *scan)
3174 struct cgroup_iter it;
3175 struct task_struct *p, *dropped;
3176 /* Never dereference latest_task, since it's not refcounted */
3177 struct task_struct *latest_task = NULL;
3178 struct ptr_heap tmp_heap;
3179 struct ptr_heap *heap;
3180 struct timespec latest_time = { 0, 0 };
3183 /* The caller supplied our heap and pre-allocated its memory */
3185 heap->gt = &started_after;
3187 /* We need to allocate our own heap memory */
3189 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
3191 /* cannot allocate the heap */
3197 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3198 * to determine which are of interest, and using the scanner's
3199 * "process_task" callback to process any of them that need an update.
3200 * Since we don't want to hold any locks during the task updates,
3201 * gather tasks to be processed in a heap structure.
3202 * The heap is sorted by descending task start time.
3203 * If the statically-sized heap fills up, we overflow tasks that
3204 * started later, and in future iterations only consider tasks that
3205 * started after the latest task in the previous pass. This
3206 * guarantees forward progress and that we don't miss any tasks.
3209 cgroup_iter_start(scan->cg, &it);
3210 while ((p = cgroup_iter_next(scan->cg, &it))) {
3212 * Only affect tasks that qualify per the caller's callback,
3213 * if he provided one
3215 if (scan->test_task && !scan->test_task(p, scan))
3218 * Only process tasks that started after the last task
3221 if (!started_after_time(p, &latest_time, latest_task))
3223 dropped = heap_insert(heap, p);
3224 if (dropped == NULL) {
3226 * The new task was inserted; the heap wasn't
3230 } else if (dropped != p) {
3232 * The new task was inserted, and pushed out a
3236 put_task_struct(dropped);
3239 * Else the new task was newer than anything already in
3240 * the heap and wasn't inserted
3243 cgroup_iter_end(scan->cg, &it);
3246 for (i = 0; i < heap->size; i++) {
3247 struct task_struct *q = heap->ptrs[i];
3249 latest_time = q->start_time;
3252 /* Process the task per the caller's callback */
3253 scan->process_task(q, scan);
3257 * If we had to process any tasks at all, scan again
3258 * in case some of them were in the middle of forking
3259 * children that didn't get processed.
3260 * Not the most efficient way to do it, but it avoids
3261 * having to take callback_mutex in the fork path
3265 if (heap == &tmp_heap)
3266 heap_free(&tmp_heap);
3270 static void cgroup_transfer_one_task(struct task_struct *task,
3271 struct cgroup_scanner *scan)
3273 struct cgroup *new_cgroup = scan->data;
3275 mutex_lock(&cgroup_mutex);
3276 cgroup_attach_task(new_cgroup, task, false);
3277 mutex_unlock(&cgroup_mutex);
3281 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3282 * @to: cgroup to which the tasks will be moved
3283 * @from: cgroup in which the tasks currently reside
3285 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
3287 struct cgroup_scanner scan;
3290 scan.test_task = NULL; /* select all tasks in cgroup */
3291 scan.process_task = cgroup_transfer_one_task;
3295 return cgroup_scan_tasks(&scan);
3299 * Stuff for reading the 'tasks'/'procs' files.
3301 * Reading this file can return large amounts of data if a cgroup has
3302 * *lots* of attached tasks. So it may need several calls to read(),
3303 * but we cannot guarantee that the information we produce is correct
3304 * unless we produce it entirely atomically.
3308 /* which pidlist file are we talking about? */
3309 enum cgroup_filetype {
3315 * A pidlist is a list of pids that virtually represents the contents of one
3316 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3317 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3320 struct cgroup_pidlist {
3322 * used to find which pidlist is wanted. doesn't change as long as
3323 * this particular list stays in the list.
3325 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
3328 /* how many elements the above list has */
3330 /* how many files are using the current array */
3332 /* each of these stored in a list by its cgroup */
3333 struct list_head links;
3334 /* pointer to the cgroup we belong to, for list removal purposes */
3335 struct cgroup *owner;
3336 /* protects the other fields */
3337 struct rw_semaphore mutex;
3341 * The following two functions "fix" the issue where there are more pids
3342 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3343 * TODO: replace with a kernel-wide solution to this problem
3345 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3346 static void *pidlist_allocate(int count)
3348 if (PIDLIST_TOO_LARGE(count))
3349 return vmalloc(count * sizeof(pid_t));
3351 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3353 static void pidlist_free(void *p)
3355 if (is_vmalloc_addr(p))
3362 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3363 * Returns the number of unique elements.
3365 static int pidlist_uniq(pid_t *list, int length)
3370 * we presume the 0th element is unique, so i starts at 1. trivial
3371 * edge cases first; no work needs to be done for either
3373 if (length == 0 || length == 1)
3375 /* src and dest walk down the list; dest counts unique elements */
3376 for (src = 1; src < length; src++) {
3377 /* find next unique element */
3378 while (list[src] == list[src-1]) {
3383 /* dest always points to where the next unique element goes */
3384 list[dest] = list[src];
3391 static int cmppid(const void *a, const void *b)
3393 return *(pid_t *)a - *(pid_t *)b;
3397 * find the appropriate pidlist for our purpose (given procs vs tasks)
3398 * returns with the lock on that pidlist already held, and takes care
3399 * of the use count, or returns NULL with no locks held if we're out of
3402 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3403 enum cgroup_filetype type)
3405 struct cgroup_pidlist *l;
3406 /* don't need task_nsproxy() if we're looking at ourself */
3407 struct pid_namespace *ns = task_active_pid_ns(current);
3410 * We can't drop the pidlist_mutex before taking the l->mutex in case
3411 * the last ref-holder is trying to remove l from the list at the same
3412 * time. Holding the pidlist_mutex precludes somebody taking whichever
3413 * list we find out from under us - compare release_pid_array().
3415 mutex_lock(&cgrp->pidlist_mutex);
3416 list_for_each_entry(l, &cgrp->pidlists, links) {
3417 if (l->key.type == type && l->key.ns == ns) {
3418 /* make sure l doesn't vanish out from under us */
3419 down_write(&l->mutex);
3420 mutex_unlock(&cgrp->pidlist_mutex);
3424 /* entry not found; create a new one */
3425 l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3427 mutex_unlock(&cgrp->pidlist_mutex);
3430 init_rwsem(&l->mutex);
3431 down_write(&l->mutex);
3433 l->key.ns = get_pid_ns(ns);
3434 l->use_count = 0; /* don't increment here */
3437 list_add(&l->links, &cgrp->pidlists);
3438 mutex_unlock(&cgrp->pidlist_mutex);
3443 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3445 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3446 struct cgroup_pidlist **lp)
3450 int pid, n = 0; /* used for populating the array */
3451 struct cgroup_iter it;
3452 struct task_struct *tsk;
3453 struct cgroup_pidlist *l;
3456 * If cgroup gets more users after we read count, we won't have
3457 * enough space - tough. This race is indistinguishable to the
3458 * caller from the case that the additional cgroup users didn't
3459 * show up until sometime later on.
3461 length = cgroup_task_count(cgrp);
3462 array = pidlist_allocate(length);
3465 /* now, populate the array */
3466 cgroup_iter_start(cgrp, &it);
3467 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3468 if (unlikely(n == length))
3470 /* get tgid or pid for procs or tasks file respectively */
3471 if (type == CGROUP_FILE_PROCS)
3472 pid = task_tgid_vnr(tsk);
3474 pid = task_pid_vnr(tsk);
3475 if (pid > 0) /* make sure to only use valid results */
3478 cgroup_iter_end(cgrp, &it);
3480 /* now sort & (if procs) strip out duplicates */
3481 sort(array, length, sizeof(pid_t), cmppid, NULL);
3482 if (type == CGROUP_FILE_PROCS)
3483 length = pidlist_uniq(array, length);
3484 l = cgroup_pidlist_find(cgrp, type);
3486 pidlist_free(array);
3489 /* store array, freeing old if necessary - lock already held */
3490 pidlist_free(l->list);
3494 up_write(&l->mutex);
3500 * cgroupstats_build - build and fill cgroupstats
3501 * @stats: cgroupstats to fill information into
3502 * @dentry: A dentry entry belonging to the cgroup for which stats have
3505 * Build and fill cgroupstats so that taskstats can export it to user
3508 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3511 struct cgroup *cgrp;
3512 struct cgroup_iter it;
3513 struct task_struct *tsk;
3516 * Validate dentry by checking the superblock operations,
3517 * and make sure it's a directory.
3519 if (dentry->d_sb->s_op != &cgroup_ops ||
3520 !S_ISDIR(dentry->d_inode->i_mode))
3524 cgrp = dentry->d_fsdata;
3526 cgroup_iter_start(cgrp, &it);
3527 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3528 switch (tsk->state) {
3530 stats->nr_running++;
3532 case TASK_INTERRUPTIBLE:
3533 stats->nr_sleeping++;
3535 case TASK_UNINTERRUPTIBLE:
3536 stats->nr_uninterruptible++;
3539 stats->nr_stopped++;
3542 if (delayacct_is_task_waiting_on_io(tsk))
3543 stats->nr_io_wait++;
3547 cgroup_iter_end(cgrp, &it);
3555 * seq_file methods for the tasks/procs files. The seq_file position is the
3556 * next pid to display; the seq_file iterator is a pointer to the pid
3557 * in the cgroup->l->list array.
3560 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3563 * Initially we receive a position value that corresponds to
3564 * one more than the last pid shown (or 0 on the first call or
3565 * after a seek to the start). Use a binary-search to find the
3566 * next pid to display, if any
3568 struct cgroup_pidlist *l = s->private;
3569 int index = 0, pid = *pos;
3572 down_read(&l->mutex);
3574 int end = l->length;
3576 while (index < end) {
3577 int mid = (index + end) / 2;
3578 if (l->list[mid] == pid) {
3581 } else if (l->list[mid] <= pid)
3587 /* If we're off the end of the array, we're done */
3588 if (index >= l->length)
3590 /* Update the abstract position to be the actual pid that we found */
3591 iter = l->list + index;
3596 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3598 struct cgroup_pidlist *l = s->private;
3602 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3604 struct cgroup_pidlist *l = s->private;
3606 pid_t *end = l->list + l->length;
3608 * Advance to the next pid in the array. If this goes off the
3620 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3622 return seq_printf(s, "%d\n", *(int *)v);
3626 * seq_operations functions for iterating on pidlists through seq_file -
3627 * independent of whether it's tasks or procs
3629 static const struct seq_operations cgroup_pidlist_seq_operations = {
3630 .start = cgroup_pidlist_start,
3631 .stop = cgroup_pidlist_stop,
3632 .next = cgroup_pidlist_next,
3633 .show = cgroup_pidlist_show,
3636 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3639 * the case where we're the last user of this particular pidlist will
3640 * have us remove it from the cgroup's list, which entails taking the
3641 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3642 * pidlist_mutex, we have to take pidlist_mutex first.
3644 mutex_lock(&l->owner->pidlist_mutex);
3645 down_write(&l->mutex);
3646 BUG_ON(!l->use_count);
3647 if (!--l->use_count) {
3648 /* we're the last user if refcount is 0; remove and free */
3649 list_del(&l->links);
3650 mutex_unlock(&l->owner->pidlist_mutex);
3651 pidlist_free(l->list);
3652 put_pid_ns(l->key.ns);
3653 up_write(&l->mutex);
3657 mutex_unlock(&l->owner->pidlist_mutex);
3658 up_write(&l->mutex);
3661 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3663 struct cgroup_pidlist *l;
3664 if (!(file->f_mode & FMODE_READ))
3667 * the seq_file will only be initialized if the file was opened for
3668 * reading; hence we check if it's not null only in that case.
3670 l = ((struct seq_file *)file->private_data)->private;
3671 cgroup_release_pid_array(l);
3672 return seq_release(inode, file);
3675 static const struct file_operations cgroup_pidlist_operations = {
3677 .llseek = seq_lseek,
3678 .write = cgroup_file_write,
3679 .release = cgroup_pidlist_release,
3683 * The following functions handle opens on a file that displays a pidlist
3684 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3687 /* helper function for the two below it */
3688 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3690 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3691 struct cgroup_pidlist *l;
3694 /* Nothing to do for write-only files */
3695 if (!(file->f_mode & FMODE_READ))
3698 /* have the array populated */
3699 retval = pidlist_array_load(cgrp, type, &l);
3702 /* configure file information */
3703 file->f_op = &cgroup_pidlist_operations;
3705 retval = seq_open(file, &cgroup_pidlist_seq_operations);
3707 cgroup_release_pid_array(l);
3710 ((struct seq_file *)file->private_data)->private = l;
3713 static int cgroup_tasks_open(struct inode *unused, struct file *file)
3715 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3717 static int cgroup_procs_open(struct inode *unused, struct file *file)
3719 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3722 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3725 return notify_on_release(cgrp);
3728 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3732 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3734 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3736 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3741 * Unregister event and free resources.
3743 * Gets called from workqueue.
3745 static void cgroup_event_remove(struct work_struct *work)
3747 struct cgroup_event *event = container_of(work, struct cgroup_event,
3749 struct cgroup *cgrp = event->cgrp;
3751 remove_wait_queue(event->wqh, &event->wait);
3753 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3755 /* Notify userspace the event is going away. */
3756 eventfd_signal(event->eventfd, 1);
3758 eventfd_ctx_put(event->eventfd);
3764 * Gets called on POLLHUP on eventfd when user closes it.
3766 * Called with wqh->lock held and interrupts disabled.
3768 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3769 int sync, void *key)
3771 struct cgroup_event *event = container_of(wait,
3772 struct cgroup_event, wait);
3773 struct cgroup *cgrp = event->cgrp;
3774 unsigned long flags = (unsigned long)key;
3776 if (flags & POLLHUP) {
3778 * If the event has been detached at cgroup removal, we
3779 * can simply return knowing the other side will cleanup
3782 * We can't race against event freeing since the other
3783 * side will require wqh->lock via remove_wait_queue(),
3786 spin_lock(&cgrp->event_list_lock);
3787 if (!list_empty(&event->list)) {
3788 list_del_init(&event->list);
3790 * We are in atomic context, but cgroup_event_remove()
3791 * may sleep, so we have to call it in workqueue.
3793 schedule_work(&event->remove);
3795 spin_unlock(&cgrp->event_list_lock);
3801 static void cgroup_event_ptable_queue_proc(struct file *file,
3802 wait_queue_head_t *wqh, poll_table *pt)
3804 struct cgroup_event *event = container_of(pt,
3805 struct cgroup_event, pt);
3808 add_wait_queue(wqh, &event->wait);
3812 * Parse input and register new cgroup event handler.
3814 * Input must be in format '<event_fd> <control_fd> <args>'.
3815 * Interpretation of args is defined by control file implementation.
3817 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3820 struct cgroup_event *event = NULL;
3821 struct cgroup *cgrp_cfile;
3822 unsigned int efd, cfd;
3823 struct file *efile = NULL;
3824 struct file *cfile = NULL;
3828 efd = simple_strtoul(buffer, &endp, 10);
3833 cfd = simple_strtoul(buffer, &endp, 10);
3834 if ((*endp != ' ') && (*endp != '\0'))
3838 event = kzalloc(sizeof(*event), GFP_KERNEL);
3842 INIT_LIST_HEAD(&event->list);
3843 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3844 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3845 INIT_WORK(&event->remove, cgroup_event_remove);
3847 efile = eventfd_fget(efd);
3848 if (IS_ERR(efile)) {
3849 ret = PTR_ERR(efile);
3853 event->eventfd = eventfd_ctx_fileget(efile);
3854 if (IS_ERR(event->eventfd)) {
3855 ret = PTR_ERR(event->eventfd);
3865 /* the process need read permission on control file */
3866 /* AV: shouldn't we check that it's been opened for read instead? */
3867 ret = inode_permission(file_inode(cfile), MAY_READ);
3871 event->cft = __file_cft(cfile);
3872 if (IS_ERR(event->cft)) {
3873 ret = PTR_ERR(event->cft);
3878 * The file to be monitored must be in the same cgroup as
3879 * cgroup.event_control is.
3881 cgrp_cfile = __d_cgrp(cfile->f_dentry->d_parent);
3882 if (cgrp_cfile != cgrp) {
3887 if (!event->cft->register_event || !event->cft->unregister_event) {
3892 ret = event->cft->register_event(cgrp, event->cft,
3893 event->eventfd, buffer);
3897 efile->f_op->poll(efile, &event->pt);
3900 * Events should be removed after rmdir of cgroup directory, but before
3901 * destroying subsystem state objects. Let's take reference to cgroup
3902 * directory dentry to do that.
3906 spin_lock(&cgrp->event_list_lock);
3907 list_add(&event->list, &cgrp->event_list);
3908 spin_unlock(&cgrp->event_list_lock);
3919 if (event && event->eventfd && !IS_ERR(event->eventfd))
3920 eventfd_ctx_put(event->eventfd);
3922 if (!IS_ERR_OR_NULL(efile))
3930 static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3933 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3936 static int cgroup_clone_children_write(struct cgroup *cgrp,
3941 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3943 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3948 * for the common functions, 'private' gives the type of file
3950 /* for hysterical raisins, we can't put this on the older files */
3951 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3952 static struct cftype files[] = {
3955 .open = cgroup_tasks_open,
3956 .write_u64 = cgroup_tasks_write,
3957 .release = cgroup_pidlist_release,
3958 .mode = S_IRUGO | S_IWUSR,
3961 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3962 .open = cgroup_procs_open,
3963 .write_u64 = cgroup_procs_write,
3964 .release = cgroup_pidlist_release,
3965 .mode = S_IRUGO | S_IWUSR,
3968 .name = "notify_on_release",
3969 .read_u64 = cgroup_read_notify_on_release,
3970 .write_u64 = cgroup_write_notify_on_release,
3973 .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3974 .write_string = cgroup_write_event_control,
3978 .name = "cgroup.clone_children",
3979 .flags = CFTYPE_INSANE,
3980 .read_u64 = cgroup_clone_children_read,
3981 .write_u64 = cgroup_clone_children_write,
3984 .name = "cgroup.sane_behavior",
3985 .flags = CFTYPE_ONLY_ON_ROOT,
3986 .read_seq_string = cgroup_sane_behavior_show,
3989 .name = "release_agent",
3990 .flags = CFTYPE_ONLY_ON_ROOT,
3991 .read_seq_string = cgroup_release_agent_show,
3992 .write_string = cgroup_release_agent_write,
3993 .max_write_len = PATH_MAX,
3999 * cgroup_populate_dir - selectively creation of files in a directory
4000 * @cgrp: target cgroup
4001 * @base_files: true if the base files should be added
4002 * @subsys_mask: mask of the subsystem ids whose files should be added
4004 static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
4005 unsigned long subsys_mask)
4008 struct cgroup_subsys *ss;
4011 err = cgroup_addrm_files(cgrp, NULL, files, true);
4016 /* process cftsets of each subsystem */
4017 for_each_subsys(cgrp->root, ss) {
4018 struct cftype_set *set;
4019 if (!test_bit(ss->subsys_id, &subsys_mask))
4022 list_for_each_entry(set, &ss->cftsets, node)
4023 cgroup_addrm_files(cgrp, ss, set->cfts, true);
4026 /* This cgroup is ready now */
4027 for_each_subsys(cgrp->root, ss) {
4028 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4030 * Update id->css pointer and make this css visible from
4031 * CSS ID functions. This pointer will be dereferened
4032 * from RCU-read-side without locks.
4035 rcu_assign_pointer(css->id->css, css);
4041 static void css_dput_fn(struct work_struct *work)
4043 struct cgroup_subsys_state *css =
4044 container_of(work, struct cgroup_subsys_state, dput_work);
4045 struct dentry *dentry = css->cgroup->dentry;
4046 struct super_block *sb = dentry->d_sb;
4048 atomic_inc(&sb->s_active);
4050 deactivate_super(sb);
4053 static void init_cgroup_css(struct cgroup_subsys_state *css,
4054 struct cgroup_subsys *ss,
4055 struct cgroup *cgrp)
4058 atomic_set(&css->refcnt, 1);
4061 if (cgrp == dummytop)
4062 css->flags |= CSS_ROOT;
4063 BUG_ON(cgrp->subsys[ss->subsys_id]);
4064 cgrp->subsys[ss->subsys_id] = css;
4067 * css holds an extra ref to @cgrp->dentry which is put on the last
4068 * css_put(). dput() requires process context, which css_put() may
4069 * be called without. @css->dput_work will be used to invoke
4070 * dput() asynchronously from css_put().
4072 INIT_WORK(&css->dput_work, css_dput_fn);
4075 /* invoke ->post_create() on a new CSS and mark it online if successful */
4076 static int online_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4080 lockdep_assert_held(&cgroup_mutex);
4083 ret = ss->css_online(cgrp);
4085 cgrp->subsys[ss->subsys_id]->flags |= CSS_ONLINE;
4089 /* if the CSS is online, invoke ->pre_destory() on it and mark it offline */
4090 static void offline_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4091 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4093 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4095 lockdep_assert_held(&cgroup_mutex);
4097 if (!(css->flags & CSS_ONLINE))
4100 if (ss->css_offline)
4101 ss->css_offline(cgrp);
4103 cgrp->subsys[ss->subsys_id]->flags &= ~CSS_ONLINE;
4107 * cgroup_create - create a cgroup
4108 * @parent: cgroup that will be parent of the new cgroup
4109 * @dentry: dentry of the new cgroup
4110 * @mode: mode to set on new inode
4112 * Must be called with the mutex on the parent inode held
4114 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
4117 struct cgroup *cgrp;
4118 struct cgroup_name *name;
4119 struct cgroupfs_root *root = parent->root;
4121 struct cgroup_subsys *ss;
4122 struct super_block *sb = root->sb;
4124 /* allocate the cgroup and its ID, 0 is reserved for the root */
4125 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
4129 name = cgroup_alloc_name(dentry);
4132 rcu_assign_pointer(cgrp->name, name);
4134 cgrp->id = ida_simple_get(&root->cgroup_ida, 1, 0, GFP_KERNEL);
4139 * Only live parents can have children. Note that the liveliness
4140 * check isn't strictly necessary because cgroup_mkdir() and
4141 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4142 * anyway so that locking is contained inside cgroup proper and we
4143 * don't get nasty surprises if we ever grow another caller.
4145 if (!cgroup_lock_live_group(parent)) {
4150 /* Grab a reference on the superblock so the hierarchy doesn't
4151 * get deleted on unmount if there are child cgroups. This
4152 * can be done outside cgroup_mutex, since the sb can't
4153 * disappear while someone has an open control file on the
4155 atomic_inc(&sb->s_active);
4157 init_cgroup_housekeeping(cgrp);
4159 dentry->d_fsdata = cgrp;
4160 cgrp->dentry = dentry;
4162 cgrp->parent = parent;
4163 cgrp->root = parent->root;
4165 if (notify_on_release(parent))
4166 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
4168 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
4169 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4171 for_each_subsys(root, ss) {
4172 struct cgroup_subsys_state *css;
4174 css = ss->css_alloc(cgrp);
4179 init_cgroup_css(css, ss, cgrp);
4181 err = alloc_css_id(ss, parent, cgrp);
4188 * Create directory. cgroup_create_file() returns with the new
4189 * directory locked on success so that it can be populated without
4190 * dropping cgroup_mutex.
4192 err = cgroup_create_file(dentry, S_IFDIR | mode, sb);
4195 lockdep_assert_held(&dentry->d_inode->i_mutex);
4197 /* allocation complete, commit to creation */
4198 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
4199 list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children);
4200 root->number_of_cgroups++;
4202 /* each css holds a ref to the cgroup's dentry */
4203 for_each_subsys(root, ss)
4206 /* hold a ref to the parent's dentry */
4207 dget(parent->dentry);
4209 /* creation succeeded, notify subsystems */
4210 for_each_subsys(root, ss) {
4211 err = online_css(ss, cgrp);
4215 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
4217 pr_warning("cgroup: %s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
4218 current->comm, current->pid, ss->name);
4219 if (!strcmp(ss->name, "memory"))
4220 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4221 ss->warned_broken_hierarchy = true;
4225 err = cgroup_populate_dir(cgrp, true, root->subsys_mask);
4229 mutex_unlock(&cgroup_mutex);
4230 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
4235 for_each_subsys(root, ss) {
4236 if (cgrp->subsys[ss->subsys_id])
4239 mutex_unlock(&cgroup_mutex);
4240 /* Release the reference count that we took on the superblock */
4241 deactivate_super(sb);
4243 ida_simple_remove(&root->cgroup_ida, cgrp->id);
4245 kfree(rcu_dereference_raw(cgrp->name));
4251 cgroup_destroy_locked(cgrp);
4252 mutex_unlock(&cgroup_mutex);
4253 mutex_unlock(&dentry->d_inode->i_mutex);
4257 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4259 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
4261 /* the vfs holds inode->i_mutex already */
4262 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
4265 static int cgroup_destroy_locked(struct cgroup *cgrp)
4266 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4268 struct dentry *d = cgrp->dentry;
4269 struct cgroup *parent = cgrp->parent;
4270 struct cgroup_event *event, *tmp;
4271 struct cgroup_subsys *ss;
4273 lockdep_assert_held(&d->d_inode->i_mutex);
4274 lockdep_assert_held(&cgroup_mutex);
4276 if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children))
4280 * Block new css_tryget() by deactivating refcnt and mark @cgrp
4281 * removed. This makes future css_tryget() and child creation
4282 * attempts fail thus maintaining the removal conditions verified
4285 for_each_subsys(cgrp->root, ss) {
4286 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4288 WARN_ON(atomic_read(&css->refcnt) < 0);
4289 atomic_add(CSS_DEACT_BIAS, &css->refcnt);
4291 set_bit(CGRP_REMOVED, &cgrp->flags);
4293 /* tell subsystems to initate destruction */
4294 for_each_subsys(cgrp->root, ss)
4295 offline_css(ss, cgrp);
4298 * Put all the base refs. Each css holds an extra reference to the
4299 * cgroup's dentry and cgroup removal proceeds regardless of css
4300 * refs. On the last put of each css, whenever that may be, the
4301 * extra dentry ref is put so that dentry destruction happens only
4302 * after all css's are released.
4304 for_each_subsys(cgrp->root, ss)
4305 css_put(cgrp->subsys[ss->subsys_id]);
4307 raw_spin_lock(&release_list_lock);
4308 if (!list_empty(&cgrp->release_list))
4309 list_del_init(&cgrp->release_list);
4310 raw_spin_unlock(&release_list_lock);
4312 /* delete this cgroup from parent->children */
4313 list_del_rcu(&cgrp->sibling);
4314 list_del_init(&cgrp->allcg_node);
4317 cgroup_d_remove_dir(d);
4320 set_bit(CGRP_RELEASABLE, &parent->flags);
4321 check_for_release(parent);
4324 * Unregister events and notify userspace.
4325 * Notify userspace about cgroup removing only after rmdir of cgroup
4326 * directory to avoid race between userspace and kernelspace.
4328 spin_lock(&cgrp->event_list_lock);
4329 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4330 list_del_init(&event->list);
4331 schedule_work(&event->remove);
4333 spin_unlock(&cgrp->event_list_lock);
4338 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
4342 mutex_lock(&cgroup_mutex);
4343 ret = cgroup_destroy_locked(dentry->d_fsdata);
4344 mutex_unlock(&cgroup_mutex);
4349 static void __init_or_module cgroup_init_cftsets(struct cgroup_subsys *ss)
4351 INIT_LIST_HEAD(&ss->cftsets);
4354 * base_cftset is embedded in subsys itself, no need to worry about
4357 if (ss->base_cftypes) {
4358 ss->base_cftset.cfts = ss->base_cftypes;
4359 list_add_tail(&ss->base_cftset.node, &ss->cftsets);
4363 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4365 struct cgroup_subsys_state *css;
4367 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4369 mutex_lock(&cgroup_mutex);
4371 /* init base cftset */
4372 cgroup_init_cftsets(ss);
4374 /* Create the top cgroup state for this subsystem */
4375 list_add(&ss->sibling, &rootnode.subsys_list);
4376 ss->root = &rootnode;
4377 css = ss->css_alloc(dummytop);
4378 /* We don't handle early failures gracefully */
4379 BUG_ON(IS_ERR(css));
4380 init_cgroup_css(css, ss, dummytop);
4382 /* Update the init_css_set to contain a subsys
4383 * pointer to this state - since the subsystem is
4384 * newly registered, all tasks and hence the
4385 * init_css_set is in the subsystem's top cgroup. */
4386 init_css_set.subsys[ss->subsys_id] = css;
4388 need_forkexit_callback |= ss->fork || ss->exit;
4390 /* At system boot, before all subsystems have been
4391 * registered, no tasks have been forked, so we don't
4392 * need to invoke fork callbacks here. */
4393 BUG_ON(!list_empty(&init_task.tasks));
4395 BUG_ON(online_css(ss, dummytop));
4397 mutex_unlock(&cgroup_mutex);
4399 /* this function shouldn't be used with modular subsystems, since they
4400 * need to register a subsys_id, among other things */
4405 * cgroup_load_subsys: load and register a modular subsystem at runtime
4406 * @ss: the subsystem to load
4408 * This function should be called in a modular subsystem's initcall. If the
4409 * subsystem is built as a module, it will be assigned a new subsys_id and set
4410 * up for use. If the subsystem is built-in anyway, work is delegated to the
4411 * simpler cgroup_init_subsys.
4413 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4415 struct cgroup_subsys_state *css;
4417 struct hlist_node *tmp;
4421 /* check name and function validity */
4422 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4423 ss->css_alloc == NULL || ss->css_free == NULL)
4427 * we don't support callbacks in modular subsystems. this check is
4428 * before the ss->module check for consistency; a subsystem that could
4429 * be a module should still have no callbacks even if the user isn't
4430 * compiling it as one.
4432 if (ss->fork || ss->exit)
4436 * an optionally modular subsystem is built-in: we want to do nothing,
4437 * since cgroup_init_subsys will have already taken care of it.
4439 if (ss->module == NULL) {
4440 /* a sanity check */
4441 BUG_ON(subsys[ss->subsys_id] != ss);
4445 /* init base cftset */
4446 cgroup_init_cftsets(ss);
4448 mutex_lock(&cgroup_mutex);
4449 subsys[ss->subsys_id] = ss;
4452 * no ss->css_alloc seems to need anything important in the ss
4453 * struct, so this can happen first (i.e. before the rootnode
4456 css = ss->css_alloc(dummytop);
4458 /* failure case - need to deassign the subsys[] slot. */
4459 subsys[ss->subsys_id] = NULL;
4460 mutex_unlock(&cgroup_mutex);
4461 return PTR_ERR(css);
4464 list_add(&ss->sibling, &rootnode.subsys_list);
4465 ss->root = &rootnode;
4467 /* our new subsystem will be attached to the dummy hierarchy. */
4468 init_cgroup_css(css, ss, dummytop);
4469 /* init_idr must be after init_cgroup_css because it sets css->id. */
4471 ret = cgroup_init_idr(ss, css);
4477 * Now we need to entangle the css into the existing css_sets. unlike
4478 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4479 * will need a new pointer to it; done by iterating the css_set_table.
4480 * furthermore, modifying the existing css_sets will corrupt the hash
4481 * table state, so each changed css_set will need its hash recomputed.
4482 * this is all done under the css_set_lock.
4484 write_lock(&css_set_lock);
4485 hash_for_each_safe(css_set_table, i, tmp, cg, hlist) {
4486 /* skip entries that we already rehashed */
4487 if (cg->subsys[ss->subsys_id])
4489 /* remove existing entry */
4490 hash_del(&cg->hlist);
4492 cg->subsys[ss->subsys_id] = css;
4493 /* recompute hash and restore entry */
4494 key = css_set_hash(cg->subsys);
4495 hash_add(css_set_table, &cg->hlist, key);
4497 write_unlock(&css_set_lock);
4499 ret = online_css(ss, dummytop);
4504 mutex_unlock(&cgroup_mutex);
4508 mutex_unlock(&cgroup_mutex);
4509 /* @ss can't be mounted here as try_module_get() would fail */
4510 cgroup_unload_subsys(ss);
4513 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4516 * cgroup_unload_subsys: unload a modular subsystem
4517 * @ss: the subsystem to unload
4519 * This function should be called in a modular subsystem's exitcall. When this
4520 * function is invoked, the refcount on the subsystem's module will be 0, so
4521 * the subsystem will not be attached to any hierarchy.
4523 void cgroup_unload_subsys(struct cgroup_subsys *ss)
4525 struct cg_cgroup_link *link;
4527 BUG_ON(ss->module == NULL);
4530 * we shouldn't be called if the subsystem is in use, and the use of
4531 * try_module_get in parse_cgroupfs_options should ensure that it
4532 * doesn't start being used while we're killing it off.
4534 BUG_ON(ss->root != &rootnode);
4536 mutex_lock(&cgroup_mutex);
4538 offline_css(ss, dummytop);
4541 idr_destroy(&ss->idr);
4543 /* deassign the subsys_id */
4544 subsys[ss->subsys_id] = NULL;
4546 /* remove subsystem from rootnode's list of subsystems */
4547 list_del_init(&ss->sibling);
4550 * disentangle the css from all css_sets attached to the dummytop. as
4551 * in loading, we need to pay our respects to the hashtable gods.
4553 write_lock(&css_set_lock);
4554 list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
4555 struct css_set *cg = link->cg;
4558 hash_del(&cg->hlist);
4559 cg->subsys[ss->subsys_id] = NULL;
4560 key = css_set_hash(cg->subsys);
4561 hash_add(css_set_table, &cg->hlist, key);
4563 write_unlock(&css_set_lock);
4566 * remove subsystem's css from the dummytop and free it - need to
4567 * free before marking as null because ss->css_free needs the
4568 * cgrp->subsys pointer to find their state. note that this also
4569 * takes care of freeing the css_id.
4571 ss->css_free(dummytop);
4572 dummytop->subsys[ss->subsys_id] = NULL;
4574 mutex_unlock(&cgroup_mutex);
4576 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4579 * cgroup_init_early - cgroup initialization at system boot
4581 * Initialize cgroups at system boot, and initialize any
4582 * subsystems that request early init.
4584 int __init cgroup_init_early(void)
4587 atomic_set(&init_css_set.refcount, 1);
4588 INIT_LIST_HEAD(&init_css_set.cg_links);
4589 INIT_LIST_HEAD(&init_css_set.tasks);
4590 INIT_HLIST_NODE(&init_css_set.hlist);
4592 init_cgroup_root(&rootnode);
4594 init_task.cgroups = &init_css_set;
4596 init_css_set_link.cg = &init_css_set;
4597 init_css_set_link.cgrp = dummytop;
4598 list_add(&init_css_set_link.cgrp_link_list,
4599 &rootnode.top_cgroup.css_sets);
4600 list_add(&init_css_set_link.cg_link_list,
4601 &init_css_set.cg_links);
4603 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4604 struct cgroup_subsys *ss = subsys[i];
4606 /* at bootup time, we don't worry about modular subsystems */
4607 if (!ss || ss->module)
4611 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4612 BUG_ON(!ss->css_alloc);
4613 BUG_ON(!ss->css_free);
4614 if (ss->subsys_id != i) {
4615 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4616 ss->name, ss->subsys_id);
4621 cgroup_init_subsys(ss);
4627 * cgroup_init - cgroup initialization
4629 * Register cgroup filesystem and /proc file, and initialize
4630 * any subsystems that didn't request early init.
4632 int __init cgroup_init(void)
4638 err = bdi_init(&cgroup_backing_dev_info);
4642 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4643 struct cgroup_subsys *ss = subsys[i];
4645 /* at bootup time, we don't worry about modular subsystems */
4646 if (!ss || ss->module)
4648 if (!ss->early_init)
4649 cgroup_init_subsys(ss);
4651 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4654 /* Add init_css_set to the hash table */
4655 key = css_set_hash(init_css_set.subsys);
4656 hash_add(css_set_table, &init_css_set.hlist, key);
4658 /* allocate id for the dummy hierarchy */
4659 BUG_ON(cgroup_init_root_id(&rootnode));
4661 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4667 err = register_filesystem(&cgroup_fs_type);
4669 kobject_put(cgroup_kobj);
4673 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4677 bdi_destroy(&cgroup_backing_dev_info);
4683 * proc_cgroup_show()
4684 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4685 * - Used for /proc/<pid>/cgroup.
4686 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4687 * doesn't really matter if tsk->cgroup changes after we read it,
4688 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4689 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4690 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4691 * cgroup to top_cgroup.
4694 /* TODO: Use a proper seq_file iterator */
4695 int proc_cgroup_show(struct seq_file *m, void *v)
4698 struct task_struct *tsk;
4701 struct cgroupfs_root *root;
4704 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4710 tsk = get_pid_task(pid, PIDTYPE_PID);
4716 mutex_lock(&cgroup_mutex);
4718 for_each_active_root(root) {
4719 struct cgroup_subsys *ss;
4720 struct cgroup *cgrp;
4723 seq_printf(m, "%d:", root->hierarchy_id);
4724 for_each_subsys(root, ss)
4725 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4726 if (strlen(root->name))
4727 seq_printf(m, "%sname=%s", count ? "," : "",
4730 cgrp = task_cgroup_from_root(tsk, root);
4731 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4739 mutex_unlock(&cgroup_mutex);
4740 put_task_struct(tsk);
4747 /* Display information about each subsystem and each hierarchy */
4748 static int proc_cgroupstats_show(struct seq_file *m, void *v)
4752 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4754 * ideally we don't want subsystems moving around while we do this.
4755 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4756 * subsys/hierarchy state.
4758 mutex_lock(&cgroup_mutex);
4759 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4760 struct cgroup_subsys *ss = subsys[i];
4763 seq_printf(m, "%s\t%d\t%d\t%d\n",
4764 ss->name, ss->root->hierarchy_id,
4765 ss->root->number_of_cgroups, !ss->disabled);
4767 mutex_unlock(&cgroup_mutex);
4771 static int cgroupstats_open(struct inode *inode, struct file *file)
4773 return single_open(file, proc_cgroupstats_show, NULL);
4776 static const struct file_operations proc_cgroupstats_operations = {
4777 .open = cgroupstats_open,
4779 .llseek = seq_lseek,
4780 .release = single_release,
4784 * cgroup_fork - attach newly forked task to its parents cgroup.
4785 * @child: pointer to task_struct of forking parent process.
4787 * Description: A task inherits its parent's cgroup at fork().
4789 * A pointer to the shared css_set was automatically copied in
4790 * fork.c by dup_task_struct(). However, we ignore that copy, since
4791 * it was not made under the protection of RCU or cgroup_mutex, so
4792 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4793 * have already changed current->cgroups, allowing the previously
4794 * referenced cgroup group to be removed and freed.
4796 * At the point that cgroup_fork() is called, 'current' is the parent
4797 * task, and the passed argument 'child' points to the child task.
4799 void cgroup_fork(struct task_struct *child)
4802 child->cgroups = current->cgroups;
4803 get_css_set(child->cgroups);
4804 task_unlock(current);
4805 INIT_LIST_HEAD(&child->cg_list);
4809 * cgroup_post_fork - called on a new task after adding it to the task list
4810 * @child: the task in question
4812 * Adds the task to the list running through its css_set if necessary and
4813 * call the subsystem fork() callbacks. Has to be after the task is
4814 * visible on the task list in case we race with the first call to
4815 * cgroup_iter_start() - to guarantee that the new task ends up on its
4818 void cgroup_post_fork(struct task_struct *child)
4823 * use_task_css_set_links is set to 1 before we walk the tasklist
4824 * under the tasklist_lock and we read it here after we added the child
4825 * to the tasklist under the tasklist_lock as well. If the child wasn't
4826 * yet in the tasklist when we walked through it from
4827 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4828 * should be visible now due to the paired locking and barriers implied
4829 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4830 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4833 if (use_task_css_set_links) {
4834 write_lock(&css_set_lock);
4836 if (list_empty(&child->cg_list))
4837 list_add(&child->cg_list, &child->cgroups->tasks);
4839 write_unlock(&css_set_lock);
4843 * Call ss->fork(). This must happen after @child is linked on
4844 * css_set; otherwise, @child might change state between ->fork()
4845 * and addition to css_set.
4847 if (need_forkexit_callback) {
4849 * fork/exit callbacks are supported only for builtin
4850 * subsystems, and the builtin section of the subsys
4851 * array is immutable, so we don't need to lock the
4852 * subsys array here. On the other hand, modular section
4853 * of the array can be freed at module unload, so we
4856 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4857 struct cgroup_subsys *ss = subsys[i];
4866 * cgroup_exit - detach cgroup from exiting task
4867 * @tsk: pointer to task_struct of exiting process
4868 * @run_callback: run exit callbacks?
4870 * Description: Detach cgroup from @tsk and release it.
4872 * Note that cgroups marked notify_on_release force every task in
4873 * them to take the global cgroup_mutex mutex when exiting.
4874 * This could impact scaling on very large systems. Be reluctant to
4875 * use notify_on_release cgroups where very high task exit scaling
4876 * is required on large systems.
4878 * the_top_cgroup_hack:
4880 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4882 * We call cgroup_exit() while the task is still competent to
4883 * handle notify_on_release(), then leave the task attached to the
4884 * root cgroup in each hierarchy for the remainder of its exit.
4886 * To do this properly, we would increment the reference count on
4887 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4888 * code we would add a second cgroup function call, to drop that
4889 * reference. This would just create an unnecessary hot spot on
4890 * the top_cgroup reference count, to no avail.
4892 * Normally, holding a reference to a cgroup without bumping its
4893 * count is unsafe. The cgroup could go away, or someone could
4894 * attach us to a different cgroup, decrementing the count on
4895 * the first cgroup that we never incremented. But in this case,
4896 * top_cgroup isn't going away, and either task has PF_EXITING set,
4897 * which wards off any cgroup_attach_task() attempts, or task is a failed
4898 * fork, never visible to cgroup_attach_task.
4900 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4906 * Unlink from the css_set task list if necessary.
4907 * Optimistically check cg_list before taking
4910 if (!list_empty(&tsk->cg_list)) {
4911 write_lock(&css_set_lock);
4912 if (!list_empty(&tsk->cg_list))
4913 list_del_init(&tsk->cg_list);
4914 write_unlock(&css_set_lock);
4917 /* Reassign the task to the init_css_set. */
4920 tsk->cgroups = &init_css_set;
4922 if (run_callbacks && need_forkexit_callback) {
4924 * fork/exit callbacks are supported only for builtin
4925 * subsystems, see cgroup_post_fork() for details.
4927 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4928 struct cgroup_subsys *ss = subsys[i];
4931 struct cgroup *old_cgrp =
4932 rcu_dereference_raw(cg->subsys[i])->cgroup;
4933 struct cgroup *cgrp = task_cgroup(tsk, i);
4934 ss->exit(cgrp, old_cgrp, tsk);
4940 put_css_set_taskexit(cg);
4943 static void check_for_release(struct cgroup *cgrp)
4945 /* All of these checks rely on RCU to keep the cgroup
4946 * structure alive */
4947 if (cgroup_is_releasable(cgrp) &&
4948 !atomic_read(&cgrp->count) && list_empty(&cgrp->children)) {
4950 * Control Group is currently removeable. If it's not
4951 * already queued for a userspace notification, queue
4954 int need_schedule_work = 0;
4956 raw_spin_lock(&release_list_lock);
4957 if (!cgroup_is_removed(cgrp) &&
4958 list_empty(&cgrp->release_list)) {
4959 list_add(&cgrp->release_list, &release_list);
4960 need_schedule_work = 1;
4962 raw_spin_unlock(&release_list_lock);
4963 if (need_schedule_work)
4964 schedule_work(&release_agent_work);
4968 /* Caller must verify that the css is not for root cgroup */
4969 bool __css_tryget(struct cgroup_subsys_state *css)
4974 v = css_refcnt(css);
4975 t = atomic_cmpxchg(&css->refcnt, v, v + 1);
4983 EXPORT_SYMBOL_GPL(__css_tryget);
4985 /* Caller must verify that the css is not for root cgroup */
4986 void __css_put(struct cgroup_subsys_state *css)
4990 v = css_unbias_refcnt(atomic_dec_return(&css->refcnt));
4992 schedule_work(&css->dput_work);
4994 EXPORT_SYMBOL_GPL(__css_put);
4997 * Notify userspace when a cgroup is released, by running the
4998 * configured release agent with the name of the cgroup (path
4999 * relative to the root of cgroup file system) as the argument.
5001 * Most likely, this user command will try to rmdir this cgroup.
5003 * This races with the possibility that some other task will be
5004 * attached to this cgroup before it is removed, or that some other
5005 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5006 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5007 * unused, and this cgroup will be reprieved from its death sentence,
5008 * to continue to serve a useful existence. Next time it's released,
5009 * we will get notified again, if it still has 'notify_on_release' set.
5011 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5012 * means only wait until the task is successfully execve()'d. The
5013 * separate release agent task is forked by call_usermodehelper(),
5014 * then control in this thread returns here, without waiting for the
5015 * release agent task. We don't bother to wait because the caller of
5016 * this routine has no use for the exit status of the release agent
5017 * task, so no sense holding our caller up for that.
5019 static void cgroup_release_agent(struct work_struct *work)
5021 BUG_ON(work != &release_agent_work);
5022 mutex_lock(&cgroup_mutex);
5023 raw_spin_lock(&release_list_lock);
5024 while (!list_empty(&release_list)) {
5025 char *argv[3], *envp[3];
5027 char *pathbuf = NULL, *agentbuf = NULL;
5028 struct cgroup *cgrp = list_entry(release_list.next,
5031 list_del_init(&cgrp->release_list);
5032 raw_spin_unlock(&release_list_lock);
5033 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
5036 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
5038 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
5043 argv[i++] = agentbuf;
5044 argv[i++] = pathbuf;
5048 /* minimal command environment */
5049 envp[i++] = "HOME=/";
5050 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5053 /* Drop the lock while we invoke the usermode helper,
5054 * since the exec could involve hitting disk and hence
5055 * be a slow process */
5056 mutex_unlock(&cgroup_mutex);
5057 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
5058 mutex_lock(&cgroup_mutex);
5062 raw_spin_lock(&release_list_lock);
5064 raw_spin_unlock(&release_list_lock);
5065 mutex_unlock(&cgroup_mutex);
5068 static int __init cgroup_disable(char *str)
5073 while ((token = strsep(&str, ",")) != NULL) {
5076 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
5077 struct cgroup_subsys *ss = subsys[i];
5080 * cgroup_disable, being at boot time, can't
5081 * know about module subsystems, so we don't
5084 if (!ss || ss->module)
5087 if (!strcmp(token, ss->name)) {
5089 printk(KERN_INFO "Disabling %s control group"
5090 " subsystem\n", ss->name);
5097 __setup("cgroup_disable=", cgroup_disable);
5100 * Functons for CSS ID.
5104 *To get ID other than 0, this should be called when !cgroup_is_removed().
5106 unsigned short css_id(struct cgroup_subsys_state *css)
5108 struct css_id *cssid;
5111 * This css_id() can return correct value when somone has refcnt
5112 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5113 * it's unchanged until freed.
5115 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5121 EXPORT_SYMBOL_GPL(css_id);
5123 unsigned short css_depth(struct cgroup_subsys_state *css)
5125 struct css_id *cssid;
5127 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5130 return cssid->depth;
5133 EXPORT_SYMBOL_GPL(css_depth);
5136 * css_is_ancestor - test "root" css is an ancestor of "child"
5137 * @child: the css to be tested.
5138 * @root: the css supporsed to be an ancestor of the child.
5140 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5141 * this function reads css->id, the caller must hold rcu_read_lock().
5142 * But, considering usual usage, the csses should be valid objects after test.
5143 * Assuming that the caller will do some action to the child if this returns
5144 * returns true, the caller must take "child";s reference count.
5145 * If "child" is valid object and this returns true, "root" is valid, too.
5148 bool css_is_ancestor(struct cgroup_subsys_state *child,
5149 const struct cgroup_subsys_state *root)
5151 struct css_id *child_id;
5152 struct css_id *root_id;
5154 child_id = rcu_dereference(child->id);
5157 root_id = rcu_dereference(root->id);
5160 if (child_id->depth < root_id->depth)
5162 if (child_id->stack[root_id->depth] != root_id->id)
5167 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
5169 struct css_id *id = css->id;
5170 /* When this is called before css_id initialization, id can be NULL */
5174 BUG_ON(!ss->use_id);
5176 rcu_assign_pointer(id->css, NULL);
5177 rcu_assign_pointer(css->id, NULL);
5178 spin_lock(&ss->id_lock);
5179 idr_remove(&ss->idr, id->id);
5180 spin_unlock(&ss->id_lock);
5181 kfree_rcu(id, rcu_head);
5183 EXPORT_SYMBOL_GPL(free_css_id);
5186 * This is called by init or create(). Then, calls to this function are
5187 * always serialized (By cgroup_mutex() at create()).
5190 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
5192 struct css_id *newid;
5195 BUG_ON(!ss->use_id);
5197 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
5198 newid = kzalloc(size, GFP_KERNEL);
5200 return ERR_PTR(-ENOMEM);
5202 idr_preload(GFP_KERNEL);
5203 spin_lock(&ss->id_lock);
5204 /* Don't use 0. allocates an ID of 1-65535 */
5205 ret = idr_alloc(&ss->idr, newid, 1, CSS_ID_MAX + 1, GFP_NOWAIT);
5206 spin_unlock(&ss->id_lock);
5209 /* Returns error when there are no free spaces for new ID.*/
5214 newid->depth = depth;
5218 return ERR_PTR(ret);
5222 static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
5223 struct cgroup_subsys_state *rootcss)
5225 struct css_id *newid;
5227 spin_lock_init(&ss->id_lock);
5230 newid = get_new_cssid(ss, 0);
5232 return PTR_ERR(newid);
5234 newid->stack[0] = newid->id;
5235 newid->css = rootcss;
5236 rootcss->id = newid;
5240 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
5241 struct cgroup *child)
5243 int subsys_id, i, depth = 0;
5244 struct cgroup_subsys_state *parent_css, *child_css;
5245 struct css_id *child_id, *parent_id;
5247 subsys_id = ss->subsys_id;
5248 parent_css = parent->subsys[subsys_id];
5249 child_css = child->subsys[subsys_id];
5250 parent_id = parent_css->id;
5251 depth = parent_id->depth + 1;
5253 child_id = get_new_cssid(ss, depth);
5254 if (IS_ERR(child_id))
5255 return PTR_ERR(child_id);
5257 for (i = 0; i < depth; i++)
5258 child_id->stack[i] = parent_id->stack[i];
5259 child_id->stack[depth] = child_id->id;
5261 * child_id->css pointer will be set after this cgroup is available
5262 * see cgroup_populate_dir()
5264 rcu_assign_pointer(child_css->id, child_id);
5270 * css_lookup - lookup css by id
5271 * @ss: cgroup subsys to be looked into.
5274 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5275 * NULL if not. Should be called under rcu_read_lock()
5277 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5279 struct css_id *cssid = NULL;
5281 BUG_ON(!ss->use_id);
5282 cssid = idr_find(&ss->idr, id);
5284 if (unlikely(!cssid))
5287 return rcu_dereference(cssid->css);
5289 EXPORT_SYMBOL_GPL(css_lookup);
5292 * get corresponding css from file open on cgroupfs directory
5294 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5296 struct cgroup *cgrp;
5297 struct inode *inode;
5298 struct cgroup_subsys_state *css;
5300 inode = file_inode(f);
5301 /* check in cgroup filesystem dir */
5302 if (inode->i_op != &cgroup_dir_inode_operations)
5303 return ERR_PTR(-EBADF);
5305 if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5306 return ERR_PTR(-EINVAL);
5309 cgrp = __d_cgrp(f->f_dentry);
5310 css = cgrp->subsys[id];
5311 return css ? css : ERR_PTR(-ENOENT);
5314 #ifdef CONFIG_CGROUP_DEBUG
5315 static struct cgroup_subsys_state *debug_css_alloc(struct cgroup *cont)
5317 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5320 return ERR_PTR(-ENOMEM);
5325 static void debug_css_free(struct cgroup *cont)
5327 kfree(cont->subsys[debug_subsys_id]);
5330 static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
5332 return atomic_read(&cont->count);
5335 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
5337 return cgroup_task_count(cont);
5340 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
5342 return (u64)(unsigned long)current->cgroups;
5345 static u64 current_css_set_refcount_read(struct cgroup *cont,
5351 count = atomic_read(¤t->cgroups->refcount);
5356 static int current_css_set_cg_links_read(struct cgroup *cont,
5358 struct seq_file *seq)
5360 struct cg_cgroup_link *link;
5363 read_lock(&css_set_lock);
5365 cg = rcu_dereference(current->cgroups);
5366 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
5367 struct cgroup *c = link->cgrp;
5371 name = c->dentry->d_name.name;
5374 seq_printf(seq, "Root %d group %s\n",
5375 c->root->hierarchy_id, name);
5378 read_unlock(&css_set_lock);
5382 #define MAX_TASKS_SHOWN_PER_CSS 25
5383 static int cgroup_css_links_read(struct cgroup *cont,
5385 struct seq_file *seq)
5387 struct cg_cgroup_link *link;
5389 read_lock(&css_set_lock);
5390 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
5391 struct css_set *cg = link->cg;
5392 struct task_struct *task;
5394 seq_printf(seq, "css_set %p\n", cg);
5395 list_for_each_entry(task, &cg->tasks, cg_list) {
5396 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5397 seq_puts(seq, " ...\n");
5400 seq_printf(seq, " task %d\n",
5401 task_pid_vnr(task));
5405 read_unlock(&css_set_lock);
5409 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5411 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5414 static struct cftype debug_files[] = {
5416 .name = "cgroup_refcount",
5417 .read_u64 = cgroup_refcount_read,
5420 .name = "taskcount",
5421 .read_u64 = debug_taskcount_read,
5425 .name = "current_css_set",
5426 .read_u64 = current_css_set_read,
5430 .name = "current_css_set_refcount",
5431 .read_u64 = current_css_set_refcount_read,
5435 .name = "current_css_set_cg_links",
5436 .read_seq_string = current_css_set_cg_links_read,
5440 .name = "cgroup_css_links",
5441 .read_seq_string = cgroup_css_links_read,
5445 .name = "releasable",
5446 .read_u64 = releasable_read,
5452 struct cgroup_subsys debug_subsys = {
5454 .css_alloc = debug_css_alloc,
5455 .css_free = debug_css_free,
5456 .subsys_id = debug_subsys_id,
5457 .base_cftypes = debug_files,
5459 #endif /* CONFIG_CGROUP_DEBUG */