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);
852 * Drop the active superblock reference that we took when we
855 deactivate_super(cgrp->root->sb);
858 * if we're getting rid of the cgroup, refcount should ensure
859 * that there are no pidlists left.
861 BUG_ON(!list_empty(&cgrp->pidlists));
863 simple_xattrs_free(&cgrp->xattrs);
865 ida_simple_remove(&cgrp->root->cgroup_ida, cgrp->id);
866 kfree(rcu_dereference_raw(cgrp->name));
870 static void cgroup_free_rcu(struct rcu_head *head)
872 struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head);
874 schedule_work(&cgrp->free_work);
877 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
879 /* is dentry a directory ? if so, kfree() associated cgroup */
880 if (S_ISDIR(inode->i_mode)) {
881 struct cgroup *cgrp = dentry->d_fsdata;
883 BUG_ON(!(cgroup_is_removed(cgrp)));
884 call_rcu(&cgrp->rcu_head, cgroup_free_rcu);
886 struct cfent *cfe = __d_cfe(dentry);
887 struct cgroup *cgrp = dentry->d_parent->d_fsdata;
889 WARN_ONCE(!list_empty(&cfe->node) &&
890 cgrp != &cgrp->root->top_cgroup,
891 "cfe still linked for %s\n", cfe->type->name);
892 simple_xattrs_free(&cfe->xattrs);
898 static int cgroup_delete(const struct dentry *d)
903 static void remove_dir(struct dentry *d)
905 struct dentry *parent = dget(d->d_parent);
908 simple_rmdir(parent->d_inode, d);
912 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
916 lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex);
917 lockdep_assert_held(&cgroup_mutex);
920 * If we're doing cleanup due to failure of cgroup_create(),
921 * the corresponding @cfe may not exist.
923 list_for_each_entry(cfe, &cgrp->files, node) {
924 struct dentry *d = cfe->dentry;
926 if (cft && cfe->type != cft)
931 simple_unlink(cgrp->dentry->d_inode, d);
932 list_del_init(&cfe->node);
940 * cgroup_clear_directory - selective removal of base and subsystem files
941 * @dir: directory containing the files
942 * @base_files: true if the base files should be removed
943 * @subsys_mask: mask of the subsystem ids whose files should be removed
945 static void cgroup_clear_directory(struct dentry *dir, bool base_files,
946 unsigned long subsys_mask)
948 struct cgroup *cgrp = __d_cgrp(dir);
949 struct cgroup_subsys *ss;
951 for_each_subsys(cgrp->root, ss) {
952 struct cftype_set *set;
953 if (!test_bit(ss->subsys_id, &subsys_mask))
955 list_for_each_entry(set, &ss->cftsets, node)
956 cgroup_addrm_files(cgrp, NULL, set->cfts, false);
959 while (!list_empty(&cgrp->files))
960 cgroup_rm_file(cgrp, NULL);
965 * NOTE : the dentry must have been dget()'ed
967 static void cgroup_d_remove_dir(struct dentry *dentry)
969 struct dentry *parent;
970 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
972 cgroup_clear_directory(dentry, true, root->subsys_mask);
974 parent = dentry->d_parent;
975 spin_lock(&parent->d_lock);
976 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
977 list_del_init(&dentry->d_u.d_child);
978 spin_unlock(&dentry->d_lock);
979 spin_unlock(&parent->d_lock);
984 * Call with cgroup_mutex held. Drops reference counts on modules, including
985 * any duplicate ones that parse_cgroupfs_options took. If this function
986 * returns an error, no reference counts are touched.
988 static int rebind_subsystems(struct cgroupfs_root *root,
989 unsigned long final_subsys_mask)
991 unsigned long added_mask, removed_mask;
992 struct cgroup *cgrp = &root->top_cgroup;
995 BUG_ON(!mutex_is_locked(&cgroup_mutex));
996 BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
998 removed_mask = root->actual_subsys_mask & ~final_subsys_mask;
999 added_mask = final_subsys_mask & ~root->actual_subsys_mask;
1000 /* Check that any added subsystems are currently free */
1001 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1002 unsigned long bit = 1UL << i;
1003 struct cgroup_subsys *ss = subsys[i];
1004 if (!(bit & added_mask))
1007 * Nobody should tell us to do a subsys that doesn't exist:
1008 * parse_cgroupfs_options should catch that case and refcounts
1009 * ensure that subsystems won't disappear once selected.
1012 if (ss->root != &rootnode) {
1013 /* Subsystem isn't free */
1018 /* Currently we don't handle adding/removing subsystems when
1019 * any child cgroups exist. This is theoretically supportable
1020 * but involves complex error handling, so it's being left until
1022 if (root->number_of_cgroups > 1)
1025 /* Process each subsystem */
1026 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1027 struct cgroup_subsys *ss = subsys[i];
1028 unsigned long bit = 1UL << i;
1029 if (bit & added_mask) {
1030 /* We're binding this subsystem to this hierarchy */
1032 BUG_ON(cgrp->subsys[i]);
1033 BUG_ON(!dummytop->subsys[i]);
1034 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
1035 cgrp->subsys[i] = dummytop->subsys[i];
1036 cgrp->subsys[i]->cgroup = cgrp;
1037 list_move(&ss->sibling, &root->subsys_list);
1041 /* refcount was already taken, and we're keeping it */
1042 } else if (bit & removed_mask) {
1043 /* We're removing this subsystem */
1045 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1046 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1049 dummytop->subsys[i]->cgroup = dummytop;
1050 cgrp->subsys[i] = NULL;
1051 subsys[i]->root = &rootnode;
1052 list_move(&ss->sibling, &rootnode.subsys_list);
1053 /* subsystem is now free - drop reference on module */
1054 module_put(ss->module);
1055 } else if (bit & final_subsys_mask) {
1056 /* Subsystem state should already exist */
1058 BUG_ON(!cgrp->subsys[i]);
1060 * a refcount was taken, but we already had one, so
1061 * drop the extra reference.
1063 module_put(ss->module);
1064 #ifdef CONFIG_MODULE_UNLOAD
1065 BUG_ON(ss->module && !module_refcount(ss->module));
1068 /* Subsystem state shouldn't exist */
1069 BUG_ON(cgrp->subsys[i]);
1072 root->subsys_mask = root->actual_subsys_mask = final_subsys_mask;
1077 static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1079 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1080 struct cgroup_subsys *ss;
1082 mutex_lock(&cgroup_root_mutex);
1083 for_each_subsys(root, ss)
1084 seq_printf(seq, ",%s", ss->name);
1085 if (root->flags & CGRP_ROOT_SANE_BEHAVIOR)
1086 seq_puts(seq, ",sane_behavior");
1087 if (root->flags & CGRP_ROOT_NOPREFIX)
1088 seq_puts(seq, ",noprefix");
1089 if (root->flags & CGRP_ROOT_XATTR)
1090 seq_puts(seq, ",xattr");
1091 if (strlen(root->release_agent_path))
1092 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1093 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags))
1094 seq_puts(seq, ",clone_children");
1095 if (strlen(root->name))
1096 seq_printf(seq, ",name=%s", root->name);
1097 mutex_unlock(&cgroup_root_mutex);
1101 struct cgroup_sb_opts {
1102 unsigned long subsys_mask;
1103 unsigned long flags;
1104 char *release_agent;
1105 bool cpuset_clone_children;
1107 /* User explicitly requested empty subsystem */
1110 struct cgroupfs_root *new_root;
1115 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1116 * with cgroup_mutex held to protect the subsys[] array. This function takes
1117 * refcounts on subsystems to be used, unless it returns error, in which case
1118 * no refcounts are taken.
1120 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1122 char *token, *o = data;
1123 bool all_ss = false, one_ss = false;
1124 unsigned long mask = (unsigned long)-1;
1126 bool module_pin_failed = false;
1128 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1130 #ifdef CONFIG_CPUSETS
1131 mask = ~(1UL << cpuset_subsys_id);
1134 memset(opts, 0, sizeof(*opts));
1136 while ((token = strsep(&o, ",")) != NULL) {
1139 if (!strcmp(token, "none")) {
1140 /* Explicitly have no subsystems */
1144 if (!strcmp(token, "all")) {
1145 /* Mutually exclusive option 'all' + subsystem name */
1151 if (!strcmp(token, "__DEVEL__sane_behavior")) {
1152 opts->flags |= CGRP_ROOT_SANE_BEHAVIOR;
1155 if (!strcmp(token, "noprefix")) {
1156 opts->flags |= CGRP_ROOT_NOPREFIX;
1159 if (!strcmp(token, "clone_children")) {
1160 opts->cpuset_clone_children = true;
1163 if (!strcmp(token, "xattr")) {
1164 opts->flags |= CGRP_ROOT_XATTR;
1167 if (!strncmp(token, "release_agent=", 14)) {
1168 /* Specifying two release agents is forbidden */
1169 if (opts->release_agent)
1171 opts->release_agent =
1172 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1173 if (!opts->release_agent)
1177 if (!strncmp(token, "name=", 5)) {
1178 const char *name = token + 5;
1179 /* Can't specify an empty name */
1182 /* Must match [\w.-]+ */
1183 for (i = 0; i < strlen(name); i++) {
1187 if ((c == '.') || (c == '-') || (c == '_'))
1191 /* Specifying two names is forbidden */
1194 opts->name = kstrndup(name,
1195 MAX_CGROUP_ROOT_NAMELEN - 1,
1203 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1204 struct cgroup_subsys *ss = subsys[i];
1207 if (strcmp(token, ss->name))
1212 /* Mutually exclusive option 'all' + subsystem name */
1215 set_bit(i, &opts->subsys_mask);
1220 if (i == CGROUP_SUBSYS_COUNT)
1225 * If the 'all' option was specified select all the subsystems,
1226 * otherwise if 'none', 'name=' and a subsystem name options
1227 * were not specified, let's default to 'all'
1229 if (all_ss || (!one_ss && !opts->none && !opts->name)) {
1230 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1231 struct cgroup_subsys *ss = subsys[i];
1236 set_bit(i, &opts->subsys_mask);
1240 /* Consistency checks */
1242 if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) {
1243 pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1245 if (opts->flags & CGRP_ROOT_NOPREFIX) {
1246 pr_err("cgroup: sane_behavior: noprefix is not allowed\n");
1250 if (opts->cpuset_clone_children) {
1251 pr_err("cgroup: sane_behavior: clone_children is not allowed\n");
1257 * Option noprefix was introduced just for backward compatibility
1258 * with the old cpuset, so we allow noprefix only if mounting just
1259 * the cpuset subsystem.
1261 if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask))
1265 /* Can't specify "none" and some subsystems */
1266 if (opts->subsys_mask && opts->none)
1270 * We either have to specify by name or by subsystems. (So all
1271 * empty hierarchies must have a name).
1273 if (!opts->subsys_mask && !opts->name)
1277 * Grab references on all the modules we'll need, so the subsystems
1278 * don't dance around before rebind_subsystems attaches them. This may
1279 * take duplicate reference counts on a subsystem that's already used,
1280 * but rebind_subsystems handles this case.
1282 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1283 unsigned long bit = 1UL << i;
1285 if (!(bit & opts->subsys_mask))
1287 if (!try_module_get(subsys[i]->module)) {
1288 module_pin_failed = true;
1292 if (module_pin_failed) {
1294 * oops, one of the modules was going away. this means that we
1295 * raced with a module_delete call, and to the user this is
1296 * essentially a "subsystem doesn't exist" case.
1298 for (i--; i >= 0; i--) {
1299 /* drop refcounts only on the ones we took */
1300 unsigned long bit = 1UL << i;
1302 if (!(bit & opts->subsys_mask))
1304 module_put(subsys[i]->module);
1312 static void drop_parsed_module_refcounts(unsigned long subsys_mask)
1315 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1316 unsigned long bit = 1UL << i;
1318 if (!(bit & subsys_mask))
1320 module_put(subsys[i]->module);
1324 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1327 struct cgroupfs_root *root = sb->s_fs_info;
1328 struct cgroup *cgrp = &root->top_cgroup;
1329 struct cgroup_sb_opts opts;
1330 unsigned long added_mask, removed_mask;
1332 if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) {
1333 pr_err("cgroup: sane_behavior: remount is not allowed\n");
1337 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1338 mutex_lock(&cgroup_mutex);
1339 mutex_lock(&cgroup_root_mutex);
1341 /* See what subsystems are wanted */
1342 ret = parse_cgroupfs_options(data, &opts);
1346 if (opts.subsys_mask != root->actual_subsys_mask || opts.release_agent)
1347 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1348 task_tgid_nr(current), current->comm);
1350 added_mask = opts.subsys_mask & ~root->subsys_mask;
1351 removed_mask = root->subsys_mask & ~opts.subsys_mask;
1353 /* Don't allow flags or name to change at remount */
1354 if (opts.flags != root->flags ||
1355 (opts.name && strcmp(opts.name, root->name))) {
1357 drop_parsed_module_refcounts(opts.subsys_mask);
1362 * Clear out the files of subsystems that should be removed, do
1363 * this before rebind_subsystems, since rebind_subsystems may
1364 * change this hierarchy's subsys_list.
1366 cgroup_clear_directory(cgrp->dentry, false, removed_mask);
1368 ret = rebind_subsystems(root, opts.subsys_mask);
1370 /* rebind_subsystems failed, re-populate the removed files */
1371 cgroup_populate_dir(cgrp, false, removed_mask);
1372 drop_parsed_module_refcounts(opts.subsys_mask);
1376 /* re-populate subsystem files */
1377 cgroup_populate_dir(cgrp, false, added_mask);
1379 if (opts.release_agent)
1380 strcpy(root->release_agent_path, opts.release_agent);
1382 kfree(opts.release_agent);
1384 mutex_unlock(&cgroup_root_mutex);
1385 mutex_unlock(&cgroup_mutex);
1386 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1390 static const struct super_operations cgroup_ops = {
1391 .statfs = simple_statfs,
1392 .drop_inode = generic_delete_inode,
1393 .show_options = cgroup_show_options,
1394 .remount_fs = cgroup_remount,
1397 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1399 INIT_LIST_HEAD(&cgrp->sibling);
1400 INIT_LIST_HEAD(&cgrp->children);
1401 INIT_LIST_HEAD(&cgrp->files);
1402 INIT_LIST_HEAD(&cgrp->css_sets);
1403 INIT_LIST_HEAD(&cgrp->allcg_node);
1404 INIT_LIST_HEAD(&cgrp->release_list);
1405 INIT_LIST_HEAD(&cgrp->pidlists);
1406 INIT_WORK(&cgrp->free_work, cgroup_free_fn);
1407 mutex_init(&cgrp->pidlist_mutex);
1408 INIT_LIST_HEAD(&cgrp->event_list);
1409 spin_lock_init(&cgrp->event_list_lock);
1410 simple_xattrs_init(&cgrp->xattrs);
1413 static void init_cgroup_root(struct cgroupfs_root *root)
1415 struct cgroup *cgrp = &root->top_cgroup;
1417 INIT_LIST_HEAD(&root->subsys_list);
1418 INIT_LIST_HEAD(&root->root_list);
1419 INIT_LIST_HEAD(&root->allcg_list);
1420 root->number_of_cgroups = 1;
1422 cgrp->name = &root_cgroup_name;
1423 init_cgroup_housekeeping(cgrp);
1424 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
1427 static bool init_root_id(struct cgroupfs_root *root)
1432 if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1434 spin_lock(&hierarchy_id_lock);
1435 /* Try to allocate the next unused ID */
1436 ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1437 &root->hierarchy_id);
1439 /* Try again starting from 0 */
1440 ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1442 next_hierarchy_id = root->hierarchy_id + 1;
1443 } else if (ret != -EAGAIN) {
1444 /* Can only get here if the 31-bit IDR is full ... */
1447 spin_unlock(&hierarchy_id_lock);
1452 static int cgroup_test_super(struct super_block *sb, void *data)
1454 struct cgroup_sb_opts *opts = data;
1455 struct cgroupfs_root *root = sb->s_fs_info;
1457 /* If we asked for a name then it must match */
1458 if (opts->name && strcmp(opts->name, root->name))
1462 * If we asked for subsystems (or explicitly for no
1463 * subsystems) then they must match
1465 if ((opts->subsys_mask || opts->none)
1466 && (opts->subsys_mask != root->subsys_mask))
1472 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1474 struct cgroupfs_root *root;
1476 if (!opts->subsys_mask && !opts->none)
1479 root = kzalloc(sizeof(*root), GFP_KERNEL);
1481 return ERR_PTR(-ENOMEM);
1483 if (!init_root_id(root)) {
1485 return ERR_PTR(-ENOMEM);
1487 init_cgroup_root(root);
1489 root->subsys_mask = opts->subsys_mask;
1490 root->flags = opts->flags;
1491 ida_init(&root->cgroup_ida);
1492 if (opts->release_agent)
1493 strcpy(root->release_agent_path, opts->release_agent);
1495 strcpy(root->name, opts->name);
1496 if (opts->cpuset_clone_children)
1497 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags);
1501 static void cgroup_drop_root(struct cgroupfs_root *root)
1506 BUG_ON(!root->hierarchy_id);
1507 spin_lock(&hierarchy_id_lock);
1508 ida_remove(&hierarchy_ida, root->hierarchy_id);
1509 spin_unlock(&hierarchy_id_lock);
1510 ida_destroy(&root->cgroup_ida);
1514 static int cgroup_set_super(struct super_block *sb, void *data)
1517 struct cgroup_sb_opts *opts = data;
1519 /* If we don't have a new root, we can't set up a new sb */
1520 if (!opts->new_root)
1523 BUG_ON(!opts->subsys_mask && !opts->none);
1525 ret = set_anon_super(sb, NULL);
1529 sb->s_fs_info = opts->new_root;
1530 opts->new_root->sb = sb;
1532 sb->s_blocksize = PAGE_CACHE_SIZE;
1533 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1534 sb->s_magic = CGROUP_SUPER_MAGIC;
1535 sb->s_op = &cgroup_ops;
1540 static int cgroup_get_rootdir(struct super_block *sb)
1542 static const struct dentry_operations cgroup_dops = {
1543 .d_iput = cgroup_diput,
1544 .d_delete = cgroup_delete,
1547 struct inode *inode =
1548 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1553 inode->i_fop = &simple_dir_operations;
1554 inode->i_op = &cgroup_dir_inode_operations;
1555 /* directories start off with i_nlink == 2 (for "." entry) */
1557 sb->s_root = d_make_root(inode);
1560 /* for everything else we want ->d_op set */
1561 sb->s_d_op = &cgroup_dops;
1565 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1566 int flags, const char *unused_dev_name,
1569 struct cgroup_sb_opts opts;
1570 struct cgroupfs_root *root;
1572 struct super_block *sb;
1573 struct cgroupfs_root *new_root;
1574 struct inode *inode;
1576 /* First find the desired set of subsystems */
1577 mutex_lock(&cgroup_mutex);
1578 ret = parse_cgroupfs_options(data, &opts);
1579 mutex_unlock(&cgroup_mutex);
1584 * Allocate a new cgroup root. We may not need it if we're
1585 * reusing an existing hierarchy.
1587 new_root = cgroup_root_from_opts(&opts);
1588 if (IS_ERR(new_root)) {
1589 ret = PTR_ERR(new_root);
1592 opts.new_root = new_root;
1594 /* Locate an existing or new sb for this hierarchy */
1595 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, 0, &opts);
1598 cgroup_drop_root(opts.new_root);
1602 root = sb->s_fs_info;
1604 if (root == opts.new_root) {
1605 /* We used the new root structure, so this is a new hierarchy */
1606 struct list_head tmp_cg_links;
1607 struct cgroup *root_cgrp = &root->top_cgroup;
1608 struct cgroupfs_root *existing_root;
1609 const struct cred *cred;
1613 BUG_ON(sb->s_root != NULL);
1615 ret = cgroup_get_rootdir(sb);
1617 goto drop_new_super;
1618 inode = sb->s_root->d_inode;
1620 mutex_lock(&inode->i_mutex);
1621 mutex_lock(&cgroup_mutex);
1622 mutex_lock(&cgroup_root_mutex);
1624 /* Check for name clashes with existing mounts */
1626 if (strlen(root->name))
1627 for_each_active_root(existing_root)
1628 if (!strcmp(existing_root->name, root->name))
1632 * We're accessing css_set_count without locking
1633 * css_set_lock here, but that's OK - it can only be
1634 * increased by someone holding cgroup_lock, and
1635 * that's us. The worst that can happen is that we
1636 * have some link structures left over
1638 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1642 ret = rebind_subsystems(root, root->subsys_mask);
1643 if (ret == -EBUSY) {
1644 free_cg_links(&tmp_cg_links);
1648 * There must be no failure case after here, since rebinding
1649 * takes care of subsystems' refcounts, which are explicitly
1650 * dropped in the failure exit path.
1653 /* EBUSY should be the only error here */
1656 list_add(&root->root_list, &roots);
1659 sb->s_root->d_fsdata = root_cgrp;
1660 root->top_cgroup.dentry = sb->s_root;
1662 /* Link the top cgroup in this hierarchy into all
1663 * the css_set objects */
1664 write_lock(&css_set_lock);
1665 hash_for_each(css_set_table, i, cg, hlist)
1666 link_css_set(&tmp_cg_links, cg, root_cgrp);
1667 write_unlock(&css_set_lock);
1669 free_cg_links(&tmp_cg_links);
1671 BUG_ON(!list_empty(&root_cgrp->children));
1672 BUG_ON(root->number_of_cgroups != 1);
1674 cred = override_creds(&init_cred);
1675 cgroup_populate_dir(root_cgrp, true, root->subsys_mask);
1677 mutex_unlock(&cgroup_root_mutex);
1678 mutex_unlock(&cgroup_mutex);
1679 mutex_unlock(&inode->i_mutex);
1682 * We re-used an existing hierarchy - the new root (if
1683 * any) is not needed
1685 cgroup_drop_root(opts.new_root);
1687 if (((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) &&
1688 root->flags != opts.flags) {
1689 pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n");
1691 goto drop_new_super;
1694 /* no subsys rebinding, so refcounts don't change */
1695 drop_parsed_module_refcounts(opts.subsys_mask);
1698 kfree(opts.release_agent);
1700 return dget(sb->s_root);
1703 mutex_unlock(&cgroup_root_mutex);
1704 mutex_unlock(&cgroup_mutex);
1705 mutex_unlock(&inode->i_mutex);
1707 deactivate_locked_super(sb);
1709 drop_parsed_module_refcounts(opts.subsys_mask);
1711 kfree(opts.release_agent);
1713 return ERR_PTR(ret);
1716 static void cgroup_kill_sb(struct super_block *sb) {
1717 struct cgroupfs_root *root = sb->s_fs_info;
1718 struct cgroup *cgrp = &root->top_cgroup;
1720 struct cg_cgroup_link *link;
1721 struct cg_cgroup_link *saved_link;
1725 BUG_ON(root->number_of_cgroups != 1);
1726 BUG_ON(!list_empty(&cgrp->children));
1728 mutex_lock(&cgroup_mutex);
1729 mutex_lock(&cgroup_root_mutex);
1731 /* Rebind all subsystems back to the default hierarchy */
1732 ret = rebind_subsystems(root, 0);
1733 /* Shouldn't be able to fail ... */
1737 * Release all the links from css_sets to this hierarchy's
1740 write_lock(&css_set_lock);
1742 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1744 list_del(&link->cg_link_list);
1745 list_del(&link->cgrp_link_list);
1748 write_unlock(&css_set_lock);
1750 if (!list_empty(&root->root_list)) {
1751 list_del(&root->root_list);
1755 mutex_unlock(&cgroup_root_mutex);
1756 mutex_unlock(&cgroup_mutex);
1758 simple_xattrs_free(&cgrp->xattrs);
1760 kill_litter_super(sb);
1761 cgroup_drop_root(root);
1764 static struct file_system_type cgroup_fs_type = {
1766 .mount = cgroup_mount,
1767 .kill_sb = cgroup_kill_sb,
1770 static struct kobject *cgroup_kobj;
1773 * cgroup_path - generate the path of a cgroup
1774 * @cgrp: the cgroup in question
1775 * @buf: the buffer to write the path into
1776 * @buflen: the length of the buffer
1778 * Writes path of cgroup into buf. Returns 0 on success, -errno on error.
1780 * We can't generate cgroup path using dentry->d_name, as accessing
1781 * dentry->name must be protected by irq-unsafe dentry->d_lock or parent
1782 * inode's i_mutex, while on the other hand cgroup_path() can be called
1783 * with some irq-safe spinlocks held.
1785 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1787 int ret = -ENAMETOOLONG;
1790 if (!cgrp->parent) {
1791 if (strlcpy(buf, "/", buflen) >= buflen)
1792 return -ENAMETOOLONG;
1796 start = buf + buflen - 1;
1801 const char *name = cgroup_name(cgrp);
1805 if ((start -= len) < buf)
1807 memcpy(start, name, len);
1813 cgrp = cgrp->parent;
1814 } while (cgrp->parent);
1816 memmove(buf, start, buf + buflen - start);
1821 EXPORT_SYMBOL_GPL(cgroup_path);
1824 * Control Group taskset
1826 struct task_and_cgroup {
1827 struct task_struct *task;
1828 struct cgroup *cgrp;
1832 struct cgroup_taskset {
1833 struct task_and_cgroup single;
1834 struct flex_array *tc_array;
1837 struct cgroup *cur_cgrp;
1841 * cgroup_taskset_first - reset taskset and return the first task
1842 * @tset: taskset of interest
1844 * @tset iteration is initialized and the first task is returned.
1846 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1848 if (tset->tc_array) {
1850 return cgroup_taskset_next(tset);
1852 tset->cur_cgrp = tset->single.cgrp;
1853 return tset->single.task;
1856 EXPORT_SYMBOL_GPL(cgroup_taskset_first);
1859 * cgroup_taskset_next - iterate to the next task in taskset
1860 * @tset: taskset of interest
1862 * Return the next task in @tset. Iteration must have been initialized
1863 * with cgroup_taskset_first().
1865 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1867 struct task_and_cgroup *tc;
1869 if (!tset->tc_array || tset->idx >= tset->tc_array_len)
1872 tc = flex_array_get(tset->tc_array, tset->idx++);
1873 tset->cur_cgrp = tc->cgrp;
1876 EXPORT_SYMBOL_GPL(cgroup_taskset_next);
1879 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1880 * @tset: taskset of interest
1882 * Return the cgroup for the current (last returned) task of @tset. This
1883 * function must be preceded by either cgroup_taskset_first() or
1884 * cgroup_taskset_next().
1886 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset)
1888 return tset->cur_cgrp;
1890 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup);
1893 * cgroup_taskset_size - return the number of tasks in taskset
1894 * @tset: taskset of interest
1896 int cgroup_taskset_size(struct cgroup_taskset *tset)
1898 return tset->tc_array ? tset->tc_array_len : 1;
1900 EXPORT_SYMBOL_GPL(cgroup_taskset_size);
1904 * cgroup_task_migrate - move a task from one cgroup to another.
1906 * Must be called with cgroup_mutex and threadgroup locked.
1908 static void cgroup_task_migrate(struct cgroup *oldcgrp,
1909 struct task_struct *tsk, struct css_set *newcg)
1911 struct css_set *oldcg;
1914 * We are synchronized through threadgroup_lock() against PF_EXITING
1915 * setting such that we can't race against cgroup_exit() changing the
1916 * css_set to init_css_set and dropping the old one.
1918 WARN_ON_ONCE(tsk->flags & PF_EXITING);
1919 oldcg = tsk->cgroups;
1922 rcu_assign_pointer(tsk->cgroups, newcg);
1925 /* Update the css_set linked lists if we're using them */
1926 write_lock(&css_set_lock);
1927 if (!list_empty(&tsk->cg_list))
1928 list_move(&tsk->cg_list, &newcg->tasks);
1929 write_unlock(&css_set_lock);
1932 * We just gained a reference on oldcg by taking it from the task. As
1933 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1934 * it here; it will be freed under RCU.
1936 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1941 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
1942 * @cgrp: the cgroup to attach to
1943 * @tsk: the task or the leader of the threadgroup to be attached
1944 * @threadgroup: attach the whole threadgroup?
1946 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
1947 * task_lock of @tsk or each thread in the threadgroup individually in turn.
1949 static int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk,
1952 int retval, i, group_size;
1953 struct cgroup_subsys *ss, *failed_ss = NULL;
1954 struct cgroupfs_root *root = cgrp->root;
1955 /* threadgroup list cursor and array */
1956 struct task_struct *leader = tsk;
1957 struct task_and_cgroup *tc;
1958 struct flex_array *group;
1959 struct cgroup_taskset tset = { };
1962 * step 0: in order to do expensive, possibly blocking operations for
1963 * every thread, we cannot iterate the thread group list, since it needs
1964 * rcu or tasklist locked. instead, build an array of all threads in the
1965 * group - group_rwsem prevents new threads from appearing, and if
1966 * threads exit, this will just be an over-estimate.
1969 group_size = get_nr_threads(tsk);
1972 /* flex_array supports very large thread-groups better than kmalloc. */
1973 group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
1976 /* pre-allocate to guarantee space while iterating in rcu read-side. */
1977 retval = flex_array_prealloc(group, 0, group_size, GFP_KERNEL);
1979 goto out_free_group_list;
1983 * Prevent freeing of tasks while we take a snapshot. Tasks that are
1984 * already PF_EXITING could be freed from underneath us unless we
1985 * take an rcu_read_lock.
1989 struct task_and_cgroup ent;
1991 /* @tsk either already exited or can't exit until the end */
1992 if (tsk->flags & PF_EXITING)
1995 /* as per above, nr_threads may decrease, but not increase. */
1996 BUG_ON(i >= group_size);
1998 ent.cgrp = task_cgroup_from_root(tsk, root);
1999 /* nothing to do if this task is already in the cgroup */
2000 if (ent.cgrp == cgrp)
2003 * saying GFP_ATOMIC has no effect here because we did prealloc
2004 * earlier, but it's good form to communicate our expectations.
2006 retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
2007 BUG_ON(retval != 0);
2012 } while_each_thread(leader, tsk);
2014 /* remember the number of threads in the array for later. */
2016 tset.tc_array = group;
2017 tset.tc_array_len = group_size;
2019 /* methods shouldn't be called if no task is actually migrating */
2022 goto out_free_group_list;
2025 * step 1: check that we can legitimately attach to the cgroup.
2027 for_each_subsys(root, ss) {
2028 if (ss->can_attach) {
2029 retval = ss->can_attach(cgrp, &tset);
2032 goto out_cancel_attach;
2038 * step 2: make sure css_sets exist for all threads to be migrated.
2039 * we use find_css_set, which allocates a new one if necessary.
2041 for (i = 0; i < group_size; i++) {
2042 tc = flex_array_get(group, i);
2043 tc->cg = find_css_set(tc->task->cgroups, cgrp);
2046 goto out_put_css_set_refs;
2051 * step 3: now that we're guaranteed success wrt the css_sets,
2052 * proceed to move all tasks to the new cgroup. There are no
2053 * failure cases after here, so this is the commit point.
2055 for (i = 0; i < group_size; i++) {
2056 tc = flex_array_get(group, i);
2057 cgroup_task_migrate(tc->cgrp, tc->task, tc->cg);
2059 /* nothing is sensitive to fork() after this point. */
2062 * step 4: do subsystem attach callbacks.
2064 for_each_subsys(root, ss) {
2066 ss->attach(cgrp, &tset);
2070 * step 5: success! and cleanup
2073 out_put_css_set_refs:
2075 for (i = 0; i < group_size; i++) {
2076 tc = flex_array_get(group, i);
2079 put_css_set(tc->cg);
2084 for_each_subsys(root, ss) {
2085 if (ss == failed_ss)
2087 if (ss->cancel_attach)
2088 ss->cancel_attach(cgrp, &tset);
2091 out_free_group_list:
2092 flex_array_free(group);
2097 * Find the task_struct of the task to attach by vpid and pass it along to the
2098 * function to attach either it or all tasks in its threadgroup. Will lock
2099 * cgroup_mutex and threadgroup; may take task_lock of task.
2101 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2103 struct task_struct *tsk;
2104 const struct cred *cred = current_cred(), *tcred;
2107 if (!cgroup_lock_live_group(cgrp))
2113 tsk = find_task_by_vpid(pid);
2117 goto out_unlock_cgroup;
2120 * even if we're attaching all tasks in the thread group, we
2121 * only need to check permissions on one of them.
2123 tcred = __task_cred(tsk);
2124 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2125 !uid_eq(cred->euid, tcred->uid) &&
2126 !uid_eq(cred->euid, tcred->suid)) {
2129 goto out_unlock_cgroup;
2135 tsk = tsk->group_leader;
2138 * Workqueue threads may acquire PF_THREAD_BOUND and become
2139 * trapped in a cpuset, or RT worker may be born in a cgroup
2140 * with no rt_runtime allocated. Just say no.
2142 if (tsk == kthreadd_task || (tsk->flags & PF_THREAD_BOUND)) {
2145 goto out_unlock_cgroup;
2148 get_task_struct(tsk);
2151 threadgroup_lock(tsk);
2153 if (!thread_group_leader(tsk)) {
2155 * a race with de_thread from another thread's exec()
2156 * may strip us of our leadership, if this happens,
2157 * there is no choice but to throw this task away and
2158 * try again; this is
2159 * "double-double-toil-and-trouble-check locking".
2161 threadgroup_unlock(tsk);
2162 put_task_struct(tsk);
2163 goto retry_find_task;
2167 ret = cgroup_attach_task(cgrp, tsk, threadgroup);
2169 threadgroup_unlock(tsk);
2171 put_task_struct(tsk);
2173 mutex_unlock(&cgroup_mutex);
2178 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2179 * @from: attach to all cgroups of a given task
2180 * @tsk: the task to be attached
2182 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2184 struct cgroupfs_root *root;
2187 mutex_lock(&cgroup_mutex);
2188 for_each_active_root(root) {
2189 struct cgroup *from_cg = task_cgroup_from_root(from, root);
2191 retval = cgroup_attach_task(from_cg, tsk, false);
2195 mutex_unlock(&cgroup_mutex);
2199 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2201 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2203 return attach_task_by_pid(cgrp, pid, false);
2206 static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2208 return attach_task_by_pid(cgrp, tgid, true);
2211 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2214 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2215 if (strlen(buffer) >= PATH_MAX)
2217 if (!cgroup_lock_live_group(cgrp))
2219 mutex_lock(&cgroup_root_mutex);
2220 strcpy(cgrp->root->release_agent_path, buffer);
2221 mutex_unlock(&cgroup_root_mutex);
2222 mutex_unlock(&cgroup_mutex);
2226 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2227 struct seq_file *seq)
2229 if (!cgroup_lock_live_group(cgrp))
2231 seq_puts(seq, cgrp->root->release_agent_path);
2232 seq_putc(seq, '\n');
2233 mutex_unlock(&cgroup_mutex);
2237 static int cgroup_sane_behavior_show(struct cgroup *cgrp, struct cftype *cft,
2238 struct seq_file *seq)
2240 seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp));
2244 /* A buffer size big enough for numbers or short strings */
2245 #define CGROUP_LOCAL_BUFFER_SIZE 64
2247 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2249 const char __user *userbuf,
2250 size_t nbytes, loff_t *unused_ppos)
2252 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2258 if (nbytes >= sizeof(buffer))
2260 if (copy_from_user(buffer, userbuf, nbytes))
2263 buffer[nbytes] = 0; /* nul-terminate */
2264 if (cft->write_u64) {
2265 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2268 retval = cft->write_u64(cgrp, cft, val);
2270 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2273 retval = cft->write_s64(cgrp, cft, val);
2280 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2282 const char __user *userbuf,
2283 size_t nbytes, loff_t *unused_ppos)
2285 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2287 size_t max_bytes = cft->max_write_len;
2288 char *buffer = local_buffer;
2291 max_bytes = sizeof(local_buffer) - 1;
2292 if (nbytes >= max_bytes)
2294 /* Allocate a dynamic buffer if we need one */
2295 if (nbytes >= sizeof(local_buffer)) {
2296 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2300 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2305 buffer[nbytes] = 0; /* nul-terminate */
2306 retval = cft->write_string(cgrp, cft, strstrip(buffer));
2310 if (buffer != local_buffer)
2315 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2316 size_t nbytes, loff_t *ppos)
2318 struct cftype *cft = __d_cft(file->f_dentry);
2319 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2321 if (cgroup_is_removed(cgrp))
2324 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2325 if (cft->write_u64 || cft->write_s64)
2326 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2327 if (cft->write_string)
2328 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2330 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2331 return ret ? ret : nbytes;
2336 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2338 char __user *buf, size_t nbytes,
2341 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2342 u64 val = cft->read_u64(cgrp, cft);
2343 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2345 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2348 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2350 char __user *buf, size_t nbytes,
2353 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2354 s64 val = cft->read_s64(cgrp, cft);
2355 int len = sprintf(tmp, "%lld\n", (long long) val);
2357 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2360 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2361 size_t nbytes, loff_t *ppos)
2363 struct cftype *cft = __d_cft(file->f_dentry);
2364 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2366 if (cgroup_is_removed(cgrp))
2370 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2372 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2374 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2379 * seqfile ops/methods for returning structured data. Currently just
2380 * supports string->u64 maps, but can be extended in future.
2383 struct cgroup_seqfile_state {
2385 struct cgroup *cgroup;
2388 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2390 struct seq_file *sf = cb->state;
2391 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2394 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2396 struct cgroup_seqfile_state *state = m->private;
2397 struct cftype *cft = state->cft;
2398 if (cft->read_map) {
2399 struct cgroup_map_cb cb = {
2400 .fill = cgroup_map_add,
2403 return cft->read_map(state->cgroup, cft, &cb);
2405 return cft->read_seq_string(state->cgroup, cft, m);
2408 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2410 struct seq_file *seq = file->private_data;
2411 kfree(seq->private);
2412 return single_release(inode, file);
2415 static const struct file_operations cgroup_seqfile_operations = {
2417 .write = cgroup_file_write,
2418 .llseek = seq_lseek,
2419 .release = cgroup_seqfile_release,
2422 static int cgroup_file_open(struct inode *inode, struct file *file)
2427 err = generic_file_open(inode, file);
2430 cft = __d_cft(file->f_dentry);
2432 if (cft->read_map || cft->read_seq_string) {
2433 struct cgroup_seqfile_state *state =
2434 kzalloc(sizeof(*state), GFP_USER);
2438 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2439 file->f_op = &cgroup_seqfile_operations;
2440 err = single_open(file, cgroup_seqfile_show, state);
2443 } else if (cft->open)
2444 err = cft->open(inode, file);
2451 static int cgroup_file_release(struct inode *inode, struct file *file)
2453 struct cftype *cft = __d_cft(file->f_dentry);
2455 return cft->release(inode, file);
2460 * cgroup_rename - Only allow simple rename of directories in place.
2462 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2463 struct inode *new_dir, struct dentry *new_dentry)
2466 struct cgroup_name *name, *old_name;
2467 struct cgroup *cgrp;
2470 * It's convinient to use parent dir's i_mutex to protected
2473 lockdep_assert_held(&old_dir->i_mutex);
2475 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2477 if (new_dentry->d_inode)
2479 if (old_dir != new_dir)
2482 cgrp = __d_cgrp(old_dentry);
2484 name = cgroup_alloc_name(new_dentry);
2488 ret = simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2494 old_name = cgrp->name;
2495 rcu_assign_pointer(cgrp->name, name);
2497 kfree_rcu(old_name, rcu_head);
2501 static struct simple_xattrs *__d_xattrs(struct dentry *dentry)
2503 if (S_ISDIR(dentry->d_inode->i_mode))
2504 return &__d_cgrp(dentry)->xattrs;
2506 return &__d_cfe(dentry)->xattrs;
2509 static inline int xattr_enabled(struct dentry *dentry)
2511 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
2512 return root->flags & CGRP_ROOT_XATTR;
2515 static bool is_valid_xattr(const char *name)
2517 if (!strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
2518 !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
2523 static int cgroup_setxattr(struct dentry *dentry, const char *name,
2524 const void *val, size_t size, int flags)
2526 if (!xattr_enabled(dentry))
2528 if (!is_valid_xattr(name))
2530 return simple_xattr_set(__d_xattrs(dentry), name, val, size, flags);
2533 static int cgroup_removexattr(struct dentry *dentry, const char *name)
2535 if (!xattr_enabled(dentry))
2537 if (!is_valid_xattr(name))
2539 return simple_xattr_remove(__d_xattrs(dentry), name);
2542 static ssize_t cgroup_getxattr(struct dentry *dentry, const char *name,
2543 void *buf, size_t size)
2545 if (!xattr_enabled(dentry))
2547 if (!is_valid_xattr(name))
2549 return simple_xattr_get(__d_xattrs(dentry), name, buf, size);
2552 static ssize_t cgroup_listxattr(struct dentry *dentry, char *buf, size_t size)
2554 if (!xattr_enabled(dentry))
2556 return simple_xattr_list(__d_xattrs(dentry), buf, size);
2559 static const struct file_operations cgroup_file_operations = {
2560 .read = cgroup_file_read,
2561 .write = cgroup_file_write,
2562 .llseek = generic_file_llseek,
2563 .open = cgroup_file_open,
2564 .release = cgroup_file_release,
2567 static const struct inode_operations cgroup_file_inode_operations = {
2568 .setxattr = cgroup_setxattr,
2569 .getxattr = cgroup_getxattr,
2570 .listxattr = cgroup_listxattr,
2571 .removexattr = cgroup_removexattr,
2574 static const struct inode_operations cgroup_dir_inode_operations = {
2575 .lookup = cgroup_lookup,
2576 .mkdir = cgroup_mkdir,
2577 .rmdir = cgroup_rmdir,
2578 .rename = cgroup_rename,
2579 .setxattr = cgroup_setxattr,
2580 .getxattr = cgroup_getxattr,
2581 .listxattr = cgroup_listxattr,
2582 .removexattr = cgroup_removexattr,
2585 static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2587 if (dentry->d_name.len > NAME_MAX)
2588 return ERR_PTR(-ENAMETOOLONG);
2589 d_add(dentry, NULL);
2594 * Check if a file is a control file
2596 static inline struct cftype *__file_cft(struct file *file)
2598 if (file_inode(file)->i_fop != &cgroup_file_operations)
2599 return ERR_PTR(-EINVAL);
2600 return __d_cft(file->f_dentry);
2603 static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2604 struct super_block *sb)
2606 struct inode *inode;
2610 if (dentry->d_inode)
2613 inode = cgroup_new_inode(mode, sb);
2617 if (S_ISDIR(mode)) {
2618 inode->i_op = &cgroup_dir_inode_operations;
2619 inode->i_fop = &simple_dir_operations;
2621 /* start off with i_nlink == 2 (for "." entry) */
2623 inc_nlink(dentry->d_parent->d_inode);
2626 * Control reaches here with cgroup_mutex held.
2627 * @inode->i_mutex should nest outside cgroup_mutex but we
2628 * want to populate it immediately without releasing
2629 * cgroup_mutex. As @inode isn't visible to anyone else
2630 * yet, trylock will always succeed without affecting
2633 WARN_ON_ONCE(!mutex_trylock(&inode->i_mutex));
2634 } else if (S_ISREG(mode)) {
2636 inode->i_fop = &cgroup_file_operations;
2637 inode->i_op = &cgroup_file_inode_operations;
2639 d_instantiate(dentry, inode);
2640 dget(dentry); /* Extra count - pin the dentry in core */
2645 * cgroup_file_mode - deduce file mode of a control file
2646 * @cft: the control file in question
2648 * returns cft->mode if ->mode is not 0
2649 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2650 * returns S_IRUGO if it has only a read handler
2651 * returns S_IWUSR if it has only a write hander
2653 static umode_t cgroup_file_mode(const struct cftype *cft)
2660 if (cft->read || cft->read_u64 || cft->read_s64 ||
2661 cft->read_map || cft->read_seq_string)
2664 if (cft->write || cft->write_u64 || cft->write_s64 ||
2665 cft->write_string || cft->trigger)
2671 static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2674 struct dentry *dir = cgrp->dentry;
2675 struct cgroup *parent = __d_cgrp(dir);
2676 struct dentry *dentry;
2680 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2682 if (subsys && !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) {
2683 strcpy(name, subsys->name);
2686 strcat(name, cft->name);
2688 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2690 cfe = kzalloc(sizeof(*cfe), GFP_KERNEL);
2694 dentry = lookup_one_len(name, dir, strlen(name));
2695 if (IS_ERR(dentry)) {
2696 error = PTR_ERR(dentry);
2700 mode = cgroup_file_mode(cft);
2701 error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb);
2703 cfe->type = (void *)cft;
2704 cfe->dentry = dentry;
2705 dentry->d_fsdata = cfe;
2706 simple_xattrs_init(&cfe->xattrs);
2707 list_add_tail(&cfe->node, &parent->files);
2716 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2717 struct cftype cfts[], bool is_add)
2722 for (cft = cfts; cft->name[0] != '\0'; cft++) {
2723 /* does cft->flags tell us to skip this file on @cgrp? */
2724 if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp))
2726 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2728 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
2732 err = cgroup_add_file(cgrp, subsys, cft);
2734 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2738 cgroup_rm_file(cgrp, cft);
2744 static DEFINE_MUTEX(cgroup_cft_mutex);
2746 static void cgroup_cfts_prepare(void)
2747 __acquires(&cgroup_cft_mutex) __acquires(&cgroup_mutex)
2750 * Thanks to the entanglement with vfs inode locking, we can't walk
2751 * the existing cgroups under cgroup_mutex and create files.
2752 * Instead, we increment reference on all cgroups and build list of
2753 * them using @cgrp->cft_q_node. Grab cgroup_cft_mutex to ensure
2754 * exclusive access to the field.
2756 mutex_lock(&cgroup_cft_mutex);
2757 mutex_lock(&cgroup_mutex);
2760 static void cgroup_cfts_commit(struct cgroup_subsys *ss,
2761 struct cftype *cfts, bool is_add)
2762 __releases(&cgroup_mutex) __releases(&cgroup_cft_mutex)
2765 struct cgroup *cgrp, *n;
2767 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2768 if (cfts && ss->root != &rootnode) {
2769 list_for_each_entry(cgrp, &ss->root->allcg_list, allcg_node) {
2771 list_add_tail(&cgrp->cft_q_node, &pending);
2775 mutex_unlock(&cgroup_mutex);
2778 * All new cgroups will see @cfts update on @ss->cftsets. Add/rm
2779 * files for all cgroups which were created before.
2781 list_for_each_entry_safe(cgrp, n, &pending, cft_q_node) {
2782 struct inode *inode = cgrp->dentry->d_inode;
2784 mutex_lock(&inode->i_mutex);
2785 mutex_lock(&cgroup_mutex);
2786 if (!cgroup_is_removed(cgrp))
2787 cgroup_addrm_files(cgrp, ss, cfts, is_add);
2788 mutex_unlock(&cgroup_mutex);
2789 mutex_unlock(&inode->i_mutex);
2791 list_del_init(&cgrp->cft_q_node);
2795 mutex_unlock(&cgroup_cft_mutex);
2799 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2800 * @ss: target cgroup subsystem
2801 * @cfts: zero-length name terminated array of cftypes
2803 * Register @cfts to @ss. Files described by @cfts are created for all
2804 * existing cgroups to which @ss is attached and all future cgroups will
2805 * have them too. This function can be called anytime whether @ss is
2808 * Returns 0 on successful registration, -errno on failure. Note that this
2809 * function currently returns 0 as long as @cfts registration is successful
2810 * even if some file creation attempts on existing cgroups fail.
2812 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2814 struct cftype_set *set;
2816 set = kzalloc(sizeof(*set), GFP_KERNEL);
2820 cgroup_cfts_prepare();
2822 list_add_tail(&set->node, &ss->cftsets);
2823 cgroup_cfts_commit(ss, cfts, true);
2827 EXPORT_SYMBOL_GPL(cgroup_add_cftypes);
2830 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2831 * @ss: target cgroup subsystem
2832 * @cfts: zero-length name terminated array of cftypes
2834 * Unregister @cfts from @ss. Files described by @cfts are removed from
2835 * all existing cgroups to which @ss is attached and all future cgroups
2836 * won't have them either. This function can be called anytime whether @ss
2837 * is attached or not.
2839 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2840 * registered with @ss.
2842 int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2844 struct cftype_set *set;
2846 cgroup_cfts_prepare();
2848 list_for_each_entry(set, &ss->cftsets, node) {
2849 if (set->cfts == cfts) {
2850 list_del_init(&set->node);
2851 cgroup_cfts_commit(ss, cfts, false);
2856 cgroup_cfts_commit(ss, NULL, false);
2861 * cgroup_task_count - count the number of tasks in a cgroup.
2862 * @cgrp: the cgroup in question
2864 * Return the number of tasks in the cgroup.
2866 int cgroup_task_count(const struct cgroup *cgrp)
2869 struct cg_cgroup_link *link;
2871 read_lock(&css_set_lock);
2872 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2873 count += atomic_read(&link->cg->refcount);
2875 read_unlock(&css_set_lock);
2880 * Advance a list_head iterator. The iterator should be positioned at
2881 * the start of a css_set
2883 static void cgroup_advance_iter(struct cgroup *cgrp,
2884 struct cgroup_iter *it)
2886 struct list_head *l = it->cg_link;
2887 struct cg_cgroup_link *link;
2890 /* Advance to the next non-empty css_set */
2893 if (l == &cgrp->css_sets) {
2897 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2899 } while (list_empty(&cg->tasks));
2901 it->task = cg->tasks.next;
2905 * To reduce the fork() overhead for systems that are not actually
2906 * using their cgroups capability, we don't maintain the lists running
2907 * through each css_set to its tasks until we see the list actually
2908 * used - in other words after the first call to cgroup_iter_start().
2910 static void cgroup_enable_task_cg_lists(void)
2912 struct task_struct *p, *g;
2913 write_lock(&css_set_lock);
2914 use_task_css_set_links = 1;
2916 * We need tasklist_lock because RCU is not safe against
2917 * while_each_thread(). Besides, a forking task that has passed
2918 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2919 * is not guaranteed to have its child immediately visible in the
2920 * tasklist if we walk through it with RCU.
2922 read_lock(&tasklist_lock);
2923 do_each_thread(g, p) {
2926 * We should check if the process is exiting, otherwise
2927 * it will race with cgroup_exit() in that the list
2928 * entry won't be deleted though the process has exited.
2930 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2931 list_add(&p->cg_list, &p->cgroups->tasks);
2933 } while_each_thread(g, p);
2934 read_unlock(&tasklist_lock);
2935 write_unlock(&css_set_lock);
2939 * cgroup_next_descendant_pre - find the next descendant for pre-order walk
2940 * @pos: the current position (%NULL to initiate traversal)
2941 * @cgroup: cgroup whose descendants to walk
2943 * To be used by cgroup_for_each_descendant_pre(). Find the next
2944 * descendant to visit for pre-order traversal of @cgroup's descendants.
2946 struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
2947 struct cgroup *cgroup)
2949 struct cgroup *next;
2951 WARN_ON_ONCE(!rcu_read_lock_held());
2953 /* if first iteration, pretend we just visited @cgroup */
2955 if (list_empty(&cgroup->children))
2960 /* visit the first child if exists */
2961 next = list_first_or_null_rcu(&pos->children, struct cgroup, sibling);
2965 /* no child, visit my or the closest ancestor's next sibling */
2967 next = list_entry_rcu(pos->sibling.next, struct cgroup,
2969 if (&next->sibling != &pos->parent->children)
2973 } while (pos != cgroup);
2977 EXPORT_SYMBOL_GPL(cgroup_next_descendant_pre);
2980 * cgroup_rightmost_descendant - return the rightmost descendant of a cgroup
2981 * @pos: cgroup of interest
2983 * Return the rightmost descendant of @pos. If there's no descendant,
2984 * @pos is returned. This can be used during pre-order traversal to skip
2987 struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos)
2989 struct cgroup *last, *tmp;
2991 WARN_ON_ONCE(!rcu_read_lock_held());
2995 /* ->prev isn't RCU safe, walk ->next till the end */
2997 list_for_each_entry_rcu(tmp, &last->children, sibling)
3003 EXPORT_SYMBOL_GPL(cgroup_rightmost_descendant);
3005 static struct cgroup *cgroup_leftmost_descendant(struct cgroup *pos)
3007 struct cgroup *last;
3011 pos = list_first_or_null_rcu(&pos->children, struct cgroup,
3019 * cgroup_next_descendant_post - find the next descendant for post-order walk
3020 * @pos: the current position (%NULL to initiate traversal)
3021 * @cgroup: cgroup whose descendants to walk
3023 * To be used by cgroup_for_each_descendant_post(). Find the next
3024 * descendant to visit for post-order traversal of @cgroup's descendants.
3026 struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
3027 struct cgroup *cgroup)
3029 struct cgroup *next;
3031 WARN_ON_ONCE(!rcu_read_lock_held());
3033 /* if first iteration, visit the leftmost descendant */
3035 next = cgroup_leftmost_descendant(cgroup);
3036 return next != cgroup ? next : NULL;
3039 /* if there's an unvisited sibling, visit its leftmost descendant */
3040 next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling);
3041 if (&next->sibling != &pos->parent->children)
3042 return cgroup_leftmost_descendant(next);
3044 /* no sibling left, visit parent */
3046 return next != cgroup ? next : NULL;
3048 EXPORT_SYMBOL_GPL(cgroup_next_descendant_post);
3050 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
3051 __acquires(css_set_lock)
3054 * The first time anyone tries to iterate across a cgroup,
3055 * we need to enable the list linking each css_set to its
3056 * tasks, and fix up all existing tasks.
3058 if (!use_task_css_set_links)
3059 cgroup_enable_task_cg_lists();
3061 read_lock(&css_set_lock);
3062 it->cg_link = &cgrp->css_sets;
3063 cgroup_advance_iter(cgrp, it);
3066 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
3067 struct cgroup_iter *it)
3069 struct task_struct *res;
3070 struct list_head *l = it->task;
3071 struct cg_cgroup_link *link;
3073 /* If the iterator cg is NULL, we have no tasks */
3076 res = list_entry(l, struct task_struct, cg_list);
3077 /* Advance iterator to find next entry */
3079 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
3080 if (l == &link->cg->tasks) {
3081 /* We reached the end of this task list - move on to
3082 * the next cg_cgroup_link */
3083 cgroup_advance_iter(cgrp, it);
3090 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
3091 __releases(css_set_lock)
3093 read_unlock(&css_set_lock);
3096 static inline int started_after_time(struct task_struct *t1,
3097 struct timespec *time,
3098 struct task_struct *t2)
3100 int start_diff = timespec_compare(&t1->start_time, time);
3101 if (start_diff > 0) {
3103 } else if (start_diff < 0) {
3107 * Arbitrarily, if two processes started at the same
3108 * time, we'll say that the lower pointer value
3109 * started first. Note that t2 may have exited by now
3110 * so this may not be a valid pointer any longer, but
3111 * that's fine - it still serves to distinguish
3112 * between two tasks started (effectively) simultaneously.
3119 * This function is a callback from heap_insert() and is used to order
3121 * In this case we order the heap in descending task start time.
3123 static inline int started_after(void *p1, void *p2)
3125 struct task_struct *t1 = p1;
3126 struct task_struct *t2 = p2;
3127 return started_after_time(t1, &t2->start_time, t2);
3131 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3132 * @scan: struct cgroup_scanner containing arguments for the scan
3134 * Arguments include pointers to callback functions test_task() and
3136 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3137 * and if it returns true, call process_task() for it also.
3138 * The test_task pointer may be NULL, meaning always true (select all tasks).
3139 * Effectively duplicates cgroup_iter_{start,next,end}()
3140 * but does not lock css_set_lock for the call to process_task().
3141 * The struct cgroup_scanner may be embedded in any structure of the caller's
3143 * It is guaranteed that process_task() will act on every task that
3144 * is a member of the cgroup for the duration of this call. This
3145 * function may or may not call process_task() for tasks that exit
3146 * or move to a different cgroup during the call, or are forked or
3147 * move into the cgroup during the call.
3149 * Note that test_task() may be called with locks held, and may in some
3150 * situations be called multiple times for the same task, so it should
3152 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3153 * pre-allocated and will be used for heap operations (and its "gt" member will
3154 * be overwritten), else a temporary heap will be used (allocation of which
3155 * may cause this function to fail).
3157 int cgroup_scan_tasks(struct cgroup_scanner *scan)
3160 struct cgroup_iter it;
3161 struct task_struct *p, *dropped;
3162 /* Never dereference latest_task, since it's not refcounted */
3163 struct task_struct *latest_task = NULL;
3164 struct ptr_heap tmp_heap;
3165 struct ptr_heap *heap;
3166 struct timespec latest_time = { 0, 0 };
3169 /* The caller supplied our heap and pre-allocated its memory */
3171 heap->gt = &started_after;
3173 /* We need to allocate our own heap memory */
3175 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
3177 /* cannot allocate the heap */
3183 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3184 * to determine which are of interest, and using the scanner's
3185 * "process_task" callback to process any of them that need an update.
3186 * Since we don't want to hold any locks during the task updates,
3187 * gather tasks to be processed in a heap structure.
3188 * The heap is sorted by descending task start time.
3189 * If the statically-sized heap fills up, we overflow tasks that
3190 * started later, and in future iterations only consider tasks that
3191 * started after the latest task in the previous pass. This
3192 * guarantees forward progress and that we don't miss any tasks.
3195 cgroup_iter_start(scan->cg, &it);
3196 while ((p = cgroup_iter_next(scan->cg, &it))) {
3198 * Only affect tasks that qualify per the caller's callback,
3199 * if he provided one
3201 if (scan->test_task && !scan->test_task(p, scan))
3204 * Only process tasks that started after the last task
3207 if (!started_after_time(p, &latest_time, latest_task))
3209 dropped = heap_insert(heap, p);
3210 if (dropped == NULL) {
3212 * The new task was inserted; the heap wasn't
3216 } else if (dropped != p) {
3218 * The new task was inserted, and pushed out a
3222 put_task_struct(dropped);
3225 * Else the new task was newer than anything already in
3226 * the heap and wasn't inserted
3229 cgroup_iter_end(scan->cg, &it);
3232 for (i = 0; i < heap->size; i++) {
3233 struct task_struct *q = heap->ptrs[i];
3235 latest_time = q->start_time;
3238 /* Process the task per the caller's callback */
3239 scan->process_task(q, scan);
3243 * If we had to process any tasks at all, scan again
3244 * in case some of them were in the middle of forking
3245 * children that didn't get processed.
3246 * Not the most efficient way to do it, but it avoids
3247 * having to take callback_mutex in the fork path
3251 if (heap == &tmp_heap)
3252 heap_free(&tmp_heap);
3256 static void cgroup_transfer_one_task(struct task_struct *task,
3257 struct cgroup_scanner *scan)
3259 struct cgroup *new_cgroup = scan->data;
3261 mutex_lock(&cgroup_mutex);
3262 cgroup_attach_task(new_cgroup, task, false);
3263 mutex_unlock(&cgroup_mutex);
3267 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3268 * @to: cgroup to which the tasks will be moved
3269 * @from: cgroup in which the tasks currently reside
3271 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
3273 struct cgroup_scanner scan;
3276 scan.test_task = NULL; /* select all tasks in cgroup */
3277 scan.process_task = cgroup_transfer_one_task;
3281 return cgroup_scan_tasks(&scan);
3285 * Stuff for reading the 'tasks'/'procs' files.
3287 * Reading this file can return large amounts of data if a cgroup has
3288 * *lots* of attached tasks. So it may need several calls to read(),
3289 * but we cannot guarantee that the information we produce is correct
3290 * unless we produce it entirely atomically.
3294 /* which pidlist file are we talking about? */
3295 enum cgroup_filetype {
3301 * A pidlist is a list of pids that virtually represents the contents of one
3302 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3303 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3306 struct cgroup_pidlist {
3308 * used to find which pidlist is wanted. doesn't change as long as
3309 * this particular list stays in the list.
3311 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
3314 /* how many elements the above list has */
3316 /* how many files are using the current array */
3318 /* each of these stored in a list by its cgroup */
3319 struct list_head links;
3320 /* pointer to the cgroup we belong to, for list removal purposes */
3321 struct cgroup *owner;
3322 /* protects the other fields */
3323 struct rw_semaphore mutex;
3327 * The following two functions "fix" the issue where there are more pids
3328 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3329 * TODO: replace with a kernel-wide solution to this problem
3331 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3332 static void *pidlist_allocate(int count)
3334 if (PIDLIST_TOO_LARGE(count))
3335 return vmalloc(count * sizeof(pid_t));
3337 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3339 static void pidlist_free(void *p)
3341 if (is_vmalloc_addr(p))
3348 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3349 * Returns the number of unique elements.
3351 static int pidlist_uniq(pid_t *list, int length)
3356 * we presume the 0th element is unique, so i starts at 1. trivial
3357 * edge cases first; no work needs to be done for either
3359 if (length == 0 || length == 1)
3361 /* src and dest walk down the list; dest counts unique elements */
3362 for (src = 1; src < length; src++) {
3363 /* find next unique element */
3364 while (list[src] == list[src-1]) {
3369 /* dest always points to where the next unique element goes */
3370 list[dest] = list[src];
3377 static int cmppid(const void *a, const void *b)
3379 return *(pid_t *)a - *(pid_t *)b;
3383 * find the appropriate pidlist for our purpose (given procs vs tasks)
3384 * returns with the lock on that pidlist already held, and takes care
3385 * of the use count, or returns NULL with no locks held if we're out of
3388 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3389 enum cgroup_filetype type)
3391 struct cgroup_pidlist *l;
3392 /* don't need task_nsproxy() if we're looking at ourself */
3393 struct pid_namespace *ns = task_active_pid_ns(current);
3396 * We can't drop the pidlist_mutex before taking the l->mutex in case
3397 * the last ref-holder is trying to remove l from the list at the same
3398 * time. Holding the pidlist_mutex precludes somebody taking whichever
3399 * list we find out from under us - compare release_pid_array().
3401 mutex_lock(&cgrp->pidlist_mutex);
3402 list_for_each_entry(l, &cgrp->pidlists, links) {
3403 if (l->key.type == type && l->key.ns == ns) {
3404 /* make sure l doesn't vanish out from under us */
3405 down_write(&l->mutex);
3406 mutex_unlock(&cgrp->pidlist_mutex);
3410 /* entry not found; create a new one */
3411 l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3413 mutex_unlock(&cgrp->pidlist_mutex);
3416 init_rwsem(&l->mutex);
3417 down_write(&l->mutex);
3419 l->key.ns = get_pid_ns(ns);
3420 l->use_count = 0; /* don't increment here */
3423 list_add(&l->links, &cgrp->pidlists);
3424 mutex_unlock(&cgrp->pidlist_mutex);
3429 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3431 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3432 struct cgroup_pidlist **lp)
3436 int pid, n = 0; /* used for populating the array */
3437 struct cgroup_iter it;
3438 struct task_struct *tsk;
3439 struct cgroup_pidlist *l;
3442 * If cgroup gets more users after we read count, we won't have
3443 * enough space - tough. This race is indistinguishable to the
3444 * caller from the case that the additional cgroup users didn't
3445 * show up until sometime later on.
3447 length = cgroup_task_count(cgrp);
3448 array = pidlist_allocate(length);
3451 /* now, populate the array */
3452 cgroup_iter_start(cgrp, &it);
3453 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3454 if (unlikely(n == length))
3456 /* get tgid or pid for procs or tasks file respectively */
3457 if (type == CGROUP_FILE_PROCS)
3458 pid = task_tgid_vnr(tsk);
3460 pid = task_pid_vnr(tsk);
3461 if (pid > 0) /* make sure to only use valid results */
3464 cgroup_iter_end(cgrp, &it);
3466 /* now sort & (if procs) strip out duplicates */
3467 sort(array, length, sizeof(pid_t), cmppid, NULL);
3468 if (type == CGROUP_FILE_PROCS)
3469 length = pidlist_uniq(array, length);
3470 l = cgroup_pidlist_find(cgrp, type);
3472 pidlist_free(array);
3475 /* store array, freeing old if necessary - lock already held */
3476 pidlist_free(l->list);
3480 up_write(&l->mutex);
3486 * cgroupstats_build - build and fill cgroupstats
3487 * @stats: cgroupstats to fill information into
3488 * @dentry: A dentry entry belonging to the cgroup for which stats have
3491 * Build and fill cgroupstats so that taskstats can export it to user
3494 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3497 struct cgroup *cgrp;
3498 struct cgroup_iter it;
3499 struct task_struct *tsk;
3502 * Validate dentry by checking the superblock operations,
3503 * and make sure it's a directory.
3505 if (dentry->d_sb->s_op != &cgroup_ops ||
3506 !S_ISDIR(dentry->d_inode->i_mode))
3510 cgrp = dentry->d_fsdata;
3512 cgroup_iter_start(cgrp, &it);
3513 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3514 switch (tsk->state) {
3516 stats->nr_running++;
3518 case TASK_INTERRUPTIBLE:
3519 stats->nr_sleeping++;
3521 case TASK_UNINTERRUPTIBLE:
3522 stats->nr_uninterruptible++;
3525 stats->nr_stopped++;
3528 if (delayacct_is_task_waiting_on_io(tsk))
3529 stats->nr_io_wait++;
3533 cgroup_iter_end(cgrp, &it);
3541 * seq_file methods for the tasks/procs files. The seq_file position is the
3542 * next pid to display; the seq_file iterator is a pointer to the pid
3543 * in the cgroup->l->list array.
3546 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3549 * Initially we receive a position value that corresponds to
3550 * one more than the last pid shown (or 0 on the first call or
3551 * after a seek to the start). Use a binary-search to find the
3552 * next pid to display, if any
3554 struct cgroup_pidlist *l = s->private;
3555 int index = 0, pid = *pos;
3558 down_read(&l->mutex);
3560 int end = l->length;
3562 while (index < end) {
3563 int mid = (index + end) / 2;
3564 if (l->list[mid] == pid) {
3567 } else if (l->list[mid] <= pid)
3573 /* If we're off the end of the array, we're done */
3574 if (index >= l->length)
3576 /* Update the abstract position to be the actual pid that we found */
3577 iter = l->list + index;
3582 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3584 struct cgroup_pidlist *l = s->private;
3588 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3590 struct cgroup_pidlist *l = s->private;
3592 pid_t *end = l->list + l->length;
3594 * Advance to the next pid in the array. If this goes off the
3606 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3608 return seq_printf(s, "%d\n", *(int *)v);
3612 * seq_operations functions for iterating on pidlists through seq_file -
3613 * independent of whether it's tasks or procs
3615 static const struct seq_operations cgroup_pidlist_seq_operations = {
3616 .start = cgroup_pidlist_start,
3617 .stop = cgroup_pidlist_stop,
3618 .next = cgroup_pidlist_next,
3619 .show = cgroup_pidlist_show,
3622 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3625 * the case where we're the last user of this particular pidlist will
3626 * have us remove it from the cgroup's list, which entails taking the
3627 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3628 * pidlist_mutex, we have to take pidlist_mutex first.
3630 mutex_lock(&l->owner->pidlist_mutex);
3631 down_write(&l->mutex);
3632 BUG_ON(!l->use_count);
3633 if (!--l->use_count) {
3634 /* we're the last user if refcount is 0; remove and free */
3635 list_del(&l->links);
3636 mutex_unlock(&l->owner->pidlist_mutex);
3637 pidlist_free(l->list);
3638 put_pid_ns(l->key.ns);
3639 up_write(&l->mutex);
3643 mutex_unlock(&l->owner->pidlist_mutex);
3644 up_write(&l->mutex);
3647 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3649 struct cgroup_pidlist *l;
3650 if (!(file->f_mode & FMODE_READ))
3653 * the seq_file will only be initialized if the file was opened for
3654 * reading; hence we check if it's not null only in that case.
3656 l = ((struct seq_file *)file->private_data)->private;
3657 cgroup_release_pid_array(l);
3658 return seq_release(inode, file);
3661 static const struct file_operations cgroup_pidlist_operations = {
3663 .llseek = seq_lseek,
3664 .write = cgroup_file_write,
3665 .release = cgroup_pidlist_release,
3669 * The following functions handle opens on a file that displays a pidlist
3670 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3673 /* helper function for the two below it */
3674 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3676 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3677 struct cgroup_pidlist *l;
3680 /* Nothing to do for write-only files */
3681 if (!(file->f_mode & FMODE_READ))
3684 /* have the array populated */
3685 retval = pidlist_array_load(cgrp, type, &l);
3688 /* configure file information */
3689 file->f_op = &cgroup_pidlist_operations;
3691 retval = seq_open(file, &cgroup_pidlist_seq_operations);
3693 cgroup_release_pid_array(l);
3696 ((struct seq_file *)file->private_data)->private = l;
3699 static int cgroup_tasks_open(struct inode *unused, struct file *file)
3701 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3703 static int cgroup_procs_open(struct inode *unused, struct file *file)
3705 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3708 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3711 return notify_on_release(cgrp);
3714 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3718 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3720 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3722 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3727 * Unregister event and free resources.
3729 * Gets called from workqueue.
3731 static void cgroup_event_remove(struct work_struct *work)
3733 struct cgroup_event *event = container_of(work, struct cgroup_event,
3735 struct cgroup *cgrp = event->cgrp;
3737 remove_wait_queue(event->wqh, &event->wait);
3739 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3741 /* Notify userspace the event is going away. */
3742 eventfd_signal(event->eventfd, 1);
3744 eventfd_ctx_put(event->eventfd);
3750 * Gets called on POLLHUP on eventfd when user closes it.
3752 * Called with wqh->lock held and interrupts disabled.
3754 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3755 int sync, void *key)
3757 struct cgroup_event *event = container_of(wait,
3758 struct cgroup_event, wait);
3759 struct cgroup *cgrp = event->cgrp;
3760 unsigned long flags = (unsigned long)key;
3762 if (flags & POLLHUP) {
3764 * If the event has been detached at cgroup removal, we
3765 * can simply return knowing the other side will cleanup
3768 * We can't race against event freeing since the other
3769 * side will require wqh->lock via remove_wait_queue(),
3772 spin_lock(&cgrp->event_list_lock);
3773 if (!list_empty(&event->list)) {
3774 list_del_init(&event->list);
3776 * We are in atomic context, but cgroup_event_remove()
3777 * may sleep, so we have to call it in workqueue.
3779 schedule_work(&event->remove);
3781 spin_unlock(&cgrp->event_list_lock);
3787 static void cgroup_event_ptable_queue_proc(struct file *file,
3788 wait_queue_head_t *wqh, poll_table *pt)
3790 struct cgroup_event *event = container_of(pt,
3791 struct cgroup_event, pt);
3794 add_wait_queue(wqh, &event->wait);
3798 * Parse input and register new cgroup event handler.
3800 * Input must be in format '<event_fd> <control_fd> <args>'.
3801 * Interpretation of args is defined by control file implementation.
3803 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3806 struct cgroup_event *event = NULL;
3807 struct cgroup *cgrp_cfile;
3808 unsigned int efd, cfd;
3809 struct file *efile = NULL;
3810 struct file *cfile = NULL;
3814 efd = simple_strtoul(buffer, &endp, 10);
3819 cfd = simple_strtoul(buffer, &endp, 10);
3820 if ((*endp != ' ') && (*endp != '\0'))
3824 event = kzalloc(sizeof(*event), GFP_KERNEL);
3828 INIT_LIST_HEAD(&event->list);
3829 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3830 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3831 INIT_WORK(&event->remove, cgroup_event_remove);
3833 efile = eventfd_fget(efd);
3834 if (IS_ERR(efile)) {
3835 ret = PTR_ERR(efile);
3839 event->eventfd = eventfd_ctx_fileget(efile);
3840 if (IS_ERR(event->eventfd)) {
3841 ret = PTR_ERR(event->eventfd);
3851 /* the process need read permission on control file */
3852 /* AV: shouldn't we check that it's been opened for read instead? */
3853 ret = inode_permission(file_inode(cfile), MAY_READ);
3857 event->cft = __file_cft(cfile);
3858 if (IS_ERR(event->cft)) {
3859 ret = PTR_ERR(event->cft);
3864 * The file to be monitored must be in the same cgroup as
3865 * cgroup.event_control is.
3867 cgrp_cfile = __d_cgrp(cfile->f_dentry->d_parent);
3868 if (cgrp_cfile != cgrp) {
3873 if (!event->cft->register_event || !event->cft->unregister_event) {
3878 ret = event->cft->register_event(cgrp, event->cft,
3879 event->eventfd, buffer);
3884 * Events should be removed after rmdir of cgroup directory, but before
3885 * destroying subsystem state objects. Let's take reference to cgroup
3886 * directory dentry to do that.
3890 spin_lock(&cgrp->event_list_lock);
3891 list_add(&event->list, &cgrp->event_list);
3892 spin_unlock(&cgrp->event_list_lock);
3903 if (event && event->eventfd && !IS_ERR(event->eventfd))
3904 eventfd_ctx_put(event->eventfd);
3906 if (!IS_ERR_OR_NULL(efile))
3914 static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3917 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3920 static int cgroup_clone_children_write(struct cgroup *cgrp,
3925 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3927 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3932 * for the common functions, 'private' gives the type of file
3934 /* for hysterical raisins, we can't put this on the older files */
3935 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3936 static struct cftype files[] = {
3939 .open = cgroup_tasks_open,
3940 .write_u64 = cgroup_tasks_write,
3941 .release = cgroup_pidlist_release,
3942 .mode = S_IRUGO | S_IWUSR,
3945 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3946 .open = cgroup_procs_open,
3947 .write_u64 = cgroup_procs_write,
3948 .release = cgroup_pidlist_release,
3949 .mode = S_IRUGO | S_IWUSR,
3952 .name = "notify_on_release",
3953 .read_u64 = cgroup_read_notify_on_release,
3954 .write_u64 = cgroup_write_notify_on_release,
3957 .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3958 .write_string = cgroup_write_event_control,
3962 .name = "cgroup.clone_children",
3963 .flags = CFTYPE_INSANE,
3964 .read_u64 = cgroup_clone_children_read,
3965 .write_u64 = cgroup_clone_children_write,
3968 .name = "cgroup.sane_behavior",
3969 .flags = CFTYPE_ONLY_ON_ROOT,
3970 .read_seq_string = cgroup_sane_behavior_show,
3973 .name = "release_agent",
3974 .flags = CFTYPE_ONLY_ON_ROOT,
3975 .read_seq_string = cgroup_release_agent_show,
3976 .write_string = cgroup_release_agent_write,
3977 .max_write_len = PATH_MAX,
3983 * cgroup_populate_dir - selectively creation of files in a directory
3984 * @cgrp: target cgroup
3985 * @base_files: true if the base files should be added
3986 * @subsys_mask: mask of the subsystem ids whose files should be added
3988 static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
3989 unsigned long subsys_mask)
3992 struct cgroup_subsys *ss;
3995 err = cgroup_addrm_files(cgrp, NULL, files, true);
4000 /* process cftsets of each subsystem */
4001 for_each_subsys(cgrp->root, ss) {
4002 struct cftype_set *set;
4003 if (!test_bit(ss->subsys_id, &subsys_mask))
4006 list_for_each_entry(set, &ss->cftsets, node)
4007 cgroup_addrm_files(cgrp, ss, set->cfts, true);
4010 /* This cgroup is ready now */
4011 for_each_subsys(cgrp->root, ss) {
4012 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4014 * Update id->css pointer and make this css visible from
4015 * CSS ID functions. This pointer will be dereferened
4016 * from RCU-read-side without locks.
4019 rcu_assign_pointer(css->id->css, css);
4025 static void css_dput_fn(struct work_struct *work)
4027 struct cgroup_subsys_state *css =
4028 container_of(work, struct cgroup_subsys_state, dput_work);
4029 struct dentry *dentry = css->cgroup->dentry;
4030 struct super_block *sb = dentry->d_sb;
4032 atomic_inc(&sb->s_active);
4034 deactivate_super(sb);
4037 static void init_cgroup_css(struct cgroup_subsys_state *css,
4038 struct cgroup_subsys *ss,
4039 struct cgroup *cgrp)
4042 atomic_set(&css->refcnt, 1);
4045 if (cgrp == dummytop)
4046 css->flags |= CSS_ROOT;
4047 BUG_ON(cgrp->subsys[ss->subsys_id]);
4048 cgrp->subsys[ss->subsys_id] = css;
4051 * css holds an extra ref to @cgrp->dentry which is put on the last
4052 * css_put(). dput() requires process context, which css_put() may
4053 * be called without. @css->dput_work will be used to invoke
4054 * dput() asynchronously from css_put().
4056 INIT_WORK(&css->dput_work, css_dput_fn);
4059 /* invoke ->post_create() on a new CSS and mark it online if successful */
4060 static int online_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4064 lockdep_assert_held(&cgroup_mutex);
4067 ret = ss->css_online(cgrp);
4069 cgrp->subsys[ss->subsys_id]->flags |= CSS_ONLINE;
4073 /* if the CSS is online, invoke ->pre_destory() on it and mark it offline */
4074 static void offline_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4075 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4077 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4079 lockdep_assert_held(&cgroup_mutex);
4081 if (!(css->flags & CSS_ONLINE))
4084 if (ss->css_offline)
4085 ss->css_offline(cgrp);
4087 cgrp->subsys[ss->subsys_id]->flags &= ~CSS_ONLINE;
4091 * cgroup_create - create a cgroup
4092 * @parent: cgroup that will be parent of the new cgroup
4093 * @dentry: dentry of the new cgroup
4094 * @mode: mode to set on new inode
4096 * Must be called with the mutex on the parent inode held
4098 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
4101 struct cgroup *cgrp;
4102 struct cgroup_name *name;
4103 struct cgroupfs_root *root = parent->root;
4105 struct cgroup_subsys *ss;
4106 struct super_block *sb = root->sb;
4108 /* allocate the cgroup and its ID, 0 is reserved for the root */
4109 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
4113 name = cgroup_alloc_name(dentry);
4116 rcu_assign_pointer(cgrp->name, name);
4118 cgrp->id = ida_simple_get(&root->cgroup_ida, 1, 0, GFP_KERNEL);
4123 * Only live parents can have children. Note that the liveliness
4124 * check isn't strictly necessary because cgroup_mkdir() and
4125 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4126 * anyway so that locking is contained inside cgroup proper and we
4127 * don't get nasty surprises if we ever grow another caller.
4129 if (!cgroup_lock_live_group(parent)) {
4134 /* Grab a reference on the superblock so the hierarchy doesn't
4135 * get deleted on unmount if there are child cgroups. This
4136 * can be done outside cgroup_mutex, since the sb can't
4137 * disappear while someone has an open control file on the
4139 atomic_inc(&sb->s_active);
4141 init_cgroup_housekeeping(cgrp);
4143 dentry->d_fsdata = cgrp;
4144 cgrp->dentry = dentry;
4146 cgrp->parent = parent;
4147 cgrp->root = parent->root;
4149 if (notify_on_release(parent))
4150 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
4152 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
4153 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4155 for_each_subsys(root, ss) {
4156 struct cgroup_subsys_state *css;
4158 css = ss->css_alloc(cgrp);
4163 init_cgroup_css(css, ss, cgrp);
4165 err = alloc_css_id(ss, parent, cgrp);
4172 * Create directory. cgroup_create_file() returns with the new
4173 * directory locked on success so that it can be populated without
4174 * dropping cgroup_mutex.
4176 err = cgroup_create_file(dentry, S_IFDIR | mode, sb);
4179 lockdep_assert_held(&dentry->d_inode->i_mutex);
4181 /* allocation complete, commit to creation */
4182 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
4183 list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children);
4184 root->number_of_cgroups++;
4186 /* each css holds a ref to the cgroup's dentry */
4187 for_each_subsys(root, ss)
4190 /* hold a ref to the parent's dentry */
4191 dget(parent->dentry);
4193 /* creation succeeded, notify subsystems */
4194 for_each_subsys(root, ss) {
4195 err = online_css(ss, cgrp);
4199 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
4201 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",
4202 current->comm, current->pid, ss->name);
4203 if (!strcmp(ss->name, "memory"))
4204 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4205 ss->warned_broken_hierarchy = true;
4209 err = cgroup_populate_dir(cgrp, true, root->subsys_mask);
4213 mutex_unlock(&cgroup_mutex);
4214 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
4219 for_each_subsys(root, ss) {
4220 if (cgrp->subsys[ss->subsys_id])
4223 mutex_unlock(&cgroup_mutex);
4224 /* Release the reference count that we took on the superblock */
4225 deactivate_super(sb);
4227 ida_simple_remove(&root->cgroup_ida, cgrp->id);
4229 kfree(rcu_dereference_raw(cgrp->name));
4235 cgroup_destroy_locked(cgrp);
4236 mutex_unlock(&cgroup_mutex);
4237 mutex_unlock(&dentry->d_inode->i_mutex);
4241 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4243 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
4245 /* the vfs holds inode->i_mutex already */
4246 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
4249 static int cgroup_destroy_locked(struct cgroup *cgrp)
4250 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4252 struct dentry *d = cgrp->dentry;
4253 struct cgroup *parent = cgrp->parent;
4254 struct cgroup_event *event, *tmp;
4255 struct cgroup_subsys *ss;
4257 lockdep_assert_held(&d->d_inode->i_mutex);
4258 lockdep_assert_held(&cgroup_mutex);
4260 if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children))
4264 * Block new css_tryget() by deactivating refcnt and mark @cgrp
4265 * removed. This makes future css_tryget() and child creation
4266 * attempts fail thus maintaining the removal conditions verified
4269 for_each_subsys(cgrp->root, ss) {
4270 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4272 WARN_ON(atomic_read(&css->refcnt) < 0);
4273 atomic_add(CSS_DEACT_BIAS, &css->refcnt);
4275 set_bit(CGRP_REMOVED, &cgrp->flags);
4277 /* tell subsystems to initate destruction */
4278 for_each_subsys(cgrp->root, ss)
4279 offline_css(ss, cgrp);
4282 * Put all the base refs. Each css holds an extra reference to the
4283 * cgroup's dentry and cgroup removal proceeds regardless of css
4284 * refs. On the last put of each css, whenever that may be, the
4285 * extra dentry ref is put so that dentry destruction happens only
4286 * after all css's are released.
4288 for_each_subsys(cgrp->root, ss)
4289 css_put(cgrp->subsys[ss->subsys_id]);
4291 raw_spin_lock(&release_list_lock);
4292 if (!list_empty(&cgrp->release_list))
4293 list_del_init(&cgrp->release_list);
4294 raw_spin_unlock(&release_list_lock);
4296 /* delete this cgroup from parent->children */
4297 list_del_rcu(&cgrp->sibling);
4298 list_del_init(&cgrp->allcg_node);
4301 cgroup_d_remove_dir(d);
4304 set_bit(CGRP_RELEASABLE, &parent->flags);
4305 check_for_release(parent);
4308 * Unregister events and notify userspace.
4309 * Notify userspace about cgroup removing only after rmdir of cgroup
4310 * directory to avoid race between userspace and kernelspace.
4312 spin_lock(&cgrp->event_list_lock);
4313 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4314 list_del_init(&event->list);
4315 schedule_work(&event->remove);
4317 spin_unlock(&cgrp->event_list_lock);
4322 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
4326 mutex_lock(&cgroup_mutex);
4327 ret = cgroup_destroy_locked(dentry->d_fsdata);
4328 mutex_unlock(&cgroup_mutex);
4333 static void __init_or_module cgroup_init_cftsets(struct cgroup_subsys *ss)
4335 INIT_LIST_HEAD(&ss->cftsets);
4338 * base_cftset is embedded in subsys itself, no need to worry about
4341 if (ss->base_cftypes) {
4342 ss->base_cftset.cfts = ss->base_cftypes;
4343 list_add_tail(&ss->base_cftset.node, &ss->cftsets);
4347 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4349 struct cgroup_subsys_state *css;
4351 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4353 mutex_lock(&cgroup_mutex);
4355 /* init base cftset */
4356 cgroup_init_cftsets(ss);
4358 /* Create the top cgroup state for this subsystem */
4359 list_add(&ss->sibling, &rootnode.subsys_list);
4360 ss->root = &rootnode;
4361 css = ss->css_alloc(dummytop);
4362 /* We don't handle early failures gracefully */
4363 BUG_ON(IS_ERR(css));
4364 init_cgroup_css(css, ss, dummytop);
4366 /* Update the init_css_set to contain a subsys
4367 * pointer to this state - since the subsystem is
4368 * newly registered, all tasks and hence the
4369 * init_css_set is in the subsystem's top cgroup. */
4370 init_css_set.subsys[ss->subsys_id] = css;
4372 need_forkexit_callback |= ss->fork || ss->exit;
4374 /* At system boot, before all subsystems have been
4375 * registered, no tasks have been forked, so we don't
4376 * need to invoke fork callbacks here. */
4377 BUG_ON(!list_empty(&init_task.tasks));
4380 BUG_ON(online_css(ss, dummytop));
4382 mutex_unlock(&cgroup_mutex);
4384 /* this function shouldn't be used with modular subsystems, since they
4385 * need to register a subsys_id, among other things */
4390 * cgroup_load_subsys: load and register a modular subsystem at runtime
4391 * @ss: the subsystem to load
4393 * This function should be called in a modular subsystem's initcall. If the
4394 * subsystem is built as a module, it will be assigned a new subsys_id and set
4395 * up for use. If the subsystem is built-in anyway, work is delegated to the
4396 * simpler cgroup_init_subsys.
4398 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4400 struct cgroup_subsys_state *css;
4402 struct hlist_node *tmp;
4406 /* check name and function validity */
4407 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4408 ss->css_alloc == NULL || ss->css_free == NULL)
4412 * we don't support callbacks in modular subsystems. this check is
4413 * before the ss->module check for consistency; a subsystem that could
4414 * be a module should still have no callbacks even if the user isn't
4415 * compiling it as one.
4417 if (ss->fork || ss->exit)
4421 * an optionally modular subsystem is built-in: we want to do nothing,
4422 * since cgroup_init_subsys will have already taken care of it.
4424 if (ss->module == NULL) {
4425 /* a sanity check */
4426 BUG_ON(subsys[ss->subsys_id] != ss);
4430 /* init base cftset */
4431 cgroup_init_cftsets(ss);
4433 mutex_lock(&cgroup_mutex);
4434 subsys[ss->subsys_id] = ss;
4437 * no ss->css_alloc seems to need anything important in the ss
4438 * struct, so this can happen first (i.e. before the rootnode
4441 css = ss->css_alloc(dummytop);
4443 /* failure case - need to deassign the subsys[] slot. */
4444 subsys[ss->subsys_id] = NULL;
4445 mutex_unlock(&cgroup_mutex);
4446 return PTR_ERR(css);
4449 list_add(&ss->sibling, &rootnode.subsys_list);
4450 ss->root = &rootnode;
4452 /* our new subsystem will be attached to the dummy hierarchy. */
4453 init_cgroup_css(css, ss, dummytop);
4454 /* init_idr must be after init_cgroup_css because it sets css->id. */
4456 ret = cgroup_init_idr(ss, css);
4462 * Now we need to entangle the css into the existing css_sets. unlike
4463 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4464 * will need a new pointer to it; done by iterating the css_set_table.
4465 * furthermore, modifying the existing css_sets will corrupt the hash
4466 * table state, so each changed css_set will need its hash recomputed.
4467 * this is all done under the css_set_lock.
4469 write_lock(&css_set_lock);
4470 hash_for_each_safe(css_set_table, i, tmp, cg, hlist) {
4471 /* skip entries that we already rehashed */
4472 if (cg->subsys[ss->subsys_id])
4474 /* remove existing entry */
4475 hash_del(&cg->hlist);
4477 cg->subsys[ss->subsys_id] = css;
4478 /* recompute hash and restore entry */
4479 key = css_set_hash(cg->subsys);
4480 hash_add(css_set_table, &cg->hlist, key);
4482 write_unlock(&css_set_lock);
4485 ret = online_css(ss, dummytop);
4490 mutex_unlock(&cgroup_mutex);
4494 mutex_unlock(&cgroup_mutex);
4495 /* @ss can't be mounted here as try_module_get() would fail */
4496 cgroup_unload_subsys(ss);
4499 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4502 * cgroup_unload_subsys: unload a modular subsystem
4503 * @ss: the subsystem to unload
4505 * This function should be called in a modular subsystem's exitcall. When this
4506 * function is invoked, the refcount on the subsystem's module will be 0, so
4507 * the subsystem will not be attached to any hierarchy.
4509 void cgroup_unload_subsys(struct cgroup_subsys *ss)
4511 struct cg_cgroup_link *link;
4513 BUG_ON(ss->module == NULL);
4516 * we shouldn't be called if the subsystem is in use, and the use of
4517 * try_module_get in parse_cgroupfs_options should ensure that it
4518 * doesn't start being used while we're killing it off.
4520 BUG_ON(ss->root != &rootnode);
4522 mutex_lock(&cgroup_mutex);
4524 offline_css(ss, dummytop);
4528 idr_destroy(&ss->idr);
4530 /* deassign the subsys_id */
4531 subsys[ss->subsys_id] = NULL;
4533 /* remove subsystem from rootnode's list of subsystems */
4534 list_del_init(&ss->sibling);
4537 * disentangle the css from all css_sets attached to the dummytop. as
4538 * in loading, we need to pay our respects to the hashtable gods.
4540 write_lock(&css_set_lock);
4541 list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
4542 struct css_set *cg = link->cg;
4545 hash_del(&cg->hlist);
4546 cg->subsys[ss->subsys_id] = NULL;
4547 key = css_set_hash(cg->subsys);
4548 hash_add(css_set_table, &cg->hlist, key);
4550 write_unlock(&css_set_lock);
4553 * remove subsystem's css from the dummytop and free it - need to
4554 * free before marking as null because ss->css_free needs the
4555 * cgrp->subsys pointer to find their state. note that this also
4556 * takes care of freeing the css_id.
4558 ss->css_free(dummytop);
4559 dummytop->subsys[ss->subsys_id] = NULL;
4561 mutex_unlock(&cgroup_mutex);
4563 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4566 * cgroup_init_early - cgroup initialization at system boot
4568 * Initialize cgroups at system boot, and initialize any
4569 * subsystems that request early init.
4571 int __init cgroup_init_early(void)
4574 atomic_set(&init_css_set.refcount, 1);
4575 INIT_LIST_HEAD(&init_css_set.cg_links);
4576 INIT_LIST_HEAD(&init_css_set.tasks);
4577 INIT_HLIST_NODE(&init_css_set.hlist);
4579 init_cgroup_root(&rootnode);
4581 init_task.cgroups = &init_css_set;
4583 init_css_set_link.cg = &init_css_set;
4584 init_css_set_link.cgrp = dummytop;
4585 list_add(&init_css_set_link.cgrp_link_list,
4586 &rootnode.top_cgroup.css_sets);
4587 list_add(&init_css_set_link.cg_link_list,
4588 &init_css_set.cg_links);
4590 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4591 struct cgroup_subsys *ss = subsys[i];
4593 /* at bootup time, we don't worry about modular subsystems */
4594 if (!ss || ss->module)
4598 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4599 BUG_ON(!ss->css_alloc);
4600 BUG_ON(!ss->css_free);
4601 if (ss->subsys_id != i) {
4602 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4603 ss->name, ss->subsys_id);
4608 cgroup_init_subsys(ss);
4614 * cgroup_init - cgroup initialization
4616 * Register cgroup filesystem and /proc file, and initialize
4617 * any subsystems that didn't request early init.
4619 int __init cgroup_init(void)
4625 err = bdi_init(&cgroup_backing_dev_info);
4629 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4630 struct cgroup_subsys *ss = subsys[i];
4632 /* at bootup time, we don't worry about modular subsystems */
4633 if (!ss || ss->module)
4635 if (!ss->early_init)
4636 cgroup_init_subsys(ss);
4638 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4641 /* Add init_css_set to the hash table */
4642 key = css_set_hash(init_css_set.subsys);
4643 hash_add(css_set_table, &init_css_set.hlist, key);
4644 BUG_ON(!init_root_id(&rootnode));
4646 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4652 err = register_filesystem(&cgroup_fs_type);
4654 kobject_put(cgroup_kobj);
4658 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4662 bdi_destroy(&cgroup_backing_dev_info);
4668 * proc_cgroup_show()
4669 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4670 * - Used for /proc/<pid>/cgroup.
4671 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4672 * doesn't really matter if tsk->cgroup changes after we read it,
4673 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4674 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4675 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4676 * cgroup to top_cgroup.
4679 /* TODO: Use a proper seq_file iterator */
4680 static int proc_cgroup_show(struct seq_file *m, void *v)
4683 struct task_struct *tsk;
4686 struct cgroupfs_root *root;
4689 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4695 tsk = get_pid_task(pid, PIDTYPE_PID);
4701 mutex_lock(&cgroup_mutex);
4703 for_each_active_root(root) {
4704 struct cgroup_subsys *ss;
4705 struct cgroup *cgrp;
4708 seq_printf(m, "%d:", root->hierarchy_id);
4709 for_each_subsys(root, ss)
4710 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4711 if (strlen(root->name))
4712 seq_printf(m, "%sname=%s", count ? "," : "",
4715 cgrp = task_cgroup_from_root(tsk, root);
4716 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4724 mutex_unlock(&cgroup_mutex);
4725 put_task_struct(tsk);
4732 static int cgroup_open(struct inode *inode, struct file *file)
4734 struct pid *pid = PROC_I(inode)->pid;
4735 return single_open(file, proc_cgroup_show, pid);
4738 const struct file_operations proc_cgroup_operations = {
4739 .open = cgroup_open,
4741 .llseek = seq_lseek,
4742 .release = single_release,
4745 /* Display information about each subsystem and each hierarchy */
4746 static int proc_cgroupstats_show(struct seq_file *m, void *v)
4750 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4752 * ideally we don't want subsystems moving around while we do this.
4753 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4754 * subsys/hierarchy state.
4756 mutex_lock(&cgroup_mutex);
4757 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4758 struct cgroup_subsys *ss = subsys[i];
4761 seq_printf(m, "%s\t%d\t%d\t%d\n",
4762 ss->name, ss->root->hierarchy_id,
4763 ss->root->number_of_cgroups, !ss->disabled);
4765 mutex_unlock(&cgroup_mutex);
4769 static int cgroupstats_open(struct inode *inode, struct file *file)
4771 return single_open(file, proc_cgroupstats_show, NULL);
4774 static const struct file_operations proc_cgroupstats_operations = {
4775 .open = cgroupstats_open,
4777 .llseek = seq_lseek,
4778 .release = single_release,
4782 * cgroup_fork - attach newly forked task to its parents cgroup.
4783 * @child: pointer to task_struct of forking parent process.
4785 * Description: A task inherits its parent's cgroup at fork().
4787 * A pointer to the shared css_set was automatically copied in
4788 * fork.c by dup_task_struct(). However, we ignore that copy, since
4789 * it was not made under the protection of RCU or cgroup_mutex, so
4790 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4791 * have already changed current->cgroups, allowing the previously
4792 * referenced cgroup group to be removed and freed.
4794 * At the point that cgroup_fork() is called, 'current' is the parent
4795 * task, and the passed argument 'child' points to the child task.
4797 void cgroup_fork(struct task_struct *child)
4800 child->cgroups = current->cgroups;
4801 get_css_set(child->cgroups);
4802 task_unlock(current);
4803 INIT_LIST_HEAD(&child->cg_list);
4807 * cgroup_post_fork - called on a new task after adding it to the task list
4808 * @child: the task in question
4810 * Adds the task to the list running through its css_set if necessary and
4811 * call the subsystem fork() callbacks. Has to be after the task is
4812 * visible on the task list in case we race with the first call to
4813 * cgroup_iter_start() - to guarantee that the new task ends up on its
4816 void cgroup_post_fork(struct task_struct *child)
4821 * use_task_css_set_links is set to 1 before we walk the tasklist
4822 * under the tasklist_lock and we read it here after we added the child
4823 * to the tasklist under the tasklist_lock as well. If the child wasn't
4824 * yet in the tasklist when we walked through it from
4825 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4826 * should be visible now due to the paired locking and barriers implied
4827 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4828 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4831 if (use_task_css_set_links) {
4832 write_lock(&css_set_lock);
4834 if (list_empty(&child->cg_list))
4835 list_add(&child->cg_list, &child->cgroups->tasks);
4837 write_unlock(&css_set_lock);
4841 * Call ss->fork(). This must happen after @child is linked on
4842 * css_set; otherwise, @child might change state between ->fork()
4843 * and addition to css_set.
4845 if (need_forkexit_callback) {
4847 * fork/exit callbacks are supported only for builtin
4848 * subsystems, and the builtin section of the subsys
4849 * array is immutable, so we don't need to lock the
4850 * subsys array here. On the other hand, modular section
4851 * of the array can be freed at module unload, so we
4854 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4855 struct cgroup_subsys *ss = subsys[i];
4864 * cgroup_exit - detach cgroup from exiting task
4865 * @tsk: pointer to task_struct of exiting process
4866 * @run_callback: run exit callbacks?
4868 * Description: Detach cgroup from @tsk and release it.
4870 * Note that cgroups marked notify_on_release force every task in
4871 * them to take the global cgroup_mutex mutex when exiting.
4872 * This could impact scaling on very large systems. Be reluctant to
4873 * use notify_on_release cgroups where very high task exit scaling
4874 * is required on large systems.
4876 * the_top_cgroup_hack:
4878 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4880 * We call cgroup_exit() while the task is still competent to
4881 * handle notify_on_release(), then leave the task attached to the
4882 * root cgroup in each hierarchy for the remainder of its exit.
4884 * To do this properly, we would increment the reference count on
4885 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4886 * code we would add a second cgroup function call, to drop that
4887 * reference. This would just create an unnecessary hot spot on
4888 * the top_cgroup reference count, to no avail.
4890 * Normally, holding a reference to a cgroup without bumping its
4891 * count is unsafe. The cgroup could go away, or someone could
4892 * attach us to a different cgroup, decrementing the count on
4893 * the first cgroup that we never incremented. But in this case,
4894 * top_cgroup isn't going away, and either task has PF_EXITING set,
4895 * which wards off any cgroup_attach_task() attempts, or task is a failed
4896 * fork, never visible to cgroup_attach_task.
4898 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4904 * Unlink from the css_set task list if necessary.
4905 * Optimistically check cg_list before taking
4908 if (!list_empty(&tsk->cg_list)) {
4909 write_lock(&css_set_lock);
4910 if (!list_empty(&tsk->cg_list))
4911 list_del_init(&tsk->cg_list);
4912 write_unlock(&css_set_lock);
4915 /* Reassign the task to the init_css_set. */
4918 tsk->cgroups = &init_css_set;
4920 if (run_callbacks && need_forkexit_callback) {
4922 * fork/exit callbacks are supported only for builtin
4923 * subsystems, see cgroup_post_fork() for details.
4925 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4926 struct cgroup_subsys *ss = subsys[i];
4929 struct cgroup *old_cgrp =
4930 rcu_dereference_raw(cg->subsys[i])->cgroup;
4931 struct cgroup *cgrp = task_cgroup(tsk, i);
4932 ss->exit(cgrp, old_cgrp, tsk);
4938 put_css_set_taskexit(cg);
4941 static void check_for_release(struct cgroup *cgrp)
4943 /* All of these checks rely on RCU to keep the cgroup
4944 * structure alive */
4945 if (cgroup_is_releasable(cgrp) &&
4946 !atomic_read(&cgrp->count) && list_empty(&cgrp->children)) {
4948 * Control Group is currently removeable. If it's not
4949 * already queued for a userspace notification, queue
4952 int need_schedule_work = 0;
4954 raw_spin_lock(&release_list_lock);
4955 if (!cgroup_is_removed(cgrp) &&
4956 list_empty(&cgrp->release_list)) {
4957 list_add(&cgrp->release_list, &release_list);
4958 need_schedule_work = 1;
4960 raw_spin_unlock(&release_list_lock);
4961 if (need_schedule_work)
4962 schedule_work(&release_agent_work);
4966 /* Caller must verify that the css is not for root cgroup */
4967 bool __css_tryget(struct cgroup_subsys_state *css)
4972 v = css_refcnt(css);
4973 t = atomic_cmpxchg(&css->refcnt, v, v + 1);
4981 EXPORT_SYMBOL_GPL(__css_tryget);
4983 /* Caller must verify that the css is not for root cgroup */
4984 void __css_put(struct cgroup_subsys_state *css)
4988 v = css_unbias_refcnt(atomic_dec_return(&css->refcnt));
4990 schedule_work(&css->dput_work);
4992 EXPORT_SYMBOL_GPL(__css_put);
4995 * Notify userspace when a cgroup is released, by running the
4996 * configured release agent with the name of the cgroup (path
4997 * relative to the root of cgroup file system) as the argument.
4999 * Most likely, this user command will try to rmdir this cgroup.
5001 * This races with the possibility that some other task will be
5002 * attached to this cgroup before it is removed, or that some other
5003 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5004 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5005 * unused, and this cgroup will be reprieved from its death sentence,
5006 * to continue to serve a useful existence. Next time it's released,
5007 * we will get notified again, if it still has 'notify_on_release' set.
5009 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5010 * means only wait until the task is successfully execve()'d. The
5011 * separate release agent task is forked by call_usermodehelper(),
5012 * then control in this thread returns here, without waiting for the
5013 * release agent task. We don't bother to wait because the caller of
5014 * this routine has no use for the exit status of the release agent
5015 * task, so no sense holding our caller up for that.
5017 static void cgroup_release_agent(struct work_struct *work)
5019 BUG_ON(work != &release_agent_work);
5020 mutex_lock(&cgroup_mutex);
5021 raw_spin_lock(&release_list_lock);
5022 while (!list_empty(&release_list)) {
5023 char *argv[3], *envp[3];
5025 char *pathbuf = NULL, *agentbuf = NULL;
5026 struct cgroup *cgrp = list_entry(release_list.next,
5029 list_del_init(&cgrp->release_list);
5030 raw_spin_unlock(&release_list_lock);
5031 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
5034 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
5036 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
5041 argv[i++] = agentbuf;
5042 argv[i++] = pathbuf;
5046 /* minimal command environment */
5047 envp[i++] = "HOME=/";
5048 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5051 /* Drop the lock while we invoke the usermode helper,
5052 * since the exec could involve hitting disk and hence
5053 * be a slow process */
5054 mutex_unlock(&cgroup_mutex);
5055 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
5056 mutex_lock(&cgroup_mutex);
5060 raw_spin_lock(&release_list_lock);
5062 raw_spin_unlock(&release_list_lock);
5063 mutex_unlock(&cgroup_mutex);
5066 static int __init cgroup_disable(char *str)
5071 while ((token = strsep(&str, ",")) != NULL) {
5074 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
5075 struct cgroup_subsys *ss = subsys[i];
5078 * cgroup_disable, being at boot time, can't
5079 * know about module subsystems, so we don't
5082 if (!ss || ss->module)
5085 if (!strcmp(token, ss->name)) {
5087 printk(KERN_INFO "Disabling %s control group"
5088 " subsystem\n", ss->name);
5095 __setup("cgroup_disable=", cgroup_disable);
5098 * Functons for CSS ID.
5102 *To get ID other than 0, this should be called when !cgroup_is_removed().
5104 unsigned short css_id(struct cgroup_subsys_state *css)
5106 struct css_id *cssid;
5109 * This css_id() can return correct value when somone has refcnt
5110 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5111 * it's unchanged until freed.
5113 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5119 EXPORT_SYMBOL_GPL(css_id);
5121 unsigned short css_depth(struct cgroup_subsys_state *css)
5123 struct css_id *cssid;
5125 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5128 return cssid->depth;
5131 EXPORT_SYMBOL_GPL(css_depth);
5134 * css_is_ancestor - test "root" css is an ancestor of "child"
5135 * @child: the css to be tested.
5136 * @root: the css supporsed to be an ancestor of the child.
5138 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5139 * this function reads css->id, the caller must hold rcu_read_lock().
5140 * But, considering usual usage, the csses should be valid objects after test.
5141 * Assuming that the caller will do some action to the child if this returns
5142 * returns true, the caller must take "child";s reference count.
5143 * If "child" is valid object and this returns true, "root" is valid, too.
5146 bool css_is_ancestor(struct cgroup_subsys_state *child,
5147 const struct cgroup_subsys_state *root)
5149 struct css_id *child_id;
5150 struct css_id *root_id;
5152 child_id = rcu_dereference(child->id);
5155 root_id = rcu_dereference(root->id);
5158 if (child_id->depth < root_id->depth)
5160 if (child_id->stack[root_id->depth] != root_id->id)
5165 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
5167 struct css_id *id = css->id;
5168 /* When this is called before css_id initialization, id can be NULL */
5172 BUG_ON(!ss->use_id);
5174 rcu_assign_pointer(id->css, NULL);
5175 rcu_assign_pointer(css->id, NULL);
5176 spin_lock(&ss->id_lock);
5177 idr_remove(&ss->idr, id->id);
5178 spin_unlock(&ss->id_lock);
5179 kfree_rcu(id, rcu_head);
5181 EXPORT_SYMBOL_GPL(free_css_id);
5184 * This is called by init or create(). Then, calls to this function are
5185 * always serialized (By cgroup_mutex() at create()).
5188 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
5190 struct css_id *newid;
5193 BUG_ON(!ss->use_id);
5195 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
5196 newid = kzalloc(size, GFP_KERNEL);
5198 return ERR_PTR(-ENOMEM);
5200 idr_preload(GFP_KERNEL);
5201 spin_lock(&ss->id_lock);
5202 /* Don't use 0. allocates an ID of 1-65535 */
5203 ret = idr_alloc(&ss->idr, newid, 1, CSS_ID_MAX + 1, GFP_NOWAIT);
5204 spin_unlock(&ss->id_lock);
5207 /* Returns error when there are no free spaces for new ID.*/
5212 newid->depth = depth;
5216 return ERR_PTR(ret);
5220 static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
5221 struct cgroup_subsys_state *rootcss)
5223 struct css_id *newid;
5225 spin_lock_init(&ss->id_lock);
5228 newid = get_new_cssid(ss, 0);
5230 return PTR_ERR(newid);
5232 newid->stack[0] = newid->id;
5233 newid->css = rootcss;
5234 rootcss->id = newid;
5238 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
5239 struct cgroup *child)
5241 int subsys_id, i, depth = 0;
5242 struct cgroup_subsys_state *parent_css, *child_css;
5243 struct css_id *child_id, *parent_id;
5245 subsys_id = ss->subsys_id;
5246 parent_css = parent->subsys[subsys_id];
5247 child_css = child->subsys[subsys_id];
5248 parent_id = parent_css->id;
5249 depth = parent_id->depth + 1;
5251 child_id = get_new_cssid(ss, depth);
5252 if (IS_ERR(child_id))
5253 return PTR_ERR(child_id);
5255 for (i = 0; i < depth; i++)
5256 child_id->stack[i] = parent_id->stack[i];
5257 child_id->stack[depth] = child_id->id;
5259 * child_id->css pointer will be set after this cgroup is available
5260 * see cgroup_populate_dir()
5262 rcu_assign_pointer(child_css->id, child_id);
5268 * css_lookup - lookup css by id
5269 * @ss: cgroup subsys to be looked into.
5272 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5273 * NULL if not. Should be called under rcu_read_lock()
5275 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5277 struct css_id *cssid = NULL;
5279 BUG_ON(!ss->use_id);
5280 cssid = idr_find(&ss->idr, id);
5282 if (unlikely(!cssid))
5285 return rcu_dereference(cssid->css);
5287 EXPORT_SYMBOL_GPL(css_lookup);
5290 * css_get_next - lookup next cgroup under specified hierarchy.
5291 * @ss: pointer to subsystem
5292 * @id: current position of iteration.
5293 * @root: pointer to css. search tree under this.
5294 * @foundid: position of found object.
5296 * Search next css under the specified hierarchy of rootid. Calling under
5297 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5299 struct cgroup_subsys_state *
5300 css_get_next(struct cgroup_subsys *ss, int id,
5301 struct cgroup_subsys_state *root, int *foundid)
5303 struct cgroup_subsys_state *ret = NULL;
5306 int rootid = css_id(root);
5307 int depth = css_depth(root);
5312 BUG_ON(!ss->use_id);
5313 WARN_ON_ONCE(!rcu_read_lock_held());
5315 /* fill start point for scan */
5319 * scan next entry from bitmap(tree), tmpid is updated after
5322 tmp = idr_get_next(&ss->idr, &tmpid);
5325 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
5326 ret = rcu_dereference(tmp->css);
5332 /* continue to scan from next id */
5339 * get corresponding css from file open on cgroupfs directory
5341 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5343 struct cgroup *cgrp;
5344 struct inode *inode;
5345 struct cgroup_subsys_state *css;
5347 inode = file_inode(f);
5348 /* check in cgroup filesystem dir */
5349 if (inode->i_op != &cgroup_dir_inode_operations)
5350 return ERR_PTR(-EBADF);
5352 if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5353 return ERR_PTR(-EINVAL);
5356 cgrp = __d_cgrp(f->f_dentry);
5357 css = cgrp->subsys[id];
5358 return css ? css : ERR_PTR(-ENOENT);
5361 #ifdef CONFIG_CGROUP_DEBUG
5362 static struct cgroup_subsys_state *debug_css_alloc(struct cgroup *cont)
5364 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5367 return ERR_PTR(-ENOMEM);
5372 static void debug_css_free(struct cgroup *cont)
5374 kfree(cont->subsys[debug_subsys_id]);
5377 static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
5379 return atomic_read(&cont->count);
5382 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
5384 return cgroup_task_count(cont);
5387 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
5389 return (u64)(unsigned long)current->cgroups;
5392 static u64 current_css_set_refcount_read(struct cgroup *cont,
5398 count = atomic_read(¤t->cgroups->refcount);
5403 static int current_css_set_cg_links_read(struct cgroup *cont,
5405 struct seq_file *seq)
5407 struct cg_cgroup_link *link;
5410 read_lock(&css_set_lock);
5412 cg = rcu_dereference(current->cgroups);
5413 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
5414 struct cgroup *c = link->cgrp;
5418 name = c->dentry->d_name.name;
5421 seq_printf(seq, "Root %d group %s\n",
5422 c->root->hierarchy_id, name);
5425 read_unlock(&css_set_lock);
5429 #define MAX_TASKS_SHOWN_PER_CSS 25
5430 static int cgroup_css_links_read(struct cgroup *cont,
5432 struct seq_file *seq)
5434 struct cg_cgroup_link *link;
5436 read_lock(&css_set_lock);
5437 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
5438 struct css_set *cg = link->cg;
5439 struct task_struct *task;
5441 seq_printf(seq, "css_set %p\n", cg);
5442 list_for_each_entry(task, &cg->tasks, cg_list) {
5443 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5444 seq_puts(seq, " ...\n");
5447 seq_printf(seq, " task %d\n",
5448 task_pid_vnr(task));
5452 read_unlock(&css_set_lock);
5456 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5458 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5461 static struct cftype debug_files[] = {
5463 .name = "cgroup_refcount",
5464 .read_u64 = cgroup_refcount_read,
5467 .name = "taskcount",
5468 .read_u64 = debug_taskcount_read,
5472 .name = "current_css_set",
5473 .read_u64 = current_css_set_read,
5477 .name = "current_css_set_refcount",
5478 .read_u64 = current_css_set_refcount_read,
5482 .name = "current_css_set_cg_links",
5483 .read_seq_string = current_css_set_cg_links_read,
5487 .name = "cgroup_css_links",
5488 .read_seq_string = cgroup_css_links_read,
5492 .name = "releasable",
5493 .read_u64 = releasable_read,
5499 struct cgroup_subsys debug_subsys = {
5501 .css_alloc = debug_css_alloc,
5502 .css_free = debug_css_free,
5503 .subsys_id = debug_subsys_id,
5504 .base_cftypes = debug_files,
5506 #endif /* CONFIG_CGROUP_DEBUG */