2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Copyright notices from the original cpuset code:
8 * --------------------------------------------------
9 * Copyright (C) 2003 BULL SA.
10 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
12 * Portions derived from Patrick Mochel's sysfs code.
13 * sysfs is Copyright (c) 2001-3 Patrick Mochel
15 * 2003-10-10 Written by Simon Derr.
16 * 2003-10-22 Updates by Stephen Hemminger.
17 * 2004 May-July Rework by Paul Jackson.
18 * ---------------------------------------------------
20 * This file is subject to the terms and conditions of the GNU General Public
21 * License. See the file COPYING in the main directory of the Linux
22 * distribution for more details.
25 #include <linux/cgroup.h>
26 #include <linux/ctype.h>
27 #include <linux/errno.h>
29 #include <linux/kernel.h>
30 #include <linux/list.h>
32 #include <linux/mutex.h>
33 #include <linux/mount.h>
34 #include <linux/pagemap.h>
35 #include <linux/proc_fs.h>
36 #include <linux/rcupdate.h>
37 #include <linux/sched.h>
38 #include <linux/backing-dev.h>
39 #include <linux/seq_file.h>
40 #include <linux/slab.h>
41 #include <linux/magic.h>
42 #include <linux/spinlock.h>
43 #include <linux/string.h>
44 #include <linux/sort.h>
45 #include <linux/kmod.h>
46 #include <linux/delayacct.h>
47 #include <linux/cgroupstats.h>
48 #include <linux/hash.h>
49 #include <linux/namei.h>
50 #include <linux/smp_lock.h>
51 #include <linux/pid_namespace.h>
52 #include <linux/idr.h>
54 #include <asm/atomic.h>
56 static DEFINE_MUTEX(cgroup_mutex);
58 /* Generate an array of cgroup subsystem pointers */
59 #define SUBSYS(_x) &_x ## _subsys,
61 static struct cgroup_subsys *subsys[] = {
62 #include <linux/cgroup_subsys.h>
65 #define MAX_CGROUP_ROOT_NAMELEN 64
68 * A cgroupfs_root represents the root of a cgroup hierarchy,
69 * and may be associated with a superblock to form an active
72 struct cgroupfs_root {
73 struct super_block *sb;
76 * The bitmask of subsystems intended to be attached to this
79 unsigned long subsys_bits;
81 /* Unique id for this hierarchy. */
84 /* The bitmask of subsystems currently attached to this hierarchy */
85 unsigned long actual_subsys_bits;
87 /* A list running through the attached subsystems */
88 struct list_head subsys_list;
90 /* The root cgroup for this hierarchy */
91 struct cgroup top_cgroup;
93 /* Tracks how many cgroups are currently defined in hierarchy.*/
94 int number_of_cgroups;
96 /* A list running through the active hierarchies */
97 struct list_head root_list;
99 /* Hierarchy-specific flags */
102 /* The path to use for release notifications. */
103 char release_agent_path[PATH_MAX];
105 /* The name for this hierarchy - may be empty */
106 char name[MAX_CGROUP_ROOT_NAMELEN];
110 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
111 * subsystems that are otherwise unattached - it never has more than a
112 * single cgroup, and all tasks are part of that cgroup.
114 static struct cgroupfs_root rootnode;
117 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
118 * cgroup_subsys->use_id != 0.
120 #define CSS_ID_MAX (65535)
123 * The css to which this ID points. This pointer is set to valid value
124 * after cgroup is populated. If cgroup is removed, this will be NULL.
125 * This pointer is expected to be RCU-safe because destroy()
126 * is called after synchronize_rcu(). But for safe use, css_is_removed()
127 * css_tryget() should be used for avoiding race.
129 struct cgroup_subsys_state *css;
135 * Depth in hierarchy which this ID belongs to.
137 unsigned short depth;
139 * ID is freed by RCU. (and lookup routine is RCU safe.)
141 struct rcu_head rcu_head;
143 * Hierarchy of CSS ID belongs to.
145 unsigned short stack[0]; /* Array of Length (depth+1) */
149 /* The list of hierarchy roots */
151 static LIST_HEAD(roots);
152 static int root_count;
154 static DEFINE_IDA(hierarchy_ida);
155 static int next_hierarchy_id;
156 static DEFINE_SPINLOCK(hierarchy_id_lock);
158 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
159 #define dummytop (&rootnode.top_cgroup)
161 /* This flag indicates whether tasks in the fork and exit paths should
162 * check for fork/exit handlers to call. This avoids us having to do
163 * extra work in the fork/exit path if none of the subsystems need to
166 static int need_forkexit_callback __read_mostly;
168 /* convenient tests for these bits */
169 inline int cgroup_is_removed(const struct cgroup *cgrp)
171 return test_bit(CGRP_REMOVED, &cgrp->flags);
174 /* bits in struct cgroupfs_root flags field */
176 ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
179 static int cgroup_is_releasable(const struct cgroup *cgrp)
182 (1 << CGRP_RELEASABLE) |
183 (1 << CGRP_NOTIFY_ON_RELEASE);
184 return (cgrp->flags & bits) == bits;
187 static int notify_on_release(const struct cgroup *cgrp)
189 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
193 * for_each_subsys() allows you to iterate on each subsystem attached to
194 * an active hierarchy
196 #define for_each_subsys(_root, _ss) \
197 list_for_each_entry(_ss, &_root->subsys_list, sibling)
199 /* for_each_active_root() allows you to iterate across the active hierarchies */
200 #define for_each_active_root(_root) \
201 list_for_each_entry(_root, &roots, root_list)
203 /* the list of cgroups eligible for automatic release. Protected by
204 * release_list_lock */
205 static LIST_HEAD(release_list);
206 static DEFINE_SPINLOCK(release_list_lock);
207 static void cgroup_release_agent(struct work_struct *work);
208 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
209 static void check_for_release(struct cgroup *cgrp);
211 /* Link structure for associating css_set objects with cgroups */
212 struct cg_cgroup_link {
214 * List running through cg_cgroup_links associated with a
215 * cgroup, anchored on cgroup->css_sets
217 struct list_head cgrp_link_list;
220 * List running through cg_cgroup_links pointing at a
221 * single css_set object, anchored on css_set->cg_links
223 struct list_head cg_link_list;
227 /* The default css_set - used by init and its children prior to any
228 * hierarchies being mounted. It contains a pointer to the root state
229 * for each subsystem. Also used to anchor the list of css_sets. Not
230 * reference-counted, to improve performance when child cgroups
231 * haven't been created.
234 static struct css_set init_css_set;
235 static struct cg_cgroup_link init_css_set_link;
237 static int cgroup_subsys_init_idr(struct cgroup_subsys *ss);
239 /* css_set_lock protects the list of css_set objects, and the
240 * chain of tasks off each css_set. Nests outside task->alloc_lock
241 * due to cgroup_iter_start() */
242 static DEFINE_RWLOCK(css_set_lock);
243 static int css_set_count;
246 * hash table for cgroup groups. This improves the performance to find
247 * an existing css_set. This hash doesn't (currently) take into
248 * account cgroups in empty hierarchies.
250 #define CSS_SET_HASH_BITS 7
251 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
252 static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE];
254 static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[])
258 unsigned long tmp = 0UL;
260 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
261 tmp += (unsigned long)css[i];
262 tmp = (tmp >> 16) ^ tmp;
264 index = hash_long(tmp, CSS_SET_HASH_BITS);
266 return &css_set_table[index];
269 /* We don't maintain the lists running through each css_set to its
270 * task until after the first call to cgroup_iter_start(). This
271 * reduces the fork()/exit() overhead for people who have cgroups
272 * compiled into their kernel but not actually in use */
273 static int use_task_css_set_links __read_mostly;
275 static void __put_css_set(struct css_set *cg, int taskexit)
277 struct cg_cgroup_link *link;
278 struct cg_cgroup_link *saved_link;
280 * Ensure that the refcount doesn't hit zero while any readers
281 * can see it. Similar to atomic_dec_and_lock(), but for an
284 if (atomic_add_unless(&cg->refcount, -1, 1))
286 write_lock(&css_set_lock);
287 if (!atomic_dec_and_test(&cg->refcount)) {
288 write_unlock(&css_set_lock);
292 /* This css_set is dead. unlink it and release cgroup refcounts */
293 hlist_del(&cg->hlist);
296 list_for_each_entry_safe(link, saved_link, &cg->cg_links,
298 struct cgroup *cgrp = link->cgrp;
299 list_del(&link->cg_link_list);
300 list_del(&link->cgrp_link_list);
301 if (atomic_dec_and_test(&cgrp->count) &&
302 notify_on_release(cgrp)) {
304 set_bit(CGRP_RELEASABLE, &cgrp->flags);
305 check_for_release(cgrp);
311 write_unlock(&css_set_lock);
316 * refcounted get/put for css_set objects
318 static inline void get_css_set(struct css_set *cg)
320 atomic_inc(&cg->refcount);
323 static inline void put_css_set(struct css_set *cg)
325 __put_css_set(cg, 0);
328 static inline void put_css_set_taskexit(struct css_set *cg)
330 __put_css_set(cg, 1);
334 * compare_css_sets - helper function for find_existing_css_set().
335 * @cg: candidate css_set being tested
336 * @old_cg: existing css_set for a task
337 * @new_cgrp: cgroup that's being entered by the task
338 * @template: desired set of css pointers in css_set (pre-calculated)
340 * Returns true if "cg" matches "old_cg" except for the hierarchy
341 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
343 static bool compare_css_sets(struct css_set *cg,
344 struct css_set *old_cg,
345 struct cgroup *new_cgrp,
346 struct cgroup_subsys_state *template[])
348 struct list_head *l1, *l2;
350 if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
351 /* Not all subsystems matched */
356 * Compare cgroup pointers in order to distinguish between
357 * different cgroups in heirarchies with no subsystems. We
358 * could get by with just this check alone (and skip the
359 * memcmp above) but on most setups the memcmp check will
360 * avoid the need for this more expensive check on almost all
365 l2 = &old_cg->cg_links;
367 struct cg_cgroup_link *cgl1, *cgl2;
368 struct cgroup *cg1, *cg2;
372 /* See if we reached the end - both lists are equal length. */
373 if (l1 == &cg->cg_links) {
374 BUG_ON(l2 != &old_cg->cg_links);
377 BUG_ON(l2 == &old_cg->cg_links);
379 /* Locate the cgroups associated with these links. */
380 cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
381 cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
384 /* Hierarchies should be linked in the same order. */
385 BUG_ON(cg1->root != cg2->root);
388 * If this hierarchy is the hierarchy of the cgroup
389 * that's changing, then we need to check that this
390 * css_set points to the new cgroup; if it's any other
391 * hierarchy, then this css_set should point to the
392 * same cgroup as the old css_set.
394 if (cg1->root == new_cgrp->root) {
406 * find_existing_css_set() is a helper for
407 * find_css_set(), and checks to see whether an existing
408 * css_set is suitable.
410 * oldcg: the cgroup group that we're using before the cgroup
413 * cgrp: the cgroup that we're moving into
415 * template: location in which to build the desired set of subsystem
416 * state objects for the new cgroup group
418 static struct css_set *find_existing_css_set(
419 struct css_set *oldcg,
421 struct cgroup_subsys_state *template[])
424 struct cgroupfs_root *root = cgrp->root;
425 struct hlist_head *hhead;
426 struct hlist_node *node;
429 /* Built the set of subsystem state objects that we want to
430 * see in the new css_set */
431 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
432 if (root->subsys_bits & (1UL << i)) {
433 /* Subsystem is in this hierarchy. So we want
434 * the subsystem state from the new
436 template[i] = cgrp->subsys[i];
438 /* Subsystem is not in this hierarchy, so we
439 * don't want to change the subsystem state */
440 template[i] = oldcg->subsys[i];
444 hhead = css_set_hash(template);
445 hlist_for_each_entry(cg, node, hhead, hlist) {
446 if (!compare_css_sets(cg, oldcg, cgrp, template))
449 /* This css_set matches what we need */
453 /* No existing cgroup group matched */
457 static void free_cg_links(struct list_head *tmp)
459 struct cg_cgroup_link *link;
460 struct cg_cgroup_link *saved_link;
462 list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
463 list_del(&link->cgrp_link_list);
469 * allocate_cg_links() allocates "count" cg_cgroup_link structures
470 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
471 * success or a negative error
473 static int allocate_cg_links(int count, struct list_head *tmp)
475 struct cg_cgroup_link *link;
478 for (i = 0; i < count; i++) {
479 link = kmalloc(sizeof(*link), GFP_KERNEL);
484 list_add(&link->cgrp_link_list, tmp);
490 * link_css_set - a helper function to link a css_set to a cgroup
491 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
492 * @cg: the css_set to be linked
493 * @cgrp: the destination cgroup
495 static void link_css_set(struct list_head *tmp_cg_links,
496 struct css_set *cg, struct cgroup *cgrp)
498 struct cg_cgroup_link *link;
500 BUG_ON(list_empty(tmp_cg_links));
501 link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
505 atomic_inc(&cgrp->count);
506 list_move(&link->cgrp_link_list, &cgrp->css_sets);
508 * Always add links to the tail of the list so that the list
509 * is sorted by order of hierarchy creation
511 list_add_tail(&link->cg_link_list, &cg->cg_links);
515 * find_css_set() takes an existing cgroup group and a
516 * cgroup object, and returns a css_set object that's
517 * equivalent to the old group, but with the given cgroup
518 * substituted into the appropriate hierarchy. Must be called with
521 static struct css_set *find_css_set(
522 struct css_set *oldcg, struct cgroup *cgrp)
525 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
527 struct list_head tmp_cg_links;
529 struct hlist_head *hhead;
530 struct cg_cgroup_link *link;
532 /* First see if we already have a cgroup group that matches
534 read_lock(&css_set_lock);
535 res = find_existing_css_set(oldcg, cgrp, template);
538 read_unlock(&css_set_lock);
543 res = kmalloc(sizeof(*res), GFP_KERNEL);
547 /* Allocate all the cg_cgroup_link objects that we'll need */
548 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
553 atomic_set(&res->refcount, 1);
554 INIT_LIST_HEAD(&res->cg_links);
555 INIT_LIST_HEAD(&res->tasks);
556 INIT_HLIST_NODE(&res->hlist);
558 /* Copy the set of subsystem state objects generated in
559 * find_existing_css_set() */
560 memcpy(res->subsys, template, sizeof(res->subsys));
562 write_lock(&css_set_lock);
563 /* Add reference counts and links from the new css_set. */
564 list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
565 struct cgroup *c = link->cgrp;
566 if (c->root == cgrp->root)
568 link_css_set(&tmp_cg_links, res, c);
571 BUG_ON(!list_empty(&tmp_cg_links));
575 /* Add this cgroup group to the hash table */
576 hhead = css_set_hash(res->subsys);
577 hlist_add_head(&res->hlist, hhead);
579 write_unlock(&css_set_lock);
585 * Return the cgroup for "task" from the given hierarchy. Must be
586 * called with cgroup_mutex held.
588 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
589 struct cgroupfs_root *root)
592 struct cgroup *res = NULL;
594 BUG_ON(!mutex_is_locked(&cgroup_mutex));
595 read_lock(&css_set_lock);
597 * No need to lock the task - since we hold cgroup_mutex the
598 * task can't change groups, so the only thing that can happen
599 * is that it exits and its css is set back to init_css_set.
602 if (css == &init_css_set) {
603 res = &root->top_cgroup;
605 struct cg_cgroup_link *link;
606 list_for_each_entry(link, &css->cg_links, cg_link_list) {
607 struct cgroup *c = link->cgrp;
608 if (c->root == root) {
614 read_unlock(&css_set_lock);
620 * There is one global cgroup mutex. We also require taking
621 * task_lock() when dereferencing a task's cgroup subsys pointers.
622 * See "The task_lock() exception", at the end of this comment.
624 * A task must hold cgroup_mutex to modify cgroups.
626 * Any task can increment and decrement the count field without lock.
627 * So in general, code holding cgroup_mutex can't rely on the count
628 * field not changing. However, if the count goes to zero, then only
629 * cgroup_attach_task() can increment it again. Because a count of zero
630 * means that no tasks are currently attached, therefore there is no
631 * way a task attached to that cgroup can fork (the other way to
632 * increment the count). So code holding cgroup_mutex can safely
633 * assume that if the count is zero, it will stay zero. Similarly, if
634 * a task holds cgroup_mutex on a cgroup with zero count, it
635 * knows that the cgroup won't be removed, as cgroup_rmdir()
638 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
639 * (usually) take cgroup_mutex. These are the two most performance
640 * critical pieces of code here. The exception occurs on cgroup_exit(),
641 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
642 * is taken, and if the cgroup count is zero, a usermode call made
643 * to the release agent with the name of the cgroup (path relative to
644 * the root of cgroup file system) as the argument.
646 * A cgroup can only be deleted if both its 'count' of using tasks
647 * is zero, and its list of 'children' cgroups is empty. Since all
648 * tasks in the system use _some_ cgroup, and since there is always at
649 * least one task in the system (init, pid == 1), therefore, top_cgroup
650 * always has either children cgroups and/or using tasks. So we don't
651 * need a special hack to ensure that top_cgroup cannot be deleted.
653 * The task_lock() exception
655 * The need for this exception arises from the action of
656 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
657 * another. It does so using cgroup_mutex, however there are
658 * several performance critical places that need to reference
659 * task->cgroup without the expense of grabbing a system global
660 * mutex. Therefore except as noted below, when dereferencing or, as
661 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
662 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
663 * the task_struct routinely used for such matters.
665 * P.S. One more locking exception. RCU is used to guard the
666 * update of a tasks cgroup pointer by cgroup_attach_task()
670 * cgroup_lock - lock out any changes to cgroup structures
673 void cgroup_lock(void)
675 mutex_lock(&cgroup_mutex);
679 * cgroup_unlock - release lock on cgroup changes
681 * Undo the lock taken in a previous cgroup_lock() call.
683 void cgroup_unlock(void)
685 mutex_unlock(&cgroup_mutex);
689 * A couple of forward declarations required, due to cyclic reference loop:
690 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
691 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
695 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode);
696 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
697 static int cgroup_populate_dir(struct cgroup *cgrp);
698 static const struct inode_operations cgroup_dir_inode_operations;
699 static struct file_operations proc_cgroupstats_operations;
701 static struct backing_dev_info cgroup_backing_dev_info = {
703 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
706 static int alloc_css_id(struct cgroup_subsys *ss,
707 struct cgroup *parent, struct cgroup *child);
709 static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb)
711 struct inode *inode = new_inode(sb);
714 inode->i_mode = mode;
715 inode->i_uid = current_fsuid();
716 inode->i_gid = current_fsgid();
717 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
718 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
724 * Call subsys's pre_destroy handler.
725 * This is called before css refcnt check.
727 static int cgroup_call_pre_destroy(struct cgroup *cgrp)
729 struct cgroup_subsys *ss;
732 for_each_subsys(cgrp->root, ss)
733 if (ss->pre_destroy) {
734 ret = ss->pre_destroy(ss, cgrp);
741 static void free_cgroup_rcu(struct rcu_head *obj)
743 struct cgroup *cgrp = container_of(obj, struct cgroup, rcu_head);
748 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
750 /* is dentry a directory ? if so, kfree() associated cgroup */
751 if (S_ISDIR(inode->i_mode)) {
752 struct cgroup *cgrp = dentry->d_fsdata;
753 struct cgroup_subsys *ss;
754 BUG_ON(!(cgroup_is_removed(cgrp)));
755 /* It's possible for external users to be holding css
756 * reference counts on a cgroup; css_put() needs to
757 * be able to access the cgroup after decrementing
758 * the reference count in order to know if it needs to
759 * queue the cgroup to be handled by the release
763 mutex_lock(&cgroup_mutex);
765 * Release the subsystem state objects.
767 for_each_subsys(cgrp->root, ss)
768 ss->destroy(ss, cgrp);
770 cgrp->root->number_of_cgroups--;
771 mutex_unlock(&cgroup_mutex);
774 * Drop the active superblock reference that we took when we
777 deactivate_super(cgrp->root->sb);
779 call_rcu(&cgrp->rcu_head, free_cgroup_rcu);
784 static void remove_dir(struct dentry *d)
786 struct dentry *parent = dget(d->d_parent);
789 simple_rmdir(parent->d_inode, d);
793 static void cgroup_clear_directory(struct dentry *dentry)
795 struct list_head *node;
797 BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
798 spin_lock(&dcache_lock);
799 node = dentry->d_subdirs.next;
800 while (node != &dentry->d_subdirs) {
801 struct dentry *d = list_entry(node, struct dentry, d_u.d_child);
804 /* This should never be called on a cgroup
805 * directory with child cgroups */
806 BUG_ON(d->d_inode->i_mode & S_IFDIR);
808 spin_unlock(&dcache_lock);
810 simple_unlink(dentry->d_inode, d);
812 spin_lock(&dcache_lock);
814 node = dentry->d_subdirs.next;
816 spin_unlock(&dcache_lock);
820 * NOTE : the dentry must have been dget()'ed
822 static void cgroup_d_remove_dir(struct dentry *dentry)
824 cgroup_clear_directory(dentry);
826 spin_lock(&dcache_lock);
827 list_del_init(&dentry->d_u.d_child);
828 spin_unlock(&dcache_lock);
833 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
834 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
835 * reference to css->refcnt. In general, this refcnt is expected to goes down
838 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
840 DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq);
842 static void cgroup_wakeup_rmdir_waiter(struct cgroup *cgrp)
844 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)))
845 wake_up_all(&cgroup_rmdir_waitq);
848 void cgroup_exclude_rmdir(struct cgroup_subsys_state *css)
853 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css)
855 cgroup_wakeup_rmdir_waiter(css->cgroup);
860 static int rebind_subsystems(struct cgroupfs_root *root,
861 unsigned long final_bits)
863 unsigned long added_bits, removed_bits;
864 struct cgroup *cgrp = &root->top_cgroup;
867 removed_bits = root->actual_subsys_bits & ~final_bits;
868 added_bits = final_bits & ~root->actual_subsys_bits;
869 /* Check that any added subsystems are currently free */
870 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
871 unsigned long bit = 1UL << i;
872 struct cgroup_subsys *ss = subsys[i];
873 if (!(bit & added_bits))
875 if (ss->root != &rootnode) {
876 /* Subsystem isn't free */
881 /* Currently we don't handle adding/removing subsystems when
882 * any child cgroups exist. This is theoretically supportable
883 * but involves complex error handling, so it's being left until
885 if (root->number_of_cgroups > 1)
888 /* Process each subsystem */
889 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
890 struct cgroup_subsys *ss = subsys[i];
891 unsigned long bit = 1UL << i;
892 if (bit & added_bits) {
893 /* We're binding this subsystem to this hierarchy */
894 BUG_ON(cgrp->subsys[i]);
895 BUG_ON(!dummytop->subsys[i]);
896 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
897 mutex_lock(&ss->hierarchy_mutex);
898 cgrp->subsys[i] = dummytop->subsys[i];
899 cgrp->subsys[i]->cgroup = cgrp;
900 list_move(&ss->sibling, &root->subsys_list);
904 mutex_unlock(&ss->hierarchy_mutex);
905 } else if (bit & removed_bits) {
906 /* We're removing this subsystem */
907 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
908 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
909 mutex_lock(&ss->hierarchy_mutex);
911 ss->bind(ss, dummytop);
912 dummytop->subsys[i]->cgroup = dummytop;
913 cgrp->subsys[i] = NULL;
914 subsys[i]->root = &rootnode;
915 list_move(&ss->sibling, &rootnode.subsys_list);
916 mutex_unlock(&ss->hierarchy_mutex);
917 } else if (bit & final_bits) {
918 /* Subsystem state should already exist */
919 BUG_ON(!cgrp->subsys[i]);
921 /* Subsystem state shouldn't exist */
922 BUG_ON(cgrp->subsys[i]);
925 root->subsys_bits = root->actual_subsys_bits = final_bits;
931 static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs)
933 struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info;
934 struct cgroup_subsys *ss;
936 mutex_lock(&cgroup_mutex);
937 for_each_subsys(root, ss)
938 seq_printf(seq, ",%s", ss->name);
939 if (test_bit(ROOT_NOPREFIX, &root->flags))
940 seq_puts(seq, ",noprefix");
941 if (strlen(root->release_agent_path))
942 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
943 if (strlen(root->name))
944 seq_printf(seq, ",name=%s", root->name);
945 mutex_unlock(&cgroup_mutex);
949 struct cgroup_sb_opts {
950 unsigned long subsys_bits;
954 /* User explicitly requested empty subsystem */
957 struct cgroupfs_root *new_root;
961 /* Convert a hierarchy specifier into a bitmask of subsystems and
963 static int parse_cgroupfs_options(char *data,
964 struct cgroup_sb_opts *opts)
966 char *token, *o = data ?: "all";
967 unsigned long mask = (unsigned long)-1;
969 #ifdef CONFIG_CPUSETS
970 mask = ~(1UL << cpuset_subsys_id);
973 memset(opts, 0, sizeof(*opts));
975 while ((token = strsep(&o, ",")) != NULL) {
978 if (!strcmp(token, "all")) {
979 /* Add all non-disabled subsystems */
981 opts->subsys_bits = 0;
982 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
983 struct cgroup_subsys *ss = subsys[i];
985 opts->subsys_bits |= 1ul << i;
987 } else if (!strcmp(token, "none")) {
988 /* Explicitly have no subsystems */
990 } else if (!strcmp(token, "noprefix")) {
991 set_bit(ROOT_NOPREFIX, &opts->flags);
992 } else if (!strncmp(token, "release_agent=", 14)) {
993 /* Specifying two release agents is forbidden */
994 if (opts->release_agent)
996 opts->release_agent =
997 kstrndup(token + 14, PATH_MAX, GFP_KERNEL);
998 if (!opts->release_agent)
1000 } else if (!strncmp(token, "name=", 5)) {
1002 const char *name = token + 5;
1003 /* Can't specify an empty name */
1006 /* Must match [\w.-]+ */
1007 for (i = 0; i < strlen(name); i++) {
1011 if ((c == '.') || (c == '-') || (c == '_'))
1015 /* Specifying two names is forbidden */
1018 opts->name = kstrndup(name,
1019 MAX_CGROUP_ROOT_NAMELEN,
1024 struct cgroup_subsys *ss;
1026 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1028 if (!strcmp(token, ss->name)) {
1030 set_bit(i, &opts->subsys_bits);
1034 if (i == CGROUP_SUBSYS_COUNT)
1039 /* Consistency checks */
1042 * Option noprefix was introduced just for backward compatibility
1043 * with the old cpuset, so we allow noprefix only if mounting just
1044 * the cpuset subsystem.
1046 if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1047 (opts->subsys_bits & mask))
1051 /* Can't specify "none" and some subsystems */
1052 if (opts->subsys_bits && opts->none)
1056 * We either have to specify by name or by subsystems. (So all
1057 * empty hierarchies must have a name).
1059 if (!opts->subsys_bits && !opts->name)
1065 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1068 struct cgroupfs_root *root = sb->s_fs_info;
1069 struct cgroup *cgrp = &root->top_cgroup;
1070 struct cgroup_sb_opts opts;
1073 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1074 mutex_lock(&cgroup_mutex);
1076 /* See what subsystems are wanted */
1077 ret = parse_cgroupfs_options(data, &opts);
1081 /* Don't allow flags to change at remount */
1082 if (opts.flags != root->flags) {
1087 /* Don't allow name to change at remount */
1088 if (opts.name && strcmp(opts.name, root->name)) {
1093 ret = rebind_subsystems(root, opts.subsys_bits);
1097 /* (re)populate subsystem files */
1098 cgroup_populate_dir(cgrp);
1100 if (opts.release_agent)
1101 strcpy(root->release_agent_path, opts.release_agent);
1103 kfree(opts.release_agent);
1105 mutex_unlock(&cgroup_mutex);
1106 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1111 static const struct super_operations cgroup_ops = {
1112 .statfs = simple_statfs,
1113 .drop_inode = generic_delete_inode,
1114 .show_options = cgroup_show_options,
1115 .remount_fs = cgroup_remount,
1118 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1120 INIT_LIST_HEAD(&cgrp->sibling);
1121 INIT_LIST_HEAD(&cgrp->children);
1122 INIT_LIST_HEAD(&cgrp->css_sets);
1123 INIT_LIST_HEAD(&cgrp->release_list);
1124 init_rwsem(&(cgrp->tasks.mutex));
1125 init_rwsem(&(cgrp->procs.mutex));
1128 static void init_cgroup_root(struct cgroupfs_root *root)
1130 struct cgroup *cgrp = &root->top_cgroup;
1131 INIT_LIST_HEAD(&root->subsys_list);
1132 INIT_LIST_HEAD(&root->root_list);
1133 root->number_of_cgroups = 1;
1135 cgrp->top_cgroup = cgrp;
1136 init_cgroup_housekeeping(cgrp);
1139 static bool init_root_id(struct cgroupfs_root *root)
1144 if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1146 spin_lock(&hierarchy_id_lock);
1147 /* Try to allocate the next unused ID */
1148 ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1149 &root->hierarchy_id);
1151 /* Try again starting from 0 */
1152 ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1154 next_hierarchy_id = root->hierarchy_id + 1;
1155 } else if (ret != -EAGAIN) {
1156 /* Can only get here if the 31-bit IDR is full ... */
1159 spin_unlock(&hierarchy_id_lock);
1164 static int cgroup_test_super(struct super_block *sb, void *data)
1166 struct cgroup_sb_opts *opts = data;
1167 struct cgroupfs_root *root = sb->s_fs_info;
1169 /* If we asked for a name then it must match */
1170 if (opts->name && strcmp(opts->name, root->name))
1174 * If we asked for subsystems (or explicitly for no
1175 * subsystems) then they must match
1177 if ((opts->subsys_bits || opts->none)
1178 && (opts->subsys_bits != root->subsys_bits))
1184 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1186 struct cgroupfs_root *root;
1188 if (!opts->subsys_bits && !opts->none)
1191 root = kzalloc(sizeof(*root), GFP_KERNEL);
1193 return ERR_PTR(-ENOMEM);
1195 if (!init_root_id(root)) {
1197 return ERR_PTR(-ENOMEM);
1199 init_cgroup_root(root);
1201 root->subsys_bits = opts->subsys_bits;
1202 root->flags = opts->flags;
1203 if (opts->release_agent)
1204 strcpy(root->release_agent_path, opts->release_agent);
1206 strcpy(root->name, opts->name);
1210 static void cgroup_drop_root(struct cgroupfs_root *root)
1215 BUG_ON(!root->hierarchy_id);
1216 spin_lock(&hierarchy_id_lock);
1217 ida_remove(&hierarchy_ida, root->hierarchy_id);
1218 spin_unlock(&hierarchy_id_lock);
1222 static int cgroup_set_super(struct super_block *sb, void *data)
1225 struct cgroup_sb_opts *opts = data;
1227 /* If we don't have a new root, we can't set up a new sb */
1228 if (!opts->new_root)
1231 BUG_ON(!opts->subsys_bits && !opts->none);
1233 ret = set_anon_super(sb, NULL);
1237 sb->s_fs_info = opts->new_root;
1238 opts->new_root->sb = sb;
1240 sb->s_blocksize = PAGE_CACHE_SIZE;
1241 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1242 sb->s_magic = CGROUP_SUPER_MAGIC;
1243 sb->s_op = &cgroup_ops;
1248 static int cgroup_get_rootdir(struct super_block *sb)
1250 struct inode *inode =
1251 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1252 struct dentry *dentry;
1257 inode->i_fop = &simple_dir_operations;
1258 inode->i_op = &cgroup_dir_inode_operations;
1259 /* directories start off with i_nlink == 2 (for "." entry) */
1261 dentry = d_alloc_root(inode);
1266 sb->s_root = dentry;
1270 static int cgroup_get_sb(struct file_system_type *fs_type,
1271 int flags, const char *unused_dev_name,
1272 void *data, struct vfsmount *mnt)
1274 struct cgroup_sb_opts opts;
1275 struct cgroupfs_root *root;
1277 struct super_block *sb;
1278 struct cgroupfs_root *new_root;
1280 /* First find the desired set of subsystems */
1281 ret = parse_cgroupfs_options(data, &opts);
1286 * Allocate a new cgroup root. We may not need it if we're
1287 * reusing an existing hierarchy.
1289 new_root = cgroup_root_from_opts(&opts);
1290 if (IS_ERR(new_root)) {
1291 ret = PTR_ERR(new_root);
1294 opts.new_root = new_root;
1296 /* Locate an existing or new sb for this hierarchy */
1297 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, &opts);
1300 cgroup_drop_root(opts.new_root);
1304 root = sb->s_fs_info;
1306 if (root == opts.new_root) {
1307 /* We used the new root structure, so this is a new hierarchy */
1308 struct list_head tmp_cg_links;
1309 struct cgroup *root_cgrp = &root->top_cgroup;
1310 struct inode *inode;
1311 struct cgroupfs_root *existing_root;
1314 BUG_ON(sb->s_root != NULL);
1316 ret = cgroup_get_rootdir(sb);
1318 goto drop_new_super;
1319 inode = sb->s_root->d_inode;
1321 mutex_lock(&inode->i_mutex);
1322 mutex_lock(&cgroup_mutex);
1324 if (strlen(root->name)) {
1325 /* Check for name clashes with existing mounts */
1326 for_each_active_root(existing_root) {
1327 if (!strcmp(existing_root->name, root->name)) {
1329 mutex_unlock(&cgroup_mutex);
1330 mutex_unlock(&inode->i_mutex);
1331 goto drop_new_super;
1337 * We're accessing css_set_count without locking
1338 * css_set_lock here, but that's OK - it can only be
1339 * increased by someone holding cgroup_lock, and
1340 * that's us. The worst that can happen is that we
1341 * have some link structures left over
1343 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1345 mutex_unlock(&cgroup_mutex);
1346 mutex_unlock(&inode->i_mutex);
1347 goto drop_new_super;
1350 ret = rebind_subsystems(root, root->subsys_bits);
1351 if (ret == -EBUSY) {
1352 mutex_unlock(&cgroup_mutex);
1353 mutex_unlock(&inode->i_mutex);
1354 free_cg_links(&tmp_cg_links);
1355 goto drop_new_super;
1358 /* EBUSY should be the only error here */
1361 list_add(&root->root_list, &roots);
1364 sb->s_root->d_fsdata = root_cgrp;
1365 root->top_cgroup.dentry = sb->s_root;
1367 /* Link the top cgroup in this hierarchy into all
1368 * the css_set objects */
1369 write_lock(&css_set_lock);
1370 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
1371 struct hlist_head *hhead = &css_set_table[i];
1372 struct hlist_node *node;
1375 hlist_for_each_entry(cg, node, hhead, hlist)
1376 link_css_set(&tmp_cg_links, cg, root_cgrp);
1378 write_unlock(&css_set_lock);
1380 free_cg_links(&tmp_cg_links);
1382 BUG_ON(!list_empty(&root_cgrp->sibling));
1383 BUG_ON(!list_empty(&root_cgrp->children));
1384 BUG_ON(root->number_of_cgroups != 1);
1386 cgroup_populate_dir(root_cgrp);
1387 mutex_unlock(&cgroup_mutex);
1388 mutex_unlock(&inode->i_mutex);
1391 * We re-used an existing hierarchy - the new root (if
1392 * any) is not needed
1394 cgroup_drop_root(opts.new_root);
1397 simple_set_mnt(mnt, sb);
1398 kfree(opts.release_agent);
1403 deactivate_locked_super(sb);
1405 kfree(opts.release_agent);
1411 static void cgroup_kill_sb(struct super_block *sb) {
1412 struct cgroupfs_root *root = sb->s_fs_info;
1413 struct cgroup *cgrp = &root->top_cgroup;
1415 struct cg_cgroup_link *link;
1416 struct cg_cgroup_link *saved_link;
1420 BUG_ON(root->number_of_cgroups != 1);
1421 BUG_ON(!list_empty(&cgrp->children));
1422 BUG_ON(!list_empty(&cgrp->sibling));
1424 mutex_lock(&cgroup_mutex);
1426 /* Rebind all subsystems back to the default hierarchy */
1427 ret = rebind_subsystems(root, 0);
1428 /* Shouldn't be able to fail ... */
1432 * Release all the links from css_sets to this hierarchy's
1435 write_lock(&css_set_lock);
1437 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1439 list_del(&link->cg_link_list);
1440 list_del(&link->cgrp_link_list);
1443 write_unlock(&css_set_lock);
1445 if (!list_empty(&root->root_list)) {
1446 list_del(&root->root_list);
1450 mutex_unlock(&cgroup_mutex);
1452 kill_litter_super(sb);
1453 cgroup_drop_root(root);
1456 static struct file_system_type cgroup_fs_type = {
1458 .get_sb = cgroup_get_sb,
1459 .kill_sb = cgroup_kill_sb,
1462 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
1464 return dentry->d_fsdata;
1467 static inline struct cftype *__d_cft(struct dentry *dentry)
1469 return dentry->d_fsdata;
1473 * cgroup_path - generate the path of a cgroup
1474 * @cgrp: the cgroup in question
1475 * @buf: the buffer to write the path into
1476 * @buflen: the length of the buffer
1478 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1479 * reference. Writes path of cgroup into buf. Returns 0 on success,
1482 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1485 struct dentry *dentry = rcu_dereference(cgrp->dentry);
1487 if (!dentry || cgrp == dummytop) {
1489 * Inactive subsystems have no dentry for their root
1496 start = buf + buflen;
1500 int len = dentry->d_name.len;
1501 if ((start -= len) < buf)
1502 return -ENAMETOOLONG;
1503 memcpy(start, cgrp->dentry->d_name.name, len);
1504 cgrp = cgrp->parent;
1507 dentry = rcu_dereference(cgrp->dentry);
1511 return -ENAMETOOLONG;
1514 memmove(buf, start, buf + buflen - start);
1519 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1520 * @cgrp: the cgroup the task is attaching to
1521 * @tsk: the task to be attached
1523 * Call holding cgroup_mutex. May take task_lock of
1524 * the task 'tsk' during call.
1526 int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1529 struct cgroup_subsys *ss;
1530 struct cgroup *oldcgrp;
1532 struct css_set *newcg;
1533 struct cgroupfs_root *root = cgrp->root;
1535 /* Nothing to do if the task is already in that cgroup */
1536 oldcgrp = task_cgroup_from_root(tsk, root);
1537 if (cgrp == oldcgrp)
1540 for_each_subsys(root, ss) {
1541 if (ss->can_attach) {
1542 retval = ss->can_attach(ss, cgrp, tsk);
1553 * Locate or allocate a new css_set for this task,
1554 * based on its final set of cgroups
1556 newcg = find_css_set(cg, cgrp);
1562 if (tsk->flags & PF_EXITING) {
1567 rcu_assign_pointer(tsk->cgroups, newcg);
1570 /* Update the css_set linked lists if we're using them */
1571 write_lock(&css_set_lock);
1572 if (!list_empty(&tsk->cg_list)) {
1573 list_del(&tsk->cg_list);
1574 list_add(&tsk->cg_list, &newcg->tasks);
1576 write_unlock(&css_set_lock);
1578 for_each_subsys(root, ss) {
1580 ss->attach(ss, cgrp, oldcgrp, tsk);
1582 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1587 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1588 * is no longer empty.
1590 cgroup_wakeup_rmdir_waiter(cgrp);
1595 * Attach task with pid 'pid' to cgroup 'cgrp'. Call with cgroup_mutex
1596 * held. May take task_lock of task
1598 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid)
1600 struct task_struct *tsk;
1601 const struct cred *cred = current_cred(), *tcred;
1606 tsk = find_task_by_vpid(pid);
1607 if (!tsk || tsk->flags & PF_EXITING) {
1612 tcred = __task_cred(tsk);
1614 cred->euid != tcred->uid &&
1615 cred->euid != tcred->suid) {
1619 get_task_struct(tsk);
1623 get_task_struct(tsk);
1626 ret = cgroup_attach_task(cgrp, tsk);
1627 put_task_struct(tsk);
1631 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
1634 if (!cgroup_lock_live_group(cgrp))
1636 ret = attach_task_by_pid(cgrp, pid);
1642 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
1643 * @cgrp: the cgroup to be checked for liveness
1645 * On success, returns true; the lock should be later released with
1646 * cgroup_unlock(). On failure returns false with no lock held.
1648 bool cgroup_lock_live_group(struct cgroup *cgrp)
1650 mutex_lock(&cgroup_mutex);
1651 if (cgroup_is_removed(cgrp)) {
1652 mutex_unlock(&cgroup_mutex);
1658 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
1661 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
1662 if (!cgroup_lock_live_group(cgrp))
1664 strcpy(cgrp->root->release_agent_path, buffer);
1669 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
1670 struct seq_file *seq)
1672 if (!cgroup_lock_live_group(cgrp))
1674 seq_puts(seq, cgrp->root->release_agent_path);
1675 seq_putc(seq, '\n');
1680 /* A buffer size big enough for numbers or short strings */
1681 #define CGROUP_LOCAL_BUFFER_SIZE 64
1683 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
1685 const char __user *userbuf,
1686 size_t nbytes, loff_t *unused_ppos)
1688 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
1694 if (nbytes >= sizeof(buffer))
1696 if (copy_from_user(buffer, userbuf, nbytes))
1699 buffer[nbytes] = 0; /* nul-terminate */
1701 if (cft->write_u64) {
1702 u64 val = simple_strtoull(buffer, &end, 0);
1705 retval = cft->write_u64(cgrp, cft, val);
1707 s64 val = simple_strtoll(buffer, &end, 0);
1710 retval = cft->write_s64(cgrp, cft, val);
1717 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
1719 const char __user *userbuf,
1720 size_t nbytes, loff_t *unused_ppos)
1722 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
1724 size_t max_bytes = cft->max_write_len;
1725 char *buffer = local_buffer;
1728 max_bytes = sizeof(local_buffer) - 1;
1729 if (nbytes >= max_bytes)
1731 /* Allocate a dynamic buffer if we need one */
1732 if (nbytes >= sizeof(local_buffer)) {
1733 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
1737 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
1742 buffer[nbytes] = 0; /* nul-terminate */
1744 retval = cft->write_string(cgrp, cft, buffer);
1748 if (buffer != local_buffer)
1753 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
1754 size_t nbytes, loff_t *ppos)
1756 struct cftype *cft = __d_cft(file->f_dentry);
1757 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
1759 if (cgroup_is_removed(cgrp))
1762 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
1763 if (cft->write_u64 || cft->write_s64)
1764 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
1765 if (cft->write_string)
1766 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
1768 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
1769 return ret ? ret : nbytes;
1774 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
1776 char __user *buf, size_t nbytes,
1779 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
1780 u64 val = cft->read_u64(cgrp, cft);
1781 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
1783 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
1786 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
1788 char __user *buf, size_t nbytes,
1791 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
1792 s64 val = cft->read_s64(cgrp, cft);
1793 int len = sprintf(tmp, "%lld\n", (long long) val);
1795 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
1798 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
1799 size_t nbytes, loff_t *ppos)
1801 struct cftype *cft = __d_cft(file->f_dentry);
1802 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
1804 if (cgroup_is_removed(cgrp))
1808 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
1810 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
1812 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
1817 * seqfile ops/methods for returning structured data. Currently just
1818 * supports string->u64 maps, but can be extended in future.
1821 struct cgroup_seqfile_state {
1823 struct cgroup *cgroup;
1826 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
1828 struct seq_file *sf = cb->state;
1829 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
1832 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
1834 struct cgroup_seqfile_state *state = m->private;
1835 struct cftype *cft = state->cft;
1836 if (cft->read_map) {
1837 struct cgroup_map_cb cb = {
1838 .fill = cgroup_map_add,
1841 return cft->read_map(state->cgroup, cft, &cb);
1843 return cft->read_seq_string(state->cgroup, cft, m);
1846 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
1848 struct seq_file *seq = file->private_data;
1849 kfree(seq->private);
1850 return single_release(inode, file);
1853 static struct file_operations cgroup_seqfile_operations = {
1855 .write = cgroup_file_write,
1856 .llseek = seq_lseek,
1857 .release = cgroup_seqfile_release,
1860 static int cgroup_file_open(struct inode *inode, struct file *file)
1865 err = generic_file_open(inode, file);
1868 cft = __d_cft(file->f_dentry);
1870 if (cft->read_map || cft->read_seq_string) {
1871 struct cgroup_seqfile_state *state =
1872 kzalloc(sizeof(*state), GFP_USER);
1876 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
1877 file->f_op = &cgroup_seqfile_operations;
1878 err = single_open(file, cgroup_seqfile_show, state);
1881 } else if (cft->open)
1882 err = cft->open(inode, file);
1889 static int cgroup_file_release(struct inode *inode, struct file *file)
1891 struct cftype *cft = __d_cft(file->f_dentry);
1893 return cft->release(inode, file);
1898 * cgroup_rename - Only allow simple rename of directories in place.
1900 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
1901 struct inode *new_dir, struct dentry *new_dentry)
1903 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1905 if (new_dentry->d_inode)
1907 if (old_dir != new_dir)
1909 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
1912 static struct file_operations cgroup_file_operations = {
1913 .read = cgroup_file_read,
1914 .write = cgroup_file_write,
1915 .llseek = generic_file_llseek,
1916 .open = cgroup_file_open,
1917 .release = cgroup_file_release,
1920 static const struct inode_operations cgroup_dir_inode_operations = {
1921 .lookup = simple_lookup,
1922 .mkdir = cgroup_mkdir,
1923 .rmdir = cgroup_rmdir,
1924 .rename = cgroup_rename,
1927 static int cgroup_create_file(struct dentry *dentry, mode_t mode,
1928 struct super_block *sb)
1930 static const struct dentry_operations cgroup_dops = {
1931 .d_iput = cgroup_diput,
1934 struct inode *inode;
1938 if (dentry->d_inode)
1941 inode = cgroup_new_inode(mode, sb);
1945 if (S_ISDIR(mode)) {
1946 inode->i_op = &cgroup_dir_inode_operations;
1947 inode->i_fop = &simple_dir_operations;
1949 /* start off with i_nlink == 2 (for "." entry) */
1952 /* start with the directory inode held, so that we can
1953 * populate it without racing with another mkdir */
1954 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1955 } else if (S_ISREG(mode)) {
1957 inode->i_fop = &cgroup_file_operations;
1959 dentry->d_op = &cgroup_dops;
1960 d_instantiate(dentry, inode);
1961 dget(dentry); /* Extra count - pin the dentry in core */
1966 * cgroup_create_dir - create a directory for an object.
1967 * @cgrp: the cgroup we create the directory for. It must have a valid
1968 * ->parent field. And we are going to fill its ->dentry field.
1969 * @dentry: dentry of the new cgroup
1970 * @mode: mode to set on new directory.
1972 static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
1975 struct dentry *parent;
1978 parent = cgrp->parent->dentry;
1979 error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
1981 dentry->d_fsdata = cgrp;
1982 inc_nlink(parent->d_inode);
1983 rcu_assign_pointer(cgrp->dentry, dentry);
1992 * cgroup_file_mode - deduce file mode of a control file
1993 * @cft: the control file in question
1995 * returns cft->mode if ->mode is not 0
1996 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
1997 * returns S_IRUGO if it has only a read handler
1998 * returns S_IWUSR if it has only a write hander
2000 static mode_t cgroup_file_mode(const struct cftype *cft)
2007 if (cft->read || cft->read_u64 || cft->read_s64 ||
2008 cft->read_map || cft->read_seq_string)
2011 if (cft->write || cft->write_u64 || cft->write_s64 ||
2012 cft->write_string || cft->trigger)
2018 int cgroup_add_file(struct cgroup *cgrp,
2019 struct cgroup_subsys *subsys,
2020 const struct cftype *cft)
2022 struct dentry *dir = cgrp->dentry;
2023 struct dentry *dentry;
2027 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2028 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
2029 strcpy(name, subsys->name);
2032 strcat(name, cft->name);
2033 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2034 dentry = lookup_one_len(name, dir, strlen(name));
2035 if (!IS_ERR(dentry)) {
2036 mode = cgroup_file_mode(cft);
2037 error = cgroup_create_file(dentry, mode | S_IFREG,
2040 dentry->d_fsdata = (void *)cft;
2043 error = PTR_ERR(dentry);
2047 int cgroup_add_files(struct cgroup *cgrp,
2048 struct cgroup_subsys *subsys,
2049 const struct cftype cft[],
2053 for (i = 0; i < count; i++) {
2054 err = cgroup_add_file(cgrp, subsys, &cft[i]);
2062 * cgroup_task_count - count the number of tasks in a cgroup.
2063 * @cgrp: the cgroup in question
2065 * Return the number of tasks in the cgroup.
2067 int cgroup_task_count(const struct cgroup *cgrp)
2070 struct cg_cgroup_link *link;
2072 read_lock(&css_set_lock);
2073 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2074 count += atomic_read(&link->cg->refcount);
2076 read_unlock(&css_set_lock);
2081 * Advance a list_head iterator. The iterator should be positioned at
2082 * the start of a css_set
2084 static void cgroup_advance_iter(struct cgroup *cgrp,
2085 struct cgroup_iter *it)
2087 struct list_head *l = it->cg_link;
2088 struct cg_cgroup_link *link;
2091 /* Advance to the next non-empty css_set */
2094 if (l == &cgrp->css_sets) {
2098 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2100 } while (list_empty(&cg->tasks));
2102 it->task = cg->tasks.next;
2106 * To reduce the fork() overhead for systems that are not actually
2107 * using their cgroups capability, we don't maintain the lists running
2108 * through each css_set to its tasks until we see the list actually
2109 * used - in other words after the first call to cgroup_iter_start().
2111 * The tasklist_lock is not held here, as do_each_thread() and
2112 * while_each_thread() are protected by RCU.
2114 static void cgroup_enable_task_cg_lists(void)
2116 struct task_struct *p, *g;
2117 write_lock(&css_set_lock);
2118 use_task_css_set_links = 1;
2119 do_each_thread(g, p) {
2122 * We should check if the process is exiting, otherwise
2123 * it will race with cgroup_exit() in that the list
2124 * entry won't be deleted though the process has exited.
2126 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2127 list_add(&p->cg_list, &p->cgroups->tasks);
2129 } while_each_thread(g, p);
2130 write_unlock(&css_set_lock);
2133 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
2136 * The first time anyone tries to iterate across a cgroup,
2137 * we need to enable the list linking each css_set to its
2138 * tasks, and fix up all existing tasks.
2140 if (!use_task_css_set_links)
2141 cgroup_enable_task_cg_lists();
2143 read_lock(&css_set_lock);
2144 it->cg_link = &cgrp->css_sets;
2145 cgroup_advance_iter(cgrp, it);
2148 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
2149 struct cgroup_iter *it)
2151 struct task_struct *res;
2152 struct list_head *l = it->task;
2153 struct cg_cgroup_link *link;
2155 /* If the iterator cg is NULL, we have no tasks */
2158 res = list_entry(l, struct task_struct, cg_list);
2159 /* Advance iterator to find next entry */
2161 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
2162 if (l == &link->cg->tasks) {
2163 /* We reached the end of this task list - move on to
2164 * the next cg_cgroup_link */
2165 cgroup_advance_iter(cgrp, it);
2172 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
2174 read_unlock(&css_set_lock);
2177 static inline int started_after_time(struct task_struct *t1,
2178 struct timespec *time,
2179 struct task_struct *t2)
2181 int start_diff = timespec_compare(&t1->start_time, time);
2182 if (start_diff > 0) {
2184 } else if (start_diff < 0) {
2188 * Arbitrarily, if two processes started at the same
2189 * time, we'll say that the lower pointer value
2190 * started first. Note that t2 may have exited by now
2191 * so this may not be a valid pointer any longer, but
2192 * that's fine - it still serves to distinguish
2193 * between two tasks started (effectively) simultaneously.
2200 * This function is a callback from heap_insert() and is used to order
2202 * In this case we order the heap in descending task start time.
2204 static inline int started_after(void *p1, void *p2)
2206 struct task_struct *t1 = p1;
2207 struct task_struct *t2 = p2;
2208 return started_after_time(t1, &t2->start_time, t2);
2212 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
2213 * @scan: struct cgroup_scanner containing arguments for the scan
2215 * Arguments include pointers to callback functions test_task() and
2217 * Iterate through all the tasks in a cgroup, calling test_task() for each,
2218 * and if it returns true, call process_task() for it also.
2219 * The test_task pointer may be NULL, meaning always true (select all tasks).
2220 * Effectively duplicates cgroup_iter_{start,next,end}()
2221 * but does not lock css_set_lock for the call to process_task().
2222 * The struct cgroup_scanner may be embedded in any structure of the caller's
2224 * It is guaranteed that process_task() will act on every task that
2225 * is a member of the cgroup for the duration of this call. This
2226 * function may or may not call process_task() for tasks that exit
2227 * or move to a different cgroup during the call, or are forked or
2228 * move into the cgroup during the call.
2230 * Note that test_task() may be called with locks held, and may in some
2231 * situations be called multiple times for the same task, so it should
2233 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
2234 * pre-allocated and will be used for heap operations (and its "gt" member will
2235 * be overwritten), else a temporary heap will be used (allocation of which
2236 * may cause this function to fail).
2238 int cgroup_scan_tasks(struct cgroup_scanner *scan)
2241 struct cgroup_iter it;
2242 struct task_struct *p, *dropped;
2243 /* Never dereference latest_task, since it's not refcounted */
2244 struct task_struct *latest_task = NULL;
2245 struct ptr_heap tmp_heap;
2246 struct ptr_heap *heap;
2247 struct timespec latest_time = { 0, 0 };
2250 /* The caller supplied our heap and pre-allocated its memory */
2252 heap->gt = &started_after;
2254 /* We need to allocate our own heap memory */
2256 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
2258 /* cannot allocate the heap */
2264 * Scan tasks in the cgroup, using the scanner's "test_task" callback
2265 * to determine which are of interest, and using the scanner's
2266 * "process_task" callback to process any of them that need an update.
2267 * Since we don't want to hold any locks during the task updates,
2268 * gather tasks to be processed in a heap structure.
2269 * The heap is sorted by descending task start time.
2270 * If the statically-sized heap fills up, we overflow tasks that
2271 * started later, and in future iterations only consider tasks that
2272 * started after the latest task in the previous pass. This
2273 * guarantees forward progress and that we don't miss any tasks.
2276 cgroup_iter_start(scan->cg, &it);
2277 while ((p = cgroup_iter_next(scan->cg, &it))) {
2279 * Only affect tasks that qualify per the caller's callback,
2280 * if he provided one
2282 if (scan->test_task && !scan->test_task(p, scan))
2285 * Only process tasks that started after the last task
2288 if (!started_after_time(p, &latest_time, latest_task))
2290 dropped = heap_insert(heap, p);
2291 if (dropped == NULL) {
2293 * The new task was inserted; the heap wasn't
2297 } else if (dropped != p) {
2299 * The new task was inserted, and pushed out a
2303 put_task_struct(dropped);
2306 * Else the new task was newer than anything already in
2307 * the heap and wasn't inserted
2310 cgroup_iter_end(scan->cg, &it);
2313 for (i = 0; i < heap->size; i++) {
2314 struct task_struct *q = heap->ptrs[i];
2316 latest_time = q->start_time;
2319 /* Process the task per the caller's callback */
2320 scan->process_task(q, scan);
2324 * If we had to process any tasks at all, scan again
2325 * in case some of them were in the middle of forking
2326 * children that didn't get processed.
2327 * Not the most efficient way to do it, but it avoids
2328 * having to take callback_mutex in the fork path
2332 if (heap == &tmp_heap)
2333 heap_free(&tmp_heap);
2338 * Stuff for reading the 'tasks'/'procs' files.
2340 * Reading this file can return large amounts of data if a cgroup has
2341 * *lots* of attached tasks. So it may need several calls to read(),
2342 * but we cannot guarantee that the information we produce is correct
2343 * unless we produce it entirely atomically.
2348 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
2349 * If the new stripped list is sufficiently smaller and there's enough memory
2350 * to allocate a new buffer, will let go of the unneeded memory. Returns the
2351 * number of unique elements.
2353 /* is the size difference enough that we should re-allocate the array? */
2354 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
2355 static int pidlist_uniq(pid_t **p, int length)
2362 * we presume the 0th element is unique, so i starts at 1. trivial
2363 * edge cases first; no work needs to be done for either
2365 if (length == 0 || length == 1)
2367 /* src and dest walk down the list; dest counts unique elements */
2368 for (src = 1; src < length; src++) {
2369 /* find next unique element */
2370 while (list[src] == list[src-1]) {
2375 /* dest always points to where the next unique element goes */
2376 list[dest] = list[src];
2381 * if the length difference is large enough, we want to allocate a
2382 * smaller buffer to save memory. if this fails due to out of memory,
2383 * we'll just stay with what we've got.
2385 if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) {
2386 newlist = krealloc(list, dest * sizeof(pid_t), GFP_KERNEL);
2393 static int cmppid(const void *a, const void *b)
2395 return *(pid_t *)a - *(pid_t *)b;
2399 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
2401 static int pidlist_array_load(struct cgroup *cgrp, bool procs)
2405 int pid, n = 0; /* used for populating the array */
2406 struct cgroup_iter it;
2407 struct task_struct *tsk;
2408 struct cgroup_pidlist *l;
2411 * If cgroup gets more users after we read count, we won't have
2412 * enough space - tough. This race is indistinguishable to the
2413 * caller from the case that the additional cgroup users didn't
2414 * show up until sometime later on.
2416 length = cgroup_task_count(cgrp);
2417 array = kmalloc(length * sizeof(pid_t), GFP_KERNEL);
2420 /* now, populate the array */
2421 cgroup_iter_start(cgrp, &it);
2422 while ((tsk = cgroup_iter_next(cgrp, &it))) {
2423 if (unlikely(n == length))
2425 /* get tgid or pid for procs or tasks file respectively */
2426 pid = (procs ? task_tgid_vnr(tsk) : task_pid_vnr(tsk));
2427 if (pid > 0) /* make sure to only use valid results */
2430 cgroup_iter_end(cgrp, &it);
2432 /* now sort & (if procs) strip out duplicates */
2433 sort(array, length, sizeof(pid_t), cmppid, NULL);
2435 length = pidlist_uniq(&array, length);
2440 /* store array in cgroup, freeing old if necessary */
2441 down_write(&l->mutex);
2446 up_write(&l->mutex);
2451 * cgroupstats_build - build and fill cgroupstats
2452 * @stats: cgroupstats to fill information into
2453 * @dentry: A dentry entry belonging to the cgroup for which stats have
2456 * Build and fill cgroupstats so that taskstats can export it to user
2459 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
2462 struct cgroup *cgrp;
2463 struct cgroup_iter it;
2464 struct task_struct *tsk;
2467 * Validate dentry by checking the superblock operations,
2468 * and make sure it's a directory.
2470 if (dentry->d_sb->s_op != &cgroup_ops ||
2471 !S_ISDIR(dentry->d_inode->i_mode))
2475 cgrp = dentry->d_fsdata;
2477 cgroup_iter_start(cgrp, &it);
2478 while ((tsk = cgroup_iter_next(cgrp, &it))) {
2479 switch (tsk->state) {
2481 stats->nr_running++;
2483 case TASK_INTERRUPTIBLE:
2484 stats->nr_sleeping++;
2486 case TASK_UNINTERRUPTIBLE:
2487 stats->nr_uninterruptible++;
2490 stats->nr_stopped++;
2493 if (delayacct_is_task_waiting_on_io(tsk))
2494 stats->nr_io_wait++;
2498 cgroup_iter_end(cgrp, &it);
2506 * seq_file methods for the tasks/procs files. The seq_file position is the
2507 * next pid to display; the seq_file iterator is a pointer to the pid
2508 * in the cgroup->l->list array.
2511 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
2514 * Initially we receive a position value that corresponds to
2515 * one more than the last pid shown (or 0 on the first call or
2516 * after a seek to the start). Use a binary-search to find the
2517 * next pid to display, if any
2519 struct cgroup_pidlist *l = s->private;
2520 int index = 0, pid = *pos;
2523 down_read(&l->mutex);
2525 int end = l->length;
2527 while (index < end) {
2528 int mid = (index + end) / 2;
2529 if (l->list[mid] == pid) {
2532 } else if (l->list[mid] <= pid)
2538 /* If we're off the end of the array, we're done */
2539 if (index >= l->length)
2541 /* Update the abstract position to be the actual pid that we found */
2542 iter = l->list + index;
2547 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
2549 struct cgroup_pidlist *l = s->private;
2553 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
2555 struct cgroup_pidlist *l = s->private;
2557 pid_t *end = l->list + l->length;
2559 * Advance to the next pid in the array. If this goes off the
2571 static int cgroup_pidlist_show(struct seq_file *s, void *v)
2573 return seq_printf(s, "%d\n", *(int *)v);
2577 * seq_operations functions for iterating on pidlists through seq_file -
2578 * independent of whether it's tasks or procs
2580 static const struct seq_operations cgroup_pidlist_seq_operations = {
2581 .start = cgroup_pidlist_start,
2582 .stop = cgroup_pidlist_stop,
2583 .next = cgroup_pidlist_next,
2584 .show = cgroup_pidlist_show,
2587 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
2589 down_write(&l->mutex);
2590 BUG_ON(!l->use_count);
2591 if (!--l->use_count) {
2596 up_write(&l->mutex);
2599 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
2601 struct cgroup_pidlist *l;
2602 if (!(file->f_mode & FMODE_READ))
2605 * the seq_file will only be initialized if the file was opened for
2606 * reading; hence we check if it's not null only in that case.
2608 l = ((struct seq_file *)file->private_data)->private;
2609 cgroup_release_pid_array(l);
2610 return seq_release(inode, file);
2613 static const struct file_operations cgroup_pidlist_operations = {
2615 .llseek = seq_lseek,
2616 .write = cgroup_file_write,
2617 .release = cgroup_pidlist_release,
2621 * The following functions handle opens on a file that displays a pidlist
2622 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
2625 /* helper function for the two below it */
2626 static int cgroup_pidlist_open(struct file *file, bool procs)
2628 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2629 struct cgroup_pidlist *l = (procs ? &cgrp->procs : &cgrp->tasks);
2632 /* Nothing to do for write-only files */
2633 if (!(file->f_mode & FMODE_READ))
2636 /* have the array populated */
2637 retval = pidlist_array_load(cgrp, procs);
2640 /* configure file information */
2641 file->f_op = &cgroup_pidlist_operations;
2643 retval = seq_open(file, &cgroup_pidlist_seq_operations);
2645 cgroup_release_pid_array(l);
2648 ((struct seq_file *)file->private_data)->private = l;
2651 static int cgroup_tasks_open(struct inode *unused, struct file *file)
2653 return cgroup_pidlist_open(file, false);
2655 static int cgroup_procs_open(struct inode *unused, struct file *file)
2657 return cgroup_pidlist_open(file, true);
2660 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
2663 return notify_on_release(cgrp);
2666 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
2670 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
2672 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
2674 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
2679 * for the common functions, 'private' gives the type of file
2681 /* for hysterical raisins, we can't put this on the older files */
2682 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
2683 static struct cftype files[] = {
2686 .open = cgroup_tasks_open,
2687 .write_u64 = cgroup_tasks_write,
2688 .release = cgroup_pidlist_release,
2689 .mode = S_IRUGO | S_IWUSR,
2692 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
2693 .open = cgroup_procs_open,
2694 /* .write_u64 = cgroup_procs_write, TODO */
2695 .release = cgroup_pidlist_release,
2699 .name = "notify_on_release",
2700 .read_u64 = cgroup_read_notify_on_release,
2701 .write_u64 = cgroup_write_notify_on_release,
2705 static struct cftype cft_release_agent = {
2706 .name = "release_agent",
2707 .read_seq_string = cgroup_release_agent_show,
2708 .write_string = cgroup_release_agent_write,
2709 .max_write_len = PATH_MAX,
2712 static int cgroup_populate_dir(struct cgroup *cgrp)
2715 struct cgroup_subsys *ss;
2717 /* First clear out any existing files */
2718 cgroup_clear_directory(cgrp->dentry);
2720 err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files));
2724 if (cgrp == cgrp->top_cgroup) {
2725 if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0)
2729 for_each_subsys(cgrp->root, ss) {
2730 if (ss->populate && (err = ss->populate(ss, cgrp)) < 0)
2733 /* This cgroup is ready now */
2734 for_each_subsys(cgrp->root, ss) {
2735 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
2737 * Update id->css pointer and make this css visible from
2738 * CSS ID functions. This pointer will be dereferened
2739 * from RCU-read-side without locks.
2742 rcu_assign_pointer(css->id->css, css);
2748 static void init_cgroup_css(struct cgroup_subsys_state *css,
2749 struct cgroup_subsys *ss,
2750 struct cgroup *cgrp)
2753 atomic_set(&css->refcnt, 1);
2756 if (cgrp == dummytop)
2757 set_bit(CSS_ROOT, &css->flags);
2758 BUG_ON(cgrp->subsys[ss->subsys_id]);
2759 cgrp->subsys[ss->subsys_id] = css;
2762 static void cgroup_lock_hierarchy(struct cgroupfs_root *root)
2764 /* We need to take each hierarchy_mutex in a consistent order */
2767 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2768 struct cgroup_subsys *ss = subsys[i];
2769 if (ss->root == root)
2770 mutex_lock(&ss->hierarchy_mutex);
2774 static void cgroup_unlock_hierarchy(struct cgroupfs_root *root)
2778 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2779 struct cgroup_subsys *ss = subsys[i];
2780 if (ss->root == root)
2781 mutex_unlock(&ss->hierarchy_mutex);
2786 * cgroup_create - create a cgroup
2787 * @parent: cgroup that will be parent of the new cgroup
2788 * @dentry: dentry of the new cgroup
2789 * @mode: mode to set on new inode
2791 * Must be called with the mutex on the parent inode held
2793 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
2796 struct cgroup *cgrp;
2797 struct cgroupfs_root *root = parent->root;
2799 struct cgroup_subsys *ss;
2800 struct super_block *sb = root->sb;
2802 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
2806 /* Grab a reference on the superblock so the hierarchy doesn't
2807 * get deleted on unmount if there are child cgroups. This
2808 * can be done outside cgroup_mutex, since the sb can't
2809 * disappear while someone has an open control file on the
2811 atomic_inc(&sb->s_active);
2813 mutex_lock(&cgroup_mutex);
2815 init_cgroup_housekeeping(cgrp);
2817 cgrp->parent = parent;
2818 cgrp->root = parent->root;
2819 cgrp->top_cgroup = parent->top_cgroup;
2821 if (notify_on_release(parent))
2822 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
2824 for_each_subsys(root, ss) {
2825 struct cgroup_subsys_state *css = ss->create(ss, cgrp);
2830 init_cgroup_css(css, ss, cgrp);
2832 if (alloc_css_id(ss, parent, cgrp))
2834 /* At error, ->destroy() callback has to free assigned ID. */
2837 cgroup_lock_hierarchy(root);
2838 list_add(&cgrp->sibling, &cgrp->parent->children);
2839 cgroup_unlock_hierarchy(root);
2840 root->number_of_cgroups++;
2842 err = cgroup_create_dir(cgrp, dentry, mode);
2846 /* The cgroup directory was pre-locked for us */
2847 BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
2849 err = cgroup_populate_dir(cgrp);
2850 /* If err < 0, we have a half-filled directory - oh well ;) */
2852 mutex_unlock(&cgroup_mutex);
2853 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
2859 cgroup_lock_hierarchy(root);
2860 list_del(&cgrp->sibling);
2861 cgroup_unlock_hierarchy(root);
2862 root->number_of_cgroups--;
2866 for_each_subsys(root, ss) {
2867 if (cgrp->subsys[ss->subsys_id])
2868 ss->destroy(ss, cgrp);
2871 mutex_unlock(&cgroup_mutex);
2873 /* Release the reference count that we took on the superblock */
2874 deactivate_super(sb);
2880 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2882 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
2884 /* the vfs holds inode->i_mutex already */
2885 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
2888 static int cgroup_has_css_refs(struct cgroup *cgrp)
2890 /* Check the reference count on each subsystem. Since we
2891 * already established that there are no tasks in the
2892 * cgroup, if the css refcount is also 1, then there should
2893 * be no outstanding references, so the subsystem is safe to
2894 * destroy. We scan across all subsystems rather than using
2895 * the per-hierarchy linked list of mounted subsystems since
2896 * we can be called via check_for_release() with no
2897 * synchronization other than RCU, and the subsystem linked
2898 * list isn't RCU-safe */
2900 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2901 struct cgroup_subsys *ss = subsys[i];
2902 struct cgroup_subsys_state *css;
2903 /* Skip subsystems not in this hierarchy */
2904 if (ss->root != cgrp->root)
2906 css = cgrp->subsys[ss->subsys_id];
2907 /* When called from check_for_release() it's possible
2908 * that by this point the cgroup has been removed
2909 * and the css deleted. But a false-positive doesn't
2910 * matter, since it can only happen if the cgroup
2911 * has been deleted and hence no longer needs the
2912 * release agent to be called anyway. */
2913 if (css && (atomic_read(&css->refcnt) > 1))
2920 * Atomically mark all (or else none) of the cgroup's CSS objects as
2921 * CSS_REMOVED. Return true on success, or false if the cgroup has
2922 * busy subsystems. Call with cgroup_mutex held
2925 static int cgroup_clear_css_refs(struct cgroup *cgrp)
2927 struct cgroup_subsys *ss;
2928 unsigned long flags;
2929 bool failed = false;
2930 local_irq_save(flags);
2931 for_each_subsys(cgrp->root, ss) {
2932 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
2935 /* We can only remove a CSS with a refcnt==1 */
2936 refcnt = atomic_read(&css->refcnt);
2943 * Drop the refcnt to 0 while we check other
2944 * subsystems. This will cause any racing
2945 * css_tryget() to spin until we set the
2946 * CSS_REMOVED bits or abort
2948 if (atomic_cmpxchg(&css->refcnt, refcnt, 0) == refcnt)
2954 for_each_subsys(cgrp->root, ss) {
2955 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
2958 * Restore old refcnt if we previously managed
2959 * to clear it from 1 to 0
2961 if (!atomic_read(&css->refcnt))
2962 atomic_set(&css->refcnt, 1);
2964 /* Commit the fact that the CSS is removed */
2965 set_bit(CSS_REMOVED, &css->flags);
2968 local_irq_restore(flags);
2972 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
2974 struct cgroup *cgrp = dentry->d_fsdata;
2976 struct cgroup *parent;
2980 /* the vfs holds both inode->i_mutex already */
2982 mutex_lock(&cgroup_mutex);
2983 if (atomic_read(&cgrp->count) != 0) {
2984 mutex_unlock(&cgroup_mutex);
2987 if (!list_empty(&cgrp->children)) {
2988 mutex_unlock(&cgroup_mutex);
2991 mutex_unlock(&cgroup_mutex);
2994 * In general, subsystem has no css->refcnt after pre_destroy(). But
2995 * in racy cases, subsystem may have to get css->refcnt after
2996 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
2997 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
2998 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
2999 * and subsystem's reference count handling. Please see css_get/put
3000 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
3002 set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3005 * Call pre_destroy handlers of subsys. Notify subsystems
3006 * that rmdir() request comes.
3008 ret = cgroup_call_pre_destroy(cgrp);
3010 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3014 mutex_lock(&cgroup_mutex);
3015 parent = cgrp->parent;
3016 if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) {
3017 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3018 mutex_unlock(&cgroup_mutex);
3021 prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE);
3022 if (!cgroup_clear_css_refs(cgrp)) {
3023 mutex_unlock(&cgroup_mutex);
3025 * Because someone may call cgroup_wakeup_rmdir_waiter() before
3026 * prepare_to_wait(), we need to check this flag.
3028 if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags))
3030 finish_wait(&cgroup_rmdir_waitq, &wait);
3031 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3032 if (signal_pending(current))
3036 /* NO css_tryget() can success after here. */
3037 finish_wait(&cgroup_rmdir_waitq, &wait);
3038 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3040 spin_lock(&release_list_lock);
3041 set_bit(CGRP_REMOVED, &cgrp->flags);
3042 if (!list_empty(&cgrp->release_list))
3043 list_del(&cgrp->release_list);
3044 spin_unlock(&release_list_lock);
3046 cgroup_lock_hierarchy(cgrp->root);
3047 /* delete this cgroup from parent->children */
3048 list_del(&cgrp->sibling);
3049 cgroup_unlock_hierarchy(cgrp->root);
3051 spin_lock(&cgrp->dentry->d_lock);
3052 d = dget(cgrp->dentry);
3053 spin_unlock(&d->d_lock);
3055 cgroup_d_remove_dir(d);
3058 set_bit(CGRP_RELEASABLE, &parent->flags);
3059 check_for_release(parent);
3061 mutex_unlock(&cgroup_mutex);
3065 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
3067 struct cgroup_subsys_state *css;
3069 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
3071 /* Create the top cgroup state for this subsystem */
3072 list_add(&ss->sibling, &rootnode.subsys_list);
3073 ss->root = &rootnode;
3074 css = ss->create(ss, dummytop);
3075 /* We don't handle early failures gracefully */
3076 BUG_ON(IS_ERR(css));
3077 init_cgroup_css(css, ss, dummytop);
3079 /* Update the init_css_set to contain a subsys
3080 * pointer to this state - since the subsystem is
3081 * newly registered, all tasks and hence the
3082 * init_css_set is in the subsystem's top cgroup. */
3083 init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
3085 need_forkexit_callback |= ss->fork || ss->exit;
3087 /* At system boot, before all subsystems have been
3088 * registered, no tasks have been forked, so we don't
3089 * need to invoke fork callbacks here. */
3090 BUG_ON(!list_empty(&init_task.tasks));
3092 mutex_init(&ss->hierarchy_mutex);
3093 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
3098 * cgroup_init_early - cgroup initialization at system boot
3100 * Initialize cgroups at system boot, and initialize any
3101 * subsystems that request early init.
3103 int __init cgroup_init_early(void)
3106 atomic_set(&init_css_set.refcount, 1);
3107 INIT_LIST_HEAD(&init_css_set.cg_links);
3108 INIT_LIST_HEAD(&init_css_set.tasks);
3109 INIT_HLIST_NODE(&init_css_set.hlist);
3111 init_cgroup_root(&rootnode);
3113 init_task.cgroups = &init_css_set;
3115 init_css_set_link.cg = &init_css_set;
3116 init_css_set_link.cgrp = dummytop;
3117 list_add(&init_css_set_link.cgrp_link_list,
3118 &rootnode.top_cgroup.css_sets);
3119 list_add(&init_css_set_link.cg_link_list,
3120 &init_css_set.cg_links);
3122 for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
3123 INIT_HLIST_HEAD(&css_set_table[i]);
3125 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3126 struct cgroup_subsys *ss = subsys[i];
3129 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
3130 BUG_ON(!ss->create);
3131 BUG_ON(!ss->destroy);
3132 if (ss->subsys_id != i) {
3133 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
3134 ss->name, ss->subsys_id);
3139 cgroup_init_subsys(ss);
3145 * cgroup_init - cgroup initialization
3147 * Register cgroup filesystem and /proc file, and initialize
3148 * any subsystems that didn't request early init.
3150 int __init cgroup_init(void)
3154 struct hlist_head *hhead;
3156 err = bdi_init(&cgroup_backing_dev_info);
3160 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3161 struct cgroup_subsys *ss = subsys[i];
3162 if (!ss->early_init)
3163 cgroup_init_subsys(ss);
3165 cgroup_subsys_init_idr(ss);
3168 /* Add init_css_set to the hash table */
3169 hhead = css_set_hash(init_css_set.subsys);
3170 hlist_add_head(&init_css_set.hlist, hhead);
3171 BUG_ON(!init_root_id(&rootnode));
3172 err = register_filesystem(&cgroup_fs_type);
3176 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
3180 bdi_destroy(&cgroup_backing_dev_info);
3186 * proc_cgroup_show()
3187 * - Print task's cgroup paths into seq_file, one line for each hierarchy
3188 * - Used for /proc/<pid>/cgroup.
3189 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
3190 * doesn't really matter if tsk->cgroup changes after we read it,
3191 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
3192 * anyway. No need to check that tsk->cgroup != NULL, thanks to
3193 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
3194 * cgroup to top_cgroup.
3197 /* TODO: Use a proper seq_file iterator */
3198 static int proc_cgroup_show(struct seq_file *m, void *v)
3201 struct task_struct *tsk;
3204 struct cgroupfs_root *root;
3207 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
3213 tsk = get_pid_task(pid, PIDTYPE_PID);
3219 mutex_lock(&cgroup_mutex);
3221 for_each_active_root(root) {
3222 struct cgroup_subsys *ss;
3223 struct cgroup *cgrp;
3226 seq_printf(m, "%d:", root->hierarchy_id);
3227 for_each_subsys(root, ss)
3228 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
3229 if (strlen(root->name))
3230 seq_printf(m, "%sname=%s", count ? "," : "",
3233 cgrp = task_cgroup_from_root(tsk, root);
3234 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
3242 mutex_unlock(&cgroup_mutex);
3243 put_task_struct(tsk);
3250 static int cgroup_open(struct inode *inode, struct file *file)
3252 struct pid *pid = PROC_I(inode)->pid;
3253 return single_open(file, proc_cgroup_show, pid);
3256 struct file_operations proc_cgroup_operations = {
3257 .open = cgroup_open,
3259 .llseek = seq_lseek,
3260 .release = single_release,
3263 /* Display information about each subsystem and each hierarchy */
3264 static int proc_cgroupstats_show(struct seq_file *m, void *v)
3268 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
3269 mutex_lock(&cgroup_mutex);
3270 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3271 struct cgroup_subsys *ss = subsys[i];
3272 seq_printf(m, "%s\t%d\t%d\t%d\n",
3273 ss->name, ss->root->hierarchy_id,
3274 ss->root->number_of_cgroups, !ss->disabled);
3276 mutex_unlock(&cgroup_mutex);
3280 static int cgroupstats_open(struct inode *inode, struct file *file)
3282 return single_open(file, proc_cgroupstats_show, NULL);
3285 static struct file_operations proc_cgroupstats_operations = {
3286 .open = cgroupstats_open,
3288 .llseek = seq_lseek,
3289 .release = single_release,
3293 * cgroup_fork - attach newly forked task to its parents cgroup.
3294 * @child: pointer to task_struct of forking parent process.
3296 * Description: A task inherits its parent's cgroup at fork().
3298 * A pointer to the shared css_set was automatically copied in
3299 * fork.c by dup_task_struct(). However, we ignore that copy, since
3300 * it was not made under the protection of RCU or cgroup_mutex, so
3301 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
3302 * have already changed current->cgroups, allowing the previously
3303 * referenced cgroup group to be removed and freed.
3305 * At the point that cgroup_fork() is called, 'current' is the parent
3306 * task, and the passed argument 'child' points to the child task.
3308 void cgroup_fork(struct task_struct *child)
3311 child->cgroups = current->cgroups;
3312 get_css_set(child->cgroups);
3313 task_unlock(current);
3314 INIT_LIST_HEAD(&child->cg_list);
3318 * cgroup_fork_callbacks - run fork callbacks
3319 * @child: the new task
3321 * Called on a new task very soon before adding it to the
3322 * tasklist. No need to take any locks since no-one can
3323 * be operating on this task.
3325 void cgroup_fork_callbacks(struct task_struct *child)
3327 if (need_forkexit_callback) {
3329 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3330 struct cgroup_subsys *ss = subsys[i];
3332 ss->fork(ss, child);
3338 * cgroup_post_fork - called on a new task after adding it to the task list
3339 * @child: the task in question
3341 * Adds the task to the list running through its css_set if necessary.
3342 * Has to be after the task is visible on the task list in case we race
3343 * with the first call to cgroup_iter_start() - to guarantee that the
3344 * new task ends up on its list.
3346 void cgroup_post_fork(struct task_struct *child)
3348 if (use_task_css_set_links) {
3349 write_lock(&css_set_lock);
3351 if (list_empty(&child->cg_list))
3352 list_add(&child->cg_list, &child->cgroups->tasks);
3354 write_unlock(&css_set_lock);
3358 * cgroup_exit - detach cgroup from exiting task
3359 * @tsk: pointer to task_struct of exiting process
3360 * @run_callback: run exit callbacks?
3362 * Description: Detach cgroup from @tsk and release it.
3364 * Note that cgroups marked notify_on_release force every task in
3365 * them to take the global cgroup_mutex mutex when exiting.
3366 * This could impact scaling on very large systems. Be reluctant to
3367 * use notify_on_release cgroups where very high task exit scaling
3368 * is required on large systems.
3370 * the_top_cgroup_hack:
3372 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
3374 * We call cgroup_exit() while the task is still competent to
3375 * handle notify_on_release(), then leave the task attached to the
3376 * root cgroup in each hierarchy for the remainder of its exit.
3378 * To do this properly, we would increment the reference count on
3379 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
3380 * code we would add a second cgroup function call, to drop that
3381 * reference. This would just create an unnecessary hot spot on
3382 * the top_cgroup reference count, to no avail.
3384 * Normally, holding a reference to a cgroup without bumping its
3385 * count is unsafe. The cgroup could go away, or someone could
3386 * attach us to a different cgroup, decrementing the count on
3387 * the first cgroup that we never incremented. But in this case,
3388 * top_cgroup isn't going away, and either task has PF_EXITING set,
3389 * which wards off any cgroup_attach_task() attempts, or task is a failed
3390 * fork, never visible to cgroup_attach_task.
3392 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
3397 if (run_callbacks && need_forkexit_callback) {
3398 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3399 struct cgroup_subsys *ss = subsys[i];
3406 * Unlink from the css_set task list if necessary.
3407 * Optimistically check cg_list before taking
3410 if (!list_empty(&tsk->cg_list)) {
3411 write_lock(&css_set_lock);
3412 if (!list_empty(&tsk->cg_list))
3413 list_del(&tsk->cg_list);
3414 write_unlock(&css_set_lock);
3417 /* Reassign the task to the init_css_set. */
3420 tsk->cgroups = &init_css_set;
3423 put_css_set_taskexit(cg);
3427 * cgroup_clone - clone the cgroup the given subsystem is attached to
3428 * @tsk: the task to be moved
3429 * @subsys: the given subsystem
3430 * @nodename: the name for the new cgroup
3432 * Duplicate the current cgroup in the hierarchy that the given
3433 * subsystem is attached to, and move this task into the new
3436 int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *subsys,
3439 struct dentry *dentry;
3441 struct cgroup *parent, *child;
3442 struct inode *inode;
3444 struct cgroupfs_root *root;
3445 struct cgroup_subsys *ss;
3447 /* We shouldn't be called by an unregistered subsystem */
3448 BUG_ON(!subsys->active);
3450 /* First figure out what hierarchy and cgroup we're dealing
3451 * with, and pin them so we can drop cgroup_mutex */
3452 mutex_lock(&cgroup_mutex);
3454 root = subsys->root;
3455 if (root == &rootnode) {
3456 mutex_unlock(&cgroup_mutex);
3460 /* Pin the hierarchy */
3461 if (!atomic_inc_not_zero(&root->sb->s_active)) {
3462 /* We race with the final deactivate_super() */
3463 mutex_unlock(&cgroup_mutex);
3467 /* Keep the cgroup alive */
3469 parent = task_cgroup(tsk, subsys->subsys_id);
3474 mutex_unlock(&cgroup_mutex);
3476 /* Now do the VFS work to create a cgroup */
3477 inode = parent->dentry->d_inode;
3479 /* Hold the parent directory mutex across this operation to
3480 * stop anyone else deleting the new cgroup */
3481 mutex_lock(&inode->i_mutex);
3482 dentry = lookup_one_len(nodename, parent->dentry, strlen(nodename));
3483 if (IS_ERR(dentry)) {
3485 "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename,
3487 ret = PTR_ERR(dentry);
3491 /* Create the cgroup directory, which also creates the cgroup */
3492 ret = vfs_mkdir(inode, dentry, 0755);
3493 child = __d_cgrp(dentry);
3497 "Failed to create cgroup %s: %d\n", nodename,
3502 /* The cgroup now exists. Retake cgroup_mutex and check
3503 * that we're still in the same state that we thought we
3505 mutex_lock(&cgroup_mutex);
3506 if ((root != subsys->root) ||
3507 (parent != task_cgroup(tsk, subsys->subsys_id))) {
3508 /* Aargh, we raced ... */
3509 mutex_unlock(&inode->i_mutex);
3512 deactivate_super(root->sb);
3513 /* The cgroup is still accessible in the VFS, but
3514 * we're not going to try to rmdir() it at this
3517 "Race in cgroup_clone() - leaking cgroup %s\n",
3522 /* do any required auto-setup */
3523 for_each_subsys(root, ss) {
3525 ss->post_clone(ss, child);
3528 /* All seems fine. Finish by moving the task into the new cgroup */
3529 ret = cgroup_attach_task(child, tsk);
3530 mutex_unlock(&cgroup_mutex);
3533 mutex_unlock(&inode->i_mutex);
3535 mutex_lock(&cgroup_mutex);
3537 mutex_unlock(&cgroup_mutex);
3538 deactivate_super(root->sb);
3543 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
3544 * @cgrp: the cgroup in question
3545 * @task: the task in question
3547 * See if @cgrp is a descendant of @task's cgroup in the appropriate
3550 * If we are sending in dummytop, then presumably we are creating
3551 * the top cgroup in the subsystem.
3553 * Called only by the ns (nsproxy) cgroup.
3555 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
3558 struct cgroup *target;
3560 if (cgrp == dummytop)
3563 target = task_cgroup_from_root(task, cgrp->root);
3564 while (cgrp != target && cgrp!= cgrp->top_cgroup)
3565 cgrp = cgrp->parent;
3566 ret = (cgrp == target);
3570 static void check_for_release(struct cgroup *cgrp)
3572 /* All of these checks rely on RCU to keep the cgroup
3573 * structure alive */
3574 if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
3575 && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
3576 /* Control Group is currently removeable. If it's not
3577 * already queued for a userspace notification, queue
3579 int need_schedule_work = 0;
3580 spin_lock(&release_list_lock);
3581 if (!cgroup_is_removed(cgrp) &&
3582 list_empty(&cgrp->release_list)) {
3583 list_add(&cgrp->release_list, &release_list);
3584 need_schedule_work = 1;
3586 spin_unlock(&release_list_lock);
3587 if (need_schedule_work)
3588 schedule_work(&release_agent_work);
3592 void __css_put(struct cgroup_subsys_state *css)
3594 struct cgroup *cgrp = css->cgroup;
3596 if (atomic_dec_return(&css->refcnt) == 1) {
3597 if (notify_on_release(cgrp)) {
3598 set_bit(CGRP_RELEASABLE, &cgrp->flags);
3599 check_for_release(cgrp);
3601 cgroup_wakeup_rmdir_waiter(cgrp);
3607 * Notify userspace when a cgroup is released, by running the
3608 * configured release agent with the name of the cgroup (path
3609 * relative to the root of cgroup file system) as the argument.
3611 * Most likely, this user command will try to rmdir this cgroup.
3613 * This races with the possibility that some other task will be
3614 * attached to this cgroup before it is removed, or that some other
3615 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
3616 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
3617 * unused, and this cgroup will be reprieved from its death sentence,
3618 * to continue to serve a useful existence. Next time it's released,
3619 * we will get notified again, if it still has 'notify_on_release' set.
3621 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
3622 * means only wait until the task is successfully execve()'d. The
3623 * separate release agent task is forked by call_usermodehelper(),
3624 * then control in this thread returns here, without waiting for the
3625 * release agent task. We don't bother to wait because the caller of
3626 * this routine has no use for the exit status of the release agent
3627 * task, so no sense holding our caller up for that.
3629 static void cgroup_release_agent(struct work_struct *work)
3631 BUG_ON(work != &release_agent_work);
3632 mutex_lock(&cgroup_mutex);
3633 spin_lock(&release_list_lock);
3634 while (!list_empty(&release_list)) {
3635 char *argv[3], *envp[3];
3637 char *pathbuf = NULL, *agentbuf = NULL;
3638 struct cgroup *cgrp = list_entry(release_list.next,
3641 list_del_init(&cgrp->release_list);
3642 spin_unlock(&release_list_lock);
3643 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
3646 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
3648 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
3653 argv[i++] = agentbuf;
3654 argv[i++] = pathbuf;
3658 /* minimal command environment */
3659 envp[i++] = "HOME=/";
3660 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
3663 /* Drop the lock while we invoke the usermode helper,
3664 * since the exec could involve hitting disk and hence
3665 * be a slow process */
3666 mutex_unlock(&cgroup_mutex);
3667 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
3668 mutex_lock(&cgroup_mutex);
3672 spin_lock(&release_list_lock);
3674 spin_unlock(&release_list_lock);
3675 mutex_unlock(&cgroup_mutex);
3678 static int __init cgroup_disable(char *str)
3683 while ((token = strsep(&str, ",")) != NULL) {
3687 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3688 struct cgroup_subsys *ss = subsys[i];
3690 if (!strcmp(token, ss->name)) {
3692 printk(KERN_INFO "Disabling %s control group"
3693 " subsystem\n", ss->name);
3700 __setup("cgroup_disable=", cgroup_disable);
3703 * Functons for CSS ID.
3707 *To get ID other than 0, this should be called when !cgroup_is_removed().
3709 unsigned short css_id(struct cgroup_subsys_state *css)
3711 struct css_id *cssid = rcu_dereference(css->id);
3718 unsigned short css_depth(struct cgroup_subsys_state *css)
3720 struct css_id *cssid = rcu_dereference(css->id);
3723 return cssid->depth;
3727 bool css_is_ancestor(struct cgroup_subsys_state *child,
3728 const struct cgroup_subsys_state *root)
3730 struct css_id *child_id = rcu_dereference(child->id);
3731 struct css_id *root_id = rcu_dereference(root->id);
3733 if (!child_id || !root_id || (child_id->depth < root_id->depth))
3735 return child_id->stack[root_id->depth] == root_id->id;
3738 static void __free_css_id_cb(struct rcu_head *head)
3742 id = container_of(head, struct css_id, rcu_head);
3746 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
3748 struct css_id *id = css->id;
3749 /* When this is called before css_id initialization, id can be NULL */
3753 BUG_ON(!ss->use_id);
3755 rcu_assign_pointer(id->css, NULL);
3756 rcu_assign_pointer(css->id, NULL);
3757 spin_lock(&ss->id_lock);
3758 idr_remove(&ss->idr, id->id);
3759 spin_unlock(&ss->id_lock);
3760 call_rcu(&id->rcu_head, __free_css_id_cb);
3764 * This is called by init or create(). Then, calls to this function are
3765 * always serialized (By cgroup_mutex() at create()).
3768 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
3770 struct css_id *newid;
3771 int myid, error, size;
3773 BUG_ON(!ss->use_id);
3775 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
3776 newid = kzalloc(size, GFP_KERNEL);
3778 return ERR_PTR(-ENOMEM);
3780 if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
3784 spin_lock(&ss->id_lock);
3785 /* Don't use 0. allocates an ID of 1-65535 */
3786 error = idr_get_new_above(&ss->idr, newid, 1, &myid);
3787 spin_unlock(&ss->id_lock);
3789 /* Returns error when there are no free spaces for new ID.*/
3794 if (myid > CSS_ID_MAX)
3798 newid->depth = depth;
3802 spin_lock(&ss->id_lock);
3803 idr_remove(&ss->idr, myid);
3804 spin_unlock(&ss->id_lock);
3807 return ERR_PTR(error);
3811 static int __init cgroup_subsys_init_idr(struct cgroup_subsys *ss)
3813 struct css_id *newid;
3814 struct cgroup_subsys_state *rootcss;
3816 spin_lock_init(&ss->id_lock);
3819 rootcss = init_css_set.subsys[ss->subsys_id];
3820 newid = get_new_cssid(ss, 0);
3822 return PTR_ERR(newid);
3824 newid->stack[0] = newid->id;
3825 newid->css = rootcss;
3826 rootcss->id = newid;
3830 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
3831 struct cgroup *child)
3833 int subsys_id, i, depth = 0;
3834 struct cgroup_subsys_state *parent_css, *child_css;
3835 struct css_id *child_id, *parent_id = NULL;
3837 subsys_id = ss->subsys_id;
3838 parent_css = parent->subsys[subsys_id];
3839 child_css = child->subsys[subsys_id];
3840 depth = css_depth(parent_css) + 1;
3841 parent_id = parent_css->id;
3843 child_id = get_new_cssid(ss, depth);
3844 if (IS_ERR(child_id))
3845 return PTR_ERR(child_id);
3847 for (i = 0; i < depth; i++)
3848 child_id->stack[i] = parent_id->stack[i];
3849 child_id->stack[depth] = child_id->id;
3851 * child_id->css pointer will be set after this cgroup is available
3852 * see cgroup_populate_dir()
3854 rcu_assign_pointer(child_css->id, child_id);
3860 * css_lookup - lookup css by id
3861 * @ss: cgroup subsys to be looked into.
3864 * Returns pointer to cgroup_subsys_state if there is valid one with id.
3865 * NULL if not. Should be called under rcu_read_lock()
3867 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
3869 struct css_id *cssid = NULL;
3871 BUG_ON(!ss->use_id);
3872 cssid = idr_find(&ss->idr, id);
3874 if (unlikely(!cssid))
3877 return rcu_dereference(cssid->css);
3881 * css_get_next - lookup next cgroup under specified hierarchy.
3882 * @ss: pointer to subsystem
3883 * @id: current position of iteration.
3884 * @root: pointer to css. search tree under this.
3885 * @foundid: position of found object.
3887 * Search next css under the specified hierarchy of rootid. Calling under
3888 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
3890 struct cgroup_subsys_state *
3891 css_get_next(struct cgroup_subsys *ss, int id,
3892 struct cgroup_subsys_state *root, int *foundid)
3894 struct cgroup_subsys_state *ret = NULL;
3897 int rootid = css_id(root);
3898 int depth = css_depth(root);
3903 BUG_ON(!ss->use_id);
3904 /* fill start point for scan */
3908 * scan next entry from bitmap(tree), tmpid is updated after
3911 spin_lock(&ss->id_lock);
3912 tmp = idr_get_next(&ss->idr, &tmpid);
3913 spin_unlock(&ss->id_lock);
3917 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
3918 ret = rcu_dereference(tmp->css);
3924 /* continue to scan from next id */
3930 #ifdef CONFIG_CGROUP_DEBUG
3931 static struct cgroup_subsys_state *debug_create(struct cgroup_subsys *ss,
3932 struct cgroup *cont)
3934 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
3937 return ERR_PTR(-ENOMEM);
3942 static void debug_destroy(struct cgroup_subsys *ss, struct cgroup *cont)
3944 kfree(cont->subsys[debug_subsys_id]);
3947 static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
3949 return atomic_read(&cont->count);
3952 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
3954 return cgroup_task_count(cont);
3957 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
3959 return (u64)(unsigned long)current->cgroups;
3962 static u64 current_css_set_refcount_read(struct cgroup *cont,
3968 count = atomic_read(¤t->cgroups->refcount);
3973 static int current_css_set_cg_links_read(struct cgroup *cont,
3975 struct seq_file *seq)
3977 struct cg_cgroup_link *link;
3980 read_lock(&css_set_lock);
3982 cg = rcu_dereference(current->cgroups);
3983 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
3984 struct cgroup *c = link->cgrp;
3988 name = c->dentry->d_name.name;
3991 seq_printf(seq, "Root %d group %s\n",
3992 c->root->hierarchy_id, name);
3995 read_unlock(&css_set_lock);
3999 #define MAX_TASKS_SHOWN_PER_CSS 25
4000 static int cgroup_css_links_read(struct cgroup *cont,
4002 struct seq_file *seq)
4004 struct cg_cgroup_link *link;
4006 read_lock(&css_set_lock);
4007 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
4008 struct css_set *cg = link->cg;
4009 struct task_struct *task;
4011 seq_printf(seq, "css_set %p\n", cg);
4012 list_for_each_entry(task, &cg->tasks, cg_list) {
4013 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
4014 seq_puts(seq, " ...\n");
4017 seq_printf(seq, " task %d\n",
4018 task_pid_vnr(task));
4022 read_unlock(&css_set_lock);
4026 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
4028 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
4031 static struct cftype debug_files[] = {
4033 .name = "cgroup_refcount",
4034 .read_u64 = cgroup_refcount_read,
4037 .name = "taskcount",
4038 .read_u64 = debug_taskcount_read,
4042 .name = "current_css_set",
4043 .read_u64 = current_css_set_read,
4047 .name = "current_css_set_refcount",
4048 .read_u64 = current_css_set_refcount_read,
4052 .name = "current_css_set_cg_links",
4053 .read_seq_string = current_css_set_cg_links_read,
4057 .name = "cgroup_css_links",
4058 .read_seq_string = cgroup_css_links_read,
4062 .name = "releasable",
4063 .read_u64 = releasable_read,
4067 static int debug_populate(struct cgroup_subsys *ss, struct cgroup *cont)
4069 return cgroup_add_files(cont, ss, debug_files,
4070 ARRAY_SIZE(debug_files));
4073 struct cgroup_subsys debug_subsys = {
4075 .create = debug_create,
4076 .destroy = debug_destroy,
4077 .populate = debug_populate,
4078 .subsys_id = debug_subsys_id,
4080 #endif /* CONFIG_CGROUP_DEBUG */