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>
53 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
55 #include <asm/atomic.h>
57 static DEFINE_MUTEX(cgroup_mutex);
59 /* Generate an array of cgroup subsystem pointers */
60 #define SUBSYS(_x) &_x ## _subsys,
62 static struct cgroup_subsys *subsys[] = {
63 #include <linux/cgroup_subsys.h>
66 #define MAX_CGROUP_ROOT_NAMELEN 64
69 * A cgroupfs_root represents the root of a cgroup hierarchy,
70 * and may be associated with a superblock to form an active
73 struct cgroupfs_root {
74 struct super_block *sb;
77 * The bitmask of subsystems intended to be attached to this
80 unsigned long subsys_bits;
82 /* Unique id for this hierarchy. */
85 /* The bitmask of subsystems currently attached to this hierarchy */
86 unsigned long actual_subsys_bits;
88 /* A list running through the attached subsystems */
89 struct list_head subsys_list;
91 /* The root cgroup for this hierarchy */
92 struct cgroup top_cgroup;
94 /* Tracks how many cgroups are currently defined in hierarchy.*/
95 int number_of_cgroups;
97 /* A list running through the active hierarchies */
98 struct list_head root_list;
100 /* Hierarchy-specific flags */
103 /* The path to use for release notifications. */
104 char release_agent_path[PATH_MAX];
106 /* The name for this hierarchy - may be empty */
107 char name[MAX_CGROUP_ROOT_NAMELEN];
111 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
112 * subsystems that are otherwise unattached - it never has more than a
113 * single cgroup, and all tasks are part of that cgroup.
115 static struct cgroupfs_root rootnode;
118 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
119 * cgroup_subsys->use_id != 0.
121 #define CSS_ID_MAX (65535)
124 * The css to which this ID points. This pointer is set to valid value
125 * after cgroup is populated. If cgroup is removed, this will be NULL.
126 * This pointer is expected to be RCU-safe because destroy()
127 * is called after synchronize_rcu(). But for safe use, css_is_removed()
128 * css_tryget() should be used for avoiding race.
130 struct cgroup_subsys_state *css;
136 * Depth in hierarchy which this ID belongs to.
138 unsigned short depth;
140 * ID is freed by RCU. (and lookup routine is RCU safe.)
142 struct rcu_head rcu_head;
144 * Hierarchy of CSS ID belongs to.
146 unsigned short stack[0]; /* Array of Length (depth+1) */
150 /* The list of hierarchy roots */
152 static LIST_HEAD(roots);
153 static int root_count;
155 static DEFINE_IDA(hierarchy_ida);
156 static int next_hierarchy_id;
157 static DEFINE_SPINLOCK(hierarchy_id_lock);
159 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
160 #define dummytop (&rootnode.top_cgroup)
162 /* This flag indicates whether tasks in the fork and exit paths should
163 * check for fork/exit handlers to call. This avoids us having to do
164 * extra work in the fork/exit path if none of the subsystems need to
167 static int need_forkexit_callback __read_mostly;
169 /* convenient tests for these bits */
170 inline int cgroup_is_removed(const struct cgroup *cgrp)
172 return test_bit(CGRP_REMOVED, &cgrp->flags);
175 /* bits in struct cgroupfs_root flags field */
177 ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
180 static int cgroup_is_releasable(const struct cgroup *cgrp)
183 (1 << CGRP_RELEASABLE) |
184 (1 << CGRP_NOTIFY_ON_RELEASE);
185 return (cgrp->flags & bits) == bits;
188 static int notify_on_release(const struct cgroup *cgrp)
190 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
194 * for_each_subsys() allows you to iterate on each subsystem attached to
195 * an active hierarchy
197 #define for_each_subsys(_root, _ss) \
198 list_for_each_entry(_ss, &_root->subsys_list, sibling)
200 /* for_each_active_root() allows you to iterate across the active hierarchies */
201 #define for_each_active_root(_root) \
202 list_for_each_entry(_root, &roots, root_list)
204 /* the list of cgroups eligible for automatic release. Protected by
205 * release_list_lock */
206 static LIST_HEAD(release_list);
207 static DEFINE_SPINLOCK(release_list_lock);
208 static void cgroup_release_agent(struct work_struct *work);
209 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
210 static void check_for_release(struct cgroup *cgrp);
212 /* Link structure for associating css_set objects with cgroups */
213 struct cg_cgroup_link {
215 * List running through cg_cgroup_links associated with a
216 * cgroup, anchored on cgroup->css_sets
218 struct list_head cgrp_link_list;
221 * List running through cg_cgroup_links pointing at a
222 * single css_set object, anchored on css_set->cg_links
224 struct list_head cg_link_list;
228 /* The default css_set - used by init and its children prior to any
229 * hierarchies being mounted. It contains a pointer to the root state
230 * for each subsystem. Also used to anchor the list of css_sets. Not
231 * reference-counted, to improve performance when child cgroups
232 * haven't been created.
235 static struct css_set init_css_set;
236 static struct cg_cgroup_link init_css_set_link;
238 static int cgroup_subsys_init_idr(struct cgroup_subsys *ss);
240 /* css_set_lock protects the list of css_set objects, and the
241 * chain of tasks off each css_set. Nests outside task->alloc_lock
242 * due to cgroup_iter_start() */
243 static DEFINE_RWLOCK(css_set_lock);
244 static int css_set_count;
247 * hash table for cgroup groups. This improves the performance to find
248 * an existing css_set. This hash doesn't (currently) take into
249 * account cgroups in empty hierarchies.
251 #define CSS_SET_HASH_BITS 7
252 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
253 static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE];
255 static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[])
259 unsigned long tmp = 0UL;
261 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
262 tmp += (unsigned long)css[i];
263 tmp = (tmp >> 16) ^ tmp;
265 index = hash_long(tmp, CSS_SET_HASH_BITS);
267 return &css_set_table[index];
270 /* We don't maintain the lists running through each css_set to its
271 * task until after the first call to cgroup_iter_start(). This
272 * reduces the fork()/exit() overhead for people who have cgroups
273 * compiled into their kernel but not actually in use */
274 static int use_task_css_set_links __read_mostly;
276 static void __put_css_set(struct css_set *cg, int taskexit)
278 struct cg_cgroup_link *link;
279 struct cg_cgroup_link *saved_link;
281 * Ensure that the refcount doesn't hit zero while any readers
282 * can see it. Similar to atomic_dec_and_lock(), but for an
285 if (atomic_add_unless(&cg->refcount, -1, 1))
287 write_lock(&css_set_lock);
288 if (!atomic_dec_and_test(&cg->refcount)) {
289 write_unlock(&css_set_lock);
293 /* This css_set is dead. unlink it and release cgroup refcounts */
294 hlist_del(&cg->hlist);
297 list_for_each_entry_safe(link, saved_link, &cg->cg_links,
299 struct cgroup *cgrp = link->cgrp;
300 list_del(&link->cg_link_list);
301 list_del(&link->cgrp_link_list);
302 if (atomic_dec_and_test(&cgrp->count) &&
303 notify_on_release(cgrp)) {
305 set_bit(CGRP_RELEASABLE, &cgrp->flags);
306 check_for_release(cgrp);
312 write_unlock(&css_set_lock);
317 * refcounted get/put for css_set objects
319 static inline void get_css_set(struct css_set *cg)
321 atomic_inc(&cg->refcount);
324 static inline void put_css_set(struct css_set *cg)
326 __put_css_set(cg, 0);
329 static inline void put_css_set_taskexit(struct css_set *cg)
331 __put_css_set(cg, 1);
335 * compare_css_sets - helper function for find_existing_css_set().
336 * @cg: candidate css_set being tested
337 * @old_cg: existing css_set for a task
338 * @new_cgrp: cgroup that's being entered by the task
339 * @template: desired set of css pointers in css_set (pre-calculated)
341 * Returns true if "cg" matches "old_cg" except for the hierarchy
342 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
344 static bool compare_css_sets(struct css_set *cg,
345 struct css_set *old_cg,
346 struct cgroup *new_cgrp,
347 struct cgroup_subsys_state *template[])
349 struct list_head *l1, *l2;
351 if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
352 /* Not all subsystems matched */
357 * Compare cgroup pointers in order to distinguish between
358 * different cgroups in heirarchies with no subsystems. We
359 * could get by with just this check alone (and skip the
360 * memcmp above) but on most setups the memcmp check will
361 * avoid the need for this more expensive check on almost all
366 l2 = &old_cg->cg_links;
368 struct cg_cgroup_link *cgl1, *cgl2;
369 struct cgroup *cg1, *cg2;
373 /* See if we reached the end - both lists are equal length. */
374 if (l1 == &cg->cg_links) {
375 BUG_ON(l2 != &old_cg->cg_links);
378 BUG_ON(l2 == &old_cg->cg_links);
380 /* Locate the cgroups associated with these links. */
381 cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
382 cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
385 /* Hierarchies should be linked in the same order. */
386 BUG_ON(cg1->root != cg2->root);
389 * If this hierarchy is the hierarchy of the cgroup
390 * that's changing, then we need to check that this
391 * css_set points to the new cgroup; if it's any other
392 * hierarchy, then this css_set should point to the
393 * same cgroup as the old css_set.
395 if (cg1->root == new_cgrp->root) {
407 * find_existing_css_set() is a helper for
408 * find_css_set(), and checks to see whether an existing
409 * css_set is suitable.
411 * oldcg: the cgroup group that we're using before the cgroup
414 * cgrp: the cgroup that we're moving into
416 * template: location in which to build the desired set of subsystem
417 * state objects for the new cgroup group
419 static struct css_set *find_existing_css_set(
420 struct css_set *oldcg,
422 struct cgroup_subsys_state *template[])
425 struct cgroupfs_root *root = cgrp->root;
426 struct hlist_head *hhead;
427 struct hlist_node *node;
430 /* Built the set of subsystem state objects that we want to
431 * see in the new css_set */
432 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
433 if (root->subsys_bits & (1UL << i)) {
434 /* Subsystem is in this hierarchy. So we want
435 * the subsystem state from the new
437 template[i] = cgrp->subsys[i];
439 /* Subsystem is not in this hierarchy, so we
440 * don't want to change the subsystem state */
441 template[i] = oldcg->subsys[i];
445 hhead = css_set_hash(template);
446 hlist_for_each_entry(cg, node, hhead, hlist) {
447 if (!compare_css_sets(cg, oldcg, cgrp, template))
450 /* This css_set matches what we need */
454 /* No existing cgroup group matched */
458 static void free_cg_links(struct list_head *tmp)
460 struct cg_cgroup_link *link;
461 struct cg_cgroup_link *saved_link;
463 list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
464 list_del(&link->cgrp_link_list);
470 * allocate_cg_links() allocates "count" cg_cgroup_link structures
471 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
472 * success or a negative error
474 static int allocate_cg_links(int count, struct list_head *tmp)
476 struct cg_cgroup_link *link;
479 for (i = 0; i < count; i++) {
480 link = kmalloc(sizeof(*link), GFP_KERNEL);
485 list_add(&link->cgrp_link_list, tmp);
491 * link_css_set - a helper function to link a css_set to a cgroup
492 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
493 * @cg: the css_set to be linked
494 * @cgrp: the destination cgroup
496 static void link_css_set(struct list_head *tmp_cg_links,
497 struct css_set *cg, struct cgroup *cgrp)
499 struct cg_cgroup_link *link;
501 BUG_ON(list_empty(tmp_cg_links));
502 link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
506 atomic_inc(&cgrp->count);
507 list_move(&link->cgrp_link_list, &cgrp->css_sets);
509 * Always add links to the tail of the list so that the list
510 * is sorted by order of hierarchy creation
512 list_add_tail(&link->cg_link_list, &cg->cg_links);
516 * find_css_set() takes an existing cgroup group and a
517 * cgroup object, and returns a css_set object that's
518 * equivalent to the old group, but with the given cgroup
519 * substituted into the appropriate hierarchy. Must be called with
522 static struct css_set *find_css_set(
523 struct css_set *oldcg, struct cgroup *cgrp)
526 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
528 struct list_head tmp_cg_links;
530 struct hlist_head *hhead;
531 struct cg_cgroup_link *link;
533 /* First see if we already have a cgroup group that matches
535 read_lock(&css_set_lock);
536 res = find_existing_css_set(oldcg, cgrp, template);
539 read_unlock(&css_set_lock);
544 res = kmalloc(sizeof(*res), GFP_KERNEL);
548 /* Allocate all the cg_cgroup_link objects that we'll need */
549 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
554 atomic_set(&res->refcount, 1);
555 INIT_LIST_HEAD(&res->cg_links);
556 INIT_LIST_HEAD(&res->tasks);
557 INIT_HLIST_NODE(&res->hlist);
559 /* Copy the set of subsystem state objects generated in
560 * find_existing_css_set() */
561 memcpy(res->subsys, template, sizeof(res->subsys));
563 write_lock(&css_set_lock);
564 /* Add reference counts and links from the new css_set. */
565 list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
566 struct cgroup *c = link->cgrp;
567 if (c->root == cgrp->root)
569 link_css_set(&tmp_cg_links, res, c);
572 BUG_ON(!list_empty(&tmp_cg_links));
576 /* Add this cgroup group to the hash table */
577 hhead = css_set_hash(res->subsys);
578 hlist_add_head(&res->hlist, hhead);
580 write_unlock(&css_set_lock);
586 * Return the cgroup for "task" from the given hierarchy. Must be
587 * called with cgroup_mutex held.
589 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
590 struct cgroupfs_root *root)
593 struct cgroup *res = NULL;
595 BUG_ON(!mutex_is_locked(&cgroup_mutex));
596 read_lock(&css_set_lock);
598 * No need to lock the task - since we hold cgroup_mutex the
599 * task can't change groups, so the only thing that can happen
600 * is that it exits and its css is set back to init_css_set.
603 if (css == &init_css_set) {
604 res = &root->top_cgroup;
606 struct cg_cgroup_link *link;
607 list_for_each_entry(link, &css->cg_links, cg_link_list) {
608 struct cgroup *c = link->cgrp;
609 if (c->root == root) {
615 read_unlock(&css_set_lock);
621 * There is one global cgroup mutex. We also require taking
622 * task_lock() when dereferencing a task's cgroup subsys pointers.
623 * See "The task_lock() exception", at the end of this comment.
625 * A task must hold cgroup_mutex to modify cgroups.
627 * Any task can increment and decrement the count field without lock.
628 * So in general, code holding cgroup_mutex can't rely on the count
629 * field not changing. However, if the count goes to zero, then only
630 * cgroup_attach_task() can increment it again. Because a count of zero
631 * means that no tasks are currently attached, therefore there is no
632 * way a task attached to that cgroup can fork (the other way to
633 * increment the count). So code holding cgroup_mutex can safely
634 * assume that if the count is zero, it will stay zero. Similarly, if
635 * a task holds cgroup_mutex on a cgroup with zero count, it
636 * knows that the cgroup won't be removed, as cgroup_rmdir()
639 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
640 * (usually) take cgroup_mutex. These are the two most performance
641 * critical pieces of code here. The exception occurs on cgroup_exit(),
642 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
643 * is taken, and if the cgroup count is zero, a usermode call made
644 * to the release agent with the name of the cgroup (path relative to
645 * the root of cgroup file system) as the argument.
647 * A cgroup can only be deleted if both its 'count' of using tasks
648 * is zero, and its list of 'children' cgroups is empty. Since all
649 * tasks in the system use _some_ cgroup, and since there is always at
650 * least one task in the system (init, pid == 1), therefore, top_cgroup
651 * always has either children cgroups and/or using tasks. So we don't
652 * need a special hack to ensure that top_cgroup cannot be deleted.
654 * The task_lock() exception
656 * The need for this exception arises from the action of
657 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
658 * another. It does so using cgroup_mutex, however there are
659 * several performance critical places that need to reference
660 * task->cgroup without the expense of grabbing a system global
661 * mutex. Therefore except as noted below, when dereferencing or, as
662 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
663 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
664 * the task_struct routinely used for such matters.
666 * P.S. One more locking exception. RCU is used to guard the
667 * update of a tasks cgroup pointer by cgroup_attach_task()
671 * cgroup_lock - lock out any changes to cgroup structures
674 void cgroup_lock(void)
676 mutex_lock(&cgroup_mutex);
680 * cgroup_unlock - release lock on cgroup changes
682 * Undo the lock taken in a previous cgroup_lock() call.
684 void cgroup_unlock(void)
686 mutex_unlock(&cgroup_mutex);
690 * A couple of forward declarations required, due to cyclic reference loop:
691 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
692 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
696 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode);
697 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
698 static int cgroup_populate_dir(struct cgroup *cgrp);
699 static const struct inode_operations cgroup_dir_inode_operations;
700 static struct file_operations proc_cgroupstats_operations;
702 static struct backing_dev_info cgroup_backing_dev_info = {
704 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
707 static int alloc_css_id(struct cgroup_subsys *ss,
708 struct cgroup *parent, struct cgroup *child);
710 static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb)
712 struct inode *inode = new_inode(sb);
715 inode->i_mode = mode;
716 inode->i_uid = current_fsuid();
717 inode->i_gid = current_fsgid();
718 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
719 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
725 * Call subsys's pre_destroy handler.
726 * This is called before css refcnt check.
728 static int cgroup_call_pre_destroy(struct cgroup *cgrp)
730 struct cgroup_subsys *ss;
733 for_each_subsys(cgrp->root, ss)
734 if (ss->pre_destroy) {
735 ret = ss->pre_destroy(ss, cgrp);
742 static void free_cgroup_rcu(struct rcu_head *obj)
744 struct cgroup *cgrp = container_of(obj, struct cgroup, rcu_head);
749 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
751 /* is dentry a directory ? if so, kfree() associated cgroup */
752 if (S_ISDIR(inode->i_mode)) {
753 struct cgroup *cgrp = dentry->d_fsdata;
754 struct cgroup_subsys *ss;
755 BUG_ON(!(cgroup_is_removed(cgrp)));
756 /* It's possible for external users to be holding css
757 * reference counts on a cgroup; css_put() needs to
758 * be able to access the cgroup after decrementing
759 * the reference count in order to know if it needs to
760 * queue the cgroup to be handled by the release
764 mutex_lock(&cgroup_mutex);
766 * Release the subsystem state objects.
768 for_each_subsys(cgrp->root, ss)
769 ss->destroy(ss, cgrp);
771 cgrp->root->number_of_cgroups--;
772 mutex_unlock(&cgroup_mutex);
775 * Drop the active superblock reference that we took when we
778 deactivate_super(cgrp->root->sb);
781 * if we're getting rid of the cgroup, refcount should ensure
782 * that there are no pidlists left.
784 BUG_ON(!list_empty(&cgrp->pidlists));
786 call_rcu(&cgrp->rcu_head, free_cgroup_rcu);
791 static void remove_dir(struct dentry *d)
793 struct dentry *parent = dget(d->d_parent);
796 simple_rmdir(parent->d_inode, d);
800 static void cgroup_clear_directory(struct dentry *dentry)
802 struct list_head *node;
804 BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
805 spin_lock(&dcache_lock);
806 node = dentry->d_subdirs.next;
807 while (node != &dentry->d_subdirs) {
808 struct dentry *d = list_entry(node, struct dentry, d_u.d_child);
811 /* This should never be called on a cgroup
812 * directory with child cgroups */
813 BUG_ON(d->d_inode->i_mode & S_IFDIR);
815 spin_unlock(&dcache_lock);
817 simple_unlink(dentry->d_inode, d);
819 spin_lock(&dcache_lock);
821 node = dentry->d_subdirs.next;
823 spin_unlock(&dcache_lock);
827 * NOTE : the dentry must have been dget()'ed
829 static void cgroup_d_remove_dir(struct dentry *dentry)
831 cgroup_clear_directory(dentry);
833 spin_lock(&dcache_lock);
834 list_del_init(&dentry->d_u.d_child);
835 spin_unlock(&dcache_lock);
840 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
841 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
842 * reference to css->refcnt. In general, this refcnt is expected to goes down
845 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
847 DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq);
849 static void cgroup_wakeup_rmdir_waiter(struct cgroup *cgrp)
851 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)))
852 wake_up_all(&cgroup_rmdir_waitq);
855 void cgroup_exclude_rmdir(struct cgroup_subsys_state *css)
860 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css)
862 cgroup_wakeup_rmdir_waiter(css->cgroup);
867 static int rebind_subsystems(struct cgroupfs_root *root,
868 unsigned long final_bits)
870 unsigned long added_bits, removed_bits;
871 struct cgroup *cgrp = &root->top_cgroup;
874 removed_bits = root->actual_subsys_bits & ~final_bits;
875 added_bits = final_bits & ~root->actual_subsys_bits;
876 /* Check that any added subsystems are currently free */
877 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
878 unsigned long bit = 1UL << i;
879 struct cgroup_subsys *ss = subsys[i];
880 if (!(bit & added_bits))
882 if (ss->root != &rootnode) {
883 /* Subsystem isn't free */
888 /* Currently we don't handle adding/removing subsystems when
889 * any child cgroups exist. This is theoretically supportable
890 * but involves complex error handling, so it's being left until
892 if (root->number_of_cgroups > 1)
895 /* Process each subsystem */
896 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
897 struct cgroup_subsys *ss = subsys[i];
898 unsigned long bit = 1UL << i;
899 if (bit & added_bits) {
900 /* We're binding this subsystem to this hierarchy */
901 BUG_ON(cgrp->subsys[i]);
902 BUG_ON(!dummytop->subsys[i]);
903 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
904 mutex_lock(&ss->hierarchy_mutex);
905 cgrp->subsys[i] = dummytop->subsys[i];
906 cgrp->subsys[i]->cgroup = cgrp;
907 list_move(&ss->sibling, &root->subsys_list);
911 mutex_unlock(&ss->hierarchy_mutex);
912 } else if (bit & removed_bits) {
913 /* We're removing this subsystem */
914 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
915 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
916 mutex_lock(&ss->hierarchy_mutex);
918 ss->bind(ss, dummytop);
919 dummytop->subsys[i]->cgroup = dummytop;
920 cgrp->subsys[i] = NULL;
921 subsys[i]->root = &rootnode;
922 list_move(&ss->sibling, &rootnode.subsys_list);
923 mutex_unlock(&ss->hierarchy_mutex);
924 } else if (bit & final_bits) {
925 /* Subsystem state should already exist */
926 BUG_ON(!cgrp->subsys[i]);
928 /* Subsystem state shouldn't exist */
929 BUG_ON(cgrp->subsys[i]);
932 root->subsys_bits = root->actual_subsys_bits = final_bits;
938 static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs)
940 struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info;
941 struct cgroup_subsys *ss;
943 mutex_lock(&cgroup_mutex);
944 for_each_subsys(root, ss)
945 seq_printf(seq, ",%s", ss->name);
946 if (test_bit(ROOT_NOPREFIX, &root->flags))
947 seq_puts(seq, ",noprefix");
948 if (strlen(root->release_agent_path))
949 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
950 if (strlen(root->name))
951 seq_printf(seq, ",name=%s", root->name);
952 mutex_unlock(&cgroup_mutex);
956 struct cgroup_sb_opts {
957 unsigned long subsys_bits;
961 /* User explicitly requested empty subsystem */
964 struct cgroupfs_root *new_root;
968 /* Convert a hierarchy specifier into a bitmask of subsystems and
970 static int parse_cgroupfs_options(char *data,
971 struct cgroup_sb_opts *opts)
973 char *token, *o = data ?: "all";
974 unsigned long mask = (unsigned long)-1;
976 #ifdef CONFIG_CPUSETS
977 mask = ~(1UL << cpuset_subsys_id);
980 memset(opts, 0, sizeof(*opts));
982 while ((token = strsep(&o, ",")) != NULL) {
985 if (!strcmp(token, "all")) {
986 /* Add all non-disabled subsystems */
988 opts->subsys_bits = 0;
989 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
990 struct cgroup_subsys *ss = subsys[i];
992 opts->subsys_bits |= 1ul << i;
994 } else if (!strcmp(token, "none")) {
995 /* Explicitly have no subsystems */
997 } else if (!strcmp(token, "noprefix")) {
998 set_bit(ROOT_NOPREFIX, &opts->flags);
999 } else if (!strncmp(token, "release_agent=", 14)) {
1000 /* Specifying two release agents is forbidden */
1001 if (opts->release_agent)
1003 opts->release_agent =
1004 kstrndup(token + 14, PATH_MAX, GFP_KERNEL);
1005 if (!opts->release_agent)
1007 } else if (!strncmp(token, "name=", 5)) {
1009 const char *name = token + 5;
1010 /* Can't specify an empty name */
1013 /* Must match [\w.-]+ */
1014 for (i = 0; i < strlen(name); i++) {
1018 if ((c == '.') || (c == '-') || (c == '_'))
1022 /* Specifying two names is forbidden */
1025 opts->name = kstrndup(name,
1026 MAX_CGROUP_ROOT_NAMELEN,
1031 struct cgroup_subsys *ss;
1033 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1035 if (!strcmp(token, ss->name)) {
1037 set_bit(i, &opts->subsys_bits);
1041 if (i == CGROUP_SUBSYS_COUNT)
1046 /* Consistency checks */
1049 * Option noprefix was introduced just for backward compatibility
1050 * with the old cpuset, so we allow noprefix only if mounting just
1051 * the cpuset subsystem.
1053 if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1054 (opts->subsys_bits & mask))
1058 /* Can't specify "none" and some subsystems */
1059 if (opts->subsys_bits && opts->none)
1063 * We either have to specify by name or by subsystems. (So all
1064 * empty hierarchies must have a name).
1066 if (!opts->subsys_bits && !opts->name)
1072 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1075 struct cgroupfs_root *root = sb->s_fs_info;
1076 struct cgroup *cgrp = &root->top_cgroup;
1077 struct cgroup_sb_opts opts;
1080 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1081 mutex_lock(&cgroup_mutex);
1083 /* See what subsystems are wanted */
1084 ret = parse_cgroupfs_options(data, &opts);
1088 /* Don't allow flags to change at remount */
1089 if (opts.flags != root->flags) {
1094 /* Don't allow name to change at remount */
1095 if (opts.name && strcmp(opts.name, root->name)) {
1100 ret = rebind_subsystems(root, opts.subsys_bits);
1104 /* (re)populate subsystem files */
1105 cgroup_populate_dir(cgrp);
1107 if (opts.release_agent)
1108 strcpy(root->release_agent_path, opts.release_agent);
1110 kfree(opts.release_agent);
1112 mutex_unlock(&cgroup_mutex);
1113 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1118 static const struct super_operations cgroup_ops = {
1119 .statfs = simple_statfs,
1120 .drop_inode = generic_delete_inode,
1121 .show_options = cgroup_show_options,
1122 .remount_fs = cgroup_remount,
1125 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1127 INIT_LIST_HEAD(&cgrp->sibling);
1128 INIT_LIST_HEAD(&cgrp->children);
1129 INIT_LIST_HEAD(&cgrp->css_sets);
1130 INIT_LIST_HEAD(&cgrp->release_list);
1131 INIT_LIST_HEAD(&cgrp->pidlists);
1132 mutex_init(&cgrp->pidlist_mutex);
1135 static void init_cgroup_root(struct cgroupfs_root *root)
1137 struct cgroup *cgrp = &root->top_cgroup;
1138 INIT_LIST_HEAD(&root->subsys_list);
1139 INIT_LIST_HEAD(&root->root_list);
1140 root->number_of_cgroups = 1;
1142 cgrp->top_cgroup = cgrp;
1143 init_cgroup_housekeeping(cgrp);
1146 static bool init_root_id(struct cgroupfs_root *root)
1151 if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1153 spin_lock(&hierarchy_id_lock);
1154 /* Try to allocate the next unused ID */
1155 ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1156 &root->hierarchy_id);
1158 /* Try again starting from 0 */
1159 ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1161 next_hierarchy_id = root->hierarchy_id + 1;
1162 } else if (ret != -EAGAIN) {
1163 /* Can only get here if the 31-bit IDR is full ... */
1166 spin_unlock(&hierarchy_id_lock);
1171 static int cgroup_test_super(struct super_block *sb, void *data)
1173 struct cgroup_sb_opts *opts = data;
1174 struct cgroupfs_root *root = sb->s_fs_info;
1176 /* If we asked for a name then it must match */
1177 if (opts->name && strcmp(opts->name, root->name))
1181 * If we asked for subsystems (or explicitly for no
1182 * subsystems) then they must match
1184 if ((opts->subsys_bits || opts->none)
1185 && (opts->subsys_bits != root->subsys_bits))
1191 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1193 struct cgroupfs_root *root;
1195 if (!opts->subsys_bits && !opts->none)
1198 root = kzalloc(sizeof(*root), GFP_KERNEL);
1200 return ERR_PTR(-ENOMEM);
1202 if (!init_root_id(root)) {
1204 return ERR_PTR(-ENOMEM);
1206 init_cgroup_root(root);
1208 root->subsys_bits = opts->subsys_bits;
1209 root->flags = opts->flags;
1210 if (opts->release_agent)
1211 strcpy(root->release_agent_path, opts->release_agent);
1213 strcpy(root->name, opts->name);
1217 static void cgroup_drop_root(struct cgroupfs_root *root)
1222 BUG_ON(!root->hierarchy_id);
1223 spin_lock(&hierarchy_id_lock);
1224 ida_remove(&hierarchy_ida, root->hierarchy_id);
1225 spin_unlock(&hierarchy_id_lock);
1229 static int cgroup_set_super(struct super_block *sb, void *data)
1232 struct cgroup_sb_opts *opts = data;
1234 /* If we don't have a new root, we can't set up a new sb */
1235 if (!opts->new_root)
1238 BUG_ON(!opts->subsys_bits && !opts->none);
1240 ret = set_anon_super(sb, NULL);
1244 sb->s_fs_info = opts->new_root;
1245 opts->new_root->sb = sb;
1247 sb->s_blocksize = PAGE_CACHE_SIZE;
1248 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1249 sb->s_magic = CGROUP_SUPER_MAGIC;
1250 sb->s_op = &cgroup_ops;
1255 static int cgroup_get_rootdir(struct super_block *sb)
1257 struct inode *inode =
1258 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1259 struct dentry *dentry;
1264 inode->i_fop = &simple_dir_operations;
1265 inode->i_op = &cgroup_dir_inode_operations;
1266 /* directories start off with i_nlink == 2 (for "." entry) */
1268 dentry = d_alloc_root(inode);
1273 sb->s_root = dentry;
1277 static int cgroup_get_sb(struct file_system_type *fs_type,
1278 int flags, const char *unused_dev_name,
1279 void *data, struct vfsmount *mnt)
1281 struct cgroup_sb_opts opts;
1282 struct cgroupfs_root *root;
1284 struct super_block *sb;
1285 struct cgroupfs_root *new_root;
1287 /* First find the desired set of subsystems */
1288 ret = parse_cgroupfs_options(data, &opts);
1293 * Allocate a new cgroup root. We may not need it if we're
1294 * reusing an existing hierarchy.
1296 new_root = cgroup_root_from_opts(&opts);
1297 if (IS_ERR(new_root)) {
1298 ret = PTR_ERR(new_root);
1301 opts.new_root = new_root;
1303 /* Locate an existing or new sb for this hierarchy */
1304 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, &opts);
1307 cgroup_drop_root(opts.new_root);
1311 root = sb->s_fs_info;
1313 if (root == opts.new_root) {
1314 /* We used the new root structure, so this is a new hierarchy */
1315 struct list_head tmp_cg_links;
1316 struct cgroup *root_cgrp = &root->top_cgroup;
1317 struct inode *inode;
1318 struct cgroupfs_root *existing_root;
1321 BUG_ON(sb->s_root != NULL);
1323 ret = cgroup_get_rootdir(sb);
1325 goto drop_new_super;
1326 inode = sb->s_root->d_inode;
1328 mutex_lock(&inode->i_mutex);
1329 mutex_lock(&cgroup_mutex);
1331 if (strlen(root->name)) {
1332 /* Check for name clashes with existing mounts */
1333 for_each_active_root(existing_root) {
1334 if (!strcmp(existing_root->name, root->name)) {
1336 mutex_unlock(&cgroup_mutex);
1337 mutex_unlock(&inode->i_mutex);
1338 goto drop_new_super;
1344 * We're accessing css_set_count without locking
1345 * css_set_lock here, but that's OK - it can only be
1346 * increased by someone holding cgroup_lock, and
1347 * that's us. The worst that can happen is that we
1348 * have some link structures left over
1350 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1352 mutex_unlock(&cgroup_mutex);
1353 mutex_unlock(&inode->i_mutex);
1354 goto drop_new_super;
1357 ret = rebind_subsystems(root, root->subsys_bits);
1358 if (ret == -EBUSY) {
1359 mutex_unlock(&cgroup_mutex);
1360 mutex_unlock(&inode->i_mutex);
1361 free_cg_links(&tmp_cg_links);
1362 goto drop_new_super;
1365 /* EBUSY should be the only error here */
1368 list_add(&root->root_list, &roots);
1371 sb->s_root->d_fsdata = root_cgrp;
1372 root->top_cgroup.dentry = sb->s_root;
1374 /* Link the top cgroup in this hierarchy into all
1375 * the css_set objects */
1376 write_lock(&css_set_lock);
1377 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
1378 struct hlist_head *hhead = &css_set_table[i];
1379 struct hlist_node *node;
1382 hlist_for_each_entry(cg, node, hhead, hlist)
1383 link_css_set(&tmp_cg_links, cg, root_cgrp);
1385 write_unlock(&css_set_lock);
1387 free_cg_links(&tmp_cg_links);
1389 BUG_ON(!list_empty(&root_cgrp->sibling));
1390 BUG_ON(!list_empty(&root_cgrp->children));
1391 BUG_ON(root->number_of_cgroups != 1);
1393 cgroup_populate_dir(root_cgrp);
1394 mutex_unlock(&cgroup_mutex);
1395 mutex_unlock(&inode->i_mutex);
1398 * We re-used an existing hierarchy - the new root (if
1399 * any) is not needed
1401 cgroup_drop_root(opts.new_root);
1404 simple_set_mnt(mnt, sb);
1405 kfree(opts.release_agent);
1410 deactivate_locked_super(sb);
1412 kfree(opts.release_agent);
1418 static void cgroup_kill_sb(struct super_block *sb) {
1419 struct cgroupfs_root *root = sb->s_fs_info;
1420 struct cgroup *cgrp = &root->top_cgroup;
1422 struct cg_cgroup_link *link;
1423 struct cg_cgroup_link *saved_link;
1427 BUG_ON(root->number_of_cgroups != 1);
1428 BUG_ON(!list_empty(&cgrp->children));
1429 BUG_ON(!list_empty(&cgrp->sibling));
1431 mutex_lock(&cgroup_mutex);
1433 /* Rebind all subsystems back to the default hierarchy */
1434 ret = rebind_subsystems(root, 0);
1435 /* Shouldn't be able to fail ... */
1439 * Release all the links from css_sets to this hierarchy's
1442 write_lock(&css_set_lock);
1444 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1446 list_del(&link->cg_link_list);
1447 list_del(&link->cgrp_link_list);
1450 write_unlock(&css_set_lock);
1452 if (!list_empty(&root->root_list)) {
1453 list_del(&root->root_list);
1457 mutex_unlock(&cgroup_mutex);
1459 kill_litter_super(sb);
1460 cgroup_drop_root(root);
1463 static struct file_system_type cgroup_fs_type = {
1465 .get_sb = cgroup_get_sb,
1466 .kill_sb = cgroup_kill_sb,
1469 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
1471 return dentry->d_fsdata;
1474 static inline struct cftype *__d_cft(struct dentry *dentry)
1476 return dentry->d_fsdata;
1480 * cgroup_path - generate the path of a cgroup
1481 * @cgrp: the cgroup in question
1482 * @buf: the buffer to write the path into
1483 * @buflen: the length of the buffer
1485 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1486 * reference. Writes path of cgroup into buf. Returns 0 on success,
1489 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1492 struct dentry *dentry = rcu_dereference(cgrp->dentry);
1494 if (!dentry || cgrp == dummytop) {
1496 * Inactive subsystems have no dentry for their root
1503 start = buf + buflen;
1507 int len = dentry->d_name.len;
1508 if ((start -= len) < buf)
1509 return -ENAMETOOLONG;
1510 memcpy(start, cgrp->dentry->d_name.name, len);
1511 cgrp = cgrp->parent;
1514 dentry = rcu_dereference(cgrp->dentry);
1518 return -ENAMETOOLONG;
1521 memmove(buf, start, buf + buflen - start);
1526 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1527 * @cgrp: the cgroup the task is attaching to
1528 * @tsk: the task to be attached
1530 * Call holding cgroup_mutex. May take task_lock of
1531 * the task 'tsk' during call.
1533 int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1536 struct cgroup_subsys *ss;
1537 struct cgroup *oldcgrp;
1539 struct css_set *newcg;
1540 struct cgroupfs_root *root = cgrp->root;
1542 /* Nothing to do if the task is already in that cgroup */
1543 oldcgrp = task_cgroup_from_root(tsk, root);
1544 if (cgrp == oldcgrp)
1547 for_each_subsys(root, ss) {
1548 if (ss->can_attach) {
1549 retval = ss->can_attach(ss, cgrp, tsk);
1560 * Locate or allocate a new css_set for this task,
1561 * based on its final set of cgroups
1563 newcg = find_css_set(cg, cgrp);
1569 if (tsk->flags & PF_EXITING) {
1574 rcu_assign_pointer(tsk->cgroups, newcg);
1577 /* Update the css_set linked lists if we're using them */
1578 write_lock(&css_set_lock);
1579 if (!list_empty(&tsk->cg_list)) {
1580 list_del(&tsk->cg_list);
1581 list_add(&tsk->cg_list, &newcg->tasks);
1583 write_unlock(&css_set_lock);
1585 for_each_subsys(root, ss) {
1587 ss->attach(ss, cgrp, oldcgrp, tsk);
1589 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1594 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1595 * is no longer empty.
1597 cgroup_wakeup_rmdir_waiter(cgrp);
1602 * Attach task with pid 'pid' to cgroup 'cgrp'. Call with cgroup_mutex
1603 * held. May take task_lock of task
1605 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid)
1607 struct task_struct *tsk;
1608 const struct cred *cred = current_cred(), *tcred;
1613 tsk = find_task_by_vpid(pid);
1614 if (!tsk || tsk->flags & PF_EXITING) {
1619 tcred = __task_cred(tsk);
1621 cred->euid != tcred->uid &&
1622 cred->euid != tcred->suid) {
1626 get_task_struct(tsk);
1630 get_task_struct(tsk);
1633 ret = cgroup_attach_task(cgrp, tsk);
1634 put_task_struct(tsk);
1638 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
1641 if (!cgroup_lock_live_group(cgrp))
1643 ret = attach_task_by_pid(cgrp, pid);
1649 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
1650 * @cgrp: the cgroup to be checked for liveness
1652 * On success, returns true; the lock should be later released with
1653 * cgroup_unlock(). On failure returns false with no lock held.
1655 bool cgroup_lock_live_group(struct cgroup *cgrp)
1657 mutex_lock(&cgroup_mutex);
1658 if (cgroup_is_removed(cgrp)) {
1659 mutex_unlock(&cgroup_mutex);
1665 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
1668 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
1669 if (!cgroup_lock_live_group(cgrp))
1671 strcpy(cgrp->root->release_agent_path, buffer);
1676 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
1677 struct seq_file *seq)
1679 if (!cgroup_lock_live_group(cgrp))
1681 seq_puts(seq, cgrp->root->release_agent_path);
1682 seq_putc(seq, '\n');
1687 /* A buffer size big enough for numbers or short strings */
1688 #define CGROUP_LOCAL_BUFFER_SIZE 64
1690 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
1692 const char __user *userbuf,
1693 size_t nbytes, loff_t *unused_ppos)
1695 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
1701 if (nbytes >= sizeof(buffer))
1703 if (copy_from_user(buffer, userbuf, nbytes))
1706 buffer[nbytes] = 0; /* nul-terminate */
1708 if (cft->write_u64) {
1709 u64 val = simple_strtoull(buffer, &end, 0);
1712 retval = cft->write_u64(cgrp, cft, val);
1714 s64 val = simple_strtoll(buffer, &end, 0);
1717 retval = cft->write_s64(cgrp, cft, val);
1724 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
1726 const char __user *userbuf,
1727 size_t nbytes, loff_t *unused_ppos)
1729 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
1731 size_t max_bytes = cft->max_write_len;
1732 char *buffer = local_buffer;
1735 max_bytes = sizeof(local_buffer) - 1;
1736 if (nbytes >= max_bytes)
1738 /* Allocate a dynamic buffer if we need one */
1739 if (nbytes >= sizeof(local_buffer)) {
1740 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
1744 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
1749 buffer[nbytes] = 0; /* nul-terminate */
1751 retval = cft->write_string(cgrp, cft, buffer);
1755 if (buffer != local_buffer)
1760 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
1761 size_t nbytes, loff_t *ppos)
1763 struct cftype *cft = __d_cft(file->f_dentry);
1764 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
1766 if (cgroup_is_removed(cgrp))
1769 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
1770 if (cft->write_u64 || cft->write_s64)
1771 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
1772 if (cft->write_string)
1773 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
1775 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
1776 return ret ? ret : nbytes;
1781 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
1783 char __user *buf, size_t nbytes,
1786 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
1787 u64 val = cft->read_u64(cgrp, cft);
1788 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
1790 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
1793 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
1795 char __user *buf, size_t nbytes,
1798 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
1799 s64 val = cft->read_s64(cgrp, cft);
1800 int len = sprintf(tmp, "%lld\n", (long long) val);
1802 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
1805 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
1806 size_t nbytes, loff_t *ppos)
1808 struct cftype *cft = __d_cft(file->f_dentry);
1809 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
1811 if (cgroup_is_removed(cgrp))
1815 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
1817 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
1819 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
1824 * seqfile ops/methods for returning structured data. Currently just
1825 * supports string->u64 maps, but can be extended in future.
1828 struct cgroup_seqfile_state {
1830 struct cgroup *cgroup;
1833 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
1835 struct seq_file *sf = cb->state;
1836 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
1839 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
1841 struct cgroup_seqfile_state *state = m->private;
1842 struct cftype *cft = state->cft;
1843 if (cft->read_map) {
1844 struct cgroup_map_cb cb = {
1845 .fill = cgroup_map_add,
1848 return cft->read_map(state->cgroup, cft, &cb);
1850 return cft->read_seq_string(state->cgroup, cft, m);
1853 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
1855 struct seq_file *seq = file->private_data;
1856 kfree(seq->private);
1857 return single_release(inode, file);
1860 static struct file_operations cgroup_seqfile_operations = {
1862 .write = cgroup_file_write,
1863 .llseek = seq_lseek,
1864 .release = cgroup_seqfile_release,
1867 static int cgroup_file_open(struct inode *inode, struct file *file)
1872 err = generic_file_open(inode, file);
1875 cft = __d_cft(file->f_dentry);
1877 if (cft->read_map || cft->read_seq_string) {
1878 struct cgroup_seqfile_state *state =
1879 kzalloc(sizeof(*state), GFP_USER);
1883 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
1884 file->f_op = &cgroup_seqfile_operations;
1885 err = single_open(file, cgroup_seqfile_show, state);
1888 } else if (cft->open)
1889 err = cft->open(inode, file);
1896 static int cgroup_file_release(struct inode *inode, struct file *file)
1898 struct cftype *cft = __d_cft(file->f_dentry);
1900 return cft->release(inode, file);
1905 * cgroup_rename - Only allow simple rename of directories in place.
1907 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
1908 struct inode *new_dir, struct dentry *new_dentry)
1910 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1912 if (new_dentry->d_inode)
1914 if (old_dir != new_dir)
1916 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
1919 static struct file_operations cgroup_file_operations = {
1920 .read = cgroup_file_read,
1921 .write = cgroup_file_write,
1922 .llseek = generic_file_llseek,
1923 .open = cgroup_file_open,
1924 .release = cgroup_file_release,
1927 static const struct inode_operations cgroup_dir_inode_operations = {
1928 .lookup = simple_lookup,
1929 .mkdir = cgroup_mkdir,
1930 .rmdir = cgroup_rmdir,
1931 .rename = cgroup_rename,
1934 static int cgroup_create_file(struct dentry *dentry, mode_t mode,
1935 struct super_block *sb)
1937 static const struct dentry_operations cgroup_dops = {
1938 .d_iput = cgroup_diput,
1941 struct inode *inode;
1945 if (dentry->d_inode)
1948 inode = cgroup_new_inode(mode, sb);
1952 if (S_ISDIR(mode)) {
1953 inode->i_op = &cgroup_dir_inode_operations;
1954 inode->i_fop = &simple_dir_operations;
1956 /* start off with i_nlink == 2 (for "." entry) */
1959 /* start with the directory inode held, so that we can
1960 * populate it without racing with another mkdir */
1961 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1962 } else if (S_ISREG(mode)) {
1964 inode->i_fop = &cgroup_file_operations;
1966 dentry->d_op = &cgroup_dops;
1967 d_instantiate(dentry, inode);
1968 dget(dentry); /* Extra count - pin the dentry in core */
1973 * cgroup_create_dir - create a directory for an object.
1974 * @cgrp: the cgroup we create the directory for. It must have a valid
1975 * ->parent field. And we are going to fill its ->dentry field.
1976 * @dentry: dentry of the new cgroup
1977 * @mode: mode to set on new directory.
1979 static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
1982 struct dentry *parent;
1985 parent = cgrp->parent->dentry;
1986 error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
1988 dentry->d_fsdata = cgrp;
1989 inc_nlink(parent->d_inode);
1990 rcu_assign_pointer(cgrp->dentry, dentry);
1999 * cgroup_file_mode - deduce file mode of a control file
2000 * @cft: the control file in question
2002 * returns cft->mode if ->mode is not 0
2003 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2004 * returns S_IRUGO if it has only a read handler
2005 * returns S_IWUSR if it has only a write hander
2007 static mode_t cgroup_file_mode(const struct cftype *cft)
2014 if (cft->read || cft->read_u64 || cft->read_s64 ||
2015 cft->read_map || cft->read_seq_string)
2018 if (cft->write || cft->write_u64 || cft->write_s64 ||
2019 cft->write_string || cft->trigger)
2025 int cgroup_add_file(struct cgroup *cgrp,
2026 struct cgroup_subsys *subsys,
2027 const struct cftype *cft)
2029 struct dentry *dir = cgrp->dentry;
2030 struct dentry *dentry;
2034 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2035 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
2036 strcpy(name, subsys->name);
2039 strcat(name, cft->name);
2040 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2041 dentry = lookup_one_len(name, dir, strlen(name));
2042 if (!IS_ERR(dentry)) {
2043 mode = cgroup_file_mode(cft);
2044 error = cgroup_create_file(dentry, mode | S_IFREG,
2047 dentry->d_fsdata = (void *)cft;
2050 error = PTR_ERR(dentry);
2054 int cgroup_add_files(struct cgroup *cgrp,
2055 struct cgroup_subsys *subsys,
2056 const struct cftype cft[],
2060 for (i = 0; i < count; i++) {
2061 err = cgroup_add_file(cgrp, subsys, &cft[i]);
2069 * cgroup_task_count - count the number of tasks in a cgroup.
2070 * @cgrp: the cgroup in question
2072 * Return the number of tasks in the cgroup.
2074 int cgroup_task_count(const struct cgroup *cgrp)
2077 struct cg_cgroup_link *link;
2079 read_lock(&css_set_lock);
2080 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2081 count += atomic_read(&link->cg->refcount);
2083 read_unlock(&css_set_lock);
2088 * Advance a list_head iterator. The iterator should be positioned at
2089 * the start of a css_set
2091 static void cgroup_advance_iter(struct cgroup *cgrp,
2092 struct cgroup_iter *it)
2094 struct list_head *l = it->cg_link;
2095 struct cg_cgroup_link *link;
2098 /* Advance to the next non-empty css_set */
2101 if (l == &cgrp->css_sets) {
2105 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2107 } while (list_empty(&cg->tasks));
2109 it->task = cg->tasks.next;
2113 * To reduce the fork() overhead for systems that are not actually
2114 * using their cgroups capability, we don't maintain the lists running
2115 * through each css_set to its tasks until we see the list actually
2116 * used - in other words after the first call to cgroup_iter_start().
2118 * The tasklist_lock is not held here, as do_each_thread() and
2119 * while_each_thread() are protected by RCU.
2121 static void cgroup_enable_task_cg_lists(void)
2123 struct task_struct *p, *g;
2124 write_lock(&css_set_lock);
2125 use_task_css_set_links = 1;
2126 do_each_thread(g, p) {
2129 * We should check if the process is exiting, otherwise
2130 * it will race with cgroup_exit() in that the list
2131 * entry won't be deleted though the process has exited.
2133 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2134 list_add(&p->cg_list, &p->cgroups->tasks);
2136 } while_each_thread(g, p);
2137 write_unlock(&css_set_lock);
2140 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
2143 * The first time anyone tries to iterate across a cgroup,
2144 * we need to enable the list linking each css_set to its
2145 * tasks, and fix up all existing tasks.
2147 if (!use_task_css_set_links)
2148 cgroup_enable_task_cg_lists();
2150 read_lock(&css_set_lock);
2151 it->cg_link = &cgrp->css_sets;
2152 cgroup_advance_iter(cgrp, it);
2155 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
2156 struct cgroup_iter *it)
2158 struct task_struct *res;
2159 struct list_head *l = it->task;
2160 struct cg_cgroup_link *link;
2162 /* If the iterator cg is NULL, we have no tasks */
2165 res = list_entry(l, struct task_struct, cg_list);
2166 /* Advance iterator to find next entry */
2168 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
2169 if (l == &link->cg->tasks) {
2170 /* We reached the end of this task list - move on to
2171 * the next cg_cgroup_link */
2172 cgroup_advance_iter(cgrp, it);
2179 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
2181 read_unlock(&css_set_lock);
2184 static inline int started_after_time(struct task_struct *t1,
2185 struct timespec *time,
2186 struct task_struct *t2)
2188 int start_diff = timespec_compare(&t1->start_time, time);
2189 if (start_diff > 0) {
2191 } else if (start_diff < 0) {
2195 * Arbitrarily, if two processes started at the same
2196 * time, we'll say that the lower pointer value
2197 * started first. Note that t2 may have exited by now
2198 * so this may not be a valid pointer any longer, but
2199 * that's fine - it still serves to distinguish
2200 * between two tasks started (effectively) simultaneously.
2207 * This function is a callback from heap_insert() and is used to order
2209 * In this case we order the heap in descending task start time.
2211 static inline int started_after(void *p1, void *p2)
2213 struct task_struct *t1 = p1;
2214 struct task_struct *t2 = p2;
2215 return started_after_time(t1, &t2->start_time, t2);
2219 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
2220 * @scan: struct cgroup_scanner containing arguments for the scan
2222 * Arguments include pointers to callback functions test_task() and
2224 * Iterate through all the tasks in a cgroup, calling test_task() for each,
2225 * and if it returns true, call process_task() for it also.
2226 * The test_task pointer may be NULL, meaning always true (select all tasks).
2227 * Effectively duplicates cgroup_iter_{start,next,end}()
2228 * but does not lock css_set_lock for the call to process_task().
2229 * The struct cgroup_scanner may be embedded in any structure of the caller's
2231 * It is guaranteed that process_task() will act on every task that
2232 * is a member of the cgroup for the duration of this call. This
2233 * function may or may not call process_task() for tasks that exit
2234 * or move to a different cgroup during the call, or are forked or
2235 * move into the cgroup during the call.
2237 * Note that test_task() may be called with locks held, and may in some
2238 * situations be called multiple times for the same task, so it should
2240 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
2241 * pre-allocated and will be used for heap operations (and its "gt" member will
2242 * be overwritten), else a temporary heap will be used (allocation of which
2243 * may cause this function to fail).
2245 int cgroup_scan_tasks(struct cgroup_scanner *scan)
2248 struct cgroup_iter it;
2249 struct task_struct *p, *dropped;
2250 /* Never dereference latest_task, since it's not refcounted */
2251 struct task_struct *latest_task = NULL;
2252 struct ptr_heap tmp_heap;
2253 struct ptr_heap *heap;
2254 struct timespec latest_time = { 0, 0 };
2257 /* The caller supplied our heap and pre-allocated its memory */
2259 heap->gt = &started_after;
2261 /* We need to allocate our own heap memory */
2263 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
2265 /* cannot allocate the heap */
2271 * Scan tasks in the cgroup, using the scanner's "test_task" callback
2272 * to determine which are of interest, and using the scanner's
2273 * "process_task" callback to process any of them that need an update.
2274 * Since we don't want to hold any locks during the task updates,
2275 * gather tasks to be processed in a heap structure.
2276 * The heap is sorted by descending task start time.
2277 * If the statically-sized heap fills up, we overflow tasks that
2278 * started later, and in future iterations only consider tasks that
2279 * started after the latest task in the previous pass. This
2280 * guarantees forward progress and that we don't miss any tasks.
2283 cgroup_iter_start(scan->cg, &it);
2284 while ((p = cgroup_iter_next(scan->cg, &it))) {
2286 * Only affect tasks that qualify per the caller's callback,
2287 * if he provided one
2289 if (scan->test_task && !scan->test_task(p, scan))
2292 * Only process tasks that started after the last task
2295 if (!started_after_time(p, &latest_time, latest_task))
2297 dropped = heap_insert(heap, p);
2298 if (dropped == NULL) {
2300 * The new task was inserted; the heap wasn't
2304 } else if (dropped != p) {
2306 * The new task was inserted, and pushed out a
2310 put_task_struct(dropped);
2313 * Else the new task was newer than anything already in
2314 * the heap and wasn't inserted
2317 cgroup_iter_end(scan->cg, &it);
2320 for (i = 0; i < heap->size; i++) {
2321 struct task_struct *q = heap->ptrs[i];
2323 latest_time = q->start_time;
2326 /* Process the task per the caller's callback */
2327 scan->process_task(q, scan);
2331 * If we had to process any tasks at all, scan again
2332 * in case some of them were in the middle of forking
2333 * children that didn't get processed.
2334 * Not the most efficient way to do it, but it avoids
2335 * having to take callback_mutex in the fork path
2339 if (heap == &tmp_heap)
2340 heap_free(&tmp_heap);
2345 * Stuff for reading the 'tasks'/'procs' files.
2347 * Reading this file can return large amounts of data if a cgroup has
2348 * *lots* of attached tasks. So it may need several calls to read(),
2349 * but we cannot guarantee that the information we produce is correct
2350 * unless we produce it entirely atomically.
2355 * The following two functions "fix" the issue where there are more pids
2356 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
2357 * TODO: replace with a kernel-wide solution to this problem
2359 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
2360 static void *pidlist_allocate(int count)
2362 if (PIDLIST_TOO_LARGE(count))
2363 return vmalloc(count * sizeof(pid_t));
2365 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
2367 static void pidlist_free(void *p)
2369 if (is_vmalloc_addr(p))
2374 static void *pidlist_resize(void *p, int newcount)
2377 /* note: if new alloc fails, old p will still be valid either way */
2378 if (is_vmalloc_addr(p)) {
2379 newlist = vmalloc(newcount * sizeof(pid_t));
2382 memcpy(newlist, p, newcount * sizeof(pid_t));
2385 newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL);
2391 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
2392 * If the new stripped list is sufficiently smaller and there's enough memory
2393 * to allocate a new buffer, will let go of the unneeded memory. Returns the
2394 * number of unique elements.
2396 /* is the size difference enough that we should re-allocate the array? */
2397 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
2398 static int pidlist_uniq(pid_t **p, int length)
2405 * we presume the 0th element is unique, so i starts at 1. trivial
2406 * edge cases first; no work needs to be done for either
2408 if (length == 0 || length == 1)
2410 /* src and dest walk down the list; dest counts unique elements */
2411 for (src = 1; src < length; src++) {
2412 /* find next unique element */
2413 while (list[src] == list[src-1]) {
2418 /* dest always points to where the next unique element goes */
2419 list[dest] = list[src];
2424 * if the length difference is large enough, we want to allocate a
2425 * smaller buffer to save memory. if this fails due to out of memory,
2426 * we'll just stay with what we've got.
2428 if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) {
2429 newlist = pidlist_resize(list, dest);
2436 static int cmppid(const void *a, const void *b)
2438 return *(pid_t *)a - *(pid_t *)b;
2442 * find the appropriate pidlist for our purpose (given procs vs tasks)
2443 * returns with the lock on that pidlist already held, and takes care
2444 * of the use count, or returns NULL with no locks held if we're out of
2447 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
2448 enum cgroup_filetype type)
2450 struct cgroup_pidlist *l;
2451 /* don't need task_nsproxy() if we're looking at ourself */
2452 struct pid_namespace *ns = get_pid_ns(current->nsproxy->pid_ns);
2454 * We can't drop the pidlist_mutex before taking the l->mutex in case
2455 * the last ref-holder is trying to remove l from the list at the same
2456 * time. Holding the pidlist_mutex precludes somebody taking whichever
2457 * list we find out from under us - compare release_pid_array().
2459 mutex_lock(&cgrp->pidlist_mutex);
2460 list_for_each_entry(l, &cgrp->pidlists, links) {
2461 if (l->key.type == type && l->key.ns == ns) {
2462 /* found a matching list - drop the extra refcount */
2464 /* make sure l doesn't vanish out from under us */
2465 down_write(&l->mutex);
2466 mutex_unlock(&cgrp->pidlist_mutex);
2471 /* entry not found; create a new one */
2472 l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
2474 mutex_unlock(&cgrp->pidlist_mutex);
2478 init_rwsem(&l->mutex);
2479 down_write(&l->mutex);
2482 l->use_count = 0; /* don't increment here */
2485 list_add(&l->links, &cgrp->pidlists);
2486 mutex_unlock(&cgrp->pidlist_mutex);
2491 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
2493 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
2494 struct cgroup_pidlist **lp)
2498 int pid, n = 0; /* used for populating the array */
2499 struct cgroup_iter it;
2500 struct task_struct *tsk;
2501 struct cgroup_pidlist *l;
2504 * If cgroup gets more users after we read count, we won't have
2505 * enough space - tough. This race is indistinguishable to the
2506 * caller from the case that the additional cgroup users didn't
2507 * show up until sometime later on.
2509 length = cgroup_task_count(cgrp);
2510 array = pidlist_allocate(length);
2513 /* now, populate the array */
2514 cgroup_iter_start(cgrp, &it);
2515 while ((tsk = cgroup_iter_next(cgrp, &it))) {
2516 if (unlikely(n == length))
2518 /* get tgid or pid for procs or tasks file respectively */
2519 if (type == CGROUP_FILE_PROCS)
2520 pid = task_tgid_vnr(tsk);
2522 pid = task_pid_vnr(tsk);
2523 if (pid > 0) /* make sure to only use valid results */
2526 cgroup_iter_end(cgrp, &it);
2528 /* now sort & (if procs) strip out duplicates */
2529 sort(array, length, sizeof(pid_t), cmppid, NULL);
2530 if (type == CGROUP_FILE_PROCS)
2531 length = pidlist_uniq(&array, length);
2532 l = cgroup_pidlist_find(cgrp, type);
2534 pidlist_free(array);
2537 /* store array, freeing old if necessary - lock already held */
2538 pidlist_free(l->list);
2542 up_write(&l->mutex);
2548 * cgroupstats_build - build and fill cgroupstats
2549 * @stats: cgroupstats to fill information into
2550 * @dentry: A dentry entry belonging to the cgroup for which stats have
2553 * Build and fill cgroupstats so that taskstats can export it to user
2556 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
2559 struct cgroup *cgrp;
2560 struct cgroup_iter it;
2561 struct task_struct *tsk;
2564 * Validate dentry by checking the superblock operations,
2565 * and make sure it's a directory.
2567 if (dentry->d_sb->s_op != &cgroup_ops ||
2568 !S_ISDIR(dentry->d_inode->i_mode))
2572 cgrp = dentry->d_fsdata;
2574 cgroup_iter_start(cgrp, &it);
2575 while ((tsk = cgroup_iter_next(cgrp, &it))) {
2576 switch (tsk->state) {
2578 stats->nr_running++;
2580 case TASK_INTERRUPTIBLE:
2581 stats->nr_sleeping++;
2583 case TASK_UNINTERRUPTIBLE:
2584 stats->nr_uninterruptible++;
2587 stats->nr_stopped++;
2590 if (delayacct_is_task_waiting_on_io(tsk))
2591 stats->nr_io_wait++;
2595 cgroup_iter_end(cgrp, &it);
2603 * seq_file methods for the tasks/procs files. The seq_file position is the
2604 * next pid to display; the seq_file iterator is a pointer to the pid
2605 * in the cgroup->l->list array.
2608 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
2611 * Initially we receive a position value that corresponds to
2612 * one more than the last pid shown (or 0 on the first call or
2613 * after a seek to the start). Use a binary-search to find the
2614 * next pid to display, if any
2616 struct cgroup_pidlist *l = s->private;
2617 int index = 0, pid = *pos;
2620 down_read(&l->mutex);
2622 int end = l->length;
2624 while (index < end) {
2625 int mid = (index + end) / 2;
2626 if (l->list[mid] == pid) {
2629 } else if (l->list[mid] <= pid)
2635 /* If we're off the end of the array, we're done */
2636 if (index >= l->length)
2638 /* Update the abstract position to be the actual pid that we found */
2639 iter = l->list + index;
2644 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
2646 struct cgroup_pidlist *l = s->private;
2650 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
2652 struct cgroup_pidlist *l = s->private;
2654 pid_t *end = l->list + l->length;
2656 * Advance to the next pid in the array. If this goes off the
2668 static int cgroup_pidlist_show(struct seq_file *s, void *v)
2670 return seq_printf(s, "%d\n", *(int *)v);
2674 * seq_operations functions for iterating on pidlists through seq_file -
2675 * independent of whether it's tasks or procs
2677 static const struct seq_operations cgroup_pidlist_seq_operations = {
2678 .start = cgroup_pidlist_start,
2679 .stop = cgroup_pidlist_stop,
2680 .next = cgroup_pidlist_next,
2681 .show = cgroup_pidlist_show,
2684 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
2687 * the case where we're the last user of this particular pidlist will
2688 * have us remove it from the cgroup's list, which entails taking the
2689 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
2690 * pidlist_mutex, we have to take pidlist_mutex first.
2692 mutex_lock(&l->owner->pidlist_mutex);
2693 down_write(&l->mutex);
2694 BUG_ON(!l->use_count);
2695 if (!--l->use_count) {
2696 /* we're the last user if refcount is 0; remove and free */
2697 list_del(&l->links);
2698 mutex_unlock(&l->owner->pidlist_mutex);
2699 pidlist_free(l->list);
2700 put_pid_ns(l->key.ns);
2701 up_write(&l->mutex);
2705 mutex_unlock(&l->owner->pidlist_mutex);
2706 up_write(&l->mutex);
2709 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
2711 struct cgroup_pidlist *l;
2712 if (!(file->f_mode & FMODE_READ))
2715 * the seq_file will only be initialized if the file was opened for
2716 * reading; hence we check if it's not null only in that case.
2718 l = ((struct seq_file *)file->private_data)->private;
2719 cgroup_release_pid_array(l);
2720 return seq_release(inode, file);
2723 static const struct file_operations cgroup_pidlist_operations = {
2725 .llseek = seq_lseek,
2726 .write = cgroup_file_write,
2727 .release = cgroup_pidlist_release,
2731 * The following functions handle opens on a file that displays a pidlist
2732 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
2735 /* helper function for the two below it */
2736 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
2738 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2739 struct cgroup_pidlist *l;
2742 /* Nothing to do for write-only files */
2743 if (!(file->f_mode & FMODE_READ))
2746 /* have the array populated */
2747 retval = pidlist_array_load(cgrp, type, &l);
2750 /* configure file information */
2751 file->f_op = &cgroup_pidlist_operations;
2753 retval = seq_open(file, &cgroup_pidlist_seq_operations);
2755 cgroup_release_pid_array(l);
2758 ((struct seq_file *)file->private_data)->private = l;
2761 static int cgroup_tasks_open(struct inode *unused, struct file *file)
2763 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
2765 static int cgroup_procs_open(struct inode *unused, struct file *file)
2767 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
2770 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
2773 return notify_on_release(cgrp);
2776 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
2780 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
2782 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
2784 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
2789 * for the common functions, 'private' gives the type of file
2791 /* for hysterical raisins, we can't put this on the older files */
2792 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
2793 static struct cftype files[] = {
2796 .open = cgroup_tasks_open,
2797 .write_u64 = cgroup_tasks_write,
2798 .release = cgroup_pidlist_release,
2799 .mode = S_IRUGO | S_IWUSR,
2802 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
2803 .open = cgroup_procs_open,
2804 /* .write_u64 = cgroup_procs_write, TODO */
2805 .release = cgroup_pidlist_release,
2809 .name = "notify_on_release",
2810 .read_u64 = cgroup_read_notify_on_release,
2811 .write_u64 = cgroup_write_notify_on_release,
2815 static struct cftype cft_release_agent = {
2816 .name = "release_agent",
2817 .read_seq_string = cgroup_release_agent_show,
2818 .write_string = cgroup_release_agent_write,
2819 .max_write_len = PATH_MAX,
2822 static int cgroup_populate_dir(struct cgroup *cgrp)
2825 struct cgroup_subsys *ss;
2827 /* First clear out any existing files */
2828 cgroup_clear_directory(cgrp->dentry);
2830 err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files));
2834 if (cgrp == cgrp->top_cgroup) {
2835 if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0)
2839 for_each_subsys(cgrp->root, ss) {
2840 if (ss->populate && (err = ss->populate(ss, cgrp)) < 0)
2843 /* This cgroup is ready now */
2844 for_each_subsys(cgrp->root, ss) {
2845 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
2847 * Update id->css pointer and make this css visible from
2848 * CSS ID functions. This pointer will be dereferened
2849 * from RCU-read-side without locks.
2852 rcu_assign_pointer(css->id->css, css);
2858 static void init_cgroup_css(struct cgroup_subsys_state *css,
2859 struct cgroup_subsys *ss,
2860 struct cgroup *cgrp)
2863 atomic_set(&css->refcnt, 1);
2866 if (cgrp == dummytop)
2867 set_bit(CSS_ROOT, &css->flags);
2868 BUG_ON(cgrp->subsys[ss->subsys_id]);
2869 cgrp->subsys[ss->subsys_id] = css;
2872 static void cgroup_lock_hierarchy(struct cgroupfs_root *root)
2874 /* We need to take each hierarchy_mutex in a consistent order */
2877 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2878 struct cgroup_subsys *ss = subsys[i];
2879 if (ss->root == root)
2880 mutex_lock(&ss->hierarchy_mutex);
2884 static void cgroup_unlock_hierarchy(struct cgroupfs_root *root)
2888 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2889 struct cgroup_subsys *ss = subsys[i];
2890 if (ss->root == root)
2891 mutex_unlock(&ss->hierarchy_mutex);
2896 * cgroup_create - create a cgroup
2897 * @parent: cgroup that will be parent of the new cgroup
2898 * @dentry: dentry of the new cgroup
2899 * @mode: mode to set on new inode
2901 * Must be called with the mutex on the parent inode held
2903 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
2906 struct cgroup *cgrp;
2907 struct cgroupfs_root *root = parent->root;
2909 struct cgroup_subsys *ss;
2910 struct super_block *sb = root->sb;
2912 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
2916 /* Grab a reference on the superblock so the hierarchy doesn't
2917 * get deleted on unmount if there are child cgroups. This
2918 * can be done outside cgroup_mutex, since the sb can't
2919 * disappear while someone has an open control file on the
2921 atomic_inc(&sb->s_active);
2923 mutex_lock(&cgroup_mutex);
2925 init_cgroup_housekeeping(cgrp);
2927 cgrp->parent = parent;
2928 cgrp->root = parent->root;
2929 cgrp->top_cgroup = parent->top_cgroup;
2931 if (notify_on_release(parent))
2932 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
2934 for_each_subsys(root, ss) {
2935 struct cgroup_subsys_state *css = ss->create(ss, cgrp);
2940 init_cgroup_css(css, ss, cgrp);
2942 if (alloc_css_id(ss, parent, cgrp))
2944 /* At error, ->destroy() callback has to free assigned ID. */
2947 cgroup_lock_hierarchy(root);
2948 list_add(&cgrp->sibling, &cgrp->parent->children);
2949 cgroup_unlock_hierarchy(root);
2950 root->number_of_cgroups++;
2952 err = cgroup_create_dir(cgrp, dentry, mode);
2956 /* The cgroup directory was pre-locked for us */
2957 BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
2959 err = cgroup_populate_dir(cgrp);
2960 /* If err < 0, we have a half-filled directory - oh well ;) */
2962 mutex_unlock(&cgroup_mutex);
2963 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
2969 cgroup_lock_hierarchy(root);
2970 list_del(&cgrp->sibling);
2971 cgroup_unlock_hierarchy(root);
2972 root->number_of_cgroups--;
2976 for_each_subsys(root, ss) {
2977 if (cgrp->subsys[ss->subsys_id])
2978 ss->destroy(ss, cgrp);
2981 mutex_unlock(&cgroup_mutex);
2983 /* Release the reference count that we took on the superblock */
2984 deactivate_super(sb);
2990 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2992 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
2994 /* the vfs holds inode->i_mutex already */
2995 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
2998 static int cgroup_has_css_refs(struct cgroup *cgrp)
3000 /* Check the reference count on each subsystem. Since we
3001 * already established that there are no tasks in the
3002 * cgroup, if the css refcount is also 1, then there should
3003 * be no outstanding references, so the subsystem is safe to
3004 * destroy. We scan across all subsystems rather than using
3005 * the per-hierarchy linked list of mounted subsystems since
3006 * we can be called via check_for_release() with no
3007 * synchronization other than RCU, and the subsystem linked
3008 * list isn't RCU-safe */
3010 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3011 struct cgroup_subsys *ss = subsys[i];
3012 struct cgroup_subsys_state *css;
3013 /* Skip subsystems not in this hierarchy */
3014 if (ss->root != cgrp->root)
3016 css = cgrp->subsys[ss->subsys_id];
3017 /* When called from check_for_release() it's possible
3018 * that by this point the cgroup has been removed
3019 * and the css deleted. But a false-positive doesn't
3020 * matter, since it can only happen if the cgroup
3021 * has been deleted and hence no longer needs the
3022 * release agent to be called anyway. */
3023 if (css && (atomic_read(&css->refcnt) > 1))
3030 * Atomically mark all (or else none) of the cgroup's CSS objects as
3031 * CSS_REMOVED. Return true on success, or false if the cgroup has
3032 * busy subsystems. Call with cgroup_mutex held
3035 static int cgroup_clear_css_refs(struct cgroup *cgrp)
3037 struct cgroup_subsys *ss;
3038 unsigned long flags;
3039 bool failed = false;
3040 local_irq_save(flags);
3041 for_each_subsys(cgrp->root, ss) {
3042 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3045 /* We can only remove a CSS with a refcnt==1 */
3046 refcnt = atomic_read(&css->refcnt);
3053 * Drop the refcnt to 0 while we check other
3054 * subsystems. This will cause any racing
3055 * css_tryget() to spin until we set the
3056 * CSS_REMOVED bits or abort
3058 if (atomic_cmpxchg(&css->refcnt, refcnt, 0) == refcnt)
3064 for_each_subsys(cgrp->root, ss) {
3065 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3068 * Restore old refcnt if we previously managed
3069 * to clear it from 1 to 0
3071 if (!atomic_read(&css->refcnt))
3072 atomic_set(&css->refcnt, 1);
3074 /* Commit the fact that the CSS is removed */
3075 set_bit(CSS_REMOVED, &css->flags);
3078 local_irq_restore(flags);
3082 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
3084 struct cgroup *cgrp = dentry->d_fsdata;
3086 struct cgroup *parent;
3090 /* the vfs holds both inode->i_mutex already */
3092 mutex_lock(&cgroup_mutex);
3093 if (atomic_read(&cgrp->count) != 0) {
3094 mutex_unlock(&cgroup_mutex);
3097 if (!list_empty(&cgrp->children)) {
3098 mutex_unlock(&cgroup_mutex);
3101 mutex_unlock(&cgroup_mutex);
3104 * In general, subsystem has no css->refcnt after pre_destroy(). But
3105 * in racy cases, subsystem may have to get css->refcnt after
3106 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
3107 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
3108 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
3109 * and subsystem's reference count handling. Please see css_get/put
3110 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
3112 set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3115 * Call pre_destroy handlers of subsys. Notify subsystems
3116 * that rmdir() request comes.
3118 ret = cgroup_call_pre_destroy(cgrp);
3120 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3124 mutex_lock(&cgroup_mutex);
3125 parent = cgrp->parent;
3126 if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) {
3127 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3128 mutex_unlock(&cgroup_mutex);
3131 prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE);
3132 if (!cgroup_clear_css_refs(cgrp)) {
3133 mutex_unlock(&cgroup_mutex);
3135 * Because someone may call cgroup_wakeup_rmdir_waiter() before
3136 * prepare_to_wait(), we need to check this flag.
3138 if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags))
3140 finish_wait(&cgroup_rmdir_waitq, &wait);
3141 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3142 if (signal_pending(current))
3146 /* NO css_tryget() can success after here. */
3147 finish_wait(&cgroup_rmdir_waitq, &wait);
3148 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3150 spin_lock(&release_list_lock);
3151 set_bit(CGRP_REMOVED, &cgrp->flags);
3152 if (!list_empty(&cgrp->release_list))
3153 list_del(&cgrp->release_list);
3154 spin_unlock(&release_list_lock);
3156 cgroup_lock_hierarchy(cgrp->root);
3157 /* delete this cgroup from parent->children */
3158 list_del(&cgrp->sibling);
3159 cgroup_unlock_hierarchy(cgrp->root);
3161 spin_lock(&cgrp->dentry->d_lock);
3162 d = dget(cgrp->dentry);
3163 spin_unlock(&d->d_lock);
3165 cgroup_d_remove_dir(d);
3168 set_bit(CGRP_RELEASABLE, &parent->flags);
3169 check_for_release(parent);
3171 mutex_unlock(&cgroup_mutex);
3175 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
3177 struct cgroup_subsys_state *css;
3179 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
3181 /* Create the top cgroup state for this subsystem */
3182 list_add(&ss->sibling, &rootnode.subsys_list);
3183 ss->root = &rootnode;
3184 css = ss->create(ss, dummytop);
3185 /* We don't handle early failures gracefully */
3186 BUG_ON(IS_ERR(css));
3187 init_cgroup_css(css, ss, dummytop);
3189 /* Update the init_css_set to contain a subsys
3190 * pointer to this state - since the subsystem is
3191 * newly registered, all tasks and hence the
3192 * init_css_set is in the subsystem's top cgroup. */
3193 init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
3195 need_forkexit_callback |= ss->fork || ss->exit;
3197 /* At system boot, before all subsystems have been
3198 * registered, no tasks have been forked, so we don't
3199 * need to invoke fork callbacks here. */
3200 BUG_ON(!list_empty(&init_task.tasks));
3202 mutex_init(&ss->hierarchy_mutex);
3203 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
3208 * cgroup_init_early - cgroup initialization at system boot
3210 * Initialize cgroups at system boot, and initialize any
3211 * subsystems that request early init.
3213 int __init cgroup_init_early(void)
3216 atomic_set(&init_css_set.refcount, 1);
3217 INIT_LIST_HEAD(&init_css_set.cg_links);
3218 INIT_LIST_HEAD(&init_css_set.tasks);
3219 INIT_HLIST_NODE(&init_css_set.hlist);
3221 init_cgroup_root(&rootnode);
3223 init_task.cgroups = &init_css_set;
3225 init_css_set_link.cg = &init_css_set;
3226 init_css_set_link.cgrp = dummytop;
3227 list_add(&init_css_set_link.cgrp_link_list,
3228 &rootnode.top_cgroup.css_sets);
3229 list_add(&init_css_set_link.cg_link_list,
3230 &init_css_set.cg_links);
3232 for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
3233 INIT_HLIST_HEAD(&css_set_table[i]);
3235 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3236 struct cgroup_subsys *ss = subsys[i];
3239 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
3240 BUG_ON(!ss->create);
3241 BUG_ON(!ss->destroy);
3242 if (ss->subsys_id != i) {
3243 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
3244 ss->name, ss->subsys_id);
3249 cgroup_init_subsys(ss);
3255 * cgroup_init - cgroup initialization
3257 * Register cgroup filesystem and /proc file, and initialize
3258 * any subsystems that didn't request early init.
3260 int __init cgroup_init(void)
3264 struct hlist_head *hhead;
3266 err = bdi_init(&cgroup_backing_dev_info);
3270 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3271 struct cgroup_subsys *ss = subsys[i];
3272 if (!ss->early_init)
3273 cgroup_init_subsys(ss);
3275 cgroup_subsys_init_idr(ss);
3278 /* Add init_css_set to the hash table */
3279 hhead = css_set_hash(init_css_set.subsys);
3280 hlist_add_head(&init_css_set.hlist, hhead);
3281 BUG_ON(!init_root_id(&rootnode));
3282 err = register_filesystem(&cgroup_fs_type);
3286 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
3290 bdi_destroy(&cgroup_backing_dev_info);
3296 * proc_cgroup_show()
3297 * - Print task's cgroup paths into seq_file, one line for each hierarchy
3298 * - Used for /proc/<pid>/cgroup.
3299 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
3300 * doesn't really matter if tsk->cgroup changes after we read it,
3301 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
3302 * anyway. No need to check that tsk->cgroup != NULL, thanks to
3303 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
3304 * cgroup to top_cgroup.
3307 /* TODO: Use a proper seq_file iterator */
3308 static int proc_cgroup_show(struct seq_file *m, void *v)
3311 struct task_struct *tsk;
3314 struct cgroupfs_root *root;
3317 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
3323 tsk = get_pid_task(pid, PIDTYPE_PID);
3329 mutex_lock(&cgroup_mutex);
3331 for_each_active_root(root) {
3332 struct cgroup_subsys *ss;
3333 struct cgroup *cgrp;
3336 seq_printf(m, "%d:", root->hierarchy_id);
3337 for_each_subsys(root, ss)
3338 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
3339 if (strlen(root->name))
3340 seq_printf(m, "%sname=%s", count ? "," : "",
3343 cgrp = task_cgroup_from_root(tsk, root);
3344 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
3352 mutex_unlock(&cgroup_mutex);
3353 put_task_struct(tsk);
3360 static int cgroup_open(struct inode *inode, struct file *file)
3362 struct pid *pid = PROC_I(inode)->pid;
3363 return single_open(file, proc_cgroup_show, pid);
3366 struct file_operations proc_cgroup_operations = {
3367 .open = cgroup_open,
3369 .llseek = seq_lseek,
3370 .release = single_release,
3373 /* Display information about each subsystem and each hierarchy */
3374 static int proc_cgroupstats_show(struct seq_file *m, void *v)
3378 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
3379 mutex_lock(&cgroup_mutex);
3380 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3381 struct cgroup_subsys *ss = subsys[i];
3382 seq_printf(m, "%s\t%d\t%d\t%d\n",
3383 ss->name, ss->root->hierarchy_id,
3384 ss->root->number_of_cgroups, !ss->disabled);
3386 mutex_unlock(&cgroup_mutex);
3390 static int cgroupstats_open(struct inode *inode, struct file *file)
3392 return single_open(file, proc_cgroupstats_show, NULL);
3395 static struct file_operations proc_cgroupstats_operations = {
3396 .open = cgroupstats_open,
3398 .llseek = seq_lseek,
3399 .release = single_release,
3403 * cgroup_fork - attach newly forked task to its parents cgroup.
3404 * @child: pointer to task_struct of forking parent process.
3406 * Description: A task inherits its parent's cgroup at fork().
3408 * A pointer to the shared css_set was automatically copied in
3409 * fork.c by dup_task_struct(). However, we ignore that copy, since
3410 * it was not made under the protection of RCU or cgroup_mutex, so
3411 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
3412 * have already changed current->cgroups, allowing the previously
3413 * referenced cgroup group to be removed and freed.
3415 * At the point that cgroup_fork() is called, 'current' is the parent
3416 * task, and the passed argument 'child' points to the child task.
3418 void cgroup_fork(struct task_struct *child)
3421 child->cgroups = current->cgroups;
3422 get_css_set(child->cgroups);
3423 task_unlock(current);
3424 INIT_LIST_HEAD(&child->cg_list);
3428 * cgroup_fork_callbacks - run fork callbacks
3429 * @child: the new task
3431 * Called on a new task very soon before adding it to the
3432 * tasklist. No need to take any locks since no-one can
3433 * be operating on this task.
3435 void cgroup_fork_callbacks(struct task_struct *child)
3437 if (need_forkexit_callback) {
3439 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3440 struct cgroup_subsys *ss = subsys[i];
3442 ss->fork(ss, child);
3448 * cgroup_post_fork - called on a new task after adding it to the task list
3449 * @child: the task in question
3451 * Adds the task to the list running through its css_set if necessary.
3452 * Has to be after the task is visible on the task list in case we race
3453 * with the first call to cgroup_iter_start() - to guarantee that the
3454 * new task ends up on its list.
3456 void cgroup_post_fork(struct task_struct *child)
3458 if (use_task_css_set_links) {
3459 write_lock(&css_set_lock);
3461 if (list_empty(&child->cg_list))
3462 list_add(&child->cg_list, &child->cgroups->tasks);
3464 write_unlock(&css_set_lock);
3468 * cgroup_exit - detach cgroup from exiting task
3469 * @tsk: pointer to task_struct of exiting process
3470 * @run_callback: run exit callbacks?
3472 * Description: Detach cgroup from @tsk and release it.
3474 * Note that cgroups marked notify_on_release force every task in
3475 * them to take the global cgroup_mutex mutex when exiting.
3476 * This could impact scaling on very large systems. Be reluctant to
3477 * use notify_on_release cgroups where very high task exit scaling
3478 * is required on large systems.
3480 * the_top_cgroup_hack:
3482 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
3484 * We call cgroup_exit() while the task is still competent to
3485 * handle notify_on_release(), then leave the task attached to the
3486 * root cgroup in each hierarchy for the remainder of its exit.
3488 * To do this properly, we would increment the reference count on
3489 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
3490 * code we would add a second cgroup function call, to drop that
3491 * reference. This would just create an unnecessary hot spot on
3492 * the top_cgroup reference count, to no avail.
3494 * Normally, holding a reference to a cgroup without bumping its
3495 * count is unsafe. The cgroup could go away, or someone could
3496 * attach us to a different cgroup, decrementing the count on
3497 * the first cgroup that we never incremented. But in this case,
3498 * top_cgroup isn't going away, and either task has PF_EXITING set,
3499 * which wards off any cgroup_attach_task() attempts, or task is a failed
3500 * fork, never visible to cgroup_attach_task.
3502 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
3507 if (run_callbacks && need_forkexit_callback) {
3508 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3509 struct cgroup_subsys *ss = subsys[i];
3516 * Unlink from the css_set task list if necessary.
3517 * Optimistically check cg_list before taking
3520 if (!list_empty(&tsk->cg_list)) {
3521 write_lock(&css_set_lock);
3522 if (!list_empty(&tsk->cg_list))
3523 list_del(&tsk->cg_list);
3524 write_unlock(&css_set_lock);
3527 /* Reassign the task to the init_css_set. */
3530 tsk->cgroups = &init_css_set;
3533 put_css_set_taskexit(cg);
3537 * cgroup_clone - clone the cgroup the given subsystem is attached to
3538 * @tsk: the task to be moved
3539 * @subsys: the given subsystem
3540 * @nodename: the name for the new cgroup
3542 * Duplicate the current cgroup in the hierarchy that the given
3543 * subsystem is attached to, and move this task into the new
3546 int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *subsys,
3549 struct dentry *dentry;
3551 struct cgroup *parent, *child;
3552 struct inode *inode;
3554 struct cgroupfs_root *root;
3555 struct cgroup_subsys *ss;
3557 /* We shouldn't be called by an unregistered subsystem */
3558 BUG_ON(!subsys->active);
3560 /* First figure out what hierarchy and cgroup we're dealing
3561 * with, and pin them so we can drop cgroup_mutex */
3562 mutex_lock(&cgroup_mutex);
3564 root = subsys->root;
3565 if (root == &rootnode) {
3566 mutex_unlock(&cgroup_mutex);
3570 /* Pin the hierarchy */
3571 if (!atomic_inc_not_zero(&root->sb->s_active)) {
3572 /* We race with the final deactivate_super() */
3573 mutex_unlock(&cgroup_mutex);
3577 /* Keep the cgroup alive */
3579 parent = task_cgroup(tsk, subsys->subsys_id);
3584 mutex_unlock(&cgroup_mutex);
3586 /* Now do the VFS work to create a cgroup */
3587 inode = parent->dentry->d_inode;
3589 /* Hold the parent directory mutex across this operation to
3590 * stop anyone else deleting the new cgroup */
3591 mutex_lock(&inode->i_mutex);
3592 dentry = lookup_one_len(nodename, parent->dentry, strlen(nodename));
3593 if (IS_ERR(dentry)) {
3595 "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename,
3597 ret = PTR_ERR(dentry);
3601 /* Create the cgroup directory, which also creates the cgroup */
3602 ret = vfs_mkdir(inode, dentry, 0755);
3603 child = __d_cgrp(dentry);
3607 "Failed to create cgroup %s: %d\n", nodename,
3612 /* The cgroup now exists. Retake cgroup_mutex and check
3613 * that we're still in the same state that we thought we
3615 mutex_lock(&cgroup_mutex);
3616 if ((root != subsys->root) ||
3617 (parent != task_cgroup(tsk, subsys->subsys_id))) {
3618 /* Aargh, we raced ... */
3619 mutex_unlock(&inode->i_mutex);
3622 deactivate_super(root->sb);
3623 /* The cgroup is still accessible in the VFS, but
3624 * we're not going to try to rmdir() it at this
3627 "Race in cgroup_clone() - leaking cgroup %s\n",
3632 /* do any required auto-setup */
3633 for_each_subsys(root, ss) {
3635 ss->post_clone(ss, child);
3638 /* All seems fine. Finish by moving the task into the new cgroup */
3639 ret = cgroup_attach_task(child, tsk);
3640 mutex_unlock(&cgroup_mutex);
3643 mutex_unlock(&inode->i_mutex);
3645 mutex_lock(&cgroup_mutex);
3647 mutex_unlock(&cgroup_mutex);
3648 deactivate_super(root->sb);
3653 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
3654 * @cgrp: the cgroup in question
3655 * @task: the task in question
3657 * See if @cgrp is a descendant of @task's cgroup in the appropriate
3660 * If we are sending in dummytop, then presumably we are creating
3661 * the top cgroup in the subsystem.
3663 * Called only by the ns (nsproxy) cgroup.
3665 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
3668 struct cgroup *target;
3670 if (cgrp == dummytop)
3673 target = task_cgroup_from_root(task, cgrp->root);
3674 while (cgrp != target && cgrp!= cgrp->top_cgroup)
3675 cgrp = cgrp->parent;
3676 ret = (cgrp == target);
3680 static void check_for_release(struct cgroup *cgrp)
3682 /* All of these checks rely on RCU to keep the cgroup
3683 * structure alive */
3684 if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
3685 && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
3686 /* Control Group is currently removeable. If it's not
3687 * already queued for a userspace notification, queue
3689 int need_schedule_work = 0;
3690 spin_lock(&release_list_lock);
3691 if (!cgroup_is_removed(cgrp) &&
3692 list_empty(&cgrp->release_list)) {
3693 list_add(&cgrp->release_list, &release_list);
3694 need_schedule_work = 1;
3696 spin_unlock(&release_list_lock);
3697 if (need_schedule_work)
3698 schedule_work(&release_agent_work);
3702 void __css_put(struct cgroup_subsys_state *css)
3704 struct cgroup *cgrp = css->cgroup;
3706 if (atomic_dec_return(&css->refcnt) == 1) {
3707 if (notify_on_release(cgrp)) {
3708 set_bit(CGRP_RELEASABLE, &cgrp->flags);
3709 check_for_release(cgrp);
3711 cgroup_wakeup_rmdir_waiter(cgrp);
3717 * Notify userspace when a cgroup is released, by running the
3718 * configured release agent with the name of the cgroup (path
3719 * relative to the root of cgroup file system) as the argument.
3721 * Most likely, this user command will try to rmdir this cgroup.
3723 * This races with the possibility that some other task will be
3724 * attached to this cgroup before it is removed, or that some other
3725 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
3726 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
3727 * unused, and this cgroup will be reprieved from its death sentence,
3728 * to continue to serve a useful existence. Next time it's released,
3729 * we will get notified again, if it still has 'notify_on_release' set.
3731 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
3732 * means only wait until the task is successfully execve()'d. The
3733 * separate release agent task is forked by call_usermodehelper(),
3734 * then control in this thread returns here, without waiting for the
3735 * release agent task. We don't bother to wait because the caller of
3736 * this routine has no use for the exit status of the release agent
3737 * task, so no sense holding our caller up for that.
3739 static void cgroup_release_agent(struct work_struct *work)
3741 BUG_ON(work != &release_agent_work);
3742 mutex_lock(&cgroup_mutex);
3743 spin_lock(&release_list_lock);
3744 while (!list_empty(&release_list)) {
3745 char *argv[3], *envp[3];
3747 char *pathbuf = NULL, *agentbuf = NULL;
3748 struct cgroup *cgrp = list_entry(release_list.next,
3751 list_del_init(&cgrp->release_list);
3752 spin_unlock(&release_list_lock);
3753 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
3756 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
3758 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
3763 argv[i++] = agentbuf;
3764 argv[i++] = pathbuf;
3768 /* minimal command environment */
3769 envp[i++] = "HOME=/";
3770 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
3773 /* Drop the lock while we invoke the usermode helper,
3774 * since the exec could involve hitting disk and hence
3775 * be a slow process */
3776 mutex_unlock(&cgroup_mutex);
3777 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
3778 mutex_lock(&cgroup_mutex);
3782 spin_lock(&release_list_lock);
3784 spin_unlock(&release_list_lock);
3785 mutex_unlock(&cgroup_mutex);
3788 static int __init cgroup_disable(char *str)
3793 while ((token = strsep(&str, ",")) != NULL) {
3797 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3798 struct cgroup_subsys *ss = subsys[i];
3800 if (!strcmp(token, ss->name)) {
3802 printk(KERN_INFO "Disabling %s control group"
3803 " subsystem\n", ss->name);
3810 __setup("cgroup_disable=", cgroup_disable);
3813 * Functons for CSS ID.
3817 *To get ID other than 0, this should be called when !cgroup_is_removed().
3819 unsigned short css_id(struct cgroup_subsys_state *css)
3821 struct css_id *cssid = rcu_dereference(css->id);
3828 unsigned short css_depth(struct cgroup_subsys_state *css)
3830 struct css_id *cssid = rcu_dereference(css->id);
3833 return cssid->depth;
3837 bool css_is_ancestor(struct cgroup_subsys_state *child,
3838 const struct cgroup_subsys_state *root)
3840 struct css_id *child_id = rcu_dereference(child->id);
3841 struct css_id *root_id = rcu_dereference(root->id);
3843 if (!child_id || !root_id || (child_id->depth < root_id->depth))
3845 return child_id->stack[root_id->depth] == root_id->id;
3848 static void __free_css_id_cb(struct rcu_head *head)
3852 id = container_of(head, struct css_id, rcu_head);
3856 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
3858 struct css_id *id = css->id;
3859 /* When this is called before css_id initialization, id can be NULL */
3863 BUG_ON(!ss->use_id);
3865 rcu_assign_pointer(id->css, NULL);
3866 rcu_assign_pointer(css->id, NULL);
3867 spin_lock(&ss->id_lock);
3868 idr_remove(&ss->idr, id->id);
3869 spin_unlock(&ss->id_lock);
3870 call_rcu(&id->rcu_head, __free_css_id_cb);
3874 * This is called by init or create(). Then, calls to this function are
3875 * always serialized (By cgroup_mutex() at create()).
3878 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
3880 struct css_id *newid;
3881 int myid, error, size;
3883 BUG_ON(!ss->use_id);
3885 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
3886 newid = kzalloc(size, GFP_KERNEL);
3888 return ERR_PTR(-ENOMEM);
3890 if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
3894 spin_lock(&ss->id_lock);
3895 /* Don't use 0. allocates an ID of 1-65535 */
3896 error = idr_get_new_above(&ss->idr, newid, 1, &myid);
3897 spin_unlock(&ss->id_lock);
3899 /* Returns error when there are no free spaces for new ID.*/
3904 if (myid > CSS_ID_MAX)
3908 newid->depth = depth;
3912 spin_lock(&ss->id_lock);
3913 idr_remove(&ss->idr, myid);
3914 spin_unlock(&ss->id_lock);
3917 return ERR_PTR(error);
3921 static int __init cgroup_subsys_init_idr(struct cgroup_subsys *ss)
3923 struct css_id *newid;
3924 struct cgroup_subsys_state *rootcss;
3926 spin_lock_init(&ss->id_lock);
3929 rootcss = init_css_set.subsys[ss->subsys_id];
3930 newid = get_new_cssid(ss, 0);
3932 return PTR_ERR(newid);
3934 newid->stack[0] = newid->id;
3935 newid->css = rootcss;
3936 rootcss->id = newid;
3940 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
3941 struct cgroup *child)
3943 int subsys_id, i, depth = 0;
3944 struct cgroup_subsys_state *parent_css, *child_css;
3945 struct css_id *child_id, *parent_id = NULL;
3947 subsys_id = ss->subsys_id;
3948 parent_css = parent->subsys[subsys_id];
3949 child_css = child->subsys[subsys_id];
3950 depth = css_depth(parent_css) + 1;
3951 parent_id = parent_css->id;
3953 child_id = get_new_cssid(ss, depth);
3954 if (IS_ERR(child_id))
3955 return PTR_ERR(child_id);
3957 for (i = 0; i < depth; i++)
3958 child_id->stack[i] = parent_id->stack[i];
3959 child_id->stack[depth] = child_id->id;
3961 * child_id->css pointer will be set after this cgroup is available
3962 * see cgroup_populate_dir()
3964 rcu_assign_pointer(child_css->id, child_id);
3970 * css_lookup - lookup css by id
3971 * @ss: cgroup subsys to be looked into.
3974 * Returns pointer to cgroup_subsys_state if there is valid one with id.
3975 * NULL if not. Should be called under rcu_read_lock()
3977 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
3979 struct css_id *cssid = NULL;
3981 BUG_ON(!ss->use_id);
3982 cssid = idr_find(&ss->idr, id);
3984 if (unlikely(!cssid))
3987 return rcu_dereference(cssid->css);
3991 * css_get_next - lookup next cgroup under specified hierarchy.
3992 * @ss: pointer to subsystem
3993 * @id: current position of iteration.
3994 * @root: pointer to css. search tree under this.
3995 * @foundid: position of found object.
3997 * Search next css under the specified hierarchy of rootid. Calling under
3998 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
4000 struct cgroup_subsys_state *
4001 css_get_next(struct cgroup_subsys *ss, int id,
4002 struct cgroup_subsys_state *root, int *foundid)
4004 struct cgroup_subsys_state *ret = NULL;
4007 int rootid = css_id(root);
4008 int depth = css_depth(root);
4013 BUG_ON(!ss->use_id);
4014 /* fill start point for scan */
4018 * scan next entry from bitmap(tree), tmpid is updated after
4021 spin_lock(&ss->id_lock);
4022 tmp = idr_get_next(&ss->idr, &tmpid);
4023 spin_unlock(&ss->id_lock);
4027 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
4028 ret = rcu_dereference(tmp->css);
4034 /* continue to scan from next id */
4040 #ifdef CONFIG_CGROUP_DEBUG
4041 static struct cgroup_subsys_state *debug_create(struct cgroup_subsys *ss,
4042 struct cgroup *cont)
4044 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
4047 return ERR_PTR(-ENOMEM);
4052 static void debug_destroy(struct cgroup_subsys *ss, struct cgroup *cont)
4054 kfree(cont->subsys[debug_subsys_id]);
4057 static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
4059 return atomic_read(&cont->count);
4062 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
4064 return cgroup_task_count(cont);
4067 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
4069 return (u64)(unsigned long)current->cgroups;
4072 static u64 current_css_set_refcount_read(struct cgroup *cont,
4078 count = atomic_read(¤t->cgroups->refcount);
4083 static int current_css_set_cg_links_read(struct cgroup *cont,
4085 struct seq_file *seq)
4087 struct cg_cgroup_link *link;
4090 read_lock(&css_set_lock);
4092 cg = rcu_dereference(current->cgroups);
4093 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
4094 struct cgroup *c = link->cgrp;
4098 name = c->dentry->d_name.name;
4101 seq_printf(seq, "Root %d group %s\n",
4102 c->root->hierarchy_id, name);
4105 read_unlock(&css_set_lock);
4109 #define MAX_TASKS_SHOWN_PER_CSS 25
4110 static int cgroup_css_links_read(struct cgroup *cont,
4112 struct seq_file *seq)
4114 struct cg_cgroup_link *link;
4116 read_lock(&css_set_lock);
4117 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
4118 struct css_set *cg = link->cg;
4119 struct task_struct *task;
4121 seq_printf(seq, "css_set %p\n", cg);
4122 list_for_each_entry(task, &cg->tasks, cg_list) {
4123 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
4124 seq_puts(seq, " ...\n");
4127 seq_printf(seq, " task %d\n",
4128 task_pid_vnr(task));
4132 read_unlock(&css_set_lock);
4136 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
4138 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
4141 static struct cftype debug_files[] = {
4143 .name = "cgroup_refcount",
4144 .read_u64 = cgroup_refcount_read,
4147 .name = "taskcount",
4148 .read_u64 = debug_taskcount_read,
4152 .name = "current_css_set",
4153 .read_u64 = current_css_set_read,
4157 .name = "current_css_set_refcount",
4158 .read_u64 = current_css_set_refcount_read,
4162 .name = "current_css_set_cg_links",
4163 .read_seq_string = current_css_set_cg_links_read,
4167 .name = "cgroup_css_links",
4168 .read_seq_string = cgroup_css_links_read,
4172 .name = "releasable",
4173 .read_u64 = releasable_read,
4177 static int debug_populate(struct cgroup_subsys *ss, struct cgroup *cont)
4179 return cgroup_add_files(cont, ss, debug_files,
4180 ARRAY_SIZE(debug_files));
4183 struct cgroup_subsys debug_subsys = {
4185 .create = debug_create,
4186 .destroy = debug_destroy,
4187 .populate = debug_populate,
4188 .subsys_id = debug_subsys_id,
4190 #endif /* CONFIG_CGROUP_DEBUG */