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 static void free_css_set_rcu(struct rcu_head *obj)
272 struct css_set *cg = container_of(obj, struct css_set, rcu_head);
276 /* We don't maintain the lists running through each css_set to its
277 * task until after the first call to cgroup_iter_start(). This
278 * reduces the fork()/exit() overhead for people who have cgroups
279 * compiled into their kernel but not actually in use */
280 static int use_task_css_set_links __read_mostly;
282 static void __put_css_set(struct css_set *cg, int taskexit)
284 struct cg_cgroup_link *link;
285 struct cg_cgroup_link *saved_link;
287 * Ensure that the refcount doesn't hit zero while any readers
288 * can see it. Similar to atomic_dec_and_lock(), but for an
291 if (atomic_add_unless(&cg->refcount, -1, 1))
293 write_lock(&css_set_lock);
294 if (!atomic_dec_and_test(&cg->refcount)) {
295 write_unlock(&css_set_lock);
299 /* This css_set is dead. unlink it and release cgroup refcounts */
300 hlist_del(&cg->hlist);
303 list_for_each_entry_safe(link, saved_link, &cg->cg_links,
305 struct cgroup *cgrp = link->cgrp;
306 list_del(&link->cg_link_list);
307 list_del(&link->cgrp_link_list);
308 if (atomic_dec_and_test(&cgrp->count) &&
309 notify_on_release(cgrp)) {
311 set_bit(CGRP_RELEASABLE, &cgrp->flags);
312 check_for_release(cgrp);
318 write_unlock(&css_set_lock);
319 call_rcu(&cg->rcu_head, free_css_set_rcu);
323 * refcounted get/put for css_set objects
325 static inline void get_css_set(struct css_set *cg)
327 atomic_inc(&cg->refcount);
330 static inline void put_css_set(struct css_set *cg)
332 __put_css_set(cg, 0);
335 static inline void put_css_set_taskexit(struct css_set *cg)
337 __put_css_set(cg, 1);
341 * compare_css_sets - helper function for find_existing_css_set().
342 * @cg: candidate css_set being tested
343 * @old_cg: existing css_set for a task
344 * @new_cgrp: cgroup that's being entered by the task
345 * @template: desired set of css pointers in css_set (pre-calculated)
347 * Returns true if "cg" matches "old_cg" except for the hierarchy
348 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
350 static bool compare_css_sets(struct css_set *cg,
351 struct css_set *old_cg,
352 struct cgroup *new_cgrp,
353 struct cgroup_subsys_state *template[])
355 struct list_head *l1, *l2;
357 if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
358 /* Not all subsystems matched */
363 * Compare cgroup pointers in order to distinguish between
364 * different cgroups in heirarchies with no subsystems. We
365 * could get by with just this check alone (and skip the
366 * memcmp above) but on most setups the memcmp check will
367 * avoid the need for this more expensive check on almost all
372 l2 = &old_cg->cg_links;
374 struct cg_cgroup_link *cgl1, *cgl2;
375 struct cgroup *cg1, *cg2;
379 /* See if we reached the end - both lists are equal length. */
380 if (l1 == &cg->cg_links) {
381 BUG_ON(l2 != &old_cg->cg_links);
384 BUG_ON(l2 == &old_cg->cg_links);
386 /* Locate the cgroups associated with these links. */
387 cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
388 cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
391 /* Hierarchies should be linked in the same order. */
392 BUG_ON(cg1->root != cg2->root);
395 * If this hierarchy is the hierarchy of the cgroup
396 * that's changing, then we need to check that this
397 * css_set points to the new cgroup; if it's any other
398 * hierarchy, then this css_set should point to the
399 * same cgroup as the old css_set.
401 if (cg1->root == new_cgrp->root) {
413 * find_existing_css_set() is a helper for
414 * find_css_set(), and checks to see whether an existing
415 * css_set is suitable.
417 * oldcg: the cgroup group that we're using before the cgroup
420 * cgrp: the cgroup that we're moving into
422 * template: location in which to build the desired set of subsystem
423 * state objects for the new cgroup group
425 static struct css_set *find_existing_css_set(
426 struct css_set *oldcg,
428 struct cgroup_subsys_state *template[])
431 struct cgroupfs_root *root = cgrp->root;
432 struct hlist_head *hhead;
433 struct hlist_node *node;
436 /* Built the set of subsystem state objects that we want to
437 * see in the new css_set */
438 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
439 if (root->subsys_bits & (1UL << i)) {
440 /* Subsystem is in this hierarchy. So we want
441 * the subsystem state from the new
443 template[i] = cgrp->subsys[i];
445 /* Subsystem is not in this hierarchy, so we
446 * don't want to change the subsystem state */
447 template[i] = oldcg->subsys[i];
451 hhead = css_set_hash(template);
452 hlist_for_each_entry(cg, node, hhead, hlist) {
453 if (!compare_css_sets(cg, oldcg, cgrp, template))
456 /* This css_set matches what we need */
460 /* No existing cgroup group matched */
464 static void free_cg_links(struct list_head *tmp)
466 struct cg_cgroup_link *link;
467 struct cg_cgroup_link *saved_link;
469 list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
470 list_del(&link->cgrp_link_list);
476 * allocate_cg_links() allocates "count" cg_cgroup_link structures
477 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
478 * success or a negative error
480 static int allocate_cg_links(int count, struct list_head *tmp)
482 struct cg_cgroup_link *link;
485 for (i = 0; i < count; i++) {
486 link = kmalloc(sizeof(*link), GFP_KERNEL);
491 list_add(&link->cgrp_link_list, tmp);
497 * link_css_set - a helper function to link a css_set to a cgroup
498 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
499 * @cg: the css_set to be linked
500 * @cgrp: the destination cgroup
502 static void link_css_set(struct list_head *tmp_cg_links,
503 struct css_set *cg, struct cgroup *cgrp)
505 struct cg_cgroup_link *link;
507 BUG_ON(list_empty(tmp_cg_links));
508 link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
512 atomic_inc(&cgrp->count);
513 list_move(&link->cgrp_link_list, &cgrp->css_sets);
515 * Always add links to the tail of the list so that the list
516 * is sorted by order of hierarchy creation
518 list_add_tail(&link->cg_link_list, &cg->cg_links);
522 * find_css_set() takes an existing cgroup group and a
523 * cgroup object, and returns a css_set object that's
524 * equivalent to the old group, but with the given cgroup
525 * substituted into the appropriate hierarchy. Must be called with
528 static struct css_set *find_css_set(
529 struct css_set *oldcg, struct cgroup *cgrp)
532 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
534 struct list_head tmp_cg_links;
536 struct hlist_head *hhead;
537 struct cg_cgroup_link *link;
539 /* First see if we already have a cgroup group that matches
541 read_lock(&css_set_lock);
542 res = find_existing_css_set(oldcg, cgrp, template);
545 read_unlock(&css_set_lock);
550 res = kmalloc(sizeof(*res), GFP_KERNEL);
554 /* Allocate all the cg_cgroup_link objects that we'll need */
555 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
560 atomic_set(&res->refcount, 1);
561 INIT_LIST_HEAD(&res->cg_links);
562 INIT_LIST_HEAD(&res->tasks);
563 INIT_HLIST_NODE(&res->hlist);
565 /* Copy the set of subsystem state objects generated in
566 * find_existing_css_set() */
567 memcpy(res->subsys, template, sizeof(res->subsys));
569 write_lock(&css_set_lock);
570 /* Add reference counts and links from the new css_set. */
571 list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
572 struct cgroup *c = link->cgrp;
573 if (c->root == cgrp->root)
575 link_css_set(&tmp_cg_links, res, c);
578 BUG_ON(!list_empty(&tmp_cg_links));
582 /* Add this cgroup group to the hash table */
583 hhead = css_set_hash(res->subsys);
584 hlist_add_head(&res->hlist, hhead);
586 write_unlock(&css_set_lock);
592 * Return the cgroup for "task" from the given hierarchy. Must be
593 * called with cgroup_mutex held.
595 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
596 struct cgroupfs_root *root)
599 struct cgroup *res = NULL;
601 BUG_ON(!mutex_is_locked(&cgroup_mutex));
602 read_lock(&css_set_lock);
604 * No need to lock the task - since we hold cgroup_mutex the
605 * task can't change groups, so the only thing that can happen
606 * is that it exits and its css is set back to init_css_set.
609 if (css == &init_css_set) {
610 res = &root->top_cgroup;
612 struct cg_cgroup_link *link;
613 list_for_each_entry(link, &css->cg_links, cg_link_list) {
614 struct cgroup *c = link->cgrp;
615 if (c->root == root) {
621 read_unlock(&css_set_lock);
627 * There is one global cgroup mutex. We also require taking
628 * task_lock() when dereferencing a task's cgroup subsys pointers.
629 * See "The task_lock() exception", at the end of this comment.
631 * A task must hold cgroup_mutex to modify cgroups.
633 * Any task can increment and decrement the count field without lock.
634 * So in general, code holding cgroup_mutex can't rely on the count
635 * field not changing. However, if the count goes to zero, then only
636 * cgroup_attach_task() can increment it again. Because a count of zero
637 * means that no tasks are currently attached, therefore there is no
638 * way a task attached to that cgroup can fork (the other way to
639 * increment the count). So code holding cgroup_mutex can safely
640 * assume that if the count is zero, it will stay zero. Similarly, if
641 * a task holds cgroup_mutex on a cgroup with zero count, it
642 * knows that the cgroup won't be removed, as cgroup_rmdir()
645 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
646 * (usually) take cgroup_mutex. These are the two most performance
647 * critical pieces of code here. The exception occurs on cgroup_exit(),
648 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
649 * is taken, and if the cgroup count is zero, a usermode call made
650 * to the release agent with the name of the cgroup (path relative to
651 * the root of cgroup file system) as the argument.
653 * A cgroup can only be deleted if both its 'count' of using tasks
654 * is zero, and its list of 'children' cgroups is empty. Since all
655 * tasks in the system use _some_ cgroup, and since there is always at
656 * least one task in the system (init, pid == 1), therefore, top_cgroup
657 * always has either children cgroups and/or using tasks. So we don't
658 * need a special hack to ensure that top_cgroup cannot be deleted.
660 * The task_lock() exception
662 * The need for this exception arises from the action of
663 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
664 * another. It does so using cgroup_mutex, however there are
665 * several performance critical places that need to reference
666 * task->cgroup without the expense of grabbing a system global
667 * mutex. Therefore except as noted below, when dereferencing or, as
668 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
669 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
670 * the task_struct routinely used for such matters.
672 * P.S. One more locking exception. RCU is used to guard the
673 * update of a tasks cgroup pointer by cgroup_attach_task()
677 * cgroup_lock - lock out any changes to cgroup structures
680 void cgroup_lock(void)
682 mutex_lock(&cgroup_mutex);
686 * cgroup_unlock - release lock on cgroup changes
688 * Undo the lock taken in a previous cgroup_lock() call.
690 void cgroup_unlock(void)
692 mutex_unlock(&cgroup_mutex);
696 * A couple of forward declarations required, due to cyclic reference loop:
697 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
698 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
702 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode);
703 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
704 static int cgroup_populate_dir(struct cgroup *cgrp);
705 static const struct inode_operations cgroup_dir_inode_operations;
706 static const struct file_operations proc_cgroupstats_operations;
708 static struct backing_dev_info cgroup_backing_dev_info = {
710 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
713 static int alloc_css_id(struct cgroup_subsys *ss,
714 struct cgroup *parent, struct cgroup *child);
716 static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb)
718 struct inode *inode = new_inode(sb);
721 inode->i_mode = mode;
722 inode->i_uid = current_fsuid();
723 inode->i_gid = current_fsgid();
724 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
725 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
731 * Call subsys's pre_destroy handler.
732 * This is called before css refcnt check.
734 static int cgroup_call_pre_destroy(struct cgroup *cgrp)
736 struct cgroup_subsys *ss;
739 for_each_subsys(cgrp->root, ss)
740 if (ss->pre_destroy) {
741 ret = ss->pre_destroy(ss, cgrp);
748 static void free_cgroup_rcu(struct rcu_head *obj)
750 struct cgroup *cgrp = container_of(obj, struct cgroup, rcu_head);
755 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
757 /* is dentry a directory ? if so, kfree() associated cgroup */
758 if (S_ISDIR(inode->i_mode)) {
759 struct cgroup *cgrp = dentry->d_fsdata;
760 struct cgroup_subsys *ss;
761 BUG_ON(!(cgroup_is_removed(cgrp)));
762 /* It's possible for external users to be holding css
763 * reference counts on a cgroup; css_put() needs to
764 * be able to access the cgroup after decrementing
765 * the reference count in order to know if it needs to
766 * queue the cgroup to be handled by the release
770 mutex_lock(&cgroup_mutex);
772 * Release the subsystem state objects.
774 for_each_subsys(cgrp->root, ss)
775 ss->destroy(ss, cgrp);
777 cgrp->root->number_of_cgroups--;
778 mutex_unlock(&cgroup_mutex);
781 * Drop the active superblock reference that we took when we
784 deactivate_super(cgrp->root->sb);
787 * if we're getting rid of the cgroup, refcount should ensure
788 * that there are no pidlists left.
790 BUG_ON(!list_empty(&cgrp->pidlists));
792 call_rcu(&cgrp->rcu_head, free_cgroup_rcu);
797 static void remove_dir(struct dentry *d)
799 struct dentry *parent = dget(d->d_parent);
802 simple_rmdir(parent->d_inode, d);
806 static void cgroup_clear_directory(struct dentry *dentry)
808 struct list_head *node;
810 BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
811 spin_lock(&dcache_lock);
812 node = dentry->d_subdirs.next;
813 while (node != &dentry->d_subdirs) {
814 struct dentry *d = list_entry(node, struct dentry, d_u.d_child);
817 /* This should never be called on a cgroup
818 * directory with child cgroups */
819 BUG_ON(d->d_inode->i_mode & S_IFDIR);
821 spin_unlock(&dcache_lock);
823 simple_unlink(dentry->d_inode, d);
825 spin_lock(&dcache_lock);
827 node = dentry->d_subdirs.next;
829 spin_unlock(&dcache_lock);
833 * NOTE : the dentry must have been dget()'ed
835 static void cgroup_d_remove_dir(struct dentry *dentry)
837 cgroup_clear_directory(dentry);
839 spin_lock(&dcache_lock);
840 list_del_init(&dentry->d_u.d_child);
841 spin_unlock(&dcache_lock);
846 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
847 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
848 * reference to css->refcnt. In general, this refcnt is expected to goes down
851 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
853 DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq);
855 static void cgroup_wakeup_rmdir_waiter(struct cgroup *cgrp)
857 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)))
858 wake_up_all(&cgroup_rmdir_waitq);
861 void cgroup_exclude_rmdir(struct cgroup_subsys_state *css)
866 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css)
868 cgroup_wakeup_rmdir_waiter(css->cgroup);
873 static int rebind_subsystems(struct cgroupfs_root *root,
874 unsigned long final_bits)
876 unsigned long added_bits, removed_bits;
877 struct cgroup *cgrp = &root->top_cgroup;
880 removed_bits = root->actual_subsys_bits & ~final_bits;
881 added_bits = final_bits & ~root->actual_subsys_bits;
882 /* Check that any added subsystems are currently free */
883 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
884 unsigned long bit = 1UL << i;
885 struct cgroup_subsys *ss = subsys[i];
886 if (!(bit & added_bits))
888 if (ss->root != &rootnode) {
889 /* Subsystem isn't free */
894 /* Currently we don't handle adding/removing subsystems when
895 * any child cgroups exist. This is theoretically supportable
896 * but involves complex error handling, so it's being left until
898 if (root->number_of_cgroups > 1)
901 /* Process each subsystem */
902 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
903 struct cgroup_subsys *ss = subsys[i];
904 unsigned long bit = 1UL << i;
905 if (bit & added_bits) {
906 /* We're binding this subsystem to this hierarchy */
907 BUG_ON(cgrp->subsys[i]);
908 BUG_ON(!dummytop->subsys[i]);
909 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
910 mutex_lock(&ss->hierarchy_mutex);
911 cgrp->subsys[i] = dummytop->subsys[i];
912 cgrp->subsys[i]->cgroup = cgrp;
913 list_move(&ss->sibling, &root->subsys_list);
917 mutex_unlock(&ss->hierarchy_mutex);
918 } else if (bit & removed_bits) {
919 /* We're removing this subsystem */
920 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
921 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
922 mutex_lock(&ss->hierarchy_mutex);
924 ss->bind(ss, dummytop);
925 dummytop->subsys[i]->cgroup = dummytop;
926 cgrp->subsys[i] = NULL;
927 subsys[i]->root = &rootnode;
928 list_move(&ss->sibling, &rootnode.subsys_list);
929 mutex_unlock(&ss->hierarchy_mutex);
930 } else if (bit & final_bits) {
931 /* Subsystem state should already exist */
932 BUG_ON(!cgrp->subsys[i]);
934 /* Subsystem state shouldn't exist */
935 BUG_ON(cgrp->subsys[i]);
938 root->subsys_bits = root->actual_subsys_bits = final_bits;
944 static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs)
946 struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info;
947 struct cgroup_subsys *ss;
949 mutex_lock(&cgroup_mutex);
950 for_each_subsys(root, ss)
951 seq_printf(seq, ",%s", ss->name);
952 if (test_bit(ROOT_NOPREFIX, &root->flags))
953 seq_puts(seq, ",noprefix");
954 if (strlen(root->release_agent_path))
955 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
956 if (strlen(root->name))
957 seq_printf(seq, ",name=%s", root->name);
958 mutex_unlock(&cgroup_mutex);
962 struct cgroup_sb_opts {
963 unsigned long subsys_bits;
967 /* User explicitly requested empty subsystem */
970 struct cgroupfs_root *new_root;
974 /* Convert a hierarchy specifier into a bitmask of subsystems and
976 static int parse_cgroupfs_options(char *data,
977 struct cgroup_sb_opts *opts)
979 char *token, *o = data ?: "all";
980 unsigned long mask = (unsigned long)-1;
982 #ifdef CONFIG_CPUSETS
983 mask = ~(1UL << cpuset_subsys_id);
986 memset(opts, 0, sizeof(*opts));
988 while ((token = strsep(&o, ",")) != NULL) {
991 if (!strcmp(token, "all")) {
992 /* Add all non-disabled subsystems */
994 opts->subsys_bits = 0;
995 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
996 struct cgroup_subsys *ss = subsys[i];
998 opts->subsys_bits |= 1ul << i;
1000 } else if (!strcmp(token, "none")) {
1001 /* Explicitly have no subsystems */
1003 } else if (!strcmp(token, "noprefix")) {
1004 set_bit(ROOT_NOPREFIX, &opts->flags);
1005 } else if (!strncmp(token, "release_agent=", 14)) {
1006 /* Specifying two release agents is forbidden */
1007 if (opts->release_agent)
1009 opts->release_agent =
1010 kstrndup(token + 14, PATH_MAX, GFP_KERNEL);
1011 if (!opts->release_agent)
1013 } else if (!strncmp(token, "name=", 5)) {
1015 const char *name = token + 5;
1016 /* Can't specify an empty name */
1019 /* Must match [\w.-]+ */
1020 for (i = 0; i < strlen(name); i++) {
1024 if ((c == '.') || (c == '-') || (c == '_'))
1028 /* Specifying two names is forbidden */
1031 opts->name = kstrndup(name,
1032 MAX_CGROUP_ROOT_NAMELEN,
1037 struct cgroup_subsys *ss;
1039 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1041 if (!strcmp(token, ss->name)) {
1043 set_bit(i, &opts->subsys_bits);
1047 if (i == CGROUP_SUBSYS_COUNT)
1052 /* Consistency checks */
1055 * Option noprefix was introduced just for backward compatibility
1056 * with the old cpuset, so we allow noprefix only if mounting just
1057 * the cpuset subsystem.
1059 if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1060 (opts->subsys_bits & mask))
1064 /* Can't specify "none" and some subsystems */
1065 if (opts->subsys_bits && opts->none)
1069 * We either have to specify by name or by subsystems. (So all
1070 * empty hierarchies must have a name).
1072 if (!opts->subsys_bits && !opts->name)
1078 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1081 struct cgroupfs_root *root = sb->s_fs_info;
1082 struct cgroup *cgrp = &root->top_cgroup;
1083 struct cgroup_sb_opts opts;
1086 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1087 mutex_lock(&cgroup_mutex);
1089 /* See what subsystems are wanted */
1090 ret = parse_cgroupfs_options(data, &opts);
1094 /* Don't allow flags to change at remount */
1095 if (opts.flags != root->flags) {
1100 /* Don't allow name to change at remount */
1101 if (opts.name && strcmp(opts.name, root->name)) {
1106 ret = rebind_subsystems(root, opts.subsys_bits);
1110 /* (re)populate subsystem files */
1111 cgroup_populate_dir(cgrp);
1113 if (opts.release_agent)
1114 strcpy(root->release_agent_path, opts.release_agent);
1116 kfree(opts.release_agent);
1118 mutex_unlock(&cgroup_mutex);
1119 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1124 static const struct super_operations cgroup_ops = {
1125 .statfs = simple_statfs,
1126 .drop_inode = generic_delete_inode,
1127 .show_options = cgroup_show_options,
1128 .remount_fs = cgroup_remount,
1131 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1133 INIT_LIST_HEAD(&cgrp->sibling);
1134 INIT_LIST_HEAD(&cgrp->children);
1135 INIT_LIST_HEAD(&cgrp->css_sets);
1136 INIT_LIST_HEAD(&cgrp->release_list);
1137 INIT_LIST_HEAD(&cgrp->pidlists);
1138 mutex_init(&cgrp->pidlist_mutex);
1141 static void init_cgroup_root(struct cgroupfs_root *root)
1143 struct cgroup *cgrp = &root->top_cgroup;
1144 INIT_LIST_HEAD(&root->subsys_list);
1145 INIT_LIST_HEAD(&root->root_list);
1146 root->number_of_cgroups = 1;
1148 cgrp->top_cgroup = cgrp;
1149 init_cgroup_housekeeping(cgrp);
1152 static bool init_root_id(struct cgroupfs_root *root)
1157 if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1159 spin_lock(&hierarchy_id_lock);
1160 /* Try to allocate the next unused ID */
1161 ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1162 &root->hierarchy_id);
1164 /* Try again starting from 0 */
1165 ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1167 next_hierarchy_id = root->hierarchy_id + 1;
1168 } else if (ret != -EAGAIN) {
1169 /* Can only get here if the 31-bit IDR is full ... */
1172 spin_unlock(&hierarchy_id_lock);
1177 static int cgroup_test_super(struct super_block *sb, void *data)
1179 struct cgroup_sb_opts *opts = data;
1180 struct cgroupfs_root *root = sb->s_fs_info;
1182 /* If we asked for a name then it must match */
1183 if (opts->name && strcmp(opts->name, root->name))
1187 * If we asked for subsystems (or explicitly for no
1188 * subsystems) then they must match
1190 if ((opts->subsys_bits || opts->none)
1191 && (opts->subsys_bits != root->subsys_bits))
1197 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1199 struct cgroupfs_root *root;
1201 if (!opts->subsys_bits && !opts->none)
1204 root = kzalloc(sizeof(*root), GFP_KERNEL);
1206 return ERR_PTR(-ENOMEM);
1208 if (!init_root_id(root)) {
1210 return ERR_PTR(-ENOMEM);
1212 init_cgroup_root(root);
1214 root->subsys_bits = opts->subsys_bits;
1215 root->flags = opts->flags;
1216 if (opts->release_agent)
1217 strcpy(root->release_agent_path, opts->release_agent);
1219 strcpy(root->name, opts->name);
1223 static void cgroup_drop_root(struct cgroupfs_root *root)
1228 BUG_ON(!root->hierarchy_id);
1229 spin_lock(&hierarchy_id_lock);
1230 ida_remove(&hierarchy_ida, root->hierarchy_id);
1231 spin_unlock(&hierarchy_id_lock);
1235 static int cgroup_set_super(struct super_block *sb, void *data)
1238 struct cgroup_sb_opts *opts = data;
1240 /* If we don't have a new root, we can't set up a new sb */
1241 if (!opts->new_root)
1244 BUG_ON(!opts->subsys_bits && !opts->none);
1246 ret = set_anon_super(sb, NULL);
1250 sb->s_fs_info = opts->new_root;
1251 opts->new_root->sb = sb;
1253 sb->s_blocksize = PAGE_CACHE_SIZE;
1254 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1255 sb->s_magic = CGROUP_SUPER_MAGIC;
1256 sb->s_op = &cgroup_ops;
1261 static int cgroup_get_rootdir(struct super_block *sb)
1263 struct inode *inode =
1264 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1265 struct dentry *dentry;
1270 inode->i_fop = &simple_dir_operations;
1271 inode->i_op = &cgroup_dir_inode_operations;
1272 /* directories start off with i_nlink == 2 (for "." entry) */
1274 dentry = d_alloc_root(inode);
1279 sb->s_root = dentry;
1283 static int cgroup_get_sb(struct file_system_type *fs_type,
1284 int flags, const char *unused_dev_name,
1285 void *data, struct vfsmount *mnt)
1287 struct cgroup_sb_opts opts;
1288 struct cgroupfs_root *root;
1290 struct super_block *sb;
1291 struct cgroupfs_root *new_root;
1293 /* First find the desired set of subsystems */
1294 ret = parse_cgroupfs_options(data, &opts);
1299 * Allocate a new cgroup root. We may not need it if we're
1300 * reusing an existing hierarchy.
1302 new_root = cgroup_root_from_opts(&opts);
1303 if (IS_ERR(new_root)) {
1304 ret = PTR_ERR(new_root);
1307 opts.new_root = new_root;
1309 /* Locate an existing or new sb for this hierarchy */
1310 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, &opts);
1313 cgroup_drop_root(opts.new_root);
1317 root = sb->s_fs_info;
1319 if (root == opts.new_root) {
1320 /* We used the new root structure, so this is a new hierarchy */
1321 struct list_head tmp_cg_links;
1322 struct cgroup *root_cgrp = &root->top_cgroup;
1323 struct inode *inode;
1324 struct cgroupfs_root *existing_root;
1327 BUG_ON(sb->s_root != NULL);
1329 ret = cgroup_get_rootdir(sb);
1331 goto drop_new_super;
1332 inode = sb->s_root->d_inode;
1334 mutex_lock(&inode->i_mutex);
1335 mutex_lock(&cgroup_mutex);
1337 if (strlen(root->name)) {
1338 /* Check for name clashes with existing mounts */
1339 for_each_active_root(existing_root) {
1340 if (!strcmp(existing_root->name, root->name)) {
1342 mutex_unlock(&cgroup_mutex);
1343 mutex_unlock(&inode->i_mutex);
1344 goto drop_new_super;
1350 * We're accessing css_set_count without locking
1351 * css_set_lock here, but that's OK - it can only be
1352 * increased by someone holding cgroup_lock, and
1353 * that's us. The worst that can happen is that we
1354 * have some link structures left over
1356 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1358 mutex_unlock(&cgroup_mutex);
1359 mutex_unlock(&inode->i_mutex);
1360 goto drop_new_super;
1363 ret = rebind_subsystems(root, root->subsys_bits);
1364 if (ret == -EBUSY) {
1365 mutex_unlock(&cgroup_mutex);
1366 mutex_unlock(&inode->i_mutex);
1367 free_cg_links(&tmp_cg_links);
1368 goto drop_new_super;
1371 /* EBUSY should be the only error here */
1374 list_add(&root->root_list, &roots);
1377 sb->s_root->d_fsdata = root_cgrp;
1378 root->top_cgroup.dentry = sb->s_root;
1380 /* Link the top cgroup in this hierarchy into all
1381 * the css_set objects */
1382 write_lock(&css_set_lock);
1383 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
1384 struct hlist_head *hhead = &css_set_table[i];
1385 struct hlist_node *node;
1388 hlist_for_each_entry(cg, node, hhead, hlist)
1389 link_css_set(&tmp_cg_links, cg, root_cgrp);
1391 write_unlock(&css_set_lock);
1393 free_cg_links(&tmp_cg_links);
1395 BUG_ON(!list_empty(&root_cgrp->sibling));
1396 BUG_ON(!list_empty(&root_cgrp->children));
1397 BUG_ON(root->number_of_cgroups != 1);
1399 cgroup_populate_dir(root_cgrp);
1400 mutex_unlock(&cgroup_mutex);
1401 mutex_unlock(&inode->i_mutex);
1404 * We re-used an existing hierarchy - the new root (if
1405 * any) is not needed
1407 cgroup_drop_root(opts.new_root);
1410 simple_set_mnt(mnt, sb);
1411 kfree(opts.release_agent);
1416 deactivate_locked_super(sb);
1418 kfree(opts.release_agent);
1424 static void cgroup_kill_sb(struct super_block *sb) {
1425 struct cgroupfs_root *root = sb->s_fs_info;
1426 struct cgroup *cgrp = &root->top_cgroup;
1428 struct cg_cgroup_link *link;
1429 struct cg_cgroup_link *saved_link;
1433 BUG_ON(root->number_of_cgroups != 1);
1434 BUG_ON(!list_empty(&cgrp->children));
1435 BUG_ON(!list_empty(&cgrp->sibling));
1437 mutex_lock(&cgroup_mutex);
1439 /* Rebind all subsystems back to the default hierarchy */
1440 ret = rebind_subsystems(root, 0);
1441 /* Shouldn't be able to fail ... */
1445 * Release all the links from css_sets to this hierarchy's
1448 write_lock(&css_set_lock);
1450 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1452 list_del(&link->cg_link_list);
1453 list_del(&link->cgrp_link_list);
1456 write_unlock(&css_set_lock);
1458 if (!list_empty(&root->root_list)) {
1459 list_del(&root->root_list);
1463 mutex_unlock(&cgroup_mutex);
1465 kill_litter_super(sb);
1466 cgroup_drop_root(root);
1469 static struct file_system_type cgroup_fs_type = {
1471 .get_sb = cgroup_get_sb,
1472 .kill_sb = cgroup_kill_sb,
1475 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
1477 return dentry->d_fsdata;
1480 static inline struct cftype *__d_cft(struct dentry *dentry)
1482 return dentry->d_fsdata;
1486 * cgroup_path - generate the path of a cgroup
1487 * @cgrp: the cgroup in question
1488 * @buf: the buffer to write the path into
1489 * @buflen: the length of the buffer
1491 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1492 * reference. Writes path of cgroup into buf. Returns 0 on success,
1495 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1498 struct dentry *dentry = rcu_dereference(cgrp->dentry);
1500 if (!dentry || cgrp == dummytop) {
1502 * Inactive subsystems have no dentry for their root
1509 start = buf + buflen;
1513 int len = dentry->d_name.len;
1514 if ((start -= len) < buf)
1515 return -ENAMETOOLONG;
1516 memcpy(start, cgrp->dentry->d_name.name, len);
1517 cgrp = cgrp->parent;
1520 dentry = rcu_dereference(cgrp->dentry);
1524 return -ENAMETOOLONG;
1527 memmove(buf, start, buf + buflen - start);
1532 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1533 * @cgrp: the cgroup the task is attaching to
1534 * @tsk: the task to be attached
1536 * Call holding cgroup_mutex. May take task_lock of
1537 * the task 'tsk' during call.
1539 int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1542 struct cgroup_subsys *ss;
1543 struct cgroup *oldcgrp;
1545 struct css_set *newcg;
1546 struct cgroupfs_root *root = cgrp->root;
1548 /* Nothing to do if the task is already in that cgroup */
1549 oldcgrp = task_cgroup_from_root(tsk, root);
1550 if (cgrp == oldcgrp)
1553 for_each_subsys(root, ss) {
1554 if (ss->can_attach) {
1555 retval = ss->can_attach(ss, cgrp, tsk, false);
1566 * Locate or allocate a new css_set for this task,
1567 * based on its final set of cgroups
1569 newcg = find_css_set(cg, cgrp);
1575 if (tsk->flags & PF_EXITING) {
1580 rcu_assign_pointer(tsk->cgroups, newcg);
1583 /* Update the css_set linked lists if we're using them */
1584 write_lock(&css_set_lock);
1585 if (!list_empty(&tsk->cg_list)) {
1586 list_del(&tsk->cg_list);
1587 list_add(&tsk->cg_list, &newcg->tasks);
1589 write_unlock(&css_set_lock);
1591 for_each_subsys(root, ss) {
1593 ss->attach(ss, cgrp, oldcgrp, tsk, false);
1595 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1600 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1601 * is no longer empty.
1603 cgroup_wakeup_rmdir_waiter(cgrp);
1608 * Attach task with pid 'pid' to cgroup 'cgrp'. Call with cgroup_mutex
1609 * held. May take task_lock of task
1611 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid)
1613 struct task_struct *tsk;
1614 const struct cred *cred = current_cred(), *tcred;
1619 tsk = find_task_by_vpid(pid);
1620 if (!tsk || tsk->flags & PF_EXITING) {
1625 tcred = __task_cred(tsk);
1627 cred->euid != tcred->uid &&
1628 cred->euid != tcred->suid) {
1632 get_task_struct(tsk);
1636 get_task_struct(tsk);
1639 ret = cgroup_attach_task(cgrp, tsk);
1640 put_task_struct(tsk);
1644 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
1647 if (!cgroup_lock_live_group(cgrp))
1649 ret = attach_task_by_pid(cgrp, pid);
1655 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
1656 * @cgrp: the cgroup to be checked for liveness
1658 * On success, returns true; the lock should be later released with
1659 * cgroup_unlock(). On failure returns false with no lock held.
1661 bool cgroup_lock_live_group(struct cgroup *cgrp)
1663 mutex_lock(&cgroup_mutex);
1664 if (cgroup_is_removed(cgrp)) {
1665 mutex_unlock(&cgroup_mutex);
1671 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
1674 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
1675 if (!cgroup_lock_live_group(cgrp))
1677 strcpy(cgrp->root->release_agent_path, buffer);
1682 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
1683 struct seq_file *seq)
1685 if (!cgroup_lock_live_group(cgrp))
1687 seq_puts(seq, cgrp->root->release_agent_path);
1688 seq_putc(seq, '\n');
1693 /* A buffer size big enough for numbers or short strings */
1694 #define CGROUP_LOCAL_BUFFER_SIZE 64
1696 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
1698 const char __user *userbuf,
1699 size_t nbytes, loff_t *unused_ppos)
1701 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
1707 if (nbytes >= sizeof(buffer))
1709 if (copy_from_user(buffer, userbuf, nbytes))
1712 buffer[nbytes] = 0; /* nul-terminate */
1713 if (cft->write_u64) {
1714 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
1717 retval = cft->write_u64(cgrp, cft, val);
1719 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
1722 retval = cft->write_s64(cgrp, cft, val);
1729 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
1731 const char __user *userbuf,
1732 size_t nbytes, loff_t *unused_ppos)
1734 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
1736 size_t max_bytes = cft->max_write_len;
1737 char *buffer = local_buffer;
1740 max_bytes = sizeof(local_buffer) - 1;
1741 if (nbytes >= max_bytes)
1743 /* Allocate a dynamic buffer if we need one */
1744 if (nbytes >= sizeof(local_buffer)) {
1745 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
1749 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
1754 buffer[nbytes] = 0; /* nul-terminate */
1755 retval = cft->write_string(cgrp, cft, strstrip(buffer));
1759 if (buffer != local_buffer)
1764 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
1765 size_t nbytes, loff_t *ppos)
1767 struct cftype *cft = __d_cft(file->f_dentry);
1768 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
1770 if (cgroup_is_removed(cgrp))
1773 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
1774 if (cft->write_u64 || cft->write_s64)
1775 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
1776 if (cft->write_string)
1777 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
1779 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
1780 return ret ? ret : nbytes;
1785 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
1787 char __user *buf, size_t nbytes,
1790 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
1791 u64 val = cft->read_u64(cgrp, cft);
1792 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
1794 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
1797 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
1799 char __user *buf, size_t nbytes,
1802 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
1803 s64 val = cft->read_s64(cgrp, cft);
1804 int len = sprintf(tmp, "%lld\n", (long long) val);
1806 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
1809 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
1810 size_t nbytes, loff_t *ppos)
1812 struct cftype *cft = __d_cft(file->f_dentry);
1813 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
1815 if (cgroup_is_removed(cgrp))
1819 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
1821 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
1823 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
1828 * seqfile ops/methods for returning structured data. Currently just
1829 * supports string->u64 maps, but can be extended in future.
1832 struct cgroup_seqfile_state {
1834 struct cgroup *cgroup;
1837 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
1839 struct seq_file *sf = cb->state;
1840 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
1843 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
1845 struct cgroup_seqfile_state *state = m->private;
1846 struct cftype *cft = state->cft;
1847 if (cft->read_map) {
1848 struct cgroup_map_cb cb = {
1849 .fill = cgroup_map_add,
1852 return cft->read_map(state->cgroup, cft, &cb);
1854 return cft->read_seq_string(state->cgroup, cft, m);
1857 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
1859 struct seq_file *seq = file->private_data;
1860 kfree(seq->private);
1861 return single_release(inode, file);
1864 static const struct file_operations cgroup_seqfile_operations = {
1866 .write = cgroup_file_write,
1867 .llseek = seq_lseek,
1868 .release = cgroup_seqfile_release,
1871 static int cgroup_file_open(struct inode *inode, struct file *file)
1876 err = generic_file_open(inode, file);
1879 cft = __d_cft(file->f_dentry);
1881 if (cft->read_map || cft->read_seq_string) {
1882 struct cgroup_seqfile_state *state =
1883 kzalloc(sizeof(*state), GFP_USER);
1887 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
1888 file->f_op = &cgroup_seqfile_operations;
1889 err = single_open(file, cgroup_seqfile_show, state);
1892 } else if (cft->open)
1893 err = cft->open(inode, file);
1900 static int cgroup_file_release(struct inode *inode, struct file *file)
1902 struct cftype *cft = __d_cft(file->f_dentry);
1904 return cft->release(inode, file);
1909 * cgroup_rename - Only allow simple rename of directories in place.
1911 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
1912 struct inode *new_dir, struct dentry *new_dentry)
1914 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1916 if (new_dentry->d_inode)
1918 if (old_dir != new_dir)
1920 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
1923 static const struct file_operations cgroup_file_operations = {
1924 .read = cgroup_file_read,
1925 .write = cgroup_file_write,
1926 .llseek = generic_file_llseek,
1927 .open = cgroup_file_open,
1928 .release = cgroup_file_release,
1931 static const struct inode_operations cgroup_dir_inode_operations = {
1932 .lookup = simple_lookup,
1933 .mkdir = cgroup_mkdir,
1934 .rmdir = cgroup_rmdir,
1935 .rename = cgroup_rename,
1938 static int cgroup_create_file(struct dentry *dentry, mode_t mode,
1939 struct super_block *sb)
1941 static const struct dentry_operations cgroup_dops = {
1942 .d_iput = cgroup_diput,
1945 struct inode *inode;
1949 if (dentry->d_inode)
1952 inode = cgroup_new_inode(mode, sb);
1956 if (S_ISDIR(mode)) {
1957 inode->i_op = &cgroup_dir_inode_operations;
1958 inode->i_fop = &simple_dir_operations;
1960 /* start off with i_nlink == 2 (for "." entry) */
1963 /* start with the directory inode held, so that we can
1964 * populate it without racing with another mkdir */
1965 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1966 } else if (S_ISREG(mode)) {
1968 inode->i_fop = &cgroup_file_operations;
1970 dentry->d_op = &cgroup_dops;
1971 d_instantiate(dentry, inode);
1972 dget(dentry); /* Extra count - pin the dentry in core */
1977 * cgroup_create_dir - create a directory for an object.
1978 * @cgrp: the cgroup we create the directory for. It must have a valid
1979 * ->parent field. And we are going to fill its ->dentry field.
1980 * @dentry: dentry of the new cgroup
1981 * @mode: mode to set on new directory.
1983 static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
1986 struct dentry *parent;
1989 parent = cgrp->parent->dentry;
1990 error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
1992 dentry->d_fsdata = cgrp;
1993 inc_nlink(parent->d_inode);
1994 rcu_assign_pointer(cgrp->dentry, dentry);
2003 * cgroup_file_mode - deduce file mode of a control file
2004 * @cft: the control file in question
2006 * returns cft->mode if ->mode is not 0
2007 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2008 * returns S_IRUGO if it has only a read handler
2009 * returns S_IWUSR if it has only a write hander
2011 static mode_t cgroup_file_mode(const struct cftype *cft)
2018 if (cft->read || cft->read_u64 || cft->read_s64 ||
2019 cft->read_map || cft->read_seq_string)
2022 if (cft->write || cft->write_u64 || cft->write_s64 ||
2023 cft->write_string || cft->trigger)
2029 int cgroup_add_file(struct cgroup *cgrp,
2030 struct cgroup_subsys *subsys,
2031 const struct cftype *cft)
2033 struct dentry *dir = cgrp->dentry;
2034 struct dentry *dentry;
2038 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2039 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
2040 strcpy(name, subsys->name);
2043 strcat(name, cft->name);
2044 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2045 dentry = lookup_one_len(name, dir, strlen(name));
2046 if (!IS_ERR(dentry)) {
2047 mode = cgroup_file_mode(cft);
2048 error = cgroup_create_file(dentry, mode | S_IFREG,
2051 dentry->d_fsdata = (void *)cft;
2054 error = PTR_ERR(dentry);
2058 int cgroup_add_files(struct cgroup *cgrp,
2059 struct cgroup_subsys *subsys,
2060 const struct cftype cft[],
2064 for (i = 0; i < count; i++) {
2065 err = cgroup_add_file(cgrp, subsys, &cft[i]);
2073 * cgroup_task_count - count the number of tasks in a cgroup.
2074 * @cgrp: the cgroup in question
2076 * Return the number of tasks in the cgroup.
2078 int cgroup_task_count(const struct cgroup *cgrp)
2081 struct cg_cgroup_link *link;
2083 read_lock(&css_set_lock);
2084 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2085 count += atomic_read(&link->cg->refcount);
2087 read_unlock(&css_set_lock);
2092 * Advance a list_head iterator. The iterator should be positioned at
2093 * the start of a css_set
2095 static void cgroup_advance_iter(struct cgroup *cgrp,
2096 struct cgroup_iter *it)
2098 struct list_head *l = it->cg_link;
2099 struct cg_cgroup_link *link;
2102 /* Advance to the next non-empty css_set */
2105 if (l == &cgrp->css_sets) {
2109 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2111 } while (list_empty(&cg->tasks));
2113 it->task = cg->tasks.next;
2117 * To reduce the fork() overhead for systems that are not actually
2118 * using their cgroups capability, we don't maintain the lists running
2119 * through each css_set to its tasks until we see the list actually
2120 * used - in other words after the first call to cgroup_iter_start().
2122 * The tasklist_lock is not held here, as do_each_thread() and
2123 * while_each_thread() are protected by RCU.
2125 static void cgroup_enable_task_cg_lists(void)
2127 struct task_struct *p, *g;
2128 write_lock(&css_set_lock);
2129 use_task_css_set_links = 1;
2130 do_each_thread(g, p) {
2133 * We should check if the process is exiting, otherwise
2134 * it will race with cgroup_exit() in that the list
2135 * entry won't be deleted though the process has exited.
2137 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2138 list_add(&p->cg_list, &p->cgroups->tasks);
2140 } while_each_thread(g, p);
2141 write_unlock(&css_set_lock);
2144 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
2147 * The first time anyone tries to iterate across a cgroup,
2148 * we need to enable the list linking each css_set to its
2149 * tasks, and fix up all existing tasks.
2151 if (!use_task_css_set_links)
2152 cgroup_enable_task_cg_lists();
2154 read_lock(&css_set_lock);
2155 it->cg_link = &cgrp->css_sets;
2156 cgroup_advance_iter(cgrp, it);
2159 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
2160 struct cgroup_iter *it)
2162 struct task_struct *res;
2163 struct list_head *l = it->task;
2164 struct cg_cgroup_link *link;
2166 /* If the iterator cg is NULL, we have no tasks */
2169 res = list_entry(l, struct task_struct, cg_list);
2170 /* Advance iterator to find next entry */
2172 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
2173 if (l == &link->cg->tasks) {
2174 /* We reached the end of this task list - move on to
2175 * the next cg_cgroup_link */
2176 cgroup_advance_iter(cgrp, it);
2183 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
2185 read_unlock(&css_set_lock);
2188 static inline int started_after_time(struct task_struct *t1,
2189 struct timespec *time,
2190 struct task_struct *t2)
2192 int start_diff = timespec_compare(&t1->start_time, time);
2193 if (start_diff > 0) {
2195 } else if (start_diff < 0) {
2199 * Arbitrarily, if two processes started at the same
2200 * time, we'll say that the lower pointer value
2201 * started first. Note that t2 may have exited by now
2202 * so this may not be a valid pointer any longer, but
2203 * that's fine - it still serves to distinguish
2204 * between two tasks started (effectively) simultaneously.
2211 * This function is a callback from heap_insert() and is used to order
2213 * In this case we order the heap in descending task start time.
2215 static inline int started_after(void *p1, void *p2)
2217 struct task_struct *t1 = p1;
2218 struct task_struct *t2 = p2;
2219 return started_after_time(t1, &t2->start_time, t2);
2223 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
2224 * @scan: struct cgroup_scanner containing arguments for the scan
2226 * Arguments include pointers to callback functions test_task() and
2228 * Iterate through all the tasks in a cgroup, calling test_task() for each,
2229 * and if it returns true, call process_task() for it also.
2230 * The test_task pointer may be NULL, meaning always true (select all tasks).
2231 * Effectively duplicates cgroup_iter_{start,next,end}()
2232 * but does not lock css_set_lock for the call to process_task().
2233 * The struct cgroup_scanner may be embedded in any structure of the caller's
2235 * It is guaranteed that process_task() will act on every task that
2236 * is a member of the cgroup for the duration of this call. This
2237 * function may or may not call process_task() for tasks that exit
2238 * or move to a different cgroup during the call, or are forked or
2239 * move into the cgroup during the call.
2241 * Note that test_task() may be called with locks held, and may in some
2242 * situations be called multiple times for the same task, so it should
2244 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
2245 * pre-allocated and will be used for heap operations (and its "gt" member will
2246 * be overwritten), else a temporary heap will be used (allocation of which
2247 * may cause this function to fail).
2249 int cgroup_scan_tasks(struct cgroup_scanner *scan)
2252 struct cgroup_iter it;
2253 struct task_struct *p, *dropped;
2254 /* Never dereference latest_task, since it's not refcounted */
2255 struct task_struct *latest_task = NULL;
2256 struct ptr_heap tmp_heap;
2257 struct ptr_heap *heap;
2258 struct timespec latest_time = { 0, 0 };
2261 /* The caller supplied our heap and pre-allocated its memory */
2263 heap->gt = &started_after;
2265 /* We need to allocate our own heap memory */
2267 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
2269 /* cannot allocate the heap */
2275 * Scan tasks in the cgroup, using the scanner's "test_task" callback
2276 * to determine which are of interest, and using the scanner's
2277 * "process_task" callback to process any of them that need an update.
2278 * Since we don't want to hold any locks during the task updates,
2279 * gather tasks to be processed in a heap structure.
2280 * The heap is sorted by descending task start time.
2281 * If the statically-sized heap fills up, we overflow tasks that
2282 * started later, and in future iterations only consider tasks that
2283 * started after the latest task in the previous pass. This
2284 * guarantees forward progress and that we don't miss any tasks.
2287 cgroup_iter_start(scan->cg, &it);
2288 while ((p = cgroup_iter_next(scan->cg, &it))) {
2290 * Only affect tasks that qualify per the caller's callback,
2291 * if he provided one
2293 if (scan->test_task && !scan->test_task(p, scan))
2296 * Only process tasks that started after the last task
2299 if (!started_after_time(p, &latest_time, latest_task))
2301 dropped = heap_insert(heap, p);
2302 if (dropped == NULL) {
2304 * The new task was inserted; the heap wasn't
2308 } else if (dropped != p) {
2310 * The new task was inserted, and pushed out a
2314 put_task_struct(dropped);
2317 * Else the new task was newer than anything already in
2318 * the heap and wasn't inserted
2321 cgroup_iter_end(scan->cg, &it);
2324 for (i = 0; i < heap->size; i++) {
2325 struct task_struct *q = heap->ptrs[i];
2327 latest_time = q->start_time;
2330 /* Process the task per the caller's callback */
2331 scan->process_task(q, scan);
2335 * If we had to process any tasks at all, scan again
2336 * in case some of them were in the middle of forking
2337 * children that didn't get processed.
2338 * Not the most efficient way to do it, but it avoids
2339 * having to take callback_mutex in the fork path
2343 if (heap == &tmp_heap)
2344 heap_free(&tmp_heap);
2349 * Stuff for reading the 'tasks'/'procs' files.
2351 * Reading this file can return large amounts of data if a cgroup has
2352 * *lots* of attached tasks. So it may need several calls to read(),
2353 * but we cannot guarantee that the information we produce is correct
2354 * unless we produce it entirely atomically.
2359 * The following two functions "fix" the issue where there are more pids
2360 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
2361 * TODO: replace with a kernel-wide solution to this problem
2363 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
2364 static void *pidlist_allocate(int count)
2366 if (PIDLIST_TOO_LARGE(count))
2367 return vmalloc(count * sizeof(pid_t));
2369 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
2371 static void pidlist_free(void *p)
2373 if (is_vmalloc_addr(p))
2378 static void *pidlist_resize(void *p, int newcount)
2381 /* note: if new alloc fails, old p will still be valid either way */
2382 if (is_vmalloc_addr(p)) {
2383 newlist = vmalloc(newcount * sizeof(pid_t));
2386 memcpy(newlist, p, newcount * sizeof(pid_t));
2389 newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL);
2395 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
2396 * If the new stripped list is sufficiently smaller and there's enough memory
2397 * to allocate a new buffer, will let go of the unneeded memory. Returns the
2398 * number of unique elements.
2400 /* is the size difference enough that we should re-allocate the array? */
2401 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
2402 static int pidlist_uniq(pid_t **p, int length)
2409 * we presume the 0th element is unique, so i starts at 1. trivial
2410 * edge cases first; no work needs to be done for either
2412 if (length == 0 || length == 1)
2414 /* src and dest walk down the list; dest counts unique elements */
2415 for (src = 1; src < length; src++) {
2416 /* find next unique element */
2417 while (list[src] == list[src-1]) {
2422 /* dest always points to where the next unique element goes */
2423 list[dest] = list[src];
2428 * if the length difference is large enough, we want to allocate a
2429 * smaller buffer to save memory. if this fails due to out of memory,
2430 * we'll just stay with what we've got.
2432 if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) {
2433 newlist = pidlist_resize(list, dest);
2440 static int cmppid(const void *a, const void *b)
2442 return *(pid_t *)a - *(pid_t *)b;
2446 * find the appropriate pidlist for our purpose (given procs vs tasks)
2447 * returns with the lock on that pidlist already held, and takes care
2448 * of the use count, or returns NULL with no locks held if we're out of
2451 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
2452 enum cgroup_filetype type)
2454 struct cgroup_pidlist *l;
2455 /* don't need task_nsproxy() if we're looking at ourself */
2456 struct pid_namespace *ns = get_pid_ns(current->nsproxy->pid_ns);
2458 * We can't drop the pidlist_mutex before taking the l->mutex in case
2459 * the last ref-holder is trying to remove l from the list at the same
2460 * time. Holding the pidlist_mutex precludes somebody taking whichever
2461 * list we find out from under us - compare release_pid_array().
2463 mutex_lock(&cgrp->pidlist_mutex);
2464 list_for_each_entry(l, &cgrp->pidlists, links) {
2465 if (l->key.type == type && l->key.ns == ns) {
2466 /* found a matching list - drop the extra refcount */
2468 /* make sure l doesn't vanish out from under us */
2469 down_write(&l->mutex);
2470 mutex_unlock(&cgrp->pidlist_mutex);
2474 /* entry not found; create a new one */
2475 l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
2477 mutex_unlock(&cgrp->pidlist_mutex);
2481 init_rwsem(&l->mutex);
2482 down_write(&l->mutex);
2485 l->use_count = 0; /* don't increment here */
2488 list_add(&l->links, &cgrp->pidlists);
2489 mutex_unlock(&cgrp->pidlist_mutex);
2494 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
2496 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
2497 struct cgroup_pidlist **lp)
2501 int pid, n = 0; /* used for populating the array */
2502 struct cgroup_iter it;
2503 struct task_struct *tsk;
2504 struct cgroup_pidlist *l;
2507 * If cgroup gets more users after we read count, we won't have
2508 * enough space - tough. This race is indistinguishable to the
2509 * caller from the case that the additional cgroup users didn't
2510 * show up until sometime later on.
2512 length = cgroup_task_count(cgrp);
2513 array = pidlist_allocate(length);
2516 /* now, populate the array */
2517 cgroup_iter_start(cgrp, &it);
2518 while ((tsk = cgroup_iter_next(cgrp, &it))) {
2519 if (unlikely(n == length))
2521 /* get tgid or pid for procs or tasks file respectively */
2522 if (type == CGROUP_FILE_PROCS)
2523 pid = task_tgid_vnr(tsk);
2525 pid = task_pid_vnr(tsk);
2526 if (pid > 0) /* make sure to only use valid results */
2529 cgroup_iter_end(cgrp, &it);
2531 /* now sort & (if procs) strip out duplicates */
2532 sort(array, length, sizeof(pid_t), cmppid, NULL);
2533 if (type == CGROUP_FILE_PROCS)
2534 length = pidlist_uniq(&array, length);
2535 l = cgroup_pidlist_find(cgrp, type);
2537 pidlist_free(array);
2540 /* store array, freeing old if necessary - lock already held */
2541 pidlist_free(l->list);
2545 up_write(&l->mutex);
2551 * cgroupstats_build - build and fill cgroupstats
2552 * @stats: cgroupstats to fill information into
2553 * @dentry: A dentry entry belonging to the cgroup for which stats have
2556 * Build and fill cgroupstats so that taskstats can export it to user
2559 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
2562 struct cgroup *cgrp;
2563 struct cgroup_iter it;
2564 struct task_struct *tsk;
2567 * Validate dentry by checking the superblock operations,
2568 * and make sure it's a directory.
2570 if (dentry->d_sb->s_op != &cgroup_ops ||
2571 !S_ISDIR(dentry->d_inode->i_mode))
2575 cgrp = dentry->d_fsdata;
2577 cgroup_iter_start(cgrp, &it);
2578 while ((tsk = cgroup_iter_next(cgrp, &it))) {
2579 switch (tsk->state) {
2581 stats->nr_running++;
2583 case TASK_INTERRUPTIBLE:
2584 stats->nr_sleeping++;
2586 case TASK_UNINTERRUPTIBLE:
2587 stats->nr_uninterruptible++;
2590 stats->nr_stopped++;
2593 if (delayacct_is_task_waiting_on_io(tsk))
2594 stats->nr_io_wait++;
2598 cgroup_iter_end(cgrp, &it);
2606 * seq_file methods for the tasks/procs files. The seq_file position is the
2607 * next pid to display; the seq_file iterator is a pointer to the pid
2608 * in the cgroup->l->list array.
2611 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
2614 * Initially we receive a position value that corresponds to
2615 * one more than the last pid shown (or 0 on the first call or
2616 * after a seek to the start). Use a binary-search to find the
2617 * next pid to display, if any
2619 struct cgroup_pidlist *l = s->private;
2620 int index = 0, pid = *pos;
2623 down_read(&l->mutex);
2625 int end = l->length;
2627 while (index < end) {
2628 int mid = (index + end) / 2;
2629 if (l->list[mid] == pid) {
2632 } else if (l->list[mid] <= pid)
2638 /* If we're off the end of the array, we're done */
2639 if (index >= l->length)
2641 /* Update the abstract position to be the actual pid that we found */
2642 iter = l->list + index;
2647 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
2649 struct cgroup_pidlist *l = s->private;
2653 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
2655 struct cgroup_pidlist *l = s->private;
2657 pid_t *end = l->list + l->length;
2659 * Advance to the next pid in the array. If this goes off the
2671 static int cgroup_pidlist_show(struct seq_file *s, void *v)
2673 return seq_printf(s, "%d\n", *(int *)v);
2677 * seq_operations functions for iterating on pidlists through seq_file -
2678 * independent of whether it's tasks or procs
2680 static const struct seq_operations cgroup_pidlist_seq_operations = {
2681 .start = cgroup_pidlist_start,
2682 .stop = cgroup_pidlist_stop,
2683 .next = cgroup_pidlist_next,
2684 .show = cgroup_pidlist_show,
2687 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
2690 * the case where we're the last user of this particular pidlist will
2691 * have us remove it from the cgroup's list, which entails taking the
2692 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
2693 * pidlist_mutex, we have to take pidlist_mutex first.
2695 mutex_lock(&l->owner->pidlist_mutex);
2696 down_write(&l->mutex);
2697 BUG_ON(!l->use_count);
2698 if (!--l->use_count) {
2699 /* we're the last user if refcount is 0; remove and free */
2700 list_del(&l->links);
2701 mutex_unlock(&l->owner->pidlist_mutex);
2702 pidlist_free(l->list);
2703 put_pid_ns(l->key.ns);
2704 up_write(&l->mutex);
2708 mutex_unlock(&l->owner->pidlist_mutex);
2709 up_write(&l->mutex);
2712 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
2714 struct cgroup_pidlist *l;
2715 if (!(file->f_mode & FMODE_READ))
2718 * the seq_file will only be initialized if the file was opened for
2719 * reading; hence we check if it's not null only in that case.
2721 l = ((struct seq_file *)file->private_data)->private;
2722 cgroup_release_pid_array(l);
2723 return seq_release(inode, file);
2726 static const struct file_operations cgroup_pidlist_operations = {
2728 .llseek = seq_lseek,
2729 .write = cgroup_file_write,
2730 .release = cgroup_pidlist_release,
2734 * The following functions handle opens on a file that displays a pidlist
2735 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
2738 /* helper function for the two below it */
2739 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
2741 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2742 struct cgroup_pidlist *l;
2745 /* Nothing to do for write-only files */
2746 if (!(file->f_mode & FMODE_READ))
2749 /* have the array populated */
2750 retval = pidlist_array_load(cgrp, type, &l);
2753 /* configure file information */
2754 file->f_op = &cgroup_pidlist_operations;
2756 retval = seq_open(file, &cgroup_pidlist_seq_operations);
2758 cgroup_release_pid_array(l);
2761 ((struct seq_file *)file->private_data)->private = l;
2764 static int cgroup_tasks_open(struct inode *unused, struct file *file)
2766 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
2768 static int cgroup_procs_open(struct inode *unused, struct file *file)
2770 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
2773 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
2776 return notify_on_release(cgrp);
2779 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
2783 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
2785 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
2787 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
2792 * for the common functions, 'private' gives the type of file
2794 /* for hysterical raisins, we can't put this on the older files */
2795 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
2796 static struct cftype files[] = {
2799 .open = cgroup_tasks_open,
2800 .write_u64 = cgroup_tasks_write,
2801 .release = cgroup_pidlist_release,
2802 .mode = S_IRUGO | S_IWUSR,
2805 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
2806 .open = cgroup_procs_open,
2807 /* .write_u64 = cgroup_procs_write, TODO */
2808 .release = cgroup_pidlist_release,
2812 .name = "notify_on_release",
2813 .read_u64 = cgroup_read_notify_on_release,
2814 .write_u64 = cgroup_write_notify_on_release,
2818 static struct cftype cft_release_agent = {
2819 .name = "release_agent",
2820 .read_seq_string = cgroup_release_agent_show,
2821 .write_string = cgroup_release_agent_write,
2822 .max_write_len = PATH_MAX,
2825 static int cgroup_populate_dir(struct cgroup *cgrp)
2828 struct cgroup_subsys *ss;
2830 /* First clear out any existing files */
2831 cgroup_clear_directory(cgrp->dentry);
2833 err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files));
2837 if (cgrp == cgrp->top_cgroup) {
2838 if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0)
2842 for_each_subsys(cgrp->root, ss) {
2843 if (ss->populate && (err = ss->populate(ss, cgrp)) < 0)
2846 /* This cgroup is ready now */
2847 for_each_subsys(cgrp->root, ss) {
2848 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
2850 * Update id->css pointer and make this css visible from
2851 * CSS ID functions. This pointer will be dereferened
2852 * from RCU-read-side without locks.
2855 rcu_assign_pointer(css->id->css, css);
2861 static void init_cgroup_css(struct cgroup_subsys_state *css,
2862 struct cgroup_subsys *ss,
2863 struct cgroup *cgrp)
2866 atomic_set(&css->refcnt, 1);
2869 if (cgrp == dummytop)
2870 set_bit(CSS_ROOT, &css->flags);
2871 BUG_ON(cgrp->subsys[ss->subsys_id]);
2872 cgrp->subsys[ss->subsys_id] = css;
2875 static void cgroup_lock_hierarchy(struct cgroupfs_root *root)
2877 /* We need to take each hierarchy_mutex in a consistent order */
2880 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2881 struct cgroup_subsys *ss = subsys[i];
2882 if (ss->root == root)
2883 mutex_lock(&ss->hierarchy_mutex);
2887 static void cgroup_unlock_hierarchy(struct cgroupfs_root *root)
2891 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2892 struct cgroup_subsys *ss = subsys[i];
2893 if (ss->root == root)
2894 mutex_unlock(&ss->hierarchy_mutex);
2899 * cgroup_create - create a cgroup
2900 * @parent: cgroup that will be parent of the new cgroup
2901 * @dentry: dentry of the new cgroup
2902 * @mode: mode to set on new inode
2904 * Must be called with the mutex on the parent inode held
2906 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
2909 struct cgroup *cgrp;
2910 struct cgroupfs_root *root = parent->root;
2912 struct cgroup_subsys *ss;
2913 struct super_block *sb = root->sb;
2915 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
2919 /* Grab a reference on the superblock so the hierarchy doesn't
2920 * get deleted on unmount if there are child cgroups. This
2921 * can be done outside cgroup_mutex, since the sb can't
2922 * disappear while someone has an open control file on the
2924 atomic_inc(&sb->s_active);
2926 mutex_lock(&cgroup_mutex);
2928 init_cgroup_housekeeping(cgrp);
2930 cgrp->parent = parent;
2931 cgrp->root = parent->root;
2932 cgrp->top_cgroup = parent->top_cgroup;
2934 if (notify_on_release(parent))
2935 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
2937 for_each_subsys(root, ss) {
2938 struct cgroup_subsys_state *css = ss->create(ss, cgrp);
2944 init_cgroup_css(css, ss, cgrp);
2946 err = alloc_css_id(ss, parent, cgrp);
2950 /* At error, ->destroy() callback has to free assigned ID. */
2953 cgroup_lock_hierarchy(root);
2954 list_add(&cgrp->sibling, &cgrp->parent->children);
2955 cgroup_unlock_hierarchy(root);
2956 root->number_of_cgroups++;
2958 err = cgroup_create_dir(cgrp, dentry, mode);
2962 /* The cgroup directory was pre-locked for us */
2963 BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
2965 err = cgroup_populate_dir(cgrp);
2966 /* If err < 0, we have a half-filled directory - oh well ;) */
2968 mutex_unlock(&cgroup_mutex);
2969 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
2975 cgroup_lock_hierarchy(root);
2976 list_del(&cgrp->sibling);
2977 cgroup_unlock_hierarchy(root);
2978 root->number_of_cgroups--;
2982 for_each_subsys(root, ss) {
2983 if (cgrp->subsys[ss->subsys_id])
2984 ss->destroy(ss, cgrp);
2987 mutex_unlock(&cgroup_mutex);
2989 /* Release the reference count that we took on the superblock */
2990 deactivate_super(sb);
2996 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2998 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
3000 /* the vfs holds inode->i_mutex already */
3001 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
3004 static int cgroup_has_css_refs(struct cgroup *cgrp)
3006 /* Check the reference count on each subsystem. Since we
3007 * already established that there are no tasks in the
3008 * cgroup, if the css refcount is also 1, then there should
3009 * be no outstanding references, so the subsystem is safe to
3010 * destroy. We scan across all subsystems rather than using
3011 * the per-hierarchy linked list of mounted subsystems since
3012 * we can be called via check_for_release() with no
3013 * synchronization other than RCU, and the subsystem linked
3014 * list isn't RCU-safe */
3016 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3017 struct cgroup_subsys *ss = subsys[i];
3018 struct cgroup_subsys_state *css;
3019 /* Skip subsystems not in this hierarchy */
3020 if (ss->root != cgrp->root)
3022 css = cgrp->subsys[ss->subsys_id];
3023 /* When called from check_for_release() it's possible
3024 * that by this point the cgroup has been removed
3025 * and the css deleted. But a false-positive doesn't
3026 * matter, since it can only happen if the cgroup
3027 * has been deleted and hence no longer needs the
3028 * release agent to be called anyway. */
3029 if (css && (atomic_read(&css->refcnt) > 1))
3036 * Atomically mark all (or else none) of the cgroup's CSS objects as
3037 * CSS_REMOVED. Return true on success, or false if the cgroup has
3038 * busy subsystems. Call with cgroup_mutex held
3041 static int cgroup_clear_css_refs(struct cgroup *cgrp)
3043 struct cgroup_subsys *ss;
3044 unsigned long flags;
3045 bool failed = false;
3046 local_irq_save(flags);
3047 for_each_subsys(cgrp->root, ss) {
3048 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3051 /* We can only remove a CSS with a refcnt==1 */
3052 refcnt = atomic_read(&css->refcnt);
3059 * Drop the refcnt to 0 while we check other
3060 * subsystems. This will cause any racing
3061 * css_tryget() to spin until we set the
3062 * CSS_REMOVED bits or abort
3064 if (atomic_cmpxchg(&css->refcnt, refcnt, 0) == refcnt)
3070 for_each_subsys(cgrp->root, ss) {
3071 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3074 * Restore old refcnt if we previously managed
3075 * to clear it from 1 to 0
3077 if (!atomic_read(&css->refcnt))
3078 atomic_set(&css->refcnt, 1);
3080 /* Commit the fact that the CSS is removed */
3081 set_bit(CSS_REMOVED, &css->flags);
3084 local_irq_restore(flags);
3088 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
3090 struct cgroup *cgrp = dentry->d_fsdata;
3092 struct cgroup *parent;
3096 /* the vfs holds both inode->i_mutex already */
3098 mutex_lock(&cgroup_mutex);
3099 if (atomic_read(&cgrp->count) != 0) {
3100 mutex_unlock(&cgroup_mutex);
3103 if (!list_empty(&cgrp->children)) {
3104 mutex_unlock(&cgroup_mutex);
3107 mutex_unlock(&cgroup_mutex);
3110 * In general, subsystem has no css->refcnt after pre_destroy(). But
3111 * in racy cases, subsystem may have to get css->refcnt after
3112 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
3113 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
3114 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
3115 * and subsystem's reference count handling. Please see css_get/put
3116 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
3118 set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3121 * Call pre_destroy handlers of subsys. Notify subsystems
3122 * that rmdir() request comes.
3124 ret = cgroup_call_pre_destroy(cgrp);
3126 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3130 mutex_lock(&cgroup_mutex);
3131 parent = cgrp->parent;
3132 if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) {
3133 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3134 mutex_unlock(&cgroup_mutex);
3137 prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE);
3138 if (!cgroup_clear_css_refs(cgrp)) {
3139 mutex_unlock(&cgroup_mutex);
3141 * Because someone may call cgroup_wakeup_rmdir_waiter() before
3142 * prepare_to_wait(), we need to check this flag.
3144 if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags))
3146 finish_wait(&cgroup_rmdir_waitq, &wait);
3147 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3148 if (signal_pending(current))
3152 /* NO css_tryget() can success after here. */
3153 finish_wait(&cgroup_rmdir_waitq, &wait);
3154 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3156 spin_lock(&release_list_lock);
3157 set_bit(CGRP_REMOVED, &cgrp->flags);
3158 if (!list_empty(&cgrp->release_list))
3159 list_del(&cgrp->release_list);
3160 spin_unlock(&release_list_lock);
3162 cgroup_lock_hierarchy(cgrp->root);
3163 /* delete this cgroup from parent->children */
3164 list_del(&cgrp->sibling);
3165 cgroup_unlock_hierarchy(cgrp->root);
3167 spin_lock(&cgrp->dentry->d_lock);
3168 d = dget(cgrp->dentry);
3169 spin_unlock(&d->d_lock);
3171 cgroup_d_remove_dir(d);
3174 set_bit(CGRP_RELEASABLE, &parent->flags);
3175 check_for_release(parent);
3177 mutex_unlock(&cgroup_mutex);
3181 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
3183 struct cgroup_subsys_state *css;
3185 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
3187 /* Create the top cgroup state for this subsystem */
3188 list_add(&ss->sibling, &rootnode.subsys_list);
3189 ss->root = &rootnode;
3190 css = ss->create(ss, dummytop);
3191 /* We don't handle early failures gracefully */
3192 BUG_ON(IS_ERR(css));
3193 init_cgroup_css(css, ss, dummytop);
3195 /* Update the init_css_set to contain a subsys
3196 * pointer to this state - since the subsystem is
3197 * newly registered, all tasks and hence the
3198 * init_css_set is in the subsystem's top cgroup. */
3199 init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
3201 need_forkexit_callback |= ss->fork || ss->exit;
3203 /* At system boot, before all subsystems have been
3204 * registered, no tasks have been forked, so we don't
3205 * need to invoke fork callbacks here. */
3206 BUG_ON(!list_empty(&init_task.tasks));
3208 mutex_init(&ss->hierarchy_mutex);
3209 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
3214 * cgroup_init_early - cgroup initialization at system boot
3216 * Initialize cgroups at system boot, and initialize any
3217 * subsystems that request early init.
3219 int __init cgroup_init_early(void)
3222 atomic_set(&init_css_set.refcount, 1);
3223 INIT_LIST_HEAD(&init_css_set.cg_links);
3224 INIT_LIST_HEAD(&init_css_set.tasks);
3225 INIT_HLIST_NODE(&init_css_set.hlist);
3227 init_cgroup_root(&rootnode);
3229 init_task.cgroups = &init_css_set;
3231 init_css_set_link.cg = &init_css_set;
3232 init_css_set_link.cgrp = dummytop;
3233 list_add(&init_css_set_link.cgrp_link_list,
3234 &rootnode.top_cgroup.css_sets);
3235 list_add(&init_css_set_link.cg_link_list,
3236 &init_css_set.cg_links);
3238 for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
3239 INIT_HLIST_HEAD(&css_set_table[i]);
3241 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3242 struct cgroup_subsys *ss = subsys[i];
3245 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
3246 BUG_ON(!ss->create);
3247 BUG_ON(!ss->destroy);
3248 if (ss->subsys_id != i) {
3249 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
3250 ss->name, ss->subsys_id);
3255 cgroup_init_subsys(ss);
3261 * cgroup_init - cgroup initialization
3263 * Register cgroup filesystem and /proc file, and initialize
3264 * any subsystems that didn't request early init.
3266 int __init cgroup_init(void)
3270 struct hlist_head *hhead;
3272 err = bdi_init(&cgroup_backing_dev_info);
3276 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3277 struct cgroup_subsys *ss = subsys[i];
3278 if (!ss->early_init)
3279 cgroup_init_subsys(ss);
3281 cgroup_subsys_init_idr(ss);
3284 /* Add init_css_set to the hash table */
3285 hhead = css_set_hash(init_css_set.subsys);
3286 hlist_add_head(&init_css_set.hlist, hhead);
3287 BUG_ON(!init_root_id(&rootnode));
3288 err = register_filesystem(&cgroup_fs_type);
3292 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
3296 bdi_destroy(&cgroup_backing_dev_info);
3302 * proc_cgroup_show()
3303 * - Print task's cgroup paths into seq_file, one line for each hierarchy
3304 * - Used for /proc/<pid>/cgroup.
3305 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
3306 * doesn't really matter if tsk->cgroup changes after we read it,
3307 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
3308 * anyway. No need to check that tsk->cgroup != NULL, thanks to
3309 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
3310 * cgroup to top_cgroup.
3313 /* TODO: Use a proper seq_file iterator */
3314 static int proc_cgroup_show(struct seq_file *m, void *v)
3317 struct task_struct *tsk;
3320 struct cgroupfs_root *root;
3323 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
3329 tsk = get_pid_task(pid, PIDTYPE_PID);
3335 mutex_lock(&cgroup_mutex);
3337 for_each_active_root(root) {
3338 struct cgroup_subsys *ss;
3339 struct cgroup *cgrp;
3342 seq_printf(m, "%d:", root->hierarchy_id);
3343 for_each_subsys(root, ss)
3344 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
3345 if (strlen(root->name))
3346 seq_printf(m, "%sname=%s", count ? "," : "",
3349 cgrp = task_cgroup_from_root(tsk, root);
3350 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
3358 mutex_unlock(&cgroup_mutex);
3359 put_task_struct(tsk);
3366 static int cgroup_open(struct inode *inode, struct file *file)
3368 struct pid *pid = PROC_I(inode)->pid;
3369 return single_open(file, proc_cgroup_show, pid);
3372 const struct file_operations proc_cgroup_operations = {
3373 .open = cgroup_open,
3375 .llseek = seq_lseek,
3376 .release = single_release,
3379 /* Display information about each subsystem and each hierarchy */
3380 static int proc_cgroupstats_show(struct seq_file *m, void *v)
3384 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
3385 mutex_lock(&cgroup_mutex);
3386 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3387 struct cgroup_subsys *ss = subsys[i];
3388 seq_printf(m, "%s\t%d\t%d\t%d\n",
3389 ss->name, ss->root->hierarchy_id,
3390 ss->root->number_of_cgroups, !ss->disabled);
3392 mutex_unlock(&cgroup_mutex);
3396 static int cgroupstats_open(struct inode *inode, struct file *file)
3398 return single_open(file, proc_cgroupstats_show, NULL);
3401 static const struct file_operations proc_cgroupstats_operations = {
3402 .open = cgroupstats_open,
3404 .llseek = seq_lseek,
3405 .release = single_release,
3409 * cgroup_fork - attach newly forked task to its parents cgroup.
3410 * @child: pointer to task_struct of forking parent process.
3412 * Description: A task inherits its parent's cgroup at fork().
3414 * A pointer to the shared css_set was automatically copied in
3415 * fork.c by dup_task_struct(). However, we ignore that copy, since
3416 * it was not made under the protection of RCU or cgroup_mutex, so
3417 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
3418 * have already changed current->cgroups, allowing the previously
3419 * referenced cgroup group to be removed and freed.
3421 * At the point that cgroup_fork() is called, 'current' is the parent
3422 * task, and the passed argument 'child' points to the child task.
3424 void cgroup_fork(struct task_struct *child)
3427 child->cgroups = current->cgroups;
3428 get_css_set(child->cgroups);
3429 task_unlock(current);
3430 INIT_LIST_HEAD(&child->cg_list);
3434 * cgroup_fork_callbacks - run fork callbacks
3435 * @child: the new task
3437 * Called on a new task very soon before adding it to the
3438 * tasklist. No need to take any locks since no-one can
3439 * be operating on this task.
3441 void cgroup_fork_callbacks(struct task_struct *child)
3443 if (need_forkexit_callback) {
3445 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3446 struct cgroup_subsys *ss = subsys[i];
3448 ss->fork(ss, child);
3454 * cgroup_post_fork - called on a new task after adding it to the task list
3455 * @child: the task in question
3457 * Adds the task to the list running through its css_set if necessary.
3458 * Has to be after the task is visible on the task list in case we race
3459 * with the first call to cgroup_iter_start() - to guarantee that the
3460 * new task ends up on its list.
3462 void cgroup_post_fork(struct task_struct *child)
3464 if (use_task_css_set_links) {
3465 write_lock(&css_set_lock);
3467 if (list_empty(&child->cg_list))
3468 list_add(&child->cg_list, &child->cgroups->tasks);
3470 write_unlock(&css_set_lock);
3474 * cgroup_exit - detach cgroup from exiting task
3475 * @tsk: pointer to task_struct of exiting process
3476 * @run_callback: run exit callbacks?
3478 * Description: Detach cgroup from @tsk and release it.
3480 * Note that cgroups marked notify_on_release force every task in
3481 * them to take the global cgroup_mutex mutex when exiting.
3482 * This could impact scaling on very large systems. Be reluctant to
3483 * use notify_on_release cgroups where very high task exit scaling
3484 * is required on large systems.
3486 * the_top_cgroup_hack:
3488 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
3490 * We call cgroup_exit() while the task is still competent to
3491 * handle notify_on_release(), then leave the task attached to the
3492 * root cgroup in each hierarchy for the remainder of its exit.
3494 * To do this properly, we would increment the reference count on
3495 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
3496 * code we would add a second cgroup function call, to drop that
3497 * reference. This would just create an unnecessary hot spot on
3498 * the top_cgroup reference count, to no avail.
3500 * Normally, holding a reference to a cgroup without bumping its
3501 * count is unsafe. The cgroup could go away, or someone could
3502 * attach us to a different cgroup, decrementing the count on
3503 * the first cgroup that we never incremented. But in this case,
3504 * top_cgroup isn't going away, and either task has PF_EXITING set,
3505 * which wards off any cgroup_attach_task() attempts, or task is a failed
3506 * fork, never visible to cgroup_attach_task.
3508 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
3513 if (run_callbacks && need_forkexit_callback) {
3514 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3515 struct cgroup_subsys *ss = subsys[i];
3522 * Unlink from the css_set task list if necessary.
3523 * Optimistically check cg_list before taking
3526 if (!list_empty(&tsk->cg_list)) {
3527 write_lock(&css_set_lock);
3528 if (!list_empty(&tsk->cg_list))
3529 list_del(&tsk->cg_list);
3530 write_unlock(&css_set_lock);
3533 /* Reassign the task to the init_css_set. */
3536 tsk->cgroups = &init_css_set;
3539 put_css_set_taskexit(cg);
3543 * cgroup_clone - clone the cgroup the given subsystem is attached to
3544 * @tsk: the task to be moved
3545 * @subsys: the given subsystem
3546 * @nodename: the name for the new cgroup
3548 * Duplicate the current cgroup in the hierarchy that the given
3549 * subsystem is attached to, and move this task into the new
3552 int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *subsys,
3555 struct dentry *dentry;
3557 struct cgroup *parent, *child;
3558 struct inode *inode;
3560 struct cgroupfs_root *root;
3561 struct cgroup_subsys *ss;
3563 /* We shouldn't be called by an unregistered subsystem */
3564 BUG_ON(!subsys->active);
3566 /* First figure out what hierarchy and cgroup we're dealing
3567 * with, and pin them so we can drop cgroup_mutex */
3568 mutex_lock(&cgroup_mutex);
3570 root = subsys->root;
3571 if (root == &rootnode) {
3572 mutex_unlock(&cgroup_mutex);
3576 /* Pin the hierarchy */
3577 if (!atomic_inc_not_zero(&root->sb->s_active)) {
3578 /* We race with the final deactivate_super() */
3579 mutex_unlock(&cgroup_mutex);
3583 /* Keep the cgroup alive */
3585 parent = task_cgroup(tsk, subsys->subsys_id);
3590 mutex_unlock(&cgroup_mutex);
3592 /* Now do the VFS work to create a cgroup */
3593 inode = parent->dentry->d_inode;
3595 /* Hold the parent directory mutex across this operation to
3596 * stop anyone else deleting the new cgroup */
3597 mutex_lock(&inode->i_mutex);
3598 dentry = lookup_one_len(nodename, parent->dentry, strlen(nodename));
3599 if (IS_ERR(dentry)) {
3601 "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename,
3603 ret = PTR_ERR(dentry);
3607 /* Create the cgroup directory, which also creates the cgroup */
3608 ret = vfs_mkdir(inode, dentry, 0755);
3609 child = __d_cgrp(dentry);
3613 "Failed to create cgroup %s: %d\n", nodename,
3618 /* The cgroup now exists. Retake cgroup_mutex and check
3619 * that we're still in the same state that we thought we
3621 mutex_lock(&cgroup_mutex);
3622 if ((root != subsys->root) ||
3623 (parent != task_cgroup(tsk, subsys->subsys_id))) {
3624 /* Aargh, we raced ... */
3625 mutex_unlock(&inode->i_mutex);
3628 deactivate_super(root->sb);
3629 /* The cgroup is still accessible in the VFS, but
3630 * we're not going to try to rmdir() it at this
3633 "Race in cgroup_clone() - leaking cgroup %s\n",
3638 /* do any required auto-setup */
3639 for_each_subsys(root, ss) {
3641 ss->post_clone(ss, child);
3644 /* All seems fine. Finish by moving the task into the new cgroup */
3645 ret = cgroup_attach_task(child, tsk);
3646 mutex_unlock(&cgroup_mutex);
3649 mutex_unlock(&inode->i_mutex);
3651 mutex_lock(&cgroup_mutex);
3653 mutex_unlock(&cgroup_mutex);
3654 deactivate_super(root->sb);
3659 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
3660 * @cgrp: the cgroup in question
3661 * @task: the task in question
3663 * See if @cgrp is a descendant of @task's cgroup in the appropriate
3666 * If we are sending in dummytop, then presumably we are creating
3667 * the top cgroup in the subsystem.
3669 * Called only by the ns (nsproxy) cgroup.
3671 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
3674 struct cgroup *target;
3676 if (cgrp == dummytop)
3679 target = task_cgroup_from_root(task, cgrp->root);
3680 while (cgrp != target && cgrp!= cgrp->top_cgroup)
3681 cgrp = cgrp->parent;
3682 ret = (cgrp == target);
3686 static void check_for_release(struct cgroup *cgrp)
3688 /* All of these checks rely on RCU to keep the cgroup
3689 * structure alive */
3690 if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
3691 && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
3692 /* Control Group is currently removeable. If it's not
3693 * already queued for a userspace notification, queue
3695 int need_schedule_work = 0;
3696 spin_lock(&release_list_lock);
3697 if (!cgroup_is_removed(cgrp) &&
3698 list_empty(&cgrp->release_list)) {
3699 list_add(&cgrp->release_list, &release_list);
3700 need_schedule_work = 1;
3702 spin_unlock(&release_list_lock);
3703 if (need_schedule_work)
3704 schedule_work(&release_agent_work);
3708 void __css_put(struct cgroup_subsys_state *css)
3710 struct cgroup *cgrp = css->cgroup;
3713 val = atomic_dec_return(&css->refcnt);
3715 if (notify_on_release(cgrp)) {
3716 set_bit(CGRP_RELEASABLE, &cgrp->flags);
3717 check_for_release(cgrp);
3719 cgroup_wakeup_rmdir_waiter(cgrp);
3722 WARN_ON_ONCE(val < 1);
3726 * Notify userspace when a cgroup is released, by running the
3727 * configured release agent with the name of the cgroup (path
3728 * relative to the root of cgroup file system) as the argument.
3730 * Most likely, this user command will try to rmdir this cgroup.
3732 * This races with the possibility that some other task will be
3733 * attached to this cgroup before it is removed, or that some other
3734 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
3735 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
3736 * unused, and this cgroup will be reprieved from its death sentence,
3737 * to continue to serve a useful existence. Next time it's released,
3738 * we will get notified again, if it still has 'notify_on_release' set.
3740 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
3741 * means only wait until the task is successfully execve()'d. The
3742 * separate release agent task is forked by call_usermodehelper(),
3743 * then control in this thread returns here, without waiting for the
3744 * release agent task. We don't bother to wait because the caller of
3745 * this routine has no use for the exit status of the release agent
3746 * task, so no sense holding our caller up for that.
3748 static void cgroup_release_agent(struct work_struct *work)
3750 BUG_ON(work != &release_agent_work);
3751 mutex_lock(&cgroup_mutex);
3752 spin_lock(&release_list_lock);
3753 while (!list_empty(&release_list)) {
3754 char *argv[3], *envp[3];
3756 char *pathbuf = NULL, *agentbuf = NULL;
3757 struct cgroup *cgrp = list_entry(release_list.next,
3760 list_del_init(&cgrp->release_list);
3761 spin_unlock(&release_list_lock);
3762 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
3765 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
3767 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
3772 argv[i++] = agentbuf;
3773 argv[i++] = pathbuf;
3777 /* minimal command environment */
3778 envp[i++] = "HOME=/";
3779 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
3782 /* Drop the lock while we invoke the usermode helper,
3783 * since the exec could involve hitting disk and hence
3784 * be a slow process */
3785 mutex_unlock(&cgroup_mutex);
3786 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
3787 mutex_lock(&cgroup_mutex);
3791 spin_lock(&release_list_lock);
3793 spin_unlock(&release_list_lock);
3794 mutex_unlock(&cgroup_mutex);
3797 static int __init cgroup_disable(char *str)
3802 while ((token = strsep(&str, ",")) != NULL) {
3806 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3807 struct cgroup_subsys *ss = subsys[i];
3809 if (!strcmp(token, ss->name)) {
3811 printk(KERN_INFO "Disabling %s control group"
3812 " subsystem\n", ss->name);
3819 __setup("cgroup_disable=", cgroup_disable);
3822 * Functons for CSS ID.
3826 *To get ID other than 0, this should be called when !cgroup_is_removed().
3828 unsigned short css_id(struct cgroup_subsys_state *css)
3830 struct css_id *cssid = rcu_dereference(css->id);
3837 unsigned short css_depth(struct cgroup_subsys_state *css)
3839 struct css_id *cssid = rcu_dereference(css->id);
3842 return cssid->depth;
3846 bool css_is_ancestor(struct cgroup_subsys_state *child,
3847 const struct cgroup_subsys_state *root)
3849 struct css_id *child_id = rcu_dereference(child->id);
3850 struct css_id *root_id = rcu_dereference(root->id);
3852 if (!child_id || !root_id || (child_id->depth < root_id->depth))
3854 return child_id->stack[root_id->depth] == root_id->id;
3857 static void __free_css_id_cb(struct rcu_head *head)
3861 id = container_of(head, struct css_id, rcu_head);
3865 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
3867 struct css_id *id = css->id;
3868 /* When this is called before css_id initialization, id can be NULL */
3872 BUG_ON(!ss->use_id);
3874 rcu_assign_pointer(id->css, NULL);
3875 rcu_assign_pointer(css->id, NULL);
3876 spin_lock(&ss->id_lock);
3877 idr_remove(&ss->idr, id->id);
3878 spin_unlock(&ss->id_lock);
3879 call_rcu(&id->rcu_head, __free_css_id_cb);
3883 * This is called by init or create(). Then, calls to this function are
3884 * always serialized (By cgroup_mutex() at create()).
3887 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
3889 struct css_id *newid;
3890 int myid, error, size;
3892 BUG_ON(!ss->use_id);
3894 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
3895 newid = kzalloc(size, GFP_KERNEL);
3897 return ERR_PTR(-ENOMEM);
3899 if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
3903 spin_lock(&ss->id_lock);
3904 /* Don't use 0. allocates an ID of 1-65535 */
3905 error = idr_get_new_above(&ss->idr, newid, 1, &myid);
3906 spin_unlock(&ss->id_lock);
3908 /* Returns error when there are no free spaces for new ID.*/
3913 if (myid > CSS_ID_MAX)
3917 newid->depth = depth;
3921 spin_lock(&ss->id_lock);
3922 idr_remove(&ss->idr, myid);
3923 spin_unlock(&ss->id_lock);
3926 return ERR_PTR(error);
3930 static int __init cgroup_subsys_init_idr(struct cgroup_subsys *ss)
3932 struct css_id *newid;
3933 struct cgroup_subsys_state *rootcss;
3935 spin_lock_init(&ss->id_lock);
3938 rootcss = init_css_set.subsys[ss->subsys_id];
3939 newid = get_new_cssid(ss, 0);
3941 return PTR_ERR(newid);
3943 newid->stack[0] = newid->id;
3944 newid->css = rootcss;
3945 rootcss->id = newid;
3949 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
3950 struct cgroup *child)
3952 int subsys_id, i, depth = 0;
3953 struct cgroup_subsys_state *parent_css, *child_css;
3954 struct css_id *child_id, *parent_id = NULL;
3956 subsys_id = ss->subsys_id;
3957 parent_css = parent->subsys[subsys_id];
3958 child_css = child->subsys[subsys_id];
3959 depth = css_depth(parent_css) + 1;
3960 parent_id = parent_css->id;
3962 child_id = get_new_cssid(ss, depth);
3963 if (IS_ERR(child_id))
3964 return PTR_ERR(child_id);
3966 for (i = 0; i < depth; i++)
3967 child_id->stack[i] = parent_id->stack[i];
3968 child_id->stack[depth] = child_id->id;
3970 * child_id->css pointer will be set after this cgroup is available
3971 * see cgroup_populate_dir()
3973 rcu_assign_pointer(child_css->id, child_id);
3979 * css_lookup - lookup css by id
3980 * @ss: cgroup subsys to be looked into.
3983 * Returns pointer to cgroup_subsys_state if there is valid one with id.
3984 * NULL if not. Should be called under rcu_read_lock()
3986 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
3988 struct css_id *cssid = NULL;
3990 BUG_ON(!ss->use_id);
3991 cssid = idr_find(&ss->idr, id);
3993 if (unlikely(!cssid))
3996 return rcu_dereference(cssid->css);
4000 * css_get_next - lookup next cgroup under specified hierarchy.
4001 * @ss: pointer to subsystem
4002 * @id: current position of iteration.
4003 * @root: pointer to css. search tree under this.
4004 * @foundid: position of found object.
4006 * Search next css under the specified hierarchy of rootid. Calling under
4007 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
4009 struct cgroup_subsys_state *
4010 css_get_next(struct cgroup_subsys *ss, int id,
4011 struct cgroup_subsys_state *root, int *foundid)
4013 struct cgroup_subsys_state *ret = NULL;
4016 int rootid = css_id(root);
4017 int depth = css_depth(root);
4022 BUG_ON(!ss->use_id);
4023 /* fill start point for scan */
4027 * scan next entry from bitmap(tree), tmpid is updated after
4030 spin_lock(&ss->id_lock);
4031 tmp = idr_get_next(&ss->idr, &tmpid);
4032 spin_unlock(&ss->id_lock);
4036 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
4037 ret = rcu_dereference(tmp->css);
4043 /* continue to scan from next id */
4049 #ifdef CONFIG_CGROUP_DEBUG
4050 static struct cgroup_subsys_state *debug_create(struct cgroup_subsys *ss,
4051 struct cgroup *cont)
4053 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
4056 return ERR_PTR(-ENOMEM);
4061 static void debug_destroy(struct cgroup_subsys *ss, struct cgroup *cont)
4063 kfree(cont->subsys[debug_subsys_id]);
4066 static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
4068 return atomic_read(&cont->count);
4071 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
4073 return cgroup_task_count(cont);
4076 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
4078 return (u64)(unsigned long)current->cgroups;
4081 static u64 current_css_set_refcount_read(struct cgroup *cont,
4087 count = atomic_read(¤t->cgroups->refcount);
4092 static int current_css_set_cg_links_read(struct cgroup *cont,
4094 struct seq_file *seq)
4096 struct cg_cgroup_link *link;
4099 read_lock(&css_set_lock);
4101 cg = rcu_dereference(current->cgroups);
4102 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
4103 struct cgroup *c = link->cgrp;
4107 name = c->dentry->d_name.name;
4110 seq_printf(seq, "Root %d group %s\n",
4111 c->root->hierarchy_id, name);
4114 read_unlock(&css_set_lock);
4118 #define MAX_TASKS_SHOWN_PER_CSS 25
4119 static int cgroup_css_links_read(struct cgroup *cont,
4121 struct seq_file *seq)
4123 struct cg_cgroup_link *link;
4125 read_lock(&css_set_lock);
4126 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
4127 struct css_set *cg = link->cg;
4128 struct task_struct *task;
4130 seq_printf(seq, "css_set %p\n", cg);
4131 list_for_each_entry(task, &cg->tasks, cg_list) {
4132 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
4133 seq_puts(seq, " ...\n");
4136 seq_printf(seq, " task %d\n",
4137 task_pid_vnr(task));
4141 read_unlock(&css_set_lock);
4145 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
4147 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
4150 static struct cftype debug_files[] = {
4152 .name = "cgroup_refcount",
4153 .read_u64 = cgroup_refcount_read,
4156 .name = "taskcount",
4157 .read_u64 = debug_taskcount_read,
4161 .name = "current_css_set",
4162 .read_u64 = current_css_set_read,
4166 .name = "current_css_set_refcount",
4167 .read_u64 = current_css_set_refcount_read,
4171 .name = "current_css_set_cg_links",
4172 .read_seq_string = current_css_set_cg_links_read,
4176 .name = "cgroup_css_links",
4177 .read_seq_string = cgroup_css_links_read,
4181 .name = "releasable",
4182 .read_u64 = releasable_read,
4186 static int debug_populate(struct cgroup_subsys *ss, struct cgroup *cont)
4188 return cgroup_add_files(cont, ss, debug_files,
4189 ARRAY_SIZE(debug_files));
4192 struct cgroup_subsys debug_subsys = {
4194 .create = debug_create,
4195 .destroy = debug_destroy,
4196 .populate = debug_populate,
4197 .subsys_id = debug_subsys_id,
4199 #endif /* CONFIG_CGROUP_DEBUG */