2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
11 * Copyright notices from the original cpuset code:
12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
19 * 2003-10-10 Written by Simon Derr.
20 * 2003-10-22 Updates by Stephen Hemminger.
21 * 2004 May-July Rework by Paul Jackson.
22 * ---------------------------------------------------
24 * This file is subject to the terms and conditions of the GNU General Public
25 * License. See the file COPYING in the main directory of the Linux
26 * distribution for more details.
29 #include <linux/cgroup.h>
30 #include <linux/cred.h>
31 #include <linux/ctype.h>
32 #include <linux/errno.h>
34 #include <linux/init_task.h>
35 #include <linux/kernel.h>
36 #include <linux/list.h>
38 #include <linux/mutex.h>
39 #include <linux/mount.h>
40 #include <linux/pagemap.h>
41 #include <linux/proc_fs.h>
42 #include <linux/rcupdate.h>
43 #include <linux/sched.h>
44 #include <linux/backing-dev.h>
45 #include <linux/seq_file.h>
46 #include <linux/slab.h>
47 #include <linux/magic.h>
48 #include <linux/spinlock.h>
49 #include <linux/string.h>
50 #include <linux/sort.h>
51 #include <linux/kmod.h>
52 #include <linux/module.h>
53 #include <linux/delayacct.h>
54 #include <linux/cgroupstats.h>
55 #include <linux/hashtable.h>
56 #include <linux/namei.h>
57 #include <linux/pid_namespace.h>
58 #include <linux/idr.h>
59 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
60 #include <linux/eventfd.h>
61 #include <linux/poll.h>
62 #include <linux/flex_array.h> /* used in cgroup_attach_task */
63 #include <linux/kthread.h>
65 #include <linux/atomic.h>
67 /* css deactivation bias, makes css->refcnt negative to deny new trygets */
68 #define CSS_DEACT_BIAS INT_MIN
71 * cgroup_mutex is the master lock. Any modification to cgroup or its
72 * hierarchy must be performed while holding it.
74 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
75 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
76 * release_agent_path and so on. Modifying requires both cgroup_mutex and
77 * cgroup_root_mutex. Readers can acquire either of the two. This is to
78 * break the following locking order cycle.
80 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
81 * B. namespace_sem -> cgroup_mutex
83 * B happens only through cgroup_show_options() and using cgroup_root_mutex
86 #ifdef CONFIG_PROVE_RCU
87 DEFINE_MUTEX(cgroup_mutex);
88 EXPORT_SYMBOL_GPL(cgroup_mutex); /* only for task_subsys_state_check() */
90 static DEFINE_MUTEX(cgroup_mutex);
93 static DEFINE_MUTEX(cgroup_root_mutex);
96 * Generate an array of cgroup subsystem pointers. At boot time, this is
97 * populated with the built in subsystems, and modular subsystems are
98 * registered after that. The mutable section of this array is protected by
101 #define SUBSYS(_x) [_x ## _subsys_id] = &_x ## _subsys,
102 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
103 static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
104 #include <linux/cgroup_subsys.h>
107 #define MAX_CGROUP_ROOT_NAMELEN 64
110 * A cgroupfs_root represents the root of a cgroup hierarchy,
111 * and may be associated with a superblock to form an active
114 struct cgroupfs_root {
115 struct super_block *sb;
118 * The bitmask of subsystems intended to be attached to this
121 unsigned long subsys_mask;
123 /* Unique id for this hierarchy. */
126 /* The bitmask of subsystems currently attached to this hierarchy */
127 unsigned long actual_subsys_mask;
129 /* A list running through the attached subsystems */
130 struct list_head subsys_list;
132 /* The root cgroup for this hierarchy */
133 struct cgroup top_cgroup;
135 /* Tracks how many cgroups are currently defined in hierarchy.*/
136 int number_of_cgroups;
138 /* A list running through the active hierarchies */
139 struct list_head root_list;
141 /* All cgroups on this root, cgroup_mutex protected */
142 struct list_head allcg_list;
144 /* Hierarchy-specific flags */
147 /* IDs for cgroups in this hierarchy */
148 struct ida cgroup_ida;
150 /* The path to use for release notifications. */
151 char release_agent_path[PATH_MAX];
153 /* The name for this hierarchy - may be empty */
154 char name[MAX_CGROUP_ROOT_NAMELEN];
158 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
159 * subsystems that are otherwise unattached - it never has more than a
160 * single cgroup, and all tasks are part of that cgroup.
162 static struct cgroupfs_root rootnode;
165 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
168 struct list_head node;
169 struct dentry *dentry;
174 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
175 * cgroup_subsys->use_id != 0.
177 #define CSS_ID_MAX (65535)
180 * The css to which this ID points. This pointer is set to valid value
181 * after cgroup is populated. If cgroup is removed, this will be NULL.
182 * This pointer is expected to be RCU-safe because destroy()
183 * is called after synchronize_rcu(). But for safe use, css_tryget()
184 * should be used for avoiding race.
186 struct cgroup_subsys_state __rcu *css;
192 * Depth in hierarchy which this ID belongs to.
194 unsigned short depth;
196 * ID is freed by RCU. (and lookup routine is RCU safe.)
198 struct rcu_head rcu_head;
200 * Hierarchy of CSS ID belongs to.
202 unsigned short stack[0]; /* Array of Length (depth+1) */
206 * cgroup_event represents events which userspace want to receive.
208 struct cgroup_event {
210 * Cgroup which the event belongs to.
214 * Control file which the event associated.
218 * eventfd to signal userspace about the event.
220 struct eventfd_ctx *eventfd;
222 * Each of these stored in a list by the cgroup.
224 struct list_head list;
226 * All fields below needed to unregister event when
227 * userspace closes eventfd.
230 wait_queue_head_t *wqh;
232 struct work_struct remove;
235 /* The list of hierarchy roots */
237 static LIST_HEAD(roots);
238 static int root_count;
240 static DEFINE_IDA(hierarchy_ida);
241 static int next_hierarchy_id;
242 static DEFINE_SPINLOCK(hierarchy_id_lock);
244 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
245 #define dummytop (&rootnode.top_cgroup)
247 static struct cgroup_name root_cgroup_name = { .name = "/" };
249 /* This flag indicates whether tasks in the fork and exit paths should
250 * check for fork/exit handlers to call. This avoids us having to do
251 * extra work in the fork/exit path if none of the subsystems need to
254 static int need_forkexit_callback __read_mostly;
256 static int cgroup_destroy_locked(struct cgroup *cgrp);
257 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
258 struct cftype cfts[], bool is_add);
260 static int css_unbias_refcnt(int refcnt)
262 return refcnt >= 0 ? refcnt : refcnt - CSS_DEACT_BIAS;
265 /* the current nr of refs, always >= 0 whether @css is deactivated or not */
266 static int css_refcnt(struct cgroup_subsys_state *css)
268 int v = atomic_read(&css->refcnt);
270 return css_unbias_refcnt(v);
273 /* convenient tests for these bits */
274 inline int cgroup_is_removed(const struct cgroup *cgrp)
276 return test_bit(CGRP_REMOVED, &cgrp->flags);
280 * cgroup_is_descendant - test ancestry
281 * @cgrp: the cgroup to be tested
282 * @ancestor: possible ancestor of @cgrp
284 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
285 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
286 * and @ancestor are accessible.
288 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor)
291 if (cgrp == ancestor)
297 EXPORT_SYMBOL_GPL(cgroup_is_descendant);
299 /* cgroupfs_root->flags */
301 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
302 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
305 static int cgroup_is_releasable(const struct cgroup *cgrp)
308 (1 << CGRP_RELEASABLE) |
309 (1 << CGRP_NOTIFY_ON_RELEASE);
310 return (cgrp->flags & bits) == bits;
313 static int notify_on_release(const struct cgroup *cgrp)
315 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
319 * for_each_subsys() allows you to iterate on each subsystem attached to
320 * an active hierarchy
322 #define for_each_subsys(_root, _ss) \
323 list_for_each_entry(_ss, &_root->subsys_list, sibling)
325 /* for_each_active_root() allows you to iterate across the active hierarchies */
326 #define for_each_active_root(_root) \
327 list_for_each_entry(_root, &roots, root_list)
329 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
331 return dentry->d_fsdata;
334 static inline struct cfent *__d_cfe(struct dentry *dentry)
336 return dentry->d_fsdata;
339 static inline struct cftype *__d_cft(struct dentry *dentry)
341 return __d_cfe(dentry)->type;
345 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
346 * @cgrp: the cgroup to be checked for liveness
348 * On success, returns true; the mutex should be later unlocked. On
349 * failure returns false with no lock held.
351 static bool cgroup_lock_live_group(struct cgroup *cgrp)
353 mutex_lock(&cgroup_mutex);
354 if (cgroup_is_removed(cgrp)) {
355 mutex_unlock(&cgroup_mutex);
361 /* the list of cgroups eligible for automatic release. Protected by
362 * release_list_lock */
363 static LIST_HEAD(release_list);
364 static DEFINE_RAW_SPINLOCK(release_list_lock);
365 static void cgroup_release_agent(struct work_struct *work);
366 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
367 static void check_for_release(struct cgroup *cgrp);
369 /* Link structure for associating css_set objects with cgroups */
370 struct cg_cgroup_link {
372 * List running through cg_cgroup_links associated with a
373 * cgroup, anchored on cgroup->css_sets
375 struct list_head cgrp_link_list;
378 * List running through cg_cgroup_links pointing at a
379 * single css_set object, anchored on css_set->cg_links
381 struct list_head cg_link_list;
385 /* The default css_set - used by init and its children prior to any
386 * hierarchies being mounted. It contains a pointer to the root state
387 * for each subsystem. Also used to anchor the list of css_sets. Not
388 * reference-counted, to improve performance when child cgroups
389 * haven't been created.
392 static struct css_set init_css_set;
393 static struct cg_cgroup_link init_css_set_link;
395 static int cgroup_init_idr(struct cgroup_subsys *ss,
396 struct cgroup_subsys_state *css);
398 /* css_set_lock protects the list of css_set objects, and the
399 * chain of tasks off each css_set. Nests outside task->alloc_lock
400 * due to cgroup_iter_start() */
401 static DEFINE_RWLOCK(css_set_lock);
402 static int css_set_count;
405 * hash table for cgroup groups. This improves the performance to find
406 * an existing css_set. This hash doesn't (currently) take into
407 * account cgroups in empty hierarchies.
409 #define CSS_SET_HASH_BITS 7
410 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
412 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
415 unsigned long key = 0UL;
417 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
418 key += (unsigned long)css[i];
419 key = (key >> 16) ^ key;
424 /* We don't maintain the lists running through each css_set to its
425 * task until after the first call to cgroup_iter_start(). This
426 * reduces the fork()/exit() overhead for people who have cgroups
427 * compiled into their kernel but not actually in use */
428 static int use_task_css_set_links __read_mostly;
430 static void __put_css_set(struct css_set *cg, int taskexit)
432 struct cg_cgroup_link *link;
433 struct cg_cgroup_link *saved_link;
435 * Ensure that the refcount doesn't hit zero while any readers
436 * can see it. Similar to atomic_dec_and_lock(), but for an
439 if (atomic_add_unless(&cg->refcount, -1, 1))
441 write_lock(&css_set_lock);
442 if (!atomic_dec_and_test(&cg->refcount)) {
443 write_unlock(&css_set_lock);
447 /* This css_set is dead. unlink it and release cgroup refcounts */
448 hash_del(&cg->hlist);
451 list_for_each_entry_safe(link, saved_link, &cg->cg_links,
453 struct cgroup *cgrp = link->cgrp;
454 list_del(&link->cg_link_list);
455 list_del(&link->cgrp_link_list);
458 * We may not be holding cgroup_mutex, and if cgrp->count is
459 * dropped to 0 the cgroup can be destroyed at any time, hence
460 * rcu_read_lock is used to keep it alive.
463 if (atomic_dec_and_test(&cgrp->count) &&
464 notify_on_release(cgrp)) {
466 set_bit(CGRP_RELEASABLE, &cgrp->flags);
467 check_for_release(cgrp);
474 write_unlock(&css_set_lock);
475 kfree_rcu(cg, rcu_head);
479 * refcounted get/put for css_set objects
481 static inline void get_css_set(struct css_set *cg)
483 atomic_inc(&cg->refcount);
486 static inline void put_css_set(struct css_set *cg)
488 __put_css_set(cg, 0);
491 static inline void put_css_set_taskexit(struct css_set *cg)
493 __put_css_set(cg, 1);
497 * compare_css_sets - helper function for find_existing_css_set().
498 * @cg: candidate css_set being tested
499 * @old_cg: existing css_set for a task
500 * @new_cgrp: cgroup that's being entered by the task
501 * @template: desired set of css pointers in css_set (pre-calculated)
503 * Returns true if "cg" matches "old_cg" except for the hierarchy
504 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
506 static bool compare_css_sets(struct css_set *cg,
507 struct css_set *old_cg,
508 struct cgroup *new_cgrp,
509 struct cgroup_subsys_state *template[])
511 struct list_head *l1, *l2;
513 if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
514 /* Not all subsystems matched */
519 * Compare cgroup pointers in order to distinguish between
520 * different cgroups in heirarchies with no subsystems. We
521 * could get by with just this check alone (and skip the
522 * memcmp above) but on most setups the memcmp check will
523 * avoid the need for this more expensive check on almost all
528 l2 = &old_cg->cg_links;
530 struct cg_cgroup_link *cgl1, *cgl2;
531 struct cgroup *cg1, *cg2;
535 /* See if we reached the end - both lists are equal length. */
536 if (l1 == &cg->cg_links) {
537 BUG_ON(l2 != &old_cg->cg_links);
540 BUG_ON(l2 == &old_cg->cg_links);
542 /* Locate the cgroups associated with these links. */
543 cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
544 cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
547 /* Hierarchies should be linked in the same order. */
548 BUG_ON(cg1->root != cg2->root);
551 * If this hierarchy is the hierarchy of the cgroup
552 * that's changing, then we need to check that this
553 * css_set points to the new cgroup; if it's any other
554 * hierarchy, then this css_set should point to the
555 * same cgroup as the old css_set.
557 if (cg1->root == new_cgrp->root) {
569 * find_existing_css_set() is a helper for
570 * find_css_set(), and checks to see whether an existing
571 * css_set is suitable.
573 * oldcg: the cgroup group that we're using before the cgroup
576 * cgrp: the cgroup that we're moving into
578 * template: location in which to build the desired set of subsystem
579 * state objects for the new cgroup group
581 static struct css_set *find_existing_css_set(
582 struct css_set *oldcg,
584 struct cgroup_subsys_state *template[])
587 struct cgroupfs_root *root = cgrp->root;
592 * Build the set of subsystem state objects that we want to see in the
593 * new css_set. while subsystems can change globally, the entries here
594 * won't change, so no need for locking.
596 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
597 if (root->subsys_mask & (1UL << i)) {
598 /* Subsystem is in this hierarchy. So we want
599 * the subsystem state from the new
601 template[i] = cgrp->subsys[i];
603 /* Subsystem is not in this hierarchy, so we
604 * don't want to change the subsystem state */
605 template[i] = oldcg->subsys[i];
609 key = css_set_hash(template);
610 hash_for_each_possible(css_set_table, cg, hlist, key) {
611 if (!compare_css_sets(cg, oldcg, cgrp, template))
614 /* This css_set matches what we need */
618 /* No existing cgroup group matched */
622 static void free_cg_links(struct list_head *tmp)
624 struct cg_cgroup_link *link;
625 struct cg_cgroup_link *saved_link;
627 list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
628 list_del(&link->cgrp_link_list);
634 * allocate_cg_links() allocates "count" cg_cgroup_link structures
635 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
636 * success or a negative error
638 static int allocate_cg_links(int count, struct list_head *tmp)
640 struct cg_cgroup_link *link;
643 for (i = 0; i < count; i++) {
644 link = kmalloc(sizeof(*link), GFP_KERNEL);
649 list_add(&link->cgrp_link_list, tmp);
655 * link_css_set - a helper function to link a css_set to a cgroup
656 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
657 * @cg: the css_set to be linked
658 * @cgrp: the destination cgroup
660 static void link_css_set(struct list_head *tmp_cg_links,
661 struct css_set *cg, struct cgroup *cgrp)
663 struct cg_cgroup_link *link;
665 BUG_ON(list_empty(tmp_cg_links));
666 link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
670 atomic_inc(&cgrp->count);
671 list_move(&link->cgrp_link_list, &cgrp->css_sets);
673 * Always add links to the tail of the list so that the list
674 * is sorted by order of hierarchy creation
676 list_add_tail(&link->cg_link_list, &cg->cg_links);
680 * find_css_set() takes an existing cgroup group and a
681 * cgroup object, and returns a css_set object that's
682 * equivalent to the old group, but with the given cgroup
683 * substituted into the appropriate hierarchy. Must be called with
686 static struct css_set *find_css_set(
687 struct css_set *oldcg, struct cgroup *cgrp)
690 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
692 struct list_head tmp_cg_links;
694 struct cg_cgroup_link *link;
697 /* First see if we already have a cgroup group that matches
699 read_lock(&css_set_lock);
700 res = find_existing_css_set(oldcg, cgrp, template);
703 read_unlock(&css_set_lock);
708 res = kmalloc(sizeof(*res), GFP_KERNEL);
712 /* Allocate all the cg_cgroup_link objects that we'll need */
713 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
718 atomic_set(&res->refcount, 1);
719 INIT_LIST_HEAD(&res->cg_links);
720 INIT_LIST_HEAD(&res->tasks);
721 INIT_HLIST_NODE(&res->hlist);
723 /* Copy the set of subsystem state objects generated in
724 * find_existing_css_set() */
725 memcpy(res->subsys, template, sizeof(res->subsys));
727 write_lock(&css_set_lock);
728 /* Add reference counts and links from the new css_set. */
729 list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
730 struct cgroup *c = link->cgrp;
731 if (c->root == cgrp->root)
733 link_css_set(&tmp_cg_links, res, c);
736 BUG_ON(!list_empty(&tmp_cg_links));
740 /* Add this cgroup group to the hash table */
741 key = css_set_hash(res->subsys);
742 hash_add(css_set_table, &res->hlist, key);
744 write_unlock(&css_set_lock);
750 * Return the cgroup for "task" from the given hierarchy. Must be
751 * called with cgroup_mutex held.
753 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
754 struct cgroupfs_root *root)
757 struct cgroup *res = NULL;
759 BUG_ON(!mutex_is_locked(&cgroup_mutex));
760 read_lock(&css_set_lock);
762 * No need to lock the task - since we hold cgroup_mutex the
763 * task can't change groups, so the only thing that can happen
764 * is that it exits and its css is set back to init_css_set.
767 if (css == &init_css_set) {
768 res = &root->top_cgroup;
770 struct cg_cgroup_link *link;
771 list_for_each_entry(link, &css->cg_links, cg_link_list) {
772 struct cgroup *c = link->cgrp;
773 if (c->root == root) {
779 read_unlock(&css_set_lock);
785 * There is one global cgroup mutex. We also require taking
786 * task_lock() when dereferencing a task's cgroup subsys pointers.
787 * See "The task_lock() exception", at the end of this comment.
789 * A task must hold cgroup_mutex to modify cgroups.
791 * Any task can increment and decrement the count field without lock.
792 * So in general, code holding cgroup_mutex can't rely on the count
793 * field not changing. However, if the count goes to zero, then only
794 * cgroup_attach_task() can increment it again. Because a count of zero
795 * means that no tasks are currently attached, therefore there is no
796 * way a task attached to that cgroup can fork (the other way to
797 * increment the count). So code holding cgroup_mutex can safely
798 * assume that if the count is zero, it will stay zero. Similarly, if
799 * a task holds cgroup_mutex on a cgroup with zero count, it
800 * knows that the cgroup won't be removed, as cgroup_rmdir()
803 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
804 * (usually) take cgroup_mutex. These are the two most performance
805 * critical pieces of code here. The exception occurs on cgroup_exit(),
806 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
807 * is taken, and if the cgroup count is zero, a usermode call made
808 * to the release agent with the name of the cgroup (path relative to
809 * the root of cgroup file system) as the argument.
811 * A cgroup can only be deleted if both its 'count' of using tasks
812 * is zero, and its list of 'children' cgroups is empty. Since all
813 * tasks in the system use _some_ cgroup, and since there is always at
814 * least one task in the system (init, pid == 1), therefore, top_cgroup
815 * always has either children cgroups and/or using tasks. So we don't
816 * need a special hack to ensure that top_cgroup cannot be deleted.
818 * The task_lock() exception
820 * The need for this exception arises from the action of
821 * cgroup_attach_task(), which overwrites one task's cgroup pointer with
822 * another. It does so using cgroup_mutex, however there are
823 * several performance critical places that need to reference
824 * task->cgroup without the expense of grabbing a system global
825 * mutex. Therefore except as noted below, when dereferencing or, as
826 * in cgroup_attach_task(), modifying a task's cgroup pointer we use
827 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
828 * the task_struct routinely used for such matters.
830 * P.S. One more locking exception. RCU is used to guard the
831 * update of a tasks cgroup pointer by cgroup_attach_task()
835 * A couple of forward declarations required, due to cyclic reference loop:
836 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
837 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
841 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
842 static struct dentry *cgroup_lookup(struct inode *, struct dentry *, unsigned int);
843 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
844 static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
845 unsigned long subsys_mask);
846 static const struct inode_operations cgroup_dir_inode_operations;
847 static const struct file_operations proc_cgroupstats_operations;
849 static struct backing_dev_info cgroup_backing_dev_info = {
851 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
854 static int alloc_css_id(struct cgroup_subsys *ss,
855 struct cgroup *parent, struct cgroup *child);
857 static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb)
859 struct inode *inode = new_inode(sb);
862 inode->i_ino = get_next_ino();
863 inode->i_mode = mode;
864 inode->i_uid = current_fsuid();
865 inode->i_gid = current_fsgid();
866 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
867 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
872 static struct cgroup_name *cgroup_alloc_name(struct dentry *dentry)
874 struct cgroup_name *name;
876 name = kmalloc(sizeof(*name) + dentry->d_name.len + 1, GFP_KERNEL);
879 strcpy(name->name, dentry->d_name.name);
883 static void cgroup_free_fn(struct work_struct *work)
885 struct cgroup *cgrp = container_of(work, struct cgroup, free_work);
886 struct cgroup_subsys *ss;
888 mutex_lock(&cgroup_mutex);
890 * Release the subsystem state objects.
892 for_each_subsys(cgrp->root, ss)
895 cgrp->root->number_of_cgroups--;
896 mutex_unlock(&cgroup_mutex);
899 * We get a ref to the parent's dentry, and put the ref when
900 * this cgroup is being freed, so it's guaranteed that the
901 * parent won't be destroyed before its children.
903 dput(cgrp->parent->dentry);
906 * Drop the active superblock reference that we took when we
909 deactivate_super(cgrp->root->sb);
912 * if we're getting rid of the cgroup, refcount should ensure
913 * that there are no pidlists left.
915 BUG_ON(!list_empty(&cgrp->pidlists));
917 simple_xattrs_free(&cgrp->xattrs);
919 ida_simple_remove(&cgrp->root->cgroup_ida, cgrp->id);
920 kfree(rcu_dereference_raw(cgrp->name));
924 static void cgroup_free_rcu(struct rcu_head *head)
926 struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head);
928 schedule_work(&cgrp->free_work);
931 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
933 /* is dentry a directory ? if so, kfree() associated cgroup */
934 if (S_ISDIR(inode->i_mode)) {
935 struct cgroup *cgrp = dentry->d_fsdata;
937 BUG_ON(!(cgroup_is_removed(cgrp)));
938 call_rcu(&cgrp->rcu_head, cgroup_free_rcu);
940 struct cfent *cfe = __d_cfe(dentry);
941 struct cgroup *cgrp = dentry->d_parent->d_fsdata;
942 struct cftype *cft = cfe->type;
944 WARN_ONCE(!list_empty(&cfe->node) &&
945 cgrp != &cgrp->root->top_cgroup,
946 "cfe still linked for %s\n", cfe->type->name);
948 simple_xattrs_free(&cft->xattrs);
953 static int cgroup_delete(const struct dentry *d)
958 static void remove_dir(struct dentry *d)
960 struct dentry *parent = dget(d->d_parent);
963 simple_rmdir(parent->d_inode, d);
967 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
971 lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex);
972 lockdep_assert_held(&cgroup_mutex);
975 * If we're doing cleanup due to failure of cgroup_create(),
976 * the corresponding @cfe may not exist.
978 list_for_each_entry(cfe, &cgrp->files, node) {
979 struct dentry *d = cfe->dentry;
981 if (cft && cfe->type != cft)
986 simple_unlink(cgrp->dentry->d_inode, d);
987 list_del_init(&cfe->node);
995 * cgroup_clear_directory - selective removal of base and subsystem files
996 * @dir: directory containing the files
997 * @base_files: true if the base files should be removed
998 * @subsys_mask: mask of the subsystem ids whose files should be removed
1000 static void cgroup_clear_directory(struct dentry *dir, bool base_files,
1001 unsigned long subsys_mask)
1003 struct cgroup *cgrp = __d_cgrp(dir);
1004 struct cgroup_subsys *ss;
1006 for_each_subsys(cgrp->root, ss) {
1007 struct cftype_set *set;
1008 if (!test_bit(ss->subsys_id, &subsys_mask))
1010 list_for_each_entry(set, &ss->cftsets, node)
1011 cgroup_addrm_files(cgrp, NULL, set->cfts, false);
1014 while (!list_empty(&cgrp->files))
1015 cgroup_rm_file(cgrp, NULL);
1020 * NOTE : the dentry must have been dget()'ed
1022 static void cgroup_d_remove_dir(struct dentry *dentry)
1024 struct dentry *parent;
1025 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1027 cgroup_clear_directory(dentry, true, root->subsys_mask);
1029 parent = dentry->d_parent;
1030 spin_lock(&parent->d_lock);
1031 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1032 list_del_init(&dentry->d_u.d_child);
1033 spin_unlock(&dentry->d_lock);
1034 spin_unlock(&parent->d_lock);
1039 * Call with cgroup_mutex held. Drops reference counts on modules, including
1040 * any duplicate ones that parse_cgroupfs_options took. If this function
1041 * returns an error, no reference counts are touched.
1043 static int rebind_subsystems(struct cgroupfs_root *root,
1044 unsigned long final_subsys_mask)
1046 unsigned long added_mask, removed_mask;
1047 struct cgroup *cgrp = &root->top_cgroup;
1050 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1051 BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
1053 removed_mask = root->actual_subsys_mask & ~final_subsys_mask;
1054 added_mask = final_subsys_mask & ~root->actual_subsys_mask;
1055 /* Check that any added subsystems are currently free */
1056 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1057 unsigned long bit = 1UL << i;
1058 struct cgroup_subsys *ss = subsys[i];
1059 if (!(bit & added_mask))
1062 * Nobody should tell us to do a subsys that doesn't exist:
1063 * parse_cgroupfs_options should catch that case and refcounts
1064 * ensure that subsystems won't disappear once selected.
1067 if (ss->root != &rootnode) {
1068 /* Subsystem isn't free */
1073 /* Currently we don't handle adding/removing subsystems when
1074 * any child cgroups exist. This is theoretically supportable
1075 * but involves complex error handling, so it's being left until
1077 if (root->number_of_cgroups > 1)
1080 /* Process each subsystem */
1081 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1082 struct cgroup_subsys *ss = subsys[i];
1083 unsigned long bit = 1UL << i;
1084 if (bit & added_mask) {
1085 /* We're binding this subsystem to this hierarchy */
1087 BUG_ON(cgrp->subsys[i]);
1088 BUG_ON(!dummytop->subsys[i]);
1089 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
1090 cgrp->subsys[i] = dummytop->subsys[i];
1091 cgrp->subsys[i]->cgroup = cgrp;
1092 list_move(&ss->sibling, &root->subsys_list);
1096 /* refcount was already taken, and we're keeping it */
1097 } else if (bit & removed_mask) {
1098 /* We're removing this subsystem */
1100 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1101 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1104 dummytop->subsys[i]->cgroup = dummytop;
1105 cgrp->subsys[i] = NULL;
1106 subsys[i]->root = &rootnode;
1107 list_move(&ss->sibling, &rootnode.subsys_list);
1108 /* subsystem is now free - drop reference on module */
1109 module_put(ss->module);
1110 } else if (bit & final_subsys_mask) {
1111 /* Subsystem state should already exist */
1113 BUG_ON(!cgrp->subsys[i]);
1115 * a refcount was taken, but we already had one, so
1116 * drop the extra reference.
1118 module_put(ss->module);
1119 #ifdef CONFIG_MODULE_UNLOAD
1120 BUG_ON(ss->module && !module_refcount(ss->module));
1123 /* Subsystem state shouldn't exist */
1124 BUG_ON(cgrp->subsys[i]);
1127 root->subsys_mask = root->actual_subsys_mask = final_subsys_mask;
1132 static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1134 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1135 struct cgroup_subsys *ss;
1137 mutex_lock(&cgroup_root_mutex);
1138 for_each_subsys(root, ss)
1139 seq_printf(seq, ",%s", ss->name);
1140 if (root->flags & CGRP_ROOT_NOPREFIX)
1141 seq_puts(seq, ",noprefix");
1142 if (root->flags & CGRP_ROOT_XATTR)
1143 seq_puts(seq, ",xattr");
1144 if (strlen(root->release_agent_path))
1145 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1146 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags))
1147 seq_puts(seq, ",clone_children");
1148 if (strlen(root->name))
1149 seq_printf(seq, ",name=%s", root->name);
1150 mutex_unlock(&cgroup_root_mutex);
1154 struct cgroup_sb_opts {
1155 unsigned long subsys_mask;
1156 unsigned long flags;
1157 char *release_agent;
1158 bool cpuset_clone_children;
1160 /* User explicitly requested empty subsystem */
1163 struct cgroupfs_root *new_root;
1168 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1169 * with cgroup_mutex held to protect the subsys[] array. This function takes
1170 * refcounts on subsystems to be used, unless it returns error, in which case
1171 * no refcounts are taken.
1173 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1175 char *token, *o = data;
1176 bool all_ss = false, one_ss = false;
1177 unsigned long mask = (unsigned long)-1;
1179 bool module_pin_failed = false;
1181 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1183 #ifdef CONFIG_CPUSETS
1184 mask = ~(1UL << cpuset_subsys_id);
1187 memset(opts, 0, sizeof(*opts));
1189 while ((token = strsep(&o, ",")) != NULL) {
1192 if (!strcmp(token, "none")) {
1193 /* Explicitly have no subsystems */
1197 if (!strcmp(token, "all")) {
1198 /* Mutually exclusive option 'all' + subsystem name */
1204 if (!strcmp(token, "noprefix")) {
1205 opts->flags |= CGRP_ROOT_NOPREFIX;
1208 if (!strcmp(token, "clone_children")) {
1209 opts->cpuset_clone_children = true;
1212 if (!strcmp(token, "xattr")) {
1213 opts->flags |= CGRP_ROOT_XATTR;
1216 if (!strncmp(token, "release_agent=", 14)) {
1217 /* Specifying two release agents is forbidden */
1218 if (opts->release_agent)
1220 opts->release_agent =
1221 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1222 if (!opts->release_agent)
1226 if (!strncmp(token, "name=", 5)) {
1227 const char *name = token + 5;
1228 /* Can't specify an empty name */
1231 /* Must match [\w.-]+ */
1232 for (i = 0; i < strlen(name); i++) {
1236 if ((c == '.') || (c == '-') || (c == '_'))
1240 /* Specifying two names is forbidden */
1243 opts->name = kstrndup(name,
1244 MAX_CGROUP_ROOT_NAMELEN - 1,
1252 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1253 struct cgroup_subsys *ss = subsys[i];
1256 if (strcmp(token, ss->name))
1261 /* Mutually exclusive option 'all' + subsystem name */
1264 set_bit(i, &opts->subsys_mask);
1269 if (i == CGROUP_SUBSYS_COUNT)
1274 * If the 'all' option was specified select all the subsystems,
1275 * otherwise if 'none', 'name=' and a subsystem name options
1276 * were not specified, let's default to 'all'
1278 if (all_ss || (!one_ss && !opts->none && !opts->name)) {
1279 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1280 struct cgroup_subsys *ss = subsys[i];
1285 set_bit(i, &opts->subsys_mask);
1289 /* Consistency checks */
1292 * Option noprefix was introduced just for backward compatibility
1293 * with the old cpuset, so we allow noprefix only if mounting just
1294 * the cpuset subsystem.
1296 if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask))
1300 /* Can't specify "none" and some subsystems */
1301 if (opts->subsys_mask && opts->none)
1305 * We either have to specify by name or by subsystems. (So all
1306 * empty hierarchies must have a name).
1308 if (!opts->subsys_mask && !opts->name)
1312 * Grab references on all the modules we'll need, so the subsystems
1313 * don't dance around before rebind_subsystems attaches them. This may
1314 * take duplicate reference counts on a subsystem that's already used,
1315 * but rebind_subsystems handles this case.
1317 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1318 unsigned long bit = 1UL << i;
1320 if (!(bit & opts->subsys_mask))
1322 if (!try_module_get(subsys[i]->module)) {
1323 module_pin_failed = true;
1327 if (module_pin_failed) {
1329 * oops, one of the modules was going away. this means that we
1330 * raced with a module_delete call, and to the user this is
1331 * essentially a "subsystem doesn't exist" case.
1333 for (i--; i >= 0; i--) {
1334 /* drop refcounts only on the ones we took */
1335 unsigned long bit = 1UL << i;
1337 if (!(bit & opts->subsys_mask))
1339 module_put(subsys[i]->module);
1347 static void drop_parsed_module_refcounts(unsigned long subsys_mask)
1350 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1351 unsigned long bit = 1UL << i;
1353 if (!(bit & subsys_mask))
1355 module_put(subsys[i]->module);
1359 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1362 struct cgroupfs_root *root = sb->s_fs_info;
1363 struct cgroup *cgrp = &root->top_cgroup;
1364 struct cgroup_sb_opts opts;
1365 unsigned long added_mask, removed_mask;
1367 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1368 mutex_lock(&cgroup_mutex);
1369 mutex_lock(&cgroup_root_mutex);
1371 /* See what subsystems are wanted */
1372 ret = parse_cgroupfs_options(data, &opts);
1376 if (opts.subsys_mask != root->actual_subsys_mask || opts.release_agent)
1377 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1378 task_tgid_nr(current), current->comm);
1380 added_mask = opts.subsys_mask & ~root->subsys_mask;
1381 removed_mask = root->subsys_mask & ~opts.subsys_mask;
1383 /* Don't allow flags or name to change at remount */
1384 if (opts.flags != root->flags ||
1385 (opts.name && strcmp(opts.name, root->name))) {
1387 drop_parsed_module_refcounts(opts.subsys_mask);
1392 * Clear out the files of subsystems that should be removed, do
1393 * this before rebind_subsystems, since rebind_subsystems may
1394 * change this hierarchy's subsys_list.
1396 cgroup_clear_directory(cgrp->dentry, false, removed_mask);
1398 ret = rebind_subsystems(root, opts.subsys_mask);
1400 /* rebind_subsystems failed, re-populate the removed files */
1401 cgroup_populate_dir(cgrp, false, removed_mask);
1402 drop_parsed_module_refcounts(opts.subsys_mask);
1406 /* re-populate subsystem files */
1407 cgroup_populate_dir(cgrp, false, added_mask);
1409 if (opts.release_agent)
1410 strcpy(root->release_agent_path, opts.release_agent);
1412 kfree(opts.release_agent);
1414 mutex_unlock(&cgroup_root_mutex);
1415 mutex_unlock(&cgroup_mutex);
1416 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1420 static const struct super_operations cgroup_ops = {
1421 .statfs = simple_statfs,
1422 .drop_inode = generic_delete_inode,
1423 .show_options = cgroup_show_options,
1424 .remount_fs = cgroup_remount,
1427 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1429 INIT_LIST_HEAD(&cgrp->sibling);
1430 INIT_LIST_HEAD(&cgrp->children);
1431 INIT_LIST_HEAD(&cgrp->files);
1432 INIT_LIST_HEAD(&cgrp->css_sets);
1433 INIT_LIST_HEAD(&cgrp->allcg_node);
1434 INIT_LIST_HEAD(&cgrp->release_list);
1435 INIT_LIST_HEAD(&cgrp->pidlists);
1436 INIT_WORK(&cgrp->free_work, cgroup_free_fn);
1437 mutex_init(&cgrp->pidlist_mutex);
1438 INIT_LIST_HEAD(&cgrp->event_list);
1439 spin_lock_init(&cgrp->event_list_lock);
1440 simple_xattrs_init(&cgrp->xattrs);
1443 static void init_cgroup_root(struct cgroupfs_root *root)
1445 struct cgroup *cgrp = &root->top_cgroup;
1447 INIT_LIST_HEAD(&root->subsys_list);
1448 INIT_LIST_HEAD(&root->root_list);
1449 INIT_LIST_HEAD(&root->allcg_list);
1450 root->number_of_cgroups = 1;
1452 cgrp->name = &root_cgroup_name;
1453 cgrp->top_cgroup = cgrp;
1454 init_cgroup_housekeeping(cgrp);
1455 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
1458 static bool init_root_id(struct cgroupfs_root *root)
1463 if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1465 spin_lock(&hierarchy_id_lock);
1466 /* Try to allocate the next unused ID */
1467 ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1468 &root->hierarchy_id);
1470 /* Try again starting from 0 */
1471 ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1473 next_hierarchy_id = root->hierarchy_id + 1;
1474 } else if (ret != -EAGAIN) {
1475 /* Can only get here if the 31-bit IDR is full ... */
1478 spin_unlock(&hierarchy_id_lock);
1483 static int cgroup_test_super(struct super_block *sb, void *data)
1485 struct cgroup_sb_opts *opts = data;
1486 struct cgroupfs_root *root = sb->s_fs_info;
1488 /* If we asked for a name then it must match */
1489 if (opts->name && strcmp(opts->name, root->name))
1493 * If we asked for subsystems (or explicitly for no
1494 * subsystems) then they must match
1496 if ((opts->subsys_mask || opts->none)
1497 && (opts->subsys_mask != root->subsys_mask))
1503 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1505 struct cgroupfs_root *root;
1507 if (!opts->subsys_mask && !opts->none)
1510 root = kzalloc(sizeof(*root), GFP_KERNEL);
1512 return ERR_PTR(-ENOMEM);
1514 if (!init_root_id(root)) {
1516 return ERR_PTR(-ENOMEM);
1518 init_cgroup_root(root);
1520 root->subsys_mask = opts->subsys_mask;
1521 root->flags = opts->flags;
1522 ida_init(&root->cgroup_ida);
1523 if (opts->release_agent)
1524 strcpy(root->release_agent_path, opts->release_agent);
1526 strcpy(root->name, opts->name);
1527 if (opts->cpuset_clone_children)
1528 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags);
1532 static void cgroup_drop_root(struct cgroupfs_root *root)
1537 BUG_ON(!root->hierarchy_id);
1538 spin_lock(&hierarchy_id_lock);
1539 ida_remove(&hierarchy_ida, root->hierarchy_id);
1540 spin_unlock(&hierarchy_id_lock);
1541 ida_destroy(&root->cgroup_ida);
1545 static int cgroup_set_super(struct super_block *sb, void *data)
1548 struct cgroup_sb_opts *opts = data;
1550 /* If we don't have a new root, we can't set up a new sb */
1551 if (!opts->new_root)
1554 BUG_ON(!opts->subsys_mask && !opts->none);
1556 ret = set_anon_super(sb, NULL);
1560 sb->s_fs_info = opts->new_root;
1561 opts->new_root->sb = sb;
1563 sb->s_blocksize = PAGE_CACHE_SIZE;
1564 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1565 sb->s_magic = CGROUP_SUPER_MAGIC;
1566 sb->s_op = &cgroup_ops;
1571 static int cgroup_get_rootdir(struct super_block *sb)
1573 static const struct dentry_operations cgroup_dops = {
1574 .d_iput = cgroup_diput,
1575 .d_delete = cgroup_delete,
1578 struct inode *inode =
1579 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1584 inode->i_fop = &simple_dir_operations;
1585 inode->i_op = &cgroup_dir_inode_operations;
1586 /* directories start off with i_nlink == 2 (for "." entry) */
1588 sb->s_root = d_make_root(inode);
1591 /* for everything else we want ->d_op set */
1592 sb->s_d_op = &cgroup_dops;
1596 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1597 int flags, const char *unused_dev_name,
1600 struct cgroup_sb_opts opts;
1601 struct cgroupfs_root *root;
1603 struct super_block *sb;
1604 struct cgroupfs_root *new_root;
1605 struct inode *inode;
1607 /* First find the desired set of subsystems */
1608 mutex_lock(&cgroup_mutex);
1609 ret = parse_cgroupfs_options(data, &opts);
1610 mutex_unlock(&cgroup_mutex);
1615 * Allocate a new cgroup root. We may not need it if we're
1616 * reusing an existing hierarchy.
1618 new_root = cgroup_root_from_opts(&opts);
1619 if (IS_ERR(new_root)) {
1620 ret = PTR_ERR(new_root);
1623 opts.new_root = new_root;
1625 /* Locate an existing or new sb for this hierarchy */
1626 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, 0, &opts);
1629 cgroup_drop_root(opts.new_root);
1633 root = sb->s_fs_info;
1635 if (root == opts.new_root) {
1636 /* We used the new root structure, so this is a new hierarchy */
1637 struct list_head tmp_cg_links;
1638 struct cgroup *root_cgrp = &root->top_cgroup;
1639 struct cgroupfs_root *existing_root;
1640 const struct cred *cred;
1644 BUG_ON(sb->s_root != NULL);
1646 ret = cgroup_get_rootdir(sb);
1648 goto drop_new_super;
1649 inode = sb->s_root->d_inode;
1651 mutex_lock(&inode->i_mutex);
1652 mutex_lock(&cgroup_mutex);
1653 mutex_lock(&cgroup_root_mutex);
1655 /* Check for name clashes with existing mounts */
1657 if (strlen(root->name))
1658 for_each_active_root(existing_root)
1659 if (!strcmp(existing_root->name, root->name))
1663 * We're accessing css_set_count without locking
1664 * css_set_lock here, but that's OK - it can only be
1665 * increased by someone holding cgroup_lock, and
1666 * that's us. The worst that can happen is that we
1667 * have some link structures left over
1669 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1673 ret = rebind_subsystems(root, root->subsys_mask);
1674 if (ret == -EBUSY) {
1675 free_cg_links(&tmp_cg_links);
1679 * There must be no failure case after here, since rebinding
1680 * takes care of subsystems' refcounts, which are explicitly
1681 * dropped in the failure exit path.
1684 /* EBUSY should be the only error here */
1687 list_add(&root->root_list, &roots);
1690 sb->s_root->d_fsdata = root_cgrp;
1691 root->top_cgroup.dentry = sb->s_root;
1693 /* Link the top cgroup in this hierarchy into all
1694 * the css_set objects */
1695 write_lock(&css_set_lock);
1696 hash_for_each(css_set_table, i, cg, hlist)
1697 link_css_set(&tmp_cg_links, cg, root_cgrp);
1698 write_unlock(&css_set_lock);
1700 free_cg_links(&tmp_cg_links);
1702 BUG_ON(!list_empty(&root_cgrp->children));
1703 BUG_ON(root->number_of_cgroups != 1);
1705 cred = override_creds(&init_cred);
1706 cgroup_populate_dir(root_cgrp, true, root->subsys_mask);
1708 mutex_unlock(&cgroup_root_mutex);
1709 mutex_unlock(&cgroup_mutex);
1710 mutex_unlock(&inode->i_mutex);
1713 * We re-used an existing hierarchy - the new root (if
1714 * any) is not needed
1716 cgroup_drop_root(opts.new_root);
1717 /* no subsys rebinding, so refcounts don't change */
1718 drop_parsed_module_refcounts(opts.subsys_mask);
1721 kfree(opts.release_agent);
1723 return dget(sb->s_root);
1726 mutex_unlock(&cgroup_root_mutex);
1727 mutex_unlock(&cgroup_mutex);
1728 mutex_unlock(&inode->i_mutex);
1730 deactivate_locked_super(sb);
1732 drop_parsed_module_refcounts(opts.subsys_mask);
1734 kfree(opts.release_agent);
1736 return ERR_PTR(ret);
1739 static void cgroup_kill_sb(struct super_block *sb) {
1740 struct cgroupfs_root *root = sb->s_fs_info;
1741 struct cgroup *cgrp = &root->top_cgroup;
1743 struct cg_cgroup_link *link;
1744 struct cg_cgroup_link *saved_link;
1748 BUG_ON(root->number_of_cgroups != 1);
1749 BUG_ON(!list_empty(&cgrp->children));
1751 mutex_lock(&cgroup_mutex);
1752 mutex_lock(&cgroup_root_mutex);
1754 /* Rebind all subsystems back to the default hierarchy */
1755 ret = rebind_subsystems(root, 0);
1756 /* Shouldn't be able to fail ... */
1760 * Release all the links from css_sets to this hierarchy's
1763 write_lock(&css_set_lock);
1765 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1767 list_del(&link->cg_link_list);
1768 list_del(&link->cgrp_link_list);
1771 write_unlock(&css_set_lock);
1773 if (!list_empty(&root->root_list)) {
1774 list_del(&root->root_list);
1778 mutex_unlock(&cgroup_root_mutex);
1779 mutex_unlock(&cgroup_mutex);
1781 simple_xattrs_free(&cgrp->xattrs);
1783 kill_litter_super(sb);
1784 cgroup_drop_root(root);
1787 static struct file_system_type cgroup_fs_type = {
1789 .mount = cgroup_mount,
1790 .kill_sb = cgroup_kill_sb,
1793 static struct kobject *cgroup_kobj;
1796 * cgroup_path - generate the path of a cgroup
1797 * @cgrp: the cgroup in question
1798 * @buf: the buffer to write the path into
1799 * @buflen: the length of the buffer
1801 * Writes path of cgroup into buf. Returns 0 on success, -errno on error.
1803 * We can't generate cgroup path using dentry->d_name, as accessing
1804 * dentry->name must be protected by irq-unsafe dentry->d_lock or parent
1805 * inode's i_mutex, while on the other hand cgroup_path() can be called
1806 * with some irq-safe spinlocks held.
1808 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1810 int ret = -ENAMETOOLONG;
1813 if (!cgrp->parent) {
1814 if (strlcpy(buf, "/", buflen) >= buflen)
1815 return -ENAMETOOLONG;
1819 start = buf + buflen - 1;
1824 const char *name = cgroup_name(cgrp);
1828 if ((start -= len) < buf)
1830 memcpy(start, name, len);
1836 cgrp = cgrp->parent;
1837 } while (cgrp->parent);
1839 memmove(buf, start, buf + buflen - start);
1844 EXPORT_SYMBOL_GPL(cgroup_path);
1847 * Control Group taskset
1849 struct task_and_cgroup {
1850 struct task_struct *task;
1851 struct cgroup *cgrp;
1855 struct cgroup_taskset {
1856 struct task_and_cgroup single;
1857 struct flex_array *tc_array;
1860 struct cgroup *cur_cgrp;
1864 * cgroup_taskset_first - reset taskset and return the first task
1865 * @tset: taskset of interest
1867 * @tset iteration is initialized and the first task is returned.
1869 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1871 if (tset->tc_array) {
1873 return cgroup_taskset_next(tset);
1875 tset->cur_cgrp = tset->single.cgrp;
1876 return tset->single.task;
1879 EXPORT_SYMBOL_GPL(cgroup_taskset_first);
1882 * cgroup_taskset_next - iterate to the next task in taskset
1883 * @tset: taskset of interest
1885 * Return the next task in @tset. Iteration must have been initialized
1886 * with cgroup_taskset_first().
1888 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1890 struct task_and_cgroup *tc;
1892 if (!tset->tc_array || tset->idx >= tset->tc_array_len)
1895 tc = flex_array_get(tset->tc_array, tset->idx++);
1896 tset->cur_cgrp = tc->cgrp;
1899 EXPORT_SYMBOL_GPL(cgroup_taskset_next);
1902 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1903 * @tset: taskset of interest
1905 * Return the cgroup for the current (last returned) task of @tset. This
1906 * function must be preceded by either cgroup_taskset_first() or
1907 * cgroup_taskset_next().
1909 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset)
1911 return tset->cur_cgrp;
1913 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup);
1916 * cgroup_taskset_size - return the number of tasks in taskset
1917 * @tset: taskset of interest
1919 int cgroup_taskset_size(struct cgroup_taskset *tset)
1921 return tset->tc_array ? tset->tc_array_len : 1;
1923 EXPORT_SYMBOL_GPL(cgroup_taskset_size);
1927 * cgroup_task_migrate - move a task from one cgroup to another.
1929 * Must be called with cgroup_mutex and threadgroup locked.
1931 static void cgroup_task_migrate(struct cgroup *oldcgrp,
1932 struct task_struct *tsk, struct css_set *newcg)
1934 struct css_set *oldcg;
1937 * We are synchronized through threadgroup_lock() against PF_EXITING
1938 * setting such that we can't race against cgroup_exit() changing the
1939 * css_set to init_css_set and dropping the old one.
1941 WARN_ON_ONCE(tsk->flags & PF_EXITING);
1942 oldcg = tsk->cgroups;
1945 rcu_assign_pointer(tsk->cgroups, newcg);
1948 /* Update the css_set linked lists if we're using them */
1949 write_lock(&css_set_lock);
1950 if (!list_empty(&tsk->cg_list))
1951 list_move(&tsk->cg_list, &newcg->tasks);
1952 write_unlock(&css_set_lock);
1955 * We just gained a reference on oldcg by taking it from the task. As
1956 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1957 * it here; it will be freed under RCU.
1959 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1964 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
1965 * @cgrp: the cgroup to attach to
1966 * @tsk: the task or the leader of the threadgroup to be attached
1967 * @threadgroup: attach the whole threadgroup?
1969 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
1970 * task_lock of @tsk or each thread in the threadgroup individually in turn.
1972 static int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk,
1975 int retval, i, group_size;
1976 struct cgroup_subsys *ss, *failed_ss = NULL;
1977 struct cgroupfs_root *root = cgrp->root;
1978 /* threadgroup list cursor and array */
1979 struct task_struct *leader = tsk;
1980 struct task_and_cgroup *tc;
1981 struct flex_array *group;
1982 struct cgroup_taskset tset = { };
1985 * step 0: in order to do expensive, possibly blocking operations for
1986 * every thread, we cannot iterate the thread group list, since it needs
1987 * rcu or tasklist locked. instead, build an array of all threads in the
1988 * group - group_rwsem prevents new threads from appearing, and if
1989 * threads exit, this will just be an over-estimate.
1992 group_size = get_nr_threads(tsk);
1995 /* flex_array supports very large thread-groups better than kmalloc. */
1996 group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
1999 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2000 retval = flex_array_prealloc(group, 0, group_size, GFP_KERNEL);
2002 goto out_free_group_list;
2006 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2007 * already PF_EXITING could be freed from underneath us unless we
2008 * take an rcu_read_lock.
2012 struct task_and_cgroup ent;
2014 /* @tsk either already exited or can't exit until the end */
2015 if (tsk->flags & PF_EXITING)
2018 /* as per above, nr_threads may decrease, but not increase. */
2019 BUG_ON(i >= group_size);
2021 ent.cgrp = task_cgroup_from_root(tsk, root);
2022 /* nothing to do if this task is already in the cgroup */
2023 if (ent.cgrp == cgrp)
2026 * saying GFP_ATOMIC has no effect here because we did prealloc
2027 * earlier, but it's good form to communicate our expectations.
2029 retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
2030 BUG_ON(retval != 0);
2035 } while_each_thread(leader, tsk);
2037 /* remember the number of threads in the array for later. */
2039 tset.tc_array = group;
2040 tset.tc_array_len = group_size;
2042 /* methods shouldn't be called if no task is actually migrating */
2045 goto out_free_group_list;
2048 * step 1: check that we can legitimately attach to the cgroup.
2050 for_each_subsys(root, ss) {
2051 if (ss->can_attach) {
2052 retval = ss->can_attach(cgrp, &tset);
2055 goto out_cancel_attach;
2061 * step 2: make sure css_sets exist for all threads to be migrated.
2062 * we use find_css_set, which allocates a new one if necessary.
2064 for (i = 0; i < group_size; i++) {
2065 tc = flex_array_get(group, i);
2066 tc->cg = find_css_set(tc->task->cgroups, cgrp);
2069 goto out_put_css_set_refs;
2074 * step 3: now that we're guaranteed success wrt the css_sets,
2075 * proceed to move all tasks to the new cgroup. There are no
2076 * failure cases after here, so this is the commit point.
2078 for (i = 0; i < group_size; i++) {
2079 tc = flex_array_get(group, i);
2080 cgroup_task_migrate(tc->cgrp, tc->task, tc->cg);
2082 /* nothing is sensitive to fork() after this point. */
2085 * step 4: do subsystem attach callbacks.
2087 for_each_subsys(root, ss) {
2089 ss->attach(cgrp, &tset);
2093 * step 5: success! and cleanup
2096 out_put_css_set_refs:
2098 for (i = 0; i < group_size; i++) {
2099 tc = flex_array_get(group, i);
2102 put_css_set(tc->cg);
2107 for_each_subsys(root, ss) {
2108 if (ss == failed_ss)
2110 if (ss->cancel_attach)
2111 ss->cancel_attach(cgrp, &tset);
2114 out_free_group_list:
2115 flex_array_free(group);
2120 * Find the task_struct of the task to attach by vpid and pass it along to the
2121 * function to attach either it or all tasks in its threadgroup. Will lock
2122 * cgroup_mutex and threadgroup; may take task_lock of task.
2124 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2126 struct task_struct *tsk;
2127 const struct cred *cred = current_cred(), *tcred;
2130 if (!cgroup_lock_live_group(cgrp))
2136 tsk = find_task_by_vpid(pid);
2140 goto out_unlock_cgroup;
2143 * even if we're attaching all tasks in the thread group, we
2144 * only need to check permissions on one of them.
2146 tcred = __task_cred(tsk);
2147 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2148 !uid_eq(cred->euid, tcred->uid) &&
2149 !uid_eq(cred->euid, tcred->suid)) {
2152 goto out_unlock_cgroup;
2158 tsk = tsk->group_leader;
2161 * Workqueue threads may acquire PF_THREAD_BOUND and become
2162 * trapped in a cpuset, or RT worker may be born in a cgroup
2163 * with no rt_runtime allocated. Just say no.
2165 if (tsk == kthreadd_task || (tsk->flags & PF_THREAD_BOUND)) {
2168 goto out_unlock_cgroup;
2171 get_task_struct(tsk);
2174 threadgroup_lock(tsk);
2176 if (!thread_group_leader(tsk)) {
2178 * a race with de_thread from another thread's exec()
2179 * may strip us of our leadership, if this happens,
2180 * there is no choice but to throw this task away and
2181 * try again; this is
2182 * "double-double-toil-and-trouble-check locking".
2184 threadgroup_unlock(tsk);
2185 put_task_struct(tsk);
2186 goto retry_find_task;
2190 ret = cgroup_attach_task(cgrp, tsk, threadgroup);
2192 threadgroup_unlock(tsk);
2194 put_task_struct(tsk);
2196 mutex_unlock(&cgroup_mutex);
2201 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2202 * @from: attach to all cgroups of a given task
2203 * @tsk: the task to be attached
2205 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2207 struct cgroupfs_root *root;
2210 mutex_lock(&cgroup_mutex);
2211 for_each_active_root(root) {
2212 struct cgroup *from_cg = task_cgroup_from_root(from, root);
2214 retval = cgroup_attach_task(from_cg, tsk, false);
2218 mutex_unlock(&cgroup_mutex);
2222 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2224 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2226 return attach_task_by_pid(cgrp, pid, false);
2229 static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2231 return attach_task_by_pid(cgrp, tgid, true);
2234 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2237 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2238 if (strlen(buffer) >= PATH_MAX)
2240 if (!cgroup_lock_live_group(cgrp))
2242 mutex_lock(&cgroup_root_mutex);
2243 strcpy(cgrp->root->release_agent_path, buffer);
2244 mutex_unlock(&cgroup_root_mutex);
2245 mutex_unlock(&cgroup_mutex);
2249 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2250 struct seq_file *seq)
2252 if (!cgroup_lock_live_group(cgrp))
2254 seq_puts(seq, cgrp->root->release_agent_path);
2255 seq_putc(seq, '\n');
2256 mutex_unlock(&cgroup_mutex);
2260 /* A buffer size big enough for numbers or short strings */
2261 #define CGROUP_LOCAL_BUFFER_SIZE 64
2263 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2265 const char __user *userbuf,
2266 size_t nbytes, loff_t *unused_ppos)
2268 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2274 if (nbytes >= sizeof(buffer))
2276 if (copy_from_user(buffer, userbuf, nbytes))
2279 buffer[nbytes] = 0; /* nul-terminate */
2280 if (cft->write_u64) {
2281 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2284 retval = cft->write_u64(cgrp, cft, val);
2286 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2289 retval = cft->write_s64(cgrp, cft, val);
2296 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2298 const char __user *userbuf,
2299 size_t nbytes, loff_t *unused_ppos)
2301 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2303 size_t max_bytes = cft->max_write_len;
2304 char *buffer = local_buffer;
2307 max_bytes = sizeof(local_buffer) - 1;
2308 if (nbytes >= max_bytes)
2310 /* Allocate a dynamic buffer if we need one */
2311 if (nbytes >= sizeof(local_buffer)) {
2312 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2316 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2321 buffer[nbytes] = 0; /* nul-terminate */
2322 retval = cft->write_string(cgrp, cft, strstrip(buffer));
2326 if (buffer != local_buffer)
2331 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2332 size_t nbytes, loff_t *ppos)
2334 struct cftype *cft = __d_cft(file->f_dentry);
2335 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2337 if (cgroup_is_removed(cgrp))
2340 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2341 if (cft->write_u64 || cft->write_s64)
2342 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2343 if (cft->write_string)
2344 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2346 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2347 return ret ? ret : nbytes;
2352 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2354 char __user *buf, size_t nbytes,
2357 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2358 u64 val = cft->read_u64(cgrp, cft);
2359 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2361 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2364 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2366 char __user *buf, size_t nbytes,
2369 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2370 s64 val = cft->read_s64(cgrp, cft);
2371 int len = sprintf(tmp, "%lld\n", (long long) val);
2373 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2376 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2377 size_t nbytes, loff_t *ppos)
2379 struct cftype *cft = __d_cft(file->f_dentry);
2380 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2382 if (cgroup_is_removed(cgrp))
2386 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2388 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2390 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2395 * seqfile ops/methods for returning structured data. Currently just
2396 * supports string->u64 maps, but can be extended in future.
2399 struct cgroup_seqfile_state {
2401 struct cgroup *cgroup;
2404 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2406 struct seq_file *sf = cb->state;
2407 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2410 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2412 struct cgroup_seqfile_state *state = m->private;
2413 struct cftype *cft = state->cft;
2414 if (cft->read_map) {
2415 struct cgroup_map_cb cb = {
2416 .fill = cgroup_map_add,
2419 return cft->read_map(state->cgroup, cft, &cb);
2421 return cft->read_seq_string(state->cgroup, cft, m);
2424 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2426 struct seq_file *seq = file->private_data;
2427 kfree(seq->private);
2428 return single_release(inode, file);
2431 static const struct file_operations cgroup_seqfile_operations = {
2433 .write = cgroup_file_write,
2434 .llseek = seq_lseek,
2435 .release = cgroup_seqfile_release,
2438 static int cgroup_file_open(struct inode *inode, struct file *file)
2443 err = generic_file_open(inode, file);
2446 cft = __d_cft(file->f_dentry);
2448 if (cft->read_map || cft->read_seq_string) {
2449 struct cgroup_seqfile_state *state =
2450 kzalloc(sizeof(*state), GFP_USER);
2454 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2455 file->f_op = &cgroup_seqfile_operations;
2456 err = single_open(file, cgroup_seqfile_show, state);
2459 } else if (cft->open)
2460 err = cft->open(inode, file);
2467 static int cgroup_file_release(struct inode *inode, struct file *file)
2469 struct cftype *cft = __d_cft(file->f_dentry);
2471 return cft->release(inode, file);
2476 * cgroup_rename - Only allow simple rename of directories in place.
2478 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2479 struct inode *new_dir, struct dentry *new_dentry)
2482 struct cgroup_name *name, *old_name;
2483 struct cgroup *cgrp;
2486 * It's convinient to use parent dir's i_mutex to protected
2489 lockdep_assert_held(&old_dir->i_mutex);
2491 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2493 if (new_dentry->d_inode)
2495 if (old_dir != new_dir)
2498 cgrp = __d_cgrp(old_dentry);
2500 name = cgroup_alloc_name(new_dentry);
2504 ret = simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2510 old_name = cgrp->name;
2511 rcu_assign_pointer(cgrp->name, name);
2513 kfree_rcu(old_name, rcu_head);
2517 static struct simple_xattrs *__d_xattrs(struct dentry *dentry)
2519 if (S_ISDIR(dentry->d_inode->i_mode))
2520 return &__d_cgrp(dentry)->xattrs;
2522 return &__d_cft(dentry)->xattrs;
2525 static inline int xattr_enabled(struct dentry *dentry)
2527 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
2528 return root->flags & CGRP_ROOT_XATTR;
2531 static bool is_valid_xattr(const char *name)
2533 if (!strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
2534 !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
2539 static int cgroup_setxattr(struct dentry *dentry, const char *name,
2540 const void *val, size_t size, int flags)
2542 if (!xattr_enabled(dentry))
2544 if (!is_valid_xattr(name))
2546 return simple_xattr_set(__d_xattrs(dentry), name, val, size, flags);
2549 static int cgroup_removexattr(struct dentry *dentry, const char *name)
2551 if (!xattr_enabled(dentry))
2553 if (!is_valid_xattr(name))
2555 return simple_xattr_remove(__d_xattrs(dentry), name);
2558 static ssize_t cgroup_getxattr(struct dentry *dentry, const char *name,
2559 void *buf, size_t size)
2561 if (!xattr_enabled(dentry))
2563 if (!is_valid_xattr(name))
2565 return simple_xattr_get(__d_xattrs(dentry), name, buf, size);
2568 static ssize_t cgroup_listxattr(struct dentry *dentry, char *buf, size_t size)
2570 if (!xattr_enabled(dentry))
2572 return simple_xattr_list(__d_xattrs(dentry), buf, size);
2575 static const struct file_operations cgroup_file_operations = {
2576 .read = cgroup_file_read,
2577 .write = cgroup_file_write,
2578 .llseek = generic_file_llseek,
2579 .open = cgroup_file_open,
2580 .release = cgroup_file_release,
2583 static const struct inode_operations cgroup_file_inode_operations = {
2584 .setxattr = cgroup_setxattr,
2585 .getxattr = cgroup_getxattr,
2586 .listxattr = cgroup_listxattr,
2587 .removexattr = cgroup_removexattr,
2590 static const struct inode_operations cgroup_dir_inode_operations = {
2591 .lookup = cgroup_lookup,
2592 .mkdir = cgroup_mkdir,
2593 .rmdir = cgroup_rmdir,
2594 .rename = cgroup_rename,
2595 .setxattr = cgroup_setxattr,
2596 .getxattr = cgroup_getxattr,
2597 .listxattr = cgroup_listxattr,
2598 .removexattr = cgroup_removexattr,
2601 static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2603 if (dentry->d_name.len > NAME_MAX)
2604 return ERR_PTR(-ENAMETOOLONG);
2605 d_add(dentry, NULL);
2610 * Check if a file is a control file
2612 static inline struct cftype *__file_cft(struct file *file)
2614 if (file_inode(file)->i_fop != &cgroup_file_operations)
2615 return ERR_PTR(-EINVAL);
2616 return __d_cft(file->f_dentry);
2619 static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2620 struct super_block *sb)
2622 struct inode *inode;
2626 if (dentry->d_inode)
2629 inode = cgroup_new_inode(mode, sb);
2633 if (S_ISDIR(mode)) {
2634 inode->i_op = &cgroup_dir_inode_operations;
2635 inode->i_fop = &simple_dir_operations;
2637 /* start off with i_nlink == 2 (for "." entry) */
2639 inc_nlink(dentry->d_parent->d_inode);
2642 * Control reaches here with cgroup_mutex held.
2643 * @inode->i_mutex should nest outside cgroup_mutex but we
2644 * want to populate it immediately without releasing
2645 * cgroup_mutex. As @inode isn't visible to anyone else
2646 * yet, trylock will always succeed without affecting
2649 WARN_ON_ONCE(!mutex_trylock(&inode->i_mutex));
2650 } else if (S_ISREG(mode)) {
2652 inode->i_fop = &cgroup_file_operations;
2653 inode->i_op = &cgroup_file_inode_operations;
2655 d_instantiate(dentry, inode);
2656 dget(dentry); /* Extra count - pin the dentry in core */
2661 * cgroup_file_mode - deduce file mode of a control file
2662 * @cft: the control file in question
2664 * returns cft->mode if ->mode is not 0
2665 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2666 * returns S_IRUGO if it has only a read handler
2667 * returns S_IWUSR if it has only a write hander
2669 static umode_t cgroup_file_mode(const struct cftype *cft)
2676 if (cft->read || cft->read_u64 || cft->read_s64 ||
2677 cft->read_map || cft->read_seq_string)
2680 if (cft->write || cft->write_u64 || cft->write_s64 ||
2681 cft->write_string || cft->trigger)
2687 static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2690 struct dentry *dir = cgrp->dentry;
2691 struct cgroup *parent = __d_cgrp(dir);
2692 struct dentry *dentry;
2696 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2698 simple_xattrs_init(&cft->xattrs);
2700 if (subsys && !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) {
2701 strcpy(name, subsys->name);
2704 strcat(name, cft->name);
2706 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2708 cfe = kzalloc(sizeof(*cfe), GFP_KERNEL);
2712 dentry = lookup_one_len(name, dir, strlen(name));
2713 if (IS_ERR(dentry)) {
2714 error = PTR_ERR(dentry);
2718 mode = cgroup_file_mode(cft);
2719 error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb);
2721 cfe->type = (void *)cft;
2722 cfe->dentry = dentry;
2723 dentry->d_fsdata = cfe;
2724 list_add_tail(&cfe->node, &parent->files);
2733 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2734 struct cftype cfts[], bool is_add)
2739 for (cft = cfts; cft->name[0] != '\0'; cft++) {
2740 /* does cft->flags tell us to skip this file on @cgrp? */
2741 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2743 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
2747 err = cgroup_add_file(cgrp, subsys, cft);
2749 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2753 cgroup_rm_file(cgrp, cft);
2759 static DEFINE_MUTEX(cgroup_cft_mutex);
2761 static void cgroup_cfts_prepare(void)
2762 __acquires(&cgroup_cft_mutex) __acquires(&cgroup_mutex)
2765 * Thanks to the entanglement with vfs inode locking, we can't walk
2766 * the existing cgroups under cgroup_mutex and create files.
2767 * Instead, we increment reference on all cgroups and build list of
2768 * them using @cgrp->cft_q_node. Grab cgroup_cft_mutex to ensure
2769 * exclusive access to the field.
2771 mutex_lock(&cgroup_cft_mutex);
2772 mutex_lock(&cgroup_mutex);
2775 static void cgroup_cfts_commit(struct cgroup_subsys *ss,
2776 struct cftype *cfts, bool is_add)
2777 __releases(&cgroup_mutex) __releases(&cgroup_cft_mutex)
2780 struct cgroup *cgrp, *n;
2782 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2783 if (cfts && ss->root != &rootnode) {
2784 list_for_each_entry(cgrp, &ss->root->allcg_list, allcg_node) {
2786 list_add_tail(&cgrp->cft_q_node, &pending);
2790 mutex_unlock(&cgroup_mutex);
2793 * All new cgroups will see @cfts update on @ss->cftsets. Add/rm
2794 * files for all cgroups which were created before.
2796 list_for_each_entry_safe(cgrp, n, &pending, cft_q_node) {
2797 struct inode *inode = cgrp->dentry->d_inode;
2799 mutex_lock(&inode->i_mutex);
2800 mutex_lock(&cgroup_mutex);
2801 if (!cgroup_is_removed(cgrp))
2802 cgroup_addrm_files(cgrp, ss, cfts, is_add);
2803 mutex_unlock(&cgroup_mutex);
2804 mutex_unlock(&inode->i_mutex);
2806 list_del_init(&cgrp->cft_q_node);
2810 mutex_unlock(&cgroup_cft_mutex);
2814 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2815 * @ss: target cgroup subsystem
2816 * @cfts: zero-length name terminated array of cftypes
2818 * Register @cfts to @ss. Files described by @cfts are created for all
2819 * existing cgroups to which @ss is attached and all future cgroups will
2820 * have them too. This function can be called anytime whether @ss is
2823 * Returns 0 on successful registration, -errno on failure. Note that this
2824 * function currently returns 0 as long as @cfts registration is successful
2825 * even if some file creation attempts on existing cgroups fail.
2827 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2829 struct cftype_set *set;
2831 set = kzalloc(sizeof(*set), GFP_KERNEL);
2835 cgroup_cfts_prepare();
2837 list_add_tail(&set->node, &ss->cftsets);
2838 cgroup_cfts_commit(ss, cfts, true);
2842 EXPORT_SYMBOL_GPL(cgroup_add_cftypes);
2845 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2846 * @ss: target cgroup subsystem
2847 * @cfts: zero-length name terminated array of cftypes
2849 * Unregister @cfts from @ss. Files described by @cfts are removed from
2850 * all existing cgroups to which @ss is attached and all future cgroups
2851 * won't have them either. This function can be called anytime whether @ss
2852 * is attached or not.
2854 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2855 * registered with @ss.
2857 int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2859 struct cftype_set *set;
2861 cgroup_cfts_prepare();
2863 list_for_each_entry(set, &ss->cftsets, node) {
2864 if (set->cfts == cfts) {
2865 list_del_init(&set->node);
2866 cgroup_cfts_commit(ss, cfts, false);
2871 cgroup_cfts_commit(ss, NULL, false);
2876 * cgroup_task_count - count the number of tasks in a cgroup.
2877 * @cgrp: the cgroup in question
2879 * Return the number of tasks in the cgroup.
2881 int cgroup_task_count(const struct cgroup *cgrp)
2884 struct cg_cgroup_link *link;
2886 read_lock(&css_set_lock);
2887 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2888 count += atomic_read(&link->cg->refcount);
2890 read_unlock(&css_set_lock);
2895 * Advance a list_head iterator. The iterator should be positioned at
2896 * the start of a css_set
2898 static void cgroup_advance_iter(struct cgroup *cgrp,
2899 struct cgroup_iter *it)
2901 struct list_head *l = it->cg_link;
2902 struct cg_cgroup_link *link;
2905 /* Advance to the next non-empty css_set */
2908 if (l == &cgrp->css_sets) {
2912 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2914 } while (list_empty(&cg->tasks));
2916 it->task = cg->tasks.next;
2920 * To reduce the fork() overhead for systems that are not actually
2921 * using their cgroups capability, we don't maintain the lists running
2922 * through each css_set to its tasks until we see the list actually
2923 * used - in other words after the first call to cgroup_iter_start().
2925 static void cgroup_enable_task_cg_lists(void)
2927 struct task_struct *p, *g;
2928 write_lock(&css_set_lock);
2929 use_task_css_set_links = 1;
2931 * We need tasklist_lock because RCU is not safe against
2932 * while_each_thread(). Besides, a forking task that has passed
2933 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2934 * is not guaranteed to have its child immediately visible in the
2935 * tasklist if we walk through it with RCU.
2937 read_lock(&tasklist_lock);
2938 do_each_thread(g, p) {
2941 * We should check if the process is exiting, otherwise
2942 * it will race with cgroup_exit() in that the list
2943 * entry won't be deleted though the process has exited.
2945 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2946 list_add(&p->cg_list, &p->cgroups->tasks);
2948 } while_each_thread(g, p);
2949 read_unlock(&tasklist_lock);
2950 write_unlock(&css_set_lock);
2954 * cgroup_next_descendant_pre - find the next descendant for pre-order walk
2955 * @pos: the current position (%NULL to initiate traversal)
2956 * @cgroup: cgroup whose descendants to walk
2958 * To be used by cgroup_for_each_descendant_pre(). Find the next
2959 * descendant to visit for pre-order traversal of @cgroup's descendants.
2961 struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
2962 struct cgroup *cgroup)
2964 struct cgroup *next;
2966 WARN_ON_ONCE(!rcu_read_lock_held());
2968 /* if first iteration, pretend we just visited @cgroup */
2970 if (list_empty(&cgroup->children))
2975 /* visit the first child if exists */
2976 next = list_first_or_null_rcu(&pos->children, struct cgroup, sibling);
2980 /* no child, visit my or the closest ancestor's next sibling */
2982 next = list_entry_rcu(pos->sibling.next, struct cgroup,
2984 if (&next->sibling != &pos->parent->children)
2988 } while (pos != cgroup);
2992 EXPORT_SYMBOL_GPL(cgroup_next_descendant_pre);
2995 * cgroup_rightmost_descendant - return the rightmost descendant of a cgroup
2996 * @pos: cgroup of interest
2998 * Return the rightmost descendant of @pos. If there's no descendant,
2999 * @pos is returned. This can be used during pre-order traversal to skip
3002 struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos)
3004 struct cgroup *last, *tmp;
3006 WARN_ON_ONCE(!rcu_read_lock_held());
3010 /* ->prev isn't RCU safe, walk ->next till the end */
3012 list_for_each_entry_rcu(tmp, &last->children, sibling)
3018 EXPORT_SYMBOL_GPL(cgroup_rightmost_descendant);
3020 static struct cgroup *cgroup_leftmost_descendant(struct cgroup *pos)
3022 struct cgroup *last;
3026 pos = list_first_or_null_rcu(&pos->children, struct cgroup,
3034 * cgroup_next_descendant_post - find the next descendant for post-order walk
3035 * @pos: the current position (%NULL to initiate traversal)
3036 * @cgroup: cgroup whose descendants to walk
3038 * To be used by cgroup_for_each_descendant_post(). Find the next
3039 * descendant to visit for post-order traversal of @cgroup's descendants.
3041 struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
3042 struct cgroup *cgroup)
3044 struct cgroup *next;
3046 WARN_ON_ONCE(!rcu_read_lock_held());
3048 /* if first iteration, visit the leftmost descendant */
3050 next = cgroup_leftmost_descendant(cgroup);
3051 return next != cgroup ? next : NULL;
3054 /* if there's an unvisited sibling, visit its leftmost descendant */
3055 next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling);
3056 if (&next->sibling != &pos->parent->children)
3057 return cgroup_leftmost_descendant(next);
3059 /* no sibling left, visit parent */
3061 return next != cgroup ? next : NULL;
3063 EXPORT_SYMBOL_GPL(cgroup_next_descendant_post);
3065 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
3066 __acquires(css_set_lock)
3069 * The first time anyone tries to iterate across a cgroup,
3070 * we need to enable the list linking each css_set to its
3071 * tasks, and fix up all existing tasks.
3073 if (!use_task_css_set_links)
3074 cgroup_enable_task_cg_lists();
3076 read_lock(&css_set_lock);
3077 it->cg_link = &cgrp->css_sets;
3078 cgroup_advance_iter(cgrp, it);
3081 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
3082 struct cgroup_iter *it)
3084 struct task_struct *res;
3085 struct list_head *l = it->task;
3086 struct cg_cgroup_link *link;
3088 /* If the iterator cg is NULL, we have no tasks */
3091 res = list_entry(l, struct task_struct, cg_list);
3092 /* Advance iterator to find next entry */
3094 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
3095 if (l == &link->cg->tasks) {
3096 /* We reached the end of this task list - move on to
3097 * the next cg_cgroup_link */
3098 cgroup_advance_iter(cgrp, it);
3105 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
3106 __releases(css_set_lock)
3108 read_unlock(&css_set_lock);
3111 static inline int started_after_time(struct task_struct *t1,
3112 struct timespec *time,
3113 struct task_struct *t2)
3115 int start_diff = timespec_compare(&t1->start_time, time);
3116 if (start_diff > 0) {
3118 } else if (start_diff < 0) {
3122 * Arbitrarily, if two processes started at the same
3123 * time, we'll say that the lower pointer value
3124 * started first. Note that t2 may have exited by now
3125 * so this may not be a valid pointer any longer, but
3126 * that's fine - it still serves to distinguish
3127 * between two tasks started (effectively) simultaneously.
3134 * This function is a callback from heap_insert() and is used to order
3136 * In this case we order the heap in descending task start time.
3138 static inline int started_after(void *p1, void *p2)
3140 struct task_struct *t1 = p1;
3141 struct task_struct *t2 = p2;
3142 return started_after_time(t1, &t2->start_time, t2);
3146 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3147 * @scan: struct cgroup_scanner containing arguments for the scan
3149 * Arguments include pointers to callback functions test_task() and
3151 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3152 * and if it returns true, call process_task() for it also.
3153 * The test_task pointer may be NULL, meaning always true (select all tasks).
3154 * Effectively duplicates cgroup_iter_{start,next,end}()
3155 * but does not lock css_set_lock for the call to process_task().
3156 * The struct cgroup_scanner may be embedded in any structure of the caller's
3158 * It is guaranteed that process_task() will act on every task that
3159 * is a member of the cgroup for the duration of this call. This
3160 * function may or may not call process_task() for tasks that exit
3161 * or move to a different cgroup during the call, or are forked or
3162 * move into the cgroup during the call.
3164 * Note that test_task() may be called with locks held, and may in some
3165 * situations be called multiple times for the same task, so it should
3167 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3168 * pre-allocated and will be used for heap operations (and its "gt" member will
3169 * be overwritten), else a temporary heap will be used (allocation of which
3170 * may cause this function to fail).
3172 int cgroup_scan_tasks(struct cgroup_scanner *scan)
3175 struct cgroup_iter it;
3176 struct task_struct *p, *dropped;
3177 /* Never dereference latest_task, since it's not refcounted */
3178 struct task_struct *latest_task = NULL;
3179 struct ptr_heap tmp_heap;
3180 struct ptr_heap *heap;
3181 struct timespec latest_time = { 0, 0 };
3184 /* The caller supplied our heap and pre-allocated its memory */
3186 heap->gt = &started_after;
3188 /* We need to allocate our own heap memory */
3190 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
3192 /* cannot allocate the heap */
3198 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3199 * to determine which are of interest, and using the scanner's
3200 * "process_task" callback to process any of them that need an update.
3201 * Since we don't want to hold any locks during the task updates,
3202 * gather tasks to be processed in a heap structure.
3203 * The heap is sorted by descending task start time.
3204 * If the statically-sized heap fills up, we overflow tasks that
3205 * started later, and in future iterations only consider tasks that
3206 * started after the latest task in the previous pass. This
3207 * guarantees forward progress and that we don't miss any tasks.
3210 cgroup_iter_start(scan->cg, &it);
3211 while ((p = cgroup_iter_next(scan->cg, &it))) {
3213 * Only affect tasks that qualify per the caller's callback,
3214 * if he provided one
3216 if (scan->test_task && !scan->test_task(p, scan))
3219 * Only process tasks that started after the last task
3222 if (!started_after_time(p, &latest_time, latest_task))
3224 dropped = heap_insert(heap, p);
3225 if (dropped == NULL) {
3227 * The new task was inserted; the heap wasn't
3231 } else if (dropped != p) {
3233 * The new task was inserted, and pushed out a
3237 put_task_struct(dropped);
3240 * Else the new task was newer than anything already in
3241 * the heap and wasn't inserted
3244 cgroup_iter_end(scan->cg, &it);
3247 for (i = 0; i < heap->size; i++) {
3248 struct task_struct *q = heap->ptrs[i];
3250 latest_time = q->start_time;
3253 /* Process the task per the caller's callback */
3254 scan->process_task(q, scan);
3258 * If we had to process any tasks at all, scan again
3259 * in case some of them were in the middle of forking
3260 * children that didn't get processed.
3261 * Not the most efficient way to do it, but it avoids
3262 * having to take callback_mutex in the fork path
3266 if (heap == &tmp_heap)
3267 heap_free(&tmp_heap);
3271 static void cgroup_transfer_one_task(struct task_struct *task,
3272 struct cgroup_scanner *scan)
3274 struct cgroup *new_cgroup = scan->data;
3276 mutex_lock(&cgroup_mutex);
3277 cgroup_attach_task(new_cgroup, task, false);
3278 mutex_unlock(&cgroup_mutex);
3282 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3283 * @to: cgroup to which the tasks will be moved
3284 * @from: cgroup in which the tasks currently reside
3286 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
3288 struct cgroup_scanner scan;
3291 scan.test_task = NULL; /* select all tasks in cgroup */
3292 scan.process_task = cgroup_transfer_one_task;
3296 return cgroup_scan_tasks(&scan);
3300 * Stuff for reading the 'tasks'/'procs' files.
3302 * Reading this file can return large amounts of data if a cgroup has
3303 * *lots* of attached tasks. So it may need several calls to read(),
3304 * but we cannot guarantee that the information we produce is correct
3305 * unless we produce it entirely atomically.
3309 /* which pidlist file are we talking about? */
3310 enum cgroup_filetype {
3316 * A pidlist is a list of pids that virtually represents the contents of one
3317 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3318 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3321 struct cgroup_pidlist {
3323 * used to find which pidlist is wanted. doesn't change as long as
3324 * this particular list stays in the list.
3326 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
3329 /* how many elements the above list has */
3331 /* how many files are using the current array */
3333 /* each of these stored in a list by its cgroup */
3334 struct list_head links;
3335 /* pointer to the cgroup we belong to, for list removal purposes */
3336 struct cgroup *owner;
3337 /* protects the other fields */
3338 struct rw_semaphore mutex;
3342 * The following two functions "fix" the issue where there are more pids
3343 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3344 * TODO: replace with a kernel-wide solution to this problem
3346 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3347 static void *pidlist_allocate(int count)
3349 if (PIDLIST_TOO_LARGE(count))
3350 return vmalloc(count * sizeof(pid_t));
3352 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3354 static void pidlist_free(void *p)
3356 if (is_vmalloc_addr(p))
3363 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3364 * Returns the number of unique elements.
3366 static int pidlist_uniq(pid_t *list, int length)
3371 * we presume the 0th element is unique, so i starts at 1. trivial
3372 * edge cases first; no work needs to be done for either
3374 if (length == 0 || length == 1)
3376 /* src and dest walk down the list; dest counts unique elements */
3377 for (src = 1; src < length; src++) {
3378 /* find next unique element */
3379 while (list[src] == list[src-1]) {
3384 /* dest always points to where the next unique element goes */
3385 list[dest] = list[src];
3392 static int cmppid(const void *a, const void *b)
3394 return *(pid_t *)a - *(pid_t *)b;
3398 * find the appropriate pidlist for our purpose (given procs vs tasks)
3399 * returns with the lock on that pidlist already held, and takes care
3400 * of the use count, or returns NULL with no locks held if we're out of
3403 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3404 enum cgroup_filetype type)
3406 struct cgroup_pidlist *l;
3407 /* don't need task_nsproxy() if we're looking at ourself */
3408 struct pid_namespace *ns = task_active_pid_ns(current);
3411 * We can't drop the pidlist_mutex before taking the l->mutex in case
3412 * the last ref-holder is trying to remove l from the list at the same
3413 * time. Holding the pidlist_mutex precludes somebody taking whichever
3414 * list we find out from under us - compare release_pid_array().
3416 mutex_lock(&cgrp->pidlist_mutex);
3417 list_for_each_entry(l, &cgrp->pidlists, links) {
3418 if (l->key.type == type && l->key.ns == ns) {
3419 /* make sure l doesn't vanish out from under us */
3420 down_write(&l->mutex);
3421 mutex_unlock(&cgrp->pidlist_mutex);
3425 /* entry not found; create a new one */
3426 l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3428 mutex_unlock(&cgrp->pidlist_mutex);
3431 init_rwsem(&l->mutex);
3432 down_write(&l->mutex);
3434 l->key.ns = get_pid_ns(ns);
3435 l->use_count = 0; /* don't increment here */
3438 list_add(&l->links, &cgrp->pidlists);
3439 mutex_unlock(&cgrp->pidlist_mutex);
3444 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3446 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3447 struct cgroup_pidlist **lp)
3451 int pid, n = 0; /* used for populating the array */
3452 struct cgroup_iter it;
3453 struct task_struct *tsk;
3454 struct cgroup_pidlist *l;
3457 * If cgroup gets more users after we read count, we won't have
3458 * enough space - tough. This race is indistinguishable to the
3459 * caller from the case that the additional cgroup users didn't
3460 * show up until sometime later on.
3462 length = cgroup_task_count(cgrp);
3463 array = pidlist_allocate(length);
3466 /* now, populate the array */
3467 cgroup_iter_start(cgrp, &it);
3468 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3469 if (unlikely(n == length))
3471 /* get tgid or pid for procs or tasks file respectively */
3472 if (type == CGROUP_FILE_PROCS)
3473 pid = task_tgid_vnr(tsk);
3475 pid = task_pid_vnr(tsk);
3476 if (pid > 0) /* make sure to only use valid results */
3479 cgroup_iter_end(cgrp, &it);
3481 /* now sort & (if procs) strip out duplicates */
3482 sort(array, length, sizeof(pid_t), cmppid, NULL);
3483 if (type == CGROUP_FILE_PROCS)
3484 length = pidlist_uniq(array, length);
3485 l = cgroup_pidlist_find(cgrp, type);
3487 pidlist_free(array);
3490 /* store array, freeing old if necessary - lock already held */
3491 pidlist_free(l->list);
3495 up_write(&l->mutex);
3501 * cgroupstats_build - build and fill cgroupstats
3502 * @stats: cgroupstats to fill information into
3503 * @dentry: A dentry entry belonging to the cgroup for which stats have
3506 * Build and fill cgroupstats so that taskstats can export it to user
3509 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3512 struct cgroup *cgrp;
3513 struct cgroup_iter it;
3514 struct task_struct *tsk;
3517 * Validate dentry by checking the superblock operations,
3518 * and make sure it's a directory.
3520 if (dentry->d_sb->s_op != &cgroup_ops ||
3521 !S_ISDIR(dentry->d_inode->i_mode))
3525 cgrp = dentry->d_fsdata;
3527 cgroup_iter_start(cgrp, &it);
3528 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3529 switch (tsk->state) {
3531 stats->nr_running++;
3533 case TASK_INTERRUPTIBLE:
3534 stats->nr_sleeping++;
3536 case TASK_UNINTERRUPTIBLE:
3537 stats->nr_uninterruptible++;
3540 stats->nr_stopped++;
3543 if (delayacct_is_task_waiting_on_io(tsk))
3544 stats->nr_io_wait++;
3548 cgroup_iter_end(cgrp, &it);
3556 * seq_file methods for the tasks/procs files. The seq_file position is the
3557 * next pid to display; the seq_file iterator is a pointer to the pid
3558 * in the cgroup->l->list array.
3561 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3564 * Initially we receive a position value that corresponds to
3565 * one more than the last pid shown (or 0 on the first call or
3566 * after a seek to the start). Use a binary-search to find the
3567 * next pid to display, if any
3569 struct cgroup_pidlist *l = s->private;
3570 int index = 0, pid = *pos;
3573 down_read(&l->mutex);
3575 int end = l->length;
3577 while (index < end) {
3578 int mid = (index + end) / 2;
3579 if (l->list[mid] == pid) {
3582 } else if (l->list[mid] <= pid)
3588 /* If we're off the end of the array, we're done */
3589 if (index >= l->length)
3591 /* Update the abstract position to be the actual pid that we found */
3592 iter = l->list + index;
3597 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3599 struct cgroup_pidlist *l = s->private;
3603 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3605 struct cgroup_pidlist *l = s->private;
3607 pid_t *end = l->list + l->length;
3609 * Advance to the next pid in the array. If this goes off the
3621 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3623 return seq_printf(s, "%d\n", *(int *)v);
3627 * seq_operations functions for iterating on pidlists through seq_file -
3628 * independent of whether it's tasks or procs
3630 static const struct seq_operations cgroup_pidlist_seq_operations = {
3631 .start = cgroup_pidlist_start,
3632 .stop = cgroup_pidlist_stop,
3633 .next = cgroup_pidlist_next,
3634 .show = cgroup_pidlist_show,
3637 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3640 * the case where we're the last user of this particular pidlist will
3641 * have us remove it from the cgroup's list, which entails taking the
3642 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3643 * pidlist_mutex, we have to take pidlist_mutex first.
3645 mutex_lock(&l->owner->pidlist_mutex);
3646 down_write(&l->mutex);
3647 BUG_ON(!l->use_count);
3648 if (!--l->use_count) {
3649 /* we're the last user if refcount is 0; remove and free */
3650 list_del(&l->links);
3651 mutex_unlock(&l->owner->pidlist_mutex);
3652 pidlist_free(l->list);
3653 put_pid_ns(l->key.ns);
3654 up_write(&l->mutex);
3658 mutex_unlock(&l->owner->pidlist_mutex);
3659 up_write(&l->mutex);
3662 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3664 struct cgroup_pidlist *l;
3665 if (!(file->f_mode & FMODE_READ))
3668 * the seq_file will only be initialized if the file was opened for
3669 * reading; hence we check if it's not null only in that case.
3671 l = ((struct seq_file *)file->private_data)->private;
3672 cgroup_release_pid_array(l);
3673 return seq_release(inode, file);
3676 static const struct file_operations cgroup_pidlist_operations = {
3678 .llseek = seq_lseek,
3679 .write = cgroup_file_write,
3680 .release = cgroup_pidlist_release,
3684 * The following functions handle opens on a file that displays a pidlist
3685 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3688 /* helper function for the two below it */
3689 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3691 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3692 struct cgroup_pidlist *l;
3695 /* Nothing to do for write-only files */
3696 if (!(file->f_mode & FMODE_READ))
3699 /* have the array populated */
3700 retval = pidlist_array_load(cgrp, type, &l);
3703 /* configure file information */
3704 file->f_op = &cgroup_pidlist_operations;
3706 retval = seq_open(file, &cgroup_pidlist_seq_operations);
3708 cgroup_release_pid_array(l);
3711 ((struct seq_file *)file->private_data)->private = l;
3714 static int cgroup_tasks_open(struct inode *unused, struct file *file)
3716 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3718 static int cgroup_procs_open(struct inode *unused, struct file *file)
3720 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3723 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3726 return notify_on_release(cgrp);
3729 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3733 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3735 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3737 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3742 * Unregister event and free resources.
3744 * Gets called from workqueue.
3746 static void cgroup_event_remove(struct work_struct *work)
3748 struct cgroup_event *event = container_of(work, struct cgroup_event,
3750 struct cgroup *cgrp = event->cgrp;
3752 remove_wait_queue(event->wqh, &event->wait);
3754 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3756 /* Notify userspace the event is going away. */
3757 eventfd_signal(event->eventfd, 1);
3759 eventfd_ctx_put(event->eventfd);
3765 * Gets called on POLLHUP on eventfd when user closes it.
3767 * Called with wqh->lock held and interrupts disabled.
3769 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3770 int sync, void *key)
3772 struct cgroup_event *event = container_of(wait,
3773 struct cgroup_event, wait);
3774 struct cgroup *cgrp = event->cgrp;
3775 unsigned long flags = (unsigned long)key;
3777 if (flags & POLLHUP) {
3779 * If the event has been detached at cgroup removal, we
3780 * can simply return knowing the other side will cleanup
3783 * We can't race against event freeing since the other
3784 * side will require wqh->lock via remove_wait_queue(),
3787 spin_lock(&cgrp->event_list_lock);
3788 if (!list_empty(&event->list)) {
3789 list_del_init(&event->list);
3791 * We are in atomic context, but cgroup_event_remove()
3792 * may sleep, so we have to call it in workqueue.
3794 schedule_work(&event->remove);
3796 spin_unlock(&cgrp->event_list_lock);
3802 static void cgroup_event_ptable_queue_proc(struct file *file,
3803 wait_queue_head_t *wqh, poll_table *pt)
3805 struct cgroup_event *event = container_of(pt,
3806 struct cgroup_event, pt);
3809 add_wait_queue(wqh, &event->wait);
3813 * Parse input and register new cgroup event handler.
3815 * Input must be in format '<event_fd> <control_fd> <args>'.
3816 * Interpretation of args is defined by control file implementation.
3818 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3821 struct cgroup_event *event = NULL;
3822 struct cgroup *cgrp_cfile;
3823 unsigned int efd, cfd;
3824 struct file *efile = NULL;
3825 struct file *cfile = NULL;
3829 efd = simple_strtoul(buffer, &endp, 10);
3834 cfd = simple_strtoul(buffer, &endp, 10);
3835 if ((*endp != ' ') && (*endp != '\0'))
3839 event = kzalloc(sizeof(*event), GFP_KERNEL);
3843 INIT_LIST_HEAD(&event->list);
3844 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3845 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3846 INIT_WORK(&event->remove, cgroup_event_remove);
3848 efile = eventfd_fget(efd);
3849 if (IS_ERR(efile)) {
3850 ret = PTR_ERR(efile);
3854 event->eventfd = eventfd_ctx_fileget(efile);
3855 if (IS_ERR(event->eventfd)) {
3856 ret = PTR_ERR(event->eventfd);
3866 /* the process need read permission on control file */
3867 /* AV: shouldn't we check that it's been opened for read instead? */
3868 ret = inode_permission(file_inode(cfile), MAY_READ);
3872 event->cft = __file_cft(cfile);
3873 if (IS_ERR(event->cft)) {
3874 ret = PTR_ERR(event->cft);
3879 * The file to be monitored must be in the same cgroup as
3880 * cgroup.event_control is.
3882 cgrp_cfile = __d_cgrp(cfile->f_dentry->d_parent);
3883 if (cgrp_cfile != cgrp) {
3888 if (!event->cft->register_event || !event->cft->unregister_event) {
3893 ret = event->cft->register_event(cgrp, event->cft,
3894 event->eventfd, buffer);
3899 * Events should be removed after rmdir of cgroup directory, but before
3900 * destroying subsystem state objects. Let's take reference to cgroup
3901 * directory dentry to do that.
3905 spin_lock(&cgrp->event_list_lock);
3906 list_add(&event->list, &cgrp->event_list);
3907 spin_unlock(&cgrp->event_list_lock);
3918 if (event && event->eventfd && !IS_ERR(event->eventfd))
3919 eventfd_ctx_put(event->eventfd);
3921 if (!IS_ERR_OR_NULL(efile))
3929 static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3932 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3935 static int cgroup_clone_children_write(struct cgroup *cgrp,
3940 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3942 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3947 * for the common functions, 'private' gives the type of file
3949 /* for hysterical raisins, we can't put this on the older files */
3950 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3951 static struct cftype files[] = {
3954 .open = cgroup_tasks_open,
3955 .write_u64 = cgroup_tasks_write,
3956 .release = cgroup_pidlist_release,
3957 .mode = S_IRUGO | S_IWUSR,
3960 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3961 .open = cgroup_procs_open,
3962 .write_u64 = cgroup_procs_write,
3963 .release = cgroup_pidlist_release,
3964 .mode = S_IRUGO | S_IWUSR,
3967 .name = "notify_on_release",
3968 .read_u64 = cgroup_read_notify_on_release,
3969 .write_u64 = cgroup_write_notify_on_release,
3972 .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3973 .write_string = cgroup_write_event_control,
3977 .name = "cgroup.clone_children",
3978 .read_u64 = cgroup_clone_children_read,
3979 .write_u64 = cgroup_clone_children_write,
3982 .name = "release_agent",
3983 .flags = CFTYPE_ONLY_ON_ROOT,
3984 .read_seq_string = cgroup_release_agent_show,
3985 .write_string = cgroup_release_agent_write,
3986 .max_write_len = PATH_MAX,
3992 * cgroup_populate_dir - selectively creation of files in a directory
3993 * @cgrp: target cgroup
3994 * @base_files: true if the base files should be added
3995 * @subsys_mask: mask of the subsystem ids whose files should be added
3997 static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
3998 unsigned long subsys_mask)
4001 struct cgroup_subsys *ss;
4004 err = cgroup_addrm_files(cgrp, NULL, files, true);
4009 /* process cftsets of each subsystem */
4010 for_each_subsys(cgrp->root, ss) {
4011 struct cftype_set *set;
4012 if (!test_bit(ss->subsys_id, &subsys_mask))
4015 list_for_each_entry(set, &ss->cftsets, node)
4016 cgroup_addrm_files(cgrp, ss, set->cfts, true);
4019 /* This cgroup is ready now */
4020 for_each_subsys(cgrp->root, ss) {
4021 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4023 * Update id->css pointer and make this css visible from
4024 * CSS ID functions. This pointer will be dereferened
4025 * from RCU-read-side without locks.
4028 rcu_assign_pointer(css->id->css, css);
4034 static void css_dput_fn(struct work_struct *work)
4036 struct cgroup_subsys_state *css =
4037 container_of(work, struct cgroup_subsys_state, dput_work);
4038 struct dentry *dentry = css->cgroup->dentry;
4039 struct super_block *sb = dentry->d_sb;
4041 atomic_inc(&sb->s_active);
4043 deactivate_super(sb);
4046 static void init_cgroup_css(struct cgroup_subsys_state *css,
4047 struct cgroup_subsys *ss,
4048 struct cgroup *cgrp)
4051 atomic_set(&css->refcnt, 1);
4054 if (cgrp == dummytop)
4055 css->flags |= CSS_ROOT;
4056 BUG_ON(cgrp->subsys[ss->subsys_id]);
4057 cgrp->subsys[ss->subsys_id] = css;
4060 * css holds an extra ref to @cgrp->dentry which is put on the last
4061 * css_put(). dput() requires process context, which css_put() may
4062 * be called without. @css->dput_work will be used to invoke
4063 * dput() asynchronously from css_put().
4065 INIT_WORK(&css->dput_work, css_dput_fn);
4068 /* invoke ->post_create() on a new CSS and mark it online if successful */
4069 static int online_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4073 lockdep_assert_held(&cgroup_mutex);
4076 ret = ss->css_online(cgrp);
4078 cgrp->subsys[ss->subsys_id]->flags |= CSS_ONLINE;
4082 /* if the CSS is online, invoke ->pre_destory() on it and mark it offline */
4083 static void offline_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4084 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4086 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4088 lockdep_assert_held(&cgroup_mutex);
4090 if (!(css->flags & CSS_ONLINE))
4093 if (ss->css_offline)
4094 ss->css_offline(cgrp);
4096 cgrp->subsys[ss->subsys_id]->flags &= ~CSS_ONLINE;
4100 * cgroup_create - create a cgroup
4101 * @parent: cgroup that will be parent of the new cgroup
4102 * @dentry: dentry of the new cgroup
4103 * @mode: mode to set on new inode
4105 * Must be called with the mutex on the parent inode held
4107 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
4110 struct cgroup *cgrp;
4111 struct cgroup_name *name;
4112 struct cgroupfs_root *root = parent->root;
4114 struct cgroup_subsys *ss;
4115 struct super_block *sb = root->sb;
4117 /* allocate the cgroup and its ID, 0 is reserved for the root */
4118 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
4122 name = cgroup_alloc_name(dentry);
4125 rcu_assign_pointer(cgrp->name, name);
4127 cgrp->id = ida_simple_get(&root->cgroup_ida, 1, 0, GFP_KERNEL);
4132 * Only live parents can have children. Note that the liveliness
4133 * check isn't strictly necessary because cgroup_mkdir() and
4134 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4135 * anyway so that locking is contained inside cgroup proper and we
4136 * don't get nasty surprises if we ever grow another caller.
4138 if (!cgroup_lock_live_group(parent)) {
4143 /* Grab a reference on the superblock so the hierarchy doesn't
4144 * get deleted on unmount if there are child cgroups. This
4145 * can be done outside cgroup_mutex, since the sb can't
4146 * disappear while someone has an open control file on the
4148 atomic_inc(&sb->s_active);
4150 init_cgroup_housekeeping(cgrp);
4152 dentry->d_fsdata = cgrp;
4153 cgrp->dentry = dentry;
4155 cgrp->parent = parent;
4156 cgrp->root = parent->root;
4157 cgrp->top_cgroup = parent->top_cgroup;
4159 if (notify_on_release(parent))
4160 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
4162 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
4163 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4165 for_each_subsys(root, ss) {
4166 struct cgroup_subsys_state *css;
4168 css = ss->css_alloc(cgrp);
4173 init_cgroup_css(css, ss, cgrp);
4175 err = alloc_css_id(ss, parent, cgrp);
4182 * Create directory. cgroup_create_file() returns with the new
4183 * directory locked on success so that it can be populated without
4184 * dropping cgroup_mutex.
4186 err = cgroup_create_file(dentry, S_IFDIR | mode, sb);
4189 lockdep_assert_held(&dentry->d_inode->i_mutex);
4191 /* allocation complete, commit to creation */
4192 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
4193 list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children);
4194 root->number_of_cgroups++;
4196 /* each css holds a ref to the cgroup's dentry */
4197 for_each_subsys(root, ss)
4200 /* hold a ref to the parent's dentry */
4201 dget(parent->dentry);
4203 /* creation succeeded, notify subsystems */
4204 for_each_subsys(root, ss) {
4205 err = online_css(ss, cgrp);
4209 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
4211 pr_warning("cgroup: %s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
4212 current->comm, current->pid, ss->name);
4213 if (!strcmp(ss->name, "memory"))
4214 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4215 ss->warned_broken_hierarchy = true;
4219 err = cgroup_populate_dir(cgrp, true, root->subsys_mask);
4223 mutex_unlock(&cgroup_mutex);
4224 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
4229 for_each_subsys(root, ss) {
4230 if (cgrp->subsys[ss->subsys_id])
4233 mutex_unlock(&cgroup_mutex);
4234 /* Release the reference count that we took on the superblock */
4235 deactivate_super(sb);
4237 ida_simple_remove(&root->cgroup_ida, cgrp->id);
4239 kfree(rcu_dereference_raw(cgrp->name));
4245 cgroup_destroy_locked(cgrp);
4246 mutex_unlock(&cgroup_mutex);
4247 mutex_unlock(&dentry->d_inode->i_mutex);
4251 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4253 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
4255 /* the vfs holds inode->i_mutex already */
4256 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
4259 static int cgroup_destroy_locked(struct cgroup *cgrp)
4260 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4262 struct dentry *d = cgrp->dentry;
4263 struct cgroup *parent = cgrp->parent;
4264 struct cgroup_event *event, *tmp;
4265 struct cgroup_subsys *ss;
4267 lockdep_assert_held(&d->d_inode->i_mutex);
4268 lockdep_assert_held(&cgroup_mutex);
4270 if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children))
4274 * Block new css_tryget() by deactivating refcnt and mark @cgrp
4275 * removed. This makes future css_tryget() and child creation
4276 * attempts fail thus maintaining the removal conditions verified
4279 for_each_subsys(cgrp->root, ss) {
4280 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4282 WARN_ON(atomic_read(&css->refcnt) < 0);
4283 atomic_add(CSS_DEACT_BIAS, &css->refcnt);
4285 set_bit(CGRP_REMOVED, &cgrp->flags);
4287 /* tell subsystems to initate destruction */
4288 for_each_subsys(cgrp->root, ss)
4289 offline_css(ss, cgrp);
4292 * Put all the base refs. Each css holds an extra reference to the
4293 * cgroup's dentry and cgroup removal proceeds regardless of css
4294 * refs. On the last put of each css, whenever that may be, the
4295 * extra dentry ref is put so that dentry destruction happens only
4296 * after all css's are released.
4298 for_each_subsys(cgrp->root, ss)
4299 css_put(cgrp->subsys[ss->subsys_id]);
4301 raw_spin_lock(&release_list_lock);
4302 if (!list_empty(&cgrp->release_list))
4303 list_del_init(&cgrp->release_list);
4304 raw_spin_unlock(&release_list_lock);
4306 /* delete this cgroup from parent->children */
4307 list_del_rcu(&cgrp->sibling);
4308 list_del_init(&cgrp->allcg_node);
4311 cgroup_d_remove_dir(d);
4314 set_bit(CGRP_RELEASABLE, &parent->flags);
4315 check_for_release(parent);
4318 * Unregister events and notify userspace.
4319 * Notify userspace about cgroup removing only after rmdir of cgroup
4320 * directory to avoid race between userspace and kernelspace.
4322 spin_lock(&cgrp->event_list_lock);
4323 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4324 list_del_init(&event->list);
4325 schedule_work(&event->remove);
4327 spin_unlock(&cgrp->event_list_lock);
4332 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
4336 mutex_lock(&cgroup_mutex);
4337 ret = cgroup_destroy_locked(dentry->d_fsdata);
4338 mutex_unlock(&cgroup_mutex);
4343 static void __init_or_module cgroup_init_cftsets(struct cgroup_subsys *ss)
4345 INIT_LIST_HEAD(&ss->cftsets);
4348 * base_cftset is embedded in subsys itself, no need to worry about
4351 if (ss->base_cftypes) {
4352 ss->base_cftset.cfts = ss->base_cftypes;
4353 list_add_tail(&ss->base_cftset.node, &ss->cftsets);
4357 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4359 struct cgroup_subsys_state *css;
4361 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4363 mutex_lock(&cgroup_mutex);
4365 /* init base cftset */
4366 cgroup_init_cftsets(ss);
4368 /* Create the top cgroup state for this subsystem */
4369 list_add(&ss->sibling, &rootnode.subsys_list);
4370 ss->root = &rootnode;
4371 css = ss->css_alloc(dummytop);
4372 /* We don't handle early failures gracefully */
4373 BUG_ON(IS_ERR(css));
4374 init_cgroup_css(css, ss, dummytop);
4376 /* Update the init_css_set to contain a subsys
4377 * pointer to this state - since the subsystem is
4378 * newly registered, all tasks and hence the
4379 * init_css_set is in the subsystem's top cgroup. */
4380 init_css_set.subsys[ss->subsys_id] = css;
4382 need_forkexit_callback |= ss->fork || ss->exit;
4384 /* At system boot, before all subsystems have been
4385 * registered, no tasks have been forked, so we don't
4386 * need to invoke fork callbacks here. */
4387 BUG_ON(!list_empty(&init_task.tasks));
4390 BUG_ON(online_css(ss, dummytop));
4392 mutex_unlock(&cgroup_mutex);
4394 /* this function shouldn't be used with modular subsystems, since they
4395 * need to register a subsys_id, among other things */
4400 * cgroup_load_subsys: load and register a modular subsystem at runtime
4401 * @ss: the subsystem to load
4403 * This function should be called in a modular subsystem's initcall. If the
4404 * subsystem is built as a module, it will be assigned a new subsys_id and set
4405 * up for use. If the subsystem is built-in anyway, work is delegated to the
4406 * simpler cgroup_init_subsys.
4408 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4410 struct cgroup_subsys_state *css;
4412 struct hlist_node *tmp;
4416 /* check name and function validity */
4417 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4418 ss->css_alloc == NULL || ss->css_free == NULL)
4422 * we don't support callbacks in modular subsystems. this check is
4423 * before the ss->module check for consistency; a subsystem that could
4424 * be a module should still have no callbacks even if the user isn't
4425 * compiling it as one.
4427 if (ss->fork || ss->exit)
4431 * an optionally modular subsystem is built-in: we want to do nothing,
4432 * since cgroup_init_subsys will have already taken care of it.
4434 if (ss->module == NULL) {
4435 /* a sanity check */
4436 BUG_ON(subsys[ss->subsys_id] != ss);
4440 /* init base cftset */
4441 cgroup_init_cftsets(ss);
4443 mutex_lock(&cgroup_mutex);
4444 subsys[ss->subsys_id] = ss;
4447 * no ss->css_alloc seems to need anything important in the ss
4448 * struct, so this can happen first (i.e. before the rootnode
4451 css = ss->css_alloc(dummytop);
4453 /* failure case - need to deassign the subsys[] slot. */
4454 subsys[ss->subsys_id] = NULL;
4455 mutex_unlock(&cgroup_mutex);
4456 return PTR_ERR(css);
4459 list_add(&ss->sibling, &rootnode.subsys_list);
4460 ss->root = &rootnode;
4462 /* our new subsystem will be attached to the dummy hierarchy. */
4463 init_cgroup_css(css, ss, dummytop);
4464 /* init_idr must be after init_cgroup_css because it sets css->id. */
4466 ret = cgroup_init_idr(ss, css);
4472 * Now we need to entangle the css into the existing css_sets. unlike
4473 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4474 * will need a new pointer to it; done by iterating the css_set_table.
4475 * furthermore, modifying the existing css_sets will corrupt the hash
4476 * table state, so each changed css_set will need its hash recomputed.
4477 * this is all done under the css_set_lock.
4479 write_lock(&css_set_lock);
4480 hash_for_each_safe(css_set_table, i, tmp, cg, hlist) {
4481 /* skip entries that we already rehashed */
4482 if (cg->subsys[ss->subsys_id])
4484 /* remove existing entry */
4485 hash_del(&cg->hlist);
4487 cg->subsys[ss->subsys_id] = css;
4488 /* recompute hash and restore entry */
4489 key = css_set_hash(cg->subsys);
4490 hash_add(css_set_table, &cg->hlist, key);
4492 write_unlock(&css_set_lock);
4495 ret = online_css(ss, dummytop);
4500 mutex_unlock(&cgroup_mutex);
4504 mutex_unlock(&cgroup_mutex);
4505 /* @ss can't be mounted here as try_module_get() would fail */
4506 cgroup_unload_subsys(ss);
4509 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4512 * cgroup_unload_subsys: unload a modular subsystem
4513 * @ss: the subsystem to unload
4515 * This function should be called in a modular subsystem's exitcall. When this
4516 * function is invoked, the refcount on the subsystem's module will be 0, so
4517 * the subsystem will not be attached to any hierarchy.
4519 void cgroup_unload_subsys(struct cgroup_subsys *ss)
4521 struct cg_cgroup_link *link;
4523 BUG_ON(ss->module == NULL);
4526 * we shouldn't be called if the subsystem is in use, and the use of
4527 * try_module_get in parse_cgroupfs_options should ensure that it
4528 * doesn't start being used while we're killing it off.
4530 BUG_ON(ss->root != &rootnode);
4532 mutex_lock(&cgroup_mutex);
4534 offline_css(ss, dummytop);
4538 idr_destroy(&ss->idr);
4540 /* deassign the subsys_id */
4541 subsys[ss->subsys_id] = NULL;
4543 /* remove subsystem from rootnode's list of subsystems */
4544 list_del_init(&ss->sibling);
4547 * disentangle the css from all css_sets attached to the dummytop. as
4548 * in loading, we need to pay our respects to the hashtable gods.
4550 write_lock(&css_set_lock);
4551 list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
4552 struct css_set *cg = link->cg;
4555 hash_del(&cg->hlist);
4556 cg->subsys[ss->subsys_id] = NULL;
4557 key = css_set_hash(cg->subsys);
4558 hash_add(css_set_table, &cg->hlist, key);
4560 write_unlock(&css_set_lock);
4563 * remove subsystem's css from the dummytop and free it - need to
4564 * free before marking as null because ss->css_free needs the
4565 * cgrp->subsys pointer to find their state. note that this also
4566 * takes care of freeing the css_id.
4568 ss->css_free(dummytop);
4569 dummytop->subsys[ss->subsys_id] = NULL;
4571 mutex_unlock(&cgroup_mutex);
4573 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4576 * cgroup_init_early - cgroup initialization at system boot
4578 * Initialize cgroups at system boot, and initialize any
4579 * subsystems that request early init.
4581 int __init cgroup_init_early(void)
4584 atomic_set(&init_css_set.refcount, 1);
4585 INIT_LIST_HEAD(&init_css_set.cg_links);
4586 INIT_LIST_HEAD(&init_css_set.tasks);
4587 INIT_HLIST_NODE(&init_css_set.hlist);
4589 init_cgroup_root(&rootnode);
4591 init_task.cgroups = &init_css_set;
4593 init_css_set_link.cg = &init_css_set;
4594 init_css_set_link.cgrp = dummytop;
4595 list_add(&init_css_set_link.cgrp_link_list,
4596 &rootnode.top_cgroup.css_sets);
4597 list_add(&init_css_set_link.cg_link_list,
4598 &init_css_set.cg_links);
4600 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4601 struct cgroup_subsys *ss = subsys[i];
4603 /* at bootup time, we don't worry about modular subsystems */
4604 if (!ss || ss->module)
4608 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4609 BUG_ON(!ss->css_alloc);
4610 BUG_ON(!ss->css_free);
4611 if (ss->subsys_id != i) {
4612 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4613 ss->name, ss->subsys_id);
4618 cgroup_init_subsys(ss);
4624 * cgroup_init - cgroup initialization
4626 * Register cgroup filesystem and /proc file, and initialize
4627 * any subsystems that didn't request early init.
4629 int __init cgroup_init(void)
4635 err = bdi_init(&cgroup_backing_dev_info);
4639 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4640 struct cgroup_subsys *ss = subsys[i];
4642 /* at bootup time, we don't worry about modular subsystems */
4643 if (!ss || ss->module)
4645 if (!ss->early_init)
4646 cgroup_init_subsys(ss);
4648 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4651 /* Add init_css_set to the hash table */
4652 key = css_set_hash(init_css_set.subsys);
4653 hash_add(css_set_table, &init_css_set.hlist, key);
4654 BUG_ON(!init_root_id(&rootnode));
4656 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4662 err = register_filesystem(&cgroup_fs_type);
4664 kobject_put(cgroup_kobj);
4668 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4672 bdi_destroy(&cgroup_backing_dev_info);
4678 * proc_cgroup_show()
4679 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4680 * - Used for /proc/<pid>/cgroup.
4681 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4682 * doesn't really matter if tsk->cgroup changes after we read it,
4683 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4684 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4685 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4686 * cgroup to top_cgroup.
4689 /* TODO: Use a proper seq_file iterator */
4690 static int proc_cgroup_show(struct seq_file *m, void *v)
4693 struct task_struct *tsk;
4696 struct cgroupfs_root *root;
4699 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4705 tsk = get_pid_task(pid, PIDTYPE_PID);
4711 mutex_lock(&cgroup_mutex);
4713 for_each_active_root(root) {
4714 struct cgroup_subsys *ss;
4715 struct cgroup *cgrp;
4718 seq_printf(m, "%d:", root->hierarchy_id);
4719 for_each_subsys(root, ss)
4720 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4721 if (strlen(root->name))
4722 seq_printf(m, "%sname=%s", count ? "," : "",
4725 cgrp = task_cgroup_from_root(tsk, root);
4726 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4734 mutex_unlock(&cgroup_mutex);
4735 put_task_struct(tsk);
4742 static int cgroup_open(struct inode *inode, struct file *file)
4744 struct pid *pid = PROC_I(inode)->pid;
4745 return single_open(file, proc_cgroup_show, pid);
4748 const struct file_operations proc_cgroup_operations = {
4749 .open = cgroup_open,
4751 .llseek = seq_lseek,
4752 .release = single_release,
4755 /* Display information about each subsystem and each hierarchy */
4756 static int proc_cgroupstats_show(struct seq_file *m, void *v)
4760 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4762 * ideally we don't want subsystems moving around while we do this.
4763 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4764 * subsys/hierarchy state.
4766 mutex_lock(&cgroup_mutex);
4767 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4768 struct cgroup_subsys *ss = subsys[i];
4771 seq_printf(m, "%s\t%d\t%d\t%d\n",
4772 ss->name, ss->root->hierarchy_id,
4773 ss->root->number_of_cgroups, !ss->disabled);
4775 mutex_unlock(&cgroup_mutex);
4779 static int cgroupstats_open(struct inode *inode, struct file *file)
4781 return single_open(file, proc_cgroupstats_show, NULL);
4784 static const struct file_operations proc_cgroupstats_operations = {
4785 .open = cgroupstats_open,
4787 .llseek = seq_lseek,
4788 .release = single_release,
4792 * cgroup_fork - attach newly forked task to its parents cgroup.
4793 * @child: pointer to task_struct of forking parent process.
4795 * Description: A task inherits its parent's cgroup at fork().
4797 * A pointer to the shared css_set was automatically copied in
4798 * fork.c by dup_task_struct(). However, we ignore that copy, since
4799 * it was not made under the protection of RCU or cgroup_mutex, so
4800 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4801 * have already changed current->cgroups, allowing the previously
4802 * referenced cgroup group to be removed and freed.
4804 * At the point that cgroup_fork() is called, 'current' is the parent
4805 * task, and the passed argument 'child' points to the child task.
4807 void cgroup_fork(struct task_struct *child)
4810 child->cgroups = current->cgroups;
4811 get_css_set(child->cgroups);
4812 task_unlock(current);
4813 INIT_LIST_HEAD(&child->cg_list);
4817 * cgroup_post_fork - called on a new task after adding it to the task list
4818 * @child: the task in question
4820 * Adds the task to the list running through its css_set if necessary and
4821 * call the subsystem fork() callbacks. Has to be after the task is
4822 * visible on the task list in case we race with the first call to
4823 * cgroup_iter_start() - to guarantee that the new task ends up on its
4826 void cgroup_post_fork(struct task_struct *child)
4831 * use_task_css_set_links is set to 1 before we walk the tasklist
4832 * under the tasklist_lock and we read it here after we added the child
4833 * to the tasklist under the tasklist_lock as well. If the child wasn't
4834 * yet in the tasklist when we walked through it from
4835 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4836 * should be visible now due to the paired locking and barriers implied
4837 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4838 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4841 if (use_task_css_set_links) {
4842 write_lock(&css_set_lock);
4844 if (list_empty(&child->cg_list))
4845 list_add(&child->cg_list, &child->cgroups->tasks);
4847 write_unlock(&css_set_lock);
4851 * Call ss->fork(). This must happen after @child is linked on
4852 * css_set; otherwise, @child might change state between ->fork()
4853 * and addition to css_set.
4855 if (need_forkexit_callback) {
4857 * fork/exit callbacks are supported only for builtin
4858 * subsystems, and the builtin section of the subsys
4859 * array is immutable, so we don't need to lock the
4860 * subsys array here. On the other hand, modular section
4861 * of the array can be freed at module unload, so we
4864 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4865 struct cgroup_subsys *ss = subsys[i];
4874 * cgroup_exit - detach cgroup from exiting task
4875 * @tsk: pointer to task_struct of exiting process
4876 * @run_callback: run exit callbacks?
4878 * Description: Detach cgroup from @tsk and release it.
4880 * Note that cgroups marked notify_on_release force every task in
4881 * them to take the global cgroup_mutex mutex when exiting.
4882 * This could impact scaling on very large systems. Be reluctant to
4883 * use notify_on_release cgroups where very high task exit scaling
4884 * is required on large systems.
4886 * the_top_cgroup_hack:
4888 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4890 * We call cgroup_exit() while the task is still competent to
4891 * handle notify_on_release(), then leave the task attached to the
4892 * root cgroup in each hierarchy for the remainder of its exit.
4894 * To do this properly, we would increment the reference count on
4895 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4896 * code we would add a second cgroup function call, to drop that
4897 * reference. This would just create an unnecessary hot spot on
4898 * the top_cgroup reference count, to no avail.
4900 * Normally, holding a reference to a cgroup without bumping its
4901 * count is unsafe. The cgroup could go away, or someone could
4902 * attach us to a different cgroup, decrementing the count on
4903 * the first cgroup that we never incremented. But in this case,
4904 * top_cgroup isn't going away, and either task has PF_EXITING set,
4905 * which wards off any cgroup_attach_task() attempts, or task is a failed
4906 * fork, never visible to cgroup_attach_task.
4908 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4914 * Unlink from the css_set task list if necessary.
4915 * Optimistically check cg_list before taking
4918 if (!list_empty(&tsk->cg_list)) {
4919 write_lock(&css_set_lock);
4920 if (!list_empty(&tsk->cg_list))
4921 list_del_init(&tsk->cg_list);
4922 write_unlock(&css_set_lock);
4925 /* Reassign the task to the init_css_set. */
4928 tsk->cgroups = &init_css_set;
4930 if (run_callbacks && need_forkexit_callback) {
4932 * fork/exit callbacks are supported only for builtin
4933 * subsystems, see cgroup_post_fork() for details.
4935 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4936 struct cgroup_subsys *ss = subsys[i];
4939 struct cgroup *old_cgrp =
4940 rcu_dereference_raw(cg->subsys[i])->cgroup;
4941 struct cgroup *cgrp = task_cgroup(tsk, i);
4942 ss->exit(cgrp, old_cgrp, tsk);
4948 put_css_set_taskexit(cg);
4951 static void check_for_release(struct cgroup *cgrp)
4953 /* All of these checks rely on RCU to keep the cgroup
4954 * structure alive */
4955 if (cgroup_is_releasable(cgrp) &&
4956 !atomic_read(&cgrp->count) && list_empty(&cgrp->children)) {
4958 * Control Group is currently removeable. If it's not
4959 * already queued for a userspace notification, queue
4962 int need_schedule_work = 0;
4964 raw_spin_lock(&release_list_lock);
4965 if (!cgroup_is_removed(cgrp) &&
4966 list_empty(&cgrp->release_list)) {
4967 list_add(&cgrp->release_list, &release_list);
4968 need_schedule_work = 1;
4970 raw_spin_unlock(&release_list_lock);
4971 if (need_schedule_work)
4972 schedule_work(&release_agent_work);
4976 /* Caller must verify that the css is not for root cgroup */
4977 bool __css_tryget(struct cgroup_subsys_state *css)
4982 v = css_refcnt(css);
4983 t = atomic_cmpxchg(&css->refcnt, v, v + 1);
4991 EXPORT_SYMBOL_GPL(__css_tryget);
4993 /* Caller must verify that the css is not for root cgroup */
4994 void __css_put(struct cgroup_subsys_state *css)
4998 v = css_unbias_refcnt(atomic_dec_return(&css->refcnt));
5000 schedule_work(&css->dput_work);
5002 EXPORT_SYMBOL_GPL(__css_put);
5005 * Notify userspace when a cgroup is released, by running the
5006 * configured release agent with the name of the cgroup (path
5007 * relative to the root of cgroup file system) as the argument.
5009 * Most likely, this user command will try to rmdir this cgroup.
5011 * This races with the possibility that some other task will be
5012 * attached to this cgroup before it is removed, or that some other
5013 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5014 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5015 * unused, and this cgroup will be reprieved from its death sentence,
5016 * to continue to serve a useful existence. Next time it's released,
5017 * we will get notified again, if it still has 'notify_on_release' set.
5019 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5020 * means only wait until the task is successfully execve()'d. The
5021 * separate release agent task is forked by call_usermodehelper(),
5022 * then control in this thread returns here, without waiting for the
5023 * release agent task. We don't bother to wait because the caller of
5024 * this routine has no use for the exit status of the release agent
5025 * task, so no sense holding our caller up for that.
5027 static void cgroup_release_agent(struct work_struct *work)
5029 BUG_ON(work != &release_agent_work);
5030 mutex_lock(&cgroup_mutex);
5031 raw_spin_lock(&release_list_lock);
5032 while (!list_empty(&release_list)) {
5033 char *argv[3], *envp[3];
5035 char *pathbuf = NULL, *agentbuf = NULL;
5036 struct cgroup *cgrp = list_entry(release_list.next,
5039 list_del_init(&cgrp->release_list);
5040 raw_spin_unlock(&release_list_lock);
5041 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
5044 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
5046 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
5051 argv[i++] = agentbuf;
5052 argv[i++] = pathbuf;
5056 /* minimal command environment */
5057 envp[i++] = "HOME=/";
5058 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5061 /* Drop the lock while we invoke the usermode helper,
5062 * since the exec could involve hitting disk and hence
5063 * be a slow process */
5064 mutex_unlock(&cgroup_mutex);
5065 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
5066 mutex_lock(&cgroup_mutex);
5070 raw_spin_lock(&release_list_lock);
5072 raw_spin_unlock(&release_list_lock);
5073 mutex_unlock(&cgroup_mutex);
5076 static int __init cgroup_disable(char *str)
5081 while ((token = strsep(&str, ",")) != NULL) {
5084 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
5085 struct cgroup_subsys *ss = subsys[i];
5088 * cgroup_disable, being at boot time, can't
5089 * know about module subsystems, so we don't
5092 if (!ss || ss->module)
5095 if (!strcmp(token, ss->name)) {
5097 printk(KERN_INFO "Disabling %s control group"
5098 " subsystem\n", ss->name);
5105 __setup("cgroup_disable=", cgroup_disable);
5108 * Functons for CSS ID.
5112 *To get ID other than 0, this should be called when !cgroup_is_removed().
5114 unsigned short css_id(struct cgroup_subsys_state *css)
5116 struct css_id *cssid;
5119 * This css_id() can return correct value when somone has refcnt
5120 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5121 * it's unchanged until freed.
5123 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5129 EXPORT_SYMBOL_GPL(css_id);
5131 unsigned short css_depth(struct cgroup_subsys_state *css)
5133 struct css_id *cssid;
5135 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5138 return cssid->depth;
5141 EXPORT_SYMBOL_GPL(css_depth);
5144 * css_is_ancestor - test "root" css is an ancestor of "child"
5145 * @child: the css to be tested.
5146 * @root: the css supporsed to be an ancestor of the child.
5148 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5149 * this function reads css->id, the caller must hold rcu_read_lock().
5150 * But, considering usual usage, the csses should be valid objects after test.
5151 * Assuming that the caller will do some action to the child if this returns
5152 * returns true, the caller must take "child";s reference count.
5153 * If "child" is valid object and this returns true, "root" is valid, too.
5156 bool css_is_ancestor(struct cgroup_subsys_state *child,
5157 const struct cgroup_subsys_state *root)
5159 struct css_id *child_id;
5160 struct css_id *root_id;
5162 child_id = rcu_dereference(child->id);
5165 root_id = rcu_dereference(root->id);
5168 if (child_id->depth < root_id->depth)
5170 if (child_id->stack[root_id->depth] != root_id->id)
5175 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
5177 struct css_id *id = css->id;
5178 /* When this is called before css_id initialization, id can be NULL */
5182 BUG_ON(!ss->use_id);
5184 rcu_assign_pointer(id->css, NULL);
5185 rcu_assign_pointer(css->id, NULL);
5186 spin_lock(&ss->id_lock);
5187 idr_remove(&ss->idr, id->id);
5188 spin_unlock(&ss->id_lock);
5189 kfree_rcu(id, rcu_head);
5191 EXPORT_SYMBOL_GPL(free_css_id);
5194 * This is called by init or create(). Then, calls to this function are
5195 * always serialized (By cgroup_mutex() at create()).
5198 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
5200 struct css_id *newid;
5203 BUG_ON(!ss->use_id);
5205 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
5206 newid = kzalloc(size, GFP_KERNEL);
5208 return ERR_PTR(-ENOMEM);
5210 idr_preload(GFP_KERNEL);
5211 spin_lock(&ss->id_lock);
5212 /* Don't use 0. allocates an ID of 1-65535 */
5213 ret = idr_alloc(&ss->idr, newid, 1, CSS_ID_MAX + 1, GFP_NOWAIT);
5214 spin_unlock(&ss->id_lock);
5217 /* Returns error when there are no free spaces for new ID.*/
5222 newid->depth = depth;
5226 return ERR_PTR(ret);
5230 static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
5231 struct cgroup_subsys_state *rootcss)
5233 struct css_id *newid;
5235 spin_lock_init(&ss->id_lock);
5238 newid = get_new_cssid(ss, 0);
5240 return PTR_ERR(newid);
5242 newid->stack[0] = newid->id;
5243 newid->css = rootcss;
5244 rootcss->id = newid;
5248 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
5249 struct cgroup *child)
5251 int subsys_id, i, depth = 0;
5252 struct cgroup_subsys_state *parent_css, *child_css;
5253 struct css_id *child_id, *parent_id;
5255 subsys_id = ss->subsys_id;
5256 parent_css = parent->subsys[subsys_id];
5257 child_css = child->subsys[subsys_id];
5258 parent_id = parent_css->id;
5259 depth = parent_id->depth + 1;
5261 child_id = get_new_cssid(ss, depth);
5262 if (IS_ERR(child_id))
5263 return PTR_ERR(child_id);
5265 for (i = 0; i < depth; i++)
5266 child_id->stack[i] = parent_id->stack[i];
5267 child_id->stack[depth] = child_id->id;
5269 * child_id->css pointer will be set after this cgroup is available
5270 * see cgroup_populate_dir()
5272 rcu_assign_pointer(child_css->id, child_id);
5278 * css_lookup - lookup css by id
5279 * @ss: cgroup subsys to be looked into.
5282 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5283 * NULL if not. Should be called under rcu_read_lock()
5285 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5287 struct css_id *cssid = NULL;
5289 BUG_ON(!ss->use_id);
5290 cssid = idr_find(&ss->idr, id);
5292 if (unlikely(!cssid))
5295 return rcu_dereference(cssid->css);
5297 EXPORT_SYMBOL_GPL(css_lookup);
5300 * css_get_next - lookup next cgroup under specified hierarchy.
5301 * @ss: pointer to subsystem
5302 * @id: current position of iteration.
5303 * @root: pointer to css. search tree under this.
5304 * @foundid: position of found object.
5306 * Search next css under the specified hierarchy of rootid. Calling under
5307 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5309 struct cgroup_subsys_state *
5310 css_get_next(struct cgroup_subsys *ss, int id,
5311 struct cgroup_subsys_state *root, int *foundid)
5313 struct cgroup_subsys_state *ret = NULL;
5316 int rootid = css_id(root);
5317 int depth = css_depth(root);
5322 BUG_ON(!ss->use_id);
5323 WARN_ON_ONCE(!rcu_read_lock_held());
5325 /* fill start point for scan */
5329 * scan next entry from bitmap(tree), tmpid is updated after
5332 tmp = idr_get_next(&ss->idr, &tmpid);
5335 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
5336 ret = rcu_dereference(tmp->css);
5342 /* continue to scan from next id */
5349 * get corresponding css from file open on cgroupfs directory
5351 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5353 struct cgroup *cgrp;
5354 struct inode *inode;
5355 struct cgroup_subsys_state *css;
5357 inode = file_inode(f);
5358 /* check in cgroup filesystem dir */
5359 if (inode->i_op != &cgroup_dir_inode_operations)
5360 return ERR_PTR(-EBADF);
5362 if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5363 return ERR_PTR(-EINVAL);
5366 cgrp = __d_cgrp(f->f_dentry);
5367 css = cgrp->subsys[id];
5368 return css ? css : ERR_PTR(-ENOENT);
5371 #ifdef CONFIG_CGROUP_DEBUG
5372 static struct cgroup_subsys_state *debug_css_alloc(struct cgroup *cont)
5374 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5377 return ERR_PTR(-ENOMEM);
5382 static void debug_css_free(struct cgroup *cont)
5384 kfree(cont->subsys[debug_subsys_id]);
5387 static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
5389 return atomic_read(&cont->count);
5392 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
5394 return cgroup_task_count(cont);
5397 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
5399 return (u64)(unsigned long)current->cgroups;
5402 static u64 current_css_set_refcount_read(struct cgroup *cont,
5408 count = atomic_read(¤t->cgroups->refcount);
5413 static int current_css_set_cg_links_read(struct cgroup *cont,
5415 struct seq_file *seq)
5417 struct cg_cgroup_link *link;
5420 read_lock(&css_set_lock);
5422 cg = rcu_dereference(current->cgroups);
5423 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
5424 struct cgroup *c = link->cgrp;
5428 name = c->dentry->d_name.name;
5431 seq_printf(seq, "Root %d group %s\n",
5432 c->root->hierarchy_id, name);
5435 read_unlock(&css_set_lock);
5439 #define MAX_TASKS_SHOWN_PER_CSS 25
5440 static int cgroup_css_links_read(struct cgroup *cont,
5442 struct seq_file *seq)
5444 struct cg_cgroup_link *link;
5446 read_lock(&css_set_lock);
5447 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
5448 struct css_set *cg = link->cg;
5449 struct task_struct *task;
5451 seq_printf(seq, "css_set %p\n", cg);
5452 list_for_each_entry(task, &cg->tasks, cg_list) {
5453 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5454 seq_puts(seq, " ...\n");
5457 seq_printf(seq, " task %d\n",
5458 task_pid_vnr(task));
5462 read_unlock(&css_set_lock);
5466 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5468 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5471 static struct cftype debug_files[] = {
5473 .name = "cgroup_refcount",
5474 .read_u64 = cgroup_refcount_read,
5477 .name = "taskcount",
5478 .read_u64 = debug_taskcount_read,
5482 .name = "current_css_set",
5483 .read_u64 = current_css_set_read,
5487 .name = "current_css_set_refcount",
5488 .read_u64 = current_css_set_refcount_read,
5492 .name = "current_css_set_cg_links",
5493 .read_seq_string = current_css_set_cg_links_read,
5497 .name = "cgroup_css_links",
5498 .read_seq_string = cgroup_css_links_read,
5502 .name = "releasable",
5503 .read_u64 = releasable_read,
5509 struct cgroup_subsys debug_subsys = {
5511 .css_alloc = debug_css_alloc,
5512 .css_free = debug_css_free,
5513 .subsys_id = debug_subsys_id,
5514 .base_cftypes = debug_files,
5516 #endif /* CONFIG_CGROUP_DEBUG */