2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
11 * Copyright notices from the original cpuset code:
12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
19 * 2003-10-10 Written by Simon Derr.
20 * 2003-10-22 Updates by Stephen Hemminger.
21 * 2004 May-July Rework by Paul Jackson.
22 * ---------------------------------------------------
24 * This file is subject to the terms and conditions of the GNU General Public
25 * License. See the file COPYING in the main directory of the Linux
26 * distribution for more details.
29 #include <linux/cgroup.h>
30 #include <linux/cred.h>
31 #include <linux/ctype.h>
32 #include <linux/errno.h>
33 #include <linux/init_task.h>
34 #include <linux/kernel.h>
35 #include <linux/list.h>
37 #include <linux/mutex.h>
38 #include <linux/mount.h>
39 #include <linux/pagemap.h>
40 #include <linux/proc_fs.h>
41 #include <linux/rcupdate.h>
42 #include <linux/sched.h>
43 #include <linux/backing-dev.h>
44 #include <linux/seq_file.h>
45 #include <linux/slab.h>
46 #include <linux/magic.h>
47 #include <linux/spinlock.h>
48 #include <linux/string.h>
49 #include <linux/sort.h>
50 #include <linux/kmod.h>
51 #include <linux/module.h>
52 #include <linux/delayacct.h>
53 #include <linux/cgroupstats.h>
54 #include <linux/hashtable.h>
55 #include <linux/namei.h>
56 #include <linux/pid_namespace.h>
57 #include <linux/idr.h>
58 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
59 #include <linux/eventfd.h>
60 #include <linux/poll.h>
61 #include <linux/flex_array.h> /* used in cgroup_attach_task */
62 #include <linux/kthread.h>
64 #include <linux/atomic.h>
67 * cgroup_mutex is the master lock. Any modification to cgroup or its
68 * hierarchy must be performed while holding it.
70 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
71 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
72 * release_agent_path and so on. Modifying requires both cgroup_mutex and
73 * cgroup_root_mutex. Readers can acquire either of the two. This is to
74 * break the following locking order cycle.
76 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
77 * B. namespace_sem -> cgroup_mutex
79 * B happens only through cgroup_show_options() and using cgroup_root_mutex
82 #ifdef CONFIG_PROVE_RCU
83 DEFINE_MUTEX(cgroup_mutex);
84 EXPORT_SYMBOL_GPL(cgroup_mutex); /* only for task_subsys_state_check() */
86 static DEFINE_MUTEX(cgroup_mutex);
89 static DEFINE_MUTEX(cgroup_root_mutex);
92 * Generate an array of cgroup subsystem pointers. At boot time, this is
93 * populated with the built in subsystems, and modular subsystems are
94 * registered after that. The mutable section of this array is protected by
97 #define SUBSYS(_x) [_x ## _subsys_id] = &_x ## _subsys,
98 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
99 static struct cgroup_subsys *cgroup_subsys[CGROUP_SUBSYS_COUNT] = {
100 #include <linux/cgroup_subsys.h>
104 * The dummy hierarchy, reserved for the subsystems that are otherwise
105 * unattached - it never has more than a single cgroup, and all tasks are
106 * part of that cgroup.
108 static struct cgroupfs_root cgroup_dummy_root;
110 /* dummy_top is a shorthand for the dummy hierarchy's top cgroup */
111 static struct cgroup * const cgroup_dummy_top = &cgroup_dummy_root.top_cgroup;
114 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
117 struct list_head node;
118 struct dentry *dentry;
122 struct simple_xattrs xattrs;
126 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
127 * cgroup_subsys->use_id != 0.
129 #define CSS_ID_MAX (65535)
132 * The css to which this ID points. This pointer is set to valid value
133 * after cgroup is populated. If cgroup is removed, this will be NULL.
134 * This pointer is expected to be RCU-safe because destroy()
135 * is called after synchronize_rcu(). But for safe use, css_tryget()
136 * should be used for avoiding race.
138 struct cgroup_subsys_state __rcu *css;
144 * Depth in hierarchy which this ID belongs to.
146 unsigned short depth;
148 * ID is freed by RCU. (and lookup routine is RCU safe.)
150 struct rcu_head rcu_head;
152 * Hierarchy of CSS ID belongs to.
154 unsigned short stack[0]; /* Array of Length (depth+1) */
158 * cgroup_event represents events which userspace want to receive.
160 struct cgroup_event {
162 * Cgroup which the event belongs to.
166 * Control file which the event associated.
170 * eventfd to signal userspace about the event.
172 struct eventfd_ctx *eventfd;
174 * Each of these stored in a list by the cgroup.
176 struct list_head list;
178 * All fields below needed to unregister event when
179 * userspace closes eventfd.
182 wait_queue_head_t *wqh;
184 struct work_struct remove;
187 /* The list of hierarchy roots */
189 static LIST_HEAD(cgroup_roots);
190 static int cgroup_root_count;
193 * Hierarchy ID allocation and mapping. It follows the same exclusion
194 * rules as other root ops - both cgroup_mutex and cgroup_root_mutex for
195 * writes, either for reads.
197 static DEFINE_IDR(cgroup_hierarchy_idr);
199 static struct cgroup_name root_cgroup_name = { .name = "/" };
202 * Assign a monotonically increasing serial number to cgroups. It
203 * guarantees cgroups with bigger numbers are newer than those with smaller
204 * numbers. Also, as cgroups are always appended to the parent's
205 * ->children list, it guarantees that sibling cgroups are always sorted in
206 * the ascending serial number order on the list. Protected by
209 static u64 cgroup_serial_nr_next = 1;
211 /* This flag indicates whether tasks in the fork and exit paths should
212 * check for fork/exit handlers to call. This avoids us having to do
213 * extra work in the fork/exit path if none of the subsystems need to
216 static int need_forkexit_callback __read_mostly;
218 static void cgroup_offline_fn(struct work_struct *work);
219 static int cgroup_destroy_locked(struct cgroup *cgrp);
220 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
221 struct cftype cfts[], bool is_add);
223 /* convenient tests for these bits */
224 static inline bool cgroup_is_dead(const struct cgroup *cgrp)
226 return test_bit(CGRP_DEAD, &cgrp->flags);
230 * cgroup_is_descendant - test ancestry
231 * @cgrp: the cgroup to be tested
232 * @ancestor: possible ancestor of @cgrp
234 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
235 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
236 * and @ancestor are accessible.
238 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor)
241 if (cgrp == ancestor)
247 EXPORT_SYMBOL_GPL(cgroup_is_descendant);
249 static int cgroup_is_releasable(const struct cgroup *cgrp)
252 (1 << CGRP_RELEASABLE) |
253 (1 << CGRP_NOTIFY_ON_RELEASE);
254 return (cgrp->flags & bits) == bits;
257 static int notify_on_release(const struct cgroup *cgrp)
259 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
263 * for_each_subsys - iterate all loaded cgroup subsystems
264 * @ss: the iteration cursor
265 * @i: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
267 * Should be called under cgroup_mutex.
269 #define for_each_subsys(ss, i) \
270 for ((i) = 0; (i) < CGROUP_SUBSYS_COUNT; (i)++) \
271 if (({ lockdep_assert_held(&cgroup_mutex); \
272 !((ss) = cgroup_subsys[i]); })) { } \
276 * for_each_builtin_subsys - iterate all built-in cgroup subsystems
277 * @ss: the iteration cursor
278 * @i: the index of @ss, CGROUP_BUILTIN_SUBSYS_COUNT after reaching the end
280 * Bulit-in subsystems are always present and iteration itself doesn't
281 * require any synchronization.
283 #define for_each_builtin_subsys(ss, i) \
284 for ((i) = 0; (i) < CGROUP_BUILTIN_SUBSYS_COUNT && \
285 (((ss) = cgroup_subsys[i]) || true); (i)++)
287 /* iterate each subsystem attached to a hierarchy */
288 #define for_each_root_subsys(root, ss) \
289 list_for_each_entry((ss), &(root)->subsys_list, sibling)
291 /* iterate across the active hierarchies */
292 #define for_each_active_root(root) \
293 list_for_each_entry((root), &cgroup_roots, root_list)
295 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
297 return dentry->d_fsdata;
300 static inline struct cfent *__d_cfe(struct dentry *dentry)
302 return dentry->d_fsdata;
305 static inline struct cftype *__d_cft(struct dentry *dentry)
307 return __d_cfe(dentry)->type;
311 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
312 * @cgrp: the cgroup to be checked for liveness
314 * On success, returns true; the mutex should be later unlocked. On
315 * failure returns false with no lock held.
317 static bool cgroup_lock_live_group(struct cgroup *cgrp)
319 mutex_lock(&cgroup_mutex);
320 if (cgroup_is_dead(cgrp)) {
321 mutex_unlock(&cgroup_mutex);
327 /* the list of cgroups eligible for automatic release. Protected by
328 * release_list_lock */
329 static LIST_HEAD(release_list);
330 static DEFINE_RAW_SPINLOCK(release_list_lock);
331 static void cgroup_release_agent(struct work_struct *work);
332 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
333 static void check_for_release(struct cgroup *cgrp);
336 * A cgroup can be associated with multiple css_sets as different tasks may
337 * belong to different cgroups on different hierarchies. In the other
338 * direction, a css_set is naturally associated with multiple cgroups.
339 * This M:N relationship is represented by the following link structure
340 * which exists for each association and allows traversing the associations
343 struct cgrp_cset_link {
344 /* the cgroup and css_set this link associates */
346 struct css_set *cset;
348 /* list of cgrp_cset_links anchored at cgrp->cset_links */
349 struct list_head cset_link;
351 /* list of cgrp_cset_links anchored at css_set->cgrp_links */
352 struct list_head cgrp_link;
355 /* The default css_set - used by init and its children prior to any
356 * hierarchies being mounted. It contains a pointer to the root state
357 * for each subsystem. Also used to anchor the list of css_sets. Not
358 * reference-counted, to improve performance when child cgroups
359 * haven't been created.
362 static struct css_set init_css_set;
363 static struct cgrp_cset_link init_cgrp_cset_link;
365 static int cgroup_init_idr(struct cgroup_subsys *ss,
366 struct cgroup_subsys_state *css);
368 /* css_set_lock protects the list of css_set objects, and the
369 * chain of tasks off each css_set. Nests outside task->alloc_lock
370 * due to cgroup_iter_start() */
371 static DEFINE_RWLOCK(css_set_lock);
372 static int css_set_count;
375 * hash table for cgroup groups. This improves the performance to find
376 * an existing css_set. This hash doesn't (currently) take into
377 * account cgroups in empty hierarchies.
379 #define CSS_SET_HASH_BITS 7
380 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
382 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
384 unsigned long key = 0UL;
385 struct cgroup_subsys *ss;
388 for_each_subsys(ss, i)
389 key += (unsigned long)css[i];
390 key = (key >> 16) ^ key;
395 /* We don't maintain the lists running through each css_set to its
396 * task until after the first call to cgroup_iter_start(). This
397 * reduces the fork()/exit() overhead for people who have cgroups
398 * compiled into their kernel but not actually in use */
399 static int use_task_css_set_links __read_mostly;
401 static void __put_css_set(struct css_set *cset, int taskexit)
403 struct cgrp_cset_link *link, *tmp_link;
406 * Ensure that the refcount doesn't hit zero while any readers
407 * can see it. Similar to atomic_dec_and_lock(), but for an
410 if (atomic_add_unless(&cset->refcount, -1, 1))
412 write_lock(&css_set_lock);
413 if (!atomic_dec_and_test(&cset->refcount)) {
414 write_unlock(&css_set_lock);
418 /* This css_set is dead. unlink it and release cgroup refcounts */
419 hash_del(&cset->hlist);
422 list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
423 struct cgroup *cgrp = link->cgrp;
425 list_del(&link->cset_link);
426 list_del(&link->cgrp_link);
428 /* @cgrp can't go away while we're holding css_set_lock */
429 if (list_empty(&cgrp->cset_links) && notify_on_release(cgrp)) {
431 set_bit(CGRP_RELEASABLE, &cgrp->flags);
432 check_for_release(cgrp);
438 write_unlock(&css_set_lock);
439 kfree_rcu(cset, rcu_head);
443 * refcounted get/put for css_set objects
445 static inline void get_css_set(struct css_set *cset)
447 atomic_inc(&cset->refcount);
450 static inline void put_css_set(struct css_set *cset)
452 __put_css_set(cset, 0);
455 static inline void put_css_set_taskexit(struct css_set *cset)
457 __put_css_set(cset, 1);
461 * compare_css_sets - helper function for find_existing_css_set().
462 * @cset: candidate css_set being tested
463 * @old_cset: existing css_set for a task
464 * @new_cgrp: cgroup that's being entered by the task
465 * @template: desired set of css pointers in css_set (pre-calculated)
467 * Returns true if "cg" matches "old_cg" except for the hierarchy
468 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
470 static bool compare_css_sets(struct css_set *cset,
471 struct css_set *old_cset,
472 struct cgroup *new_cgrp,
473 struct cgroup_subsys_state *template[])
475 struct list_head *l1, *l2;
477 if (memcmp(template, cset->subsys, sizeof(cset->subsys))) {
478 /* Not all subsystems matched */
483 * Compare cgroup pointers in order to distinguish between
484 * different cgroups in heirarchies with no subsystems. We
485 * could get by with just this check alone (and skip the
486 * memcmp above) but on most setups the memcmp check will
487 * avoid the need for this more expensive check on almost all
491 l1 = &cset->cgrp_links;
492 l2 = &old_cset->cgrp_links;
494 struct cgrp_cset_link *link1, *link2;
495 struct cgroup *cgrp1, *cgrp2;
499 /* See if we reached the end - both lists are equal length. */
500 if (l1 == &cset->cgrp_links) {
501 BUG_ON(l2 != &old_cset->cgrp_links);
504 BUG_ON(l2 == &old_cset->cgrp_links);
506 /* Locate the cgroups associated with these links. */
507 link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
508 link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
511 /* Hierarchies should be linked in the same order. */
512 BUG_ON(cgrp1->root != cgrp2->root);
515 * If this hierarchy is the hierarchy of the cgroup
516 * that's changing, then we need to check that this
517 * css_set points to the new cgroup; if it's any other
518 * hierarchy, then this css_set should point to the
519 * same cgroup as the old css_set.
521 if (cgrp1->root == new_cgrp->root) {
522 if (cgrp1 != new_cgrp)
533 * find_existing_css_set - init css array and find the matching css_set
534 * @old_cset: the css_set that we're using before the cgroup transition
535 * @cgrp: the cgroup that we're moving into
536 * @template: out param for the new set of csses, should be clear on entry
538 static struct css_set *find_existing_css_set(struct css_set *old_cset,
540 struct cgroup_subsys_state *template[])
542 struct cgroupfs_root *root = cgrp->root;
543 struct cgroup_subsys *ss;
544 struct css_set *cset;
549 * Build the set of subsystem state objects that we want to see in the
550 * new css_set. while subsystems can change globally, the entries here
551 * won't change, so no need for locking.
553 for_each_subsys(ss, i) {
554 if (root->subsys_mask & (1UL << i)) {
555 /* Subsystem is in this hierarchy. So we want
556 * the subsystem state from the new
558 template[i] = cgrp->subsys[i];
560 /* Subsystem is not in this hierarchy, so we
561 * don't want to change the subsystem state */
562 template[i] = old_cset->subsys[i];
566 key = css_set_hash(template);
567 hash_for_each_possible(css_set_table, cset, hlist, key) {
568 if (!compare_css_sets(cset, old_cset, cgrp, template))
571 /* This css_set matches what we need */
575 /* No existing cgroup group matched */
579 static void free_cgrp_cset_links(struct list_head *links_to_free)
581 struct cgrp_cset_link *link, *tmp_link;
583 list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
584 list_del(&link->cset_link);
590 * allocate_cgrp_cset_links - allocate cgrp_cset_links
591 * @count: the number of links to allocate
592 * @tmp_links: list_head the allocated links are put on
594 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
595 * through ->cset_link. Returns 0 on success or -errno.
597 static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
599 struct cgrp_cset_link *link;
602 INIT_LIST_HEAD(tmp_links);
604 for (i = 0; i < count; i++) {
605 link = kzalloc(sizeof(*link), GFP_KERNEL);
607 free_cgrp_cset_links(tmp_links);
610 list_add(&link->cset_link, tmp_links);
616 * link_css_set - a helper function to link a css_set to a cgroup
617 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
618 * @cset: the css_set to be linked
619 * @cgrp: the destination cgroup
621 static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
624 struct cgrp_cset_link *link;
626 BUG_ON(list_empty(tmp_links));
627 link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
630 list_move(&link->cset_link, &cgrp->cset_links);
632 * Always add links to the tail of the list so that the list
633 * is sorted by order of hierarchy creation
635 list_add_tail(&link->cgrp_link, &cset->cgrp_links);
639 * find_css_set - return a new css_set with one cgroup updated
640 * @old_cset: the baseline css_set
641 * @cgrp: the cgroup to be updated
643 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
644 * substituted into the appropriate hierarchy.
646 static struct css_set *find_css_set(struct css_set *old_cset,
649 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
650 struct css_set *cset;
651 struct list_head tmp_links;
652 struct cgrp_cset_link *link;
655 lockdep_assert_held(&cgroup_mutex);
657 /* First see if we already have a cgroup group that matches
659 read_lock(&css_set_lock);
660 cset = find_existing_css_set(old_cset, cgrp, template);
663 read_unlock(&css_set_lock);
668 cset = kzalloc(sizeof(*cset), GFP_KERNEL);
672 /* Allocate all the cgrp_cset_link objects that we'll need */
673 if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
678 atomic_set(&cset->refcount, 1);
679 INIT_LIST_HEAD(&cset->cgrp_links);
680 INIT_LIST_HEAD(&cset->tasks);
681 INIT_HLIST_NODE(&cset->hlist);
683 /* Copy the set of subsystem state objects generated in
684 * find_existing_css_set() */
685 memcpy(cset->subsys, template, sizeof(cset->subsys));
687 write_lock(&css_set_lock);
688 /* Add reference counts and links from the new css_set. */
689 list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
690 struct cgroup *c = link->cgrp;
692 if (c->root == cgrp->root)
694 link_css_set(&tmp_links, cset, c);
697 BUG_ON(!list_empty(&tmp_links));
701 /* Add this cgroup group to the hash table */
702 key = css_set_hash(cset->subsys);
703 hash_add(css_set_table, &cset->hlist, key);
705 write_unlock(&css_set_lock);
711 * Return the cgroup for "task" from the given hierarchy. Must be
712 * called with cgroup_mutex held.
714 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
715 struct cgroupfs_root *root)
717 struct css_set *cset;
718 struct cgroup *res = NULL;
720 BUG_ON(!mutex_is_locked(&cgroup_mutex));
721 read_lock(&css_set_lock);
723 * No need to lock the task - since we hold cgroup_mutex the
724 * task can't change groups, so the only thing that can happen
725 * is that it exits and its css is set back to init_css_set.
727 cset = task->cgroups;
728 if (cset == &init_css_set) {
729 res = &root->top_cgroup;
731 struct cgrp_cset_link *link;
733 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
734 struct cgroup *c = link->cgrp;
736 if (c->root == root) {
742 read_unlock(&css_set_lock);
748 * There is one global cgroup mutex. We also require taking
749 * task_lock() when dereferencing a task's cgroup subsys pointers.
750 * See "The task_lock() exception", at the end of this comment.
752 * A task must hold cgroup_mutex to modify cgroups.
754 * Any task can increment and decrement the count field without lock.
755 * So in general, code holding cgroup_mutex can't rely on the count
756 * field not changing. However, if the count goes to zero, then only
757 * cgroup_attach_task() can increment it again. Because a count of zero
758 * means that no tasks are currently attached, therefore there is no
759 * way a task attached to that cgroup can fork (the other way to
760 * increment the count). So code holding cgroup_mutex can safely
761 * assume that if the count is zero, it will stay zero. Similarly, if
762 * a task holds cgroup_mutex on a cgroup with zero count, it
763 * knows that the cgroup won't be removed, as cgroup_rmdir()
766 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
767 * (usually) take cgroup_mutex. These are the two most performance
768 * critical pieces of code here. The exception occurs on cgroup_exit(),
769 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
770 * is taken, and if the cgroup count is zero, a usermode call made
771 * to the release agent with the name of the cgroup (path relative to
772 * the root of cgroup file system) as the argument.
774 * A cgroup can only be deleted if both its 'count' of using tasks
775 * is zero, and its list of 'children' cgroups is empty. Since all
776 * tasks in the system use _some_ cgroup, and since there is always at
777 * least one task in the system (init, pid == 1), therefore, top_cgroup
778 * always has either children cgroups and/or using tasks. So we don't
779 * need a special hack to ensure that top_cgroup cannot be deleted.
781 * The task_lock() exception
783 * The need for this exception arises from the action of
784 * cgroup_attach_task(), which overwrites one task's cgroup pointer with
785 * another. It does so using cgroup_mutex, however there are
786 * several performance critical places that need to reference
787 * task->cgroup without the expense of grabbing a system global
788 * mutex. Therefore except as noted below, when dereferencing or, as
789 * in cgroup_attach_task(), modifying a task's cgroup pointer we use
790 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
791 * the task_struct routinely used for such matters.
793 * P.S. One more locking exception. RCU is used to guard the
794 * update of a tasks cgroup pointer by cgroup_attach_task()
798 * A couple of forward declarations required, due to cyclic reference loop:
799 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
800 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
804 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
805 static struct dentry *cgroup_lookup(struct inode *, struct dentry *, unsigned int);
806 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
807 static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
808 unsigned long subsys_mask);
809 static const struct inode_operations cgroup_dir_inode_operations;
810 static const struct file_operations proc_cgroupstats_operations;
812 static struct backing_dev_info cgroup_backing_dev_info = {
814 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
817 static int alloc_css_id(struct cgroup_subsys *ss,
818 struct cgroup *parent, struct cgroup *child);
820 static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb)
822 struct inode *inode = new_inode(sb);
825 inode->i_ino = get_next_ino();
826 inode->i_mode = mode;
827 inode->i_uid = current_fsuid();
828 inode->i_gid = current_fsgid();
829 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
830 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
835 static struct cgroup_name *cgroup_alloc_name(struct dentry *dentry)
837 struct cgroup_name *name;
839 name = kmalloc(sizeof(*name) + dentry->d_name.len + 1, GFP_KERNEL);
842 strcpy(name->name, dentry->d_name.name);
846 static void cgroup_free_fn(struct work_struct *work)
848 struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work);
849 struct cgroup_subsys *ss;
851 mutex_lock(&cgroup_mutex);
853 * Release the subsystem state objects.
855 for_each_root_subsys(cgrp->root, ss)
858 cgrp->root->number_of_cgroups--;
859 mutex_unlock(&cgroup_mutex);
862 * We get a ref to the parent's dentry, and put the ref when
863 * this cgroup is being freed, so it's guaranteed that the
864 * parent won't be destroyed before its children.
866 dput(cgrp->parent->dentry);
868 ida_simple_remove(&cgrp->root->cgroup_ida, cgrp->id);
871 * Drop the active superblock reference that we took when we
872 * created the cgroup. This will free cgrp->root, if we are
873 * holding the last reference to @sb.
875 deactivate_super(cgrp->root->sb);
878 * if we're getting rid of the cgroup, refcount should ensure
879 * that there are no pidlists left.
881 BUG_ON(!list_empty(&cgrp->pidlists));
883 simple_xattrs_free(&cgrp->xattrs);
885 kfree(rcu_dereference_raw(cgrp->name));
889 static void cgroup_free_rcu(struct rcu_head *head)
891 struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head);
893 INIT_WORK(&cgrp->destroy_work, cgroup_free_fn);
894 schedule_work(&cgrp->destroy_work);
897 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
899 /* is dentry a directory ? if so, kfree() associated cgroup */
900 if (S_ISDIR(inode->i_mode)) {
901 struct cgroup *cgrp = dentry->d_fsdata;
903 BUG_ON(!(cgroup_is_dead(cgrp)));
904 call_rcu(&cgrp->rcu_head, cgroup_free_rcu);
906 struct cfent *cfe = __d_cfe(dentry);
907 struct cgroup *cgrp = dentry->d_parent->d_fsdata;
909 WARN_ONCE(!list_empty(&cfe->node) &&
910 cgrp != &cgrp->root->top_cgroup,
911 "cfe still linked for %s\n", cfe->type->name);
912 simple_xattrs_free(&cfe->xattrs);
918 static int cgroup_delete(const struct dentry *d)
923 static void remove_dir(struct dentry *d)
925 struct dentry *parent = dget(d->d_parent);
928 simple_rmdir(parent->d_inode, d);
932 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
936 lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex);
937 lockdep_assert_held(&cgroup_mutex);
940 * If we're doing cleanup due to failure of cgroup_create(),
941 * the corresponding @cfe may not exist.
943 list_for_each_entry(cfe, &cgrp->files, node) {
944 struct dentry *d = cfe->dentry;
946 if (cft && cfe->type != cft)
951 simple_unlink(cgrp->dentry->d_inode, d);
952 list_del_init(&cfe->node);
960 * cgroup_clear_directory - selective removal of base and subsystem files
961 * @dir: directory containing the files
962 * @base_files: true if the base files should be removed
963 * @subsys_mask: mask of the subsystem ids whose files should be removed
965 static void cgroup_clear_directory(struct dentry *dir, bool base_files,
966 unsigned long subsys_mask)
968 struct cgroup *cgrp = __d_cgrp(dir);
969 struct cgroup_subsys *ss;
971 for_each_root_subsys(cgrp->root, ss) {
972 struct cftype_set *set;
973 if (!test_bit(ss->subsys_id, &subsys_mask))
975 list_for_each_entry(set, &ss->cftsets, node)
976 cgroup_addrm_files(cgrp, NULL, set->cfts, false);
979 while (!list_empty(&cgrp->files))
980 cgroup_rm_file(cgrp, NULL);
985 * NOTE : the dentry must have been dget()'ed
987 static void cgroup_d_remove_dir(struct dentry *dentry)
989 struct dentry *parent;
990 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
992 cgroup_clear_directory(dentry, true, root->subsys_mask);
994 parent = dentry->d_parent;
995 spin_lock(&parent->d_lock);
996 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
997 list_del_init(&dentry->d_u.d_child);
998 spin_unlock(&dentry->d_lock);
999 spin_unlock(&parent->d_lock);
1004 * Call with cgroup_mutex held. Drops reference counts on modules, including
1005 * any duplicate ones that parse_cgroupfs_options took. If this function
1006 * returns an error, no reference counts are touched.
1008 static int rebind_subsystems(struct cgroupfs_root *root,
1009 unsigned long added_mask, unsigned removed_mask)
1011 struct cgroup *cgrp = &root->top_cgroup;
1012 struct cgroup_subsys *ss;
1015 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1016 BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
1018 /* Check that any added subsystems are currently free */
1019 for_each_subsys(ss, i) {
1020 unsigned long bit = 1UL << i;
1022 if (!(bit & added_mask))
1025 if (ss->root != &cgroup_dummy_root) {
1026 /* Subsystem isn't free */
1031 /* Currently we don't handle adding/removing subsystems when
1032 * any child cgroups exist. This is theoretically supportable
1033 * but involves complex error handling, so it's being left until
1035 if (root->number_of_cgroups > 1)
1038 /* Process each subsystem */
1039 for_each_subsys(ss, i) {
1040 unsigned long bit = 1UL << i;
1042 if (bit & added_mask) {
1043 /* We're binding this subsystem to this hierarchy */
1044 BUG_ON(cgrp->subsys[i]);
1045 BUG_ON(!cgroup_dummy_top->subsys[i]);
1046 BUG_ON(cgroup_dummy_top->subsys[i]->cgroup != cgroup_dummy_top);
1048 cgrp->subsys[i] = cgroup_dummy_top->subsys[i];
1049 cgrp->subsys[i]->cgroup = cgrp;
1050 list_move(&ss->sibling, &root->subsys_list);
1055 /* refcount was already taken, and we're keeping it */
1056 root->subsys_mask |= bit;
1057 } else if (bit & removed_mask) {
1058 /* We're removing this subsystem */
1059 BUG_ON(cgrp->subsys[i] != cgroup_dummy_top->subsys[i]);
1060 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1063 ss->bind(cgroup_dummy_top);
1064 cgroup_dummy_top->subsys[i]->cgroup = cgroup_dummy_top;
1065 cgrp->subsys[i] = NULL;
1066 cgroup_subsys[i]->root = &cgroup_dummy_root;
1067 list_move(&ss->sibling, &cgroup_dummy_root.subsys_list);
1069 /* subsystem is now free - drop reference on module */
1070 module_put(ss->module);
1071 root->subsys_mask &= ~bit;
1072 } else if (bit & root->subsys_mask) {
1073 /* Subsystem state should already exist */
1074 BUG_ON(!cgrp->subsys[i]);
1076 * a refcount was taken, but we already had one, so
1077 * drop the extra reference.
1079 module_put(ss->module);
1080 #ifdef CONFIG_MODULE_UNLOAD
1081 BUG_ON(ss->module && !module_refcount(ss->module));
1084 /* Subsystem state shouldn't exist */
1085 BUG_ON(cgrp->subsys[i]);
1092 static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1094 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1095 struct cgroup_subsys *ss;
1097 mutex_lock(&cgroup_root_mutex);
1098 for_each_root_subsys(root, ss)
1099 seq_printf(seq, ",%s", ss->name);
1100 if (root->flags & CGRP_ROOT_SANE_BEHAVIOR)
1101 seq_puts(seq, ",sane_behavior");
1102 if (root->flags & CGRP_ROOT_NOPREFIX)
1103 seq_puts(seq, ",noprefix");
1104 if (root->flags & CGRP_ROOT_XATTR)
1105 seq_puts(seq, ",xattr");
1106 if (strlen(root->release_agent_path))
1107 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1108 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags))
1109 seq_puts(seq, ",clone_children");
1110 if (strlen(root->name))
1111 seq_printf(seq, ",name=%s", root->name);
1112 mutex_unlock(&cgroup_root_mutex);
1116 struct cgroup_sb_opts {
1117 unsigned long subsys_mask;
1118 unsigned long flags;
1119 char *release_agent;
1120 bool cpuset_clone_children;
1122 /* User explicitly requested empty subsystem */
1125 struct cgroupfs_root *new_root;
1130 * Convert a hierarchy specifier into a bitmask of subsystems and
1131 * flags. Call with cgroup_mutex held to protect the cgroup_subsys[]
1132 * array. This function takes refcounts on subsystems to be used, unless it
1133 * returns error, in which case no refcounts are taken.
1135 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1137 char *token, *o = data;
1138 bool all_ss = false, one_ss = false;
1139 unsigned long mask = (unsigned long)-1;
1140 bool module_pin_failed = false;
1141 struct cgroup_subsys *ss;
1144 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1146 #ifdef CONFIG_CPUSETS
1147 mask = ~(1UL << cpuset_subsys_id);
1150 memset(opts, 0, sizeof(*opts));
1152 while ((token = strsep(&o, ",")) != NULL) {
1155 if (!strcmp(token, "none")) {
1156 /* Explicitly have no subsystems */
1160 if (!strcmp(token, "all")) {
1161 /* Mutually exclusive option 'all' + subsystem name */
1167 if (!strcmp(token, "__DEVEL__sane_behavior")) {
1168 opts->flags |= CGRP_ROOT_SANE_BEHAVIOR;
1171 if (!strcmp(token, "noprefix")) {
1172 opts->flags |= CGRP_ROOT_NOPREFIX;
1175 if (!strcmp(token, "clone_children")) {
1176 opts->cpuset_clone_children = true;
1179 if (!strcmp(token, "xattr")) {
1180 opts->flags |= CGRP_ROOT_XATTR;
1183 if (!strncmp(token, "release_agent=", 14)) {
1184 /* Specifying two release agents is forbidden */
1185 if (opts->release_agent)
1187 opts->release_agent =
1188 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1189 if (!opts->release_agent)
1193 if (!strncmp(token, "name=", 5)) {
1194 const char *name = token + 5;
1195 /* Can't specify an empty name */
1198 /* Must match [\w.-]+ */
1199 for (i = 0; i < strlen(name); i++) {
1203 if ((c == '.') || (c == '-') || (c == '_'))
1207 /* Specifying two names is forbidden */
1210 opts->name = kstrndup(name,
1211 MAX_CGROUP_ROOT_NAMELEN - 1,
1219 for_each_subsys(ss, i) {
1220 if (strcmp(token, ss->name))
1225 /* Mutually exclusive option 'all' + subsystem name */
1228 set_bit(i, &opts->subsys_mask);
1233 if (i == CGROUP_SUBSYS_COUNT)
1238 * If the 'all' option was specified select all the subsystems,
1239 * otherwise if 'none', 'name=' and a subsystem name options
1240 * were not specified, let's default to 'all'
1242 if (all_ss || (!one_ss && !opts->none && !opts->name))
1243 for_each_subsys(ss, i)
1245 set_bit(i, &opts->subsys_mask);
1247 /* Consistency checks */
1249 if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) {
1250 pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1252 if (opts->flags & CGRP_ROOT_NOPREFIX) {
1253 pr_err("cgroup: sane_behavior: noprefix is not allowed\n");
1257 if (opts->cpuset_clone_children) {
1258 pr_err("cgroup: sane_behavior: clone_children is not allowed\n");
1264 * Option noprefix was introduced just for backward compatibility
1265 * with the old cpuset, so we allow noprefix only if mounting just
1266 * the cpuset subsystem.
1268 if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask))
1272 /* Can't specify "none" and some subsystems */
1273 if (opts->subsys_mask && opts->none)
1277 * We either have to specify by name or by subsystems. (So all
1278 * empty hierarchies must have a name).
1280 if (!opts->subsys_mask && !opts->name)
1284 * Grab references on all the modules we'll need, so the subsystems
1285 * don't dance around before rebind_subsystems attaches them. This may
1286 * take duplicate reference counts on a subsystem that's already used,
1287 * but rebind_subsystems handles this case.
1289 for_each_subsys(ss, i) {
1290 if (!(opts->subsys_mask & (1UL << i)))
1292 if (!try_module_get(cgroup_subsys[i]->module)) {
1293 module_pin_failed = true;
1297 if (module_pin_failed) {
1299 * oops, one of the modules was going away. this means that we
1300 * raced with a module_delete call, and to the user this is
1301 * essentially a "subsystem doesn't exist" case.
1303 for (i--; i >= 0; i--) {
1304 /* drop refcounts only on the ones we took */
1305 unsigned long bit = 1UL << i;
1307 if (!(bit & opts->subsys_mask))
1309 module_put(cgroup_subsys[i]->module);
1317 static void drop_parsed_module_refcounts(unsigned long subsys_mask)
1319 struct cgroup_subsys *ss;
1322 for_each_subsys(ss, i) {
1323 if (!(subsys_mask & (1UL << i)))
1325 module_put(cgroup_subsys[i]->module);
1329 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1332 struct cgroupfs_root *root = sb->s_fs_info;
1333 struct cgroup *cgrp = &root->top_cgroup;
1334 struct cgroup_sb_opts opts;
1335 unsigned long added_mask, removed_mask;
1337 if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) {
1338 pr_err("cgroup: sane_behavior: remount is not allowed\n");
1342 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1343 mutex_lock(&cgroup_mutex);
1344 mutex_lock(&cgroup_root_mutex);
1346 /* See what subsystems are wanted */
1347 ret = parse_cgroupfs_options(data, &opts);
1351 if (opts.subsys_mask != root->subsys_mask || opts.release_agent)
1352 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1353 task_tgid_nr(current), current->comm);
1355 added_mask = opts.subsys_mask & ~root->subsys_mask;
1356 removed_mask = root->subsys_mask & ~opts.subsys_mask;
1358 /* Don't allow flags or name to change at remount */
1359 if (opts.flags != root->flags ||
1360 (opts.name && strcmp(opts.name, root->name))) {
1362 drop_parsed_module_refcounts(opts.subsys_mask);
1367 * Clear out the files of subsystems that should be removed, do
1368 * this before rebind_subsystems, since rebind_subsystems may
1369 * change this hierarchy's subsys_list.
1371 cgroup_clear_directory(cgrp->dentry, false, removed_mask);
1373 ret = rebind_subsystems(root, added_mask, removed_mask);
1375 /* rebind_subsystems failed, re-populate the removed files */
1376 cgroup_populate_dir(cgrp, false, removed_mask);
1377 drop_parsed_module_refcounts(opts.subsys_mask);
1381 /* re-populate subsystem files */
1382 cgroup_populate_dir(cgrp, false, added_mask);
1384 if (opts.release_agent)
1385 strcpy(root->release_agent_path, opts.release_agent);
1387 kfree(opts.release_agent);
1389 mutex_unlock(&cgroup_root_mutex);
1390 mutex_unlock(&cgroup_mutex);
1391 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1395 static const struct super_operations cgroup_ops = {
1396 .statfs = simple_statfs,
1397 .drop_inode = generic_delete_inode,
1398 .show_options = cgroup_show_options,
1399 .remount_fs = cgroup_remount,
1402 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1404 INIT_LIST_HEAD(&cgrp->sibling);
1405 INIT_LIST_HEAD(&cgrp->children);
1406 INIT_LIST_HEAD(&cgrp->files);
1407 INIT_LIST_HEAD(&cgrp->cset_links);
1408 INIT_LIST_HEAD(&cgrp->release_list);
1409 INIT_LIST_HEAD(&cgrp->pidlists);
1410 mutex_init(&cgrp->pidlist_mutex);
1411 INIT_LIST_HEAD(&cgrp->event_list);
1412 spin_lock_init(&cgrp->event_list_lock);
1413 simple_xattrs_init(&cgrp->xattrs);
1416 static void init_cgroup_root(struct cgroupfs_root *root)
1418 struct cgroup *cgrp = &root->top_cgroup;
1420 INIT_LIST_HEAD(&root->subsys_list);
1421 INIT_LIST_HEAD(&root->root_list);
1422 root->number_of_cgroups = 1;
1424 cgrp->name = &root_cgroup_name;
1425 init_cgroup_housekeeping(cgrp);
1428 static int cgroup_init_root_id(struct cgroupfs_root *root, int start, int end)
1432 lockdep_assert_held(&cgroup_mutex);
1433 lockdep_assert_held(&cgroup_root_mutex);
1435 id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, start, end,
1440 root->hierarchy_id = id;
1444 static void cgroup_exit_root_id(struct cgroupfs_root *root)
1446 lockdep_assert_held(&cgroup_mutex);
1447 lockdep_assert_held(&cgroup_root_mutex);
1449 if (root->hierarchy_id) {
1450 idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
1451 root->hierarchy_id = 0;
1455 static int cgroup_test_super(struct super_block *sb, void *data)
1457 struct cgroup_sb_opts *opts = data;
1458 struct cgroupfs_root *root = sb->s_fs_info;
1460 /* If we asked for a name then it must match */
1461 if (opts->name && strcmp(opts->name, root->name))
1465 * If we asked for subsystems (or explicitly for no
1466 * subsystems) then they must match
1468 if ((opts->subsys_mask || opts->none)
1469 && (opts->subsys_mask != root->subsys_mask))
1475 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1477 struct cgroupfs_root *root;
1479 if (!opts->subsys_mask && !opts->none)
1482 root = kzalloc(sizeof(*root), GFP_KERNEL);
1484 return ERR_PTR(-ENOMEM);
1486 init_cgroup_root(root);
1488 root->subsys_mask = opts->subsys_mask;
1489 root->flags = opts->flags;
1490 ida_init(&root->cgroup_ida);
1491 if (opts->release_agent)
1492 strcpy(root->release_agent_path, opts->release_agent);
1494 strcpy(root->name, opts->name);
1495 if (opts->cpuset_clone_children)
1496 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags);
1500 static void cgroup_free_root(struct cgroupfs_root *root)
1503 /* hierarhcy ID shoulid already have been released */
1504 WARN_ON_ONCE(root->hierarchy_id);
1506 ida_destroy(&root->cgroup_ida);
1511 static int cgroup_set_super(struct super_block *sb, void *data)
1514 struct cgroup_sb_opts *opts = data;
1516 /* If we don't have a new root, we can't set up a new sb */
1517 if (!opts->new_root)
1520 BUG_ON(!opts->subsys_mask && !opts->none);
1522 ret = set_anon_super(sb, NULL);
1526 sb->s_fs_info = opts->new_root;
1527 opts->new_root->sb = sb;
1529 sb->s_blocksize = PAGE_CACHE_SIZE;
1530 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1531 sb->s_magic = CGROUP_SUPER_MAGIC;
1532 sb->s_op = &cgroup_ops;
1537 static int cgroup_get_rootdir(struct super_block *sb)
1539 static const struct dentry_operations cgroup_dops = {
1540 .d_iput = cgroup_diput,
1541 .d_delete = cgroup_delete,
1544 struct inode *inode =
1545 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1550 inode->i_fop = &simple_dir_operations;
1551 inode->i_op = &cgroup_dir_inode_operations;
1552 /* directories start off with i_nlink == 2 (for "." entry) */
1554 sb->s_root = d_make_root(inode);
1557 /* for everything else we want ->d_op set */
1558 sb->s_d_op = &cgroup_dops;
1562 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1563 int flags, const char *unused_dev_name,
1566 struct cgroup_sb_opts opts;
1567 struct cgroupfs_root *root;
1569 struct super_block *sb;
1570 struct cgroupfs_root *new_root;
1571 struct inode *inode;
1573 /* First find the desired set of subsystems */
1574 mutex_lock(&cgroup_mutex);
1575 ret = parse_cgroupfs_options(data, &opts);
1576 mutex_unlock(&cgroup_mutex);
1581 * Allocate a new cgroup root. We may not need it if we're
1582 * reusing an existing hierarchy.
1584 new_root = cgroup_root_from_opts(&opts);
1585 if (IS_ERR(new_root)) {
1586 ret = PTR_ERR(new_root);
1589 opts.new_root = new_root;
1591 /* Locate an existing or new sb for this hierarchy */
1592 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, 0, &opts);
1595 cgroup_free_root(opts.new_root);
1599 root = sb->s_fs_info;
1601 if (root == opts.new_root) {
1602 /* We used the new root structure, so this is a new hierarchy */
1603 struct list_head tmp_links;
1604 struct cgroup *root_cgrp = &root->top_cgroup;
1605 struct cgroupfs_root *existing_root;
1606 const struct cred *cred;
1608 struct css_set *cset;
1610 BUG_ON(sb->s_root != NULL);
1612 ret = cgroup_get_rootdir(sb);
1614 goto drop_new_super;
1615 inode = sb->s_root->d_inode;
1617 mutex_lock(&inode->i_mutex);
1618 mutex_lock(&cgroup_mutex);
1619 mutex_lock(&cgroup_root_mutex);
1621 /* Check for name clashes with existing mounts */
1623 if (strlen(root->name))
1624 for_each_active_root(existing_root)
1625 if (!strcmp(existing_root->name, root->name))
1629 * We're accessing css_set_count without locking
1630 * css_set_lock here, but that's OK - it can only be
1631 * increased by someone holding cgroup_lock, and
1632 * that's us. The worst that can happen is that we
1633 * have some link structures left over
1635 ret = allocate_cgrp_cset_links(css_set_count, &tmp_links);
1639 /* ID 0 is reserved for dummy root, 1 for unified hierarchy */
1640 ret = cgroup_init_root_id(root, 2, 0);
1644 ret = rebind_subsystems(root, root->subsys_mask, 0);
1645 if (ret == -EBUSY) {
1646 free_cgrp_cset_links(&tmp_links);
1650 * There must be no failure case after here, since rebinding
1651 * takes care of subsystems' refcounts, which are explicitly
1652 * dropped in the failure exit path.
1655 /* EBUSY should be the only error here */
1658 list_add(&root->root_list, &cgroup_roots);
1659 cgroup_root_count++;
1661 sb->s_root->d_fsdata = root_cgrp;
1662 root->top_cgroup.dentry = sb->s_root;
1664 /* Link the top cgroup in this hierarchy into all
1665 * the css_set objects */
1666 write_lock(&css_set_lock);
1667 hash_for_each(css_set_table, i, cset, hlist)
1668 link_css_set(&tmp_links, cset, root_cgrp);
1669 write_unlock(&css_set_lock);
1671 free_cgrp_cset_links(&tmp_links);
1673 BUG_ON(!list_empty(&root_cgrp->children));
1674 BUG_ON(root->number_of_cgroups != 1);
1676 cred = override_creds(&init_cred);
1677 cgroup_populate_dir(root_cgrp, true, root->subsys_mask);
1679 mutex_unlock(&cgroup_root_mutex);
1680 mutex_unlock(&cgroup_mutex);
1681 mutex_unlock(&inode->i_mutex);
1684 * We re-used an existing hierarchy - the new root (if
1685 * any) is not needed
1687 cgroup_free_root(opts.new_root);
1689 if (root->flags != opts.flags) {
1690 if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) {
1691 pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n");
1693 goto drop_new_super;
1695 pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n");
1699 /* no subsys rebinding, so refcounts don't change */
1700 drop_parsed_module_refcounts(opts.subsys_mask);
1703 kfree(opts.release_agent);
1705 return dget(sb->s_root);
1708 cgroup_exit_root_id(root);
1709 mutex_unlock(&cgroup_root_mutex);
1710 mutex_unlock(&cgroup_mutex);
1711 mutex_unlock(&inode->i_mutex);
1713 deactivate_locked_super(sb);
1715 drop_parsed_module_refcounts(opts.subsys_mask);
1717 kfree(opts.release_agent);
1719 return ERR_PTR(ret);
1722 static void cgroup_kill_sb(struct super_block *sb) {
1723 struct cgroupfs_root *root = sb->s_fs_info;
1724 struct cgroup *cgrp = &root->top_cgroup;
1725 struct cgrp_cset_link *link, *tmp_link;
1730 BUG_ON(root->number_of_cgroups != 1);
1731 BUG_ON(!list_empty(&cgrp->children));
1733 mutex_lock(&cgroup_mutex);
1734 mutex_lock(&cgroup_root_mutex);
1736 /* Rebind all subsystems back to the default hierarchy */
1737 ret = rebind_subsystems(root, 0, root->subsys_mask);
1738 /* Shouldn't be able to fail ... */
1742 * Release all the links from cset_links to this hierarchy's
1745 write_lock(&css_set_lock);
1747 list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
1748 list_del(&link->cset_link);
1749 list_del(&link->cgrp_link);
1752 write_unlock(&css_set_lock);
1754 if (!list_empty(&root->root_list)) {
1755 list_del(&root->root_list);
1756 cgroup_root_count--;
1759 cgroup_exit_root_id(root);
1761 mutex_unlock(&cgroup_root_mutex);
1762 mutex_unlock(&cgroup_mutex);
1764 simple_xattrs_free(&cgrp->xattrs);
1766 kill_litter_super(sb);
1767 cgroup_free_root(root);
1770 static struct file_system_type cgroup_fs_type = {
1772 .mount = cgroup_mount,
1773 .kill_sb = cgroup_kill_sb,
1776 static struct kobject *cgroup_kobj;
1779 * cgroup_path - generate the path of a cgroup
1780 * @cgrp: the cgroup in question
1781 * @buf: the buffer to write the path into
1782 * @buflen: the length of the buffer
1784 * Writes path of cgroup into buf. Returns 0 on success, -errno on error.
1786 * We can't generate cgroup path using dentry->d_name, as accessing
1787 * dentry->name must be protected by irq-unsafe dentry->d_lock or parent
1788 * inode's i_mutex, while on the other hand cgroup_path() can be called
1789 * with some irq-safe spinlocks held.
1791 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1793 int ret = -ENAMETOOLONG;
1796 if (!cgrp->parent) {
1797 if (strlcpy(buf, "/", buflen) >= buflen)
1798 return -ENAMETOOLONG;
1802 start = buf + buflen - 1;
1807 const char *name = cgroup_name(cgrp);
1811 if ((start -= len) < buf)
1813 memcpy(start, name, len);
1819 cgrp = cgrp->parent;
1820 } while (cgrp->parent);
1822 memmove(buf, start, buf + buflen - start);
1827 EXPORT_SYMBOL_GPL(cgroup_path);
1830 * task_cgroup_path_from_hierarchy - cgroup path of a task on a hierarchy
1831 * @task: target task
1832 * @hierarchy_id: the hierarchy to look up @task's cgroup from
1833 * @buf: the buffer to write the path into
1834 * @buflen: the length of the buffer
1836 * Determine @task's cgroup on the hierarchy specified by @hierarchy_id and
1837 * copy its path into @buf. This function grabs cgroup_mutex and shouldn't
1838 * be used inside locks used by cgroup controller callbacks.
1840 int task_cgroup_path_from_hierarchy(struct task_struct *task, int hierarchy_id,
1841 char *buf, size_t buflen)
1843 struct cgroupfs_root *root;
1844 struct cgroup *cgrp = NULL;
1847 mutex_lock(&cgroup_mutex);
1849 root = idr_find(&cgroup_hierarchy_idr, hierarchy_id);
1851 cgrp = task_cgroup_from_root(task, root);
1852 ret = cgroup_path(cgrp, buf, buflen);
1855 mutex_unlock(&cgroup_mutex);
1859 EXPORT_SYMBOL_GPL(task_cgroup_path_from_hierarchy);
1862 * Control Group taskset
1864 struct task_and_cgroup {
1865 struct task_struct *task;
1866 struct cgroup *cgrp;
1870 struct cgroup_taskset {
1871 struct task_and_cgroup single;
1872 struct flex_array *tc_array;
1875 struct cgroup *cur_cgrp;
1879 * cgroup_taskset_first - reset taskset and return the first task
1880 * @tset: taskset of interest
1882 * @tset iteration is initialized and the first task is returned.
1884 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1886 if (tset->tc_array) {
1888 return cgroup_taskset_next(tset);
1890 tset->cur_cgrp = tset->single.cgrp;
1891 return tset->single.task;
1894 EXPORT_SYMBOL_GPL(cgroup_taskset_first);
1897 * cgroup_taskset_next - iterate to the next task in taskset
1898 * @tset: taskset of interest
1900 * Return the next task in @tset. Iteration must have been initialized
1901 * with cgroup_taskset_first().
1903 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1905 struct task_and_cgroup *tc;
1907 if (!tset->tc_array || tset->idx >= tset->tc_array_len)
1910 tc = flex_array_get(tset->tc_array, tset->idx++);
1911 tset->cur_cgrp = tc->cgrp;
1914 EXPORT_SYMBOL_GPL(cgroup_taskset_next);
1917 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1918 * @tset: taskset of interest
1920 * Return the cgroup for the current (last returned) task of @tset. This
1921 * function must be preceded by either cgroup_taskset_first() or
1922 * cgroup_taskset_next().
1924 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset)
1926 return tset->cur_cgrp;
1928 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup);
1931 * cgroup_taskset_size - return the number of tasks in taskset
1932 * @tset: taskset of interest
1934 int cgroup_taskset_size(struct cgroup_taskset *tset)
1936 return tset->tc_array ? tset->tc_array_len : 1;
1938 EXPORT_SYMBOL_GPL(cgroup_taskset_size);
1942 * cgroup_task_migrate - move a task from one cgroup to another.
1944 * Must be called with cgroup_mutex and threadgroup locked.
1946 static void cgroup_task_migrate(struct cgroup *old_cgrp,
1947 struct task_struct *tsk,
1948 struct css_set *new_cset)
1950 struct css_set *old_cset;
1953 * We are synchronized through threadgroup_lock() against PF_EXITING
1954 * setting such that we can't race against cgroup_exit() changing the
1955 * css_set to init_css_set and dropping the old one.
1957 WARN_ON_ONCE(tsk->flags & PF_EXITING);
1958 old_cset = tsk->cgroups;
1961 rcu_assign_pointer(tsk->cgroups, new_cset);
1964 /* Update the css_set linked lists if we're using them */
1965 write_lock(&css_set_lock);
1966 if (!list_empty(&tsk->cg_list))
1967 list_move(&tsk->cg_list, &new_cset->tasks);
1968 write_unlock(&css_set_lock);
1971 * We just gained a reference on old_cset by taking it from the
1972 * task. As trading it for new_cset is protected by cgroup_mutex,
1973 * we're safe to drop it here; it will be freed under RCU.
1975 set_bit(CGRP_RELEASABLE, &old_cgrp->flags);
1976 put_css_set(old_cset);
1980 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
1981 * @cgrp: the cgroup to attach to
1982 * @tsk: the task or the leader of the threadgroup to be attached
1983 * @threadgroup: attach the whole threadgroup?
1985 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
1986 * task_lock of @tsk or each thread in the threadgroup individually in turn.
1988 static int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk,
1991 int retval, i, group_size;
1992 struct cgroup_subsys *ss, *failed_ss = NULL;
1993 struct cgroupfs_root *root = cgrp->root;
1994 /* threadgroup list cursor and array */
1995 struct task_struct *leader = tsk;
1996 struct task_and_cgroup *tc;
1997 struct flex_array *group;
1998 struct cgroup_taskset tset = { };
2001 * step 0: in order to do expensive, possibly blocking operations for
2002 * every thread, we cannot iterate the thread group list, since it needs
2003 * rcu or tasklist locked. instead, build an array of all threads in the
2004 * group - group_rwsem prevents new threads from appearing, and if
2005 * threads exit, this will just be an over-estimate.
2008 group_size = get_nr_threads(tsk);
2011 /* flex_array supports very large thread-groups better than kmalloc. */
2012 group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
2015 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2016 retval = flex_array_prealloc(group, 0, group_size, GFP_KERNEL);
2018 goto out_free_group_list;
2022 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2023 * already PF_EXITING could be freed from underneath us unless we
2024 * take an rcu_read_lock.
2028 struct task_and_cgroup ent;
2030 /* @tsk either already exited or can't exit until the end */
2031 if (tsk->flags & PF_EXITING)
2034 /* as per above, nr_threads may decrease, but not increase. */
2035 BUG_ON(i >= group_size);
2037 ent.cgrp = task_cgroup_from_root(tsk, root);
2038 /* nothing to do if this task is already in the cgroup */
2039 if (ent.cgrp == cgrp)
2042 * saying GFP_ATOMIC has no effect here because we did prealloc
2043 * earlier, but it's good form to communicate our expectations.
2045 retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
2046 BUG_ON(retval != 0);
2051 } while_each_thread(leader, tsk);
2053 /* remember the number of threads in the array for later. */
2055 tset.tc_array = group;
2056 tset.tc_array_len = group_size;
2058 /* methods shouldn't be called if no task is actually migrating */
2061 goto out_free_group_list;
2064 * step 1: check that we can legitimately attach to the cgroup.
2066 for_each_root_subsys(root, ss) {
2067 if (ss->can_attach) {
2068 retval = ss->can_attach(cgrp, &tset);
2071 goto out_cancel_attach;
2077 * step 2: make sure css_sets exist for all threads to be migrated.
2078 * we use find_css_set, which allocates a new one if necessary.
2080 for (i = 0; i < group_size; i++) {
2081 tc = flex_array_get(group, i);
2082 tc->cg = find_css_set(tc->task->cgroups, cgrp);
2085 goto out_put_css_set_refs;
2090 * step 3: now that we're guaranteed success wrt the css_sets,
2091 * proceed to move all tasks to the new cgroup. There are no
2092 * failure cases after here, so this is the commit point.
2094 for (i = 0; i < group_size; i++) {
2095 tc = flex_array_get(group, i);
2096 cgroup_task_migrate(tc->cgrp, tc->task, tc->cg);
2098 /* nothing is sensitive to fork() after this point. */
2101 * step 4: do subsystem attach callbacks.
2103 for_each_root_subsys(root, ss) {
2105 ss->attach(cgrp, &tset);
2109 * step 5: success! and cleanup
2112 out_put_css_set_refs:
2114 for (i = 0; i < group_size; i++) {
2115 tc = flex_array_get(group, i);
2118 put_css_set(tc->cg);
2123 for_each_root_subsys(root, ss) {
2124 if (ss == failed_ss)
2126 if (ss->cancel_attach)
2127 ss->cancel_attach(cgrp, &tset);
2130 out_free_group_list:
2131 flex_array_free(group);
2136 * Find the task_struct of the task to attach by vpid and pass it along to the
2137 * function to attach either it or all tasks in its threadgroup. Will lock
2138 * cgroup_mutex and threadgroup; may take task_lock of task.
2140 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2142 struct task_struct *tsk;
2143 const struct cred *cred = current_cred(), *tcred;
2146 if (!cgroup_lock_live_group(cgrp))
2152 tsk = find_task_by_vpid(pid);
2156 goto out_unlock_cgroup;
2159 * even if we're attaching all tasks in the thread group, we
2160 * only need to check permissions on one of them.
2162 tcred = __task_cred(tsk);
2163 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2164 !uid_eq(cred->euid, tcred->uid) &&
2165 !uid_eq(cred->euid, tcred->suid)) {
2168 goto out_unlock_cgroup;
2174 tsk = tsk->group_leader;
2177 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2178 * trapped in a cpuset, or RT worker may be born in a cgroup
2179 * with no rt_runtime allocated. Just say no.
2181 if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) {
2184 goto out_unlock_cgroup;
2187 get_task_struct(tsk);
2190 threadgroup_lock(tsk);
2192 if (!thread_group_leader(tsk)) {
2194 * a race with de_thread from another thread's exec()
2195 * may strip us of our leadership, if this happens,
2196 * there is no choice but to throw this task away and
2197 * try again; this is
2198 * "double-double-toil-and-trouble-check locking".
2200 threadgroup_unlock(tsk);
2201 put_task_struct(tsk);
2202 goto retry_find_task;
2206 ret = cgroup_attach_task(cgrp, tsk, threadgroup);
2208 threadgroup_unlock(tsk);
2210 put_task_struct(tsk);
2212 mutex_unlock(&cgroup_mutex);
2217 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2218 * @from: attach to all cgroups of a given task
2219 * @tsk: the task to be attached
2221 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2223 struct cgroupfs_root *root;
2226 mutex_lock(&cgroup_mutex);
2227 for_each_active_root(root) {
2228 struct cgroup *from_cg = task_cgroup_from_root(from, root);
2230 retval = cgroup_attach_task(from_cg, tsk, false);
2234 mutex_unlock(&cgroup_mutex);
2238 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2240 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2242 return attach_task_by_pid(cgrp, pid, false);
2245 static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2247 return attach_task_by_pid(cgrp, tgid, true);
2250 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2253 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2254 if (strlen(buffer) >= PATH_MAX)
2256 if (!cgroup_lock_live_group(cgrp))
2258 mutex_lock(&cgroup_root_mutex);
2259 strcpy(cgrp->root->release_agent_path, buffer);
2260 mutex_unlock(&cgroup_root_mutex);
2261 mutex_unlock(&cgroup_mutex);
2265 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2266 struct seq_file *seq)
2268 if (!cgroup_lock_live_group(cgrp))
2270 seq_puts(seq, cgrp->root->release_agent_path);
2271 seq_putc(seq, '\n');
2272 mutex_unlock(&cgroup_mutex);
2276 static int cgroup_sane_behavior_show(struct cgroup *cgrp, struct cftype *cft,
2277 struct seq_file *seq)
2279 seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp));
2283 /* A buffer size big enough for numbers or short strings */
2284 #define CGROUP_LOCAL_BUFFER_SIZE 64
2286 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2288 const char __user *userbuf,
2289 size_t nbytes, loff_t *unused_ppos)
2291 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2297 if (nbytes >= sizeof(buffer))
2299 if (copy_from_user(buffer, userbuf, nbytes))
2302 buffer[nbytes] = 0; /* nul-terminate */
2303 if (cft->write_u64) {
2304 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2307 retval = cft->write_u64(cgrp, cft, val);
2309 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2312 retval = cft->write_s64(cgrp, cft, val);
2319 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2321 const char __user *userbuf,
2322 size_t nbytes, loff_t *unused_ppos)
2324 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2326 size_t max_bytes = cft->max_write_len;
2327 char *buffer = local_buffer;
2330 max_bytes = sizeof(local_buffer) - 1;
2331 if (nbytes >= max_bytes)
2333 /* Allocate a dynamic buffer if we need one */
2334 if (nbytes >= sizeof(local_buffer)) {
2335 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2339 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2344 buffer[nbytes] = 0; /* nul-terminate */
2345 retval = cft->write_string(cgrp, cft, strstrip(buffer));
2349 if (buffer != local_buffer)
2354 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2355 size_t nbytes, loff_t *ppos)
2357 struct cftype *cft = __d_cft(file->f_dentry);
2358 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2360 if (cgroup_is_dead(cgrp))
2363 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2364 if (cft->write_u64 || cft->write_s64)
2365 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2366 if (cft->write_string)
2367 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2369 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2370 return ret ? ret : nbytes;
2375 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2377 char __user *buf, size_t nbytes,
2380 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2381 u64 val = cft->read_u64(cgrp, cft);
2382 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2384 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2387 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2389 char __user *buf, size_t nbytes,
2392 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2393 s64 val = cft->read_s64(cgrp, cft);
2394 int len = sprintf(tmp, "%lld\n", (long long) val);
2396 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2399 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2400 size_t nbytes, loff_t *ppos)
2402 struct cftype *cft = __d_cft(file->f_dentry);
2403 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2405 if (cgroup_is_dead(cgrp))
2409 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2411 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2413 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2418 * seqfile ops/methods for returning structured data. Currently just
2419 * supports string->u64 maps, but can be extended in future.
2422 struct cgroup_seqfile_state {
2424 struct cgroup *cgroup;
2427 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2429 struct seq_file *sf = cb->state;
2430 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2433 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2435 struct cgroup_seqfile_state *state = m->private;
2436 struct cftype *cft = state->cft;
2437 if (cft->read_map) {
2438 struct cgroup_map_cb cb = {
2439 .fill = cgroup_map_add,
2442 return cft->read_map(state->cgroup, cft, &cb);
2444 return cft->read_seq_string(state->cgroup, cft, m);
2447 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2449 struct seq_file *seq = file->private_data;
2450 kfree(seq->private);
2451 return single_release(inode, file);
2454 static const struct file_operations cgroup_seqfile_operations = {
2456 .write = cgroup_file_write,
2457 .llseek = seq_lseek,
2458 .release = cgroup_seqfile_release,
2461 static int cgroup_file_open(struct inode *inode, struct file *file)
2466 err = generic_file_open(inode, file);
2469 cft = __d_cft(file->f_dentry);
2471 if (cft->read_map || cft->read_seq_string) {
2472 struct cgroup_seqfile_state *state;
2474 state = kzalloc(sizeof(*state), GFP_USER);
2479 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2480 file->f_op = &cgroup_seqfile_operations;
2481 err = single_open(file, cgroup_seqfile_show, state);
2484 } else if (cft->open)
2485 err = cft->open(inode, file);
2492 static int cgroup_file_release(struct inode *inode, struct file *file)
2494 struct cftype *cft = __d_cft(file->f_dentry);
2496 return cft->release(inode, file);
2501 * cgroup_rename - Only allow simple rename of directories in place.
2503 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2504 struct inode *new_dir, struct dentry *new_dentry)
2507 struct cgroup_name *name, *old_name;
2508 struct cgroup *cgrp;
2511 * It's convinient to use parent dir's i_mutex to protected
2514 lockdep_assert_held(&old_dir->i_mutex);
2516 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2518 if (new_dentry->d_inode)
2520 if (old_dir != new_dir)
2523 cgrp = __d_cgrp(old_dentry);
2526 * This isn't a proper migration and its usefulness is very
2527 * limited. Disallow if sane_behavior.
2529 if (cgroup_sane_behavior(cgrp))
2532 name = cgroup_alloc_name(new_dentry);
2536 ret = simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2542 old_name = cgrp->name;
2543 rcu_assign_pointer(cgrp->name, name);
2545 kfree_rcu(old_name, rcu_head);
2549 static struct simple_xattrs *__d_xattrs(struct dentry *dentry)
2551 if (S_ISDIR(dentry->d_inode->i_mode))
2552 return &__d_cgrp(dentry)->xattrs;
2554 return &__d_cfe(dentry)->xattrs;
2557 static inline int xattr_enabled(struct dentry *dentry)
2559 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
2560 return root->flags & CGRP_ROOT_XATTR;
2563 static bool is_valid_xattr(const char *name)
2565 if (!strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
2566 !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
2571 static int cgroup_setxattr(struct dentry *dentry, const char *name,
2572 const void *val, size_t size, int flags)
2574 if (!xattr_enabled(dentry))
2576 if (!is_valid_xattr(name))
2578 return simple_xattr_set(__d_xattrs(dentry), name, val, size, flags);
2581 static int cgroup_removexattr(struct dentry *dentry, const char *name)
2583 if (!xattr_enabled(dentry))
2585 if (!is_valid_xattr(name))
2587 return simple_xattr_remove(__d_xattrs(dentry), name);
2590 static ssize_t cgroup_getxattr(struct dentry *dentry, const char *name,
2591 void *buf, size_t size)
2593 if (!xattr_enabled(dentry))
2595 if (!is_valid_xattr(name))
2597 return simple_xattr_get(__d_xattrs(dentry), name, buf, size);
2600 static ssize_t cgroup_listxattr(struct dentry *dentry, char *buf, size_t size)
2602 if (!xattr_enabled(dentry))
2604 return simple_xattr_list(__d_xattrs(dentry), buf, size);
2607 static const struct file_operations cgroup_file_operations = {
2608 .read = cgroup_file_read,
2609 .write = cgroup_file_write,
2610 .llseek = generic_file_llseek,
2611 .open = cgroup_file_open,
2612 .release = cgroup_file_release,
2615 static const struct inode_operations cgroup_file_inode_operations = {
2616 .setxattr = cgroup_setxattr,
2617 .getxattr = cgroup_getxattr,
2618 .listxattr = cgroup_listxattr,
2619 .removexattr = cgroup_removexattr,
2622 static const struct inode_operations cgroup_dir_inode_operations = {
2623 .lookup = cgroup_lookup,
2624 .mkdir = cgroup_mkdir,
2625 .rmdir = cgroup_rmdir,
2626 .rename = cgroup_rename,
2627 .setxattr = cgroup_setxattr,
2628 .getxattr = cgroup_getxattr,
2629 .listxattr = cgroup_listxattr,
2630 .removexattr = cgroup_removexattr,
2633 static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2635 if (dentry->d_name.len > NAME_MAX)
2636 return ERR_PTR(-ENAMETOOLONG);
2637 d_add(dentry, NULL);
2642 * Check if a file is a control file
2644 static inline struct cftype *__file_cft(struct file *file)
2646 if (file_inode(file)->i_fop != &cgroup_file_operations)
2647 return ERR_PTR(-EINVAL);
2648 return __d_cft(file->f_dentry);
2651 static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2652 struct super_block *sb)
2654 struct inode *inode;
2658 if (dentry->d_inode)
2661 inode = cgroup_new_inode(mode, sb);
2665 if (S_ISDIR(mode)) {
2666 inode->i_op = &cgroup_dir_inode_operations;
2667 inode->i_fop = &simple_dir_operations;
2669 /* start off with i_nlink == 2 (for "." entry) */
2671 inc_nlink(dentry->d_parent->d_inode);
2674 * Control reaches here with cgroup_mutex held.
2675 * @inode->i_mutex should nest outside cgroup_mutex but we
2676 * want to populate it immediately without releasing
2677 * cgroup_mutex. As @inode isn't visible to anyone else
2678 * yet, trylock will always succeed without affecting
2681 WARN_ON_ONCE(!mutex_trylock(&inode->i_mutex));
2682 } else if (S_ISREG(mode)) {
2684 inode->i_fop = &cgroup_file_operations;
2685 inode->i_op = &cgroup_file_inode_operations;
2687 d_instantiate(dentry, inode);
2688 dget(dentry); /* Extra count - pin the dentry in core */
2693 * cgroup_file_mode - deduce file mode of a control file
2694 * @cft: the control file in question
2696 * returns cft->mode if ->mode is not 0
2697 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2698 * returns S_IRUGO if it has only a read handler
2699 * returns S_IWUSR if it has only a write hander
2701 static umode_t cgroup_file_mode(const struct cftype *cft)
2708 if (cft->read || cft->read_u64 || cft->read_s64 ||
2709 cft->read_map || cft->read_seq_string)
2712 if (cft->write || cft->write_u64 || cft->write_s64 ||
2713 cft->write_string || cft->trigger)
2719 static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2722 struct dentry *dir = cgrp->dentry;
2723 struct cgroup *parent = __d_cgrp(dir);
2724 struct dentry *dentry;
2728 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2730 if (subsys && !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) {
2731 strcpy(name, subsys->name);
2734 strcat(name, cft->name);
2736 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2738 cfe = kzalloc(sizeof(*cfe), GFP_KERNEL);
2742 dentry = lookup_one_len(name, dir, strlen(name));
2743 if (IS_ERR(dentry)) {
2744 error = PTR_ERR(dentry);
2748 cfe->type = (void *)cft;
2749 cfe->dentry = dentry;
2750 dentry->d_fsdata = cfe;
2751 simple_xattrs_init(&cfe->xattrs);
2753 mode = cgroup_file_mode(cft);
2754 error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb);
2756 list_add_tail(&cfe->node, &parent->files);
2765 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2766 struct cftype cfts[], bool is_add)
2771 for (cft = cfts; cft->name[0] != '\0'; cft++) {
2772 /* does cft->flags tell us to skip this file on @cgrp? */
2773 if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp))
2775 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2777 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
2781 err = cgroup_add_file(cgrp, subsys, cft);
2783 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2787 cgroup_rm_file(cgrp, cft);
2793 static void cgroup_cfts_prepare(void)
2794 __acquires(&cgroup_mutex)
2797 * Thanks to the entanglement with vfs inode locking, we can't walk
2798 * the existing cgroups under cgroup_mutex and create files.
2799 * Instead, we use cgroup_for_each_descendant_pre() and drop RCU
2800 * read lock before calling cgroup_addrm_files().
2802 mutex_lock(&cgroup_mutex);
2805 static void cgroup_cfts_commit(struct cgroup_subsys *ss,
2806 struct cftype *cfts, bool is_add)
2807 __releases(&cgroup_mutex)
2810 struct cgroup *cgrp, *root = &ss->root->top_cgroup;
2811 struct super_block *sb = ss->root->sb;
2812 struct dentry *prev = NULL;
2813 struct inode *inode;
2816 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2817 if (!cfts || ss->root == &cgroup_dummy_root ||
2818 !atomic_inc_not_zero(&sb->s_active)) {
2819 mutex_unlock(&cgroup_mutex);
2824 * All cgroups which are created after we drop cgroup_mutex will
2825 * have the updated set of files, so we only need to update the
2826 * cgroups created before the current @cgroup_serial_nr_next.
2828 update_before = cgroup_serial_nr_next;
2830 mutex_unlock(&cgroup_mutex);
2832 /* @root always needs to be updated */
2833 inode = root->dentry->d_inode;
2834 mutex_lock(&inode->i_mutex);
2835 mutex_lock(&cgroup_mutex);
2836 cgroup_addrm_files(root, ss, cfts, is_add);
2837 mutex_unlock(&cgroup_mutex);
2838 mutex_unlock(&inode->i_mutex);
2840 /* add/rm files for all cgroups created before */
2842 cgroup_for_each_descendant_pre(cgrp, root) {
2843 if (cgroup_is_dead(cgrp))
2846 inode = cgrp->dentry->d_inode;
2851 prev = cgrp->dentry;
2853 mutex_lock(&inode->i_mutex);
2854 mutex_lock(&cgroup_mutex);
2855 if (cgrp->serial_nr < update_before && !cgroup_is_dead(cgrp))
2856 cgroup_addrm_files(cgrp, ss, cfts, is_add);
2857 mutex_unlock(&cgroup_mutex);
2858 mutex_unlock(&inode->i_mutex);
2864 deactivate_super(sb);
2868 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2869 * @ss: target cgroup subsystem
2870 * @cfts: zero-length name terminated array of cftypes
2872 * Register @cfts to @ss. Files described by @cfts are created for all
2873 * existing cgroups to which @ss is attached and all future cgroups will
2874 * have them too. This function can be called anytime whether @ss is
2877 * Returns 0 on successful registration, -errno on failure. Note that this
2878 * function currently returns 0 as long as @cfts registration is successful
2879 * even if some file creation attempts on existing cgroups fail.
2881 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2883 struct cftype_set *set;
2885 set = kzalloc(sizeof(*set), GFP_KERNEL);
2889 cgroup_cfts_prepare();
2891 list_add_tail(&set->node, &ss->cftsets);
2892 cgroup_cfts_commit(ss, cfts, true);
2896 EXPORT_SYMBOL_GPL(cgroup_add_cftypes);
2899 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2900 * @ss: target cgroup subsystem
2901 * @cfts: zero-length name terminated array of cftypes
2903 * Unregister @cfts from @ss. Files described by @cfts are removed from
2904 * all existing cgroups to which @ss is attached and all future cgroups
2905 * won't have them either. This function can be called anytime whether @ss
2906 * is attached or not.
2908 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2909 * registered with @ss.
2911 int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2913 struct cftype_set *set;
2915 cgroup_cfts_prepare();
2917 list_for_each_entry(set, &ss->cftsets, node) {
2918 if (set->cfts == cfts) {
2919 list_del(&set->node);
2921 cgroup_cfts_commit(ss, cfts, false);
2926 cgroup_cfts_commit(ss, NULL, false);
2931 * cgroup_task_count - count the number of tasks in a cgroup.
2932 * @cgrp: the cgroup in question
2934 * Return the number of tasks in the cgroup.
2936 int cgroup_task_count(const struct cgroup *cgrp)
2939 struct cgrp_cset_link *link;
2941 read_lock(&css_set_lock);
2942 list_for_each_entry(link, &cgrp->cset_links, cset_link)
2943 count += atomic_read(&link->cset->refcount);
2944 read_unlock(&css_set_lock);
2949 * Advance a list_head iterator. The iterator should be positioned at
2950 * the start of a css_set
2952 static void cgroup_advance_iter(struct cgroup *cgrp, struct cgroup_iter *it)
2954 struct list_head *l = it->cset_link;
2955 struct cgrp_cset_link *link;
2956 struct css_set *cset;
2958 /* Advance to the next non-empty css_set */
2961 if (l == &cgrp->cset_links) {
2962 it->cset_link = NULL;
2965 link = list_entry(l, struct cgrp_cset_link, cset_link);
2967 } while (list_empty(&cset->tasks));
2969 it->task = cset->tasks.next;
2973 * To reduce the fork() overhead for systems that are not actually
2974 * using their cgroups capability, we don't maintain the lists running
2975 * through each css_set to its tasks until we see the list actually
2976 * used - in other words after the first call to cgroup_iter_start().
2978 static void cgroup_enable_task_cg_lists(void)
2980 struct task_struct *p, *g;
2981 write_lock(&css_set_lock);
2982 use_task_css_set_links = 1;
2984 * We need tasklist_lock because RCU is not safe against
2985 * while_each_thread(). Besides, a forking task that has passed
2986 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2987 * is not guaranteed to have its child immediately visible in the
2988 * tasklist if we walk through it with RCU.
2990 read_lock(&tasklist_lock);
2991 do_each_thread(g, p) {
2994 * We should check if the process is exiting, otherwise
2995 * it will race with cgroup_exit() in that the list
2996 * entry won't be deleted though the process has exited.
2998 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2999 list_add(&p->cg_list, &p->cgroups->tasks);
3001 } while_each_thread(g, p);
3002 read_unlock(&tasklist_lock);
3003 write_unlock(&css_set_lock);
3007 * cgroup_next_sibling - find the next sibling of a given cgroup
3008 * @pos: the current cgroup
3010 * This function returns the next sibling of @pos and should be called
3011 * under RCU read lock. The only requirement is that @pos is accessible.
3012 * The next sibling is guaranteed to be returned regardless of @pos's
3015 struct cgroup *cgroup_next_sibling(struct cgroup *pos)
3017 struct cgroup *next;
3019 WARN_ON_ONCE(!rcu_read_lock_held());
3022 * @pos could already have been removed. Once a cgroup is removed,
3023 * its ->sibling.next is no longer updated when its next sibling
3024 * changes. As CGRP_DEAD assertion is serialized and happens
3025 * before the cgroup is taken off the ->sibling list, if we see it
3026 * unasserted, it's guaranteed that the next sibling hasn't
3027 * finished its grace period even if it's already removed, and thus
3028 * safe to dereference from this RCU critical section. If
3029 * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed
3030 * to be visible as %true here.
3032 if (likely(!cgroup_is_dead(pos))) {
3033 next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling);
3034 if (&next->sibling != &pos->parent->children)
3040 * Can't dereference the next pointer. Each cgroup is given a
3041 * monotonically increasing unique serial number and always
3042 * appended to the sibling list, so the next one can be found by
3043 * walking the parent's children until we see a cgroup with higher
3044 * serial number than @pos's.
3046 * While this path can be slow, it's taken only when either the
3047 * current cgroup is removed or iteration and removal race.
3049 list_for_each_entry_rcu(next, &pos->parent->children, sibling)
3050 if (next->serial_nr > pos->serial_nr)
3054 EXPORT_SYMBOL_GPL(cgroup_next_sibling);
3057 * cgroup_next_descendant_pre - find the next descendant for pre-order walk
3058 * @pos: the current position (%NULL to initiate traversal)
3059 * @cgroup: cgroup whose descendants to walk
3061 * To be used by cgroup_for_each_descendant_pre(). Find the next
3062 * descendant to visit for pre-order traversal of @cgroup's descendants.
3064 * While this function requires RCU read locking, it doesn't require the
3065 * whole traversal to be contained in a single RCU critical section. This
3066 * function will return the correct next descendant as long as both @pos
3067 * and @cgroup are accessible and @pos is a descendant of @cgroup.
3069 struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
3070 struct cgroup *cgroup)
3072 struct cgroup *next;
3074 WARN_ON_ONCE(!rcu_read_lock_held());
3076 /* if first iteration, pretend we just visited @cgroup */
3080 /* visit the first child if exists */
3081 next = list_first_or_null_rcu(&pos->children, struct cgroup, sibling);
3085 /* no child, visit my or the closest ancestor's next sibling */
3086 while (pos != cgroup) {
3087 next = cgroup_next_sibling(pos);
3095 EXPORT_SYMBOL_GPL(cgroup_next_descendant_pre);
3098 * cgroup_rightmost_descendant - return the rightmost descendant of a cgroup
3099 * @pos: cgroup of interest
3101 * Return the rightmost descendant of @pos. If there's no descendant,
3102 * @pos is returned. This can be used during pre-order traversal to skip
3105 * While this function requires RCU read locking, it doesn't require the
3106 * whole traversal to be contained in a single RCU critical section. This
3107 * function will return the correct rightmost descendant as long as @pos is
3110 struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos)
3112 struct cgroup *last, *tmp;
3114 WARN_ON_ONCE(!rcu_read_lock_held());
3118 /* ->prev isn't RCU safe, walk ->next till the end */
3120 list_for_each_entry_rcu(tmp, &last->children, sibling)
3126 EXPORT_SYMBOL_GPL(cgroup_rightmost_descendant);
3128 static struct cgroup *cgroup_leftmost_descendant(struct cgroup *pos)
3130 struct cgroup *last;
3134 pos = list_first_or_null_rcu(&pos->children, struct cgroup,
3142 * cgroup_next_descendant_post - find the next descendant for post-order walk
3143 * @pos: the current position (%NULL to initiate traversal)
3144 * @cgroup: cgroup whose descendants to walk
3146 * To be used by cgroup_for_each_descendant_post(). Find the next
3147 * descendant to visit for post-order traversal of @cgroup's descendants.
3149 * While this function requires RCU read locking, it doesn't require the
3150 * whole traversal to be contained in a single RCU critical section. This
3151 * function will return the correct next descendant as long as both @pos
3152 * and @cgroup are accessible and @pos is a descendant of @cgroup.
3154 struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
3155 struct cgroup *cgroup)
3157 struct cgroup *next;
3159 WARN_ON_ONCE(!rcu_read_lock_held());
3161 /* if first iteration, visit the leftmost descendant */
3163 next = cgroup_leftmost_descendant(cgroup);
3164 return next != cgroup ? next : NULL;
3167 /* if there's an unvisited sibling, visit its leftmost descendant */
3168 next = cgroup_next_sibling(pos);
3170 return cgroup_leftmost_descendant(next);
3172 /* no sibling left, visit parent */
3174 return next != cgroup ? next : NULL;
3176 EXPORT_SYMBOL_GPL(cgroup_next_descendant_post);
3178 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
3179 __acquires(css_set_lock)
3182 * The first time anyone tries to iterate across a cgroup,
3183 * we need to enable the list linking each css_set to its
3184 * tasks, and fix up all existing tasks.
3186 if (!use_task_css_set_links)
3187 cgroup_enable_task_cg_lists();
3189 read_lock(&css_set_lock);
3190 it->cset_link = &cgrp->cset_links;
3191 cgroup_advance_iter(cgrp, it);
3194 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
3195 struct cgroup_iter *it)
3197 struct task_struct *res;
3198 struct list_head *l = it->task;
3199 struct cgrp_cset_link *link;
3201 /* If the iterator cg is NULL, we have no tasks */
3204 res = list_entry(l, struct task_struct, cg_list);
3205 /* Advance iterator to find next entry */
3207 link = list_entry(it->cset_link, struct cgrp_cset_link, cset_link);
3208 if (l == &link->cset->tasks) {
3209 /* We reached the end of this task list - move on to
3210 * the next cg_cgroup_link */
3211 cgroup_advance_iter(cgrp, it);
3218 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
3219 __releases(css_set_lock)
3221 read_unlock(&css_set_lock);
3224 static inline int started_after_time(struct task_struct *t1,
3225 struct timespec *time,
3226 struct task_struct *t2)
3228 int start_diff = timespec_compare(&t1->start_time, time);
3229 if (start_diff > 0) {
3231 } else if (start_diff < 0) {
3235 * Arbitrarily, if two processes started at the same
3236 * time, we'll say that the lower pointer value
3237 * started first. Note that t2 may have exited by now
3238 * so this may not be a valid pointer any longer, but
3239 * that's fine - it still serves to distinguish
3240 * between two tasks started (effectively) simultaneously.
3247 * This function is a callback from heap_insert() and is used to order
3249 * In this case we order the heap in descending task start time.
3251 static inline int started_after(void *p1, void *p2)
3253 struct task_struct *t1 = p1;
3254 struct task_struct *t2 = p2;
3255 return started_after_time(t1, &t2->start_time, t2);
3259 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3260 * @scan: struct cgroup_scanner containing arguments for the scan
3262 * Arguments include pointers to callback functions test_task() and
3264 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3265 * and if it returns true, call process_task() for it also.
3266 * The test_task pointer may be NULL, meaning always true (select all tasks).
3267 * Effectively duplicates cgroup_iter_{start,next,end}()
3268 * but does not lock css_set_lock for the call to process_task().
3269 * The struct cgroup_scanner may be embedded in any structure of the caller's
3271 * It is guaranteed that process_task() will act on every task that
3272 * is a member of the cgroup for the duration of this call. This
3273 * function may or may not call process_task() for tasks that exit
3274 * or move to a different cgroup during the call, or are forked or
3275 * move into the cgroup during the call.
3277 * Note that test_task() may be called with locks held, and may in some
3278 * situations be called multiple times for the same task, so it should
3280 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3281 * pre-allocated and will be used for heap operations (and its "gt" member will
3282 * be overwritten), else a temporary heap will be used (allocation of which
3283 * may cause this function to fail).
3285 int cgroup_scan_tasks(struct cgroup_scanner *scan)
3288 struct cgroup_iter it;
3289 struct task_struct *p, *dropped;
3290 /* Never dereference latest_task, since it's not refcounted */
3291 struct task_struct *latest_task = NULL;
3292 struct ptr_heap tmp_heap;
3293 struct ptr_heap *heap;
3294 struct timespec latest_time = { 0, 0 };
3297 /* The caller supplied our heap and pre-allocated its memory */
3299 heap->gt = &started_after;
3301 /* We need to allocate our own heap memory */
3303 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
3305 /* cannot allocate the heap */
3311 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3312 * to determine which are of interest, and using the scanner's
3313 * "process_task" callback to process any of them that need an update.
3314 * Since we don't want to hold any locks during the task updates,
3315 * gather tasks to be processed in a heap structure.
3316 * The heap is sorted by descending task start time.
3317 * If the statically-sized heap fills up, we overflow tasks that
3318 * started later, and in future iterations only consider tasks that
3319 * started after the latest task in the previous pass. This
3320 * guarantees forward progress and that we don't miss any tasks.
3323 cgroup_iter_start(scan->cg, &it);
3324 while ((p = cgroup_iter_next(scan->cg, &it))) {
3326 * Only affect tasks that qualify per the caller's callback,
3327 * if he provided one
3329 if (scan->test_task && !scan->test_task(p, scan))
3332 * Only process tasks that started after the last task
3335 if (!started_after_time(p, &latest_time, latest_task))
3337 dropped = heap_insert(heap, p);
3338 if (dropped == NULL) {
3340 * The new task was inserted; the heap wasn't
3344 } else if (dropped != p) {
3346 * The new task was inserted, and pushed out a
3350 put_task_struct(dropped);
3353 * Else the new task was newer than anything already in
3354 * the heap and wasn't inserted
3357 cgroup_iter_end(scan->cg, &it);
3360 for (i = 0; i < heap->size; i++) {
3361 struct task_struct *q = heap->ptrs[i];
3363 latest_time = q->start_time;
3366 /* Process the task per the caller's callback */
3367 scan->process_task(q, scan);
3371 * If we had to process any tasks at all, scan again
3372 * in case some of them were in the middle of forking
3373 * children that didn't get processed.
3374 * Not the most efficient way to do it, but it avoids
3375 * having to take callback_mutex in the fork path
3379 if (heap == &tmp_heap)
3380 heap_free(&tmp_heap);
3384 static void cgroup_transfer_one_task(struct task_struct *task,
3385 struct cgroup_scanner *scan)
3387 struct cgroup *new_cgroup = scan->data;
3389 mutex_lock(&cgroup_mutex);
3390 cgroup_attach_task(new_cgroup, task, false);
3391 mutex_unlock(&cgroup_mutex);
3395 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3396 * @to: cgroup to which the tasks will be moved
3397 * @from: cgroup in which the tasks currently reside
3399 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
3401 struct cgroup_scanner scan;
3404 scan.test_task = NULL; /* select all tasks in cgroup */
3405 scan.process_task = cgroup_transfer_one_task;
3409 return cgroup_scan_tasks(&scan);
3413 * Stuff for reading the 'tasks'/'procs' files.
3415 * Reading this file can return large amounts of data if a cgroup has
3416 * *lots* of attached tasks. So it may need several calls to read(),
3417 * but we cannot guarantee that the information we produce is correct
3418 * unless we produce it entirely atomically.
3422 /* which pidlist file are we talking about? */
3423 enum cgroup_filetype {
3429 * A pidlist is a list of pids that virtually represents the contents of one
3430 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3431 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3434 struct cgroup_pidlist {
3436 * used to find which pidlist is wanted. doesn't change as long as
3437 * this particular list stays in the list.
3439 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
3442 /* how many elements the above list has */
3444 /* how many files are using the current array */
3446 /* each of these stored in a list by its cgroup */
3447 struct list_head links;
3448 /* pointer to the cgroup we belong to, for list removal purposes */
3449 struct cgroup *owner;
3450 /* protects the other fields */
3451 struct rw_semaphore mutex;
3455 * The following two functions "fix" the issue where there are more pids
3456 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3457 * TODO: replace with a kernel-wide solution to this problem
3459 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3460 static void *pidlist_allocate(int count)
3462 if (PIDLIST_TOO_LARGE(count))
3463 return vmalloc(count * sizeof(pid_t));
3465 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3467 static void pidlist_free(void *p)
3469 if (is_vmalloc_addr(p))
3476 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3477 * Returns the number of unique elements.
3479 static int pidlist_uniq(pid_t *list, int length)
3484 * we presume the 0th element is unique, so i starts at 1. trivial
3485 * edge cases first; no work needs to be done for either
3487 if (length == 0 || length == 1)
3489 /* src and dest walk down the list; dest counts unique elements */
3490 for (src = 1; src < length; src++) {
3491 /* find next unique element */
3492 while (list[src] == list[src-1]) {
3497 /* dest always points to where the next unique element goes */
3498 list[dest] = list[src];
3505 static int cmppid(const void *a, const void *b)
3507 return *(pid_t *)a - *(pid_t *)b;
3511 * find the appropriate pidlist for our purpose (given procs vs tasks)
3512 * returns with the lock on that pidlist already held, and takes care
3513 * of the use count, or returns NULL with no locks held if we're out of
3516 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3517 enum cgroup_filetype type)
3519 struct cgroup_pidlist *l;
3520 /* don't need task_nsproxy() if we're looking at ourself */
3521 struct pid_namespace *ns = task_active_pid_ns(current);
3524 * We can't drop the pidlist_mutex before taking the l->mutex in case
3525 * the last ref-holder is trying to remove l from the list at the same
3526 * time. Holding the pidlist_mutex precludes somebody taking whichever
3527 * list we find out from under us - compare release_pid_array().
3529 mutex_lock(&cgrp->pidlist_mutex);
3530 list_for_each_entry(l, &cgrp->pidlists, links) {
3531 if (l->key.type == type && l->key.ns == ns) {
3532 /* make sure l doesn't vanish out from under us */
3533 down_write(&l->mutex);
3534 mutex_unlock(&cgrp->pidlist_mutex);
3538 /* entry not found; create a new one */
3539 l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3541 mutex_unlock(&cgrp->pidlist_mutex);
3544 init_rwsem(&l->mutex);
3545 down_write(&l->mutex);
3547 l->key.ns = get_pid_ns(ns);
3549 list_add(&l->links, &cgrp->pidlists);
3550 mutex_unlock(&cgrp->pidlist_mutex);
3555 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3557 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3558 struct cgroup_pidlist **lp)
3562 int pid, n = 0; /* used for populating the array */
3563 struct cgroup_iter it;
3564 struct task_struct *tsk;
3565 struct cgroup_pidlist *l;
3568 * If cgroup gets more users after we read count, we won't have
3569 * enough space - tough. This race is indistinguishable to the
3570 * caller from the case that the additional cgroup users didn't
3571 * show up until sometime later on.
3573 length = cgroup_task_count(cgrp);
3574 array = pidlist_allocate(length);
3577 /* now, populate the array */
3578 cgroup_iter_start(cgrp, &it);
3579 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3580 if (unlikely(n == length))
3582 /* get tgid or pid for procs or tasks file respectively */
3583 if (type == CGROUP_FILE_PROCS)
3584 pid = task_tgid_vnr(tsk);
3586 pid = task_pid_vnr(tsk);
3587 if (pid > 0) /* make sure to only use valid results */
3590 cgroup_iter_end(cgrp, &it);
3592 /* now sort & (if procs) strip out duplicates */
3593 sort(array, length, sizeof(pid_t), cmppid, NULL);
3594 if (type == CGROUP_FILE_PROCS)
3595 length = pidlist_uniq(array, length);
3596 l = cgroup_pidlist_find(cgrp, type);
3598 pidlist_free(array);
3601 /* store array, freeing old if necessary - lock already held */
3602 pidlist_free(l->list);
3606 up_write(&l->mutex);
3612 * cgroupstats_build - build and fill cgroupstats
3613 * @stats: cgroupstats to fill information into
3614 * @dentry: A dentry entry belonging to the cgroup for which stats have
3617 * Build and fill cgroupstats so that taskstats can export it to user
3620 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3623 struct cgroup *cgrp;
3624 struct cgroup_iter it;
3625 struct task_struct *tsk;
3628 * Validate dentry by checking the superblock operations,
3629 * and make sure it's a directory.
3631 if (dentry->d_sb->s_op != &cgroup_ops ||
3632 !S_ISDIR(dentry->d_inode->i_mode))
3636 cgrp = dentry->d_fsdata;
3638 cgroup_iter_start(cgrp, &it);
3639 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3640 switch (tsk->state) {
3642 stats->nr_running++;
3644 case TASK_INTERRUPTIBLE:
3645 stats->nr_sleeping++;
3647 case TASK_UNINTERRUPTIBLE:
3648 stats->nr_uninterruptible++;
3651 stats->nr_stopped++;
3654 if (delayacct_is_task_waiting_on_io(tsk))
3655 stats->nr_io_wait++;
3659 cgroup_iter_end(cgrp, &it);
3667 * seq_file methods for the tasks/procs files. The seq_file position is the
3668 * next pid to display; the seq_file iterator is a pointer to the pid
3669 * in the cgroup->l->list array.
3672 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3675 * Initially we receive a position value that corresponds to
3676 * one more than the last pid shown (or 0 on the first call or
3677 * after a seek to the start). Use a binary-search to find the
3678 * next pid to display, if any
3680 struct cgroup_pidlist *l = s->private;
3681 int index = 0, pid = *pos;
3684 down_read(&l->mutex);
3686 int end = l->length;
3688 while (index < end) {
3689 int mid = (index + end) / 2;
3690 if (l->list[mid] == pid) {
3693 } else if (l->list[mid] <= pid)
3699 /* If we're off the end of the array, we're done */
3700 if (index >= l->length)
3702 /* Update the abstract position to be the actual pid that we found */
3703 iter = l->list + index;
3708 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3710 struct cgroup_pidlist *l = s->private;
3714 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3716 struct cgroup_pidlist *l = s->private;
3718 pid_t *end = l->list + l->length;
3720 * Advance to the next pid in the array. If this goes off the
3732 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3734 return seq_printf(s, "%d\n", *(int *)v);
3738 * seq_operations functions for iterating on pidlists through seq_file -
3739 * independent of whether it's tasks or procs
3741 static const struct seq_operations cgroup_pidlist_seq_operations = {
3742 .start = cgroup_pidlist_start,
3743 .stop = cgroup_pidlist_stop,
3744 .next = cgroup_pidlist_next,
3745 .show = cgroup_pidlist_show,
3748 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3751 * the case where we're the last user of this particular pidlist will
3752 * have us remove it from the cgroup's list, which entails taking the
3753 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3754 * pidlist_mutex, we have to take pidlist_mutex first.
3756 mutex_lock(&l->owner->pidlist_mutex);
3757 down_write(&l->mutex);
3758 BUG_ON(!l->use_count);
3759 if (!--l->use_count) {
3760 /* we're the last user if refcount is 0; remove and free */
3761 list_del(&l->links);
3762 mutex_unlock(&l->owner->pidlist_mutex);
3763 pidlist_free(l->list);
3764 put_pid_ns(l->key.ns);
3765 up_write(&l->mutex);
3769 mutex_unlock(&l->owner->pidlist_mutex);
3770 up_write(&l->mutex);
3773 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3775 struct cgroup_pidlist *l;
3776 if (!(file->f_mode & FMODE_READ))
3779 * the seq_file will only be initialized if the file was opened for
3780 * reading; hence we check if it's not null only in that case.
3782 l = ((struct seq_file *)file->private_data)->private;
3783 cgroup_release_pid_array(l);
3784 return seq_release(inode, file);
3787 static const struct file_operations cgroup_pidlist_operations = {
3789 .llseek = seq_lseek,
3790 .write = cgroup_file_write,
3791 .release = cgroup_pidlist_release,
3795 * The following functions handle opens on a file that displays a pidlist
3796 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3799 /* helper function for the two below it */
3800 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3802 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3803 struct cgroup_pidlist *l;
3806 /* Nothing to do for write-only files */
3807 if (!(file->f_mode & FMODE_READ))
3810 /* have the array populated */
3811 retval = pidlist_array_load(cgrp, type, &l);
3814 /* configure file information */
3815 file->f_op = &cgroup_pidlist_operations;
3817 retval = seq_open(file, &cgroup_pidlist_seq_operations);
3819 cgroup_release_pid_array(l);
3822 ((struct seq_file *)file->private_data)->private = l;
3825 static int cgroup_tasks_open(struct inode *unused, struct file *file)
3827 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3829 static int cgroup_procs_open(struct inode *unused, struct file *file)
3831 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3834 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3837 return notify_on_release(cgrp);
3840 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3844 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3846 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3848 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3853 * When dput() is called asynchronously, if umount has been done and
3854 * then deactivate_super() in cgroup_free_fn() kills the superblock,
3855 * there's a small window that vfs will see the root dentry with non-zero
3856 * refcnt and trigger BUG().
3858 * That's why we hold a reference before dput() and drop it right after.
3860 static void cgroup_dput(struct cgroup *cgrp)
3862 struct super_block *sb = cgrp->root->sb;
3864 atomic_inc(&sb->s_active);
3866 deactivate_super(sb);
3870 * Unregister event and free resources.
3872 * Gets called from workqueue.
3874 static void cgroup_event_remove(struct work_struct *work)
3876 struct cgroup_event *event = container_of(work, struct cgroup_event,
3878 struct cgroup *cgrp = event->cgrp;
3880 remove_wait_queue(event->wqh, &event->wait);
3882 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3884 /* Notify userspace the event is going away. */
3885 eventfd_signal(event->eventfd, 1);
3887 eventfd_ctx_put(event->eventfd);
3893 * Gets called on POLLHUP on eventfd when user closes it.
3895 * Called with wqh->lock held and interrupts disabled.
3897 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3898 int sync, void *key)
3900 struct cgroup_event *event = container_of(wait,
3901 struct cgroup_event, wait);
3902 struct cgroup *cgrp = event->cgrp;
3903 unsigned long flags = (unsigned long)key;
3905 if (flags & POLLHUP) {
3907 * If the event has been detached at cgroup removal, we
3908 * can simply return knowing the other side will cleanup
3911 * We can't race against event freeing since the other
3912 * side will require wqh->lock via remove_wait_queue(),
3915 spin_lock(&cgrp->event_list_lock);
3916 if (!list_empty(&event->list)) {
3917 list_del_init(&event->list);
3919 * We are in atomic context, but cgroup_event_remove()
3920 * may sleep, so we have to call it in workqueue.
3922 schedule_work(&event->remove);
3924 spin_unlock(&cgrp->event_list_lock);
3930 static void cgroup_event_ptable_queue_proc(struct file *file,
3931 wait_queue_head_t *wqh, poll_table *pt)
3933 struct cgroup_event *event = container_of(pt,
3934 struct cgroup_event, pt);
3937 add_wait_queue(wqh, &event->wait);
3941 * Parse input and register new cgroup event handler.
3943 * Input must be in format '<event_fd> <control_fd> <args>'.
3944 * Interpretation of args is defined by control file implementation.
3946 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3949 struct cgroup_event *event = NULL;
3950 struct cgroup *cgrp_cfile;
3951 unsigned int efd, cfd;
3952 struct file *efile = NULL;
3953 struct file *cfile = NULL;
3957 efd = simple_strtoul(buffer, &endp, 10);
3962 cfd = simple_strtoul(buffer, &endp, 10);
3963 if ((*endp != ' ') && (*endp != '\0'))
3967 event = kzalloc(sizeof(*event), GFP_KERNEL);
3971 INIT_LIST_HEAD(&event->list);
3972 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3973 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3974 INIT_WORK(&event->remove, cgroup_event_remove);
3976 efile = eventfd_fget(efd);
3977 if (IS_ERR(efile)) {
3978 ret = PTR_ERR(efile);
3982 event->eventfd = eventfd_ctx_fileget(efile);
3983 if (IS_ERR(event->eventfd)) {
3984 ret = PTR_ERR(event->eventfd);
3994 /* the process need read permission on control file */
3995 /* AV: shouldn't we check that it's been opened for read instead? */
3996 ret = inode_permission(file_inode(cfile), MAY_READ);
4000 event->cft = __file_cft(cfile);
4001 if (IS_ERR(event->cft)) {
4002 ret = PTR_ERR(event->cft);
4007 * The file to be monitored must be in the same cgroup as
4008 * cgroup.event_control is.
4010 cgrp_cfile = __d_cgrp(cfile->f_dentry->d_parent);
4011 if (cgrp_cfile != cgrp) {
4016 if (!event->cft->register_event || !event->cft->unregister_event) {
4021 ret = event->cft->register_event(cgrp, event->cft,
4022 event->eventfd, buffer);
4026 efile->f_op->poll(efile, &event->pt);
4029 * Events should be removed after rmdir of cgroup directory, but before
4030 * destroying subsystem state objects. Let's take reference to cgroup
4031 * directory dentry to do that.
4035 spin_lock(&cgrp->event_list_lock);
4036 list_add(&event->list, &cgrp->event_list);
4037 spin_unlock(&cgrp->event_list_lock);
4048 if (event && event->eventfd && !IS_ERR(event->eventfd))
4049 eventfd_ctx_put(event->eventfd);
4051 if (!IS_ERR_OR_NULL(efile))
4059 static u64 cgroup_clone_children_read(struct cgroup *cgrp,
4062 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4065 static int cgroup_clone_children_write(struct cgroup *cgrp,
4070 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4072 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4076 static struct cftype cgroup_base_files[] = {
4078 .name = "cgroup.procs",
4079 .open = cgroup_procs_open,
4080 .write_u64 = cgroup_procs_write,
4081 .release = cgroup_pidlist_release,
4082 .mode = S_IRUGO | S_IWUSR,
4085 .name = "cgroup.event_control",
4086 .write_string = cgroup_write_event_control,
4090 .name = "cgroup.clone_children",
4091 .flags = CFTYPE_INSANE,
4092 .read_u64 = cgroup_clone_children_read,
4093 .write_u64 = cgroup_clone_children_write,
4096 .name = "cgroup.sane_behavior",
4097 .flags = CFTYPE_ONLY_ON_ROOT,
4098 .read_seq_string = cgroup_sane_behavior_show,
4102 * Historical crazy stuff. These don't have "cgroup." prefix and
4103 * don't exist if sane_behavior. If you're depending on these, be
4104 * prepared to be burned.
4108 .flags = CFTYPE_INSANE, /* use "procs" instead */
4109 .open = cgroup_tasks_open,
4110 .write_u64 = cgroup_tasks_write,
4111 .release = cgroup_pidlist_release,
4112 .mode = S_IRUGO | S_IWUSR,
4115 .name = "notify_on_release",
4116 .flags = CFTYPE_INSANE,
4117 .read_u64 = cgroup_read_notify_on_release,
4118 .write_u64 = cgroup_write_notify_on_release,
4121 .name = "release_agent",
4122 .flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT,
4123 .read_seq_string = cgroup_release_agent_show,
4124 .write_string = cgroup_release_agent_write,
4125 .max_write_len = PATH_MAX,
4131 * cgroup_populate_dir - selectively creation of files in a directory
4132 * @cgrp: target cgroup
4133 * @base_files: true if the base files should be added
4134 * @subsys_mask: mask of the subsystem ids whose files should be added
4136 static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
4137 unsigned long subsys_mask)
4140 struct cgroup_subsys *ss;
4143 err = cgroup_addrm_files(cgrp, NULL, cgroup_base_files, true);
4148 /* process cftsets of each subsystem */
4149 for_each_root_subsys(cgrp->root, ss) {
4150 struct cftype_set *set;
4151 if (!test_bit(ss->subsys_id, &subsys_mask))
4154 list_for_each_entry(set, &ss->cftsets, node)
4155 cgroup_addrm_files(cgrp, ss, set->cfts, true);
4158 /* This cgroup is ready now */
4159 for_each_root_subsys(cgrp->root, ss) {
4160 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4162 * Update id->css pointer and make this css visible from
4163 * CSS ID functions. This pointer will be dereferened
4164 * from RCU-read-side without locks.
4167 rcu_assign_pointer(css->id->css, css);
4173 static void css_dput_fn(struct work_struct *work)
4175 struct cgroup_subsys_state *css =
4176 container_of(work, struct cgroup_subsys_state, dput_work);
4178 cgroup_dput(css->cgroup);
4181 static void css_release(struct percpu_ref *ref)
4183 struct cgroup_subsys_state *css =
4184 container_of(ref, struct cgroup_subsys_state, refcnt);
4186 schedule_work(&css->dput_work);
4189 static void init_cgroup_css(struct cgroup_subsys_state *css,
4190 struct cgroup_subsys *ss,
4191 struct cgroup *cgrp)
4196 if (cgrp == cgroup_dummy_top)
4197 css->flags |= CSS_ROOT;
4198 BUG_ON(cgrp->subsys[ss->subsys_id]);
4199 cgrp->subsys[ss->subsys_id] = css;
4202 * css holds an extra ref to @cgrp->dentry which is put on the last
4203 * css_put(). dput() requires process context, which css_put() may
4204 * be called without. @css->dput_work will be used to invoke
4205 * dput() asynchronously from css_put().
4207 INIT_WORK(&css->dput_work, css_dput_fn);
4210 /* invoke ->post_create() on a new CSS and mark it online if successful */
4211 static int online_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4215 lockdep_assert_held(&cgroup_mutex);
4218 ret = ss->css_online(cgrp);
4220 cgrp->subsys[ss->subsys_id]->flags |= CSS_ONLINE;
4224 /* if the CSS is online, invoke ->pre_destory() on it and mark it offline */
4225 static void offline_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4226 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4228 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4230 lockdep_assert_held(&cgroup_mutex);
4232 if (!(css->flags & CSS_ONLINE))
4235 if (ss->css_offline)
4236 ss->css_offline(cgrp);
4238 cgrp->subsys[ss->subsys_id]->flags &= ~CSS_ONLINE;
4242 * cgroup_create - create a cgroup
4243 * @parent: cgroup that will be parent of the new cgroup
4244 * @dentry: dentry of the new cgroup
4245 * @mode: mode to set on new inode
4247 * Must be called with the mutex on the parent inode held
4249 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
4252 struct cgroup *cgrp;
4253 struct cgroup_name *name;
4254 struct cgroupfs_root *root = parent->root;
4256 struct cgroup_subsys *ss;
4257 struct super_block *sb = root->sb;
4259 /* allocate the cgroup and its ID, 0 is reserved for the root */
4260 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
4264 name = cgroup_alloc_name(dentry);
4267 rcu_assign_pointer(cgrp->name, name);
4269 cgrp->id = ida_simple_get(&root->cgroup_ida, 1, 0, GFP_KERNEL);
4274 * Only live parents can have children. Note that the liveliness
4275 * check isn't strictly necessary because cgroup_mkdir() and
4276 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4277 * anyway so that locking is contained inside cgroup proper and we
4278 * don't get nasty surprises if we ever grow another caller.
4280 if (!cgroup_lock_live_group(parent)) {
4285 /* Grab a reference on the superblock so the hierarchy doesn't
4286 * get deleted on unmount if there are child cgroups. This
4287 * can be done outside cgroup_mutex, since the sb can't
4288 * disappear while someone has an open control file on the
4290 atomic_inc(&sb->s_active);
4292 init_cgroup_housekeeping(cgrp);
4294 dentry->d_fsdata = cgrp;
4295 cgrp->dentry = dentry;
4297 cgrp->parent = parent;
4298 cgrp->root = parent->root;
4300 if (notify_on_release(parent))
4301 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
4303 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
4304 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4306 for_each_root_subsys(root, ss) {
4307 struct cgroup_subsys_state *css;
4309 css = ss->css_alloc(cgrp);
4315 err = percpu_ref_init(&css->refcnt, css_release);
4319 init_cgroup_css(css, ss, cgrp);
4322 err = alloc_css_id(ss, parent, cgrp);
4329 * Create directory. cgroup_create_file() returns with the new
4330 * directory locked on success so that it can be populated without
4331 * dropping cgroup_mutex.
4333 err = cgroup_create_file(dentry, S_IFDIR | mode, sb);
4336 lockdep_assert_held(&dentry->d_inode->i_mutex);
4338 cgrp->serial_nr = cgroup_serial_nr_next++;
4340 /* allocation complete, commit to creation */
4341 list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children);
4342 root->number_of_cgroups++;
4344 /* each css holds a ref to the cgroup's dentry */
4345 for_each_root_subsys(root, ss)
4348 /* hold a ref to the parent's dentry */
4349 dget(parent->dentry);
4351 /* creation succeeded, notify subsystems */
4352 for_each_root_subsys(root, ss) {
4353 err = online_css(ss, cgrp);
4357 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
4359 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",
4360 current->comm, current->pid, ss->name);
4361 if (!strcmp(ss->name, "memory"))
4362 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4363 ss->warned_broken_hierarchy = true;
4367 err = cgroup_populate_dir(cgrp, true, root->subsys_mask);
4371 mutex_unlock(&cgroup_mutex);
4372 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
4377 for_each_root_subsys(root, ss) {
4378 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4381 percpu_ref_cancel_init(&css->refcnt);
4385 mutex_unlock(&cgroup_mutex);
4386 /* Release the reference count that we took on the superblock */
4387 deactivate_super(sb);
4389 ida_simple_remove(&root->cgroup_ida, cgrp->id);
4391 kfree(rcu_dereference_raw(cgrp->name));
4397 cgroup_destroy_locked(cgrp);
4398 mutex_unlock(&cgroup_mutex);
4399 mutex_unlock(&dentry->d_inode->i_mutex);
4403 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4405 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
4407 /* the vfs holds inode->i_mutex already */
4408 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
4411 static void cgroup_css_killed(struct cgroup *cgrp)
4413 if (!atomic_dec_and_test(&cgrp->css_kill_cnt))
4416 /* percpu ref's of all css's are killed, kick off the next step */
4417 INIT_WORK(&cgrp->destroy_work, cgroup_offline_fn);
4418 schedule_work(&cgrp->destroy_work);
4421 static void css_ref_killed_fn(struct percpu_ref *ref)
4423 struct cgroup_subsys_state *css =
4424 container_of(ref, struct cgroup_subsys_state, refcnt);
4426 cgroup_css_killed(css->cgroup);
4430 * cgroup_destroy_locked - the first stage of cgroup destruction
4431 * @cgrp: cgroup to be destroyed
4433 * css's make use of percpu refcnts whose killing latency shouldn't be
4434 * exposed to userland and are RCU protected. Also, cgroup core needs to
4435 * guarantee that css_tryget() won't succeed by the time ->css_offline() is
4436 * invoked. To satisfy all the requirements, destruction is implemented in
4437 * the following two steps.
4439 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
4440 * userland visible parts and start killing the percpu refcnts of
4441 * css's. Set up so that the next stage will be kicked off once all
4442 * the percpu refcnts are confirmed to be killed.
4444 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
4445 * rest of destruction. Once all cgroup references are gone, the
4446 * cgroup is RCU-freed.
4448 * This function implements s1. After this step, @cgrp is gone as far as
4449 * the userland is concerned and a new cgroup with the same name may be
4450 * created. As cgroup doesn't care about the names internally, this
4451 * doesn't cause any problem.
4453 static int cgroup_destroy_locked(struct cgroup *cgrp)
4454 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4456 struct dentry *d = cgrp->dentry;
4457 struct cgroup_event *event, *tmp;
4458 struct cgroup_subsys *ss;
4461 lockdep_assert_held(&d->d_inode->i_mutex);
4462 lockdep_assert_held(&cgroup_mutex);
4465 * css_set_lock synchronizes access to ->cset_links and prevents
4466 * @cgrp from being removed while __put_css_set() is in progress.
4468 read_lock(&css_set_lock);
4469 empty = list_empty(&cgrp->cset_links) && list_empty(&cgrp->children);
4470 read_unlock(&css_set_lock);
4475 * Block new css_tryget() by killing css refcnts. cgroup core
4476 * guarantees that, by the time ->css_offline() is invoked, no new
4477 * css reference will be given out via css_tryget(). We can't
4478 * simply call percpu_ref_kill() and proceed to offlining css's
4479 * because percpu_ref_kill() doesn't guarantee that the ref is seen
4480 * as killed on all CPUs on return.
4482 * Use percpu_ref_kill_and_confirm() to get notifications as each
4483 * css is confirmed to be seen as killed on all CPUs. The
4484 * notification callback keeps track of the number of css's to be
4485 * killed and schedules cgroup_offline_fn() to perform the rest of
4486 * destruction once the percpu refs of all css's are confirmed to
4489 atomic_set(&cgrp->css_kill_cnt, 1);
4490 for_each_root_subsys(cgrp->root, ss) {
4491 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4494 * Killing would put the base ref, but we need to keep it
4495 * alive until after ->css_offline.
4497 percpu_ref_get(&css->refcnt);
4499 atomic_inc(&cgrp->css_kill_cnt);
4500 percpu_ref_kill_and_confirm(&css->refcnt, css_ref_killed_fn);
4502 cgroup_css_killed(cgrp);
4505 * Mark @cgrp dead. This prevents further task migration and child
4506 * creation by disabling cgroup_lock_live_group(). Note that
4507 * CGRP_DEAD assertion is depended upon by cgroup_next_sibling() to
4508 * resume iteration after dropping RCU read lock. See
4509 * cgroup_next_sibling() for details.
4511 set_bit(CGRP_DEAD, &cgrp->flags);
4513 /* CGRP_DEAD is set, remove from ->release_list for the last time */
4514 raw_spin_lock(&release_list_lock);
4515 if (!list_empty(&cgrp->release_list))
4516 list_del_init(&cgrp->release_list);
4517 raw_spin_unlock(&release_list_lock);
4520 * Remove @cgrp directory. The removal puts the base ref but we
4521 * aren't quite done with @cgrp yet, so hold onto it.
4524 cgroup_d_remove_dir(d);
4527 * Unregister events and notify userspace.
4528 * Notify userspace about cgroup removing only after rmdir of cgroup
4529 * directory to avoid race between userspace and kernelspace.
4531 spin_lock(&cgrp->event_list_lock);
4532 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4533 list_del_init(&event->list);
4534 schedule_work(&event->remove);
4536 spin_unlock(&cgrp->event_list_lock);
4542 * cgroup_offline_fn - the second step of cgroup destruction
4543 * @work: cgroup->destroy_free_work
4545 * This function is invoked from a work item for a cgroup which is being
4546 * destroyed after the percpu refcnts of all css's are guaranteed to be
4547 * seen as killed on all CPUs, and performs the rest of destruction. This
4548 * is the second step of destruction described in the comment above
4549 * cgroup_destroy_locked().
4551 static void cgroup_offline_fn(struct work_struct *work)
4553 struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work);
4554 struct cgroup *parent = cgrp->parent;
4555 struct dentry *d = cgrp->dentry;
4556 struct cgroup_subsys *ss;
4558 mutex_lock(&cgroup_mutex);
4561 * css_tryget() is guaranteed to fail now. Tell subsystems to
4562 * initate destruction.
4564 for_each_root_subsys(cgrp->root, ss)
4565 offline_css(ss, cgrp);
4568 * Put the css refs from cgroup_destroy_locked(). Each css holds
4569 * an extra reference to the cgroup's dentry and cgroup removal
4570 * proceeds regardless of css refs. On the last put of each css,
4571 * whenever that may be, the extra dentry ref is put so that dentry
4572 * destruction happens only after all css's are released.
4574 for_each_root_subsys(cgrp->root, ss)
4575 css_put(cgrp->subsys[ss->subsys_id]);
4577 /* delete this cgroup from parent->children */
4578 list_del_rcu(&cgrp->sibling);
4582 set_bit(CGRP_RELEASABLE, &parent->flags);
4583 check_for_release(parent);
4585 mutex_unlock(&cgroup_mutex);
4588 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
4592 mutex_lock(&cgroup_mutex);
4593 ret = cgroup_destroy_locked(dentry->d_fsdata);
4594 mutex_unlock(&cgroup_mutex);
4599 static void __init_or_module cgroup_init_cftsets(struct cgroup_subsys *ss)
4601 INIT_LIST_HEAD(&ss->cftsets);
4604 * base_cftset is embedded in subsys itself, no need to worry about
4607 if (ss->base_cftypes) {
4608 ss->base_cftset.cfts = ss->base_cftypes;
4609 list_add_tail(&ss->base_cftset.node, &ss->cftsets);
4613 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4615 struct cgroup_subsys_state *css;
4617 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4619 mutex_lock(&cgroup_mutex);
4621 /* init base cftset */
4622 cgroup_init_cftsets(ss);
4624 /* Create the top cgroup state for this subsystem */
4625 list_add(&ss->sibling, &cgroup_dummy_root.subsys_list);
4626 ss->root = &cgroup_dummy_root;
4627 css = ss->css_alloc(cgroup_dummy_top);
4628 /* We don't handle early failures gracefully */
4629 BUG_ON(IS_ERR(css));
4630 init_cgroup_css(css, ss, cgroup_dummy_top);
4632 /* Update the init_css_set to contain a subsys
4633 * pointer to this state - since the subsystem is
4634 * newly registered, all tasks and hence the
4635 * init_css_set is in the subsystem's top cgroup. */
4636 init_css_set.subsys[ss->subsys_id] = css;
4638 need_forkexit_callback |= ss->fork || ss->exit;
4640 /* At system boot, before all subsystems have been
4641 * registered, no tasks have been forked, so we don't
4642 * need to invoke fork callbacks here. */
4643 BUG_ON(!list_empty(&init_task.tasks));
4645 BUG_ON(online_css(ss, cgroup_dummy_top));
4647 mutex_unlock(&cgroup_mutex);
4649 /* this function shouldn't be used with modular subsystems, since they
4650 * need to register a subsys_id, among other things */
4655 * cgroup_load_subsys: load and register a modular subsystem at runtime
4656 * @ss: the subsystem to load
4658 * This function should be called in a modular subsystem's initcall. If the
4659 * subsystem is built as a module, it will be assigned a new subsys_id and set
4660 * up for use. If the subsystem is built-in anyway, work is delegated to the
4661 * simpler cgroup_init_subsys.
4663 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4665 struct cgroup_subsys_state *css;
4667 struct hlist_node *tmp;
4668 struct css_set *cset;
4671 /* check name and function validity */
4672 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4673 ss->css_alloc == NULL || ss->css_free == NULL)
4677 * we don't support callbacks in modular subsystems. this check is
4678 * before the ss->module check for consistency; a subsystem that could
4679 * be a module should still have no callbacks even if the user isn't
4680 * compiling it as one.
4682 if (ss->fork || ss->exit)
4686 * an optionally modular subsystem is built-in: we want to do nothing,
4687 * since cgroup_init_subsys will have already taken care of it.
4689 if (ss->module == NULL) {
4690 /* a sanity check */
4691 BUG_ON(cgroup_subsys[ss->subsys_id] != ss);
4695 /* init base cftset */
4696 cgroup_init_cftsets(ss);
4698 mutex_lock(&cgroup_mutex);
4699 cgroup_subsys[ss->subsys_id] = ss;
4702 * no ss->css_alloc seems to need anything important in the ss
4703 * struct, so this can happen first (i.e. before the dummy root
4706 css = ss->css_alloc(cgroup_dummy_top);
4708 /* failure case - need to deassign the cgroup_subsys[] slot. */
4709 cgroup_subsys[ss->subsys_id] = NULL;
4710 mutex_unlock(&cgroup_mutex);
4711 return PTR_ERR(css);
4714 list_add(&ss->sibling, &cgroup_dummy_root.subsys_list);
4715 ss->root = &cgroup_dummy_root;
4717 /* our new subsystem will be attached to the dummy hierarchy. */
4718 init_cgroup_css(css, ss, cgroup_dummy_top);
4719 /* init_idr must be after init_cgroup_css because it sets css->id. */
4721 ret = cgroup_init_idr(ss, css);
4727 * Now we need to entangle the css into the existing css_sets. unlike
4728 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4729 * will need a new pointer to it; done by iterating the css_set_table.
4730 * furthermore, modifying the existing css_sets will corrupt the hash
4731 * table state, so each changed css_set will need its hash recomputed.
4732 * this is all done under the css_set_lock.
4734 write_lock(&css_set_lock);
4735 hash_for_each_safe(css_set_table, i, tmp, cset, hlist) {
4736 /* skip entries that we already rehashed */
4737 if (cset->subsys[ss->subsys_id])
4739 /* remove existing entry */
4740 hash_del(&cset->hlist);
4742 cset->subsys[ss->subsys_id] = css;
4743 /* recompute hash and restore entry */
4744 key = css_set_hash(cset->subsys);
4745 hash_add(css_set_table, &cset->hlist, key);
4747 write_unlock(&css_set_lock);
4749 ret = online_css(ss, cgroup_dummy_top);
4754 mutex_unlock(&cgroup_mutex);
4758 mutex_unlock(&cgroup_mutex);
4759 /* @ss can't be mounted here as try_module_get() would fail */
4760 cgroup_unload_subsys(ss);
4763 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4766 * cgroup_unload_subsys: unload a modular subsystem
4767 * @ss: the subsystem to unload
4769 * This function should be called in a modular subsystem's exitcall. When this
4770 * function is invoked, the refcount on the subsystem's module will be 0, so
4771 * the subsystem will not be attached to any hierarchy.
4773 void cgroup_unload_subsys(struct cgroup_subsys *ss)
4775 struct cgrp_cset_link *link;
4777 BUG_ON(ss->module == NULL);
4780 * we shouldn't be called if the subsystem is in use, and the use of
4781 * try_module_get in parse_cgroupfs_options should ensure that it
4782 * doesn't start being used while we're killing it off.
4784 BUG_ON(ss->root != &cgroup_dummy_root);
4786 mutex_lock(&cgroup_mutex);
4788 offline_css(ss, cgroup_dummy_top);
4791 idr_destroy(&ss->idr);
4793 /* deassign the subsys_id */
4794 cgroup_subsys[ss->subsys_id] = NULL;
4796 /* remove subsystem from the dummy root's list of subsystems */
4797 list_del_init(&ss->sibling);
4800 * disentangle the css from all css_sets attached to the dummy
4801 * top. as in loading, we need to pay our respects to the hashtable
4804 write_lock(&css_set_lock);
4805 list_for_each_entry(link, &cgroup_dummy_top->cset_links, cset_link) {
4806 struct css_set *cset = link->cset;
4809 hash_del(&cset->hlist);
4810 cset->subsys[ss->subsys_id] = NULL;
4811 key = css_set_hash(cset->subsys);
4812 hash_add(css_set_table, &cset->hlist, key);
4814 write_unlock(&css_set_lock);
4817 * remove subsystem's css from the cgroup_dummy_top and free it -
4818 * need to free before marking as null because ss->css_free needs
4819 * the cgrp->subsys pointer to find their state. note that this
4820 * also takes care of freeing the css_id.
4822 ss->css_free(cgroup_dummy_top);
4823 cgroup_dummy_top->subsys[ss->subsys_id] = NULL;
4825 mutex_unlock(&cgroup_mutex);
4827 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4830 * cgroup_init_early - cgroup initialization at system boot
4832 * Initialize cgroups at system boot, and initialize any
4833 * subsystems that request early init.
4835 int __init cgroup_init_early(void)
4837 struct cgroup_subsys *ss;
4840 atomic_set(&init_css_set.refcount, 1);
4841 INIT_LIST_HEAD(&init_css_set.cgrp_links);
4842 INIT_LIST_HEAD(&init_css_set.tasks);
4843 INIT_HLIST_NODE(&init_css_set.hlist);
4845 init_cgroup_root(&cgroup_dummy_root);
4846 cgroup_root_count = 1;
4847 init_task.cgroups = &init_css_set;
4849 init_cgrp_cset_link.cset = &init_css_set;
4850 init_cgrp_cset_link.cgrp = cgroup_dummy_top;
4851 list_add(&init_cgrp_cset_link.cset_link, &cgroup_dummy_top->cset_links);
4852 list_add(&init_cgrp_cset_link.cgrp_link, &init_css_set.cgrp_links);
4854 /* at bootup time, we don't worry about modular subsystems */
4855 for_each_builtin_subsys(ss, i) {
4857 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4858 BUG_ON(!ss->css_alloc);
4859 BUG_ON(!ss->css_free);
4860 if (ss->subsys_id != i) {
4861 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4862 ss->name, ss->subsys_id);
4867 cgroup_init_subsys(ss);
4873 * cgroup_init - cgroup initialization
4875 * Register cgroup filesystem and /proc file, and initialize
4876 * any subsystems that didn't request early init.
4878 int __init cgroup_init(void)
4880 struct cgroup_subsys *ss;
4884 err = bdi_init(&cgroup_backing_dev_info);
4888 for_each_builtin_subsys(ss, i) {
4889 if (!ss->early_init)
4890 cgroup_init_subsys(ss);
4892 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4895 /* allocate id for the dummy hierarchy */
4896 mutex_lock(&cgroup_mutex);
4897 mutex_lock(&cgroup_root_mutex);
4899 /* Add init_css_set to the hash table */
4900 key = css_set_hash(init_css_set.subsys);
4901 hash_add(css_set_table, &init_css_set.hlist, key);
4903 BUG_ON(cgroup_init_root_id(&cgroup_dummy_root, 0, 1));
4905 mutex_unlock(&cgroup_root_mutex);
4906 mutex_unlock(&cgroup_mutex);
4908 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4914 err = register_filesystem(&cgroup_fs_type);
4916 kobject_put(cgroup_kobj);
4920 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4924 bdi_destroy(&cgroup_backing_dev_info);
4930 * proc_cgroup_show()
4931 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4932 * - Used for /proc/<pid>/cgroup.
4933 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4934 * doesn't really matter if tsk->cgroup changes after we read it,
4935 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4936 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4937 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4938 * cgroup to top_cgroup.
4941 /* TODO: Use a proper seq_file iterator */
4942 int proc_cgroup_show(struct seq_file *m, void *v)
4945 struct task_struct *tsk;
4948 struct cgroupfs_root *root;
4951 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4957 tsk = get_pid_task(pid, PIDTYPE_PID);
4963 mutex_lock(&cgroup_mutex);
4965 for_each_active_root(root) {
4966 struct cgroup_subsys *ss;
4967 struct cgroup *cgrp;
4970 seq_printf(m, "%d:", root->hierarchy_id);
4971 for_each_root_subsys(root, ss)
4972 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4973 if (strlen(root->name))
4974 seq_printf(m, "%sname=%s", count ? "," : "",
4977 cgrp = task_cgroup_from_root(tsk, root);
4978 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4986 mutex_unlock(&cgroup_mutex);
4987 put_task_struct(tsk);
4994 /* Display information about each subsystem and each hierarchy */
4995 static int proc_cgroupstats_show(struct seq_file *m, void *v)
4997 struct cgroup_subsys *ss;
5000 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
5002 * ideally we don't want subsystems moving around while we do this.
5003 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
5004 * subsys/hierarchy state.
5006 mutex_lock(&cgroup_mutex);
5008 for_each_subsys(ss, i)
5009 seq_printf(m, "%s\t%d\t%d\t%d\n",
5010 ss->name, ss->root->hierarchy_id,
5011 ss->root->number_of_cgroups, !ss->disabled);
5013 mutex_unlock(&cgroup_mutex);
5017 static int cgroupstats_open(struct inode *inode, struct file *file)
5019 return single_open(file, proc_cgroupstats_show, NULL);
5022 static const struct file_operations proc_cgroupstats_operations = {
5023 .open = cgroupstats_open,
5025 .llseek = seq_lseek,
5026 .release = single_release,
5030 * cgroup_fork - attach newly forked task to its parents cgroup.
5031 * @child: pointer to task_struct of forking parent process.
5033 * Description: A task inherits its parent's cgroup at fork().
5035 * A pointer to the shared css_set was automatically copied in
5036 * fork.c by dup_task_struct(). However, we ignore that copy, since
5037 * it was not made under the protection of RCU or cgroup_mutex, so
5038 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
5039 * have already changed current->cgroups, allowing the previously
5040 * referenced cgroup group to be removed and freed.
5042 * At the point that cgroup_fork() is called, 'current' is the parent
5043 * task, and the passed argument 'child' points to the child task.
5045 void cgroup_fork(struct task_struct *child)
5048 child->cgroups = current->cgroups;
5049 get_css_set(child->cgroups);
5050 task_unlock(current);
5051 INIT_LIST_HEAD(&child->cg_list);
5055 * cgroup_post_fork - called on a new task after adding it to the task list
5056 * @child: the task in question
5058 * Adds the task to the list running through its css_set if necessary and
5059 * call the subsystem fork() callbacks. Has to be after the task is
5060 * visible on the task list in case we race with the first call to
5061 * cgroup_iter_start() - to guarantee that the new task ends up on its
5064 void cgroup_post_fork(struct task_struct *child)
5066 struct cgroup_subsys *ss;
5070 * use_task_css_set_links is set to 1 before we walk the tasklist
5071 * under the tasklist_lock and we read it here after we added the child
5072 * to the tasklist under the tasklist_lock as well. If the child wasn't
5073 * yet in the tasklist when we walked through it from
5074 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
5075 * should be visible now due to the paired locking and barriers implied
5076 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
5077 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
5080 if (use_task_css_set_links) {
5081 write_lock(&css_set_lock);
5083 if (list_empty(&child->cg_list))
5084 list_add(&child->cg_list, &child->cgroups->tasks);
5086 write_unlock(&css_set_lock);
5090 * Call ss->fork(). This must happen after @child is linked on
5091 * css_set; otherwise, @child might change state between ->fork()
5092 * and addition to css_set.
5094 if (need_forkexit_callback) {
5096 * fork/exit callbacks are supported only for builtin
5097 * subsystems, and the builtin section of the subsys
5098 * array is immutable, so we don't need to lock the
5099 * subsys array here. On the other hand, modular section
5100 * of the array can be freed at module unload, so we
5103 for_each_builtin_subsys(ss, i)
5110 * cgroup_exit - detach cgroup from exiting task
5111 * @tsk: pointer to task_struct of exiting process
5112 * @run_callback: run exit callbacks?
5114 * Description: Detach cgroup from @tsk and release it.
5116 * Note that cgroups marked notify_on_release force every task in
5117 * them to take the global cgroup_mutex mutex when exiting.
5118 * This could impact scaling on very large systems. Be reluctant to
5119 * use notify_on_release cgroups where very high task exit scaling
5120 * is required on large systems.
5122 * the_top_cgroup_hack:
5124 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
5126 * We call cgroup_exit() while the task is still competent to
5127 * handle notify_on_release(), then leave the task attached to the
5128 * root cgroup in each hierarchy for the remainder of its exit.
5130 * To do this properly, we would increment the reference count on
5131 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
5132 * code we would add a second cgroup function call, to drop that
5133 * reference. This would just create an unnecessary hot spot on
5134 * the top_cgroup reference count, to no avail.
5136 * Normally, holding a reference to a cgroup without bumping its
5137 * count is unsafe. The cgroup could go away, or someone could
5138 * attach us to a different cgroup, decrementing the count on
5139 * the first cgroup that we never incremented. But in this case,
5140 * top_cgroup isn't going away, and either task has PF_EXITING set,
5141 * which wards off any cgroup_attach_task() attempts, or task is a failed
5142 * fork, never visible to cgroup_attach_task.
5144 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
5146 struct cgroup_subsys *ss;
5147 struct css_set *cset;
5151 * Unlink from the css_set task list if necessary.
5152 * Optimistically check cg_list before taking
5155 if (!list_empty(&tsk->cg_list)) {
5156 write_lock(&css_set_lock);
5157 if (!list_empty(&tsk->cg_list))
5158 list_del_init(&tsk->cg_list);
5159 write_unlock(&css_set_lock);
5162 /* Reassign the task to the init_css_set. */
5164 cset = tsk->cgroups;
5165 tsk->cgroups = &init_css_set;
5167 if (run_callbacks && need_forkexit_callback) {
5169 * fork/exit callbacks are supported only for builtin
5170 * subsystems, see cgroup_post_fork() for details.
5172 for_each_builtin_subsys(ss, i) {
5174 struct cgroup *old_cgrp =
5175 rcu_dereference_raw(cset->subsys[i])->cgroup;
5176 struct cgroup *cgrp = task_cgroup(tsk, i);
5178 ss->exit(cgrp, old_cgrp, tsk);
5184 put_css_set_taskexit(cset);
5187 static void check_for_release(struct cgroup *cgrp)
5189 if (cgroup_is_releasable(cgrp) &&
5190 list_empty(&cgrp->cset_links) && list_empty(&cgrp->children)) {
5192 * Control Group is currently removeable. If it's not
5193 * already queued for a userspace notification, queue
5196 int need_schedule_work = 0;
5198 raw_spin_lock(&release_list_lock);
5199 if (!cgroup_is_dead(cgrp) &&
5200 list_empty(&cgrp->release_list)) {
5201 list_add(&cgrp->release_list, &release_list);
5202 need_schedule_work = 1;
5204 raw_spin_unlock(&release_list_lock);
5205 if (need_schedule_work)
5206 schedule_work(&release_agent_work);
5211 * Notify userspace when a cgroup is released, by running the
5212 * configured release agent with the name of the cgroup (path
5213 * relative to the root of cgroup file system) as the argument.
5215 * Most likely, this user command will try to rmdir this cgroup.
5217 * This races with the possibility that some other task will be
5218 * attached to this cgroup before it is removed, or that some other
5219 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5220 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5221 * unused, and this cgroup will be reprieved from its death sentence,
5222 * to continue to serve a useful existence. Next time it's released,
5223 * we will get notified again, if it still has 'notify_on_release' set.
5225 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5226 * means only wait until the task is successfully execve()'d. The
5227 * separate release agent task is forked by call_usermodehelper(),
5228 * then control in this thread returns here, without waiting for the
5229 * release agent task. We don't bother to wait because the caller of
5230 * this routine has no use for the exit status of the release agent
5231 * task, so no sense holding our caller up for that.
5233 static void cgroup_release_agent(struct work_struct *work)
5235 BUG_ON(work != &release_agent_work);
5236 mutex_lock(&cgroup_mutex);
5237 raw_spin_lock(&release_list_lock);
5238 while (!list_empty(&release_list)) {
5239 char *argv[3], *envp[3];
5241 char *pathbuf = NULL, *agentbuf = NULL;
5242 struct cgroup *cgrp = list_entry(release_list.next,
5245 list_del_init(&cgrp->release_list);
5246 raw_spin_unlock(&release_list_lock);
5247 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
5250 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
5252 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
5257 argv[i++] = agentbuf;
5258 argv[i++] = pathbuf;
5262 /* minimal command environment */
5263 envp[i++] = "HOME=/";
5264 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5267 /* Drop the lock while we invoke the usermode helper,
5268 * since the exec could involve hitting disk and hence
5269 * be a slow process */
5270 mutex_unlock(&cgroup_mutex);
5271 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
5272 mutex_lock(&cgroup_mutex);
5276 raw_spin_lock(&release_list_lock);
5278 raw_spin_unlock(&release_list_lock);
5279 mutex_unlock(&cgroup_mutex);
5282 static int __init cgroup_disable(char *str)
5284 struct cgroup_subsys *ss;
5288 while ((token = strsep(&str, ",")) != NULL) {
5293 * cgroup_disable, being at boot time, can't know about
5294 * module subsystems, so we don't worry about them.
5296 for_each_builtin_subsys(ss, i) {
5297 if (!strcmp(token, ss->name)) {
5299 printk(KERN_INFO "Disabling %s control group"
5300 " subsystem\n", ss->name);
5307 __setup("cgroup_disable=", cgroup_disable);
5310 * Functons for CSS ID.
5313 /* to get ID other than 0, this should be called when !cgroup_is_dead() */
5314 unsigned short css_id(struct cgroup_subsys_state *css)
5316 struct css_id *cssid;
5319 * This css_id() can return correct value when somone has refcnt
5320 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5321 * it's unchanged until freed.
5323 cssid = rcu_dereference_raw(css->id);
5329 EXPORT_SYMBOL_GPL(css_id);
5332 * css_is_ancestor - test "root" css is an ancestor of "child"
5333 * @child: the css to be tested.
5334 * @root: the css supporsed to be an ancestor of the child.
5336 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5337 * this function reads css->id, the caller must hold rcu_read_lock().
5338 * But, considering usual usage, the csses should be valid objects after test.
5339 * Assuming that the caller will do some action to the child if this returns
5340 * returns true, the caller must take "child";s reference count.
5341 * If "child" is valid object and this returns true, "root" is valid, too.
5344 bool css_is_ancestor(struct cgroup_subsys_state *child,
5345 const struct cgroup_subsys_state *root)
5347 struct css_id *child_id;
5348 struct css_id *root_id;
5350 child_id = rcu_dereference(child->id);
5353 root_id = rcu_dereference(root->id);
5356 if (child_id->depth < root_id->depth)
5358 if (child_id->stack[root_id->depth] != root_id->id)
5363 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
5365 struct css_id *id = css->id;
5366 /* When this is called before css_id initialization, id can be NULL */
5370 BUG_ON(!ss->use_id);
5372 rcu_assign_pointer(id->css, NULL);
5373 rcu_assign_pointer(css->id, NULL);
5374 spin_lock(&ss->id_lock);
5375 idr_remove(&ss->idr, id->id);
5376 spin_unlock(&ss->id_lock);
5377 kfree_rcu(id, rcu_head);
5379 EXPORT_SYMBOL_GPL(free_css_id);
5382 * This is called by init or create(). Then, calls to this function are
5383 * always serialized (By cgroup_mutex() at create()).
5386 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
5388 struct css_id *newid;
5391 BUG_ON(!ss->use_id);
5393 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
5394 newid = kzalloc(size, GFP_KERNEL);
5396 return ERR_PTR(-ENOMEM);
5398 idr_preload(GFP_KERNEL);
5399 spin_lock(&ss->id_lock);
5400 /* Don't use 0. allocates an ID of 1-65535 */
5401 ret = idr_alloc(&ss->idr, newid, 1, CSS_ID_MAX + 1, GFP_NOWAIT);
5402 spin_unlock(&ss->id_lock);
5405 /* Returns error when there are no free spaces for new ID.*/
5410 newid->depth = depth;
5414 return ERR_PTR(ret);
5418 static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
5419 struct cgroup_subsys_state *rootcss)
5421 struct css_id *newid;
5423 spin_lock_init(&ss->id_lock);
5426 newid = get_new_cssid(ss, 0);
5428 return PTR_ERR(newid);
5430 newid->stack[0] = newid->id;
5431 newid->css = rootcss;
5432 rootcss->id = newid;
5436 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
5437 struct cgroup *child)
5439 int subsys_id, i, depth = 0;
5440 struct cgroup_subsys_state *parent_css, *child_css;
5441 struct css_id *child_id, *parent_id;
5443 subsys_id = ss->subsys_id;
5444 parent_css = parent->subsys[subsys_id];
5445 child_css = child->subsys[subsys_id];
5446 parent_id = parent_css->id;
5447 depth = parent_id->depth + 1;
5449 child_id = get_new_cssid(ss, depth);
5450 if (IS_ERR(child_id))
5451 return PTR_ERR(child_id);
5453 for (i = 0; i < depth; i++)
5454 child_id->stack[i] = parent_id->stack[i];
5455 child_id->stack[depth] = child_id->id;
5457 * child_id->css pointer will be set after this cgroup is available
5458 * see cgroup_populate_dir()
5460 rcu_assign_pointer(child_css->id, child_id);
5466 * css_lookup - lookup css by id
5467 * @ss: cgroup subsys to be looked into.
5470 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5471 * NULL if not. Should be called under rcu_read_lock()
5473 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5475 struct css_id *cssid = NULL;
5477 BUG_ON(!ss->use_id);
5478 cssid = idr_find(&ss->idr, id);
5480 if (unlikely(!cssid))
5483 return rcu_dereference(cssid->css);
5485 EXPORT_SYMBOL_GPL(css_lookup);
5488 * get corresponding css from file open on cgroupfs directory
5490 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5492 struct cgroup *cgrp;
5493 struct inode *inode;
5494 struct cgroup_subsys_state *css;
5496 inode = file_inode(f);
5497 /* check in cgroup filesystem dir */
5498 if (inode->i_op != &cgroup_dir_inode_operations)
5499 return ERR_PTR(-EBADF);
5501 if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5502 return ERR_PTR(-EINVAL);
5505 cgrp = __d_cgrp(f->f_dentry);
5506 css = cgrp->subsys[id];
5507 return css ? css : ERR_PTR(-ENOENT);
5510 #ifdef CONFIG_CGROUP_DEBUG
5511 static struct cgroup_subsys_state *debug_css_alloc(struct cgroup *cgrp)
5513 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5516 return ERR_PTR(-ENOMEM);
5521 static void debug_css_free(struct cgroup *cgrp)
5523 kfree(cgrp->subsys[debug_subsys_id]);
5526 static u64 debug_taskcount_read(struct cgroup *cgrp, struct cftype *cft)
5528 return cgroup_task_count(cgrp);
5531 static u64 current_css_set_read(struct cgroup *cgrp, struct cftype *cft)
5533 return (u64)(unsigned long)current->cgroups;
5536 static u64 current_css_set_refcount_read(struct cgroup *cgrp,
5542 count = atomic_read(¤t->cgroups->refcount);
5547 static int current_css_set_cg_links_read(struct cgroup *cgrp,
5549 struct seq_file *seq)
5551 struct cgrp_cset_link *link;
5552 struct css_set *cset;
5554 read_lock(&css_set_lock);
5556 cset = rcu_dereference(current->cgroups);
5557 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
5558 struct cgroup *c = link->cgrp;
5562 name = c->dentry->d_name.name;
5565 seq_printf(seq, "Root %d group %s\n",
5566 c->root->hierarchy_id, name);
5569 read_unlock(&css_set_lock);
5573 #define MAX_TASKS_SHOWN_PER_CSS 25
5574 static int cgroup_css_links_read(struct cgroup *cgrp,
5576 struct seq_file *seq)
5578 struct cgrp_cset_link *link;
5580 read_lock(&css_set_lock);
5581 list_for_each_entry(link, &cgrp->cset_links, cset_link) {
5582 struct css_set *cset = link->cset;
5583 struct task_struct *task;
5585 seq_printf(seq, "css_set %p\n", cset);
5586 list_for_each_entry(task, &cset->tasks, cg_list) {
5587 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5588 seq_puts(seq, " ...\n");
5591 seq_printf(seq, " task %d\n",
5592 task_pid_vnr(task));
5596 read_unlock(&css_set_lock);
5600 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5602 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5605 static struct cftype debug_files[] = {
5607 .name = "taskcount",
5608 .read_u64 = debug_taskcount_read,
5612 .name = "current_css_set",
5613 .read_u64 = current_css_set_read,
5617 .name = "current_css_set_refcount",
5618 .read_u64 = current_css_set_refcount_read,
5622 .name = "current_css_set_cg_links",
5623 .read_seq_string = current_css_set_cg_links_read,
5627 .name = "cgroup_css_links",
5628 .read_seq_string = cgroup_css_links_read,
5632 .name = "releasable",
5633 .read_u64 = releasable_read,
5639 struct cgroup_subsys debug_subsys = {
5641 .css_alloc = debug_css_alloc,
5642 .css_free = debug_css_free,
5643 .subsys_id = debug_subsys_id,
5644 .base_cftypes = debug_files,
5646 #endif /* CONFIG_CGROUP_DEBUG */