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/flex_array.h> /* used in cgroup_attach_task */
60 #include <linux/kthread.h>
62 #include <linux/atomic.h>
65 * cgroup_mutex is the master lock. Any modification to cgroup or its
66 * hierarchy must be performed while holding it.
68 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
69 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
70 * release_agent_path and so on. Modifying requires both cgroup_mutex and
71 * cgroup_root_mutex. Readers can acquire either of the two. This is to
72 * break the following locking order cycle.
74 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
75 * B. namespace_sem -> cgroup_mutex
77 * B happens only through cgroup_show_options() and using cgroup_root_mutex
80 #ifdef CONFIG_PROVE_RCU
81 DEFINE_MUTEX(cgroup_mutex
);
82 EXPORT_SYMBOL_GPL(cgroup_mutex
); /* only for lockdep */
84 static DEFINE_MUTEX(cgroup_mutex
);
87 static DEFINE_MUTEX(cgroup_root_mutex
);
90 * Generate an array of cgroup subsystem pointers. At boot time, this is
91 * populated with the built in subsystems, and modular subsystems are
92 * registered after that. The mutable section of this array is protected by
95 #define SUBSYS(_x) [_x ## _subsys_id] = &_x ## _subsys,
96 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
97 static struct cgroup_subsys
*cgroup_subsys
[CGROUP_SUBSYS_COUNT
] = {
98 #include <linux/cgroup_subsys.h>
102 * The dummy hierarchy, reserved for the subsystems that are otherwise
103 * unattached - it never has more than a single cgroup, and all tasks are
104 * part of that cgroup.
106 static struct cgroupfs_root cgroup_dummy_root
;
108 /* dummy_top is a shorthand for the dummy hierarchy's top cgroup */
109 static struct cgroup
* const cgroup_dummy_top
= &cgroup_dummy_root
.top_cgroup
;
112 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
115 struct list_head node
;
116 struct dentry
*dentry
;
118 struct cgroup_subsys_state
*css
;
121 struct simple_xattrs xattrs
;
125 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
126 * cgroup_subsys->use_id != 0.
128 #define CSS_ID_MAX (65535)
131 * The css to which this ID points. This pointer is set to valid value
132 * after cgroup is populated. If cgroup is removed, this will be NULL.
133 * This pointer is expected to be RCU-safe because destroy()
134 * is called after synchronize_rcu(). But for safe use, css_tryget()
135 * should be used for avoiding race.
137 struct cgroup_subsys_state __rcu
*css
;
143 * Depth in hierarchy which this ID belongs to.
145 unsigned short depth
;
147 * ID is freed by RCU. (and lookup routine is RCU safe.)
149 struct rcu_head rcu_head
;
151 * Hierarchy of CSS ID belongs to.
153 unsigned short stack
[0]; /* Array of Length (depth+1) */
156 /* The list of hierarchy roots */
158 static LIST_HEAD(cgroup_roots
);
159 static int cgroup_root_count
;
162 * Hierarchy ID allocation and mapping. It follows the same exclusion
163 * rules as other root ops - both cgroup_mutex and cgroup_root_mutex for
164 * writes, either for reads.
166 static DEFINE_IDR(cgroup_hierarchy_idr
);
168 static struct cgroup_name root_cgroup_name
= { .name
= "/" };
171 * Assign a monotonically increasing serial number to cgroups. It
172 * guarantees cgroups with bigger numbers are newer than those with smaller
173 * numbers. Also, as cgroups are always appended to the parent's
174 * ->children list, it guarantees that sibling cgroups are always sorted in
175 * the ascending serial number order on the list. Protected by
178 static u64 cgroup_serial_nr_next
= 1;
180 /* This flag indicates whether tasks in the fork and exit paths should
181 * check for fork/exit handlers to call. This avoids us having to do
182 * extra work in the fork/exit path if none of the subsystems need to
185 static int need_forkexit_callback __read_mostly
;
187 static struct cftype cgroup_base_files
[];
189 static void cgroup_destroy_css_killed(struct cgroup
*cgrp
);
190 static int cgroup_destroy_locked(struct cgroup
*cgrp
);
191 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cftype cfts
[],
195 * cgroup_css - obtain a cgroup's css for the specified subsystem
196 * @cgrp: the cgroup of interest
197 * @ss: the subsystem of interest (%NULL returns the dummy_css)
199 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
200 * function must be called either under cgroup_mutex or rcu_read_lock() and
201 * the caller is responsible for pinning the returned css if it wants to
202 * keep accessing it outside the said locks. This function may return
203 * %NULL if @cgrp doesn't have @subsys_id enabled.
205 struct cgroup_subsys_state
*cgroup_css(struct cgroup
*cgrp
,
206 struct cgroup_subsys
*ss
)
209 return rcu_dereference_check(cgrp
->subsys
[ss
->subsys_id
],
210 lockdep_is_held(&cgroup_mutex
));
212 return &cgrp
->dummy_css
;
215 /* convenient tests for these bits */
216 static inline bool cgroup_is_dead(const struct cgroup
*cgrp
)
218 return test_bit(CGRP_DEAD
, &cgrp
->flags
);
222 * cgroup_is_descendant - test ancestry
223 * @cgrp: the cgroup to be tested
224 * @ancestor: possible ancestor of @cgrp
226 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
227 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
228 * and @ancestor are accessible.
230 bool cgroup_is_descendant(struct cgroup
*cgrp
, struct cgroup
*ancestor
)
233 if (cgrp
== ancestor
)
239 EXPORT_SYMBOL_GPL(cgroup_is_descendant
);
241 static int cgroup_is_releasable(const struct cgroup
*cgrp
)
244 (1 << CGRP_RELEASABLE
) |
245 (1 << CGRP_NOTIFY_ON_RELEASE
);
246 return (cgrp
->flags
& bits
) == bits
;
249 static int notify_on_release(const struct cgroup
*cgrp
)
251 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
255 * for_each_subsys - iterate all loaded cgroup subsystems
256 * @ss: the iteration cursor
257 * @i: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
259 * Should be called under cgroup_mutex.
261 #define for_each_subsys(ss, i) \
262 for ((i) = 0; (i) < CGROUP_SUBSYS_COUNT; (i)++) \
263 if (({ lockdep_assert_held(&cgroup_mutex); \
264 !((ss) = cgroup_subsys[i]); })) { } \
268 * for_each_builtin_subsys - iterate all built-in cgroup subsystems
269 * @ss: the iteration cursor
270 * @i: the index of @ss, CGROUP_BUILTIN_SUBSYS_COUNT after reaching the end
272 * Bulit-in subsystems are always present and iteration itself doesn't
273 * require any synchronization.
275 #define for_each_builtin_subsys(ss, i) \
276 for ((i) = 0; (i) < CGROUP_BUILTIN_SUBSYS_COUNT && \
277 (((ss) = cgroup_subsys[i]) || true); (i)++)
279 /* iterate each subsystem attached to a hierarchy */
280 #define for_each_root_subsys(root, ss) \
281 list_for_each_entry((ss), &(root)->subsys_list, sibling)
283 /* iterate across the active hierarchies */
284 #define for_each_active_root(root) \
285 list_for_each_entry((root), &cgroup_roots, root_list)
287 static inline struct cgroup
*__d_cgrp(struct dentry
*dentry
)
289 return dentry
->d_fsdata
;
292 static inline struct cfent
*__d_cfe(struct dentry
*dentry
)
294 return dentry
->d_fsdata
;
297 static inline struct cftype
*__d_cft(struct dentry
*dentry
)
299 return __d_cfe(dentry
)->type
;
303 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
304 * @cgrp: the cgroup to be checked for liveness
306 * On success, returns true; the mutex should be later unlocked. On
307 * failure returns false with no lock held.
309 static bool cgroup_lock_live_group(struct cgroup
*cgrp
)
311 mutex_lock(&cgroup_mutex
);
312 if (cgroup_is_dead(cgrp
)) {
313 mutex_unlock(&cgroup_mutex
);
319 /* the list of cgroups eligible for automatic release. Protected by
320 * release_list_lock */
321 static LIST_HEAD(release_list
);
322 static DEFINE_RAW_SPINLOCK(release_list_lock
);
323 static void cgroup_release_agent(struct work_struct
*work
);
324 static DECLARE_WORK(release_agent_work
, cgroup_release_agent
);
325 static void check_for_release(struct cgroup
*cgrp
);
328 * A cgroup can be associated with multiple css_sets as different tasks may
329 * belong to different cgroups on different hierarchies. In the other
330 * direction, a css_set is naturally associated with multiple cgroups.
331 * This M:N relationship is represented by the following link structure
332 * which exists for each association and allows traversing the associations
335 struct cgrp_cset_link
{
336 /* the cgroup and css_set this link associates */
338 struct css_set
*cset
;
340 /* list of cgrp_cset_links anchored at cgrp->cset_links */
341 struct list_head cset_link
;
343 /* list of cgrp_cset_links anchored at css_set->cgrp_links */
344 struct list_head cgrp_link
;
347 /* The default css_set - used by init and its children prior to any
348 * hierarchies being mounted. It contains a pointer to the root state
349 * for each subsystem. Also used to anchor the list of css_sets. Not
350 * reference-counted, to improve performance when child cgroups
351 * haven't been created.
354 static struct css_set init_css_set
;
355 static struct cgrp_cset_link init_cgrp_cset_link
;
357 static int cgroup_init_idr(struct cgroup_subsys
*ss
,
358 struct cgroup_subsys_state
*css
);
361 * css_set_lock protects the list of css_set objects, and the chain of
362 * tasks off each css_set. Nests outside task->alloc_lock due to
363 * css_task_iter_start().
365 static DEFINE_RWLOCK(css_set_lock
);
366 static int css_set_count
;
369 * hash table for cgroup groups. This improves the performance to find
370 * an existing css_set. This hash doesn't (currently) take into
371 * account cgroups in empty hierarchies.
373 #define CSS_SET_HASH_BITS 7
374 static DEFINE_HASHTABLE(css_set_table
, CSS_SET_HASH_BITS
);
376 static unsigned long css_set_hash(struct cgroup_subsys_state
*css
[])
378 unsigned long key
= 0UL;
379 struct cgroup_subsys
*ss
;
382 for_each_subsys(ss
, i
)
383 key
+= (unsigned long)css
[i
];
384 key
= (key
>> 16) ^ key
;
390 * We don't maintain the lists running through each css_set to its task
391 * until after the first call to css_task_iter_start(). This reduces the
392 * fork()/exit() overhead for people who have cgroups compiled into their
393 * kernel but not actually in use.
395 static int use_task_css_set_links __read_mostly
;
397 static void __put_css_set(struct css_set
*cset
, int taskexit
)
399 struct cgrp_cset_link
*link
, *tmp_link
;
402 * Ensure that the refcount doesn't hit zero while any readers
403 * can see it. Similar to atomic_dec_and_lock(), but for an
406 if (atomic_add_unless(&cset
->refcount
, -1, 1))
408 write_lock(&css_set_lock
);
409 if (!atomic_dec_and_test(&cset
->refcount
)) {
410 write_unlock(&css_set_lock
);
414 /* This css_set is dead. unlink it and release cgroup refcounts */
415 hash_del(&cset
->hlist
);
418 list_for_each_entry_safe(link
, tmp_link
, &cset
->cgrp_links
, cgrp_link
) {
419 struct cgroup
*cgrp
= link
->cgrp
;
421 list_del(&link
->cset_link
);
422 list_del(&link
->cgrp_link
);
424 /* @cgrp can't go away while we're holding css_set_lock */
425 if (list_empty(&cgrp
->cset_links
) && notify_on_release(cgrp
)) {
427 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
428 check_for_release(cgrp
);
434 write_unlock(&css_set_lock
);
435 kfree_rcu(cset
, rcu_head
);
439 * refcounted get/put for css_set objects
441 static inline void get_css_set(struct css_set
*cset
)
443 atomic_inc(&cset
->refcount
);
446 static inline void put_css_set(struct css_set
*cset
)
448 __put_css_set(cset
, 0);
451 static inline void put_css_set_taskexit(struct css_set
*cset
)
453 __put_css_set(cset
, 1);
457 * compare_css_sets - helper function for find_existing_css_set().
458 * @cset: candidate css_set being tested
459 * @old_cset: existing css_set for a task
460 * @new_cgrp: cgroup that's being entered by the task
461 * @template: desired set of css pointers in css_set (pre-calculated)
463 * Returns true if "cset" matches "old_cset" except for the hierarchy
464 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
466 static bool compare_css_sets(struct css_set
*cset
,
467 struct css_set
*old_cset
,
468 struct cgroup
*new_cgrp
,
469 struct cgroup_subsys_state
*template[])
471 struct list_head
*l1
, *l2
;
473 if (memcmp(template, cset
->subsys
, sizeof(cset
->subsys
))) {
474 /* Not all subsystems matched */
479 * Compare cgroup pointers in order to distinguish between
480 * different cgroups in heirarchies with no subsystems. We
481 * could get by with just this check alone (and skip the
482 * memcmp above) but on most setups the memcmp check will
483 * avoid the need for this more expensive check on almost all
487 l1
= &cset
->cgrp_links
;
488 l2
= &old_cset
->cgrp_links
;
490 struct cgrp_cset_link
*link1
, *link2
;
491 struct cgroup
*cgrp1
, *cgrp2
;
495 /* See if we reached the end - both lists are equal length. */
496 if (l1
== &cset
->cgrp_links
) {
497 BUG_ON(l2
!= &old_cset
->cgrp_links
);
500 BUG_ON(l2
== &old_cset
->cgrp_links
);
502 /* Locate the cgroups associated with these links. */
503 link1
= list_entry(l1
, struct cgrp_cset_link
, cgrp_link
);
504 link2
= list_entry(l2
, struct cgrp_cset_link
, cgrp_link
);
507 /* Hierarchies should be linked in the same order. */
508 BUG_ON(cgrp1
->root
!= cgrp2
->root
);
511 * If this hierarchy is the hierarchy of the cgroup
512 * that's changing, then we need to check that this
513 * css_set points to the new cgroup; if it's any other
514 * hierarchy, then this css_set should point to the
515 * same cgroup as the old css_set.
517 if (cgrp1
->root
== new_cgrp
->root
) {
518 if (cgrp1
!= new_cgrp
)
529 * find_existing_css_set - init css array and find the matching css_set
530 * @old_cset: the css_set that we're using before the cgroup transition
531 * @cgrp: the cgroup that we're moving into
532 * @template: out param for the new set of csses, should be clear on entry
534 static struct css_set
*find_existing_css_set(struct css_set
*old_cset
,
536 struct cgroup_subsys_state
*template[])
538 struct cgroupfs_root
*root
= cgrp
->root
;
539 struct cgroup_subsys
*ss
;
540 struct css_set
*cset
;
545 * Build the set of subsystem state objects that we want to see in the
546 * new css_set. while subsystems can change globally, the entries here
547 * won't change, so no need for locking.
549 for_each_subsys(ss
, i
) {
550 if (root
->subsys_mask
& (1UL << i
)) {
551 /* Subsystem is in this hierarchy. So we want
552 * the subsystem state from the new
554 template[i
] = cgroup_css(cgrp
, ss
);
556 /* Subsystem is not in this hierarchy, so we
557 * don't want to change the subsystem state */
558 template[i
] = old_cset
->subsys
[i
];
562 key
= css_set_hash(template);
563 hash_for_each_possible(css_set_table
, cset
, hlist
, key
) {
564 if (!compare_css_sets(cset
, old_cset
, cgrp
, template))
567 /* This css_set matches what we need */
571 /* No existing cgroup group matched */
575 static void free_cgrp_cset_links(struct list_head
*links_to_free
)
577 struct cgrp_cset_link
*link
, *tmp_link
;
579 list_for_each_entry_safe(link
, tmp_link
, links_to_free
, cset_link
) {
580 list_del(&link
->cset_link
);
586 * allocate_cgrp_cset_links - allocate cgrp_cset_links
587 * @count: the number of links to allocate
588 * @tmp_links: list_head the allocated links are put on
590 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
591 * through ->cset_link. Returns 0 on success or -errno.
593 static int allocate_cgrp_cset_links(int count
, struct list_head
*tmp_links
)
595 struct cgrp_cset_link
*link
;
598 INIT_LIST_HEAD(tmp_links
);
600 for (i
= 0; i
< count
; i
++) {
601 link
= kzalloc(sizeof(*link
), GFP_KERNEL
);
603 free_cgrp_cset_links(tmp_links
);
606 list_add(&link
->cset_link
, tmp_links
);
612 * link_css_set - a helper function to link a css_set to a cgroup
613 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
614 * @cset: the css_set to be linked
615 * @cgrp: the destination cgroup
617 static void link_css_set(struct list_head
*tmp_links
, struct css_set
*cset
,
620 struct cgrp_cset_link
*link
;
622 BUG_ON(list_empty(tmp_links
));
623 link
= list_first_entry(tmp_links
, struct cgrp_cset_link
, cset_link
);
626 list_move(&link
->cset_link
, &cgrp
->cset_links
);
628 * Always add links to the tail of the list so that the list
629 * is sorted by order of hierarchy creation
631 list_add_tail(&link
->cgrp_link
, &cset
->cgrp_links
);
635 * find_css_set - return a new css_set with one cgroup updated
636 * @old_cset: the baseline css_set
637 * @cgrp: the cgroup to be updated
639 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
640 * substituted into the appropriate hierarchy.
642 static struct css_set
*find_css_set(struct css_set
*old_cset
,
645 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
] = { };
646 struct css_set
*cset
;
647 struct list_head tmp_links
;
648 struct cgrp_cset_link
*link
;
651 lockdep_assert_held(&cgroup_mutex
);
653 /* First see if we already have a cgroup group that matches
655 read_lock(&css_set_lock
);
656 cset
= find_existing_css_set(old_cset
, cgrp
, template);
659 read_unlock(&css_set_lock
);
664 cset
= kzalloc(sizeof(*cset
), GFP_KERNEL
);
668 /* Allocate all the cgrp_cset_link objects that we'll need */
669 if (allocate_cgrp_cset_links(cgroup_root_count
, &tmp_links
) < 0) {
674 atomic_set(&cset
->refcount
, 1);
675 INIT_LIST_HEAD(&cset
->cgrp_links
);
676 INIT_LIST_HEAD(&cset
->tasks
);
677 INIT_HLIST_NODE(&cset
->hlist
);
679 /* Copy the set of subsystem state objects generated in
680 * find_existing_css_set() */
681 memcpy(cset
->subsys
, template, sizeof(cset
->subsys
));
683 write_lock(&css_set_lock
);
684 /* Add reference counts and links from the new css_set. */
685 list_for_each_entry(link
, &old_cset
->cgrp_links
, cgrp_link
) {
686 struct cgroup
*c
= link
->cgrp
;
688 if (c
->root
== cgrp
->root
)
690 link_css_set(&tmp_links
, cset
, c
);
693 BUG_ON(!list_empty(&tmp_links
));
697 /* Add this cgroup group to the hash table */
698 key
= css_set_hash(cset
->subsys
);
699 hash_add(css_set_table
, &cset
->hlist
, key
);
701 write_unlock(&css_set_lock
);
707 * Return the cgroup for "task" from the given hierarchy. Must be
708 * called with cgroup_mutex held.
710 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
711 struct cgroupfs_root
*root
)
713 struct css_set
*cset
;
714 struct cgroup
*res
= NULL
;
716 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
717 read_lock(&css_set_lock
);
719 * No need to lock the task - since we hold cgroup_mutex the
720 * task can't change groups, so the only thing that can happen
721 * is that it exits and its css is set back to init_css_set.
723 cset
= task_css_set(task
);
724 if (cset
== &init_css_set
) {
725 res
= &root
->top_cgroup
;
727 struct cgrp_cset_link
*link
;
729 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
730 struct cgroup
*c
= link
->cgrp
;
732 if (c
->root
== root
) {
738 read_unlock(&css_set_lock
);
744 * There is one global cgroup mutex. We also require taking
745 * task_lock() when dereferencing a task's cgroup subsys pointers.
746 * See "The task_lock() exception", at the end of this comment.
748 * A task must hold cgroup_mutex to modify cgroups.
750 * Any task can increment and decrement the count field without lock.
751 * So in general, code holding cgroup_mutex can't rely on the count
752 * field not changing. However, if the count goes to zero, then only
753 * cgroup_attach_task() can increment it again. Because a count of zero
754 * means that no tasks are currently attached, therefore there is no
755 * way a task attached to that cgroup can fork (the other way to
756 * increment the count). So code holding cgroup_mutex can safely
757 * assume that if the count is zero, it will stay zero. Similarly, if
758 * a task holds cgroup_mutex on a cgroup with zero count, it
759 * knows that the cgroup won't be removed, as cgroup_rmdir()
762 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
763 * (usually) take cgroup_mutex. These are the two most performance
764 * critical pieces of code here. The exception occurs on cgroup_exit(),
765 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
766 * is taken, and if the cgroup count is zero, a usermode call made
767 * to the release agent with the name of the cgroup (path relative to
768 * the root of cgroup file system) as the argument.
770 * A cgroup can only be deleted if both its 'count' of using tasks
771 * is zero, and its list of 'children' cgroups is empty. Since all
772 * tasks in the system use _some_ cgroup, and since there is always at
773 * least one task in the system (init, pid == 1), therefore, top_cgroup
774 * always has either children cgroups and/or using tasks. So we don't
775 * need a special hack to ensure that top_cgroup cannot be deleted.
777 * The task_lock() exception
779 * The need for this exception arises from the action of
780 * cgroup_attach_task(), which overwrites one task's cgroup pointer with
781 * another. It does so using cgroup_mutex, however there are
782 * several performance critical places that need to reference
783 * task->cgroup without the expense of grabbing a system global
784 * mutex. Therefore except as noted below, when dereferencing or, as
785 * in cgroup_attach_task(), modifying a task's cgroup pointer we use
786 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
787 * the task_struct routinely used for such matters.
789 * P.S. One more locking exception. RCU is used to guard the
790 * update of a tasks cgroup pointer by cgroup_attach_task()
794 * A couple of forward declarations required, due to cyclic reference loop:
795 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
796 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
800 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
);
801 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
);
802 static int cgroup_populate_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
);
803 static const struct inode_operations cgroup_dir_inode_operations
;
804 static const struct file_operations proc_cgroupstats_operations
;
806 static struct backing_dev_info cgroup_backing_dev_info
= {
808 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
811 static int alloc_css_id(struct cgroup_subsys_state
*child_css
);
813 static struct inode
*cgroup_new_inode(umode_t mode
, struct super_block
*sb
)
815 struct inode
*inode
= new_inode(sb
);
818 inode
->i_ino
= get_next_ino();
819 inode
->i_mode
= mode
;
820 inode
->i_uid
= current_fsuid();
821 inode
->i_gid
= current_fsgid();
822 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
823 inode
->i_mapping
->backing_dev_info
= &cgroup_backing_dev_info
;
828 static struct cgroup_name
*cgroup_alloc_name(struct dentry
*dentry
)
830 struct cgroup_name
*name
;
832 name
= kmalloc(sizeof(*name
) + dentry
->d_name
.len
+ 1, GFP_KERNEL
);
835 strcpy(name
->name
, dentry
->d_name
.name
);
839 static void cgroup_free_fn(struct work_struct
*work
)
841 struct cgroup
*cgrp
= container_of(work
, struct cgroup
, destroy_work
);
843 mutex_lock(&cgroup_mutex
);
844 cgrp
->root
->number_of_cgroups
--;
845 mutex_unlock(&cgroup_mutex
);
848 * We get a ref to the parent's dentry, and put the ref when
849 * this cgroup is being freed, so it's guaranteed that the
850 * parent won't be destroyed before its children.
852 dput(cgrp
->parent
->dentry
);
855 * Drop the active superblock reference that we took when we
856 * created the cgroup. This will free cgrp->root, if we are
857 * holding the last reference to @sb.
859 deactivate_super(cgrp
->root
->sb
);
862 * if we're getting rid of the cgroup, refcount should ensure
863 * that there are no pidlists left.
865 BUG_ON(!list_empty(&cgrp
->pidlists
));
867 simple_xattrs_free(&cgrp
->xattrs
);
869 kfree(rcu_dereference_raw(cgrp
->name
));
873 static void cgroup_free_rcu(struct rcu_head
*head
)
875 struct cgroup
*cgrp
= container_of(head
, struct cgroup
, rcu_head
);
877 INIT_WORK(&cgrp
->destroy_work
, cgroup_free_fn
);
878 schedule_work(&cgrp
->destroy_work
);
881 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
883 /* is dentry a directory ? if so, kfree() associated cgroup */
884 if (S_ISDIR(inode
->i_mode
)) {
885 struct cgroup
*cgrp
= dentry
->d_fsdata
;
887 BUG_ON(!(cgroup_is_dead(cgrp
)));
888 call_rcu(&cgrp
->rcu_head
, cgroup_free_rcu
);
890 struct cfent
*cfe
= __d_cfe(dentry
);
891 struct cgroup
*cgrp
= dentry
->d_parent
->d_fsdata
;
893 WARN_ONCE(!list_empty(&cfe
->node
) &&
894 cgrp
!= &cgrp
->root
->top_cgroup
,
895 "cfe still linked for %s\n", cfe
->type
->name
);
896 simple_xattrs_free(&cfe
->xattrs
);
902 static int cgroup_delete(const struct dentry
*d
)
907 static void remove_dir(struct dentry
*d
)
909 struct dentry
*parent
= dget(d
->d_parent
);
912 simple_rmdir(parent
->d_inode
, d
);
916 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
920 lockdep_assert_held(&cgrp
->dentry
->d_inode
->i_mutex
);
921 lockdep_assert_held(&cgroup_mutex
);
924 * If we're doing cleanup due to failure of cgroup_create(),
925 * the corresponding @cfe may not exist.
927 list_for_each_entry(cfe
, &cgrp
->files
, node
) {
928 struct dentry
*d
= cfe
->dentry
;
930 if (cft
&& cfe
->type
!= cft
)
935 simple_unlink(cgrp
->dentry
->d_inode
, d
);
936 list_del_init(&cfe
->node
);
944 * cgroup_clear_dir - remove subsys files in a cgroup directory
945 * @cgrp: target cgroup
946 * @subsys_mask: mask of the subsystem ids whose files should be removed
948 static void cgroup_clear_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
)
950 struct cgroup_subsys
*ss
;
953 for_each_subsys(ss
, i
) {
954 struct cftype_set
*set
;
956 if (!test_bit(i
, &subsys_mask
))
958 list_for_each_entry(set
, &ss
->cftsets
, node
)
959 cgroup_addrm_files(cgrp
, set
->cfts
, false);
964 * NOTE : the dentry must have been dget()'ed
966 static void cgroup_d_remove_dir(struct dentry
*dentry
)
968 struct dentry
*parent
;
970 parent
= dentry
->d_parent
;
971 spin_lock(&parent
->d_lock
);
972 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
973 list_del_init(&dentry
->d_u
.d_child
);
974 spin_unlock(&dentry
->d_lock
);
975 spin_unlock(&parent
->d_lock
);
980 * Call with cgroup_mutex held. Drops reference counts on modules, including
981 * any duplicate ones that parse_cgroupfs_options took. If this function
982 * returns an error, no reference counts are touched.
984 static int rebind_subsystems(struct cgroupfs_root
*root
,
985 unsigned long added_mask
, unsigned removed_mask
)
987 struct cgroup
*cgrp
= &root
->top_cgroup
;
988 struct cgroup_subsys
*ss
;
989 unsigned long pinned
= 0;
992 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
993 BUG_ON(!mutex_is_locked(&cgroup_root_mutex
));
995 /* Check that any added subsystems are currently free */
996 for_each_subsys(ss
, i
) {
997 if (!(added_mask
& (1 << i
)))
1000 /* is the subsystem mounted elsewhere? */
1001 if (ss
->root
!= &cgroup_dummy_root
) {
1006 /* pin the module */
1007 if (!try_module_get(ss
->module
)) {
1014 /* subsys could be missing if unloaded between parsing and here */
1015 if (added_mask
!= pinned
) {
1020 ret
= cgroup_populate_dir(cgrp
, added_mask
);
1025 * Nothing can fail from this point on. Remove files for the
1026 * removed subsystems and rebind each subsystem.
1028 cgroup_clear_dir(cgrp
, removed_mask
);
1030 for_each_subsys(ss
, i
) {
1031 unsigned long bit
= 1UL << i
;
1033 if (bit
& added_mask
) {
1034 /* We're binding this subsystem to this hierarchy */
1035 BUG_ON(cgroup_css(cgrp
, ss
));
1036 BUG_ON(!cgroup_css(cgroup_dummy_top
, ss
));
1037 BUG_ON(cgroup_css(cgroup_dummy_top
, ss
)->cgroup
!= cgroup_dummy_top
);
1039 rcu_assign_pointer(cgrp
->subsys
[i
],
1040 cgroup_css(cgroup_dummy_top
, ss
));
1041 cgroup_css(cgrp
, ss
)->cgroup
= cgrp
;
1043 list_move(&ss
->sibling
, &root
->subsys_list
);
1046 ss
->bind(cgroup_css(cgrp
, ss
));
1048 /* refcount was already taken, and we're keeping it */
1049 root
->subsys_mask
|= bit
;
1050 } else if (bit
& removed_mask
) {
1051 /* We're removing this subsystem */
1052 BUG_ON(cgroup_css(cgrp
, ss
) != cgroup_css(cgroup_dummy_top
, ss
));
1053 BUG_ON(cgroup_css(cgrp
, ss
)->cgroup
!= cgrp
);
1056 ss
->bind(cgroup_css(cgroup_dummy_top
, ss
));
1058 cgroup_css(cgroup_dummy_top
, ss
)->cgroup
= cgroup_dummy_top
;
1059 RCU_INIT_POINTER(cgrp
->subsys
[i
], NULL
);
1061 cgroup_subsys
[i
]->root
= &cgroup_dummy_root
;
1062 list_move(&ss
->sibling
, &cgroup_dummy_root
.subsys_list
);
1064 /* subsystem is now free - drop reference on module */
1065 module_put(ss
->module
);
1066 root
->subsys_mask
&= ~bit
;
1071 * Mark @root has finished binding subsystems. @root->subsys_mask
1072 * now matches the bound subsystems.
1074 root
->flags
|= CGRP_ROOT_SUBSYS_BOUND
;
1079 for_each_subsys(ss
, i
)
1080 if (pinned
& (1 << i
))
1081 module_put(ss
->module
);
1085 static int cgroup_show_options(struct seq_file
*seq
, struct dentry
*dentry
)
1087 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
1088 struct cgroup_subsys
*ss
;
1090 mutex_lock(&cgroup_root_mutex
);
1091 for_each_root_subsys(root
, ss
)
1092 seq_printf(seq
, ",%s", ss
->name
);
1093 if (root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
)
1094 seq_puts(seq
, ",sane_behavior");
1095 if (root
->flags
& CGRP_ROOT_NOPREFIX
)
1096 seq_puts(seq
, ",noprefix");
1097 if (root
->flags
& CGRP_ROOT_XATTR
)
1098 seq_puts(seq
, ",xattr");
1099 if (strlen(root
->release_agent_path
))
1100 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1101 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->top_cgroup
.flags
))
1102 seq_puts(seq
, ",clone_children");
1103 if (strlen(root
->name
))
1104 seq_printf(seq
, ",name=%s", root
->name
);
1105 mutex_unlock(&cgroup_root_mutex
);
1109 struct cgroup_sb_opts
{
1110 unsigned long subsys_mask
;
1111 unsigned long flags
;
1112 char *release_agent
;
1113 bool cpuset_clone_children
;
1115 /* User explicitly requested empty subsystem */
1118 struct cgroupfs_root
*new_root
;
1123 * Convert a hierarchy specifier into a bitmask of subsystems and
1124 * flags. Call with cgroup_mutex held to protect the cgroup_subsys[]
1125 * array. This function takes refcounts on subsystems to be used, unless it
1126 * returns error, in which case no refcounts are taken.
1128 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1130 char *token
, *o
= data
;
1131 bool all_ss
= false, one_ss
= false;
1132 unsigned long mask
= (unsigned long)-1;
1133 struct cgroup_subsys
*ss
;
1136 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1138 #ifdef CONFIG_CPUSETS
1139 mask
= ~(1UL << cpuset_subsys_id
);
1142 memset(opts
, 0, sizeof(*opts
));
1144 while ((token
= strsep(&o
, ",")) != NULL
) {
1147 if (!strcmp(token
, "none")) {
1148 /* Explicitly have no subsystems */
1152 if (!strcmp(token
, "all")) {
1153 /* Mutually exclusive option 'all' + subsystem name */
1159 if (!strcmp(token
, "__DEVEL__sane_behavior")) {
1160 opts
->flags
|= CGRP_ROOT_SANE_BEHAVIOR
;
1163 if (!strcmp(token
, "noprefix")) {
1164 opts
->flags
|= CGRP_ROOT_NOPREFIX
;
1167 if (!strcmp(token
, "clone_children")) {
1168 opts
->cpuset_clone_children
= true;
1171 if (!strcmp(token
, "xattr")) {
1172 opts
->flags
|= CGRP_ROOT_XATTR
;
1175 if (!strncmp(token
, "release_agent=", 14)) {
1176 /* Specifying two release agents is forbidden */
1177 if (opts
->release_agent
)
1179 opts
->release_agent
=
1180 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1181 if (!opts
->release_agent
)
1185 if (!strncmp(token
, "name=", 5)) {
1186 const char *name
= token
+ 5;
1187 /* Can't specify an empty name */
1190 /* Must match [\w.-]+ */
1191 for (i
= 0; i
< strlen(name
); i
++) {
1195 if ((c
== '.') || (c
== '-') || (c
== '_'))
1199 /* Specifying two names is forbidden */
1202 opts
->name
= kstrndup(name
,
1203 MAX_CGROUP_ROOT_NAMELEN
- 1,
1211 for_each_subsys(ss
, i
) {
1212 if (strcmp(token
, ss
->name
))
1217 /* Mutually exclusive option 'all' + subsystem name */
1220 set_bit(i
, &opts
->subsys_mask
);
1225 if (i
== CGROUP_SUBSYS_COUNT
)
1230 * If the 'all' option was specified select all the subsystems,
1231 * otherwise if 'none', 'name=' and a subsystem name options
1232 * were not specified, let's default to 'all'
1234 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
))
1235 for_each_subsys(ss
, i
)
1237 set_bit(i
, &opts
->subsys_mask
);
1239 /* Consistency checks */
1241 if (opts
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1242 pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1244 if (opts
->flags
& CGRP_ROOT_NOPREFIX
) {
1245 pr_err("cgroup: sane_behavior: noprefix is not allowed\n");
1249 if (opts
->cpuset_clone_children
) {
1250 pr_err("cgroup: sane_behavior: clone_children is not allowed\n");
1256 * Option noprefix was introduced just for backward compatibility
1257 * with the old cpuset, so we allow noprefix only if mounting just
1258 * the cpuset subsystem.
1260 if ((opts
->flags
& CGRP_ROOT_NOPREFIX
) && (opts
->subsys_mask
& mask
))
1264 /* Can't specify "none" and some subsystems */
1265 if (opts
->subsys_mask
&& opts
->none
)
1269 * We either have to specify by name or by subsystems. (So all
1270 * empty hierarchies must have a name).
1272 if (!opts
->subsys_mask
&& !opts
->name
)
1278 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
1281 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1282 struct cgroup
*cgrp
= &root
->top_cgroup
;
1283 struct cgroup_sb_opts opts
;
1284 unsigned long added_mask
, removed_mask
;
1286 if (root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1287 pr_err("cgroup: sane_behavior: remount is not allowed\n");
1291 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1292 mutex_lock(&cgroup_mutex
);
1293 mutex_lock(&cgroup_root_mutex
);
1295 /* See what subsystems are wanted */
1296 ret
= parse_cgroupfs_options(data
, &opts
);
1300 if (opts
.subsys_mask
!= root
->subsys_mask
|| opts
.release_agent
)
1301 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1302 task_tgid_nr(current
), current
->comm
);
1304 added_mask
= opts
.subsys_mask
& ~root
->subsys_mask
;
1305 removed_mask
= root
->subsys_mask
& ~opts
.subsys_mask
;
1307 /* Don't allow flags or name to change at remount */
1308 if (((opts
.flags
^ root
->flags
) & CGRP_ROOT_OPTION_MASK
) ||
1309 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1310 pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n",
1311 opts
.flags
& CGRP_ROOT_OPTION_MASK
, opts
.name
?: "",
1312 root
->flags
& CGRP_ROOT_OPTION_MASK
, root
->name
);
1317 /* remounting is not allowed for populated hierarchies */
1318 if (root
->number_of_cgroups
> 1) {
1323 ret
= rebind_subsystems(root
, added_mask
, removed_mask
);
1327 if (opts
.release_agent
)
1328 strcpy(root
->release_agent_path
, opts
.release_agent
);
1330 kfree(opts
.release_agent
);
1332 mutex_unlock(&cgroup_root_mutex
);
1333 mutex_unlock(&cgroup_mutex
);
1334 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1338 static const struct super_operations cgroup_ops
= {
1339 .statfs
= simple_statfs
,
1340 .drop_inode
= generic_delete_inode
,
1341 .show_options
= cgroup_show_options
,
1342 .remount_fs
= cgroup_remount
,
1345 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1347 INIT_LIST_HEAD(&cgrp
->sibling
);
1348 INIT_LIST_HEAD(&cgrp
->children
);
1349 INIT_LIST_HEAD(&cgrp
->files
);
1350 INIT_LIST_HEAD(&cgrp
->cset_links
);
1351 INIT_LIST_HEAD(&cgrp
->release_list
);
1352 INIT_LIST_HEAD(&cgrp
->pidlists
);
1353 mutex_init(&cgrp
->pidlist_mutex
);
1354 cgrp
->dummy_css
.cgroup
= cgrp
;
1355 simple_xattrs_init(&cgrp
->xattrs
);
1358 static void init_cgroup_root(struct cgroupfs_root
*root
)
1360 struct cgroup
*cgrp
= &root
->top_cgroup
;
1362 INIT_LIST_HEAD(&root
->subsys_list
);
1363 INIT_LIST_HEAD(&root
->root_list
);
1364 root
->number_of_cgroups
= 1;
1366 RCU_INIT_POINTER(cgrp
->name
, &root_cgroup_name
);
1367 init_cgroup_housekeeping(cgrp
);
1368 idr_init(&root
->cgroup_idr
);
1371 static int cgroup_init_root_id(struct cgroupfs_root
*root
, int start
, int end
)
1375 lockdep_assert_held(&cgroup_mutex
);
1376 lockdep_assert_held(&cgroup_root_mutex
);
1378 id
= idr_alloc_cyclic(&cgroup_hierarchy_idr
, root
, start
, end
,
1383 root
->hierarchy_id
= id
;
1387 static void cgroup_exit_root_id(struct cgroupfs_root
*root
)
1389 lockdep_assert_held(&cgroup_mutex
);
1390 lockdep_assert_held(&cgroup_root_mutex
);
1392 if (root
->hierarchy_id
) {
1393 idr_remove(&cgroup_hierarchy_idr
, root
->hierarchy_id
);
1394 root
->hierarchy_id
= 0;
1398 static int cgroup_test_super(struct super_block
*sb
, void *data
)
1400 struct cgroup_sb_opts
*opts
= data
;
1401 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1403 /* If we asked for a name then it must match */
1404 if (opts
->name
&& strcmp(opts
->name
, root
->name
))
1408 * If we asked for subsystems (or explicitly for no
1409 * subsystems) then they must match
1411 if ((opts
->subsys_mask
|| opts
->none
)
1412 && (opts
->subsys_mask
!= root
->subsys_mask
))
1418 static struct cgroupfs_root
*cgroup_root_from_opts(struct cgroup_sb_opts
*opts
)
1420 struct cgroupfs_root
*root
;
1422 if (!opts
->subsys_mask
&& !opts
->none
)
1425 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1427 return ERR_PTR(-ENOMEM
);
1429 init_cgroup_root(root
);
1432 * We need to set @root->subsys_mask now so that @root can be
1433 * matched by cgroup_test_super() before it finishes
1434 * initialization; otherwise, competing mounts with the same
1435 * options may try to bind the same subsystems instead of waiting
1436 * for the first one leading to unexpected mount errors.
1437 * SUBSYS_BOUND will be set once actual binding is complete.
1439 root
->subsys_mask
= opts
->subsys_mask
;
1440 root
->flags
= opts
->flags
;
1441 if (opts
->release_agent
)
1442 strcpy(root
->release_agent_path
, opts
->release_agent
);
1444 strcpy(root
->name
, opts
->name
);
1445 if (opts
->cpuset_clone_children
)
1446 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->top_cgroup
.flags
);
1450 static void cgroup_free_root(struct cgroupfs_root
*root
)
1453 /* hierarhcy ID shoulid already have been released */
1454 WARN_ON_ONCE(root
->hierarchy_id
);
1456 idr_destroy(&root
->cgroup_idr
);
1461 static int cgroup_set_super(struct super_block
*sb
, void *data
)
1464 struct cgroup_sb_opts
*opts
= data
;
1466 /* If we don't have a new root, we can't set up a new sb */
1467 if (!opts
->new_root
)
1470 BUG_ON(!opts
->subsys_mask
&& !opts
->none
);
1472 ret
= set_anon_super(sb
, NULL
);
1476 sb
->s_fs_info
= opts
->new_root
;
1477 opts
->new_root
->sb
= sb
;
1479 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
1480 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
1481 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
1482 sb
->s_op
= &cgroup_ops
;
1487 static int cgroup_get_rootdir(struct super_block
*sb
)
1489 static const struct dentry_operations cgroup_dops
= {
1490 .d_iput
= cgroup_diput
,
1491 .d_delete
= cgroup_delete
,
1494 struct inode
*inode
=
1495 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
1500 inode
->i_fop
= &simple_dir_operations
;
1501 inode
->i_op
= &cgroup_dir_inode_operations
;
1502 /* directories start off with i_nlink == 2 (for "." entry) */
1504 sb
->s_root
= d_make_root(inode
);
1507 /* for everything else we want ->d_op set */
1508 sb
->s_d_op
= &cgroup_dops
;
1512 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1513 int flags
, const char *unused_dev_name
,
1516 struct cgroup_sb_opts opts
;
1517 struct cgroupfs_root
*root
;
1519 struct super_block
*sb
;
1520 struct cgroupfs_root
*new_root
;
1521 struct list_head tmp_links
;
1522 struct inode
*inode
;
1523 const struct cred
*cred
;
1525 /* First find the desired set of subsystems */
1526 mutex_lock(&cgroup_mutex
);
1527 ret
= parse_cgroupfs_options(data
, &opts
);
1528 mutex_unlock(&cgroup_mutex
);
1533 * Allocate a new cgroup root. We may not need it if we're
1534 * reusing an existing hierarchy.
1536 new_root
= cgroup_root_from_opts(&opts
);
1537 if (IS_ERR(new_root
)) {
1538 ret
= PTR_ERR(new_root
);
1541 opts
.new_root
= new_root
;
1543 /* Locate an existing or new sb for this hierarchy */
1544 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, 0, &opts
);
1547 cgroup_free_root(opts
.new_root
);
1551 root
= sb
->s_fs_info
;
1553 if (root
== opts
.new_root
) {
1554 /* We used the new root structure, so this is a new hierarchy */
1555 struct cgroup
*root_cgrp
= &root
->top_cgroup
;
1556 struct cgroupfs_root
*existing_root
;
1558 struct css_set
*cset
;
1560 BUG_ON(sb
->s_root
!= NULL
);
1562 ret
= cgroup_get_rootdir(sb
);
1564 goto drop_new_super
;
1565 inode
= sb
->s_root
->d_inode
;
1567 mutex_lock(&inode
->i_mutex
);
1568 mutex_lock(&cgroup_mutex
);
1569 mutex_lock(&cgroup_root_mutex
);
1571 root_cgrp
->id
= idr_alloc(&root
->cgroup_idr
, root_cgrp
,
1573 if (root_cgrp
->id
< 0)
1576 /* Check for name clashes with existing mounts */
1578 if (strlen(root
->name
))
1579 for_each_active_root(existing_root
)
1580 if (!strcmp(existing_root
->name
, root
->name
))
1584 * We're accessing css_set_count without locking
1585 * css_set_lock here, but that's OK - it can only be
1586 * increased by someone holding cgroup_lock, and
1587 * that's us. The worst that can happen is that we
1588 * have some link structures left over
1590 ret
= allocate_cgrp_cset_links(css_set_count
, &tmp_links
);
1594 /* ID 0 is reserved for dummy root, 1 for unified hierarchy */
1595 ret
= cgroup_init_root_id(root
, 2, 0);
1599 sb
->s_root
->d_fsdata
= root_cgrp
;
1600 root_cgrp
->dentry
= sb
->s_root
;
1603 * We're inside get_sb() and will call lookup_one_len() to
1604 * create the root files, which doesn't work if SELinux is
1605 * in use. The following cred dancing somehow works around
1606 * it. See 2ce9738ba ("cgroupfs: use init_cred when
1607 * populating new cgroupfs mount") for more details.
1609 cred
= override_creds(&init_cred
);
1611 ret
= cgroup_addrm_files(root_cgrp
, cgroup_base_files
, true);
1615 ret
= rebind_subsystems(root
, root
->subsys_mask
, 0);
1622 * There must be no failure case after here, since rebinding
1623 * takes care of subsystems' refcounts, which are explicitly
1624 * dropped in the failure exit path.
1627 list_add(&root
->root_list
, &cgroup_roots
);
1628 cgroup_root_count
++;
1630 /* Link the top cgroup in this hierarchy into all
1631 * the css_set objects */
1632 write_lock(&css_set_lock
);
1633 hash_for_each(css_set_table
, i
, cset
, hlist
)
1634 link_css_set(&tmp_links
, cset
, root_cgrp
);
1635 write_unlock(&css_set_lock
);
1637 free_cgrp_cset_links(&tmp_links
);
1639 BUG_ON(!list_empty(&root_cgrp
->children
));
1640 BUG_ON(root
->number_of_cgroups
!= 1);
1642 mutex_unlock(&cgroup_root_mutex
);
1643 mutex_unlock(&cgroup_mutex
);
1644 mutex_unlock(&inode
->i_mutex
);
1647 * We re-used an existing hierarchy - the new root (if
1648 * any) is not needed
1650 cgroup_free_root(opts
.new_root
);
1652 if ((root
->flags
^ opts
.flags
) & CGRP_ROOT_OPTION_MASK
) {
1653 if ((root
->flags
| opts
.flags
) & CGRP_ROOT_SANE_BEHAVIOR
) {
1654 pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n");
1656 goto drop_new_super
;
1658 pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n");
1663 kfree(opts
.release_agent
);
1665 return dget(sb
->s_root
);
1668 free_cgrp_cset_links(&tmp_links
);
1669 cgroup_addrm_files(&root
->top_cgroup
, cgroup_base_files
, false);
1672 cgroup_exit_root_id(root
);
1673 mutex_unlock(&cgroup_root_mutex
);
1674 mutex_unlock(&cgroup_mutex
);
1675 mutex_unlock(&inode
->i_mutex
);
1677 deactivate_locked_super(sb
);
1679 kfree(opts
.release_agent
);
1681 return ERR_PTR(ret
);
1684 static void cgroup_kill_sb(struct super_block
*sb
) {
1685 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1686 struct cgroup
*cgrp
= &root
->top_cgroup
;
1687 struct cgrp_cset_link
*link
, *tmp_link
;
1692 BUG_ON(root
->number_of_cgroups
!= 1);
1693 BUG_ON(!list_empty(&cgrp
->children
));
1695 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1696 mutex_lock(&cgroup_mutex
);
1697 mutex_lock(&cgroup_root_mutex
);
1699 /* Rebind all subsystems back to the default hierarchy */
1700 if (root
->flags
& CGRP_ROOT_SUBSYS_BOUND
) {
1701 ret
= rebind_subsystems(root
, 0, root
->subsys_mask
);
1702 /* Shouldn't be able to fail ... */
1707 * Release all the links from cset_links to this hierarchy's
1710 write_lock(&css_set_lock
);
1712 list_for_each_entry_safe(link
, tmp_link
, &cgrp
->cset_links
, cset_link
) {
1713 list_del(&link
->cset_link
);
1714 list_del(&link
->cgrp_link
);
1717 write_unlock(&css_set_lock
);
1719 if (!list_empty(&root
->root_list
)) {
1720 list_del(&root
->root_list
);
1721 cgroup_root_count
--;
1724 cgroup_exit_root_id(root
);
1726 mutex_unlock(&cgroup_root_mutex
);
1727 mutex_unlock(&cgroup_mutex
);
1728 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1730 simple_xattrs_free(&cgrp
->xattrs
);
1732 kill_litter_super(sb
);
1733 cgroup_free_root(root
);
1736 static struct file_system_type cgroup_fs_type
= {
1738 .mount
= cgroup_mount
,
1739 .kill_sb
= cgroup_kill_sb
,
1742 static struct kobject
*cgroup_kobj
;
1745 * cgroup_path - generate the path of a cgroup
1746 * @cgrp: the cgroup in question
1747 * @buf: the buffer to write the path into
1748 * @buflen: the length of the buffer
1750 * Writes path of cgroup into buf. Returns 0 on success, -errno on error.
1752 * We can't generate cgroup path using dentry->d_name, as accessing
1753 * dentry->name must be protected by irq-unsafe dentry->d_lock or parent
1754 * inode's i_mutex, while on the other hand cgroup_path() can be called
1755 * with some irq-safe spinlocks held.
1757 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1759 int ret
= -ENAMETOOLONG
;
1762 if (!cgrp
->parent
) {
1763 if (strlcpy(buf
, "/", buflen
) >= buflen
)
1764 return -ENAMETOOLONG
;
1768 start
= buf
+ buflen
- 1;
1773 const char *name
= cgroup_name(cgrp
);
1777 if ((start
-= len
) < buf
)
1779 memcpy(start
, name
, len
);
1785 cgrp
= cgrp
->parent
;
1786 } while (cgrp
->parent
);
1788 memmove(buf
, start
, buf
+ buflen
- start
);
1793 EXPORT_SYMBOL_GPL(cgroup_path
);
1796 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
1797 * @task: target task
1798 * @buf: the buffer to write the path into
1799 * @buflen: the length of the buffer
1801 * Determine @task's cgroup on the first (the one with the lowest non-zero
1802 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
1803 * function grabs cgroup_mutex and shouldn't be used inside locks used by
1804 * cgroup controller callbacks.
1806 * Returns 0 on success, fails with -%ENAMETOOLONG if @buflen is too short.
1808 int task_cgroup_path(struct task_struct
*task
, char *buf
, size_t buflen
)
1810 struct cgroupfs_root
*root
;
1811 struct cgroup
*cgrp
;
1812 int hierarchy_id
= 1, ret
= 0;
1815 return -ENAMETOOLONG
;
1817 mutex_lock(&cgroup_mutex
);
1819 root
= idr_get_next(&cgroup_hierarchy_idr
, &hierarchy_id
);
1822 cgrp
= task_cgroup_from_root(task
, root
);
1823 ret
= cgroup_path(cgrp
, buf
, buflen
);
1825 /* if no hierarchy exists, everyone is in "/" */
1826 memcpy(buf
, "/", 2);
1829 mutex_unlock(&cgroup_mutex
);
1832 EXPORT_SYMBOL_GPL(task_cgroup_path
);
1835 * Control Group taskset
1837 struct task_and_cgroup
{
1838 struct task_struct
*task
;
1839 struct cgroup
*cgrp
;
1840 struct css_set
*cset
;
1843 struct cgroup_taskset
{
1844 struct task_and_cgroup single
;
1845 struct flex_array
*tc_array
;
1848 struct cgroup
*cur_cgrp
;
1852 * cgroup_taskset_first - reset taskset and return the first task
1853 * @tset: taskset of interest
1855 * @tset iteration is initialized and the first task is returned.
1857 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
)
1859 if (tset
->tc_array
) {
1861 return cgroup_taskset_next(tset
);
1863 tset
->cur_cgrp
= tset
->single
.cgrp
;
1864 return tset
->single
.task
;
1867 EXPORT_SYMBOL_GPL(cgroup_taskset_first
);
1870 * cgroup_taskset_next - iterate to the next task in taskset
1871 * @tset: taskset of interest
1873 * Return the next task in @tset. Iteration must have been initialized
1874 * with cgroup_taskset_first().
1876 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
)
1878 struct task_and_cgroup
*tc
;
1880 if (!tset
->tc_array
|| tset
->idx
>= tset
->tc_array_len
)
1883 tc
= flex_array_get(tset
->tc_array
, tset
->idx
++);
1884 tset
->cur_cgrp
= tc
->cgrp
;
1887 EXPORT_SYMBOL_GPL(cgroup_taskset_next
);
1890 * cgroup_taskset_cur_css - return the matching css for the current task
1891 * @tset: taskset of interest
1892 * @subsys_id: the ID of the target subsystem
1894 * Return the css for the current (last returned) task of @tset for
1895 * subsystem specified by @subsys_id. This function must be preceded by
1896 * either cgroup_taskset_first() or cgroup_taskset_next().
1898 struct cgroup_subsys_state
*cgroup_taskset_cur_css(struct cgroup_taskset
*tset
,
1901 return cgroup_css(tset
->cur_cgrp
, cgroup_subsys
[subsys_id
]);
1903 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_css
);
1906 * cgroup_taskset_size - return the number of tasks in taskset
1907 * @tset: taskset of interest
1909 int cgroup_taskset_size(struct cgroup_taskset
*tset
)
1911 return tset
->tc_array
? tset
->tc_array_len
: 1;
1913 EXPORT_SYMBOL_GPL(cgroup_taskset_size
);
1917 * cgroup_task_migrate - move a task from one cgroup to another.
1919 * Must be called with cgroup_mutex and threadgroup locked.
1921 static void cgroup_task_migrate(struct cgroup
*old_cgrp
,
1922 struct task_struct
*tsk
,
1923 struct css_set
*new_cset
)
1925 struct css_set
*old_cset
;
1928 * We are synchronized through threadgroup_lock() against PF_EXITING
1929 * setting such that we can't race against cgroup_exit() changing the
1930 * css_set to init_css_set and dropping the old one.
1932 WARN_ON_ONCE(tsk
->flags
& PF_EXITING
);
1933 old_cset
= task_css_set(tsk
);
1936 rcu_assign_pointer(tsk
->cgroups
, new_cset
);
1939 /* Update the css_set linked lists if we're using them */
1940 write_lock(&css_set_lock
);
1941 if (!list_empty(&tsk
->cg_list
))
1942 list_move(&tsk
->cg_list
, &new_cset
->tasks
);
1943 write_unlock(&css_set_lock
);
1946 * We just gained a reference on old_cset by taking it from the
1947 * task. As trading it for new_cset is protected by cgroup_mutex,
1948 * we're safe to drop it here; it will be freed under RCU.
1950 set_bit(CGRP_RELEASABLE
, &old_cgrp
->flags
);
1951 put_css_set(old_cset
);
1955 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
1956 * @cgrp: the cgroup to attach to
1957 * @tsk: the task or the leader of the threadgroup to be attached
1958 * @threadgroup: attach the whole threadgroup?
1960 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
1961 * task_lock of @tsk or each thread in the threadgroup individually in turn.
1963 static int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
,
1966 int retval
, i
, group_size
;
1967 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
1968 struct cgroupfs_root
*root
= cgrp
->root
;
1969 /* threadgroup list cursor and array */
1970 struct task_struct
*leader
= tsk
;
1971 struct task_and_cgroup
*tc
;
1972 struct flex_array
*group
;
1973 struct cgroup_taskset tset
= { };
1976 * step 0: in order to do expensive, possibly blocking operations for
1977 * every thread, we cannot iterate the thread group list, since it needs
1978 * rcu or tasklist locked. instead, build an array of all threads in the
1979 * group - group_rwsem prevents new threads from appearing, and if
1980 * threads exit, this will just be an over-estimate.
1983 group_size
= get_nr_threads(tsk
);
1986 /* flex_array supports very large thread-groups better than kmalloc. */
1987 group
= flex_array_alloc(sizeof(*tc
), group_size
, GFP_KERNEL
);
1990 /* pre-allocate to guarantee space while iterating in rcu read-side. */
1991 retval
= flex_array_prealloc(group
, 0, group_size
, GFP_KERNEL
);
1993 goto out_free_group_list
;
1997 * Prevent freeing of tasks while we take a snapshot. Tasks that are
1998 * already PF_EXITING could be freed from underneath us unless we
1999 * take an rcu_read_lock.
2003 struct task_and_cgroup ent
;
2005 /* @tsk either already exited or can't exit until the end */
2006 if (tsk
->flags
& PF_EXITING
)
2009 /* as per above, nr_threads may decrease, but not increase. */
2010 BUG_ON(i
>= group_size
);
2012 ent
.cgrp
= task_cgroup_from_root(tsk
, root
);
2013 /* nothing to do if this task is already in the cgroup */
2014 if (ent
.cgrp
== cgrp
)
2017 * saying GFP_ATOMIC has no effect here because we did prealloc
2018 * earlier, but it's good form to communicate our expectations.
2020 retval
= flex_array_put(group
, i
, &ent
, GFP_ATOMIC
);
2021 BUG_ON(retval
!= 0);
2026 } while_each_thread(leader
, tsk
);
2028 /* remember the number of threads in the array for later. */
2030 tset
.tc_array
= group
;
2031 tset
.tc_array_len
= group_size
;
2033 /* methods shouldn't be called if no task is actually migrating */
2036 goto out_free_group_list
;
2039 * step 1: check that we can legitimately attach to the cgroup.
2041 for_each_root_subsys(root
, ss
) {
2042 struct cgroup_subsys_state
*css
= cgroup_css(cgrp
, ss
);
2044 if (ss
->can_attach
) {
2045 retval
= ss
->can_attach(css
, &tset
);
2048 goto out_cancel_attach
;
2054 * step 2: make sure css_sets exist for all threads to be migrated.
2055 * we use find_css_set, which allocates a new one if necessary.
2057 for (i
= 0; i
< group_size
; i
++) {
2058 struct css_set
*old_cset
;
2060 tc
= flex_array_get(group
, i
);
2061 old_cset
= task_css_set(tc
->task
);
2062 tc
->cset
= find_css_set(old_cset
, cgrp
);
2065 goto out_put_css_set_refs
;
2070 * step 3: now that we're guaranteed success wrt the css_sets,
2071 * proceed to move all tasks to the new cgroup. There are no
2072 * failure cases after here, so this is the commit point.
2074 for (i
= 0; i
< group_size
; i
++) {
2075 tc
= flex_array_get(group
, i
);
2076 cgroup_task_migrate(tc
->cgrp
, tc
->task
, tc
->cset
);
2078 /* nothing is sensitive to fork() after this point. */
2081 * step 4: do subsystem attach callbacks.
2083 for_each_root_subsys(root
, ss
) {
2084 struct cgroup_subsys_state
*css
= cgroup_css(cgrp
, ss
);
2087 ss
->attach(css
, &tset
);
2091 * step 5: success! and cleanup
2094 out_put_css_set_refs
:
2096 for (i
= 0; i
< group_size
; i
++) {
2097 tc
= flex_array_get(group
, i
);
2100 put_css_set(tc
->cset
);
2105 for_each_root_subsys(root
, ss
) {
2106 struct cgroup_subsys_state
*css
= cgroup_css(cgrp
, ss
);
2108 if (ss
== failed_ss
)
2110 if (ss
->cancel_attach
)
2111 ss
->cancel_attach(css
, &tset
);
2114 out_free_group_list
:
2115 flex_array_free(group
);
2120 * Find the task_struct of the task to attach by vpid and pass it along to the
2121 * function to attach either it or all tasks in its threadgroup. Will lock
2122 * cgroup_mutex and threadgroup; may take task_lock of task.
2124 static int attach_task_by_pid(struct cgroup
*cgrp
, u64 pid
, bool threadgroup
)
2126 struct task_struct
*tsk
;
2127 const struct cred
*cred
= current_cred(), *tcred
;
2130 if (!cgroup_lock_live_group(cgrp
))
2136 tsk
= find_task_by_vpid(pid
);
2140 goto out_unlock_cgroup
;
2143 * even if we're attaching all tasks in the thread group, we
2144 * only need to check permissions on one of them.
2146 tcred
= __task_cred(tsk
);
2147 if (!uid_eq(cred
->euid
, GLOBAL_ROOT_UID
) &&
2148 !uid_eq(cred
->euid
, tcred
->uid
) &&
2149 !uid_eq(cred
->euid
, tcred
->suid
)) {
2152 goto out_unlock_cgroup
;
2158 tsk
= tsk
->group_leader
;
2161 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2162 * trapped in a cpuset, or RT worker may be born in a cgroup
2163 * with no rt_runtime allocated. Just say no.
2165 if (tsk
== kthreadd_task
|| (tsk
->flags
& PF_NO_SETAFFINITY
)) {
2168 goto out_unlock_cgroup
;
2171 get_task_struct(tsk
);
2174 threadgroup_lock(tsk
);
2176 if (!thread_group_leader(tsk
)) {
2178 * a race with de_thread from another thread's exec()
2179 * may strip us of our leadership, if this happens,
2180 * there is no choice but to throw this task away and
2181 * try again; this is
2182 * "double-double-toil-and-trouble-check locking".
2184 threadgroup_unlock(tsk
);
2185 put_task_struct(tsk
);
2186 goto retry_find_task
;
2190 ret
= cgroup_attach_task(cgrp
, tsk
, threadgroup
);
2192 threadgroup_unlock(tsk
);
2194 put_task_struct(tsk
);
2196 mutex_unlock(&cgroup_mutex
);
2201 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2202 * @from: attach to all cgroups of a given task
2203 * @tsk: the task to be attached
2205 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
2207 struct cgroupfs_root
*root
;
2210 mutex_lock(&cgroup_mutex
);
2211 for_each_active_root(root
) {
2212 struct cgroup
*from_cgrp
= task_cgroup_from_root(from
, root
);
2214 retval
= cgroup_attach_task(from_cgrp
, tsk
, false);
2218 mutex_unlock(&cgroup_mutex
);
2222 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
2224 static int cgroup_tasks_write(struct cgroup_subsys_state
*css
,
2225 struct cftype
*cft
, u64 pid
)
2227 return attach_task_by_pid(css
->cgroup
, pid
, false);
2230 static int cgroup_procs_write(struct cgroup_subsys_state
*css
,
2231 struct cftype
*cft
, u64 tgid
)
2233 return attach_task_by_pid(css
->cgroup
, tgid
, true);
2236 static int cgroup_release_agent_write(struct cgroup_subsys_state
*css
,
2237 struct cftype
*cft
, const char *buffer
)
2239 BUILD_BUG_ON(sizeof(css
->cgroup
->root
->release_agent_path
) < PATH_MAX
);
2240 if (strlen(buffer
) >= PATH_MAX
)
2242 if (!cgroup_lock_live_group(css
->cgroup
))
2244 mutex_lock(&cgroup_root_mutex
);
2245 strcpy(css
->cgroup
->root
->release_agent_path
, buffer
);
2246 mutex_unlock(&cgroup_root_mutex
);
2247 mutex_unlock(&cgroup_mutex
);
2251 static int cgroup_release_agent_show(struct cgroup_subsys_state
*css
,
2252 struct cftype
*cft
, struct seq_file
*seq
)
2254 struct cgroup
*cgrp
= css
->cgroup
;
2256 if (!cgroup_lock_live_group(cgrp
))
2258 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2259 seq_putc(seq
, '\n');
2260 mutex_unlock(&cgroup_mutex
);
2264 static int cgroup_sane_behavior_show(struct cgroup_subsys_state
*css
,
2265 struct cftype
*cft
, struct seq_file
*seq
)
2267 seq_printf(seq
, "%d\n", cgroup_sane_behavior(css
->cgroup
));
2271 /* A buffer size big enough for numbers or short strings */
2272 #define CGROUP_LOCAL_BUFFER_SIZE 64
2274 static ssize_t
cgroup_write_X64(struct cgroup_subsys_state
*css
,
2275 struct cftype
*cft
, struct file
*file
,
2276 const char __user
*userbuf
, size_t nbytes
,
2277 loff_t
*unused_ppos
)
2279 char buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2285 if (nbytes
>= sizeof(buffer
))
2287 if (copy_from_user(buffer
, userbuf
, nbytes
))
2290 buffer
[nbytes
] = 0; /* nul-terminate */
2291 if (cft
->write_u64
) {
2292 u64 val
= simple_strtoull(strstrip(buffer
), &end
, 0);
2295 retval
= cft
->write_u64(css
, cft
, val
);
2297 s64 val
= simple_strtoll(strstrip(buffer
), &end
, 0);
2300 retval
= cft
->write_s64(css
, cft
, val
);
2307 static ssize_t
cgroup_write_string(struct cgroup_subsys_state
*css
,
2308 struct cftype
*cft
, struct file
*file
,
2309 const char __user
*userbuf
, size_t nbytes
,
2310 loff_t
*unused_ppos
)
2312 char local_buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2314 size_t max_bytes
= cft
->max_write_len
;
2315 char *buffer
= local_buffer
;
2318 max_bytes
= sizeof(local_buffer
) - 1;
2319 if (nbytes
>= max_bytes
)
2321 /* Allocate a dynamic buffer if we need one */
2322 if (nbytes
>= sizeof(local_buffer
)) {
2323 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
2327 if (nbytes
&& copy_from_user(buffer
, userbuf
, nbytes
)) {
2332 buffer
[nbytes
] = 0; /* nul-terminate */
2333 retval
= cft
->write_string(css
, cft
, strstrip(buffer
));
2337 if (buffer
!= local_buffer
)
2342 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
2343 size_t nbytes
, loff_t
*ppos
)
2345 struct cfent
*cfe
= __d_cfe(file
->f_dentry
);
2346 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2347 struct cgroup_subsys_state
*css
= cfe
->css
;
2350 return cft
->write(css
, cft
, file
, buf
, nbytes
, ppos
);
2351 if (cft
->write_u64
|| cft
->write_s64
)
2352 return cgroup_write_X64(css
, cft
, file
, buf
, nbytes
, ppos
);
2353 if (cft
->write_string
)
2354 return cgroup_write_string(css
, cft
, file
, buf
, nbytes
, ppos
);
2356 int ret
= cft
->trigger(css
, (unsigned int)cft
->private);
2357 return ret
? ret
: nbytes
;
2362 static ssize_t
cgroup_read_u64(struct cgroup_subsys_state
*css
,
2363 struct cftype
*cft
, struct file
*file
,
2364 char __user
*buf
, size_t nbytes
, loff_t
*ppos
)
2366 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2367 u64 val
= cft
->read_u64(css
, cft
);
2368 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
2370 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2373 static ssize_t
cgroup_read_s64(struct cgroup_subsys_state
*css
,
2374 struct cftype
*cft
, struct file
*file
,
2375 char __user
*buf
, size_t nbytes
, loff_t
*ppos
)
2377 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2378 s64 val
= cft
->read_s64(css
, cft
);
2379 int len
= sprintf(tmp
, "%lld\n", (long long) val
);
2381 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2384 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
2385 size_t nbytes
, loff_t
*ppos
)
2387 struct cfent
*cfe
= __d_cfe(file
->f_dentry
);
2388 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2389 struct cgroup_subsys_state
*css
= cfe
->css
;
2392 return cft
->read(css
, cft
, file
, buf
, nbytes
, ppos
);
2394 return cgroup_read_u64(css
, cft
, file
, buf
, nbytes
, ppos
);
2396 return cgroup_read_s64(css
, cft
, file
, buf
, nbytes
, ppos
);
2401 * seqfile ops/methods for returning structured data. Currently just
2402 * supports string->u64 maps, but can be extended in future.
2405 static int cgroup_map_add(struct cgroup_map_cb
*cb
, const char *key
, u64 value
)
2407 struct seq_file
*sf
= cb
->state
;
2408 return seq_printf(sf
, "%s %llu\n", key
, (unsigned long long)value
);
2411 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2413 struct cfent
*cfe
= m
->private;
2414 struct cftype
*cft
= cfe
->type
;
2415 struct cgroup_subsys_state
*css
= cfe
->css
;
2417 if (cft
->read_map
) {
2418 struct cgroup_map_cb cb
= {
2419 .fill
= cgroup_map_add
,
2422 return cft
->read_map(css
, cft
, &cb
);
2424 return cft
->read_seq_string(css
, cft
, m
);
2427 static const struct file_operations cgroup_seqfile_operations
= {
2429 .write
= cgroup_file_write
,
2430 .llseek
= seq_lseek
,
2431 .release
= single_release
,
2434 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
2436 struct cfent
*cfe
= __d_cfe(file
->f_dentry
);
2437 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2438 struct cgroup
*cgrp
= __d_cgrp(cfe
->dentry
->d_parent
);
2439 struct cgroup_subsys_state
*css
;
2442 err
= generic_file_open(inode
, file
);
2447 * If the file belongs to a subsystem, pin the css. Will be
2448 * unpinned either on open failure or release. This ensures that
2449 * @css stays alive for all file operations.
2452 css
= cgroup_css(cgrp
, cft
->ss
);
2453 if (cft
->ss
&& !css_tryget(css
))
2461 * @cfe->css is used by read/write/close to determine the
2462 * associated css. @file->private_data would be a better place but
2463 * that's already used by seqfile. Multiple accessors may use it
2464 * simultaneously which is okay as the association never changes.
2466 WARN_ON_ONCE(cfe
->css
&& cfe
->css
!= css
);
2469 if (cft
->read_map
|| cft
->read_seq_string
) {
2470 file
->f_op
= &cgroup_seqfile_operations
;
2471 err
= single_open(file
, cgroup_seqfile_show
, cfe
);
2472 } else if (cft
->open
) {
2473 err
= cft
->open(inode
, file
);
2481 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
2483 struct cfent
*cfe
= __d_cfe(file
->f_dentry
);
2484 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2485 struct cgroup_subsys_state
*css
= cfe
->css
;
2489 ret
= cft
->release(inode
, file
);
2496 * cgroup_rename - Only allow simple rename of directories in place.
2498 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2499 struct inode
*new_dir
, struct dentry
*new_dentry
)
2502 struct cgroup_name
*name
, *old_name
;
2503 struct cgroup
*cgrp
;
2506 * It's convinient to use parent dir's i_mutex to protected
2509 lockdep_assert_held(&old_dir
->i_mutex
);
2511 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
2513 if (new_dentry
->d_inode
)
2515 if (old_dir
!= new_dir
)
2518 cgrp
= __d_cgrp(old_dentry
);
2521 * This isn't a proper migration and its usefulness is very
2522 * limited. Disallow if sane_behavior.
2524 if (cgroup_sane_behavior(cgrp
))
2527 name
= cgroup_alloc_name(new_dentry
);
2531 ret
= simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2537 old_name
= rcu_dereference_protected(cgrp
->name
, true);
2538 rcu_assign_pointer(cgrp
->name
, name
);
2540 kfree_rcu(old_name
, rcu_head
);
2544 static struct simple_xattrs
*__d_xattrs(struct dentry
*dentry
)
2546 if (S_ISDIR(dentry
->d_inode
->i_mode
))
2547 return &__d_cgrp(dentry
)->xattrs
;
2549 return &__d_cfe(dentry
)->xattrs
;
2552 static inline int xattr_enabled(struct dentry
*dentry
)
2554 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
2555 return root
->flags
& CGRP_ROOT_XATTR
;
2558 static bool is_valid_xattr(const char *name
)
2560 if (!strncmp(name
, XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
) ||
2561 !strncmp(name
, XATTR_SECURITY_PREFIX
, XATTR_SECURITY_PREFIX_LEN
))
2566 static int cgroup_setxattr(struct dentry
*dentry
, const char *name
,
2567 const void *val
, size_t size
, int flags
)
2569 if (!xattr_enabled(dentry
))
2571 if (!is_valid_xattr(name
))
2573 return simple_xattr_set(__d_xattrs(dentry
), name
, val
, size
, flags
);
2576 static int cgroup_removexattr(struct dentry
*dentry
, const char *name
)
2578 if (!xattr_enabled(dentry
))
2580 if (!is_valid_xattr(name
))
2582 return simple_xattr_remove(__d_xattrs(dentry
), name
);
2585 static ssize_t
cgroup_getxattr(struct dentry
*dentry
, const char *name
,
2586 void *buf
, size_t size
)
2588 if (!xattr_enabled(dentry
))
2590 if (!is_valid_xattr(name
))
2592 return simple_xattr_get(__d_xattrs(dentry
), name
, buf
, size
);
2595 static ssize_t
cgroup_listxattr(struct dentry
*dentry
, char *buf
, size_t size
)
2597 if (!xattr_enabled(dentry
))
2599 return simple_xattr_list(__d_xattrs(dentry
), buf
, size
);
2602 static const struct file_operations cgroup_file_operations
= {
2603 .read
= cgroup_file_read
,
2604 .write
= cgroup_file_write
,
2605 .llseek
= generic_file_llseek
,
2606 .open
= cgroup_file_open
,
2607 .release
= cgroup_file_release
,
2610 static const struct inode_operations cgroup_file_inode_operations
= {
2611 .setxattr
= cgroup_setxattr
,
2612 .getxattr
= cgroup_getxattr
,
2613 .listxattr
= cgroup_listxattr
,
2614 .removexattr
= cgroup_removexattr
,
2617 static const struct inode_operations cgroup_dir_inode_operations
= {
2618 .lookup
= simple_lookup
,
2619 .mkdir
= cgroup_mkdir
,
2620 .rmdir
= cgroup_rmdir
,
2621 .rename
= cgroup_rename
,
2622 .setxattr
= cgroup_setxattr
,
2623 .getxattr
= cgroup_getxattr
,
2624 .listxattr
= cgroup_listxattr
,
2625 .removexattr
= cgroup_removexattr
,
2629 * Check if a file is a control file
2631 struct cftype
*__file_cft(struct file
*file
)
2633 if (file_inode(file
)->i_fop
!= &cgroup_file_operations
)
2634 return ERR_PTR(-EINVAL
);
2635 return __d_cft(file
->f_dentry
);
2638 static int cgroup_create_file(struct dentry
*dentry
, umode_t mode
,
2639 struct super_block
*sb
)
2641 struct inode
*inode
;
2645 if (dentry
->d_inode
)
2648 inode
= cgroup_new_inode(mode
, sb
);
2652 if (S_ISDIR(mode
)) {
2653 inode
->i_op
= &cgroup_dir_inode_operations
;
2654 inode
->i_fop
= &simple_dir_operations
;
2656 /* start off with i_nlink == 2 (for "." entry) */
2658 inc_nlink(dentry
->d_parent
->d_inode
);
2661 * Control reaches here with cgroup_mutex held.
2662 * @inode->i_mutex should nest outside cgroup_mutex but we
2663 * want to populate it immediately without releasing
2664 * cgroup_mutex. As @inode isn't visible to anyone else
2665 * yet, trylock will always succeed without affecting
2668 WARN_ON_ONCE(!mutex_trylock(&inode
->i_mutex
));
2669 } else if (S_ISREG(mode
)) {
2671 inode
->i_fop
= &cgroup_file_operations
;
2672 inode
->i_op
= &cgroup_file_inode_operations
;
2674 d_instantiate(dentry
, inode
);
2675 dget(dentry
); /* Extra count - pin the dentry in core */
2680 * cgroup_file_mode - deduce file mode of a control file
2681 * @cft: the control file in question
2683 * returns cft->mode if ->mode is not 0
2684 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2685 * returns S_IRUGO if it has only a read handler
2686 * returns S_IWUSR if it has only a write hander
2688 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
2695 if (cft
->read
|| cft
->read_u64
|| cft
->read_s64
||
2696 cft
->read_map
|| cft
->read_seq_string
)
2699 if (cft
->write
|| cft
->write_u64
|| cft
->write_s64
||
2700 cft
->write_string
|| cft
->trigger
)
2706 static int cgroup_add_file(struct cgroup
*cgrp
, struct cftype
*cft
)
2708 struct dentry
*dir
= cgrp
->dentry
;
2709 struct cgroup
*parent
= __d_cgrp(dir
);
2710 struct dentry
*dentry
;
2714 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
2716 if (cft
->ss
&& !(cft
->flags
& CFTYPE_NO_PREFIX
) &&
2717 !(cgrp
->root
->flags
& CGRP_ROOT_NOPREFIX
)) {
2718 strcpy(name
, cft
->ss
->name
);
2721 strcat(name
, cft
->name
);
2723 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
2725 cfe
= kzalloc(sizeof(*cfe
), GFP_KERNEL
);
2729 dentry
= lookup_one_len(name
, dir
, strlen(name
));
2730 if (IS_ERR(dentry
)) {
2731 error
= PTR_ERR(dentry
);
2735 cfe
->type
= (void *)cft
;
2736 cfe
->dentry
= dentry
;
2737 dentry
->d_fsdata
= cfe
;
2738 simple_xattrs_init(&cfe
->xattrs
);
2740 mode
= cgroup_file_mode(cft
);
2741 error
= cgroup_create_file(dentry
, mode
| S_IFREG
, cgrp
->root
->sb
);
2743 list_add_tail(&cfe
->node
, &parent
->files
);
2753 * cgroup_addrm_files - add or remove files to a cgroup directory
2754 * @cgrp: the target cgroup
2755 * @cfts: array of cftypes to be added
2756 * @is_add: whether to add or remove
2758 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
2759 * For removals, this function never fails. If addition fails, this
2760 * function doesn't remove files already added. The caller is responsible
2763 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cftype cfts
[],
2769 lockdep_assert_held(&cgrp
->dentry
->d_inode
->i_mutex
);
2770 lockdep_assert_held(&cgroup_mutex
);
2772 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
2773 /* does cft->flags tell us to skip this file on @cgrp? */
2774 if ((cft
->flags
& CFTYPE_INSANE
) && cgroup_sane_behavior(cgrp
))
2776 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgrp
->parent
)
2778 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgrp
->parent
)
2782 ret
= cgroup_add_file(cgrp
, cft
);
2784 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2789 cgroup_rm_file(cgrp
, cft
);
2795 static void cgroup_cfts_prepare(void)
2796 __acquires(&cgroup_mutex
)
2799 * Thanks to the entanglement with vfs inode locking, we can't walk
2800 * the existing cgroups under cgroup_mutex and create files.
2801 * Instead, we use css_for_each_descendant_pre() and drop RCU read
2802 * lock before calling cgroup_addrm_files().
2804 mutex_lock(&cgroup_mutex
);
2807 static int cgroup_cfts_commit(struct cftype
*cfts
, bool is_add
)
2808 __releases(&cgroup_mutex
)
2811 struct cgroup_subsys
*ss
= cfts
[0].ss
;
2812 struct cgroup
*root
= &ss
->root
->top_cgroup
;
2813 struct super_block
*sb
= ss
->root
->sb
;
2814 struct dentry
*prev
= NULL
;
2815 struct inode
*inode
;
2816 struct cgroup_subsys_state
*css
;
2820 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2821 if (!cfts
|| ss
->root
== &cgroup_dummy_root
||
2822 !atomic_inc_not_zero(&sb
->s_active
)) {
2823 mutex_unlock(&cgroup_mutex
);
2828 * All cgroups which are created after we drop cgroup_mutex will
2829 * have the updated set of files, so we only need to update the
2830 * cgroups created before the current @cgroup_serial_nr_next.
2832 update_before
= cgroup_serial_nr_next
;
2834 mutex_unlock(&cgroup_mutex
);
2836 /* add/rm files for all cgroups created before */
2838 css_for_each_descendant_pre(css
, cgroup_css(root
, ss
)) {
2839 struct cgroup
*cgrp
= css
->cgroup
;
2841 if (cgroup_is_dead(cgrp
))
2844 inode
= cgrp
->dentry
->d_inode
;
2849 prev
= cgrp
->dentry
;
2851 mutex_lock(&inode
->i_mutex
);
2852 mutex_lock(&cgroup_mutex
);
2853 if (cgrp
->serial_nr
< update_before
&& !cgroup_is_dead(cgrp
))
2854 ret
= cgroup_addrm_files(cgrp
, cfts
, is_add
);
2855 mutex_unlock(&cgroup_mutex
);
2856 mutex_unlock(&inode
->i_mutex
);
2864 deactivate_super(sb
);
2869 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2870 * @ss: target cgroup subsystem
2871 * @cfts: zero-length name terminated array of cftypes
2873 * Register @cfts to @ss. Files described by @cfts are created for all
2874 * existing cgroups to which @ss is attached and all future cgroups will
2875 * have them too. This function can be called anytime whether @ss is
2878 * Returns 0 on successful registration, -errno on failure. Note that this
2879 * function currently returns 0 as long as @cfts registration is successful
2880 * even if some file creation attempts on existing cgroups fail.
2882 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
2884 struct cftype_set
*set
;
2888 set
= kzalloc(sizeof(*set
), GFP_KERNEL
);
2892 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++)
2895 cgroup_cfts_prepare();
2897 list_add_tail(&set
->node
, &ss
->cftsets
);
2898 ret
= cgroup_cfts_commit(cfts
, true);
2900 cgroup_rm_cftypes(cfts
);
2903 EXPORT_SYMBOL_GPL(cgroup_add_cftypes
);
2906 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2907 * @cfts: zero-length name terminated array of cftypes
2909 * Unregister @cfts. Files described by @cfts are removed from all
2910 * existing cgroups and all future cgroups won't have them either. This
2911 * function can be called anytime whether @cfts' subsys is attached or not.
2913 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2916 int cgroup_rm_cftypes(struct cftype
*cfts
)
2918 struct cftype_set
*set
;
2920 if (!cfts
|| !cfts
[0].ss
)
2923 cgroup_cfts_prepare();
2925 list_for_each_entry(set
, &cfts
[0].ss
->cftsets
, node
) {
2926 if (set
->cfts
== cfts
) {
2927 list_del(&set
->node
);
2929 cgroup_cfts_commit(cfts
, false);
2934 cgroup_cfts_commit(NULL
, false);
2939 * cgroup_task_count - count the number of tasks in a cgroup.
2940 * @cgrp: the cgroup in question
2942 * Return the number of tasks in the cgroup.
2944 int cgroup_task_count(const struct cgroup
*cgrp
)
2947 struct cgrp_cset_link
*link
;
2949 read_lock(&css_set_lock
);
2950 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
2951 count
+= atomic_read(&link
->cset
->refcount
);
2952 read_unlock(&css_set_lock
);
2957 * To reduce the fork() overhead for systems that are not actually using
2958 * their cgroups capability, we don't maintain the lists running through
2959 * each css_set to its tasks until we see the list actually used - in other
2960 * words after the first call to css_task_iter_start().
2962 static void cgroup_enable_task_cg_lists(void)
2964 struct task_struct
*p
, *g
;
2965 write_lock(&css_set_lock
);
2966 use_task_css_set_links
= 1;
2968 * We need tasklist_lock because RCU is not safe against
2969 * while_each_thread(). Besides, a forking task that has passed
2970 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2971 * is not guaranteed to have its child immediately visible in the
2972 * tasklist if we walk through it with RCU.
2974 read_lock(&tasklist_lock
);
2975 do_each_thread(g
, p
) {
2978 * We should check if the process is exiting, otherwise
2979 * it will race with cgroup_exit() in that the list
2980 * entry won't be deleted though the process has exited.
2982 if (!(p
->flags
& PF_EXITING
) && list_empty(&p
->cg_list
))
2983 list_add(&p
->cg_list
, &task_css_set(p
)->tasks
);
2985 } while_each_thread(g
, p
);
2986 read_unlock(&tasklist_lock
);
2987 write_unlock(&css_set_lock
);
2991 * css_next_child - find the next child of a given css
2992 * @pos_css: the current position (%NULL to initiate traversal)
2993 * @parent_css: css whose children to walk
2995 * This function returns the next child of @parent_css and should be called
2996 * under RCU read lock. The only requirement is that @parent_css and
2997 * @pos_css are accessible. The next sibling is guaranteed to be returned
2998 * regardless of their states.
3000 struct cgroup_subsys_state
*
3001 css_next_child(struct cgroup_subsys_state
*pos_css
,
3002 struct cgroup_subsys_state
*parent_css
)
3004 struct cgroup
*pos
= pos_css
? pos_css
->cgroup
: NULL
;
3005 struct cgroup
*cgrp
= parent_css
->cgroup
;
3006 struct cgroup
*next
;
3008 WARN_ON_ONCE(!rcu_read_lock_held());
3011 * @pos could already have been removed. Once a cgroup is removed,
3012 * its ->sibling.next is no longer updated when its next sibling
3013 * changes. As CGRP_DEAD assertion is serialized and happens
3014 * before the cgroup is taken off the ->sibling list, if we see it
3015 * unasserted, it's guaranteed that the next sibling hasn't
3016 * finished its grace period even if it's already removed, and thus
3017 * safe to dereference from this RCU critical section. If
3018 * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed
3019 * to be visible as %true here.
3021 * If @pos is dead, its next pointer can't be dereferenced;
3022 * however, as each cgroup is given a monotonically increasing
3023 * unique serial number and always appended to the sibling list,
3024 * the next one can be found by walking the parent's children until
3025 * we see a cgroup with higher serial number than @pos's. While
3026 * this path can be slower, it's taken only when either the current
3027 * cgroup is removed or iteration and removal race.
3030 next
= list_entry_rcu(cgrp
->children
.next
, struct cgroup
, sibling
);
3031 } else if (likely(!cgroup_is_dead(pos
))) {
3032 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup
, sibling
);
3034 list_for_each_entry_rcu(next
, &cgrp
->children
, sibling
)
3035 if (next
->serial_nr
> pos
->serial_nr
)
3039 if (&next
->sibling
== &cgrp
->children
)
3042 return cgroup_css(next
, parent_css
->ss
);
3044 EXPORT_SYMBOL_GPL(css_next_child
);
3047 * css_next_descendant_pre - find the next descendant for pre-order walk
3048 * @pos: the current position (%NULL to initiate traversal)
3049 * @root: css whose descendants to walk
3051 * To be used by css_for_each_descendant_pre(). Find the next descendant
3052 * to visit for pre-order traversal of @root's descendants. @root is
3053 * included in the iteration and the first node to be visited.
3055 * While this function requires RCU read locking, it doesn't require the
3056 * whole traversal to be contained in a single RCU critical section. This
3057 * function will return the correct next descendant as long as both @pos
3058 * and @root are accessible and @pos is a descendant of @root.
3060 struct cgroup_subsys_state
*
3061 css_next_descendant_pre(struct cgroup_subsys_state
*pos
,
3062 struct cgroup_subsys_state
*root
)
3064 struct cgroup_subsys_state
*next
;
3066 WARN_ON_ONCE(!rcu_read_lock_held());
3068 /* if first iteration, visit @root */
3072 /* visit the first child if exists */
3073 next
= css_next_child(NULL
, pos
);
3077 /* no child, visit my or the closest ancestor's next sibling */
3078 while (pos
!= root
) {
3079 next
= css_next_child(pos
, css_parent(pos
));
3082 pos
= css_parent(pos
);
3087 EXPORT_SYMBOL_GPL(css_next_descendant_pre
);
3090 * css_rightmost_descendant - return the rightmost descendant of a css
3091 * @pos: css of interest
3093 * Return the rightmost descendant of @pos. If there's no descendant, @pos
3094 * is returned. This can be used during pre-order traversal to skip
3097 * While this function requires RCU read locking, it doesn't require the
3098 * whole traversal to be contained in a single RCU critical section. This
3099 * function will return the correct rightmost descendant as long as @pos is
3102 struct cgroup_subsys_state
*
3103 css_rightmost_descendant(struct cgroup_subsys_state
*pos
)
3105 struct cgroup_subsys_state
*last
, *tmp
;
3107 WARN_ON_ONCE(!rcu_read_lock_held());
3111 /* ->prev isn't RCU safe, walk ->next till the end */
3113 css_for_each_child(tmp
, last
)
3119 EXPORT_SYMBOL_GPL(css_rightmost_descendant
);
3121 static struct cgroup_subsys_state
*
3122 css_leftmost_descendant(struct cgroup_subsys_state
*pos
)
3124 struct cgroup_subsys_state
*last
;
3128 pos
= css_next_child(NULL
, pos
);
3135 * css_next_descendant_post - find the next descendant for post-order walk
3136 * @pos: the current position (%NULL to initiate traversal)
3137 * @root: css whose descendants to walk
3139 * To be used by css_for_each_descendant_post(). Find the next descendant
3140 * to visit for post-order traversal of @root's descendants. @root is
3141 * included in the iteration and the last node to be visited.
3143 * While this function requires RCU read locking, it doesn't require the
3144 * whole traversal to be contained in a single RCU critical section. This
3145 * function will return the correct next descendant as long as both @pos
3146 * and @cgroup are accessible and @pos is a descendant of @cgroup.
3148 struct cgroup_subsys_state
*
3149 css_next_descendant_post(struct cgroup_subsys_state
*pos
,
3150 struct cgroup_subsys_state
*root
)
3152 struct cgroup_subsys_state
*next
;
3154 WARN_ON_ONCE(!rcu_read_lock_held());
3156 /* if first iteration, visit leftmost descendant which may be @root */
3158 return css_leftmost_descendant(root
);
3160 /* if we visited @root, we're done */
3164 /* if there's an unvisited sibling, visit its leftmost descendant */
3165 next
= css_next_child(pos
, css_parent(pos
));
3167 return css_leftmost_descendant(next
);
3169 /* no sibling left, visit parent */
3170 return css_parent(pos
);
3172 EXPORT_SYMBOL_GPL(css_next_descendant_post
);
3175 * css_advance_task_iter - advance a task itererator to the next css_set
3176 * @it: the iterator to advance
3178 * Advance @it to the next css_set to walk.
3180 static void css_advance_task_iter(struct css_task_iter
*it
)
3182 struct list_head
*l
= it
->cset_link
;
3183 struct cgrp_cset_link
*link
;
3184 struct css_set
*cset
;
3186 /* Advance to the next non-empty css_set */
3189 if (l
== &it
->origin_css
->cgroup
->cset_links
) {
3190 it
->cset_link
= NULL
;
3193 link
= list_entry(l
, struct cgrp_cset_link
, cset_link
);
3195 } while (list_empty(&cset
->tasks
));
3197 it
->task
= cset
->tasks
.next
;
3201 * css_task_iter_start - initiate task iteration
3202 * @css: the css to walk tasks of
3203 * @it: the task iterator to use
3205 * Initiate iteration through the tasks of @css. The caller can call
3206 * css_task_iter_next() to walk through the tasks until the function
3207 * returns NULL. On completion of iteration, css_task_iter_end() must be
3210 * Note that this function acquires a lock which is released when the
3211 * iteration finishes. The caller can't sleep while iteration is in
3214 void css_task_iter_start(struct cgroup_subsys_state
*css
,
3215 struct css_task_iter
*it
)
3216 __acquires(css_set_lock
)
3219 * The first time anyone tries to iterate across a css, we need to
3220 * enable the list linking each css_set to its tasks, and fix up
3221 * all existing tasks.
3223 if (!use_task_css_set_links
)
3224 cgroup_enable_task_cg_lists();
3226 read_lock(&css_set_lock
);
3228 it
->origin_css
= css
;
3229 it
->cset_link
= &css
->cgroup
->cset_links
;
3231 css_advance_task_iter(it
);
3235 * css_task_iter_next - return the next task for the iterator
3236 * @it: the task iterator being iterated
3238 * The "next" function for task iteration. @it should have been
3239 * initialized via css_task_iter_start(). Returns NULL when the iteration
3242 struct task_struct
*css_task_iter_next(struct css_task_iter
*it
)
3244 struct task_struct
*res
;
3245 struct list_head
*l
= it
->task
;
3246 struct cgrp_cset_link
*link
;
3248 /* If the iterator cg is NULL, we have no tasks */
3251 res
= list_entry(l
, struct task_struct
, cg_list
);
3252 /* Advance iterator to find next entry */
3254 link
= list_entry(it
->cset_link
, struct cgrp_cset_link
, cset_link
);
3255 if (l
== &link
->cset
->tasks
) {
3257 * We reached the end of this task list - move on to the
3258 * next cgrp_cset_link.
3260 css_advance_task_iter(it
);
3268 * css_task_iter_end - finish task iteration
3269 * @it: the task iterator to finish
3271 * Finish task iteration started by css_task_iter_start().
3273 void css_task_iter_end(struct css_task_iter
*it
)
3274 __releases(css_set_lock
)
3276 read_unlock(&css_set_lock
);
3279 static inline int started_after_time(struct task_struct
*t1
,
3280 struct timespec
*time
,
3281 struct task_struct
*t2
)
3283 int start_diff
= timespec_compare(&t1
->start_time
, time
);
3284 if (start_diff
> 0) {
3286 } else if (start_diff
< 0) {
3290 * Arbitrarily, if two processes started at the same
3291 * time, we'll say that the lower pointer value
3292 * started first. Note that t2 may have exited by now
3293 * so this may not be a valid pointer any longer, but
3294 * that's fine - it still serves to distinguish
3295 * between two tasks started (effectively) simultaneously.
3302 * This function is a callback from heap_insert() and is used to order
3304 * In this case we order the heap in descending task start time.
3306 static inline int started_after(void *p1
, void *p2
)
3308 struct task_struct
*t1
= p1
;
3309 struct task_struct
*t2
= p2
;
3310 return started_after_time(t1
, &t2
->start_time
, t2
);
3314 * css_scan_tasks - iterate though all the tasks in a css
3315 * @css: the css to iterate tasks of
3316 * @test: optional test callback
3317 * @process: process callback
3318 * @data: data passed to @test and @process
3319 * @heap: optional pre-allocated heap used for task iteration
3321 * Iterate through all the tasks in @css, calling @test for each, and if it
3322 * returns %true, call @process for it also.
3324 * @test may be NULL, meaning always true (select all tasks), which
3325 * effectively duplicates css_task_iter_{start,next,end}() but does not
3326 * lock css_set_lock for the call to @process.
3328 * It is guaranteed that @process will act on every task that is a member
3329 * of @css for the duration of this call. This function may or may not
3330 * call @process for tasks that exit or move to a different css during the
3331 * call, or are forked or move into the css during the call.
3333 * Note that @test may be called with locks held, and may in some
3334 * situations be called multiple times for the same task, so it should be
3337 * If @heap is non-NULL, a heap has been pre-allocated and will be used for
3338 * heap operations (and its "gt" member will be overwritten), else a
3339 * temporary heap will be used (allocation of which may cause this function
3342 int css_scan_tasks(struct cgroup_subsys_state
*css
,
3343 bool (*test
)(struct task_struct
*, void *),
3344 void (*process
)(struct task_struct
*, void *),
3345 void *data
, struct ptr_heap
*heap
)
3348 struct css_task_iter it
;
3349 struct task_struct
*p
, *dropped
;
3350 /* Never dereference latest_task, since it's not refcounted */
3351 struct task_struct
*latest_task
= NULL
;
3352 struct ptr_heap tmp_heap
;
3353 struct timespec latest_time
= { 0, 0 };
3356 /* The caller supplied our heap and pre-allocated its memory */
3357 heap
->gt
= &started_after
;
3359 /* We need to allocate our own heap memory */
3361 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
3363 /* cannot allocate the heap */
3369 * Scan tasks in the css, using the @test callback to determine
3370 * which are of interest, and invoking @process callback on the
3371 * ones which need an update. Since we don't want to hold any
3372 * locks during the task updates, gather tasks to be processed in a
3373 * heap structure. The heap is sorted by descending task start
3374 * time. If the statically-sized heap fills up, we overflow tasks
3375 * that started later, and in future iterations only consider tasks
3376 * that started after the latest task in the previous pass. This
3377 * guarantees forward progress and that we don't miss any tasks.
3380 css_task_iter_start(css
, &it
);
3381 while ((p
= css_task_iter_next(&it
))) {
3383 * Only affect tasks that qualify per the caller's callback,
3384 * if he provided one
3386 if (test
&& !test(p
, data
))
3389 * Only process tasks that started after the last task
3392 if (!started_after_time(p
, &latest_time
, latest_task
))
3394 dropped
= heap_insert(heap
, p
);
3395 if (dropped
== NULL
) {
3397 * The new task was inserted; the heap wasn't
3401 } else if (dropped
!= p
) {
3403 * The new task was inserted, and pushed out a
3407 put_task_struct(dropped
);
3410 * Else the new task was newer than anything already in
3411 * the heap and wasn't inserted
3414 css_task_iter_end(&it
);
3417 for (i
= 0; i
< heap
->size
; i
++) {
3418 struct task_struct
*q
= heap
->ptrs
[i
];
3420 latest_time
= q
->start_time
;
3423 /* Process the task per the caller's callback */
3428 * If we had to process any tasks at all, scan again
3429 * in case some of them were in the middle of forking
3430 * children that didn't get processed.
3431 * Not the most efficient way to do it, but it avoids
3432 * having to take callback_mutex in the fork path
3436 if (heap
== &tmp_heap
)
3437 heap_free(&tmp_heap
);
3441 static void cgroup_transfer_one_task(struct task_struct
*task
, void *data
)
3443 struct cgroup
*new_cgroup
= data
;
3445 mutex_lock(&cgroup_mutex
);
3446 cgroup_attach_task(new_cgroup
, task
, false);
3447 mutex_unlock(&cgroup_mutex
);
3451 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3452 * @to: cgroup to which the tasks will be moved
3453 * @from: cgroup in which the tasks currently reside
3455 int cgroup_transfer_tasks(struct cgroup
*to
, struct cgroup
*from
)
3457 return css_scan_tasks(&from
->dummy_css
, NULL
, cgroup_transfer_one_task
,
3462 * Stuff for reading the 'tasks'/'procs' files.
3464 * Reading this file can return large amounts of data if a cgroup has
3465 * *lots* of attached tasks. So it may need several calls to read(),
3466 * but we cannot guarantee that the information we produce is correct
3467 * unless we produce it entirely atomically.
3471 /* which pidlist file are we talking about? */
3472 enum cgroup_filetype
{
3478 * A pidlist is a list of pids that virtually represents the contents of one
3479 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3480 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3483 struct cgroup_pidlist
{
3485 * used to find which pidlist is wanted. doesn't change as long as
3486 * this particular list stays in the list.
3488 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
3491 /* how many elements the above list has */
3493 /* how many files are using the current array */
3495 /* each of these stored in a list by its cgroup */
3496 struct list_head links
;
3497 /* pointer to the cgroup we belong to, for list removal purposes */
3498 struct cgroup
*owner
;
3499 /* protects the other fields */
3500 struct rw_semaphore rwsem
;
3504 * The following two functions "fix" the issue where there are more pids
3505 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3506 * TODO: replace with a kernel-wide solution to this problem
3508 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3509 static void *pidlist_allocate(int count
)
3511 if (PIDLIST_TOO_LARGE(count
))
3512 return vmalloc(count
* sizeof(pid_t
));
3514 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
3516 static void pidlist_free(void *p
)
3518 if (is_vmalloc_addr(p
))
3525 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3526 * Returns the number of unique elements.
3528 static int pidlist_uniq(pid_t
*list
, int length
)
3533 * we presume the 0th element is unique, so i starts at 1. trivial
3534 * edge cases first; no work needs to be done for either
3536 if (length
== 0 || length
== 1)
3538 /* src and dest walk down the list; dest counts unique elements */
3539 for (src
= 1; src
< length
; src
++) {
3540 /* find next unique element */
3541 while (list
[src
] == list
[src
-1]) {
3546 /* dest always points to where the next unique element goes */
3547 list
[dest
] = list
[src
];
3554 static int cmppid(const void *a
, const void *b
)
3556 return *(pid_t
*)a
- *(pid_t
*)b
;
3560 * find the appropriate pidlist for our purpose (given procs vs tasks)
3561 * returns with the lock on that pidlist already held, and takes care
3562 * of the use count, or returns NULL with no locks held if we're out of
3565 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
3566 enum cgroup_filetype type
)
3568 struct cgroup_pidlist
*l
;
3569 /* don't need task_nsproxy() if we're looking at ourself */
3570 struct pid_namespace
*ns
= task_active_pid_ns(current
);
3573 * We can't drop the pidlist_mutex before taking the l->rwsem in case
3574 * the last ref-holder is trying to remove l from the list at the same
3575 * time. Holding the pidlist_mutex precludes somebody taking whichever
3576 * list we find out from under us - compare release_pid_array().
3578 mutex_lock(&cgrp
->pidlist_mutex
);
3579 list_for_each_entry(l
, &cgrp
->pidlists
, links
) {
3580 if (l
->key
.type
== type
&& l
->key
.ns
== ns
) {
3581 /* make sure l doesn't vanish out from under us */
3582 down_write(&l
->rwsem
);
3583 mutex_unlock(&cgrp
->pidlist_mutex
);
3587 /* entry not found; create a new one */
3588 l
= kzalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
3590 mutex_unlock(&cgrp
->pidlist_mutex
);
3593 init_rwsem(&l
->rwsem
);
3594 down_write(&l
->rwsem
);
3596 l
->key
.ns
= get_pid_ns(ns
);
3598 list_add(&l
->links
, &cgrp
->pidlists
);
3599 mutex_unlock(&cgrp
->pidlist_mutex
);
3604 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3606 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
3607 struct cgroup_pidlist
**lp
)
3611 int pid
, n
= 0; /* used for populating the array */
3612 struct css_task_iter it
;
3613 struct task_struct
*tsk
;
3614 struct cgroup_pidlist
*l
;
3617 * If cgroup gets more users after we read count, we won't have
3618 * enough space - tough. This race is indistinguishable to the
3619 * caller from the case that the additional cgroup users didn't
3620 * show up until sometime later on.
3622 length
= cgroup_task_count(cgrp
);
3623 array
= pidlist_allocate(length
);
3626 /* now, populate the array */
3627 css_task_iter_start(&cgrp
->dummy_css
, &it
);
3628 while ((tsk
= css_task_iter_next(&it
))) {
3629 if (unlikely(n
== length
))
3631 /* get tgid or pid for procs or tasks file respectively */
3632 if (type
== CGROUP_FILE_PROCS
)
3633 pid
= task_tgid_vnr(tsk
);
3635 pid
= task_pid_vnr(tsk
);
3636 if (pid
> 0) /* make sure to only use valid results */
3639 css_task_iter_end(&it
);
3641 /* now sort & (if procs) strip out duplicates */
3642 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
3643 if (type
== CGROUP_FILE_PROCS
)
3644 length
= pidlist_uniq(array
, length
);
3645 l
= cgroup_pidlist_find(cgrp
, type
);
3647 pidlist_free(array
);
3650 /* store array, freeing old if necessary - lock already held */
3651 pidlist_free(l
->list
);
3655 up_write(&l
->rwsem
);
3661 * cgroupstats_build - build and fill cgroupstats
3662 * @stats: cgroupstats to fill information into
3663 * @dentry: A dentry entry belonging to the cgroup for which stats have
3666 * Build and fill cgroupstats so that taskstats can export it to user
3669 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
3672 struct cgroup
*cgrp
;
3673 struct css_task_iter it
;
3674 struct task_struct
*tsk
;
3677 * Validate dentry by checking the superblock operations,
3678 * and make sure it's a directory.
3680 if (dentry
->d_sb
->s_op
!= &cgroup_ops
||
3681 !S_ISDIR(dentry
->d_inode
->i_mode
))
3685 cgrp
= dentry
->d_fsdata
;
3687 css_task_iter_start(&cgrp
->dummy_css
, &it
);
3688 while ((tsk
= css_task_iter_next(&it
))) {
3689 switch (tsk
->state
) {
3691 stats
->nr_running
++;
3693 case TASK_INTERRUPTIBLE
:
3694 stats
->nr_sleeping
++;
3696 case TASK_UNINTERRUPTIBLE
:
3697 stats
->nr_uninterruptible
++;
3700 stats
->nr_stopped
++;
3703 if (delayacct_is_task_waiting_on_io(tsk
))
3704 stats
->nr_io_wait
++;
3708 css_task_iter_end(&it
);
3716 * seq_file methods for the tasks/procs files. The seq_file position is the
3717 * next pid to display; the seq_file iterator is a pointer to the pid
3718 * in the cgroup->l->list array.
3721 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
3724 * Initially we receive a position value that corresponds to
3725 * one more than the last pid shown (or 0 on the first call or
3726 * after a seek to the start). Use a binary-search to find the
3727 * next pid to display, if any
3729 struct cgroup_pidlist
*l
= s
->private;
3730 int index
= 0, pid
= *pos
;
3733 down_read(&l
->rwsem
);
3735 int end
= l
->length
;
3737 while (index
< end
) {
3738 int mid
= (index
+ end
) / 2;
3739 if (l
->list
[mid
] == pid
) {
3742 } else if (l
->list
[mid
] <= pid
)
3748 /* If we're off the end of the array, we're done */
3749 if (index
>= l
->length
)
3751 /* Update the abstract position to be the actual pid that we found */
3752 iter
= l
->list
+ index
;
3757 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
3759 struct cgroup_pidlist
*l
= s
->private;
3763 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
3765 struct cgroup_pidlist
*l
= s
->private;
3767 pid_t
*end
= l
->list
+ l
->length
;
3769 * Advance to the next pid in the array. If this goes off the
3781 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
3783 return seq_printf(s
, "%d\n", *(int *)v
);
3787 * seq_operations functions for iterating on pidlists through seq_file -
3788 * independent of whether it's tasks or procs
3790 static const struct seq_operations cgroup_pidlist_seq_operations
= {
3791 .start
= cgroup_pidlist_start
,
3792 .stop
= cgroup_pidlist_stop
,
3793 .next
= cgroup_pidlist_next
,
3794 .show
= cgroup_pidlist_show
,
3797 static void cgroup_release_pid_array(struct cgroup_pidlist
*l
)
3800 * the case where we're the last user of this particular pidlist will
3801 * have us remove it from the cgroup's list, which entails taking the
3802 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3803 * pidlist_mutex, we have to take pidlist_mutex first.
3805 mutex_lock(&l
->owner
->pidlist_mutex
);
3806 down_write(&l
->rwsem
);
3807 BUG_ON(!l
->use_count
);
3808 if (!--l
->use_count
) {
3809 /* we're the last user if refcount is 0; remove and free */
3810 list_del(&l
->links
);
3811 mutex_unlock(&l
->owner
->pidlist_mutex
);
3812 pidlist_free(l
->list
);
3813 put_pid_ns(l
->key
.ns
);
3814 up_write(&l
->rwsem
);
3818 mutex_unlock(&l
->owner
->pidlist_mutex
);
3819 up_write(&l
->rwsem
);
3822 static int cgroup_pidlist_release(struct inode
*inode
, struct file
*file
)
3824 struct cgroup_pidlist
*l
;
3825 if (!(file
->f_mode
& FMODE_READ
))
3828 * the seq_file will only be initialized if the file was opened for
3829 * reading; hence we check if it's not null only in that case.
3831 l
= ((struct seq_file
*)file
->private_data
)->private;
3832 cgroup_release_pid_array(l
);
3833 return seq_release(inode
, file
);
3836 static const struct file_operations cgroup_pidlist_operations
= {
3838 .llseek
= seq_lseek
,
3839 .write
= cgroup_file_write
,
3840 .release
= cgroup_pidlist_release
,
3844 * The following functions handle opens on a file that displays a pidlist
3845 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3848 /* helper function for the two below it */
3849 static int cgroup_pidlist_open(struct file
*file
, enum cgroup_filetype type
)
3851 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
3852 struct cgroup_pidlist
*l
;
3855 /* Nothing to do for write-only files */
3856 if (!(file
->f_mode
& FMODE_READ
))
3859 /* have the array populated */
3860 retval
= pidlist_array_load(cgrp
, type
, &l
);
3863 /* configure file information */
3864 file
->f_op
= &cgroup_pidlist_operations
;
3866 retval
= seq_open(file
, &cgroup_pidlist_seq_operations
);
3868 cgroup_release_pid_array(l
);
3871 ((struct seq_file
*)file
->private_data
)->private = l
;
3874 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
3876 return cgroup_pidlist_open(file
, CGROUP_FILE_TASKS
);
3878 static int cgroup_procs_open(struct inode
*unused
, struct file
*file
)
3880 return cgroup_pidlist_open(file
, CGROUP_FILE_PROCS
);
3883 static u64
cgroup_read_notify_on_release(struct cgroup_subsys_state
*css
,
3886 return notify_on_release(css
->cgroup
);
3889 static int cgroup_write_notify_on_release(struct cgroup_subsys_state
*css
,
3890 struct cftype
*cft
, u64 val
)
3892 clear_bit(CGRP_RELEASABLE
, &css
->cgroup
->flags
);
3894 set_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
3896 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
3901 * When dput() is called asynchronously, if umount has been done and
3902 * then deactivate_super() in cgroup_free_fn() kills the superblock,
3903 * there's a small window that vfs will see the root dentry with non-zero
3904 * refcnt and trigger BUG().
3906 * That's why we hold a reference before dput() and drop it right after.
3908 static void cgroup_dput(struct cgroup
*cgrp
)
3910 struct super_block
*sb
= cgrp
->root
->sb
;
3912 atomic_inc(&sb
->s_active
);
3914 deactivate_super(sb
);
3917 static u64
cgroup_clone_children_read(struct cgroup_subsys_state
*css
,
3920 return test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
3923 static int cgroup_clone_children_write(struct cgroup_subsys_state
*css
,
3924 struct cftype
*cft
, u64 val
)
3927 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
3929 clear_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
3933 static struct cftype cgroup_base_files
[] = {
3935 .name
= "cgroup.procs",
3936 .open
= cgroup_procs_open
,
3937 .write_u64
= cgroup_procs_write
,
3938 .release
= cgroup_pidlist_release
,
3939 .mode
= S_IRUGO
| S_IWUSR
,
3942 .name
= "cgroup.clone_children",
3943 .flags
= CFTYPE_INSANE
,
3944 .read_u64
= cgroup_clone_children_read
,
3945 .write_u64
= cgroup_clone_children_write
,
3948 .name
= "cgroup.sane_behavior",
3949 .flags
= CFTYPE_ONLY_ON_ROOT
,
3950 .read_seq_string
= cgroup_sane_behavior_show
,
3954 * Historical crazy stuff. These don't have "cgroup." prefix and
3955 * don't exist if sane_behavior. If you're depending on these, be
3956 * prepared to be burned.
3960 .flags
= CFTYPE_INSANE
, /* use "procs" instead */
3961 .open
= cgroup_tasks_open
,
3962 .write_u64
= cgroup_tasks_write
,
3963 .release
= cgroup_pidlist_release
,
3964 .mode
= S_IRUGO
| S_IWUSR
,
3967 .name
= "notify_on_release",
3968 .flags
= CFTYPE_INSANE
,
3969 .read_u64
= cgroup_read_notify_on_release
,
3970 .write_u64
= cgroup_write_notify_on_release
,
3973 .name
= "release_agent",
3974 .flags
= CFTYPE_INSANE
| CFTYPE_ONLY_ON_ROOT
,
3975 .read_seq_string
= cgroup_release_agent_show
,
3976 .write_string
= cgroup_release_agent_write
,
3977 .max_write_len
= PATH_MAX
,
3983 * cgroup_populate_dir - create subsys files in a cgroup directory
3984 * @cgrp: target cgroup
3985 * @subsys_mask: mask of the subsystem ids whose files should be added
3987 * On failure, no file is added.
3989 static int cgroup_populate_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
)
3991 struct cgroup_subsys
*ss
;
3994 /* process cftsets of each subsystem */
3995 for_each_subsys(ss
, i
) {
3996 struct cftype_set
*set
;
3998 if (!test_bit(i
, &subsys_mask
))
4001 list_for_each_entry(set
, &ss
->cftsets
, node
) {
4002 ret
= cgroup_addrm_files(cgrp
, set
->cfts
, true);
4008 /* This cgroup is ready now */
4009 for_each_root_subsys(cgrp
->root
, ss
) {
4010 struct cgroup_subsys_state
*css
= cgroup_css(cgrp
, ss
);
4011 struct css_id
*id
= rcu_dereference_protected(css
->id
, true);
4014 * Update id->css pointer and make this css visible from
4015 * CSS ID functions. This pointer will be dereferened
4016 * from RCU-read-side without locks.
4019 rcu_assign_pointer(id
->css
, css
);
4024 cgroup_clear_dir(cgrp
, subsys_mask
);
4029 * css destruction is four-stage process.
4031 * 1. Destruction starts. Killing of the percpu_ref is initiated.
4032 * Implemented in kill_css().
4034 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4035 * and thus css_tryget() is guaranteed to fail, the css can be offlined
4036 * by invoking offline_css(). After offlining, the base ref is put.
4037 * Implemented in css_killed_work_fn().
4039 * 3. When the percpu_ref reaches zero, the only possible remaining
4040 * accessors are inside RCU read sections. css_release() schedules the
4043 * 4. After the grace period, the css can be freed. Implemented in
4044 * css_free_work_fn().
4046 * It is actually hairier because both step 2 and 4 require process context
4047 * and thus involve punting to css->destroy_work adding two additional
4048 * steps to the already complex sequence.
4050 static void css_free_work_fn(struct work_struct
*work
)
4052 struct cgroup_subsys_state
*css
=
4053 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4054 struct cgroup
*cgrp
= css
->cgroup
;
4057 css_put(css
->parent
);
4059 css
->ss
->css_free(css
);
4063 static void css_free_rcu_fn(struct rcu_head
*rcu_head
)
4065 struct cgroup_subsys_state
*css
=
4066 container_of(rcu_head
, struct cgroup_subsys_state
, rcu_head
);
4069 * css holds an extra ref to @cgrp->dentry which is put on the last
4070 * css_put(). dput() requires process context which we don't have.
4072 INIT_WORK(&css
->destroy_work
, css_free_work_fn
);
4073 schedule_work(&css
->destroy_work
);
4076 static void css_release(struct percpu_ref
*ref
)
4078 struct cgroup_subsys_state
*css
=
4079 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4081 call_rcu(&css
->rcu_head
, css_free_rcu_fn
);
4084 static void init_css(struct cgroup_subsys_state
*css
, struct cgroup_subsys
*ss
,
4085 struct cgroup
*cgrp
)
4093 css
->parent
= cgroup_css(cgrp
->parent
, ss
);
4095 css
->flags
|= CSS_ROOT
;
4097 BUG_ON(cgroup_css(cgrp
, ss
));
4100 /* invoke ->css_online() on a new CSS and mark it online if successful */
4101 static int online_css(struct cgroup_subsys_state
*css
)
4103 struct cgroup_subsys
*ss
= css
->ss
;
4106 lockdep_assert_held(&cgroup_mutex
);
4109 ret
= ss
->css_online(css
);
4111 css
->flags
|= CSS_ONLINE
;
4112 css
->cgroup
->nr_css
++;
4113 rcu_assign_pointer(css
->cgroup
->subsys
[ss
->subsys_id
], css
);
4118 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
4119 static void offline_css(struct cgroup_subsys_state
*css
)
4121 struct cgroup_subsys
*ss
= css
->ss
;
4123 lockdep_assert_held(&cgroup_mutex
);
4125 if (!(css
->flags
& CSS_ONLINE
))
4128 if (ss
->css_offline
)
4129 ss
->css_offline(css
);
4131 css
->flags
&= ~CSS_ONLINE
;
4132 css
->cgroup
->nr_css
--;
4133 RCU_INIT_POINTER(css
->cgroup
->subsys
[ss
->subsys_id
], css
);
4137 * cgroup_create - create a cgroup
4138 * @parent: cgroup that will be parent of the new cgroup
4139 * @dentry: dentry of the new cgroup
4140 * @mode: mode to set on new inode
4142 * Must be called with the mutex on the parent inode held
4144 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
4147 struct cgroup_subsys_state
*css_ar
[CGROUP_SUBSYS_COUNT
] = { };
4148 struct cgroup
*cgrp
;
4149 struct cgroup_name
*name
;
4150 struct cgroupfs_root
*root
= parent
->root
;
4152 struct cgroup_subsys
*ss
;
4153 struct super_block
*sb
= root
->sb
;
4155 /* allocate the cgroup and its ID, 0 is reserved for the root */
4156 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
4160 name
= cgroup_alloc_name(dentry
);
4163 rcu_assign_pointer(cgrp
->name
, name
);
4166 * Temporarily set the pointer to NULL, so idr_find() won't return
4167 * a half-baked cgroup.
4169 cgrp
->id
= idr_alloc(&root
->cgroup_idr
, NULL
, 1, 0, GFP_KERNEL
);
4174 * Only live parents can have children. Note that the liveliness
4175 * check isn't strictly necessary because cgroup_mkdir() and
4176 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4177 * anyway so that locking is contained inside cgroup proper and we
4178 * don't get nasty surprises if we ever grow another caller.
4180 if (!cgroup_lock_live_group(parent
)) {
4185 /* Grab a reference on the superblock so the hierarchy doesn't
4186 * get deleted on unmount if there are child cgroups. This
4187 * can be done outside cgroup_mutex, since the sb can't
4188 * disappear while someone has an open control file on the
4190 atomic_inc(&sb
->s_active
);
4192 init_cgroup_housekeeping(cgrp
);
4194 dentry
->d_fsdata
= cgrp
;
4195 cgrp
->dentry
= dentry
;
4197 cgrp
->parent
= parent
;
4198 cgrp
->dummy_css
.parent
= &parent
->dummy_css
;
4199 cgrp
->root
= parent
->root
;
4201 if (notify_on_release(parent
))
4202 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
4204 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
4205 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4207 for_each_root_subsys(root
, ss
) {
4208 struct cgroup_subsys_state
*css
;
4210 css
= ss
->css_alloc(cgroup_css(parent
, ss
));
4215 css_ar
[ss
->subsys_id
] = css
;
4217 err
= percpu_ref_init(&css
->refcnt
, css_release
);
4221 init_css(css
, ss
, cgrp
);
4224 err
= alloc_css_id(css
);
4231 * Create directory. cgroup_create_file() returns with the new
4232 * directory locked on success so that it can be populated without
4233 * dropping cgroup_mutex.
4235 err
= cgroup_create_file(dentry
, S_IFDIR
| mode
, sb
);
4238 lockdep_assert_held(&dentry
->d_inode
->i_mutex
);
4240 cgrp
->serial_nr
= cgroup_serial_nr_next
++;
4242 /* allocation complete, commit to creation */
4243 list_add_tail_rcu(&cgrp
->sibling
, &cgrp
->parent
->children
);
4244 root
->number_of_cgroups
++;
4246 /* each css holds a ref to the cgroup's dentry and the parent css */
4247 for_each_root_subsys(root
, ss
) {
4248 struct cgroup_subsys_state
*css
= css_ar
[ss
->subsys_id
];
4251 css_get(css
->parent
);
4254 /* hold a ref to the parent's dentry */
4255 dget(parent
->dentry
);
4257 /* creation succeeded, notify subsystems */
4258 for_each_root_subsys(root
, ss
) {
4259 struct cgroup_subsys_state
*css
= css_ar
[ss
->subsys_id
];
4261 err
= online_css(css
);
4265 if (ss
->broken_hierarchy
&& !ss
->warned_broken_hierarchy
&&
4267 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",
4268 current
->comm
, current
->pid
, ss
->name
);
4269 if (!strcmp(ss
->name
, "memory"))
4270 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4271 ss
->warned_broken_hierarchy
= true;
4275 idr_replace(&root
->cgroup_idr
, cgrp
, cgrp
->id
);
4277 err
= cgroup_addrm_files(cgrp
, cgroup_base_files
, true);
4281 err
= cgroup_populate_dir(cgrp
, root
->subsys_mask
);
4285 mutex_unlock(&cgroup_mutex
);
4286 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
4291 for_each_root_subsys(root
, ss
) {
4292 struct cgroup_subsys_state
*css
= css_ar
[ss
->subsys_id
];
4295 percpu_ref_cancel_init(&css
->refcnt
);
4299 mutex_unlock(&cgroup_mutex
);
4300 /* Release the reference count that we took on the superblock */
4301 deactivate_super(sb
);
4303 idr_remove(&root
->cgroup_idr
, cgrp
->id
);
4305 kfree(rcu_dereference_raw(cgrp
->name
));
4311 cgroup_destroy_locked(cgrp
);
4312 mutex_unlock(&cgroup_mutex
);
4313 mutex_unlock(&dentry
->d_inode
->i_mutex
);
4317 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
4319 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
4321 /* the vfs holds inode->i_mutex already */
4322 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
4326 * This is called when the refcnt of a css is confirmed to be killed.
4327 * css_tryget() is now guaranteed to fail.
4329 static void css_killed_work_fn(struct work_struct
*work
)
4331 struct cgroup_subsys_state
*css
=
4332 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4333 struct cgroup
*cgrp
= css
->cgroup
;
4335 mutex_lock(&cgroup_mutex
);
4338 * css_tryget() is guaranteed to fail now. Tell subsystems to
4339 * initate destruction.
4344 * If @cgrp is marked dead, it's waiting for refs of all css's to
4345 * be disabled before proceeding to the second phase of cgroup
4346 * destruction. If we are the last one, kick it off.
4348 if (!cgrp
->nr_css
&& cgroup_is_dead(cgrp
))
4349 cgroup_destroy_css_killed(cgrp
);
4351 mutex_unlock(&cgroup_mutex
);
4354 * Put the css refs from kill_css(). Each css holds an extra
4355 * reference to the cgroup's dentry and cgroup removal proceeds
4356 * regardless of css refs. On the last put of each css, whenever
4357 * that may be, the extra dentry ref is put so that dentry
4358 * destruction happens only after all css's are released.
4363 /* css kill confirmation processing requires process context, bounce */
4364 static void css_killed_ref_fn(struct percpu_ref
*ref
)
4366 struct cgroup_subsys_state
*css
=
4367 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4369 INIT_WORK(&css
->destroy_work
, css_killed_work_fn
);
4370 schedule_work(&css
->destroy_work
);
4374 * kill_css - destroy a css
4375 * @css: css to destroy
4377 * This function initiates destruction of @css by removing cgroup interface
4378 * files and putting its base reference. ->css_offline() will be invoked
4379 * asynchronously once css_tryget() is guaranteed to fail and when the
4380 * reference count reaches zero, @css will be released.
4382 static void kill_css(struct cgroup_subsys_state
*css
)
4384 cgroup_clear_dir(css
->cgroup
, 1 << css
->ss
->subsys_id
);
4387 * Killing would put the base ref, but we need to keep it alive
4388 * until after ->css_offline().
4393 * cgroup core guarantees that, by the time ->css_offline() is
4394 * invoked, no new css reference will be given out via
4395 * css_tryget(). We can't simply call percpu_ref_kill() and
4396 * proceed to offlining css's because percpu_ref_kill() doesn't
4397 * guarantee that the ref is seen as killed on all CPUs on return.
4399 * Use percpu_ref_kill_and_confirm() to get notifications as each
4400 * css is confirmed to be seen as killed on all CPUs.
4402 percpu_ref_kill_and_confirm(&css
->refcnt
, css_killed_ref_fn
);
4406 * cgroup_destroy_locked - the first stage of cgroup destruction
4407 * @cgrp: cgroup to be destroyed
4409 * css's make use of percpu refcnts whose killing latency shouldn't be
4410 * exposed to userland and are RCU protected. Also, cgroup core needs to
4411 * guarantee that css_tryget() won't succeed by the time ->css_offline() is
4412 * invoked. To satisfy all the requirements, destruction is implemented in
4413 * the following two steps.
4415 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
4416 * userland visible parts and start killing the percpu refcnts of
4417 * css's. Set up so that the next stage will be kicked off once all
4418 * the percpu refcnts are confirmed to be killed.
4420 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
4421 * rest of destruction. Once all cgroup references are gone, the
4422 * cgroup is RCU-freed.
4424 * This function implements s1. After this step, @cgrp is gone as far as
4425 * the userland is concerned and a new cgroup with the same name may be
4426 * created. As cgroup doesn't care about the names internally, this
4427 * doesn't cause any problem.
4429 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
4430 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
4432 struct dentry
*d
= cgrp
->dentry
;
4433 struct cgroup_subsys
*ss
;
4434 struct cgroup
*child
;
4437 lockdep_assert_held(&d
->d_inode
->i_mutex
);
4438 lockdep_assert_held(&cgroup_mutex
);
4441 * css_set_lock synchronizes access to ->cset_links and prevents
4442 * @cgrp from being removed while __put_css_set() is in progress.
4444 read_lock(&css_set_lock
);
4445 empty
= list_empty(&cgrp
->cset_links
);
4446 read_unlock(&css_set_lock
);
4451 * Make sure there's no live children. We can't test ->children
4452 * emptiness as dead children linger on it while being destroyed;
4453 * otherwise, "rmdir parent/child parent" may fail with -EBUSY.
4457 list_for_each_entry_rcu(child
, &cgrp
->children
, sibling
) {
4458 empty
= cgroup_is_dead(child
);
4467 * Initiate massacre of all css's. cgroup_destroy_css_killed()
4468 * will be invoked to perform the rest of destruction once the
4469 * percpu refs of all css's are confirmed to be killed.
4471 for_each_root_subsys(cgrp
->root
, ss
)
4472 kill_css(cgroup_css(cgrp
, ss
));
4475 * Mark @cgrp dead. This prevents further task migration and child
4476 * creation by disabling cgroup_lock_live_group(). Note that
4477 * CGRP_DEAD assertion is depended upon by css_next_child() to
4478 * resume iteration after dropping RCU read lock. See
4479 * css_next_child() for details.
4481 set_bit(CGRP_DEAD
, &cgrp
->flags
);
4483 /* CGRP_DEAD is set, remove from ->release_list for the last time */
4484 raw_spin_lock(&release_list_lock
);
4485 if (!list_empty(&cgrp
->release_list
))
4486 list_del_init(&cgrp
->release_list
);
4487 raw_spin_unlock(&release_list_lock
);
4490 * If @cgrp has css's attached, the second stage of cgroup
4491 * destruction is kicked off from css_killed_work_fn() after the
4492 * refs of all attached css's are killed. If @cgrp doesn't have
4493 * any css, we kick it off here.
4496 cgroup_destroy_css_killed(cgrp
);
4499 * Clear the base files and remove @cgrp directory. The removal
4500 * puts the base ref but we aren't quite done with @cgrp yet, so
4503 cgroup_addrm_files(cgrp
, cgroup_base_files
, false);
4505 cgroup_d_remove_dir(d
);
4511 * cgroup_destroy_css_killed - the second step of cgroup destruction
4512 * @work: cgroup->destroy_free_work
4514 * This function is invoked from a work item for a cgroup which is being
4515 * destroyed after all css's are offlined and performs the rest of
4516 * destruction. This is the second step of destruction described in the
4517 * comment above cgroup_destroy_locked().
4519 static void cgroup_destroy_css_killed(struct cgroup
*cgrp
)
4521 struct cgroup
*parent
= cgrp
->parent
;
4522 struct dentry
*d
= cgrp
->dentry
;
4524 lockdep_assert_held(&cgroup_mutex
);
4526 /* delete this cgroup from parent->children */
4527 list_del_rcu(&cgrp
->sibling
);
4530 * We should remove the cgroup object from idr before its grace
4531 * period starts, so we won't be looking up a cgroup while the
4532 * cgroup is being freed.
4534 idr_remove(&cgrp
->root
->cgroup_idr
, cgrp
->id
);
4539 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
4540 check_for_release(parent
);
4543 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
4547 mutex_lock(&cgroup_mutex
);
4548 ret
= cgroup_destroy_locked(dentry
->d_fsdata
);
4549 mutex_unlock(&cgroup_mutex
);
4554 static void __init_or_module
cgroup_init_cftsets(struct cgroup_subsys
*ss
)
4556 INIT_LIST_HEAD(&ss
->cftsets
);
4559 * base_cftset is embedded in subsys itself, no need to worry about
4562 if (ss
->base_cftypes
) {
4565 for (cft
= ss
->base_cftypes
; cft
->name
[0] != '\0'; cft
++)
4568 ss
->base_cftset
.cfts
= ss
->base_cftypes
;
4569 list_add_tail(&ss
->base_cftset
.node
, &ss
->cftsets
);
4573 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
4575 struct cgroup_subsys_state
*css
;
4577 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4579 mutex_lock(&cgroup_mutex
);
4581 /* init base cftset */
4582 cgroup_init_cftsets(ss
);
4584 /* Create the top cgroup state for this subsystem */
4585 list_add(&ss
->sibling
, &cgroup_dummy_root
.subsys_list
);
4586 ss
->root
= &cgroup_dummy_root
;
4587 css
= ss
->css_alloc(cgroup_css(cgroup_dummy_top
, ss
));
4588 /* We don't handle early failures gracefully */
4589 BUG_ON(IS_ERR(css
));
4590 init_css(css
, ss
, cgroup_dummy_top
);
4592 /* Update the init_css_set to contain a subsys
4593 * pointer to this state - since the subsystem is
4594 * newly registered, all tasks and hence the
4595 * init_css_set is in the subsystem's top cgroup. */
4596 init_css_set
.subsys
[ss
->subsys_id
] = css
;
4598 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
4600 /* At system boot, before all subsystems have been
4601 * registered, no tasks have been forked, so we don't
4602 * need to invoke fork callbacks here. */
4603 BUG_ON(!list_empty(&init_task
.tasks
));
4605 BUG_ON(online_css(css
));
4607 mutex_unlock(&cgroup_mutex
);
4609 /* this function shouldn't be used with modular subsystems, since they
4610 * need to register a subsys_id, among other things */
4615 * cgroup_load_subsys: load and register a modular subsystem at runtime
4616 * @ss: the subsystem to load
4618 * This function should be called in a modular subsystem's initcall. If the
4619 * subsystem is built as a module, it will be assigned a new subsys_id and set
4620 * up for use. If the subsystem is built-in anyway, work is delegated to the
4621 * simpler cgroup_init_subsys.
4623 int __init_or_module
cgroup_load_subsys(struct cgroup_subsys
*ss
)
4625 struct cgroup_subsys_state
*css
;
4627 struct hlist_node
*tmp
;
4628 struct css_set
*cset
;
4631 /* check name and function validity */
4632 if (ss
->name
== NULL
|| strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
||
4633 ss
->css_alloc
== NULL
|| ss
->css_free
== NULL
)
4637 * we don't support callbacks in modular subsystems. this check is
4638 * before the ss->module check for consistency; a subsystem that could
4639 * be a module should still have no callbacks even if the user isn't
4640 * compiling it as one.
4642 if (ss
->fork
|| ss
->exit
)
4646 * an optionally modular subsystem is built-in: we want to do nothing,
4647 * since cgroup_init_subsys will have already taken care of it.
4649 if (ss
->module
== NULL
) {
4650 /* a sanity check */
4651 BUG_ON(cgroup_subsys
[ss
->subsys_id
] != ss
);
4655 /* init base cftset */
4656 cgroup_init_cftsets(ss
);
4658 mutex_lock(&cgroup_mutex
);
4659 cgroup_subsys
[ss
->subsys_id
] = ss
;
4662 * no ss->css_alloc seems to need anything important in the ss
4663 * struct, so this can happen first (i.e. before the dummy root
4666 css
= ss
->css_alloc(cgroup_css(cgroup_dummy_top
, ss
));
4668 /* failure case - need to deassign the cgroup_subsys[] slot. */
4669 cgroup_subsys
[ss
->subsys_id
] = NULL
;
4670 mutex_unlock(&cgroup_mutex
);
4671 return PTR_ERR(css
);
4674 list_add(&ss
->sibling
, &cgroup_dummy_root
.subsys_list
);
4675 ss
->root
= &cgroup_dummy_root
;
4677 /* our new subsystem will be attached to the dummy hierarchy. */
4678 init_css(css
, ss
, cgroup_dummy_top
);
4679 /* init_idr must be after init_css() because it sets css->id. */
4681 ret
= cgroup_init_idr(ss
, css
);
4687 * Now we need to entangle the css into the existing css_sets. unlike
4688 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4689 * will need a new pointer to it; done by iterating the css_set_table.
4690 * furthermore, modifying the existing css_sets will corrupt the hash
4691 * table state, so each changed css_set will need its hash recomputed.
4692 * this is all done under the css_set_lock.
4694 write_lock(&css_set_lock
);
4695 hash_for_each_safe(css_set_table
, i
, tmp
, cset
, hlist
) {
4696 /* skip entries that we already rehashed */
4697 if (cset
->subsys
[ss
->subsys_id
])
4699 /* remove existing entry */
4700 hash_del(&cset
->hlist
);
4702 cset
->subsys
[ss
->subsys_id
] = css
;
4703 /* recompute hash and restore entry */
4704 key
= css_set_hash(cset
->subsys
);
4705 hash_add(css_set_table
, &cset
->hlist
, key
);
4707 write_unlock(&css_set_lock
);
4709 ret
= online_css(css
);
4714 mutex_unlock(&cgroup_mutex
);
4718 mutex_unlock(&cgroup_mutex
);
4719 /* @ss can't be mounted here as try_module_get() would fail */
4720 cgroup_unload_subsys(ss
);
4723 EXPORT_SYMBOL_GPL(cgroup_load_subsys
);
4726 * cgroup_unload_subsys: unload a modular subsystem
4727 * @ss: the subsystem to unload
4729 * This function should be called in a modular subsystem's exitcall. When this
4730 * function is invoked, the refcount on the subsystem's module will be 0, so
4731 * the subsystem will not be attached to any hierarchy.
4733 void cgroup_unload_subsys(struct cgroup_subsys
*ss
)
4735 struct cgrp_cset_link
*link
;
4737 BUG_ON(ss
->module
== NULL
);
4740 * we shouldn't be called if the subsystem is in use, and the use of
4741 * try_module_get() in rebind_subsystems() should ensure that it
4742 * doesn't start being used while we're killing it off.
4744 BUG_ON(ss
->root
!= &cgroup_dummy_root
);
4746 mutex_lock(&cgroup_mutex
);
4748 offline_css(cgroup_css(cgroup_dummy_top
, ss
));
4751 idr_destroy(&ss
->idr
);
4753 /* deassign the subsys_id */
4754 cgroup_subsys
[ss
->subsys_id
] = NULL
;
4756 /* remove subsystem from the dummy root's list of subsystems */
4757 list_del_init(&ss
->sibling
);
4760 * disentangle the css from all css_sets attached to the dummy
4761 * top. as in loading, we need to pay our respects to the hashtable
4764 write_lock(&css_set_lock
);
4765 list_for_each_entry(link
, &cgroup_dummy_top
->cset_links
, cset_link
) {
4766 struct css_set
*cset
= link
->cset
;
4769 hash_del(&cset
->hlist
);
4770 cset
->subsys
[ss
->subsys_id
] = NULL
;
4771 key
= css_set_hash(cset
->subsys
);
4772 hash_add(css_set_table
, &cset
->hlist
, key
);
4774 write_unlock(&css_set_lock
);
4777 * remove subsystem's css from the cgroup_dummy_top and free it -
4778 * need to free before marking as null because ss->css_free needs
4779 * the cgrp->subsys pointer to find their state. note that this
4780 * also takes care of freeing the css_id.
4782 ss
->css_free(cgroup_css(cgroup_dummy_top
, ss
));
4783 RCU_INIT_POINTER(cgroup_dummy_top
->subsys
[ss
->subsys_id
], NULL
);
4785 mutex_unlock(&cgroup_mutex
);
4787 EXPORT_SYMBOL_GPL(cgroup_unload_subsys
);
4790 * cgroup_init_early - cgroup initialization at system boot
4792 * Initialize cgroups at system boot, and initialize any
4793 * subsystems that request early init.
4795 int __init
cgroup_init_early(void)
4797 struct cgroup_subsys
*ss
;
4800 atomic_set(&init_css_set
.refcount
, 1);
4801 INIT_LIST_HEAD(&init_css_set
.cgrp_links
);
4802 INIT_LIST_HEAD(&init_css_set
.tasks
);
4803 INIT_HLIST_NODE(&init_css_set
.hlist
);
4805 init_cgroup_root(&cgroup_dummy_root
);
4806 cgroup_root_count
= 1;
4807 RCU_INIT_POINTER(init_task
.cgroups
, &init_css_set
);
4809 init_cgrp_cset_link
.cset
= &init_css_set
;
4810 init_cgrp_cset_link
.cgrp
= cgroup_dummy_top
;
4811 list_add(&init_cgrp_cset_link
.cset_link
, &cgroup_dummy_top
->cset_links
);
4812 list_add(&init_cgrp_cset_link
.cgrp_link
, &init_css_set
.cgrp_links
);
4814 /* at bootup time, we don't worry about modular subsystems */
4815 for_each_builtin_subsys(ss
, i
) {
4817 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
4818 BUG_ON(!ss
->css_alloc
);
4819 BUG_ON(!ss
->css_free
);
4820 if (ss
->subsys_id
!= i
) {
4821 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
4822 ss
->name
, ss
->subsys_id
);
4827 cgroup_init_subsys(ss
);
4833 * cgroup_init - cgroup initialization
4835 * Register cgroup filesystem and /proc file, and initialize
4836 * any subsystems that didn't request early init.
4838 int __init
cgroup_init(void)
4840 struct cgroup_subsys
*ss
;
4844 err
= bdi_init(&cgroup_backing_dev_info
);
4848 for_each_builtin_subsys(ss
, i
) {
4849 if (!ss
->early_init
)
4850 cgroup_init_subsys(ss
);
4852 cgroup_init_idr(ss
, init_css_set
.subsys
[ss
->subsys_id
]);
4855 /* allocate id for the dummy hierarchy */
4856 mutex_lock(&cgroup_mutex
);
4857 mutex_lock(&cgroup_root_mutex
);
4859 /* Add init_css_set to the hash table */
4860 key
= css_set_hash(init_css_set
.subsys
);
4861 hash_add(css_set_table
, &init_css_set
.hlist
, key
);
4863 BUG_ON(cgroup_init_root_id(&cgroup_dummy_root
, 0, 1));
4865 err
= idr_alloc(&cgroup_dummy_root
.cgroup_idr
, cgroup_dummy_top
,
4869 mutex_unlock(&cgroup_root_mutex
);
4870 mutex_unlock(&cgroup_mutex
);
4872 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
4878 err
= register_filesystem(&cgroup_fs_type
);
4880 kobject_put(cgroup_kobj
);
4884 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
4888 bdi_destroy(&cgroup_backing_dev_info
);
4894 * proc_cgroup_show()
4895 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4896 * - Used for /proc/<pid>/cgroup.
4897 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4898 * doesn't really matter if tsk->cgroup changes after we read it,
4899 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4900 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4901 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4902 * cgroup to top_cgroup.
4905 /* TODO: Use a proper seq_file iterator */
4906 int proc_cgroup_show(struct seq_file
*m
, void *v
)
4909 struct task_struct
*tsk
;
4912 struct cgroupfs_root
*root
;
4915 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4921 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
4927 mutex_lock(&cgroup_mutex
);
4929 for_each_active_root(root
) {
4930 struct cgroup_subsys
*ss
;
4931 struct cgroup
*cgrp
;
4934 seq_printf(m
, "%d:", root
->hierarchy_id
);
4935 for_each_root_subsys(root
, ss
)
4936 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
4937 if (strlen(root
->name
))
4938 seq_printf(m
, "%sname=%s", count
? "," : "",
4941 cgrp
= task_cgroup_from_root(tsk
, root
);
4942 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
4950 mutex_unlock(&cgroup_mutex
);
4951 put_task_struct(tsk
);
4958 /* Display information about each subsystem and each hierarchy */
4959 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
4961 struct cgroup_subsys
*ss
;
4964 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4966 * ideally we don't want subsystems moving around while we do this.
4967 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4968 * subsys/hierarchy state.
4970 mutex_lock(&cgroup_mutex
);
4972 for_each_subsys(ss
, i
)
4973 seq_printf(m
, "%s\t%d\t%d\t%d\n",
4974 ss
->name
, ss
->root
->hierarchy_id
,
4975 ss
->root
->number_of_cgroups
, !ss
->disabled
);
4977 mutex_unlock(&cgroup_mutex
);
4981 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
4983 return single_open(file
, proc_cgroupstats_show
, NULL
);
4986 static const struct file_operations proc_cgroupstats_operations
= {
4987 .open
= cgroupstats_open
,
4989 .llseek
= seq_lseek
,
4990 .release
= single_release
,
4994 * cgroup_fork - attach newly forked task to its parents cgroup.
4995 * @child: pointer to task_struct of forking parent process.
4997 * Description: A task inherits its parent's cgroup at fork().
4999 * A pointer to the shared css_set was automatically copied in
5000 * fork.c by dup_task_struct(). However, we ignore that copy, since
5001 * it was not made under the protection of RCU or cgroup_mutex, so
5002 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
5003 * have already changed current->cgroups, allowing the previously
5004 * referenced cgroup group to be removed and freed.
5006 * At the point that cgroup_fork() is called, 'current' is the parent
5007 * task, and the passed argument 'child' points to the child task.
5009 void cgroup_fork(struct task_struct
*child
)
5012 get_css_set(task_css_set(current
));
5013 child
->cgroups
= current
->cgroups
;
5014 task_unlock(current
);
5015 INIT_LIST_HEAD(&child
->cg_list
);
5019 * cgroup_post_fork - called on a new task after adding it to the task list
5020 * @child: the task in question
5022 * Adds the task to the list running through its css_set if necessary and
5023 * call the subsystem fork() callbacks. Has to be after the task is
5024 * visible on the task list in case we race with the first call to
5025 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
5028 void cgroup_post_fork(struct task_struct
*child
)
5030 struct cgroup_subsys
*ss
;
5034 * use_task_css_set_links is set to 1 before we walk the tasklist
5035 * under the tasklist_lock and we read it here after we added the child
5036 * to the tasklist under the tasklist_lock as well. If the child wasn't
5037 * yet in the tasklist when we walked through it from
5038 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
5039 * should be visible now due to the paired locking and barriers implied
5040 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
5041 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
5044 if (use_task_css_set_links
) {
5045 write_lock(&css_set_lock
);
5047 if (list_empty(&child
->cg_list
))
5048 list_add(&child
->cg_list
, &task_css_set(child
)->tasks
);
5050 write_unlock(&css_set_lock
);
5054 * Call ss->fork(). This must happen after @child is linked on
5055 * css_set; otherwise, @child might change state between ->fork()
5056 * and addition to css_set.
5058 if (need_forkexit_callback
) {
5060 * fork/exit callbacks are supported only for builtin
5061 * subsystems, and the builtin section of the subsys
5062 * array is immutable, so we don't need to lock the
5063 * subsys array here. On the other hand, modular section
5064 * of the array can be freed at module unload, so we
5067 for_each_builtin_subsys(ss
, i
)
5074 * cgroup_exit - detach cgroup from exiting task
5075 * @tsk: pointer to task_struct of exiting process
5076 * @run_callback: run exit callbacks?
5078 * Description: Detach cgroup from @tsk and release it.
5080 * Note that cgroups marked notify_on_release force every task in
5081 * them to take the global cgroup_mutex mutex when exiting.
5082 * This could impact scaling on very large systems. Be reluctant to
5083 * use notify_on_release cgroups where very high task exit scaling
5084 * is required on large systems.
5086 * the_top_cgroup_hack:
5088 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
5090 * We call cgroup_exit() while the task is still competent to
5091 * handle notify_on_release(), then leave the task attached to the
5092 * root cgroup in each hierarchy for the remainder of its exit.
5094 * To do this properly, we would increment the reference count on
5095 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
5096 * code we would add a second cgroup function call, to drop that
5097 * reference. This would just create an unnecessary hot spot on
5098 * the top_cgroup reference count, to no avail.
5100 * Normally, holding a reference to a cgroup without bumping its
5101 * count is unsafe. The cgroup could go away, or someone could
5102 * attach us to a different cgroup, decrementing the count on
5103 * the first cgroup that we never incremented. But in this case,
5104 * top_cgroup isn't going away, and either task has PF_EXITING set,
5105 * which wards off any cgroup_attach_task() attempts, or task is a failed
5106 * fork, never visible to cgroup_attach_task.
5108 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
5110 struct cgroup_subsys
*ss
;
5111 struct css_set
*cset
;
5115 * Unlink from the css_set task list if necessary.
5116 * Optimistically check cg_list before taking
5119 if (!list_empty(&tsk
->cg_list
)) {
5120 write_lock(&css_set_lock
);
5121 if (!list_empty(&tsk
->cg_list
))
5122 list_del_init(&tsk
->cg_list
);
5123 write_unlock(&css_set_lock
);
5126 /* Reassign the task to the init_css_set. */
5128 cset
= task_css_set(tsk
);
5129 RCU_INIT_POINTER(tsk
->cgroups
, &init_css_set
);
5131 if (run_callbacks
&& need_forkexit_callback
) {
5133 * fork/exit callbacks are supported only for builtin
5134 * subsystems, see cgroup_post_fork() for details.
5136 for_each_builtin_subsys(ss
, i
) {
5138 struct cgroup_subsys_state
*old_css
= cset
->subsys
[i
];
5139 struct cgroup_subsys_state
*css
= task_css(tsk
, i
);
5141 ss
->exit(css
, old_css
, tsk
);
5147 put_css_set_taskexit(cset
);
5150 static void check_for_release(struct cgroup
*cgrp
)
5152 if (cgroup_is_releasable(cgrp
) &&
5153 list_empty(&cgrp
->cset_links
) && list_empty(&cgrp
->children
)) {
5155 * Control Group is currently removeable. If it's not
5156 * already queued for a userspace notification, queue
5159 int need_schedule_work
= 0;
5161 raw_spin_lock(&release_list_lock
);
5162 if (!cgroup_is_dead(cgrp
) &&
5163 list_empty(&cgrp
->release_list
)) {
5164 list_add(&cgrp
->release_list
, &release_list
);
5165 need_schedule_work
= 1;
5167 raw_spin_unlock(&release_list_lock
);
5168 if (need_schedule_work
)
5169 schedule_work(&release_agent_work
);
5174 * Notify userspace when a cgroup is released, by running the
5175 * configured release agent with the name of the cgroup (path
5176 * relative to the root of cgroup file system) as the argument.
5178 * Most likely, this user command will try to rmdir this cgroup.
5180 * This races with the possibility that some other task will be
5181 * attached to this cgroup before it is removed, or that some other
5182 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5183 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5184 * unused, and this cgroup will be reprieved from its death sentence,
5185 * to continue to serve a useful existence. Next time it's released,
5186 * we will get notified again, if it still has 'notify_on_release' set.
5188 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5189 * means only wait until the task is successfully execve()'d. The
5190 * separate release agent task is forked by call_usermodehelper(),
5191 * then control in this thread returns here, without waiting for the
5192 * release agent task. We don't bother to wait because the caller of
5193 * this routine has no use for the exit status of the release agent
5194 * task, so no sense holding our caller up for that.
5196 static void cgroup_release_agent(struct work_struct
*work
)
5198 BUG_ON(work
!= &release_agent_work
);
5199 mutex_lock(&cgroup_mutex
);
5200 raw_spin_lock(&release_list_lock
);
5201 while (!list_empty(&release_list
)) {
5202 char *argv
[3], *envp
[3];
5204 char *pathbuf
= NULL
, *agentbuf
= NULL
;
5205 struct cgroup
*cgrp
= list_entry(release_list
.next
,
5208 list_del_init(&cgrp
->release_list
);
5209 raw_spin_unlock(&release_list_lock
);
5210 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
5213 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0)
5215 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
5220 argv
[i
++] = agentbuf
;
5221 argv
[i
++] = pathbuf
;
5225 /* minimal command environment */
5226 envp
[i
++] = "HOME=/";
5227 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5230 /* Drop the lock while we invoke the usermode helper,
5231 * since the exec could involve hitting disk and hence
5232 * be a slow process */
5233 mutex_unlock(&cgroup_mutex
);
5234 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
5235 mutex_lock(&cgroup_mutex
);
5239 raw_spin_lock(&release_list_lock
);
5241 raw_spin_unlock(&release_list_lock
);
5242 mutex_unlock(&cgroup_mutex
);
5245 static int __init
cgroup_disable(char *str
)
5247 struct cgroup_subsys
*ss
;
5251 while ((token
= strsep(&str
, ",")) != NULL
) {
5256 * cgroup_disable, being at boot time, can't know about
5257 * module subsystems, so we don't worry about them.
5259 for_each_builtin_subsys(ss
, i
) {
5260 if (!strcmp(token
, ss
->name
)) {
5262 printk(KERN_INFO
"Disabling %s control group"
5263 " subsystem\n", ss
->name
);
5270 __setup("cgroup_disable=", cgroup_disable
);
5273 * Functons for CSS ID.
5276 /* to get ID other than 0, this should be called when !cgroup_is_dead() */
5277 unsigned short css_id(struct cgroup_subsys_state
*css
)
5279 struct css_id
*cssid
;
5282 * This css_id() can return correct value when somone has refcnt
5283 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5284 * it's unchanged until freed.
5286 cssid
= rcu_dereference_raw(css
->id
);
5292 EXPORT_SYMBOL_GPL(css_id
);
5295 * css_is_ancestor - test "root" css is an ancestor of "child"
5296 * @child: the css to be tested.
5297 * @root: the css supporsed to be an ancestor of the child.
5299 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5300 * this function reads css->id, the caller must hold rcu_read_lock().
5301 * But, considering usual usage, the csses should be valid objects after test.
5302 * Assuming that the caller will do some action to the child if this returns
5303 * returns true, the caller must take "child";s reference count.
5304 * If "child" is valid object and this returns true, "root" is valid, too.
5307 bool css_is_ancestor(struct cgroup_subsys_state
*child
,
5308 const struct cgroup_subsys_state
*root
)
5310 struct css_id
*child_id
;
5311 struct css_id
*root_id
;
5313 child_id
= rcu_dereference(child
->id
);
5316 root_id
= rcu_dereference(root
->id
);
5319 if (child_id
->depth
< root_id
->depth
)
5321 if (child_id
->stack
[root_id
->depth
] != root_id
->id
)
5326 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
)
5328 struct css_id
*id
= rcu_dereference_protected(css
->id
, true);
5330 /* When this is called before css_id initialization, id can be NULL */
5334 BUG_ON(!ss
->use_id
);
5336 rcu_assign_pointer(id
->css
, NULL
);
5337 rcu_assign_pointer(css
->id
, NULL
);
5338 spin_lock(&ss
->id_lock
);
5339 idr_remove(&ss
->idr
, id
->id
);
5340 spin_unlock(&ss
->id_lock
);
5341 kfree_rcu(id
, rcu_head
);
5343 EXPORT_SYMBOL_GPL(free_css_id
);
5346 * This is called by init or create(). Then, calls to this function are
5347 * always serialized (By cgroup_mutex() at create()).
5350 static struct css_id
*get_new_cssid(struct cgroup_subsys
*ss
, int depth
)
5352 struct css_id
*newid
;
5355 BUG_ON(!ss
->use_id
);
5357 size
= sizeof(*newid
) + sizeof(unsigned short) * (depth
+ 1);
5358 newid
= kzalloc(size
, GFP_KERNEL
);
5360 return ERR_PTR(-ENOMEM
);
5362 idr_preload(GFP_KERNEL
);
5363 spin_lock(&ss
->id_lock
);
5364 /* Don't use 0. allocates an ID of 1-65535 */
5365 ret
= idr_alloc(&ss
->idr
, newid
, 1, CSS_ID_MAX
+ 1, GFP_NOWAIT
);
5366 spin_unlock(&ss
->id_lock
);
5369 /* Returns error when there are no free spaces for new ID.*/
5374 newid
->depth
= depth
;
5378 return ERR_PTR(ret
);
5382 static int __init_or_module
cgroup_init_idr(struct cgroup_subsys
*ss
,
5383 struct cgroup_subsys_state
*rootcss
)
5385 struct css_id
*newid
;
5387 spin_lock_init(&ss
->id_lock
);
5390 newid
= get_new_cssid(ss
, 0);
5392 return PTR_ERR(newid
);
5394 newid
->stack
[0] = newid
->id
;
5395 RCU_INIT_POINTER(newid
->css
, rootcss
);
5396 RCU_INIT_POINTER(rootcss
->id
, newid
);
5400 static int alloc_css_id(struct cgroup_subsys_state
*child_css
)
5402 struct cgroup_subsys_state
*parent_css
= css_parent(child_css
);
5403 struct css_id
*child_id
, *parent_id
;
5406 parent_id
= rcu_dereference_protected(parent_css
->id
, true);
5407 depth
= parent_id
->depth
+ 1;
5409 child_id
= get_new_cssid(child_css
->ss
, depth
);
5410 if (IS_ERR(child_id
))
5411 return PTR_ERR(child_id
);
5413 for (i
= 0; i
< depth
; i
++)
5414 child_id
->stack
[i
] = parent_id
->stack
[i
];
5415 child_id
->stack
[depth
] = child_id
->id
;
5417 * child_id->css pointer will be set after this cgroup is available
5418 * see cgroup_populate_dir()
5420 rcu_assign_pointer(child_css
->id
, child_id
);
5426 * css_lookup - lookup css by id
5427 * @ss: cgroup subsys to be looked into.
5430 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5431 * NULL if not. Should be called under rcu_read_lock()
5433 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
)
5435 struct css_id
*cssid
= NULL
;
5437 BUG_ON(!ss
->use_id
);
5438 cssid
= idr_find(&ss
->idr
, id
);
5440 if (unlikely(!cssid
))
5443 return rcu_dereference(cssid
->css
);
5445 EXPORT_SYMBOL_GPL(css_lookup
);
5448 * css_from_dir - get corresponding css from the dentry of a cgroup dir
5449 * @dentry: directory dentry of interest
5450 * @ss: subsystem of interest
5452 * Must be called under RCU read lock. The caller is responsible for
5453 * pinning the returned css if it needs to be accessed outside the RCU
5456 struct cgroup_subsys_state
*css_from_dir(struct dentry
*dentry
,
5457 struct cgroup_subsys
*ss
)
5459 struct cgroup
*cgrp
;
5461 WARN_ON_ONCE(!rcu_read_lock_held());
5463 /* is @dentry a cgroup dir? */
5464 if (!dentry
->d_inode
||
5465 dentry
->d_inode
->i_op
!= &cgroup_dir_inode_operations
)
5466 return ERR_PTR(-EBADF
);
5468 cgrp
= __d_cgrp(dentry
);
5469 return cgroup_css(cgrp
, ss
) ?: ERR_PTR(-ENOENT
);
5473 * css_from_id - lookup css by id
5474 * @id: the cgroup id
5475 * @ss: cgroup subsys to be looked into
5477 * Returns the css if there's valid one with @id, otherwise returns NULL.
5478 * Should be called under rcu_read_lock().
5480 struct cgroup_subsys_state
*css_from_id(int id
, struct cgroup_subsys
*ss
)
5482 struct cgroup
*cgrp
;
5484 rcu_lockdep_assert(rcu_read_lock_held() ||
5485 lockdep_is_held(&cgroup_mutex
),
5486 "css_from_id() needs proper protection");
5488 cgrp
= idr_find(&ss
->root
->cgroup_idr
, id
);
5490 return cgroup_css(cgrp
, ss
);
5494 #ifdef CONFIG_CGROUP_DEBUG
5495 static struct cgroup_subsys_state
*
5496 debug_css_alloc(struct cgroup_subsys_state
*parent_css
)
5498 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5501 return ERR_PTR(-ENOMEM
);
5506 static void debug_css_free(struct cgroup_subsys_state
*css
)
5511 static u64
debug_taskcount_read(struct cgroup_subsys_state
*css
,
5514 return cgroup_task_count(css
->cgroup
);
5517 static u64
current_css_set_read(struct cgroup_subsys_state
*css
,
5520 return (u64
)(unsigned long)current
->cgroups
;
5523 static u64
current_css_set_refcount_read(struct cgroup_subsys_state
*css
,
5529 count
= atomic_read(&task_css_set(current
)->refcount
);
5534 static int current_css_set_cg_links_read(struct cgroup_subsys_state
*css
,
5536 struct seq_file
*seq
)
5538 struct cgrp_cset_link
*link
;
5539 struct css_set
*cset
;
5541 read_lock(&css_set_lock
);
5543 cset
= rcu_dereference(current
->cgroups
);
5544 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
5545 struct cgroup
*c
= link
->cgrp
;
5549 name
= c
->dentry
->d_name
.name
;
5552 seq_printf(seq
, "Root %d group %s\n",
5553 c
->root
->hierarchy_id
, name
);
5556 read_unlock(&css_set_lock
);
5560 #define MAX_TASKS_SHOWN_PER_CSS 25
5561 static int cgroup_css_links_read(struct cgroup_subsys_state
*css
,
5562 struct cftype
*cft
, struct seq_file
*seq
)
5564 struct cgrp_cset_link
*link
;
5566 read_lock(&css_set_lock
);
5567 list_for_each_entry(link
, &css
->cgroup
->cset_links
, cset_link
) {
5568 struct css_set
*cset
= link
->cset
;
5569 struct task_struct
*task
;
5571 seq_printf(seq
, "css_set %p\n", cset
);
5572 list_for_each_entry(task
, &cset
->tasks
, cg_list
) {
5573 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
) {
5574 seq_puts(seq
, " ...\n");
5577 seq_printf(seq
, " task %d\n",
5578 task_pid_vnr(task
));
5582 read_unlock(&css_set_lock
);
5586 static u64
releasable_read(struct cgroup_subsys_state
*css
, struct cftype
*cft
)
5588 return test_bit(CGRP_RELEASABLE
, &css
->cgroup
->flags
);
5591 static struct cftype debug_files
[] = {
5593 .name
= "taskcount",
5594 .read_u64
= debug_taskcount_read
,
5598 .name
= "current_css_set",
5599 .read_u64
= current_css_set_read
,
5603 .name
= "current_css_set_refcount",
5604 .read_u64
= current_css_set_refcount_read
,
5608 .name
= "current_css_set_cg_links",
5609 .read_seq_string
= current_css_set_cg_links_read
,
5613 .name
= "cgroup_css_links",
5614 .read_seq_string
= cgroup_css_links_read
,
5618 .name
= "releasable",
5619 .read_u64
= releasable_read
,
5625 struct cgroup_subsys debug_subsys
= {
5627 .css_alloc
= debug_css_alloc
,
5628 .css_free
= debug_css_free
,
5629 .subsys_id
= debug_subsys_id
,
5630 .base_cftypes
= debug_files
,
5632 #endif /* CONFIG_CGROUP_DEBUG */