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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
31 #include <linux/cgroup.h>
32 #include <linux/cred.h>
33 #include <linux/ctype.h>
34 #include <linux/errno.h>
35 #include <linux/init_task.h>
36 #include <linux/kernel.h>
37 #include <linux/list.h>
38 #include <linux/magic.h>
40 #include <linux/mutex.h>
41 #include <linux/mount.h>
42 #include <linux/pagemap.h>
43 #include <linux/proc_fs.h>
44 #include <linux/rcupdate.h>
45 #include <linux/sched.h>
46 #include <linux/slab.h>
47 #include <linux/spinlock.h>
48 #include <linux/rwsem.h>
49 #include <linux/string.h>
50 #include <linux/sort.h>
51 #include <linux/kmod.h>
52 #include <linux/delayacct.h>
53 #include <linux/cgroupstats.h>
54 #include <linux/hashtable.h>
55 #include <linux/pid_namespace.h>
56 #include <linux/idr.h>
57 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
58 #include <linux/kthread.h>
59 #include <linux/delay.h>
61 #include <linux/atomic.h>
64 * pidlists linger the following amount before being destroyed. The goal
65 * is avoiding frequent destruction in the middle of consecutive read calls
66 * Expiring in the middle is a performance problem not a correctness one.
67 * 1 sec should be enough.
69 #define CGROUP_PIDLIST_DESTROY_DELAY HZ
71 #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
75 * cgroup_mutex is the master lock. Any modification to cgroup or its
76 * hierarchy must be performed while holding it.
78 * css_set_rwsem protects task->cgroups pointer, the list of css_set
79 * objects, and the chain of tasks off each css_set.
81 * These locks are exported if CONFIG_PROVE_RCU so that accessors in
82 * cgroup.h can use them for lockdep annotations.
84 #ifdef CONFIG_PROVE_RCU
85 DEFINE_MUTEX(cgroup_mutex
);
86 DECLARE_RWSEM(css_set_rwsem
);
87 EXPORT_SYMBOL_GPL(cgroup_mutex
);
88 EXPORT_SYMBOL_GPL(css_set_rwsem
);
90 static DEFINE_MUTEX(cgroup_mutex
);
91 static DECLARE_RWSEM(css_set_rwsem
);
95 * Protects cgroup_idr and css_idr so that IDs can be released without
96 * grabbing cgroup_mutex.
98 static DEFINE_SPINLOCK(cgroup_idr_lock
);
101 * Protects cgroup_subsys->release_agent_path. Modifying it also requires
102 * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock.
104 static DEFINE_SPINLOCK(release_agent_path_lock
);
106 #define cgroup_assert_mutex_or_rcu_locked() \
107 rcu_lockdep_assert(rcu_read_lock_held() || \
108 lockdep_is_held(&cgroup_mutex), \
109 "cgroup_mutex or RCU read lock required");
112 * cgroup destruction makes heavy use of work items and there can be a lot
113 * of concurrent destructions. Use a separate workqueue so that cgroup
114 * destruction work items don't end up filling up max_active of system_wq
115 * which may lead to deadlock.
117 static struct workqueue_struct
*cgroup_destroy_wq
;
120 * pidlist destructions need to be flushed on cgroup destruction. Use a
121 * separate workqueue as flush domain.
123 static struct workqueue_struct
*cgroup_pidlist_destroy_wq
;
125 /* generate an array of cgroup subsystem pointers */
126 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
127 static struct cgroup_subsys
*cgroup_subsys
[] = {
128 #include <linux/cgroup_subsys.h>
132 /* array of cgroup subsystem names */
133 #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
134 static const char *cgroup_subsys_name
[] = {
135 #include <linux/cgroup_subsys.h>
140 * The default hierarchy, reserved for the subsystems that are otherwise
141 * unattached - it never has more than a single cgroup, and all tasks are
142 * part of that cgroup.
144 struct cgroup_root cgrp_dfl_root
;
147 * The default hierarchy always exists but is hidden until mounted for the
148 * first time. This is for backward compatibility.
150 static bool cgrp_dfl_root_visible
;
153 * Set by the boot param of the same name and makes subsystems with NULL
154 * ->dfl_files to use ->legacy_files on the default hierarchy.
156 static bool cgroup_legacy_files_on_dfl
;
158 /* some controllers are not supported in the default hierarchy */
159 static unsigned int cgrp_dfl_root_inhibit_ss_mask
;
161 /* The list of hierarchy roots */
163 static LIST_HEAD(cgroup_roots
);
164 static int cgroup_root_count
;
166 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
167 static DEFINE_IDR(cgroup_hierarchy_idr
);
170 * Assign a monotonically increasing serial number to csses. It guarantees
171 * cgroups with bigger numbers are newer than those with smaller numbers.
172 * Also, as csses are always appended to the parent's ->children list, it
173 * guarantees that sibling csses are always sorted in the ascending serial
174 * number order on the list. Protected by cgroup_mutex.
176 static u64 css_serial_nr_next
= 1;
178 /* This flag indicates whether tasks in the fork and exit paths should
179 * check for fork/exit handlers to call. This avoids us having to do
180 * extra work in the fork/exit path if none of the subsystems need to
183 static int need_forkexit_callback __read_mostly
;
185 static struct cftype cgroup_dfl_base_files
[];
186 static struct cftype cgroup_legacy_base_files
[];
188 static void cgroup_put(struct cgroup
*cgrp
);
189 static int rebind_subsystems(struct cgroup_root
*dst_root
,
190 unsigned int ss_mask
);
191 static int cgroup_destroy_locked(struct cgroup
*cgrp
);
192 static int create_css(struct cgroup
*cgrp
, struct cgroup_subsys
*ss
,
194 static void css_release(struct percpu_ref
*ref
);
195 static void kill_css(struct cgroup_subsys_state
*css
);
196 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cftype cfts
[],
198 static void cgroup_pidlist_destroy_all(struct cgroup
*cgrp
);
200 /* IDR wrappers which synchronize using cgroup_idr_lock */
201 static int cgroup_idr_alloc(struct idr
*idr
, void *ptr
, int start
, int end
,
206 idr_preload(gfp_mask
);
207 spin_lock_bh(&cgroup_idr_lock
);
208 ret
= idr_alloc(idr
, ptr
, start
, end
, gfp_mask
);
209 spin_unlock_bh(&cgroup_idr_lock
);
214 static void *cgroup_idr_replace(struct idr
*idr
, void *ptr
, int id
)
218 spin_lock_bh(&cgroup_idr_lock
);
219 ret
= idr_replace(idr
, ptr
, id
);
220 spin_unlock_bh(&cgroup_idr_lock
);
224 static void cgroup_idr_remove(struct idr
*idr
, int id
)
226 spin_lock_bh(&cgroup_idr_lock
);
228 spin_unlock_bh(&cgroup_idr_lock
);
231 static struct cgroup
*cgroup_parent(struct cgroup
*cgrp
)
233 struct cgroup_subsys_state
*parent_css
= cgrp
->self
.parent
;
236 return container_of(parent_css
, struct cgroup
, self
);
241 * cgroup_css - obtain a cgroup's css for the specified subsystem
242 * @cgrp: the cgroup of interest
243 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
245 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
246 * function must be called either under cgroup_mutex or rcu_read_lock() and
247 * the caller is responsible for pinning the returned css if it wants to
248 * keep accessing it outside the said locks. This function may return
249 * %NULL if @cgrp doesn't have @subsys_id enabled.
251 static struct cgroup_subsys_state
*cgroup_css(struct cgroup
*cgrp
,
252 struct cgroup_subsys
*ss
)
255 return rcu_dereference_check(cgrp
->subsys
[ss
->id
],
256 lockdep_is_held(&cgroup_mutex
));
262 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
263 * @cgrp: the cgroup of interest
264 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
266 * Similar to cgroup_css() but returns the effctive css, which is defined
267 * as the matching css of the nearest ancestor including self which has @ss
268 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
269 * function is guaranteed to return non-NULL css.
271 static struct cgroup_subsys_state
*cgroup_e_css(struct cgroup
*cgrp
,
272 struct cgroup_subsys
*ss
)
274 lockdep_assert_held(&cgroup_mutex
);
279 if (!(cgrp
->root
->subsys_mask
& (1 << ss
->id
)))
282 while (cgroup_parent(cgrp
) &&
283 !(cgroup_parent(cgrp
)->child_subsys_mask
& (1 << ss
->id
)))
284 cgrp
= cgroup_parent(cgrp
);
286 return cgroup_css(cgrp
, ss
);
289 /* convenient tests for these bits */
290 static inline bool cgroup_is_dead(const struct cgroup
*cgrp
)
292 return !(cgrp
->self
.flags
& CSS_ONLINE
);
295 struct cgroup_subsys_state
*of_css(struct kernfs_open_file
*of
)
297 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
298 struct cftype
*cft
= of_cft(of
);
301 * This is open and unprotected implementation of cgroup_css().
302 * seq_css() is only called from a kernfs file operation which has
303 * an active reference on the file. Because all the subsystem
304 * files are drained before a css is disassociated with a cgroup,
305 * the matching css from the cgroup's subsys table is guaranteed to
306 * be and stay valid until the enclosing operation is complete.
309 return rcu_dereference_raw(cgrp
->subsys
[cft
->ss
->id
]);
313 EXPORT_SYMBOL_GPL(of_css
);
316 * cgroup_is_descendant - test ancestry
317 * @cgrp: the cgroup to be tested
318 * @ancestor: possible ancestor of @cgrp
320 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
321 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
322 * and @ancestor are accessible.
324 bool cgroup_is_descendant(struct cgroup
*cgrp
, struct cgroup
*ancestor
)
327 if (cgrp
== ancestor
)
329 cgrp
= cgroup_parent(cgrp
);
334 static int cgroup_is_releasable(const struct cgroup
*cgrp
)
337 (1 << CGRP_RELEASABLE
) |
338 (1 << CGRP_NOTIFY_ON_RELEASE
);
339 return (cgrp
->flags
& bits
) == bits
;
342 static int notify_on_release(const struct cgroup
*cgrp
)
344 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
348 * for_each_css - iterate all css's of a cgroup
349 * @css: the iteration cursor
350 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
351 * @cgrp: the target cgroup to iterate css's of
353 * Should be called under cgroup_[tree_]mutex.
355 #define for_each_css(css, ssid, cgrp) \
356 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
357 if (!((css) = rcu_dereference_check( \
358 (cgrp)->subsys[(ssid)], \
359 lockdep_is_held(&cgroup_mutex)))) { } \
363 * for_each_e_css - iterate all effective css's of a cgroup
364 * @css: the iteration cursor
365 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
366 * @cgrp: the target cgroup to iterate css's of
368 * Should be called under cgroup_[tree_]mutex.
370 #define for_each_e_css(css, ssid, cgrp) \
371 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
372 if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \
377 * for_each_subsys - iterate all enabled cgroup subsystems
378 * @ss: the iteration cursor
379 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
381 #define for_each_subsys(ss, ssid) \
382 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \
383 (((ss) = cgroup_subsys[ssid]) || true); (ssid)++)
385 /* iterate across the hierarchies */
386 #define for_each_root(root) \
387 list_for_each_entry((root), &cgroup_roots, root_list)
389 /* iterate over child cgrps, lock should be held throughout iteration */
390 #define cgroup_for_each_live_child(child, cgrp) \
391 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
392 if (({ lockdep_assert_held(&cgroup_mutex); \
393 cgroup_is_dead(child); })) \
397 /* the list of cgroups eligible for automatic release. Protected by
398 * release_list_lock */
399 static LIST_HEAD(release_list
);
400 static DEFINE_RAW_SPINLOCK(release_list_lock
);
401 static void cgroup_release_agent(struct work_struct
*work
);
402 static DECLARE_WORK(release_agent_work
, cgroup_release_agent
);
403 static void check_for_release(struct cgroup
*cgrp
);
406 * A cgroup can be associated with multiple css_sets as different tasks may
407 * belong to different cgroups on different hierarchies. In the other
408 * direction, a css_set is naturally associated with multiple cgroups.
409 * This M:N relationship is represented by the following link structure
410 * which exists for each association and allows traversing the associations
413 struct cgrp_cset_link
{
414 /* the cgroup and css_set this link associates */
416 struct css_set
*cset
;
418 /* list of cgrp_cset_links anchored at cgrp->cset_links */
419 struct list_head cset_link
;
421 /* list of cgrp_cset_links anchored at css_set->cgrp_links */
422 struct list_head cgrp_link
;
426 * The default css_set - used by init and its children prior to any
427 * hierarchies being mounted. It contains a pointer to the root state
428 * for each subsystem. Also used to anchor the list of css_sets. Not
429 * reference-counted, to improve performance when child cgroups
430 * haven't been created.
432 struct css_set init_css_set
= {
433 .refcount
= ATOMIC_INIT(1),
434 .cgrp_links
= LIST_HEAD_INIT(init_css_set
.cgrp_links
),
435 .tasks
= LIST_HEAD_INIT(init_css_set
.tasks
),
436 .mg_tasks
= LIST_HEAD_INIT(init_css_set
.mg_tasks
),
437 .mg_preload_node
= LIST_HEAD_INIT(init_css_set
.mg_preload_node
),
438 .mg_node
= LIST_HEAD_INIT(init_css_set
.mg_node
),
441 static int css_set_count
= 1; /* 1 for init_css_set */
444 * cgroup_update_populated - updated populated count of a cgroup
445 * @cgrp: the target cgroup
446 * @populated: inc or dec populated count
448 * @cgrp is either getting the first task (css_set) or losing the last.
449 * Update @cgrp->populated_cnt accordingly. The count is propagated
450 * towards root so that a given cgroup's populated_cnt is zero iff the
451 * cgroup and all its descendants are empty.
453 * @cgrp's interface file "cgroup.populated" is zero if
454 * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt
455 * changes from or to zero, userland is notified that the content of the
456 * interface file has changed. This can be used to detect when @cgrp and
457 * its descendants become populated or empty.
459 static void cgroup_update_populated(struct cgroup
*cgrp
, bool populated
)
461 lockdep_assert_held(&css_set_rwsem
);
467 trigger
= !cgrp
->populated_cnt
++;
469 trigger
= !--cgrp
->populated_cnt
;
474 if (cgrp
->populated_kn
)
475 kernfs_notify(cgrp
->populated_kn
);
476 cgrp
= cgroup_parent(cgrp
);
481 * hash table for cgroup groups. This improves the performance to find
482 * an existing css_set. This hash doesn't (currently) take into
483 * account cgroups in empty hierarchies.
485 #define CSS_SET_HASH_BITS 7
486 static DEFINE_HASHTABLE(css_set_table
, CSS_SET_HASH_BITS
);
488 static unsigned long css_set_hash(struct cgroup_subsys_state
*css
[])
490 unsigned long key
= 0UL;
491 struct cgroup_subsys
*ss
;
494 for_each_subsys(ss
, i
)
495 key
+= (unsigned long)css
[i
];
496 key
= (key
>> 16) ^ key
;
501 static void put_css_set_locked(struct css_set
*cset
, bool taskexit
)
503 struct cgrp_cset_link
*link
, *tmp_link
;
504 struct cgroup_subsys
*ss
;
507 lockdep_assert_held(&css_set_rwsem
);
509 if (!atomic_dec_and_test(&cset
->refcount
))
512 /* This css_set is dead. unlink it and release cgroup refcounts */
513 for_each_subsys(ss
, ssid
)
514 list_del(&cset
->e_cset_node
[ssid
]);
515 hash_del(&cset
->hlist
);
518 list_for_each_entry_safe(link
, tmp_link
, &cset
->cgrp_links
, cgrp_link
) {
519 struct cgroup
*cgrp
= link
->cgrp
;
521 list_del(&link
->cset_link
);
522 list_del(&link
->cgrp_link
);
524 /* @cgrp can't go away while we're holding css_set_rwsem */
525 if (list_empty(&cgrp
->cset_links
)) {
526 cgroup_update_populated(cgrp
, false);
527 if (notify_on_release(cgrp
)) {
529 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
530 check_for_release(cgrp
);
537 kfree_rcu(cset
, rcu_head
);
540 static void put_css_set(struct css_set
*cset
, bool taskexit
)
543 * Ensure that the refcount doesn't hit zero while any readers
544 * can see it. Similar to atomic_dec_and_lock(), but for an
547 if (atomic_add_unless(&cset
->refcount
, -1, 1))
550 down_write(&css_set_rwsem
);
551 put_css_set_locked(cset
, taskexit
);
552 up_write(&css_set_rwsem
);
556 * refcounted get/put for css_set objects
558 static inline void get_css_set(struct css_set
*cset
)
560 atomic_inc(&cset
->refcount
);
564 * compare_css_sets - helper function for find_existing_css_set().
565 * @cset: candidate css_set being tested
566 * @old_cset: existing css_set for a task
567 * @new_cgrp: cgroup that's being entered by the task
568 * @template: desired set of css pointers in css_set (pre-calculated)
570 * Returns true if "cset" matches "old_cset" except for the hierarchy
571 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
573 static bool compare_css_sets(struct css_set
*cset
,
574 struct css_set
*old_cset
,
575 struct cgroup
*new_cgrp
,
576 struct cgroup_subsys_state
*template[])
578 struct list_head
*l1
, *l2
;
581 * On the default hierarchy, there can be csets which are
582 * associated with the same set of cgroups but different csses.
583 * Let's first ensure that csses match.
585 if (memcmp(template, cset
->subsys
, sizeof(cset
->subsys
)))
589 * Compare cgroup pointers in order to distinguish between
590 * different cgroups in hierarchies. As different cgroups may
591 * share the same effective css, this comparison is always
594 l1
= &cset
->cgrp_links
;
595 l2
= &old_cset
->cgrp_links
;
597 struct cgrp_cset_link
*link1
, *link2
;
598 struct cgroup
*cgrp1
, *cgrp2
;
602 /* See if we reached the end - both lists are equal length. */
603 if (l1
== &cset
->cgrp_links
) {
604 BUG_ON(l2
!= &old_cset
->cgrp_links
);
607 BUG_ON(l2
== &old_cset
->cgrp_links
);
609 /* Locate the cgroups associated with these links. */
610 link1
= list_entry(l1
, struct cgrp_cset_link
, cgrp_link
);
611 link2
= list_entry(l2
, struct cgrp_cset_link
, cgrp_link
);
614 /* Hierarchies should be linked in the same order. */
615 BUG_ON(cgrp1
->root
!= cgrp2
->root
);
618 * If this hierarchy is the hierarchy of the cgroup
619 * that's changing, then we need to check that this
620 * css_set points to the new cgroup; if it's any other
621 * hierarchy, then this css_set should point to the
622 * same cgroup as the old css_set.
624 if (cgrp1
->root
== new_cgrp
->root
) {
625 if (cgrp1
!= new_cgrp
)
636 * find_existing_css_set - init css array and find the matching css_set
637 * @old_cset: the css_set that we're using before the cgroup transition
638 * @cgrp: the cgroup that we're moving into
639 * @template: out param for the new set of csses, should be clear on entry
641 static struct css_set
*find_existing_css_set(struct css_set
*old_cset
,
643 struct cgroup_subsys_state
*template[])
645 struct cgroup_root
*root
= cgrp
->root
;
646 struct cgroup_subsys
*ss
;
647 struct css_set
*cset
;
652 * Build the set of subsystem state objects that we want to see in the
653 * new css_set. while subsystems can change globally, the entries here
654 * won't change, so no need for locking.
656 for_each_subsys(ss
, i
) {
657 if (root
->subsys_mask
& (1UL << i
)) {
659 * @ss is in this hierarchy, so we want the
660 * effective css from @cgrp.
662 template[i
] = cgroup_e_css(cgrp
, ss
);
665 * @ss is not in this hierarchy, so we don't want
668 template[i
] = old_cset
->subsys
[i
];
672 key
= css_set_hash(template);
673 hash_for_each_possible(css_set_table
, cset
, hlist
, key
) {
674 if (!compare_css_sets(cset
, old_cset
, cgrp
, template))
677 /* This css_set matches what we need */
681 /* No existing cgroup group matched */
685 static void free_cgrp_cset_links(struct list_head
*links_to_free
)
687 struct cgrp_cset_link
*link
, *tmp_link
;
689 list_for_each_entry_safe(link
, tmp_link
, links_to_free
, cset_link
) {
690 list_del(&link
->cset_link
);
696 * allocate_cgrp_cset_links - allocate cgrp_cset_links
697 * @count: the number of links to allocate
698 * @tmp_links: list_head the allocated links are put on
700 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
701 * through ->cset_link. Returns 0 on success or -errno.
703 static int allocate_cgrp_cset_links(int count
, struct list_head
*tmp_links
)
705 struct cgrp_cset_link
*link
;
708 INIT_LIST_HEAD(tmp_links
);
710 for (i
= 0; i
< count
; i
++) {
711 link
= kzalloc(sizeof(*link
), GFP_KERNEL
);
713 free_cgrp_cset_links(tmp_links
);
716 list_add(&link
->cset_link
, tmp_links
);
722 * link_css_set - a helper function to link a css_set to a cgroup
723 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
724 * @cset: the css_set to be linked
725 * @cgrp: the destination cgroup
727 static void link_css_set(struct list_head
*tmp_links
, struct css_set
*cset
,
730 struct cgrp_cset_link
*link
;
732 BUG_ON(list_empty(tmp_links
));
734 if (cgroup_on_dfl(cgrp
))
735 cset
->dfl_cgrp
= cgrp
;
737 link
= list_first_entry(tmp_links
, struct cgrp_cset_link
, cset_link
);
741 if (list_empty(&cgrp
->cset_links
))
742 cgroup_update_populated(cgrp
, true);
743 list_move(&link
->cset_link
, &cgrp
->cset_links
);
746 * Always add links to the tail of the list so that the list
747 * is sorted by order of hierarchy creation
749 list_add_tail(&link
->cgrp_link
, &cset
->cgrp_links
);
753 * find_css_set - return a new css_set with one cgroup updated
754 * @old_cset: the baseline css_set
755 * @cgrp: the cgroup to be updated
757 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
758 * substituted into the appropriate hierarchy.
760 static struct css_set
*find_css_set(struct css_set
*old_cset
,
763 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
] = { };
764 struct css_set
*cset
;
765 struct list_head tmp_links
;
766 struct cgrp_cset_link
*link
;
767 struct cgroup_subsys
*ss
;
771 lockdep_assert_held(&cgroup_mutex
);
773 /* First see if we already have a cgroup group that matches
775 down_read(&css_set_rwsem
);
776 cset
= find_existing_css_set(old_cset
, cgrp
, template);
779 up_read(&css_set_rwsem
);
784 cset
= kzalloc(sizeof(*cset
), GFP_KERNEL
);
788 /* Allocate all the cgrp_cset_link objects that we'll need */
789 if (allocate_cgrp_cset_links(cgroup_root_count
, &tmp_links
) < 0) {
794 atomic_set(&cset
->refcount
, 1);
795 INIT_LIST_HEAD(&cset
->cgrp_links
);
796 INIT_LIST_HEAD(&cset
->tasks
);
797 INIT_LIST_HEAD(&cset
->mg_tasks
);
798 INIT_LIST_HEAD(&cset
->mg_preload_node
);
799 INIT_LIST_HEAD(&cset
->mg_node
);
800 INIT_HLIST_NODE(&cset
->hlist
);
802 /* Copy the set of subsystem state objects generated in
803 * find_existing_css_set() */
804 memcpy(cset
->subsys
, template, sizeof(cset
->subsys
));
806 down_write(&css_set_rwsem
);
807 /* Add reference counts and links from the new css_set. */
808 list_for_each_entry(link
, &old_cset
->cgrp_links
, cgrp_link
) {
809 struct cgroup
*c
= link
->cgrp
;
811 if (c
->root
== cgrp
->root
)
813 link_css_set(&tmp_links
, cset
, c
);
816 BUG_ON(!list_empty(&tmp_links
));
820 /* Add @cset to the hash table */
821 key
= css_set_hash(cset
->subsys
);
822 hash_add(css_set_table
, &cset
->hlist
, key
);
824 for_each_subsys(ss
, ssid
)
825 list_add_tail(&cset
->e_cset_node
[ssid
],
826 &cset
->subsys
[ssid
]->cgroup
->e_csets
[ssid
]);
828 up_write(&css_set_rwsem
);
833 static struct cgroup_root
*cgroup_root_from_kf(struct kernfs_root
*kf_root
)
835 struct cgroup
*root_cgrp
= kf_root
->kn
->priv
;
837 return root_cgrp
->root
;
840 static int cgroup_init_root_id(struct cgroup_root
*root
)
844 lockdep_assert_held(&cgroup_mutex
);
846 id
= idr_alloc_cyclic(&cgroup_hierarchy_idr
, root
, 0, 0, GFP_KERNEL
);
850 root
->hierarchy_id
= id
;
854 static void cgroup_exit_root_id(struct cgroup_root
*root
)
856 lockdep_assert_held(&cgroup_mutex
);
858 if (root
->hierarchy_id
) {
859 idr_remove(&cgroup_hierarchy_idr
, root
->hierarchy_id
);
860 root
->hierarchy_id
= 0;
864 static void cgroup_free_root(struct cgroup_root
*root
)
867 /* hierarhcy ID shoulid already have been released */
868 WARN_ON_ONCE(root
->hierarchy_id
);
870 idr_destroy(&root
->cgroup_idr
);
875 static void cgroup_destroy_root(struct cgroup_root
*root
)
877 struct cgroup
*cgrp
= &root
->cgrp
;
878 struct cgrp_cset_link
*link
, *tmp_link
;
880 mutex_lock(&cgroup_mutex
);
882 BUG_ON(atomic_read(&root
->nr_cgrps
));
883 BUG_ON(!list_empty(&cgrp
->self
.children
));
885 /* Rebind all subsystems back to the default hierarchy */
886 rebind_subsystems(&cgrp_dfl_root
, root
->subsys_mask
);
889 * Release all the links from cset_links to this hierarchy's
892 down_write(&css_set_rwsem
);
894 list_for_each_entry_safe(link
, tmp_link
, &cgrp
->cset_links
, cset_link
) {
895 list_del(&link
->cset_link
);
896 list_del(&link
->cgrp_link
);
899 up_write(&css_set_rwsem
);
901 if (!list_empty(&root
->root_list
)) {
902 list_del(&root
->root_list
);
906 cgroup_exit_root_id(root
);
908 mutex_unlock(&cgroup_mutex
);
910 kernfs_destroy_root(root
->kf_root
);
911 cgroup_free_root(root
);
914 /* look up cgroup associated with given css_set on the specified hierarchy */
915 static struct cgroup
*cset_cgroup_from_root(struct css_set
*cset
,
916 struct cgroup_root
*root
)
918 struct cgroup
*res
= NULL
;
920 lockdep_assert_held(&cgroup_mutex
);
921 lockdep_assert_held(&css_set_rwsem
);
923 if (cset
== &init_css_set
) {
926 struct cgrp_cset_link
*link
;
928 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
929 struct cgroup
*c
= link
->cgrp
;
931 if (c
->root
== root
) {
943 * Return the cgroup for "task" from the given hierarchy. Must be
944 * called with cgroup_mutex and css_set_rwsem held.
946 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
947 struct cgroup_root
*root
)
950 * No need to lock the task - since we hold cgroup_mutex the
951 * task can't change groups, so the only thing that can happen
952 * is that it exits and its css is set back to init_css_set.
954 return cset_cgroup_from_root(task_css_set(task
), root
);
958 * A task must hold cgroup_mutex to modify cgroups.
960 * Any task can increment and decrement the count field without lock.
961 * So in general, code holding cgroup_mutex can't rely on the count
962 * field not changing. However, if the count goes to zero, then only
963 * cgroup_attach_task() can increment it again. Because a count of zero
964 * means that no tasks are currently attached, therefore there is no
965 * way a task attached to that cgroup can fork (the other way to
966 * increment the count). So code holding cgroup_mutex can safely
967 * assume that if the count is zero, it will stay zero. Similarly, if
968 * a task holds cgroup_mutex on a cgroup with zero count, it
969 * knows that the cgroup won't be removed, as cgroup_rmdir()
972 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
973 * (usually) take cgroup_mutex. These are the two most performance
974 * critical pieces of code here. The exception occurs on cgroup_exit(),
975 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
976 * is taken, and if the cgroup count is zero, a usermode call made
977 * to the release agent with the name of the cgroup (path relative to
978 * the root of cgroup file system) as the argument.
980 * A cgroup can only be deleted if both its 'count' of using tasks
981 * is zero, and its list of 'children' cgroups is empty. Since all
982 * tasks in the system use _some_ cgroup, and since there is always at
983 * least one task in the system (init, pid == 1), therefore, root cgroup
984 * always has either children cgroups and/or using tasks. So we don't
985 * need a special hack to ensure that root cgroup cannot be deleted.
987 * P.S. One more locking exception. RCU is used to guard the
988 * update of a tasks cgroup pointer by cgroup_attach_task()
991 static int cgroup_populate_dir(struct cgroup
*cgrp
, unsigned int subsys_mask
);
992 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
;
993 static const struct file_operations proc_cgroupstats_operations
;
995 static char *cgroup_file_name(struct cgroup
*cgrp
, const struct cftype
*cft
,
998 if (cft
->ss
&& !(cft
->flags
& CFTYPE_NO_PREFIX
) &&
999 !(cgrp
->root
->flags
& CGRP_ROOT_NOPREFIX
))
1000 snprintf(buf
, CGROUP_FILE_NAME_MAX
, "%s.%s",
1001 cft
->ss
->name
, cft
->name
);
1003 strncpy(buf
, cft
->name
, CGROUP_FILE_NAME_MAX
);
1008 * cgroup_file_mode - deduce file mode of a control file
1009 * @cft: the control file in question
1011 * returns cft->mode if ->mode is not 0
1012 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
1013 * returns S_IRUGO if it has only a read handler
1014 * returns S_IWUSR if it has only a write hander
1016 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
1023 if (cft
->read_u64
|| cft
->read_s64
|| cft
->seq_show
)
1026 if (cft
->write_u64
|| cft
->write_s64
|| cft
->write
)
1032 static void cgroup_get(struct cgroup
*cgrp
)
1034 WARN_ON_ONCE(cgroup_is_dead(cgrp
));
1035 css_get(&cgrp
->self
);
1038 static bool cgroup_tryget(struct cgroup
*cgrp
)
1040 return css_tryget(&cgrp
->self
);
1043 static void cgroup_put(struct cgroup
*cgrp
)
1045 css_put(&cgrp
->self
);
1049 * cgroup_refresh_child_subsys_mask - update child_subsys_mask
1050 * @cgrp: the target cgroup
1052 * On the default hierarchy, a subsystem may request other subsystems to be
1053 * enabled together through its ->depends_on mask. In such cases, more
1054 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1056 * This function determines which subsystems need to be enabled given the
1057 * current @cgrp->subtree_control and records it in
1058 * @cgrp->child_subsys_mask. The resulting mask is always a superset of
1059 * @cgrp->subtree_control and follows the usual hierarchy rules.
1061 static void cgroup_refresh_child_subsys_mask(struct cgroup
*cgrp
)
1063 struct cgroup
*parent
= cgroup_parent(cgrp
);
1064 unsigned int cur_ss_mask
= cgrp
->subtree_control
;
1065 struct cgroup_subsys
*ss
;
1068 lockdep_assert_held(&cgroup_mutex
);
1070 if (!cgroup_on_dfl(cgrp
)) {
1071 cgrp
->child_subsys_mask
= cur_ss_mask
;
1076 unsigned int new_ss_mask
= cur_ss_mask
;
1078 for_each_subsys(ss
, ssid
)
1079 if (cur_ss_mask
& (1 << ssid
))
1080 new_ss_mask
|= ss
->depends_on
;
1083 * Mask out subsystems which aren't available. This can
1084 * happen only if some depended-upon subsystems were bound
1085 * to non-default hierarchies.
1088 new_ss_mask
&= parent
->child_subsys_mask
;
1090 new_ss_mask
&= cgrp
->root
->subsys_mask
;
1092 if (new_ss_mask
== cur_ss_mask
)
1094 cur_ss_mask
= new_ss_mask
;
1097 cgrp
->child_subsys_mask
= cur_ss_mask
;
1101 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1102 * @kn: the kernfs_node being serviced
1104 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1105 * the method finishes if locking succeeded. Note that once this function
1106 * returns the cgroup returned by cgroup_kn_lock_live() may become
1107 * inaccessible any time. If the caller intends to continue to access the
1108 * cgroup, it should pin it before invoking this function.
1110 static void cgroup_kn_unlock(struct kernfs_node
*kn
)
1112 struct cgroup
*cgrp
;
1114 if (kernfs_type(kn
) == KERNFS_DIR
)
1117 cgrp
= kn
->parent
->priv
;
1119 mutex_unlock(&cgroup_mutex
);
1121 kernfs_unbreak_active_protection(kn
);
1126 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1127 * @kn: the kernfs_node being serviced
1129 * This helper is to be used by a cgroup kernfs method currently servicing
1130 * @kn. It breaks the active protection, performs cgroup locking and
1131 * verifies that the associated cgroup is alive. Returns the cgroup if
1132 * alive; otherwise, %NULL. A successful return should be undone by a
1133 * matching cgroup_kn_unlock() invocation.
1135 * Any cgroup kernfs method implementation which requires locking the
1136 * associated cgroup should use this helper. It avoids nesting cgroup
1137 * locking under kernfs active protection and allows all kernfs operations
1138 * including self-removal.
1140 static struct cgroup
*cgroup_kn_lock_live(struct kernfs_node
*kn
)
1142 struct cgroup
*cgrp
;
1144 if (kernfs_type(kn
) == KERNFS_DIR
)
1147 cgrp
= kn
->parent
->priv
;
1150 * We're gonna grab cgroup_mutex which nests outside kernfs
1151 * active_ref. cgroup liveliness check alone provides enough
1152 * protection against removal. Ensure @cgrp stays accessible and
1153 * break the active_ref protection.
1155 if (!cgroup_tryget(cgrp
))
1157 kernfs_break_active_protection(kn
);
1159 mutex_lock(&cgroup_mutex
);
1161 if (!cgroup_is_dead(cgrp
))
1164 cgroup_kn_unlock(kn
);
1168 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
1170 char name
[CGROUP_FILE_NAME_MAX
];
1172 lockdep_assert_held(&cgroup_mutex
);
1173 kernfs_remove_by_name(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
));
1177 * cgroup_clear_dir - remove subsys files in a cgroup directory
1178 * @cgrp: target cgroup
1179 * @subsys_mask: mask of the subsystem ids whose files should be removed
1181 static void cgroup_clear_dir(struct cgroup
*cgrp
, unsigned int subsys_mask
)
1183 struct cgroup_subsys
*ss
;
1186 for_each_subsys(ss
, i
) {
1187 struct cftype
*cfts
;
1189 if (!(subsys_mask
& (1 << i
)))
1191 list_for_each_entry(cfts
, &ss
->cfts
, node
)
1192 cgroup_addrm_files(cgrp
, cfts
, false);
1196 static int rebind_subsystems(struct cgroup_root
*dst_root
, unsigned int ss_mask
)
1198 struct cgroup_subsys
*ss
;
1199 unsigned int tmp_ss_mask
;
1202 lockdep_assert_held(&cgroup_mutex
);
1204 for_each_subsys(ss
, ssid
) {
1205 if (!(ss_mask
& (1 << ssid
)))
1208 /* if @ss has non-root csses attached to it, can't move */
1209 if (css_next_child(NULL
, cgroup_css(&ss
->root
->cgrp
, ss
)))
1212 /* can't move between two non-dummy roots either */
1213 if (ss
->root
!= &cgrp_dfl_root
&& dst_root
!= &cgrp_dfl_root
)
1217 /* skip creating root files on dfl_root for inhibited subsystems */
1218 tmp_ss_mask
= ss_mask
;
1219 if (dst_root
== &cgrp_dfl_root
)
1220 tmp_ss_mask
&= ~cgrp_dfl_root_inhibit_ss_mask
;
1222 ret
= cgroup_populate_dir(&dst_root
->cgrp
, tmp_ss_mask
);
1224 if (dst_root
!= &cgrp_dfl_root
)
1228 * Rebinding back to the default root is not allowed to
1229 * fail. Using both default and non-default roots should
1230 * be rare. Moving subsystems back and forth even more so.
1231 * Just warn about it and continue.
1233 if (cgrp_dfl_root_visible
) {
1234 pr_warn("failed to create files (%d) while rebinding 0x%x to default root\n",
1236 pr_warn("you may retry by moving them to a different hierarchy and unbinding\n");
1241 * Nothing can fail from this point on. Remove files for the
1242 * removed subsystems and rebind each subsystem.
1244 for_each_subsys(ss
, ssid
)
1245 if (ss_mask
& (1 << ssid
))
1246 cgroup_clear_dir(&ss
->root
->cgrp
, 1 << ssid
);
1248 for_each_subsys(ss
, ssid
) {
1249 struct cgroup_root
*src_root
;
1250 struct cgroup_subsys_state
*css
;
1251 struct css_set
*cset
;
1253 if (!(ss_mask
& (1 << ssid
)))
1256 src_root
= ss
->root
;
1257 css
= cgroup_css(&src_root
->cgrp
, ss
);
1259 WARN_ON(!css
|| cgroup_css(&dst_root
->cgrp
, ss
));
1261 RCU_INIT_POINTER(src_root
->cgrp
.subsys
[ssid
], NULL
);
1262 rcu_assign_pointer(dst_root
->cgrp
.subsys
[ssid
], css
);
1263 ss
->root
= dst_root
;
1264 css
->cgroup
= &dst_root
->cgrp
;
1266 down_write(&css_set_rwsem
);
1267 hash_for_each(css_set_table
, i
, cset
, hlist
)
1268 list_move_tail(&cset
->e_cset_node
[ss
->id
],
1269 &dst_root
->cgrp
.e_csets
[ss
->id
]);
1270 up_write(&css_set_rwsem
);
1272 src_root
->subsys_mask
&= ~(1 << ssid
);
1273 src_root
->cgrp
.subtree_control
&= ~(1 << ssid
);
1274 cgroup_refresh_child_subsys_mask(&src_root
->cgrp
);
1276 /* default hierarchy doesn't enable controllers by default */
1277 dst_root
->subsys_mask
|= 1 << ssid
;
1278 if (dst_root
!= &cgrp_dfl_root
) {
1279 dst_root
->cgrp
.subtree_control
|= 1 << ssid
;
1280 cgroup_refresh_child_subsys_mask(&dst_root
->cgrp
);
1287 kernfs_activate(dst_root
->cgrp
.kn
);
1291 static int cgroup_show_options(struct seq_file
*seq
,
1292 struct kernfs_root
*kf_root
)
1294 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1295 struct cgroup_subsys
*ss
;
1298 for_each_subsys(ss
, ssid
)
1299 if (root
->subsys_mask
& (1 << ssid
))
1300 seq_printf(seq
, ",%s", ss
->name
);
1301 if (root
->flags
& CGRP_ROOT_NOPREFIX
)
1302 seq_puts(seq
, ",noprefix");
1303 if (root
->flags
& CGRP_ROOT_XATTR
)
1304 seq_puts(seq
, ",xattr");
1306 spin_lock(&release_agent_path_lock
);
1307 if (strlen(root
->release_agent_path
))
1308 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1309 spin_unlock(&release_agent_path_lock
);
1311 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
))
1312 seq_puts(seq
, ",clone_children");
1313 if (strlen(root
->name
))
1314 seq_printf(seq
, ",name=%s", root
->name
);
1318 struct cgroup_sb_opts
{
1319 unsigned int subsys_mask
;
1321 char *release_agent
;
1322 bool cpuset_clone_children
;
1324 /* User explicitly requested empty subsystem */
1328 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1330 char *token
, *o
= data
;
1331 bool all_ss
= false, one_ss
= false;
1332 unsigned int mask
= -1U;
1333 struct cgroup_subsys
*ss
;
1337 #ifdef CONFIG_CPUSETS
1338 mask
= ~(1U << cpuset_cgrp_id
);
1341 memset(opts
, 0, sizeof(*opts
));
1343 while ((token
= strsep(&o
, ",")) != NULL
) {
1348 if (!strcmp(token
, "none")) {
1349 /* Explicitly have no subsystems */
1353 if (!strcmp(token
, "all")) {
1354 /* Mutually exclusive option 'all' + subsystem name */
1360 if (!strcmp(token
, "__DEVEL__sane_behavior")) {
1361 opts
->flags
|= CGRP_ROOT_SANE_BEHAVIOR
;
1364 if (!strcmp(token
, "noprefix")) {
1365 opts
->flags
|= CGRP_ROOT_NOPREFIX
;
1368 if (!strcmp(token
, "clone_children")) {
1369 opts
->cpuset_clone_children
= true;
1372 if (!strcmp(token
, "xattr")) {
1373 opts
->flags
|= CGRP_ROOT_XATTR
;
1376 if (!strncmp(token
, "release_agent=", 14)) {
1377 /* Specifying two release agents is forbidden */
1378 if (opts
->release_agent
)
1380 opts
->release_agent
=
1381 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1382 if (!opts
->release_agent
)
1386 if (!strncmp(token
, "name=", 5)) {
1387 const char *name
= token
+ 5;
1388 /* Can't specify an empty name */
1391 /* Must match [\w.-]+ */
1392 for (i
= 0; i
< strlen(name
); i
++) {
1396 if ((c
== '.') || (c
== '-') || (c
== '_'))
1400 /* Specifying two names is forbidden */
1403 opts
->name
= kstrndup(name
,
1404 MAX_CGROUP_ROOT_NAMELEN
- 1,
1412 for_each_subsys(ss
, i
) {
1413 if (strcmp(token
, ss
->name
))
1418 /* Mutually exclusive option 'all' + subsystem name */
1421 opts
->subsys_mask
|= (1 << i
);
1426 if (i
== CGROUP_SUBSYS_COUNT
)
1430 if (opts
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1431 pr_warn("sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1433 pr_err("sane_behavior: no other mount options allowed\n");
1440 * If the 'all' option was specified select all the subsystems,
1441 * otherwise if 'none', 'name=' and a subsystem name options were
1442 * not specified, let's default to 'all'
1444 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
))
1445 for_each_subsys(ss
, i
)
1447 opts
->subsys_mask
|= (1 << i
);
1450 * We either have to specify by name or by subsystems. (So all
1451 * empty hierarchies must have a name).
1453 if (!opts
->subsys_mask
&& !opts
->name
)
1457 * Option noprefix was introduced just for backward compatibility
1458 * with the old cpuset, so we allow noprefix only if mounting just
1459 * the cpuset subsystem.
1461 if ((opts
->flags
& CGRP_ROOT_NOPREFIX
) && (opts
->subsys_mask
& mask
))
1464 /* Can't specify "none" and some subsystems */
1465 if (opts
->subsys_mask
&& opts
->none
)
1471 static int cgroup_remount(struct kernfs_root
*kf_root
, int *flags
, char *data
)
1474 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1475 struct cgroup_sb_opts opts
;
1476 unsigned int added_mask
, removed_mask
;
1478 if (root
== &cgrp_dfl_root
) {
1479 pr_err("remount is not allowed\n");
1483 mutex_lock(&cgroup_mutex
);
1485 /* See what subsystems are wanted */
1486 ret
= parse_cgroupfs_options(data
, &opts
);
1490 if (opts
.subsys_mask
!= root
->subsys_mask
|| opts
.release_agent
)
1491 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1492 task_tgid_nr(current
), current
->comm
);
1494 added_mask
= opts
.subsys_mask
& ~root
->subsys_mask
;
1495 removed_mask
= root
->subsys_mask
& ~opts
.subsys_mask
;
1497 /* Don't allow flags or name to change at remount */
1498 if ((opts
.flags
^ root
->flags
) ||
1499 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1500 pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n",
1501 opts
.flags
, opts
.name
?: "", root
->flags
, root
->name
);
1506 /* remounting is not allowed for populated hierarchies */
1507 if (!list_empty(&root
->cgrp
.self
.children
)) {
1512 ret
= rebind_subsystems(root
, added_mask
);
1516 rebind_subsystems(&cgrp_dfl_root
, removed_mask
);
1518 if (opts
.release_agent
) {
1519 spin_lock(&release_agent_path_lock
);
1520 strcpy(root
->release_agent_path
, opts
.release_agent
);
1521 spin_unlock(&release_agent_path_lock
);
1524 kfree(opts
.release_agent
);
1526 mutex_unlock(&cgroup_mutex
);
1531 * To reduce the fork() overhead for systems that are not actually using
1532 * their cgroups capability, we don't maintain the lists running through
1533 * each css_set to its tasks until we see the list actually used - in other
1534 * words after the first mount.
1536 static bool use_task_css_set_links __read_mostly
;
1538 static void cgroup_enable_task_cg_lists(void)
1540 struct task_struct
*p
, *g
;
1542 down_write(&css_set_rwsem
);
1544 if (use_task_css_set_links
)
1547 use_task_css_set_links
= true;
1550 * We need tasklist_lock because RCU is not safe against
1551 * while_each_thread(). Besides, a forking task that has passed
1552 * cgroup_post_fork() without seeing use_task_css_set_links = 1
1553 * is not guaranteed to have its child immediately visible in the
1554 * tasklist if we walk through it with RCU.
1556 read_lock(&tasklist_lock
);
1557 do_each_thread(g
, p
) {
1558 WARN_ON_ONCE(!list_empty(&p
->cg_list
) ||
1559 task_css_set(p
) != &init_css_set
);
1562 * We should check if the process is exiting, otherwise
1563 * it will race with cgroup_exit() in that the list
1564 * entry won't be deleted though the process has exited.
1565 * Do it while holding siglock so that we don't end up
1566 * racing against cgroup_exit().
1568 spin_lock_irq(&p
->sighand
->siglock
);
1569 if (!(p
->flags
& PF_EXITING
)) {
1570 struct css_set
*cset
= task_css_set(p
);
1572 list_add(&p
->cg_list
, &cset
->tasks
);
1575 spin_unlock_irq(&p
->sighand
->siglock
);
1576 } while_each_thread(g
, p
);
1577 read_unlock(&tasklist_lock
);
1579 up_write(&css_set_rwsem
);
1582 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1584 struct cgroup_subsys
*ss
;
1587 INIT_LIST_HEAD(&cgrp
->self
.sibling
);
1588 INIT_LIST_HEAD(&cgrp
->self
.children
);
1589 INIT_LIST_HEAD(&cgrp
->cset_links
);
1590 INIT_LIST_HEAD(&cgrp
->release_list
);
1591 INIT_LIST_HEAD(&cgrp
->pidlists
);
1592 mutex_init(&cgrp
->pidlist_mutex
);
1593 cgrp
->self
.cgroup
= cgrp
;
1594 cgrp
->self
.flags
|= CSS_ONLINE
;
1596 for_each_subsys(ss
, ssid
)
1597 INIT_LIST_HEAD(&cgrp
->e_csets
[ssid
]);
1599 init_waitqueue_head(&cgrp
->offline_waitq
);
1602 static void init_cgroup_root(struct cgroup_root
*root
,
1603 struct cgroup_sb_opts
*opts
)
1605 struct cgroup
*cgrp
= &root
->cgrp
;
1607 INIT_LIST_HEAD(&root
->root_list
);
1608 atomic_set(&root
->nr_cgrps
, 1);
1610 init_cgroup_housekeeping(cgrp
);
1611 idr_init(&root
->cgroup_idr
);
1613 root
->flags
= opts
->flags
;
1614 if (opts
->release_agent
)
1615 strcpy(root
->release_agent_path
, opts
->release_agent
);
1617 strcpy(root
->name
, opts
->name
);
1618 if (opts
->cpuset_clone_children
)
1619 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
);
1622 static int cgroup_setup_root(struct cgroup_root
*root
, unsigned int ss_mask
)
1624 LIST_HEAD(tmp_links
);
1625 struct cgroup
*root_cgrp
= &root
->cgrp
;
1626 struct cftype
*base_files
;
1627 struct css_set
*cset
;
1630 lockdep_assert_held(&cgroup_mutex
);
1632 ret
= cgroup_idr_alloc(&root
->cgroup_idr
, root_cgrp
, 1, 2, GFP_NOWAIT
);
1635 root_cgrp
->id
= ret
;
1637 ret
= percpu_ref_init(&root_cgrp
->self
.refcnt
, css_release
);
1642 * We're accessing css_set_count without locking css_set_rwsem here,
1643 * but that's OK - it can only be increased by someone holding
1644 * cgroup_lock, and that's us. The worst that can happen is that we
1645 * have some link structures left over
1647 ret
= allocate_cgrp_cset_links(css_set_count
, &tmp_links
);
1651 ret
= cgroup_init_root_id(root
);
1655 root
->kf_root
= kernfs_create_root(&cgroup_kf_syscall_ops
,
1656 KERNFS_ROOT_CREATE_DEACTIVATED
,
1658 if (IS_ERR(root
->kf_root
)) {
1659 ret
= PTR_ERR(root
->kf_root
);
1662 root_cgrp
->kn
= root
->kf_root
->kn
;
1664 if (root
== &cgrp_dfl_root
)
1665 base_files
= cgroup_dfl_base_files
;
1667 base_files
= cgroup_legacy_base_files
;
1669 ret
= cgroup_addrm_files(root_cgrp
, base_files
, true);
1673 ret
= rebind_subsystems(root
, ss_mask
);
1678 * There must be no failure case after here, since rebinding takes
1679 * care of subsystems' refcounts, which are explicitly dropped in
1680 * the failure exit path.
1682 list_add(&root
->root_list
, &cgroup_roots
);
1683 cgroup_root_count
++;
1686 * Link the root cgroup in this hierarchy into all the css_set
1689 down_write(&css_set_rwsem
);
1690 hash_for_each(css_set_table
, i
, cset
, hlist
)
1691 link_css_set(&tmp_links
, cset
, root_cgrp
);
1692 up_write(&css_set_rwsem
);
1694 BUG_ON(!list_empty(&root_cgrp
->self
.children
));
1695 BUG_ON(atomic_read(&root
->nr_cgrps
) != 1);
1697 kernfs_activate(root_cgrp
->kn
);
1702 kernfs_destroy_root(root
->kf_root
);
1703 root
->kf_root
= NULL
;
1705 cgroup_exit_root_id(root
);
1707 percpu_ref_exit(&root_cgrp
->self
.refcnt
);
1709 free_cgrp_cset_links(&tmp_links
);
1713 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1714 int flags
, const char *unused_dev_name
,
1717 struct super_block
*pinned_sb
= NULL
;
1718 struct cgroup_subsys
*ss
;
1719 struct cgroup_root
*root
;
1720 struct cgroup_sb_opts opts
;
1721 struct dentry
*dentry
;
1727 * The first time anyone tries to mount a cgroup, enable the list
1728 * linking each css_set to its tasks and fix up all existing tasks.
1730 if (!use_task_css_set_links
)
1731 cgroup_enable_task_cg_lists();
1733 mutex_lock(&cgroup_mutex
);
1735 /* First find the desired set of subsystems */
1736 ret
= parse_cgroupfs_options(data
, &opts
);
1740 /* look for a matching existing root */
1741 if (opts
.flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1742 cgrp_dfl_root_visible
= true;
1743 root
= &cgrp_dfl_root
;
1744 cgroup_get(&root
->cgrp
);
1750 * Destruction of cgroup root is asynchronous, so subsystems may
1751 * still be dying after the previous unmount. Let's drain the
1752 * dying subsystems. We just need to ensure that the ones
1753 * unmounted previously finish dying and don't care about new ones
1754 * starting. Testing ref liveliness is good enough.
1756 for_each_subsys(ss
, i
) {
1757 if (!(opts
.subsys_mask
& (1 << i
)) ||
1758 ss
->root
== &cgrp_dfl_root
)
1761 if (!percpu_ref_tryget_live(&ss
->root
->cgrp
.self
.refcnt
)) {
1762 mutex_unlock(&cgroup_mutex
);
1764 ret
= restart_syscall();
1767 cgroup_put(&ss
->root
->cgrp
);
1770 for_each_root(root
) {
1771 bool name_match
= false;
1773 if (root
== &cgrp_dfl_root
)
1777 * If we asked for a name then it must match. Also, if
1778 * name matches but sybsys_mask doesn't, we should fail.
1779 * Remember whether name matched.
1782 if (strcmp(opts
.name
, root
->name
))
1788 * If we asked for subsystems (or explicitly for no
1789 * subsystems) then they must match.
1791 if ((opts
.subsys_mask
|| opts
.none
) &&
1792 (opts
.subsys_mask
!= root
->subsys_mask
)) {
1799 if (root
->flags
^ opts
.flags
)
1800 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
1803 * We want to reuse @root whose lifetime is governed by its
1804 * ->cgrp. Let's check whether @root is alive and keep it
1805 * that way. As cgroup_kill_sb() can happen anytime, we
1806 * want to block it by pinning the sb so that @root doesn't
1807 * get killed before mount is complete.
1809 * With the sb pinned, tryget_live can reliably indicate
1810 * whether @root can be reused. If it's being killed,
1811 * drain it. We can use wait_queue for the wait but this
1812 * path is super cold. Let's just sleep a bit and retry.
1814 pinned_sb
= kernfs_pin_sb(root
->kf_root
, NULL
);
1815 if (IS_ERR(pinned_sb
) ||
1816 !percpu_ref_tryget_live(&root
->cgrp
.self
.refcnt
)) {
1817 mutex_unlock(&cgroup_mutex
);
1818 if (!IS_ERR_OR_NULL(pinned_sb
))
1819 deactivate_super(pinned_sb
);
1821 ret
= restart_syscall();
1830 * No such thing, create a new one. name= matching without subsys
1831 * specification is allowed for already existing hierarchies but we
1832 * can't create new one without subsys specification.
1834 if (!opts
.subsys_mask
&& !opts
.none
) {
1839 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1845 init_cgroup_root(root
, &opts
);
1847 ret
= cgroup_setup_root(root
, opts
.subsys_mask
);
1849 cgroup_free_root(root
);
1852 mutex_unlock(&cgroup_mutex
);
1854 kfree(opts
.release_agent
);
1858 return ERR_PTR(ret
);
1860 dentry
= kernfs_mount(fs_type
, flags
, root
->kf_root
,
1861 CGROUP_SUPER_MAGIC
, &new_sb
);
1862 if (IS_ERR(dentry
) || !new_sb
)
1863 cgroup_put(&root
->cgrp
);
1866 * If @pinned_sb, we're reusing an existing root and holding an
1867 * extra ref on its sb. Mount is complete. Put the extra ref.
1871 deactivate_super(pinned_sb
);
1877 static void cgroup_kill_sb(struct super_block
*sb
)
1879 struct kernfs_root
*kf_root
= kernfs_root_from_sb(sb
);
1880 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1883 * If @root doesn't have any mounts or children, start killing it.
1884 * This prevents new mounts by disabling percpu_ref_tryget_live().
1885 * cgroup_mount() may wait for @root's release.
1887 * And don't kill the default root.
1889 if (css_has_online_children(&root
->cgrp
.self
) ||
1890 root
== &cgrp_dfl_root
)
1891 cgroup_put(&root
->cgrp
);
1893 percpu_ref_kill(&root
->cgrp
.self
.refcnt
);
1898 static struct file_system_type cgroup_fs_type
= {
1900 .mount
= cgroup_mount
,
1901 .kill_sb
= cgroup_kill_sb
,
1904 static struct kobject
*cgroup_kobj
;
1907 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
1908 * @task: target task
1909 * @buf: the buffer to write the path into
1910 * @buflen: the length of the buffer
1912 * Determine @task's cgroup on the first (the one with the lowest non-zero
1913 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
1914 * function grabs cgroup_mutex and shouldn't be used inside locks used by
1915 * cgroup controller callbacks.
1917 * Return value is the same as kernfs_path().
1919 char *task_cgroup_path(struct task_struct
*task
, char *buf
, size_t buflen
)
1921 struct cgroup_root
*root
;
1922 struct cgroup
*cgrp
;
1923 int hierarchy_id
= 1;
1926 mutex_lock(&cgroup_mutex
);
1927 down_read(&css_set_rwsem
);
1929 root
= idr_get_next(&cgroup_hierarchy_idr
, &hierarchy_id
);
1932 cgrp
= task_cgroup_from_root(task
, root
);
1933 path
= cgroup_path(cgrp
, buf
, buflen
);
1935 /* if no hierarchy exists, everyone is in "/" */
1936 if (strlcpy(buf
, "/", buflen
) < buflen
)
1940 up_read(&css_set_rwsem
);
1941 mutex_unlock(&cgroup_mutex
);
1944 EXPORT_SYMBOL_GPL(task_cgroup_path
);
1946 /* used to track tasks and other necessary states during migration */
1947 struct cgroup_taskset
{
1948 /* the src and dst cset list running through cset->mg_node */
1949 struct list_head src_csets
;
1950 struct list_head dst_csets
;
1953 * Fields for cgroup_taskset_*() iteration.
1955 * Before migration is committed, the target migration tasks are on
1956 * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of
1957 * the csets on ->dst_csets. ->csets point to either ->src_csets
1958 * or ->dst_csets depending on whether migration is committed.
1960 * ->cur_csets and ->cur_task point to the current task position
1963 struct list_head
*csets
;
1964 struct css_set
*cur_cset
;
1965 struct task_struct
*cur_task
;
1969 * cgroup_taskset_first - reset taskset and return the first task
1970 * @tset: taskset of interest
1972 * @tset iteration is initialized and the first task is returned.
1974 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
)
1976 tset
->cur_cset
= list_first_entry(tset
->csets
, struct css_set
, mg_node
);
1977 tset
->cur_task
= NULL
;
1979 return cgroup_taskset_next(tset
);
1983 * cgroup_taskset_next - iterate to the next task in taskset
1984 * @tset: taskset of interest
1986 * Return the next task in @tset. Iteration must have been initialized
1987 * with cgroup_taskset_first().
1989 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
)
1991 struct css_set
*cset
= tset
->cur_cset
;
1992 struct task_struct
*task
= tset
->cur_task
;
1994 while (&cset
->mg_node
!= tset
->csets
) {
1996 task
= list_first_entry(&cset
->mg_tasks
,
1997 struct task_struct
, cg_list
);
1999 task
= list_next_entry(task
, cg_list
);
2001 if (&task
->cg_list
!= &cset
->mg_tasks
) {
2002 tset
->cur_cset
= cset
;
2003 tset
->cur_task
= task
;
2007 cset
= list_next_entry(cset
, mg_node
);
2015 * cgroup_task_migrate - move a task from one cgroup to another.
2016 * @old_cgrp: the cgroup @tsk is being migrated from
2017 * @tsk: the task being migrated
2018 * @new_cset: the new css_set @tsk is being attached to
2020 * Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked.
2022 static void cgroup_task_migrate(struct cgroup
*old_cgrp
,
2023 struct task_struct
*tsk
,
2024 struct css_set
*new_cset
)
2026 struct css_set
*old_cset
;
2028 lockdep_assert_held(&cgroup_mutex
);
2029 lockdep_assert_held(&css_set_rwsem
);
2032 * We are synchronized through threadgroup_lock() against PF_EXITING
2033 * setting such that we can't race against cgroup_exit() changing the
2034 * css_set to init_css_set and dropping the old one.
2036 WARN_ON_ONCE(tsk
->flags
& PF_EXITING
);
2037 old_cset
= task_css_set(tsk
);
2039 get_css_set(new_cset
);
2040 rcu_assign_pointer(tsk
->cgroups
, new_cset
);
2043 * Use move_tail so that cgroup_taskset_first() still returns the
2044 * leader after migration. This works because cgroup_migrate()
2045 * ensures that the dst_cset of the leader is the first on the
2046 * tset's dst_csets list.
2048 list_move_tail(&tsk
->cg_list
, &new_cset
->mg_tasks
);
2051 * We just gained a reference on old_cset by taking it from the
2052 * task. As trading it for new_cset is protected by cgroup_mutex,
2053 * we're safe to drop it here; it will be freed under RCU.
2055 set_bit(CGRP_RELEASABLE
, &old_cgrp
->flags
);
2056 put_css_set_locked(old_cset
, false);
2060 * cgroup_migrate_finish - cleanup after attach
2061 * @preloaded_csets: list of preloaded css_sets
2063 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2064 * those functions for details.
2066 static void cgroup_migrate_finish(struct list_head
*preloaded_csets
)
2068 struct css_set
*cset
, *tmp_cset
;
2070 lockdep_assert_held(&cgroup_mutex
);
2072 down_write(&css_set_rwsem
);
2073 list_for_each_entry_safe(cset
, tmp_cset
, preloaded_csets
, mg_preload_node
) {
2074 cset
->mg_src_cgrp
= NULL
;
2075 cset
->mg_dst_cset
= NULL
;
2076 list_del_init(&cset
->mg_preload_node
);
2077 put_css_set_locked(cset
, false);
2079 up_write(&css_set_rwsem
);
2083 * cgroup_migrate_add_src - add a migration source css_set
2084 * @src_cset: the source css_set to add
2085 * @dst_cgrp: the destination cgroup
2086 * @preloaded_csets: list of preloaded css_sets
2088 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2089 * @src_cset and add it to @preloaded_csets, which should later be cleaned
2090 * up by cgroup_migrate_finish().
2092 * This function may be called without holding threadgroup_lock even if the
2093 * target is a process. Threads may be created and destroyed but as long
2094 * as cgroup_mutex is not dropped, no new css_set can be put into play and
2095 * the preloaded css_sets are guaranteed to cover all migrations.
2097 static void cgroup_migrate_add_src(struct css_set
*src_cset
,
2098 struct cgroup
*dst_cgrp
,
2099 struct list_head
*preloaded_csets
)
2101 struct cgroup
*src_cgrp
;
2103 lockdep_assert_held(&cgroup_mutex
);
2104 lockdep_assert_held(&css_set_rwsem
);
2106 src_cgrp
= cset_cgroup_from_root(src_cset
, dst_cgrp
->root
);
2108 if (!list_empty(&src_cset
->mg_preload_node
))
2111 WARN_ON(src_cset
->mg_src_cgrp
);
2112 WARN_ON(!list_empty(&src_cset
->mg_tasks
));
2113 WARN_ON(!list_empty(&src_cset
->mg_node
));
2115 src_cset
->mg_src_cgrp
= src_cgrp
;
2116 get_css_set(src_cset
);
2117 list_add(&src_cset
->mg_preload_node
, preloaded_csets
);
2121 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2122 * @dst_cgrp: the destination cgroup (may be %NULL)
2123 * @preloaded_csets: list of preloaded source css_sets
2125 * Tasks are about to be moved to @dst_cgrp and all the source css_sets
2126 * have been preloaded to @preloaded_csets. This function looks up and
2127 * pins all destination css_sets, links each to its source, and append them
2128 * to @preloaded_csets. If @dst_cgrp is %NULL, the destination of each
2129 * source css_set is assumed to be its cgroup on the default hierarchy.
2131 * This function must be called after cgroup_migrate_add_src() has been
2132 * called on each migration source css_set. After migration is performed
2133 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2136 static int cgroup_migrate_prepare_dst(struct cgroup
*dst_cgrp
,
2137 struct list_head
*preloaded_csets
)
2140 struct css_set
*src_cset
, *tmp_cset
;
2142 lockdep_assert_held(&cgroup_mutex
);
2145 * Except for the root, child_subsys_mask must be zero for a cgroup
2146 * with tasks so that child cgroups don't compete against tasks.
2148 if (dst_cgrp
&& cgroup_on_dfl(dst_cgrp
) && cgroup_parent(dst_cgrp
) &&
2149 dst_cgrp
->child_subsys_mask
)
2152 /* look up the dst cset for each src cset and link it to src */
2153 list_for_each_entry_safe(src_cset
, tmp_cset
, preloaded_csets
, mg_preload_node
) {
2154 struct css_set
*dst_cset
;
2156 dst_cset
= find_css_set(src_cset
,
2157 dst_cgrp
?: src_cset
->dfl_cgrp
);
2161 WARN_ON_ONCE(src_cset
->mg_dst_cset
|| dst_cset
->mg_dst_cset
);
2164 * If src cset equals dst, it's noop. Drop the src.
2165 * cgroup_migrate() will skip the cset too. Note that we
2166 * can't handle src == dst as some nodes are used by both.
2168 if (src_cset
== dst_cset
) {
2169 src_cset
->mg_src_cgrp
= NULL
;
2170 list_del_init(&src_cset
->mg_preload_node
);
2171 put_css_set(src_cset
, false);
2172 put_css_set(dst_cset
, false);
2176 src_cset
->mg_dst_cset
= dst_cset
;
2178 if (list_empty(&dst_cset
->mg_preload_node
))
2179 list_add(&dst_cset
->mg_preload_node
, &csets
);
2181 put_css_set(dst_cset
, false);
2184 list_splice_tail(&csets
, preloaded_csets
);
2187 cgroup_migrate_finish(&csets
);
2192 * cgroup_migrate - migrate a process or task to a cgroup
2193 * @cgrp: the destination cgroup
2194 * @leader: the leader of the process or the task to migrate
2195 * @threadgroup: whether @leader points to the whole process or a single task
2197 * Migrate a process or task denoted by @leader to @cgrp. If migrating a
2198 * process, the caller must be holding threadgroup_lock of @leader. The
2199 * caller is also responsible for invoking cgroup_migrate_add_src() and
2200 * cgroup_migrate_prepare_dst() on the targets before invoking this
2201 * function and following up with cgroup_migrate_finish().
2203 * As long as a controller's ->can_attach() doesn't fail, this function is
2204 * guaranteed to succeed. This means that, excluding ->can_attach()
2205 * failure, when migrating multiple targets, the success or failure can be
2206 * decided for all targets by invoking group_migrate_prepare_dst() before
2207 * actually starting migrating.
2209 static int cgroup_migrate(struct cgroup
*cgrp
, struct task_struct
*leader
,
2212 struct cgroup_taskset tset
= {
2213 .src_csets
= LIST_HEAD_INIT(tset
.src_csets
),
2214 .dst_csets
= LIST_HEAD_INIT(tset
.dst_csets
),
2215 .csets
= &tset
.src_csets
,
2217 struct cgroup_subsys_state
*css
, *failed_css
= NULL
;
2218 struct css_set
*cset
, *tmp_cset
;
2219 struct task_struct
*task
, *tmp_task
;
2223 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2224 * already PF_EXITING could be freed from underneath us unless we
2225 * take an rcu_read_lock.
2227 down_write(&css_set_rwsem
);
2231 /* @task either already exited or can't exit until the end */
2232 if (task
->flags
& PF_EXITING
)
2235 /* leave @task alone if post_fork() hasn't linked it yet */
2236 if (list_empty(&task
->cg_list
))
2239 cset
= task_css_set(task
);
2240 if (!cset
->mg_src_cgrp
)
2244 * cgroup_taskset_first() must always return the leader.
2245 * Take care to avoid disturbing the ordering.
2247 list_move_tail(&task
->cg_list
, &cset
->mg_tasks
);
2248 if (list_empty(&cset
->mg_node
))
2249 list_add_tail(&cset
->mg_node
, &tset
.src_csets
);
2250 if (list_empty(&cset
->mg_dst_cset
->mg_node
))
2251 list_move_tail(&cset
->mg_dst_cset
->mg_node
,
2256 } while_each_thread(leader
, task
);
2258 up_write(&css_set_rwsem
);
2260 /* methods shouldn't be called if no task is actually migrating */
2261 if (list_empty(&tset
.src_csets
))
2264 /* check that we can legitimately attach to the cgroup */
2265 for_each_e_css(css
, i
, cgrp
) {
2266 if (css
->ss
->can_attach
) {
2267 ret
= css
->ss
->can_attach(css
, &tset
);
2270 goto out_cancel_attach
;
2276 * Now that we're guaranteed success, proceed to move all tasks to
2277 * the new cgroup. There are no failure cases after here, so this
2278 * is the commit point.
2280 down_write(&css_set_rwsem
);
2281 list_for_each_entry(cset
, &tset
.src_csets
, mg_node
) {
2282 list_for_each_entry_safe(task
, tmp_task
, &cset
->mg_tasks
, cg_list
)
2283 cgroup_task_migrate(cset
->mg_src_cgrp
, task
,
2286 up_write(&css_set_rwsem
);
2289 * Migration is committed, all target tasks are now on dst_csets.
2290 * Nothing is sensitive to fork() after this point. Notify
2291 * controllers that migration is complete.
2293 tset
.csets
= &tset
.dst_csets
;
2295 for_each_e_css(css
, i
, cgrp
)
2296 if (css
->ss
->attach
)
2297 css
->ss
->attach(css
, &tset
);
2300 goto out_release_tset
;
2303 for_each_e_css(css
, i
, cgrp
) {
2304 if (css
== failed_css
)
2306 if (css
->ss
->cancel_attach
)
2307 css
->ss
->cancel_attach(css
, &tset
);
2310 down_write(&css_set_rwsem
);
2311 list_splice_init(&tset
.dst_csets
, &tset
.src_csets
);
2312 list_for_each_entry_safe(cset
, tmp_cset
, &tset
.src_csets
, mg_node
) {
2313 list_splice_tail_init(&cset
->mg_tasks
, &cset
->tasks
);
2314 list_del_init(&cset
->mg_node
);
2316 up_write(&css_set_rwsem
);
2321 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2322 * @dst_cgrp: the cgroup to attach to
2323 * @leader: the task or the leader of the threadgroup to be attached
2324 * @threadgroup: attach the whole threadgroup?
2326 * Call holding cgroup_mutex and threadgroup_lock of @leader.
2328 static int cgroup_attach_task(struct cgroup
*dst_cgrp
,
2329 struct task_struct
*leader
, bool threadgroup
)
2331 LIST_HEAD(preloaded_csets
);
2332 struct task_struct
*task
;
2335 /* look up all src csets */
2336 down_read(&css_set_rwsem
);
2340 cgroup_migrate_add_src(task_css_set(task
), dst_cgrp
,
2344 } while_each_thread(leader
, task
);
2346 up_read(&css_set_rwsem
);
2348 /* prepare dst csets and commit */
2349 ret
= cgroup_migrate_prepare_dst(dst_cgrp
, &preloaded_csets
);
2351 ret
= cgroup_migrate(dst_cgrp
, leader
, threadgroup
);
2353 cgroup_migrate_finish(&preloaded_csets
);
2358 * Find the task_struct of the task to attach by vpid and pass it along to the
2359 * function to attach either it or all tasks in its threadgroup. Will lock
2360 * cgroup_mutex and threadgroup.
2362 static ssize_t
__cgroup_procs_write(struct kernfs_open_file
*of
, char *buf
,
2363 size_t nbytes
, loff_t off
, bool threadgroup
)
2365 struct task_struct
*tsk
;
2366 const struct cred
*cred
= current_cred(), *tcred
;
2367 struct cgroup
*cgrp
;
2371 if (kstrtoint(strstrip(buf
), 0, &pid
) || pid
< 0)
2374 cgrp
= cgroup_kn_lock_live(of
->kn
);
2381 tsk
= find_task_by_vpid(pid
);
2385 goto out_unlock_cgroup
;
2388 * even if we're attaching all tasks in the thread group, we
2389 * only need to check permissions on one of them.
2391 tcred
= __task_cred(tsk
);
2392 if (!uid_eq(cred
->euid
, GLOBAL_ROOT_UID
) &&
2393 !uid_eq(cred
->euid
, tcred
->uid
) &&
2394 !uid_eq(cred
->euid
, tcred
->suid
)) {
2397 goto out_unlock_cgroup
;
2403 tsk
= tsk
->group_leader
;
2406 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2407 * trapped in a cpuset, or RT worker may be born in a cgroup
2408 * with no rt_runtime allocated. Just say no.
2410 if (tsk
== kthreadd_task
|| (tsk
->flags
& PF_NO_SETAFFINITY
)) {
2413 goto out_unlock_cgroup
;
2416 get_task_struct(tsk
);
2419 threadgroup_lock(tsk
);
2421 if (!thread_group_leader(tsk
)) {
2423 * a race with de_thread from another thread's exec()
2424 * may strip us of our leadership, if this happens,
2425 * there is no choice but to throw this task away and
2426 * try again; this is
2427 * "double-double-toil-and-trouble-check locking".
2429 threadgroup_unlock(tsk
);
2430 put_task_struct(tsk
);
2431 goto retry_find_task
;
2435 ret
= cgroup_attach_task(cgrp
, tsk
, threadgroup
);
2437 threadgroup_unlock(tsk
);
2439 put_task_struct(tsk
);
2441 cgroup_kn_unlock(of
->kn
);
2442 return ret
?: nbytes
;
2446 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2447 * @from: attach to all cgroups of a given task
2448 * @tsk: the task to be attached
2450 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
2452 struct cgroup_root
*root
;
2455 mutex_lock(&cgroup_mutex
);
2456 for_each_root(root
) {
2457 struct cgroup
*from_cgrp
;
2459 if (root
== &cgrp_dfl_root
)
2462 down_read(&css_set_rwsem
);
2463 from_cgrp
= task_cgroup_from_root(from
, root
);
2464 up_read(&css_set_rwsem
);
2466 retval
= cgroup_attach_task(from_cgrp
, tsk
, false);
2470 mutex_unlock(&cgroup_mutex
);
2474 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
2476 static ssize_t
cgroup_tasks_write(struct kernfs_open_file
*of
,
2477 char *buf
, size_t nbytes
, loff_t off
)
2479 return __cgroup_procs_write(of
, buf
, nbytes
, off
, false);
2482 static ssize_t
cgroup_procs_write(struct kernfs_open_file
*of
,
2483 char *buf
, size_t nbytes
, loff_t off
)
2485 return __cgroup_procs_write(of
, buf
, nbytes
, off
, true);
2488 static ssize_t
cgroup_release_agent_write(struct kernfs_open_file
*of
,
2489 char *buf
, size_t nbytes
, loff_t off
)
2491 struct cgroup
*cgrp
;
2493 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
2495 cgrp
= cgroup_kn_lock_live(of
->kn
);
2498 spin_lock(&release_agent_path_lock
);
2499 strlcpy(cgrp
->root
->release_agent_path
, strstrip(buf
),
2500 sizeof(cgrp
->root
->release_agent_path
));
2501 spin_unlock(&release_agent_path_lock
);
2502 cgroup_kn_unlock(of
->kn
);
2506 static int cgroup_release_agent_show(struct seq_file
*seq
, void *v
)
2508 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2510 spin_lock(&release_agent_path_lock
);
2511 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2512 spin_unlock(&release_agent_path_lock
);
2513 seq_putc(seq
, '\n');
2517 static int cgroup_sane_behavior_show(struct seq_file
*seq
, void *v
)
2519 seq_puts(seq
, "0\n");
2523 static void cgroup_print_ss_mask(struct seq_file
*seq
, unsigned int ss_mask
)
2525 struct cgroup_subsys
*ss
;
2526 bool printed
= false;
2529 for_each_subsys(ss
, ssid
) {
2530 if (ss_mask
& (1 << ssid
)) {
2533 seq_printf(seq
, "%s", ss
->name
);
2538 seq_putc(seq
, '\n');
2541 /* show controllers which are currently attached to the default hierarchy */
2542 static int cgroup_root_controllers_show(struct seq_file
*seq
, void *v
)
2544 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2546 cgroup_print_ss_mask(seq
, cgrp
->root
->subsys_mask
&
2547 ~cgrp_dfl_root_inhibit_ss_mask
);
2551 /* show controllers which are enabled from the parent */
2552 static int cgroup_controllers_show(struct seq_file
*seq
, void *v
)
2554 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2556 cgroup_print_ss_mask(seq
, cgroup_parent(cgrp
)->subtree_control
);
2560 /* show controllers which are enabled for a given cgroup's children */
2561 static int cgroup_subtree_control_show(struct seq_file
*seq
, void *v
)
2563 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2565 cgroup_print_ss_mask(seq
, cgrp
->subtree_control
);
2570 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2571 * @cgrp: root of the subtree to update csses for
2573 * @cgrp's child_subsys_mask has changed and its subtree's (self excluded)
2574 * css associations need to be updated accordingly. This function looks up
2575 * all css_sets which are attached to the subtree, creates the matching
2576 * updated css_sets and migrates the tasks to the new ones.
2578 static int cgroup_update_dfl_csses(struct cgroup
*cgrp
)
2580 LIST_HEAD(preloaded_csets
);
2581 struct cgroup_subsys_state
*css
;
2582 struct css_set
*src_cset
;
2585 lockdep_assert_held(&cgroup_mutex
);
2587 /* look up all csses currently attached to @cgrp's subtree */
2588 down_read(&css_set_rwsem
);
2589 css_for_each_descendant_pre(css
, cgroup_css(cgrp
, NULL
)) {
2590 struct cgrp_cset_link
*link
;
2592 /* self is not affected by child_subsys_mask change */
2593 if (css
->cgroup
== cgrp
)
2596 list_for_each_entry(link
, &css
->cgroup
->cset_links
, cset_link
)
2597 cgroup_migrate_add_src(link
->cset
, cgrp
,
2600 up_read(&css_set_rwsem
);
2602 /* NULL dst indicates self on default hierarchy */
2603 ret
= cgroup_migrate_prepare_dst(NULL
, &preloaded_csets
);
2607 list_for_each_entry(src_cset
, &preloaded_csets
, mg_preload_node
) {
2608 struct task_struct
*last_task
= NULL
, *task
;
2610 /* src_csets precede dst_csets, break on the first dst_cset */
2611 if (!src_cset
->mg_src_cgrp
)
2615 * All tasks in src_cset need to be migrated to the
2616 * matching dst_cset. Empty it process by process. We
2617 * walk tasks but migrate processes. The leader might even
2618 * belong to a different cset but such src_cset would also
2619 * be among the target src_csets because the default
2620 * hierarchy enforces per-process membership.
2623 down_read(&css_set_rwsem
);
2624 task
= list_first_entry_or_null(&src_cset
->tasks
,
2625 struct task_struct
, cg_list
);
2627 task
= task
->group_leader
;
2628 WARN_ON_ONCE(!task_css_set(task
)->mg_src_cgrp
);
2629 get_task_struct(task
);
2631 up_read(&css_set_rwsem
);
2636 /* guard against possible infinite loop */
2637 if (WARN(last_task
== task
,
2638 "cgroup: update_dfl_csses failed to make progress, aborting in inconsistent state\n"))
2642 threadgroup_lock(task
);
2643 /* raced against de_thread() from another thread? */
2644 if (!thread_group_leader(task
)) {
2645 threadgroup_unlock(task
);
2646 put_task_struct(task
);
2650 ret
= cgroup_migrate(src_cset
->dfl_cgrp
, task
, true);
2652 threadgroup_unlock(task
);
2653 put_task_struct(task
);
2655 if (WARN(ret
, "cgroup: failed to update controllers for the default hierarchy (%d), further operations may crash or hang\n", ret
))
2661 cgroup_migrate_finish(&preloaded_csets
);
2665 /* change the enabled child controllers for a cgroup in the default hierarchy */
2666 static ssize_t
cgroup_subtree_control_write(struct kernfs_open_file
*of
,
2667 char *buf
, size_t nbytes
,
2670 unsigned int enable
= 0, disable
= 0;
2671 unsigned int css_enable
, css_disable
, old_ctrl
, new_ctrl
;
2672 struct cgroup
*cgrp
, *child
;
2673 struct cgroup_subsys
*ss
;
2678 * Parse input - space separated list of subsystem names prefixed
2679 * with either + or -.
2681 buf
= strstrip(buf
);
2682 while ((tok
= strsep(&buf
, " "))) {
2685 for_each_subsys(ss
, ssid
) {
2686 if (ss
->disabled
|| strcmp(tok
+ 1, ss
->name
) ||
2687 ((1 << ss
->id
) & cgrp_dfl_root_inhibit_ss_mask
))
2691 enable
|= 1 << ssid
;
2692 disable
&= ~(1 << ssid
);
2693 } else if (*tok
== '-') {
2694 disable
|= 1 << ssid
;
2695 enable
&= ~(1 << ssid
);
2701 if (ssid
== CGROUP_SUBSYS_COUNT
)
2705 cgrp
= cgroup_kn_lock_live(of
->kn
);
2709 for_each_subsys(ss
, ssid
) {
2710 if (enable
& (1 << ssid
)) {
2711 if (cgrp
->subtree_control
& (1 << ssid
)) {
2712 enable
&= ~(1 << ssid
);
2716 /* unavailable or not enabled on the parent? */
2717 if (!(cgrp_dfl_root
.subsys_mask
& (1 << ssid
)) ||
2718 (cgroup_parent(cgrp
) &&
2719 !(cgroup_parent(cgrp
)->subtree_control
& (1 << ssid
)))) {
2725 * @ss is already enabled through dependency and
2726 * we'll just make it visible. Skip draining.
2728 if (cgrp
->child_subsys_mask
& (1 << ssid
))
2732 * Because css offlining is asynchronous, userland
2733 * might try to re-enable the same controller while
2734 * the previous instance is still around. In such
2735 * cases, wait till it's gone using offline_waitq.
2737 cgroup_for_each_live_child(child
, cgrp
) {
2740 if (!cgroup_css(child
, ss
))
2744 prepare_to_wait(&child
->offline_waitq
, &wait
,
2745 TASK_UNINTERRUPTIBLE
);
2746 cgroup_kn_unlock(of
->kn
);
2748 finish_wait(&child
->offline_waitq
, &wait
);
2751 return restart_syscall();
2753 } else if (disable
& (1 << ssid
)) {
2754 if (!(cgrp
->subtree_control
& (1 << ssid
))) {
2755 disable
&= ~(1 << ssid
);
2759 /* a child has it enabled? */
2760 cgroup_for_each_live_child(child
, cgrp
) {
2761 if (child
->subtree_control
& (1 << ssid
)) {
2769 if (!enable
&& !disable
) {
2775 * Except for the root, subtree_control must be zero for a cgroup
2776 * with tasks so that child cgroups don't compete against tasks.
2778 if (enable
&& cgroup_parent(cgrp
) && !list_empty(&cgrp
->cset_links
)) {
2784 * Update subsys masks and calculate what needs to be done. More
2785 * subsystems than specified may need to be enabled or disabled
2786 * depending on subsystem dependencies.
2788 cgrp
->subtree_control
|= enable
;
2789 cgrp
->subtree_control
&= ~disable
;
2791 old_ctrl
= cgrp
->child_subsys_mask
;
2792 cgroup_refresh_child_subsys_mask(cgrp
);
2793 new_ctrl
= cgrp
->child_subsys_mask
;
2795 css_enable
= ~old_ctrl
& new_ctrl
;
2796 css_disable
= old_ctrl
& ~new_ctrl
;
2797 enable
|= css_enable
;
2798 disable
|= css_disable
;
2801 * Create new csses or make the existing ones visible. A css is
2802 * created invisible if it's being implicitly enabled through
2803 * dependency. An invisible css is made visible when the userland
2804 * explicitly enables it.
2806 for_each_subsys(ss
, ssid
) {
2807 if (!(enable
& (1 << ssid
)))
2810 cgroup_for_each_live_child(child
, cgrp
) {
2811 if (css_enable
& (1 << ssid
))
2812 ret
= create_css(child
, ss
,
2813 cgrp
->subtree_control
& (1 << ssid
));
2815 ret
= cgroup_populate_dir(child
, 1 << ssid
);
2822 * At this point, cgroup_e_css() results reflect the new csses
2823 * making the following cgroup_update_dfl_csses() properly update
2824 * css associations of all tasks in the subtree.
2826 ret
= cgroup_update_dfl_csses(cgrp
);
2831 * All tasks are migrated out of disabled csses. Kill or hide
2832 * them. A css is hidden when the userland requests it to be
2833 * disabled while other subsystems are still depending on it. The
2834 * css must not actively control resources and be in the vanilla
2835 * state if it's made visible again later. Controllers which may
2836 * be depended upon should provide ->css_reset() for this purpose.
2838 for_each_subsys(ss
, ssid
) {
2839 if (!(disable
& (1 << ssid
)))
2842 cgroup_for_each_live_child(child
, cgrp
) {
2843 struct cgroup_subsys_state
*css
= cgroup_css(child
, ss
);
2845 if (css_disable
& (1 << ssid
)) {
2848 cgroup_clear_dir(child
, 1 << ssid
);
2855 kernfs_activate(cgrp
->kn
);
2858 cgroup_kn_unlock(of
->kn
);
2859 return ret
?: nbytes
;
2862 cgrp
->subtree_control
&= ~enable
;
2863 cgrp
->subtree_control
|= disable
;
2864 cgroup_refresh_child_subsys_mask(cgrp
);
2866 for_each_subsys(ss
, ssid
) {
2867 if (!(enable
& (1 << ssid
)))
2870 cgroup_for_each_live_child(child
, cgrp
) {
2871 struct cgroup_subsys_state
*css
= cgroup_css(child
, ss
);
2876 if (css_enable
& (1 << ssid
))
2879 cgroup_clear_dir(child
, 1 << ssid
);
2885 static int cgroup_populated_show(struct seq_file
*seq
, void *v
)
2887 seq_printf(seq
, "%d\n", (bool)seq_css(seq
)->cgroup
->populated_cnt
);
2891 static ssize_t
cgroup_file_write(struct kernfs_open_file
*of
, char *buf
,
2892 size_t nbytes
, loff_t off
)
2894 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
2895 struct cftype
*cft
= of
->kn
->priv
;
2896 struct cgroup_subsys_state
*css
;
2900 return cft
->write(of
, buf
, nbytes
, off
);
2903 * kernfs guarantees that a file isn't deleted with operations in
2904 * flight, which means that the matching css is and stays alive and
2905 * doesn't need to be pinned. The RCU locking is not necessary
2906 * either. It's just for the convenience of using cgroup_css().
2909 css
= cgroup_css(cgrp
, cft
->ss
);
2912 if (cft
->write_u64
) {
2913 unsigned long long v
;
2914 ret
= kstrtoull(buf
, 0, &v
);
2916 ret
= cft
->write_u64(css
, cft
, v
);
2917 } else if (cft
->write_s64
) {
2919 ret
= kstrtoll(buf
, 0, &v
);
2921 ret
= cft
->write_s64(css
, cft
, v
);
2926 return ret
?: nbytes
;
2929 static void *cgroup_seqfile_start(struct seq_file
*seq
, loff_t
*ppos
)
2931 return seq_cft(seq
)->seq_start(seq
, ppos
);
2934 static void *cgroup_seqfile_next(struct seq_file
*seq
, void *v
, loff_t
*ppos
)
2936 return seq_cft(seq
)->seq_next(seq
, v
, ppos
);
2939 static void cgroup_seqfile_stop(struct seq_file
*seq
, void *v
)
2941 seq_cft(seq
)->seq_stop(seq
, v
);
2944 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2946 struct cftype
*cft
= seq_cft(m
);
2947 struct cgroup_subsys_state
*css
= seq_css(m
);
2950 return cft
->seq_show(m
, arg
);
2953 seq_printf(m
, "%llu\n", cft
->read_u64(css
, cft
));
2954 else if (cft
->read_s64
)
2955 seq_printf(m
, "%lld\n", cft
->read_s64(css
, cft
));
2961 static struct kernfs_ops cgroup_kf_single_ops
= {
2962 .atomic_write_len
= PAGE_SIZE
,
2963 .write
= cgroup_file_write
,
2964 .seq_show
= cgroup_seqfile_show
,
2967 static struct kernfs_ops cgroup_kf_ops
= {
2968 .atomic_write_len
= PAGE_SIZE
,
2969 .write
= cgroup_file_write
,
2970 .seq_start
= cgroup_seqfile_start
,
2971 .seq_next
= cgroup_seqfile_next
,
2972 .seq_stop
= cgroup_seqfile_stop
,
2973 .seq_show
= cgroup_seqfile_show
,
2977 * cgroup_rename - Only allow simple rename of directories in place.
2979 static int cgroup_rename(struct kernfs_node
*kn
, struct kernfs_node
*new_parent
,
2980 const char *new_name_str
)
2982 struct cgroup
*cgrp
= kn
->priv
;
2985 if (kernfs_type(kn
) != KERNFS_DIR
)
2987 if (kn
->parent
!= new_parent
)
2991 * This isn't a proper migration and its usefulness is very
2992 * limited. Disallow on the default hierarchy.
2994 if (cgroup_on_dfl(cgrp
))
2998 * We're gonna grab cgroup_mutex which nests outside kernfs
2999 * active_ref. kernfs_rename() doesn't require active_ref
3000 * protection. Break them before grabbing cgroup_mutex.
3002 kernfs_break_active_protection(new_parent
);
3003 kernfs_break_active_protection(kn
);
3005 mutex_lock(&cgroup_mutex
);
3007 ret
= kernfs_rename(kn
, new_parent
, new_name_str
);
3009 mutex_unlock(&cgroup_mutex
);
3011 kernfs_unbreak_active_protection(kn
);
3012 kernfs_unbreak_active_protection(new_parent
);
3016 /* set uid and gid of cgroup dirs and files to that of the creator */
3017 static int cgroup_kn_set_ugid(struct kernfs_node
*kn
)
3019 struct iattr iattr
= { .ia_valid
= ATTR_UID
| ATTR_GID
,
3020 .ia_uid
= current_fsuid(),
3021 .ia_gid
= current_fsgid(), };
3023 if (uid_eq(iattr
.ia_uid
, GLOBAL_ROOT_UID
) &&
3024 gid_eq(iattr
.ia_gid
, GLOBAL_ROOT_GID
))
3027 return kernfs_setattr(kn
, &iattr
);
3030 static int cgroup_add_file(struct cgroup
*cgrp
, struct cftype
*cft
)
3032 char name
[CGROUP_FILE_NAME_MAX
];
3033 struct kernfs_node
*kn
;
3034 struct lock_class_key
*key
= NULL
;
3037 #ifdef CONFIG_DEBUG_LOCK_ALLOC
3038 key
= &cft
->lockdep_key
;
3040 kn
= __kernfs_create_file(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
),
3041 cgroup_file_mode(cft
), 0, cft
->kf_ops
, cft
,
3046 ret
= cgroup_kn_set_ugid(kn
);
3052 if (cft
->seq_show
== cgroup_populated_show
)
3053 cgrp
->populated_kn
= kn
;
3058 * cgroup_addrm_files - add or remove files to a cgroup directory
3059 * @cgrp: the target cgroup
3060 * @cfts: array of cftypes to be added
3061 * @is_add: whether to add or remove
3063 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3064 * For removals, this function never fails. If addition fails, this
3065 * function doesn't remove files already added. The caller is responsible
3068 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cftype cfts
[],
3074 lockdep_assert_held(&cgroup_mutex
);
3076 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3077 /* does cft->flags tell us to skip this file on @cgrp? */
3078 if ((cft
->flags
& __CFTYPE_ONLY_ON_DFL
) && !cgroup_on_dfl(cgrp
))
3080 if ((cft
->flags
& __CFTYPE_NOT_ON_DFL
) && cgroup_on_dfl(cgrp
))
3082 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgroup_parent(cgrp
))
3084 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgroup_parent(cgrp
))
3088 ret
= cgroup_add_file(cgrp
, cft
);
3090 pr_warn("%s: failed to add %s, err=%d\n",
3091 __func__
, cft
->name
, ret
);
3095 cgroup_rm_file(cgrp
, cft
);
3101 static int cgroup_apply_cftypes(struct cftype
*cfts
, bool is_add
)
3104 struct cgroup_subsys
*ss
= cfts
[0].ss
;
3105 struct cgroup
*root
= &ss
->root
->cgrp
;
3106 struct cgroup_subsys_state
*css
;
3109 lockdep_assert_held(&cgroup_mutex
);
3111 /* add/rm files for all cgroups created before */
3112 css_for_each_descendant_pre(css
, cgroup_css(root
, ss
)) {
3113 struct cgroup
*cgrp
= css
->cgroup
;
3115 if (cgroup_is_dead(cgrp
))
3118 ret
= cgroup_addrm_files(cgrp
, cfts
, is_add
);
3124 kernfs_activate(root
->kn
);
3128 static void cgroup_exit_cftypes(struct cftype
*cfts
)
3132 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3133 /* free copy for custom atomic_write_len, see init_cftypes() */
3134 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
)
3139 /* revert flags set by cgroup core while adding @cfts */
3140 cft
->flags
&= ~(__CFTYPE_ONLY_ON_DFL
| __CFTYPE_NOT_ON_DFL
);
3144 static int cgroup_init_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3148 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3149 struct kernfs_ops
*kf_ops
;
3151 WARN_ON(cft
->ss
|| cft
->kf_ops
);
3154 kf_ops
= &cgroup_kf_ops
;
3156 kf_ops
= &cgroup_kf_single_ops
;
3159 * Ugh... if @cft wants a custom max_write_len, we need to
3160 * make a copy of kf_ops to set its atomic_write_len.
3162 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
) {
3163 kf_ops
= kmemdup(kf_ops
, sizeof(*kf_ops
), GFP_KERNEL
);
3165 cgroup_exit_cftypes(cfts
);
3168 kf_ops
->atomic_write_len
= cft
->max_write_len
;
3171 cft
->kf_ops
= kf_ops
;
3178 static int cgroup_rm_cftypes_locked(struct cftype
*cfts
)
3180 lockdep_assert_held(&cgroup_mutex
);
3182 if (!cfts
|| !cfts
[0].ss
)
3185 list_del(&cfts
->node
);
3186 cgroup_apply_cftypes(cfts
, false);
3187 cgroup_exit_cftypes(cfts
);
3192 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
3193 * @cfts: zero-length name terminated array of cftypes
3195 * Unregister @cfts. Files described by @cfts are removed from all
3196 * existing cgroups and all future cgroups won't have them either. This
3197 * function can be called anytime whether @cfts' subsys is attached or not.
3199 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
3202 int cgroup_rm_cftypes(struct cftype
*cfts
)
3206 mutex_lock(&cgroup_mutex
);
3207 ret
= cgroup_rm_cftypes_locked(cfts
);
3208 mutex_unlock(&cgroup_mutex
);
3213 * cgroup_add_cftypes - add an array of cftypes to a subsystem
3214 * @ss: target cgroup subsystem
3215 * @cfts: zero-length name terminated array of cftypes
3217 * Register @cfts to @ss. Files described by @cfts are created for all
3218 * existing cgroups to which @ss is attached and all future cgroups will
3219 * have them too. This function can be called anytime whether @ss is
3222 * Returns 0 on successful registration, -errno on failure. Note that this
3223 * function currently returns 0 as long as @cfts registration is successful
3224 * even if some file creation attempts on existing cgroups fail.
3226 static int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3233 if (!cfts
|| cfts
[0].name
[0] == '\0')
3236 ret
= cgroup_init_cftypes(ss
, cfts
);
3240 mutex_lock(&cgroup_mutex
);
3242 list_add_tail(&cfts
->node
, &ss
->cfts
);
3243 ret
= cgroup_apply_cftypes(cfts
, true);
3245 cgroup_rm_cftypes_locked(cfts
);
3247 mutex_unlock(&cgroup_mutex
);
3252 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
3253 * @ss: target cgroup subsystem
3254 * @cfts: zero-length name terminated array of cftypes
3256 * Similar to cgroup_add_cftypes() but the added files are only used for
3257 * the default hierarchy.
3259 int cgroup_add_dfl_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3263 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
3264 cft
->flags
|= __CFTYPE_ONLY_ON_DFL
;
3265 return cgroup_add_cftypes(ss
, cfts
);
3269 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
3270 * @ss: target cgroup subsystem
3271 * @cfts: zero-length name terminated array of cftypes
3273 * Similar to cgroup_add_cftypes() but the added files are only used for
3274 * the legacy hierarchies.
3276 int cgroup_add_legacy_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3281 * If legacy_flies_on_dfl, we want to show the legacy files on the
3282 * dfl hierarchy but iff the target subsystem hasn't been updated
3283 * for the dfl hierarchy yet.
3285 if (!cgroup_legacy_files_on_dfl
||
3286 ss
->dfl_cftypes
!= ss
->legacy_cftypes
) {
3287 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
3288 cft
->flags
|= __CFTYPE_NOT_ON_DFL
;
3291 return cgroup_add_cftypes(ss
, cfts
);
3295 * cgroup_task_count - count the number of tasks in a cgroup.
3296 * @cgrp: the cgroup in question
3298 * Return the number of tasks in the cgroup.
3300 static int cgroup_task_count(const struct cgroup
*cgrp
)
3303 struct cgrp_cset_link
*link
;
3305 down_read(&css_set_rwsem
);
3306 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
3307 count
+= atomic_read(&link
->cset
->refcount
);
3308 up_read(&css_set_rwsem
);
3313 * css_next_child - find the next child of a given css
3314 * @pos: the current position (%NULL to initiate traversal)
3315 * @parent: css whose children to walk
3317 * This function returns the next child of @parent and should be called
3318 * under either cgroup_mutex or RCU read lock. The only requirement is
3319 * that @parent and @pos are accessible. The next sibling is guaranteed to
3320 * be returned regardless of their states.
3322 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3323 * css which finished ->css_online() is guaranteed to be visible in the
3324 * future iterations and will stay visible until the last reference is put.
3325 * A css which hasn't finished ->css_online() or already finished
3326 * ->css_offline() may show up during traversal. It's each subsystem's
3327 * responsibility to synchronize against on/offlining.
3329 struct cgroup_subsys_state
*css_next_child(struct cgroup_subsys_state
*pos
,
3330 struct cgroup_subsys_state
*parent
)
3332 struct cgroup_subsys_state
*next
;
3334 cgroup_assert_mutex_or_rcu_locked();
3337 * @pos could already have been unlinked from the sibling list.
3338 * Once a cgroup is removed, its ->sibling.next is no longer
3339 * updated when its next sibling changes. CSS_RELEASED is set when
3340 * @pos is taken off list, at which time its next pointer is valid,
3341 * and, as releases are serialized, the one pointed to by the next
3342 * pointer is guaranteed to not have started release yet. This
3343 * implies that if we observe !CSS_RELEASED on @pos in this RCU
3344 * critical section, the one pointed to by its next pointer is
3345 * guaranteed to not have finished its RCU grace period even if we
3346 * have dropped rcu_read_lock() inbetween iterations.
3348 * If @pos has CSS_RELEASED set, its next pointer can't be
3349 * dereferenced; however, as each css is given a monotonically
3350 * increasing unique serial number and always appended to the
3351 * sibling list, the next one can be found by walking the parent's
3352 * children until the first css with higher serial number than
3353 * @pos's. While this path can be slower, it happens iff iteration
3354 * races against release and the race window is very small.
3357 next
= list_entry_rcu(parent
->children
.next
, struct cgroup_subsys_state
, sibling
);
3358 } else if (likely(!(pos
->flags
& CSS_RELEASED
))) {
3359 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup_subsys_state
, sibling
);
3361 list_for_each_entry_rcu(next
, &parent
->children
, sibling
)
3362 if (next
->serial_nr
> pos
->serial_nr
)
3367 * @next, if not pointing to the head, can be dereferenced and is
3370 if (&next
->sibling
!= &parent
->children
)
3376 * css_next_descendant_pre - find the next descendant for pre-order walk
3377 * @pos: the current position (%NULL to initiate traversal)
3378 * @root: css whose descendants to walk
3380 * To be used by css_for_each_descendant_pre(). Find the next descendant
3381 * to visit for pre-order traversal of @root's descendants. @root is
3382 * included in the iteration and the first node to be visited.
3384 * While this function requires cgroup_mutex or RCU read locking, it
3385 * doesn't require the whole traversal to be contained in a single critical
3386 * section. This function will return the correct next descendant as long
3387 * as both @pos and @root are accessible and @pos is a descendant of @root.
3389 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3390 * css which finished ->css_online() is guaranteed to be visible in the
3391 * future iterations and will stay visible until the last reference is put.
3392 * A css which hasn't finished ->css_online() or already finished
3393 * ->css_offline() may show up during traversal. It's each subsystem's
3394 * responsibility to synchronize against on/offlining.
3396 struct cgroup_subsys_state
*
3397 css_next_descendant_pre(struct cgroup_subsys_state
*pos
,
3398 struct cgroup_subsys_state
*root
)
3400 struct cgroup_subsys_state
*next
;
3402 cgroup_assert_mutex_or_rcu_locked();
3404 /* if first iteration, visit @root */
3408 /* visit the first child if exists */
3409 next
= css_next_child(NULL
, pos
);
3413 /* no child, visit my or the closest ancestor's next sibling */
3414 while (pos
!= root
) {
3415 next
= css_next_child(pos
, pos
->parent
);
3425 * css_rightmost_descendant - return the rightmost descendant of a css
3426 * @pos: css of interest
3428 * Return the rightmost descendant of @pos. If there's no descendant, @pos
3429 * is returned. This can be used during pre-order traversal to skip
3432 * While this function requires cgroup_mutex or RCU read locking, it
3433 * doesn't require the whole traversal to be contained in a single critical
3434 * section. This function will return the correct rightmost descendant as
3435 * long as @pos is accessible.
3437 struct cgroup_subsys_state
*
3438 css_rightmost_descendant(struct cgroup_subsys_state
*pos
)
3440 struct cgroup_subsys_state
*last
, *tmp
;
3442 cgroup_assert_mutex_or_rcu_locked();
3446 /* ->prev isn't RCU safe, walk ->next till the end */
3448 css_for_each_child(tmp
, last
)
3455 static struct cgroup_subsys_state
*
3456 css_leftmost_descendant(struct cgroup_subsys_state
*pos
)
3458 struct cgroup_subsys_state
*last
;
3462 pos
= css_next_child(NULL
, pos
);
3469 * css_next_descendant_post - find the next descendant for post-order walk
3470 * @pos: the current position (%NULL to initiate traversal)
3471 * @root: css whose descendants to walk
3473 * To be used by css_for_each_descendant_post(). Find the next descendant
3474 * to visit for post-order traversal of @root's descendants. @root is
3475 * included in the iteration and the last node to be visited.
3477 * While this function requires cgroup_mutex or RCU read locking, it
3478 * doesn't require the whole traversal to be contained in a single critical
3479 * section. This function will return the correct next descendant as long
3480 * as both @pos and @cgroup are accessible and @pos is a descendant of
3483 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3484 * css which finished ->css_online() is guaranteed to be visible in the
3485 * future iterations and will stay visible until the last reference is put.
3486 * A css which hasn't finished ->css_online() or already finished
3487 * ->css_offline() may show up during traversal. It's each subsystem's
3488 * responsibility to synchronize against on/offlining.
3490 struct cgroup_subsys_state
*
3491 css_next_descendant_post(struct cgroup_subsys_state
*pos
,
3492 struct cgroup_subsys_state
*root
)
3494 struct cgroup_subsys_state
*next
;
3496 cgroup_assert_mutex_or_rcu_locked();
3498 /* if first iteration, visit leftmost descendant which may be @root */
3500 return css_leftmost_descendant(root
);
3502 /* if we visited @root, we're done */
3506 /* if there's an unvisited sibling, visit its leftmost descendant */
3507 next
= css_next_child(pos
, pos
->parent
);
3509 return css_leftmost_descendant(next
);
3511 /* no sibling left, visit parent */
3516 * css_has_online_children - does a css have online children
3517 * @css: the target css
3519 * Returns %true if @css has any online children; otherwise, %false. This
3520 * function can be called from any context but the caller is responsible
3521 * for synchronizing against on/offlining as necessary.
3523 bool css_has_online_children(struct cgroup_subsys_state
*css
)
3525 struct cgroup_subsys_state
*child
;
3529 css_for_each_child(child
, css
) {
3530 if (child
->flags
& CSS_ONLINE
) {
3540 * css_advance_task_iter - advance a task itererator to the next css_set
3541 * @it: the iterator to advance
3543 * Advance @it to the next css_set to walk.
3545 static void css_advance_task_iter(struct css_task_iter
*it
)
3547 struct list_head
*l
= it
->cset_pos
;
3548 struct cgrp_cset_link
*link
;
3549 struct css_set
*cset
;
3551 /* Advance to the next non-empty css_set */
3554 if (l
== it
->cset_head
) {
3555 it
->cset_pos
= NULL
;
3560 cset
= container_of(l
, struct css_set
,
3561 e_cset_node
[it
->ss
->id
]);
3563 link
= list_entry(l
, struct cgrp_cset_link
, cset_link
);
3566 } while (list_empty(&cset
->tasks
) && list_empty(&cset
->mg_tasks
));
3570 if (!list_empty(&cset
->tasks
))
3571 it
->task_pos
= cset
->tasks
.next
;
3573 it
->task_pos
= cset
->mg_tasks
.next
;
3575 it
->tasks_head
= &cset
->tasks
;
3576 it
->mg_tasks_head
= &cset
->mg_tasks
;
3580 * css_task_iter_start - initiate task iteration
3581 * @css: the css to walk tasks of
3582 * @it: the task iterator to use
3584 * Initiate iteration through the tasks of @css. The caller can call
3585 * css_task_iter_next() to walk through the tasks until the function
3586 * returns NULL. On completion of iteration, css_task_iter_end() must be
3589 * Note that this function acquires a lock which is released when the
3590 * iteration finishes. The caller can't sleep while iteration is in
3593 void css_task_iter_start(struct cgroup_subsys_state
*css
,
3594 struct css_task_iter
*it
)
3595 __acquires(css_set_rwsem
)
3597 /* no one should try to iterate before mounting cgroups */
3598 WARN_ON_ONCE(!use_task_css_set_links
);
3600 down_read(&css_set_rwsem
);
3605 it
->cset_pos
= &css
->cgroup
->e_csets
[css
->ss
->id
];
3607 it
->cset_pos
= &css
->cgroup
->cset_links
;
3609 it
->cset_head
= it
->cset_pos
;
3611 css_advance_task_iter(it
);
3615 * css_task_iter_next - return the next task for the iterator
3616 * @it: the task iterator being iterated
3618 * The "next" function for task iteration. @it should have been
3619 * initialized via css_task_iter_start(). Returns NULL when the iteration
3622 struct task_struct
*css_task_iter_next(struct css_task_iter
*it
)
3624 struct task_struct
*res
;
3625 struct list_head
*l
= it
->task_pos
;
3627 /* If the iterator cg is NULL, we have no tasks */
3630 res
= list_entry(l
, struct task_struct
, cg_list
);
3633 * Advance iterator to find next entry. cset->tasks is consumed
3634 * first and then ->mg_tasks. After ->mg_tasks, we move onto the
3639 if (l
== it
->tasks_head
)
3640 l
= it
->mg_tasks_head
->next
;
3642 if (l
== it
->mg_tasks_head
)
3643 css_advance_task_iter(it
);
3651 * css_task_iter_end - finish task iteration
3652 * @it: the task iterator to finish
3654 * Finish task iteration started by css_task_iter_start().
3656 void css_task_iter_end(struct css_task_iter
*it
)
3657 __releases(css_set_rwsem
)
3659 up_read(&css_set_rwsem
);
3663 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3664 * @to: cgroup to which the tasks will be moved
3665 * @from: cgroup in which the tasks currently reside
3667 * Locking rules between cgroup_post_fork() and the migration path
3668 * guarantee that, if a task is forking while being migrated, the new child
3669 * is guaranteed to be either visible in the source cgroup after the
3670 * parent's migration is complete or put into the target cgroup. No task
3671 * can slip out of migration through forking.
3673 int cgroup_transfer_tasks(struct cgroup
*to
, struct cgroup
*from
)
3675 LIST_HEAD(preloaded_csets
);
3676 struct cgrp_cset_link
*link
;
3677 struct css_task_iter it
;
3678 struct task_struct
*task
;
3681 mutex_lock(&cgroup_mutex
);
3683 /* all tasks in @from are being moved, all csets are source */
3684 down_read(&css_set_rwsem
);
3685 list_for_each_entry(link
, &from
->cset_links
, cset_link
)
3686 cgroup_migrate_add_src(link
->cset
, to
, &preloaded_csets
);
3687 up_read(&css_set_rwsem
);
3689 ret
= cgroup_migrate_prepare_dst(to
, &preloaded_csets
);
3694 * Migrate tasks one-by-one until @form is empty. This fails iff
3695 * ->can_attach() fails.
3698 css_task_iter_start(&from
->self
, &it
);
3699 task
= css_task_iter_next(&it
);
3701 get_task_struct(task
);
3702 css_task_iter_end(&it
);
3705 ret
= cgroup_migrate(to
, task
, false);
3706 put_task_struct(task
);
3708 } while (task
&& !ret
);
3710 cgroup_migrate_finish(&preloaded_csets
);
3711 mutex_unlock(&cgroup_mutex
);
3716 * Stuff for reading the 'tasks'/'procs' files.
3718 * Reading this file can return large amounts of data if a cgroup has
3719 * *lots* of attached tasks. So it may need several calls to read(),
3720 * but we cannot guarantee that the information we produce is correct
3721 * unless we produce it entirely atomically.
3725 /* which pidlist file are we talking about? */
3726 enum cgroup_filetype
{
3732 * A pidlist is a list of pids that virtually represents the contents of one
3733 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3734 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3737 struct cgroup_pidlist
{
3739 * used to find which pidlist is wanted. doesn't change as long as
3740 * this particular list stays in the list.
3742 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
3745 /* how many elements the above list has */
3747 /* each of these stored in a list by its cgroup */
3748 struct list_head links
;
3749 /* pointer to the cgroup we belong to, for list removal purposes */
3750 struct cgroup
*owner
;
3751 /* for delayed destruction */
3752 struct delayed_work destroy_dwork
;
3756 * The following two functions "fix" the issue where there are more pids
3757 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3758 * TODO: replace with a kernel-wide solution to this problem
3760 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3761 static void *pidlist_allocate(int count
)
3763 if (PIDLIST_TOO_LARGE(count
))
3764 return vmalloc(count
* sizeof(pid_t
));
3766 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
3769 static void pidlist_free(void *p
)
3771 if (is_vmalloc_addr(p
))
3778 * Used to destroy all pidlists lingering waiting for destroy timer. None
3779 * should be left afterwards.
3781 static void cgroup_pidlist_destroy_all(struct cgroup
*cgrp
)
3783 struct cgroup_pidlist
*l
, *tmp_l
;
3785 mutex_lock(&cgrp
->pidlist_mutex
);
3786 list_for_each_entry_safe(l
, tmp_l
, &cgrp
->pidlists
, links
)
3787 mod_delayed_work(cgroup_pidlist_destroy_wq
, &l
->destroy_dwork
, 0);
3788 mutex_unlock(&cgrp
->pidlist_mutex
);
3790 flush_workqueue(cgroup_pidlist_destroy_wq
);
3791 BUG_ON(!list_empty(&cgrp
->pidlists
));
3794 static void cgroup_pidlist_destroy_work_fn(struct work_struct
*work
)
3796 struct delayed_work
*dwork
= to_delayed_work(work
);
3797 struct cgroup_pidlist
*l
= container_of(dwork
, struct cgroup_pidlist
,
3799 struct cgroup_pidlist
*tofree
= NULL
;
3801 mutex_lock(&l
->owner
->pidlist_mutex
);
3804 * Destroy iff we didn't get queued again. The state won't change
3805 * as destroy_dwork can only be queued while locked.
3807 if (!delayed_work_pending(dwork
)) {
3808 list_del(&l
->links
);
3809 pidlist_free(l
->list
);
3810 put_pid_ns(l
->key
.ns
);
3814 mutex_unlock(&l
->owner
->pidlist_mutex
);
3819 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3820 * Returns the number of unique elements.
3822 static int pidlist_uniq(pid_t
*list
, int length
)
3827 * we presume the 0th element is unique, so i starts at 1. trivial
3828 * edge cases first; no work needs to be done for either
3830 if (length
== 0 || length
== 1)
3832 /* src and dest walk down the list; dest counts unique elements */
3833 for (src
= 1; src
< length
; src
++) {
3834 /* find next unique element */
3835 while (list
[src
] == list
[src
-1]) {
3840 /* dest always points to where the next unique element goes */
3841 list
[dest
] = list
[src
];
3849 * The two pid files - task and cgroup.procs - guaranteed that the result
3850 * is sorted, which forced this whole pidlist fiasco. As pid order is
3851 * different per namespace, each namespace needs differently sorted list,
3852 * making it impossible to use, for example, single rbtree of member tasks
3853 * sorted by task pointer. As pidlists can be fairly large, allocating one
3854 * per open file is dangerous, so cgroup had to implement shared pool of
3855 * pidlists keyed by cgroup and namespace.
3857 * All this extra complexity was caused by the original implementation
3858 * committing to an entirely unnecessary property. In the long term, we
3859 * want to do away with it. Explicitly scramble sort order if on the
3860 * default hierarchy so that no such expectation exists in the new
3863 * Scrambling is done by swapping every two consecutive bits, which is
3864 * non-identity one-to-one mapping which disturbs sort order sufficiently.
3866 static pid_t
pid_fry(pid_t pid
)
3868 unsigned a
= pid
& 0x55555555;
3869 unsigned b
= pid
& 0xAAAAAAAA;
3871 return (a
<< 1) | (b
>> 1);
3874 static pid_t
cgroup_pid_fry(struct cgroup
*cgrp
, pid_t pid
)
3876 if (cgroup_on_dfl(cgrp
))
3877 return pid_fry(pid
);
3882 static int cmppid(const void *a
, const void *b
)
3884 return *(pid_t
*)a
- *(pid_t
*)b
;
3887 static int fried_cmppid(const void *a
, const void *b
)
3889 return pid_fry(*(pid_t
*)a
) - pid_fry(*(pid_t
*)b
);
3892 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
3893 enum cgroup_filetype type
)
3895 struct cgroup_pidlist
*l
;
3896 /* don't need task_nsproxy() if we're looking at ourself */
3897 struct pid_namespace
*ns
= task_active_pid_ns(current
);
3899 lockdep_assert_held(&cgrp
->pidlist_mutex
);
3901 list_for_each_entry(l
, &cgrp
->pidlists
, links
)
3902 if (l
->key
.type
== type
&& l
->key
.ns
== ns
)
3908 * find the appropriate pidlist for our purpose (given procs vs tasks)
3909 * returns with the lock on that pidlist already held, and takes care
3910 * of the use count, or returns NULL with no locks held if we're out of
3913 static struct cgroup_pidlist
*cgroup_pidlist_find_create(struct cgroup
*cgrp
,
3914 enum cgroup_filetype type
)
3916 struct cgroup_pidlist
*l
;
3918 lockdep_assert_held(&cgrp
->pidlist_mutex
);
3920 l
= cgroup_pidlist_find(cgrp
, type
);
3924 /* entry not found; create a new one */
3925 l
= kzalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
3929 INIT_DELAYED_WORK(&l
->destroy_dwork
, cgroup_pidlist_destroy_work_fn
);
3931 /* don't need task_nsproxy() if we're looking at ourself */
3932 l
->key
.ns
= get_pid_ns(task_active_pid_ns(current
));
3934 list_add(&l
->links
, &cgrp
->pidlists
);
3939 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3941 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
3942 struct cgroup_pidlist
**lp
)
3946 int pid
, n
= 0; /* used for populating the array */
3947 struct css_task_iter it
;
3948 struct task_struct
*tsk
;
3949 struct cgroup_pidlist
*l
;
3951 lockdep_assert_held(&cgrp
->pidlist_mutex
);
3954 * If cgroup gets more users after we read count, we won't have
3955 * enough space - tough. This race is indistinguishable to the
3956 * caller from the case that the additional cgroup users didn't
3957 * show up until sometime later on.
3959 length
= cgroup_task_count(cgrp
);
3960 array
= pidlist_allocate(length
);
3963 /* now, populate the array */
3964 css_task_iter_start(&cgrp
->self
, &it
);
3965 while ((tsk
= css_task_iter_next(&it
))) {
3966 if (unlikely(n
== length
))
3968 /* get tgid or pid for procs or tasks file respectively */
3969 if (type
== CGROUP_FILE_PROCS
)
3970 pid
= task_tgid_vnr(tsk
);
3972 pid
= task_pid_vnr(tsk
);
3973 if (pid
> 0) /* make sure to only use valid results */
3976 css_task_iter_end(&it
);
3978 /* now sort & (if procs) strip out duplicates */
3979 if (cgroup_on_dfl(cgrp
))
3980 sort(array
, length
, sizeof(pid_t
), fried_cmppid
, NULL
);
3982 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
3983 if (type
== CGROUP_FILE_PROCS
)
3984 length
= pidlist_uniq(array
, length
);
3986 l
= cgroup_pidlist_find_create(cgrp
, type
);
3988 mutex_unlock(&cgrp
->pidlist_mutex
);
3989 pidlist_free(array
);
3993 /* store array, freeing old if necessary */
3994 pidlist_free(l
->list
);
4002 * cgroupstats_build - build and fill cgroupstats
4003 * @stats: cgroupstats to fill information into
4004 * @dentry: A dentry entry belonging to the cgroup for which stats have
4007 * Build and fill cgroupstats so that taskstats can export it to user
4010 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
4012 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
4013 struct cgroup
*cgrp
;
4014 struct css_task_iter it
;
4015 struct task_struct
*tsk
;
4017 /* it should be kernfs_node belonging to cgroupfs and is a directory */
4018 if (dentry
->d_sb
->s_type
!= &cgroup_fs_type
|| !kn
||
4019 kernfs_type(kn
) != KERNFS_DIR
)
4022 mutex_lock(&cgroup_mutex
);
4025 * We aren't being called from kernfs and there's no guarantee on
4026 * @kn->priv's validity. For this and css_tryget_online_from_dir(),
4027 * @kn->priv is RCU safe. Let's do the RCU dancing.
4030 cgrp
= rcu_dereference(kn
->priv
);
4031 if (!cgrp
|| cgroup_is_dead(cgrp
)) {
4033 mutex_unlock(&cgroup_mutex
);
4038 css_task_iter_start(&cgrp
->self
, &it
);
4039 while ((tsk
= css_task_iter_next(&it
))) {
4040 switch (tsk
->state
) {
4042 stats
->nr_running
++;
4044 case TASK_INTERRUPTIBLE
:
4045 stats
->nr_sleeping
++;
4047 case TASK_UNINTERRUPTIBLE
:
4048 stats
->nr_uninterruptible
++;
4051 stats
->nr_stopped
++;
4054 if (delayacct_is_task_waiting_on_io(tsk
))
4055 stats
->nr_io_wait
++;
4059 css_task_iter_end(&it
);
4061 mutex_unlock(&cgroup_mutex
);
4067 * seq_file methods for the tasks/procs files. The seq_file position is the
4068 * next pid to display; the seq_file iterator is a pointer to the pid
4069 * in the cgroup->l->list array.
4072 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
4075 * Initially we receive a position value that corresponds to
4076 * one more than the last pid shown (or 0 on the first call or
4077 * after a seek to the start). Use a binary-search to find the
4078 * next pid to display, if any
4080 struct kernfs_open_file
*of
= s
->private;
4081 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
4082 struct cgroup_pidlist
*l
;
4083 enum cgroup_filetype type
= seq_cft(s
)->private;
4084 int index
= 0, pid
= *pos
;
4087 mutex_lock(&cgrp
->pidlist_mutex
);
4090 * !NULL @of->priv indicates that this isn't the first start()
4091 * after open. If the matching pidlist is around, we can use that.
4092 * Look for it. Note that @of->priv can't be used directly. It
4093 * could already have been destroyed.
4096 of
->priv
= cgroup_pidlist_find(cgrp
, type
);
4099 * Either this is the first start() after open or the matching
4100 * pidlist has been destroyed inbetween. Create a new one.
4103 ret
= pidlist_array_load(cgrp
, type
,
4104 (struct cgroup_pidlist
**)&of
->priv
);
4106 return ERR_PTR(ret
);
4111 int end
= l
->length
;
4113 while (index
< end
) {
4114 int mid
= (index
+ end
) / 2;
4115 if (cgroup_pid_fry(cgrp
, l
->list
[mid
]) == pid
) {
4118 } else if (cgroup_pid_fry(cgrp
, l
->list
[mid
]) <= pid
)
4124 /* If we're off the end of the array, we're done */
4125 if (index
>= l
->length
)
4127 /* Update the abstract position to be the actual pid that we found */
4128 iter
= l
->list
+ index
;
4129 *pos
= cgroup_pid_fry(cgrp
, *iter
);
4133 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
4135 struct kernfs_open_file
*of
= s
->private;
4136 struct cgroup_pidlist
*l
= of
->priv
;
4139 mod_delayed_work(cgroup_pidlist_destroy_wq
, &l
->destroy_dwork
,
4140 CGROUP_PIDLIST_DESTROY_DELAY
);
4141 mutex_unlock(&seq_css(s
)->cgroup
->pidlist_mutex
);
4144 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
4146 struct kernfs_open_file
*of
= s
->private;
4147 struct cgroup_pidlist
*l
= of
->priv
;
4149 pid_t
*end
= l
->list
+ l
->length
;
4151 * Advance to the next pid in the array. If this goes off the
4158 *pos
= cgroup_pid_fry(seq_css(s
)->cgroup
, *p
);
4163 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
4165 return seq_printf(s
, "%d\n", *(int *)v
);
4168 static u64
cgroup_read_notify_on_release(struct cgroup_subsys_state
*css
,
4171 return notify_on_release(css
->cgroup
);
4174 static int cgroup_write_notify_on_release(struct cgroup_subsys_state
*css
,
4175 struct cftype
*cft
, u64 val
)
4177 clear_bit(CGRP_RELEASABLE
, &css
->cgroup
->flags
);
4179 set_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
4181 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
4185 static u64
cgroup_clone_children_read(struct cgroup_subsys_state
*css
,
4188 return test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4191 static int cgroup_clone_children_write(struct cgroup_subsys_state
*css
,
4192 struct cftype
*cft
, u64 val
)
4195 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4197 clear_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4201 /* cgroup core interface files for the default hierarchy */
4202 static struct cftype cgroup_dfl_base_files
[] = {
4204 .name
= "cgroup.procs",
4205 .seq_start
= cgroup_pidlist_start
,
4206 .seq_next
= cgroup_pidlist_next
,
4207 .seq_stop
= cgroup_pidlist_stop
,
4208 .seq_show
= cgroup_pidlist_show
,
4209 .private = CGROUP_FILE_PROCS
,
4210 .write
= cgroup_procs_write
,
4211 .mode
= S_IRUGO
| S_IWUSR
,
4214 .name
= "cgroup.controllers",
4215 .flags
= CFTYPE_ONLY_ON_ROOT
,
4216 .seq_show
= cgroup_root_controllers_show
,
4219 .name
= "cgroup.controllers",
4220 .flags
= CFTYPE_NOT_ON_ROOT
,
4221 .seq_show
= cgroup_controllers_show
,
4224 .name
= "cgroup.subtree_control",
4225 .seq_show
= cgroup_subtree_control_show
,
4226 .write
= cgroup_subtree_control_write
,
4229 .name
= "cgroup.populated",
4230 .flags
= CFTYPE_NOT_ON_ROOT
,
4231 .seq_show
= cgroup_populated_show
,
4236 /* cgroup core interface files for the legacy hierarchies */
4237 static struct cftype cgroup_legacy_base_files
[] = {
4239 .name
= "cgroup.procs",
4240 .seq_start
= cgroup_pidlist_start
,
4241 .seq_next
= cgroup_pidlist_next
,
4242 .seq_stop
= cgroup_pidlist_stop
,
4243 .seq_show
= cgroup_pidlist_show
,
4244 .private = CGROUP_FILE_PROCS
,
4245 .write
= cgroup_procs_write
,
4246 .mode
= S_IRUGO
| S_IWUSR
,
4249 .name
= "cgroup.clone_children",
4250 .read_u64
= cgroup_clone_children_read
,
4251 .write_u64
= cgroup_clone_children_write
,
4254 .name
= "cgroup.sane_behavior",
4255 .flags
= CFTYPE_ONLY_ON_ROOT
,
4256 .seq_show
= cgroup_sane_behavior_show
,
4260 .seq_start
= cgroup_pidlist_start
,
4261 .seq_next
= cgroup_pidlist_next
,
4262 .seq_stop
= cgroup_pidlist_stop
,
4263 .seq_show
= cgroup_pidlist_show
,
4264 .private = CGROUP_FILE_TASKS
,
4265 .write
= cgroup_tasks_write
,
4266 .mode
= S_IRUGO
| S_IWUSR
,
4269 .name
= "notify_on_release",
4270 .read_u64
= cgroup_read_notify_on_release
,
4271 .write_u64
= cgroup_write_notify_on_release
,
4274 .name
= "release_agent",
4275 .flags
= CFTYPE_ONLY_ON_ROOT
,
4276 .seq_show
= cgroup_release_agent_show
,
4277 .write
= cgroup_release_agent_write
,
4278 .max_write_len
= PATH_MAX
- 1,
4284 * cgroup_populate_dir - create subsys files in a cgroup directory
4285 * @cgrp: target cgroup
4286 * @subsys_mask: mask of the subsystem ids whose files should be added
4288 * On failure, no file is added.
4290 static int cgroup_populate_dir(struct cgroup
*cgrp
, unsigned int subsys_mask
)
4292 struct cgroup_subsys
*ss
;
4295 /* process cftsets of each subsystem */
4296 for_each_subsys(ss
, i
) {
4297 struct cftype
*cfts
;
4299 if (!(subsys_mask
& (1 << i
)))
4302 list_for_each_entry(cfts
, &ss
->cfts
, node
) {
4303 ret
= cgroup_addrm_files(cgrp
, cfts
, true);
4310 cgroup_clear_dir(cgrp
, subsys_mask
);
4315 * css destruction is four-stage process.
4317 * 1. Destruction starts. Killing of the percpu_ref is initiated.
4318 * Implemented in kill_css().
4320 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4321 * and thus css_tryget_online() is guaranteed to fail, the css can be
4322 * offlined by invoking offline_css(). After offlining, the base ref is
4323 * put. Implemented in css_killed_work_fn().
4325 * 3. When the percpu_ref reaches zero, the only possible remaining
4326 * accessors are inside RCU read sections. css_release() schedules the
4329 * 4. After the grace period, the css can be freed. Implemented in
4330 * css_free_work_fn().
4332 * It is actually hairier because both step 2 and 4 require process context
4333 * and thus involve punting to css->destroy_work adding two additional
4334 * steps to the already complex sequence.
4336 static void css_free_work_fn(struct work_struct
*work
)
4338 struct cgroup_subsys_state
*css
=
4339 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4340 struct cgroup
*cgrp
= css
->cgroup
;
4342 percpu_ref_exit(&css
->refcnt
);
4347 css_put(css
->parent
);
4349 css
->ss
->css_free(css
);
4352 /* cgroup free path */
4353 atomic_dec(&cgrp
->root
->nr_cgrps
);
4354 cgroup_pidlist_destroy_all(cgrp
);
4356 if (cgroup_parent(cgrp
)) {
4358 * We get a ref to the parent, and put the ref when
4359 * this cgroup is being freed, so it's guaranteed
4360 * that the parent won't be destroyed before its
4363 cgroup_put(cgroup_parent(cgrp
));
4364 kernfs_put(cgrp
->kn
);
4368 * This is root cgroup's refcnt reaching zero,
4369 * which indicates that the root should be
4372 cgroup_destroy_root(cgrp
->root
);
4377 static void css_free_rcu_fn(struct rcu_head
*rcu_head
)
4379 struct cgroup_subsys_state
*css
=
4380 container_of(rcu_head
, struct cgroup_subsys_state
, rcu_head
);
4382 INIT_WORK(&css
->destroy_work
, css_free_work_fn
);
4383 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4386 static void css_release_work_fn(struct work_struct
*work
)
4388 struct cgroup_subsys_state
*css
=
4389 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4390 struct cgroup_subsys
*ss
= css
->ss
;
4391 struct cgroup
*cgrp
= css
->cgroup
;
4393 mutex_lock(&cgroup_mutex
);
4395 css
->flags
|= CSS_RELEASED
;
4396 list_del_rcu(&css
->sibling
);
4399 /* css release path */
4400 cgroup_idr_remove(&ss
->css_idr
, css
->id
);
4402 /* cgroup release path */
4403 cgroup_idr_remove(&cgrp
->root
->cgroup_idr
, cgrp
->id
);
4407 * There are two control paths which try to determine
4408 * cgroup from dentry without going through kernfs -
4409 * cgroupstats_build() and css_tryget_online_from_dir().
4410 * Those are supported by RCU protecting clearing of
4411 * cgrp->kn->priv backpointer.
4413 RCU_INIT_POINTER(*(void __rcu __force
**)&cgrp
->kn
->priv
, NULL
);
4416 mutex_unlock(&cgroup_mutex
);
4418 call_rcu(&css
->rcu_head
, css_free_rcu_fn
);
4421 static void css_release(struct percpu_ref
*ref
)
4423 struct cgroup_subsys_state
*css
=
4424 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4426 INIT_WORK(&css
->destroy_work
, css_release_work_fn
);
4427 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4430 static void init_and_link_css(struct cgroup_subsys_state
*css
,
4431 struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4433 lockdep_assert_held(&cgroup_mutex
);
4437 memset(css
, 0, sizeof(*css
));
4440 INIT_LIST_HEAD(&css
->sibling
);
4441 INIT_LIST_HEAD(&css
->children
);
4442 css
->serial_nr
= css_serial_nr_next
++;
4444 if (cgroup_parent(cgrp
)) {
4445 css
->parent
= cgroup_css(cgroup_parent(cgrp
), ss
);
4446 css_get(css
->parent
);
4449 BUG_ON(cgroup_css(cgrp
, ss
));
4452 /* invoke ->css_online() on a new CSS and mark it online if successful */
4453 static int online_css(struct cgroup_subsys_state
*css
)
4455 struct cgroup_subsys
*ss
= css
->ss
;
4458 lockdep_assert_held(&cgroup_mutex
);
4461 ret
= ss
->css_online(css
);
4463 css
->flags
|= CSS_ONLINE
;
4464 rcu_assign_pointer(css
->cgroup
->subsys
[ss
->id
], css
);
4469 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
4470 static void offline_css(struct cgroup_subsys_state
*css
)
4472 struct cgroup_subsys
*ss
= css
->ss
;
4474 lockdep_assert_held(&cgroup_mutex
);
4476 if (!(css
->flags
& CSS_ONLINE
))
4479 if (ss
->css_offline
)
4480 ss
->css_offline(css
);
4482 css
->flags
&= ~CSS_ONLINE
;
4483 RCU_INIT_POINTER(css
->cgroup
->subsys
[ss
->id
], NULL
);
4485 wake_up_all(&css
->cgroup
->offline_waitq
);
4489 * create_css - create a cgroup_subsys_state
4490 * @cgrp: the cgroup new css will be associated with
4491 * @ss: the subsys of new css
4492 * @visible: whether to create control knobs for the new css or not
4494 * Create a new css associated with @cgrp - @ss pair. On success, the new
4495 * css is online and installed in @cgrp with all interface files created if
4496 * @visible. Returns 0 on success, -errno on failure.
4498 static int create_css(struct cgroup
*cgrp
, struct cgroup_subsys
*ss
,
4501 struct cgroup
*parent
= cgroup_parent(cgrp
);
4502 struct cgroup_subsys_state
*parent_css
= cgroup_css(parent
, ss
);
4503 struct cgroup_subsys_state
*css
;
4506 lockdep_assert_held(&cgroup_mutex
);
4508 css
= ss
->css_alloc(parent_css
);
4510 return PTR_ERR(css
);
4512 init_and_link_css(css
, ss
, cgrp
);
4514 err
= percpu_ref_init(&css
->refcnt
, css_release
);
4518 err
= cgroup_idr_alloc(&ss
->css_idr
, NULL
, 2, 0, GFP_NOWAIT
);
4520 goto err_free_percpu_ref
;
4524 err
= cgroup_populate_dir(cgrp
, 1 << ss
->id
);
4529 /* @css is ready to be brought online now, make it visible */
4530 list_add_tail_rcu(&css
->sibling
, &parent_css
->children
);
4531 cgroup_idr_replace(&ss
->css_idr
, css
, css
->id
);
4533 err
= online_css(css
);
4537 if (ss
->broken_hierarchy
&& !ss
->warned_broken_hierarchy
&&
4538 cgroup_parent(parent
)) {
4539 pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
4540 current
->comm
, current
->pid
, ss
->name
);
4541 if (!strcmp(ss
->name
, "memory"))
4542 pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
4543 ss
->warned_broken_hierarchy
= true;
4549 list_del_rcu(&css
->sibling
);
4550 cgroup_clear_dir(css
->cgroup
, 1 << css
->ss
->id
);
4552 cgroup_idr_remove(&ss
->css_idr
, css
->id
);
4553 err_free_percpu_ref
:
4554 percpu_ref_exit(&css
->refcnt
);
4556 call_rcu(&css
->rcu_head
, css_free_rcu_fn
);
4560 static int cgroup_mkdir(struct kernfs_node
*parent_kn
, const char *name
,
4563 struct cgroup
*parent
, *cgrp
;
4564 struct cgroup_root
*root
;
4565 struct cgroup_subsys
*ss
;
4566 struct kernfs_node
*kn
;
4567 struct cftype
*base_files
;
4570 /* Do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable.
4572 if (strchr(name
, '\n'))
4575 parent
= cgroup_kn_lock_live(parent_kn
);
4578 root
= parent
->root
;
4580 /* allocate the cgroup and its ID, 0 is reserved for the root */
4581 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
4587 ret
= percpu_ref_init(&cgrp
->self
.refcnt
, css_release
);
4592 * Temporarily set the pointer to NULL, so idr_find() won't return
4593 * a half-baked cgroup.
4595 cgrp
->id
= cgroup_idr_alloc(&root
->cgroup_idr
, NULL
, 2, 0, GFP_NOWAIT
);
4598 goto out_cancel_ref
;
4601 init_cgroup_housekeeping(cgrp
);
4603 cgrp
->self
.parent
= &parent
->self
;
4606 if (notify_on_release(parent
))
4607 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
4609 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
4610 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4612 /* create the directory */
4613 kn
= kernfs_create_dir(parent
->kn
, name
, mode
, cgrp
);
4621 * This extra ref will be put in cgroup_free_fn() and guarantees
4622 * that @cgrp->kn is always accessible.
4626 cgrp
->self
.serial_nr
= css_serial_nr_next
++;
4628 /* allocation complete, commit to creation */
4629 list_add_tail_rcu(&cgrp
->self
.sibling
, &cgroup_parent(cgrp
)->self
.children
);
4630 atomic_inc(&root
->nr_cgrps
);
4634 * @cgrp is now fully operational. If something fails after this
4635 * point, it'll be released via the normal destruction path.
4637 cgroup_idr_replace(&root
->cgroup_idr
, cgrp
, cgrp
->id
);
4639 ret
= cgroup_kn_set_ugid(kn
);
4643 if (cgroup_on_dfl(cgrp
))
4644 base_files
= cgroup_dfl_base_files
;
4646 base_files
= cgroup_legacy_base_files
;
4648 ret
= cgroup_addrm_files(cgrp
, base_files
, true);
4652 /* let's create and online css's */
4653 for_each_subsys(ss
, ssid
) {
4654 if (parent
->child_subsys_mask
& (1 << ssid
)) {
4655 ret
= create_css(cgrp
, ss
,
4656 parent
->subtree_control
& (1 << ssid
));
4663 * On the default hierarchy, a child doesn't automatically inherit
4664 * subtree_control from the parent. Each is configured manually.
4666 if (!cgroup_on_dfl(cgrp
)) {
4667 cgrp
->subtree_control
= parent
->subtree_control
;
4668 cgroup_refresh_child_subsys_mask(cgrp
);
4671 kernfs_activate(kn
);
4677 cgroup_idr_remove(&root
->cgroup_idr
, cgrp
->id
);
4679 percpu_ref_exit(&cgrp
->self
.refcnt
);
4683 cgroup_kn_unlock(parent_kn
);
4687 cgroup_destroy_locked(cgrp
);
4692 * This is called when the refcnt of a css is confirmed to be killed.
4693 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
4694 * initate destruction and put the css ref from kill_css().
4696 static void css_killed_work_fn(struct work_struct
*work
)
4698 struct cgroup_subsys_state
*css
=
4699 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4701 mutex_lock(&cgroup_mutex
);
4703 mutex_unlock(&cgroup_mutex
);
4708 /* css kill confirmation processing requires process context, bounce */
4709 static void css_killed_ref_fn(struct percpu_ref
*ref
)
4711 struct cgroup_subsys_state
*css
=
4712 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4714 INIT_WORK(&css
->destroy_work
, css_killed_work_fn
);
4715 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4719 * kill_css - destroy a css
4720 * @css: css to destroy
4722 * This function initiates destruction of @css by removing cgroup interface
4723 * files and putting its base reference. ->css_offline() will be invoked
4724 * asynchronously once css_tryget_online() is guaranteed to fail and when
4725 * the reference count reaches zero, @css will be released.
4727 static void kill_css(struct cgroup_subsys_state
*css
)
4729 lockdep_assert_held(&cgroup_mutex
);
4732 * This must happen before css is disassociated with its cgroup.
4733 * See seq_css() for details.
4735 cgroup_clear_dir(css
->cgroup
, 1 << css
->ss
->id
);
4738 * Killing would put the base ref, but we need to keep it alive
4739 * until after ->css_offline().
4744 * cgroup core guarantees that, by the time ->css_offline() is
4745 * invoked, no new css reference will be given out via
4746 * css_tryget_online(). We can't simply call percpu_ref_kill() and
4747 * proceed to offlining css's because percpu_ref_kill() doesn't
4748 * guarantee that the ref is seen as killed on all CPUs on return.
4750 * Use percpu_ref_kill_and_confirm() to get notifications as each
4751 * css is confirmed to be seen as killed on all CPUs.
4753 percpu_ref_kill_and_confirm(&css
->refcnt
, css_killed_ref_fn
);
4757 * cgroup_destroy_locked - the first stage of cgroup destruction
4758 * @cgrp: cgroup to be destroyed
4760 * css's make use of percpu refcnts whose killing latency shouldn't be
4761 * exposed to userland and are RCU protected. Also, cgroup core needs to
4762 * guarantee that css_tryget_online() won't succeed by the time
4763 * ->css_offline() is invoked. To satisfy all the requirements,
4764 * destruction is implemented in the following two steps.
4766 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
4767 * userland visible parts and start killing the percpu refcnts of
4768 * css's. Set up so that the next stage will be kicked off once all
4769 * the percpu refcnts are confirmed to be killed.
4771 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
4772 * rest of destruction. Once all cgroup references are gone, the
4773 * cgroup is RCU-freed.
4775 * This function implements s1. After this step, @cgrp is gone as far as
4776 * the userland is concerned and a new cgroup with the same name may be
4777 * created. As cgroup doesn't care about the names internally, this
4778 * doesn't cause any problem.
4780 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
4781 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
4783 struct cgroup_subsys_state
*css
;
4787 lockdep_assert_held(&cgroup_mutex
);
4790 * css_set_rwsem synchronizes access to ->cset_links and prevents
4791 * @cgrp from being removed while put_css_set() is in progress.
4793 down_read(&css_set_rwsem
);
4794 empty
= list_empty(&cgrp
->cset_links
);
4795 up_read(&css_set_rwsem
);
4800 * Make sure there's no live children. We can't test emptiness of
4801 * ->self.children as dead children linger on it while being
4802 * drained; otherwise, "rmdir parent/child parent" may fail.
4804 if (css_has_online_children(&cgrp
->self
))
4808 * Mark @cgrp dead. This prevents further task migration and child
4809 * creation by disabling cgroup_lock_live_group().
4811 cgrp
->self
.flags
&= ~CSS_ONLINE
;
4813 /* initiate massacre of all css's */
4814 for_each_css(css
, ssid
, cgrp
)
4817 /* CSS_ONLINE is clear, remove from ->release_list for the last time */
4818 raw_spin_lock(&release_list_lock
);
4819 if (!list_empty(&cgrp
->release_list
))
4820 list_del_init(&cgrp
->release_list
);
4821 raw_spin_unlock(&release_list_lock
);
4824 * Remove @cgrp directory along with the base files. @cgrp has an
4825 * extra ref on its kn.
4827 kernfs_remove(cgrp
->kn
);
4829 set_bit(CGRP_RELEASABLE
, &cgroup_parent(cgrp
)->flags
);
4830 check_for_release(cgroup_parent(cgrp
));
4832 /* put the base reference */
4833 percpu_ref_kill(&cgrp
->self
.refcnt
);
4838 static int cgroup_rmdir(struct kernfs_node
*kn
)
4840 struct cgroup
*cgrp
;
4843 cgrp
= cgroup_kn_lock_live(kn
);
4846 cgroup_get(cgrp
); /* for @kn->priv clearing */
4848 ret
= cgroup_destroy_locked(cgrp
);
4850 cgroup_kn_unlock(kn
);
4856 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
= {
4857 .remount_fs
= cgroup_remount
,
4858 .show_options
= cgroup_show_options
,
4859 .mkdir
= cgroup_mkdir
,
4860 .rmdir
= cgroup_rmdir
,
4861 .rename
= cgroup_rename
,
4864 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
, bool early
)
4866 struct cgroup_subsys_state
*css
;
4868 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4870 mutex_lock(&cgroup_mutex
);
4872 idr_init(&ss
->css_idr
);
4873 INIT_LIST_HEAD(&ss
->cfts
);
4875 /* Create the root cgroup state for this subsystem */
4876 ss
->root
= &cgrp_dfl_root
;
4877 css
= ss
->css_alloc(cgroup_css(&cgrp_dfl_root
.cgrp
, ss
));
4878 /* We don't handle early failures gracefully */
4879 BUG_ON(IS_ERR(css
));
4880 init_and_link_css(css
, ss
, &cgrp_dfl_root
.cgrp
);
4883 * Root csses are never destroyed and we can't initialize
4884 * percpu_ref during early init. Disable refcnting.
4886 css
->flags
|= CSS_NO_REF
;
4889 /* allocation can't be done safely during early init */
4892 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2, GFP_KERNEL
);
4893 BUG_ON(css
->id
< 0);
4896 /* Update the init_css_set to contain a subsys
4897 * pointer to this state - since the subsystem is
4898 * newly registered, all tasks and hence the
4899 * init_css_set is in the subsystem's root cgroup. */
4900 init_css_set
.subsys
[ss
->id
] = css
;
4902 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
4904 /* At system boot, before all subsystems have been
4905 * registered, no tasks have been forked, so we don't
4906 * need to invoke fork callbacks here. */
4907 BUG_ON(!list_empty(&init_task
.tasks
));
4909 BUG_ON(online_css(css
));
4911 mutex_unlock(&cgroup_mutex
);
4915 * cgroup_init_early - cgroup initialization at system boot
4917 * Initialize cgroups at system boot, and initialize any
4918 * subsystems that request early init.
4920 int __init
cgroup_init_early(void)
4922 static struct cgroup_sb_opts __initdata opts
;
4923 struct cgroup_subsys
*ss
;
4926 init_cgroup_root(&cgrp_dfl_root
, &opts
);
4927 cgrp_dfl_root
.cgrp
.self
.flags
|= CSS_NO_REF
;
4929 RCU_INIT_POINTER(init_task
.cgroups
, &init_css_set
);
4931 for_each_subsys(ss
, i
) {
4932 WARN(!ss
->css_alloc
|| !ss
->css_free
|| ss
->name
|| ss
->id
,
4933 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n",
4934 i
, cgroup_subsys_name
[i
], ss
->css_alloc
, ss
->css_free
,
4936 WARN(strlen(cgroup_subsys_name
[i
]) > MAX_CGROUP_TYPE_NAMELEN
,
4937 "cgroup_subsys_name %s too long\n", cgroup_subsys_name
[i
]);
4940 ss
->name
= cgroup_subsys_name
[i
];
4943 cgroup_init_subsys(ss
, true);
4949 * cgroup_init - cgroup initialization
4951 * Register cgroup filesystem and /proc file, and initialize
4952 * any subsystems that didn't request early init.
4954 int __init
cgroup_init(void)
4956 struct cgroup_subsys
*ss
;
4960 BUG_ON(cgroup_init_cftypes(NULL
, cgroup_dfl_base_files
));
4961 BUG_ON(cgroup_init_cftypes(NULL
, cgroup_legacy_base_files
));
4963 mutex_lock(&cgroup_mutex
);
4965 /* Add init_css_set to the hash table */
4966 key
= css_set_hash(init_css_set
.subsys
);
4967 hash_add(css_set_table
, &init_css_set
.hlist
, key
);
4969 BUG_ON(cgroup_setup_root(&cgrp_dfl_root
, 0));
4971 mutex_unlock(&cgroup_mutex
);
4973 for_each_subsys(ss
, ssid
) {
4974 if (ss
->early_init
) {
4975 struct cgroup_subsys_state
*css
=
4976 init_css_set
.subsys
[ss
->id
];
4978 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2,
4980 BUG_ON(css
->id
< 0);
4982 cgroup_init_subsys(ss
, false);
4985 list_add_tail(&init_css_set
.e_cset_node
[ssid
],
4986 &cgrp_dfl_root
.cgrp
.e_csets
[ssid
]);
4989 * Setting dfl_root subsys_mask needs to consider the
4990 * disabled flag and cftype registration needs kmalloc,
4991 * both of which aren't available during early_init.
4996 cgrp_dfl_root
.subsys_mask
|= 1 << ss
->id
;
4998 if (cgroup_legacy_files_on_dfl
&& !ss
->dfl_cftypes
)
4999 ss
->dfl_cftypes
= ss
->legacy_cftypes
;
5001 if (!ss
->dfl_cftypes
)
5002 cgrp_dfl_root_inhibit_ss_mask
|= 1 << ss
->id
;
5004 if (ss
->dfl_cftypes
== ss
->legacy_cftypes
) {
5005 WARN_ON(cgroup_add_cftypes(ss
, ss
->dfl_cftypes
));
5007 WARN_ON(cgroup_add_dfl_cftypes(ss
, ss
->dfl_cftypes
));
5008 WARN_ON(cgroup_add_legacy_cftypes(ss
, ss
->legacy_cftypes
));
5012 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
5016 err
= register_filesystem(&cgroup_fs_type
);
5018 kobject_put(cgroup_kobj
);
5022 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
5026 static int __init
cgroup_wq_init(void)
5029 * There isn't much point in executing destruction path in
5030 * parallel. Good chunk is serialized with cgroup_mutex anyway.
5031 * Use 1 for @max_active.
5033 * We would prefer to do this in cgroup_init() above, but that
5034 * is called before init_workqueues(): so leave this until after.
5036 cgroup_destroy_wq
= alloc_workqueue("cgroup_destroy", 0, 1);
5037 BUG_ON(!cgroup_destroy_wq
);
5040 * Used to destroy pidlists and separate to serve as flush domain.
5041 * Cap @max_active to 1 too.
5043 cgroup_pidlist_destroy_wq
= alloc_workqueue("cgroup_pidlist_destroy",
5045 BUG_ON(!cgroup_pidlist_destroy_wq
);
5049 core_initcall(cgroup_wq_init
);
5052 * proc_cgroup_show()
5053 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5054 * - Used for /proc/<pid>/cgroup.
5057 /* TODO: Use a proper seq_file iterator */
5058 int proc_cgroup_show(struct seq_file
*m
, void *v
)
5061 struct task_struct
*tsk
;
5064 struct cgroup_root
*root
;
5067 buf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
5073 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
5079 mutex_lock(&cgroup_mutex
);
5080 down_read(&css_set_rwsem
);
5082 for_each_root(root
) {
5083 struct cgroup_subsys
*ss
;
5084 struct cgroup
*cgrp
;
5085 int ssid
, count
= 0;
5087 if (root
== &cgrp_dfl_root
&& !cgrp_dfl_root_visible
)
5090 seq_printf(m
, "%d:", root
->hierarchy_id
);
5091 for_each_subsys(ss
, ssid
)
5092 if (root
->subsys_mask
& (1 << ssid
))
5093 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
5094 if (strlen(root
->name
))
5095 seq_printf(m
, "%sname=%s", count
? "," : "",
5098 cgrp
= task_cgroup_from_root(tsk
, root
);
5099 path
= cgroup_path(cgrp
, buf
, PATH_MAX
);
5101 retval
= -ENAMETOOLONG
;
5109 up_read(&css_set_rwsem
);
5110 mutex_unlock(&cgroup_mutex
);
5111 put_task_struct(tsk
);
5118 /* Display information about each subsystem and each hierarchy */
5119 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
5121 struct cgroup_subsys
*ss
;
5124 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
5126 * ideally we don't want subsystems moving around while we do this.
5127 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
5128 * subsys/hierarchy state.
5130 mutex_lock(&cgroup_mutex
);
5132 for_each_subsys(ss
, i
)
5133 seq_printf(m
, "%s\t%d\t%d\t%d\n",
5134 ss
->name
, ss
->root
->hierarchy_id
,
5135 atomic_read(&ss
->root
->nr_cgrps
), !ss
->disabled
);
5137 mutex_unlock(&cgroup_mutex
);
5141 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
5143 return single_open(file
, proc_cgroupstats_show
, NULL
);
5146 static const struct file_operations proc_cgroupstats_operations
= {
5147 .open
= cgroupstats_open
,
5149 .llseek
= seq_lseek
,
5150 .release
= single_release
,
5154 * cgroup_fork - initialize cgroup related fields during copy_process()
5155 * @child: pointer to task_struct of forking parent process.
5157 * A task is associated with the init_css_set until cgroup_post_fork()
5158 * attaches it to the parent's css_set. Empty cg_list indicates that
5159 * @child isn't holding reference to its css_set.
5161 void cgroup_fork(struct task_struct
*child
)
5163 RCU_INIT_POINTER(child
->cgroups
, &init_css_set
);
5164 INIT_LIST_HEAD(&child
->cg_list
);
5168 * cgroup_post_fork - called on a new task after adding it to the task list
5169 * @child: the task in question
5171 * Adds the task to the list running through its css_set if necessary and
5172 * call the subsystem fork() callbacks. Has to be after the task is
5173 * visible on the task list in case we race with the first call to
5174 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
5177 void cgroup_post_fork(struct task_struct
*child
)
5179 struct cgroup_subsys
*ss
;
5183 * This may race against cgroup_enable_task_cg_links(). As that
5184 * function sets use_task_css_set_links before grabbing
5185 * tasklist_lock and we just went through tasklist_lock to add
5186 * @child, it's guaranteed that either we see the set
5187 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
5188 * @child during its iteration.
5190 * If we won the race, @child is associated with %current's
5191 * css_set. Grabbing css_set_rwsem guarantees both that the
5192 * association is stable, and, on completion of the parent's
5193 * migration, @child is visible in the source of migration or
5194 * already in the destination cgroup. This guarantee is necessary
5195 * when implementing operations which need to migrate all tasks of
5196 * a cgroup to another.
5198 * Note that if we lose to cgroup_enable_task_cg_links(), @child
5199 * will remain in init_css_set. This is safe because all tasks are
5200 * in the init_css_set before cg_links is enabled and there's no
5201 * operation which transfers all tasks out of init_css_set.
5203 if (use_task_css_set_links
) {
5204 struct css_set
*cset
;
5206 down_write(&css_set_rwsem
);
5207 cset
= task_css_set(current
);
5208 if (list_empty(&child
->cg_list
)) {
5209 rcu_assign_pointer(child
->cgroups
, cset
);
5210 list_add(&child
->cg_list
, &cset
->tasks
);
5213 up_write(&css_set_rwsem
);
5217 * Call ss->fork(). This must happen after @child is linked on
5218 * css_set; otherwise, @child might change state between ->fork()
5219 * and addition to css_set.
5221 if (need_forkexit_callback
) {
5222 for_each_subsys(ss
, i
)
5229 * cgroup_exit - detach cgroup from exiting task
5230 * @tsk: pointer to task_struct of exiting process
5232 * Description: Detach cgroup from @tsk and release it.
5234 * Note that cgroups marked notify_on_release force every task in
5235 * them to take the global cgroup_mutex mutex when exiting.
5236 * This could impact scaling on very large systems. Be reluctant to
5237 * use notify_on_release cgroups where very high task exit scaling
5238 * is required on large systems.
5240 * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
5241 * call cgroup_exit() while the task is still competent to handle
5242 * notify_on_release(), then leave the task attached to the root cgroup in
5243 * each hierarchy for the remainder of its exit. No need to bother with
5244 * init_css_set refcnting. init_css_set never goes away and we can't race
5245 * with migration path - PF_EXITING is visible to migration path.
5247 void cgroup_exit(struct task_struct
*tsk
)
5249 struct cgroup_subsys
*ss
;
5250 struct css_set
*cset
;
5251 bool put_cset
= false;
5255 * Unlink from @tsk from its css_set. As migration path can't race
5256 * with us, we can check cg_list without grabbing css_set_rwsem.
5258 if (!list_empty(&tsk
->cg_list
)) {
5259 down_write(&css_set_rwsem
);
5260 list_del_init(&tsk
->cg_list
);
5261 up_write(&css_set_rwsem
);
5265 /* Reassign the task to the init_css_set. */
5266 cset
= task_css_set(tsk
);
5267 RCU_INIT_POINTER(tsk
->cgroups
, &init_css_set
);
5269 if (need_forkexit_callback
) {
5270 /* see cgroup_post_fork() for details */
5271 for_each_subsys(ss
, i
) {
5273 struct cgroup_subsys_state
*old_css
= cset
->subsys
[i
];
5274 struct cgroup_subsys_state
*css
= task_css(tsk
, i
);
5276 ss
->exit(css
, old_css
, tsk
);
5282 put_css_set(cset
, true);
5285 static void check_for_release(struct cgroup
*cgrp
)
5287 if (cgroup_is_releasable(cgrp
) && list_empty(&cgrp
->cset_links
) &&
5288 !css_has_online_children(&cgrp
->self
)) {
5290 * Control Group is currently removeable. If it's not
5291 * already queued for a userspace notification, queue
5294 int need_schedule_work
= 0;
5296 raw_spin_lock(&release_list_lock
);
5297 if (!cgroup_is_dead(cgrp
) &&
5298 list_empty(&cgrp
->release_list
)) {
5299 list_add(&cgrp
->release_list
, &release_list
);
5300 need_schedule_work
= 1;
5302 raw_spin_unlock(&release_list_lock
);
5303 if (need_schedule_work
)
5304 schedule_work(&release_agent_work
);
5309 * Notify userspace when a cgroup is released, by running the
5310 * configured release agent with the name of the cgroup (path
5311 * relative to the root of cgroup file system) as the argument.
5313 * Most likely, this user command will try to rmdir this cgroup.
5315 * This races with the possibility that some other task will be
5316 * attached to this cgroup before it is removed, or that some other
5317 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5318 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5319 * unused, and this cgroup will be reprieved from its death sentence,
5320 * to continue to serve a useful existence. Next time it's released,
5321 * we will get notified again, if it still has 'notify_on_release' set.
5323 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5324 * means only wait until the task is successfully execve()'d. The
5325 * separate release agent task is forked by call_usermodehelper(),
5326 * then control in this thread returns here, without waiting for the
5327 * release agent task. We don't bother to wait because the caller of
5328 * this routine has no use for the exit status of the release agent
5329 * task, so no sense holding our caller up for that.
5331 static void cgroup_release_agent(struct work_struct
*work
)
5333 BUG_ON(work
!= &release_agent_work
);
5334 mutex_lock(&cgroup_mutex
);
5335 raw_spin_lock(&release_list_lock
);
5336 while (!list_empty(&release_list
)) {
5337 char *argv
[3], *envp
[3];
5339 char *pathbuf
= NULL
, *agentbuf
= NULL
, *path
;
5340 struct cgroup
*cgrp
= list_entry(release_list
.next
,
5343 list_del_init(&cgrp
->release_list
);
5344 raw_spin_unlock(&release_list_lock
);
5345 pathbuf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
5348 path
= cgroup_path(cgrp
, pathbuf
, PATH_MAX
);
5351 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
5356 argv
[i
++] = agentbuf
;
5361 /* minimal command environment */
5362 envp
[i
++] = "HOME=/";
5363 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5366 /* Drop the lock while we invoke the usermode helper,
5367 * since the exec could involve hitting disk and hence
5368 * be a slow process */
5369 mutex_unlock(&cgroup_mutex
);
5370 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
5371 mutex_lock(&cgroup_mutex
);
5375 raw_spin_lock(&release_list_lock
);
5377 raw_spin_unlock(&release_list_lock
);
5378 mutex_unlock(&cgroup_mutex
);
5381 static int __init
cgroup_disable(char *str
)
5383 struct cgroup_subsys
*ss
;
5387 while ((token
= strsep(&str
, ",")) != NULL
) {
5391 for_each_subsys(ss
, i
) {
5392 if (!strcmp(token
, ss
->name
)) {
5394 printk(KERN_INFO
"Disabling %s control group"
5395 " subsystem\n", ss
->name
);
5402 __setup("cgroup_disable=", cgroup_disable
);
5404 static int __init
cgroup_set_legacy_files_on_dfl(char *str
)
5406 printk("cgroup: using legacy files on the default hierarchy\n");
5407 cgroup_legacy_files_on_dfl
= true;
5410 __setup("cgroup__DEVEL__legacy_files_on_dfl", cgroup_set_legacy_files_on_dfl
);
5413 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
5414 * @dentry: directory dentry of interest
5415 * @ss: subsystem of interest
5417 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
5418 * to get the corresponding css and return it. If such css doesn't exist
5419 * or can't be pinned, an ERR_PTR value is returned.
5421 struct cgroup_subsys_state
*css_tryget_online_from_dir(struct dentry
*dentry
,
5422 struct cgroup_subsys
*ss
)
5424 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
5425 struct cgroup_subsys_state
*css
= NULL
;
5426 struct cgroup
*cgrp
;
5428 /* is @dentry a cgroup dir? */
5429 if (dentry
->d_sb
->s_type
!= &cgroup_fs_type
|| !kn
||
5430 kernfs_type(kn
) != KERNFS_DIR
)
5431 return ERR_PTR(-EBADF
);
5436 * This path doesn't originate from kernfs and @kn could already
5437 * have been or be removed at any point. @kn->priv is RCU
5438 * protected for this access. See css_release_work_fn() for details.
5440 cgrp
= rcu_dereference(kn
->priv
);
5442 css
= cgroup_css(cgrp
, ss
);
5444 if (!css
|| !css_tryget_online(css
))
5445 css
= ERR_PTR(-ENOENT
);
5452 * css_from_id - lookup css by id
5453 * @id: the cgroup id
5454 * @ss: cgroup subsys to be looked into
5456 * Returns the css if there's valid one with @id, otherwise returns NULL.
5457 * Should be called under rcu_read_lock().
5459 struct cgroup_subsys_state
*css_from_id(int id
, struct cgroup_subsys
*ss
)
5461 WARN_ON_ONCE(!rcu_read_lock_held());
5462 return idr_find(&ss
->css_idr
, id
);
5465 #ifdef CONFIG_CGROUP_DEBUG
5466 static struct cgroup_subsys_state
*
5467 debug_css_alloc(struct cgroup_subsys_state
*parent_css
)
5469 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5472 return ERR_PTR(-ENOMEM
);
5477 static void debug_css_free(struct cgroup_subsys_state
*css
)
5482 static u64
debug_taskcount_read(struct cgroup_subsys_state
*css
,
5485 return cgroup_task_count(css
->cgroup
);
5488 static u64
current_css_set_read(struct cgroup_subsys_state
*css
,
5491 return (u64
)(unsigned long)current
->cgroups
;
5494 static u64
current_css_set_refcount_read(struct cgroup_subsys_state
*css
,
5500 count
= atomic_read(&task_css_set(current
)->refcount
);
5505 static int current_css_set_cg_links_read(struct seq_file
*seq
, void *v
)
5507 struct cgrp_cset_link
*link
;
5508 struct css_set
*cset
;
5511 name_buf
= kmalloc(NAME_MAX
+ 1, GFP_KERNEL
);
5515 down_read(&css_set_rwsem
);
5517 cset
= rcu_dereference(current
->cgroups
);
5518 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
5519 struct cgroup
*c
= link
->cgrp
;
5521 cgroup_name(c
, name_buf
, NAME_MAX
+ 1);
5522 seq_printf(seq
, "Root %d group %s\n",
5523 c
->root
->hierarchy_id
, name_buf
);
5526 up_read(&css_set_rwsem
);
5531 #define MAX_TASKS_SHOWN_PER_CSS 25
5532 static int cgroup_css_links_read(struct seq_file
*seq
, void *v
)
5534 struct cgroup_subsys_state
*css
= seq_css(seq
);
5535 struct cgrp_cset_link
*link
;
5537 down_read(&css_set_rwsem
);
5538 list_for_each_entry(link
, &css
->cgroup
->cset_links
, cset_link
) {
5539 struct css_set
*cset
= link
->cset
;
5540 struct task_struct
*task
;
5543 seq_printf(seq
, "css_set %p\n", cset
);
5545 list_for_each_entry(task
, &cset
->tasks
, cg_list
) {
5546 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
)
5548 seq_printf(seq
, " task %d\n", task_pid_vnr(task
));
5551 list_for_each_entry(task
, &cset
->mg_tasks
, cg_list
) {
5552 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
)
5554 seq_printf(seq
, " task %d\n", task_pid_vnr(task
));
5558 seq_puts(seq
, " ...\n");
5560 up_read(&css_set_rwsem
);
5564 static u64
releasable_read(struct cgroup_subsys_state
*css
, struct cftype
*cft
)
5566 return test_bit(CGRP_RELEASABLE
, &css
->cgroup
->flags
);
5569 static struct cftype debug_files
[] = {
5571 .name
= "taskcount",
5572 .read_u64
= debug_taskcount_read
,
5576 .name
= "current_css_set",
5577 .read_u64
= current_css_set_read
,
5581 .name
= "current_css_set_refcount",
5582 .read_u64
= current_css_set_refcount_read
,
5586 .name
= "current_css_set_cg_links",
5587 .seq_show
= current_css_set_cg_links_read
,
5591 .name
= "cgroup_css_links",
5592 .seq_show
= cgroup_css_links_read
,
5596 .name
= "releasable",
5597 .read_u64
= releasable_read
,
5603 struct cgroup_subsys debug_cgrp_subsys
= {
5604 .css_alloc
= debug_css_alloc
,
5605 .css_free
= debug_css_free
,
5606 .legacy_cftypes
= debug_files
,
5608 #endif /* CONFIG_CGROUP_DEBUG */