2 * Common Block IO controller cgroup interface
4 * Based on ideas and code from CFQ, CFS and BFQ:
5 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
7 * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
8 * Paolo Valente <paolo.valente@unimore.it>
10 * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
11 * Nauman Rafique <nauman@google.com>
13 #include <linux/ioprio.h>
14 #include <linux/kdev_t.h>
15 #include <linux/module.h>
16 #include <linux/err.h>
17 #include <linux/blkdev.h>
18 #include <linux/slab.h>
19 #include <linux/genhd.h>
20 #include <linux/delay.h>
21 #include <linux/atomic.h>
22 #include "blk-cgroup.h"
25 #define MAX_KEY_LEN 100
27 static DEFINE_MUTEX(blkcg_pol_mutex
);
29 struct blkcg blkcg_root
= { .cfq_weight
= 2 * CFQ_WEIGHT_DEFAULT
};
30 EXPORT_SYMBOL_GPL(blkcg_root
);
32 static struct blkcg_policy
*blkcg_policy
[BLKCG_MAX_POLS
];
34 static bool blkcg_policy_enabled(struct request_queue
*q
,
35 const struct blkcg_policy
*pol
)
37 return pol
&& test_bit(pol
->plid
, q
->blkcg_pols
);
41 * blkg_free - free a blkg
44 * Free @blkg which may be partially allocated.
46 static void blkg_free(struct blkcg_gq
*blkg
)
53 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
54 struct blkcg_policy
*pol
= blkcg_policy
[i
];
55 struct blkg_policy_data
*pd
= blkg
->pd
[i
];
60 if (pol
&& pol
->pd_exit_fn
)
61 pol
->pd_exit_fn(blkg
);
70 * blkg_alloc - allocate a blkg
71 * @blkcg: block cgroup the new blkg is associated with
72 * @q: request_queue the new blkg is associated with
73 * @gfp_mask: allocation mask to use
75 * Allocate a new blkg assocating @blkcg and @q.
77 static struct blkcg_gq
*blkg_alloc(struct blkcg
*blkcg
, struct request_queue
*q
,
80 struct blkcg_gq
*blkg
;
83 /* alloc and init base part */
84 blkg
= kzalloc_node(sizeof(*blkg
), gfp_mask
, q
->node
);
89 INIT_LIST_HEAD(&blkg
->q_node
);
93 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
94 struct blkcg_policy
*pol
= blkcg_policy
[i
];
95 struct blkg_policy_data
*pd
;
97 if (!blkcg_policy_enabled(q
, pol
))
100 /* alloc per-policy data and attach it to blkg */
101 pd
= kzalloc_node(pol
->pd_size
, gfp_mask
, q
->node
);
110 /* invoke per-policy init */
111 if (blkcg_policy_enabled(blkg
->q
, pol
))
112 pol
->pd_init_fn(blkg
);
118 static struct blkcg_gq
*__blkg_lookup(struct blkcg
*blkcg
,
119 struct request_queue
*q
)
121 struct blkcg_gq
*blkg
;
123 blkg
= rcu_dereference(blkcg
->blkg_hint
);
124 if (blkg
&& blkg
->q
== q
)
128 * Hint didn't match. Look up from the radix tree. Note that we
129 * may not be holding queue_lock and thus are not sure whether
130 * @blkg from blkg_tree has already been removed or not, so we
131 * can't update hint to the lookup result. Leave it to the caller.
133 blkg
= radix_tree_lookup(&blkcg
->blkg_tree
, q
->id
);
134 if (blkg
&& blkg
->q
== q
)
141 * blkg_lookup - lookup blkg for the specified blkcg - q pair
142 * @blkcg: blkcg of interest
143 * @q: request_queue of interest
145 * Lookup blkg for the @blkcg - @q pair. This function should be called
146 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
147 * - see blk_queue_bypass_start() for details.
149 struct blkcg_gq
*blkg_lookup(struct blkcg
*blkcg
, struct request_queue
*q
)
151 WARN_ON_ONCE(!rcu_read_lock_held());
153 if (unlikely(blk_queue_bypass(q
)))
155 return __blkg_lookup(blkcg
, q
);
157 EXPORT_SYMBOL_GPL(blkg_lookup
);
160 * If @new_blkg is %NULL, this function tries to allocate a new one as
161 * necessary using %GFP_ATOMIC. @new_blkg is always consumed on return.
163 static struct blkcg_gq
*__blkg_lookup_create(struct blkcg
*blkcg
,
164 struct request_queue
*q
,
165 struct blkcg_gq
*new_blkg
)
167 struct blkcg_gq
*blkg
;
170 WARN_ON_ONCE(!rcu_read_lock_held());
171 lockdep_assert_held(q
->queue_lock
);
173 /* lookup and update hint on success, see __blkg_lookup() for details */
174 blkg
= __blkg_lookup(blkcg
, q
);
176 rcu_assign_pointer(blkcg
->blkg_hint
, blkg
);
180 /* blkg holds a reference to blkcg */
181 if (!css_tryget(&blkcg
->css
)) {
182 blkg
= ERR_PTR(-EINVAL
);
188 new_blkg
= blkg_alloc(blkcg
, q
, GFP_ATOMIC
);
189 if (unlikely(!new_blkg
)) {
190 blkg
= ERR_PTR(-ENOMEM
);
197 spin_lock(&blkcg
->lock
);
198 ret
= radix_tree_insert(&blkcg
->blkg_tree
, q
->id
, blkg
);
200 hlist_add_head_rcu(&blkg
->blkcg_node
, &blkcg
->blkg_list
);
201 list_add(&blkg
->q_node
, &q
->blkg_list
);
203 spin_unlock(&blkcg
->lock
);
210 css_put(&blkcg
->css
);
216 struct blkcg_gq
*blkg_lookup_create(struct blkcg
*blkcg
,
217 struct request_queue
*q
)
220 * This could be the first entry point of blkcg implementation and
221 * we shouldn't allow anything to go through for a bypassing queue.
223 if (unlikely(blk_queue_bypass(q
)))
224 return ERR_PTR(blk_queue_dead(q
) ? -EINVAL
: -EBUSY
);
225 return __blkg_lookup_create(blkcg
, q
, NULL
);
227 EXPORT_SYMBOL_GPL(blkg_lookup_create
);
229 static void blkg_destroy(struct blkcg_gq
*blkg
)
231 struct blkcg
*blkcg
= blkg
->blkcg
;
233 lockdep_assert_held(blkg
->q
->queue_lock
);
234 lockdep_assert_held(&blkcg
->lock
);
236 /* Something wrong if we are trying to remove same group twice */
237 WARN_ON_ONCE(list_empty(&blkg
->q_node
));
238 WARN_ON_ONCE(hlist_unhashed(&blkg
->blkcg_node
));
240 radix_tree_delete(&blkcg
->blkg_tree
, blkg
->q
->id
);
241 list_del_init(&blkg
->q_node
);
242 hlist_del_init_rcu(&blkg
->blkcg_node
);
245 * Both setting lookup hint to and clearing it from @blkg are done
246 * under queue_lock. If it's not pointing to @blkg now, it never
247 * will. Hint assignment itself can race safely.
249 if (rcu_dereference_raw(blkcg
->blkg_hint
) == blkg
)
250 rcu_assign_pointer(blkcg
->blkg_hint
, NULL
);
253 * Put the reference taken at the time of creation so that when all
254 * queues are gone, group can be destroyed.
260 * blkg_destroy_all - destroy all blkgs associated with a request_queue
261 * @q: request_queue of interest
263 * Destroy all blkgs associated with @q.
265 static void blkg_destroy_all(struct request_queue
*q
)
267 struct blkcg_gq
*blkg
, *n
;
269 lockdep_assert_held(q
->queue_lock
);
271 list_for_each_entry_safe(blkg
, n
, &q
->blkg_list
, q_node
) {
272 struct blkcg
*blkcg
= blkg
->blkcg
;
274 spin_lock(&blkcg
->lock
);
276 spin_unlock(&blkcg
->lock
);
280 static void blkg_rcu_free(struct rcu_head
*rcu_head
)
282 blkg_free(container_of(rcu_head
, struct blkcg_gq
, rcu_head
));
285 void __blkg_release(struct blkcg_gq
*blkg
)
287 /* release the extra blkcg reference this blkg has been holding */
288 css_put(&blkg
->blkcg
->css
);
291 * A group is freed in rcu manner. But having an rcu lock does not
292 * mean that one can access all the fields of blkg and assume these
293 * are valid. For example, don't try to follow throtl_data and
294 * request queue links.
296 * Having a reference to blkg under an rcu allows acess to only
297 * values local to groups like group stats and group rate limits
299 call_rcu(&blkg
->rcu_head
, blkg_rcu_free
);
301 EXPORT_SYMBOL_GPL(__blkg_release
);
303 static int blkcg_reset_stats(struct cgroup
*cgroup
, struct cftype
*cftype
,
306 struct blkcg
*blkcg
= cgroup_to_blkcg(cgroup
);
307 struct blkcg_gq
*blkg
;
308 struct hlist_node
*n
;
311 mutex_lock(&blkcg_pol_mutex
);
312 spin_lock_irq(&blkcg
->lock
);
315 * Note that stat reset is racy - it doesn't synchronize against
316 * stat updates. This is a debug feature which shouldn't exist
317 * anyway. If you get hit by a race, retry.
319 hlist_for_each_entry(blkg
, n
, &blkcg
->blkg_list
, blkcg_node
) {
320 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
321 struct blkcg_policy
*pol
= blkcg_policy
[i
];
323 if (blkcg_policy_enabled(blkg
->q
, pol
) &&
324 pol
->pd_reset_stats_fn
)
325 pol
->pd_reset_stats_fn(blkg
);
329 spin_unlock_irq(&blkcg
->lock
);
330 mutex_unlock(&blkcg_pol_mutex
);
334 static const char *blkg_dev_name(struct blkcg_gq
*blkg
)
336 /* some drivers (floppy) instantiate a queue w/o disk registered */
337 if (blkg
->q
->backing_dev_info
.dev
)
338 return dev_name(blkg
->q
->backing_dev_info
.dev
);
343 * blkcg_print_blkgs - helper for printing per-blkg data
344 * @sf: seq_file to print to
345 * @blkcg: blkcg of interest
346 * @prfill: fill function to print out a blkg
347 * @pol: policy in question
348 * @data: data to be passed to @prfill
349 * @show_total: to print out sum of prfill return values or not
351 * This function invokes @prfill on each blkg of @blkcg if pd for the
352 * policy specified by @pol exists. @prfill is invoked with @sf, the
353 * policy data and @data. If @show_total is %true, the sum of the return
354 * values from @prfill is printed with "Total" label at the end.
356 * This is to be used to construct print functions for
357 * cftype->read_seq_string method.
359 void blkcg_print_blkgs(struct seq_file
*sf
, struct blkcg
*blkcg
,
360 u64 (*prfill
)(struct seq_file
*,
361 struct blkg_policy_data
*, int),
362 const struct blkcg_policy
*pol
, int data
,
365 struct blkcg_gq
*blkg
;
366 struct hlist_node
*n
;
369 spin_lock_irq(&blkcg
->lock
);
370 hlist_for_each_entry(blkg
, n
, &blkcg
->blkg_list
, blkcg_node
)
371 if (blkcg_policy_enabled(blkg
->q
, pol
))
372 total
+= prfill(sf
, blkg
->pd
[pol
->plid
], data
);
373 spin_unlock_irq(&blkcg
->lock
);
376 seq_printf(sf
, "Total %llu\n", (unsigned long long)total
);
378 EXPORT_SYMBOL_GPL(blkcg_print_blkgs
);
381 * __blkg_prfill_u64 - prfill helper for a single u64 value
382 * @sf: seq_file to print to
383 * @pd: policy private data of interest
386 * Print @v to @sf for the device assocaited with @pd.
388 u64
__blkg_prfill_u64(struct seq_file
*sf
, struct blkg_policy_data
*pd
, u64 v
)
390 const char *dname
= blkg_dev_name(pd
->blkg
);
395 seq_printf(sf
, "%s %llu\n", dname
, (unsigned long long)v
);
398 EXPORT_SYMBOL_GPL(__blkg_prfill_u64
);
401 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
402 * @sf: seq_file to print to
403 * @pd: policy private data of interest
404 * @rwstat: rwstat to print
406 * Print @rwstat to @sf for the device assocaited with @pd.
408 u64
__blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
409 const struct blkg_rwstat
*rwstat
)
411 static const char *rwstr
[] = {
412 [BLKG_RWSTAT_READ
] = "Read",
413 [BLKG_RWSTAT_WRITE
] = "Write",
414 [BLKG_RWSTAT_SYNC
] = "Sync",
415 [BLKG_RWSTAT_ASYNC
] = "Async",
417 const char *dname
= blkg_dev_name(pd
->blkg
);
424 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
425 seq_printf(sf
, "%s %s %llu\n", dname
, rwstr
[i
],
426 (unsigned long long)rwstat
->cnt
[i
]);
428 v
= rwstat
->cnt
[BLKG_RWSTAT_READ
] + rwstat
->cnt
[BLKG_RWSTAT_WRITE
];
429 seq_printf(sf
, "%s Total %llu\n", dname
, (unsigned long long)v
);
434 * blkg_prfill_stat - prfill callback for blkg_stat
435 * @sf: seq_file to print to
436 * @pd: policy private data of interest
437 * @off: offset to the blkg_stat in @pd
439 * prfill callback for printing a blkg_stat.
441 u64
blkg_prfill_stat(struct seq_file
*sf
, struct blkg_policy_data
*pd
, int off
)
443 return __blkg_prfill_u64(sf
, pd
, blkg_stat_read((void *)pd
+ off
));
445 EXPORT_SYMBOL_GPL(blkg_prfill_stat
);
448 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
449 * @sf: seq_file to print to
450 * @pd: policy private data of interest
451 * @off: offset to the blkg_rwstat in @pd
453 * prfill callback for printing a blkg_rwstat.
455 u64
blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
458 struct blkg_rwstat rwstat
= blkg_rwstat_read((void *)pd
+ off
);
460 return __blkg_prfill_rwstat(sf
, pd
, &rwstat
);
462 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat
);
465 * blkg_conf_prep - parse and prepare for per-blkg config update
466 * @blkcg: target block cgroup
467 * @pol: target policy
468 * @input: input string
469 * @ctx: blkg_conf_ctx to be filled
471 * Parse per-blkg config update from @input and initialize @ctx with the
472 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
473 * value. This function returns with RCU read lock and queue lock held and
474 * must be paired with blkg_conf_finish().
476 int blkg_conf_prep(struct blkcg
*blkcg
, const struct blkcg_policy
*pol
,
477 const char *input
, struct blkg_conf_ctx
*ctx
)
478 __acquires(rcu
) __acquires(disk
->queue
->queue_lock
)
480 struct gendisk
*disk
;
481 struct blkcg_gq
*blkg
;
482 unsigned int major
, minor
;
483 unsigned long long v
;
486 if (sscanf(input
, "%u:%u %llu", &major
, &minor
, &v
) != 3)
489 disk
= get_gendisk(MKDEV(major
, minor
), &part
);
494 spin_lock_irq(disk
->queue
->queue_lock
);
496 if (blkcg_policy_enabled(disk
->queue
, pol
))
497 blkg
= blkg_lookup_create(blkcg
, disk
->queue
);
499 blkg
= ERR_PTR(-EINVAL
);
504 spin_unlock_irq(disk
->queue
->queue_lock
);
507 * If queue was bypassing, we should retry. Do so after a
508 * short msleep(). It isn't strictly necessary but queue
509 * can be bypassing for some time and it's always nice to
510 * avoid busy looping.
514 ret
= restart_syscall();
524 EXPORT_SYMBOL_GPL(blkg_conf_prep
);
527 * blkg_conf_finish - finish up per-blkg config update
528 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
530 * Finish up after per-blkg config update. This function must be paired
531 * with blkg_conf_prep().
533 void blkg_conf_finish(struct blkg_conf_ctx
*ctx
)
534 __releases(ctx
->disk
->queue
->queue_lock
) __releases(rcu
)
536 spin_unlock_irq(ctx
->disk
->queue
->queue_lock
);
540 EXPORT_SYMBOL_GPL(blkg_conf_finish
);
542 struct cftype blkcg_files
[] = {
544 .name
= "reset_stats",
545 .write_u64
= blkcg_reset_stats
,
551 * blkcg_pre_destroy - cgroup pre_destroy callback
552 * @cgroup: cgroup of interest
554 * This function is called when @cgroup is about to go away and responsible
555 * for shooting down all blkgs associated with @cgroup. blkgs should be
556 * removed while holding both q and blkcg locks. As blkcg lock is nested
557 * inside q lock, this function performs reverse double lock dancing.
559 * This is the blkcg counterpart of ioc_release_fn().
561 static int blkcg_pre_destroy(struct cgroup
*cgroup
)
563 struct blkcg
*blkcg
= cgroup_to_blkcg(cgroup
);
565 spin_lock_irq(&blkcg
->lock
);
567 while (!hlist_empty(&blkcg
->blkg_list
)) {
568 struct blkcg_gq
*blkg
= hlist_entry(blkcg
->blkg_list
.first
,
569 struct blkcg_gq
, blkcg_node
);
570 struct request_queue
*q
= blkg
->q
;
572 if (spin_trylock(q
->queue_lock
)) {
574 spin_unlock(q
->queue_lock
);
576 spin_unlock_irq(&blkcg
->lock
);
578 spin_lock_irq(&blkcg
->lock
);
582 spin_unlock_irq(&blkcg
->lock
);
586 static void blkcg_destroy(struct cgroup
*cgroup
)
588 struct blkcg
*blkcg
= cgroup_to_blkcg(cgroup
);
590 if (blkcg
!= &blkcg_root
)
594 static struct cgroup_subsys_state
*blkcg_create(struct cgroup
*cgroup
)
596 static atomic64_t id_seq
= ATOMIC64_INIT(0);
598 struct cgroup
*parent
= cgroup
->parent
;
605 blkcg
= kzalloc(sizeof(*blkcg
), GFP_KERNEL
);
607 return ERR_PTR(-ENOMEM
);
609 blkcg
->cfq_weight
= CFQ_WEIGHT_DEFAULT
;
610 blkcg
->id
= atomic64_inc_return(&id_seq
); /* root is 0, start from 1 */
612 spin_lock_init(&blkcg
->lock
);
613 INIT_RADIX_TREE(&blkcg
->blkg_tree
, GFP_ATOMIC
);
614 INIT_HLIST_HEAD(&blkcg
->blkg_list
);
620 * blkcg_init_queue - initialize blkcg part of request queue
621 * @q: request_queue to initialize
623 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
624 * part of new request_queue @q.
627 * 0 on success, -errno on failure.
629 int blkcg_init_queue(struct request_queue
*q
)
633 return blk_throtl_init(q
);
637 * blkcg_drain_queue - drain blkcg part of request_queue
638 * @q: request_queue to drain
640 * Called from blk_drain_queue(). Responsible for draining blkcg part.
642 void blkcg_drain_queue(struct request_queue
*q
)
644 lockdep_assert_held(q
->queue_lock
);
650 * blkcg_exit_queue - exit and release blkcg part of request_queue
651 * @q: request_queue being released
653 * Called from blk_release_queue(). Responsible for exiting blkcg part.
655 void blkcg_exit_queue(struct request_queue
*q
)
657 spin_lock_irq(q
->queue_lock
);
659 spin_unlock_irq(q
->queue_lock
);
665 * We cannot support shared io contexts, as we have no mean to support
666 * two tasks with the same ioc in two different groups without major rework
667 * of the main cic data structures. For now we allow a task to change
668 * its cgroup only if it's the only owner of its ioc.
670 static int blkcg_can_attach(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
)
672 struct task_struct
*task
;
673 struct io_context
*ioc
;
676 /* task_lock() is needed to avoid races with exit_io_context() */
677 cgroup_taskset_for_each(task
, cgrp
, tset
) {
679 ioc
= task
->io_context
;
680 if (ioc
&& atomic_read(&ioc
->nr_tasks
) > 1)
689 struct cgroup_subsys blkio_subsys
= {
691 .create
= blkcg_create
,
692 .can_attach
= blkcg_can_attach
,
693 .pre_destroy
= blkcg_pre_destroy
,
694 .destroy
= blkcg_destroy
,
695 .subsys_id
= blkio_subsys_id
,
696 .base_cftypes
= blkcg_files
,
697 .module
= THIS_MODULE
,
699 EXPORT_SYMBOL_GPL(blkio_subsys
);
702 * blkcg_activate_policy - activate a blkcg policy on a request_queue
703 * @q: request_queue of interest
704 * @pol: blkcg policy to activate
706 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
707 * bypass mode to populate its blkgs with policy_data for @pol.
709 * Activation happens with @q bypassed, so nobody would be accessing blkgs
710 * from IO path. Update of each blkg is protected by both queue and blkcg
711 * locks so that holding either lock and testing blkcg_policy_enabled() is
712 * always enough for dereferencing policy data.
714 * The caller is responsible for synchronizing [de]activations and policy
715 * [un]registerations. Returns 0 on success, -errno on failure.
717 int blkcg_activate_policy(struct request_queue
*q
,
718 const struct blkcg_policy
*pol
)
721 struct blkcg_gq
*blkg
;
722 struct blkg_policy_data
*pd
, *n
;
726 if (blkcg_policy_enabled(q
, pol
))
729 /* preallocations for root blkg */
730 blkg
= blkg_alloc(&blkcg_root
, q
, GFP_KERNEL
);
734 preloaded
= !radix_tree_preload(GFP_KERNEL
);
736 blk_queue_bypass_start(q
);
738 /* make sure the root blkg exists and count the existing blkgs */
739 spin_lock_irq(q
->queue_lock
);
742 blkg
= __blkg_lookup_create(&blkcg_root
, q
, blkg
);
746 radix_tree_preload_end();
754 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
)
757 spin_unlock_irq(q
->queue_lock
);
759 /* allocate policy_data for all existing blkgs */
761 pd
= kzalloc_node(pol
->pd_size
, GFP_KERNEL
, q
->node
);
766 list_add_tail(&pd
->alloc_node
, &pds
);
770 * Install the allocated pds. With @q bypassing, no new blkg
771 * should have been created while the queue lock was dropped.
773 spin_lock_irq(q
->queue_lock
);
775 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
) {
776 if (WARN_ON(list_empty(&pds
))) {
777 /* umm... this shouldn't happen, just abort */
781 pd
= list_first_entry(&pds
, struct blkg_policy_data
, alloc_node
);
782 list_del_init(&pd
->alloc_node
);
784 /* grab blkcg lock too while installing @pd on @blkg */
785 spin_lock(&blkg
->blkcg
->lock
);
787 blkg
->pd
[pol
->plid
] = pd
;
789 pol
->pd_init_fn(blkg
);
791 spin_unlock(&blkg
->blkcg
->lock
);
794 __set_bit(pol
->plid
, q
->blkcg_pols
);
797 spin_unlock_irq(q
->queue_lock
);
799 blk_queue_bypass_end(q
);
800 list_for_each_entry_safe(pd
, n
, &pds
, alloc_node
)
804 EXPORT_SYMBOL_GPL(blkcg_activate_policy
);
807 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
808 * @q: request_queue of interest
809 * @pol: blkcg policy to deactivate
811 * Deactivate @pol on @q. Follows the same synchronization rules as
812 * blkcg_activate_policy().
814 void blkcg_deactivate_policy(struct request_queue
*q
,
815 const struct blkcg_policy
*pol
)
817 struct blkcg_gq
*blkg
;
819 if (!blkcg_policy_enabled(q
, pol
))
822 blk_queue_bypass_start(q
);
823 spin_lock_irq(q
->queue_lock
);
825 __clear_bit(pol
->plid
, q
->blkcg_pols
);
827 /* if no policy is left, no need for blkgs - shoot them down */
828 if (bitmap_empty(q
->blkcg_pols
, BLKCG_MAX_POLS
))
831 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
) {
832 /* grab blkcg lock too while removing @pd from @blkg */
833 spin_lock(&blkg
->blkcg
->lock
);
836 pol
->pd_exit_fn(blkg
);
838 kfree(blkg
->pd
[pol
->plid
]);
839 blkg
->pd
[pol
->plid
] = NULL
;
841 spin_unlock(&blkg
->blkcg
->lock
);
844 spin_unlock_irq(q
->queue_lock
);
845 blk_queue_bypass_end(q
);
847 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy
);
850 * blkcg_policy_register - register a blkcg policy
851 * @pol: blkcg policy to register
853 * Register @pol with blkcg core. Might sleep and @pol may be modified on
854 * successful registration. Returns 0 on success and -errno on failure.
856 int blkcg_policy_register(struct blkcg_policy
*pol
)
860 if (WARN_ON(pol
->pd_size
< sizeof(struct blkg_policy_data
)))
863 mutex_lock(&blkcg_pol_mutex
);
865 /* find an empty slot */
867 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++)
868 if (!blkcg_policy
[i
])
870 if (i
>= BLKCG_MAX_POLS
)
873 /* register and update blkgs */
875 blkcg_policy
[i
] = pol
;
877 /* everything is in place, add intf files for the new policy */
879 WARN_ON(cgroup_add_cftypes(&blkio_subsys
, pol
->cftypes
));
882 mutex_unlock(&blkcg_pol_mutex
);
885 EXPORT_SYMBOL_GPL(blkcg_policy_register
);
888 * blkcg_policy_unregister - unregister a blkcg policy
889 * @pol: blkcg policy to unregister
891 * Undo blkcg_policy_register(@pol). Might sleep.
893 void blkcg_policy_unregister(struct blkcg_policy
*pol
)
895 mutex_lock(&blkcg_pol_mutex
);
897 if (WARN_ON(blkcg_policy
[pol
->plid
] != pol
))
900 /* kill the intf files first */
902 cgroup_rm_cftypes(&blkio_subsys
, pol
->cftypes
);
904 /* unregister and update blkgs */
905 blkcg_policy
[pol
->plid
] = NULL
;
907 mutex_unlock(&blkcg_pol_mutex
);
909 EXPORT_SYMBOL_GPL(blkcg_policy_unregister
);