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 struct blkcg
*cgroup_to_blkcg(struct cgroup
*cgroup
)
36 return container_of(cgroup_subsys_state(cgroup
, blkio_subsys_id
),
39 EXPORT_SYMBOL_GPL(cgroup_to_blkcg
);
41 static struct blkcg
*task_blkcg(struct task_struct
*tsk
)
43 return container_of(task_subsys_state(tsk
, blkio_subsys_id
),
47 struct blkcg
*bio_blkcg(struct bio
*bio
)
49 if (bio
&& bio
->bi_css
)
50 return container_of(bio
->bi_css
, struct blkcg
, css
);
51 return task_blkcg(current
);
53 EXPORT_SYMBOL_GPL(bio_blkcg
);
55 static bool blkcg_policy_enabled(struct request_queue
*q
,
56 const struct blkcg_policy
*pol
)
58 return pol
&& test_bit(pol
->plid
, q
->blkcg_pols
);
62 * blkg_free - free a blkg
65 * Free @blkg which may be partially allocated.
67 static void blkg_free(struct blkcg_gq
*blkg
)
74 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
75 struct blkcg_policy
*pol
= blkcg_policy
[i
];
76 struct blkg_policy_data
*pd
= blkg
->pd
[i
];
81 if (pol
&& pol
->pd_exit_fn
)
82 pol
->pd_exit_fn(blkg
);
91 * blkg_alloc - allocate a blkg
92 * @blkcg: block cgroup the new blkg is associated with
93 * @q: request_queue the new blkg is associated with
95 * Allocate a new blkg assocating @blkcg and @q.
97 static struct blkcg_gq
*blkg_alloc(struct blkcg
*blkcg
, struct request_queue
*q
)
99 struct blkcg_gq
*blkg
;
102 /* alloc and init base part */
103 blkg
= kzalloc_node(sizeof(*blkg
), GFP_ATOMIC
, q
->node
);
108 INIT_LIST_HEAD(&blkg
->q_node
);
112 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
113 struct blkcg_policy
*pol
= blkcg_policy
[i
];
114 struct blkg_policy_data
*pd
;
116 if (!blkcg_policy_enabled(q
, pol
))
119 /* alloc per-policy data and attach it to blkg */
120 pd
= kzalloc_node(pol
->pd_size
, GFP_ATOMIC
, q
->node
);
129 /* invoke per-policy init */
130 if (blkcg_policy_enabled(blkg
->q
, pol
))
131 pol
->pd_init_fn(blkg
);
137 static struct blkcg_gq
*__blkg_lookup(struct blkcg
*blkcg
,
138 struct request_queue
*q
)
140 struct blkcg_gq
*blkg
;
142 blkg
= rcu_dereference(blkcg
->blkg_hint
);
143 if (blkg
&& blkg
->q
== q
)
147 * Hint didn't match. Look up from the radix tree. Note that we
148 * may not be holding queue_lock and thus are not sure whether
149 * @blkg from blkg_tree has already been removed or not, so we
150 * can't update hint to the lookup result. Leave it to the caller.
152 blkg
= radix_tree_lookup(&blkcg
->blkg_tree
, q
->id
);
153 if (blkg
&& blkg
->q
== q
)
160 * blkg_lookup - lookup blkg for the specified blkcg - q pair
161 * @blkcg: blkcg of interest
162 * @q: request_queue of interest
164 * Lookup blkg for the @blkcg - @q pair. This function should be called
165 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
166 * - see blk_queue_bypass_start() for details.
168 struct blkcg_gq
*blkg_lookup(struct blkcg
*blkcg
, struct request_queue
*q
)
170 WARN_ON_ONCE(!rcu_read_lock_held());
172 if (unlikely(blk_queue_bypass(q
)))
174 return __blkg_lookup(blkcg
, q
);
176 EXPORT_SYMBOL_GPL(blkg_lookup
);
178 static struct blkcg_gq
*__blkg_lookup_create(struct blkcg
*blkcg
,
179 struct request_queue
*q
)
180 __releases(q
->queue_lock
) __acquires(q
->queue_lock
)
182 struct blkcg_gq
*blkg
;
185 WARN_ON_ONCE(!rcu_read_lock_held());
186 lockdep_assert_held(q
->queue_lock
);
188 /* lookup and update hint on success, see __blkg_lookup() for details */
189 blkg
= __blkg_lookup(blkcg
, q
);
191 rcu_assign_pointer(blkcg
->blkg_hint
, blkg
);
195 /* blkg holds a reference to blkcg */
196 if (!css_tryget(&blkcg
->css
))
197 return ERR_PTR(-EINVAL
);
201 blkg
= blkg_alloc(blkcg
, q
);
206 ret
= radix_tree_preload(GFP_ATOMIC
);
210 spin_lock(&blkcg
->lock
);
211 ret
= radix_tree_insert(&blkcg
->blkg_tree
, q
->id
, blkg
);
213 hlist_add_head_rcu(&blkg
->blkcg_node
, &blkcg
->blkg_list
);
214 list_add(&blkg
->q_node
, &q
->blkg_list
);
216 spin_unlock(&blkcg
->lock
);
218 radix_tree_preload_end();
225 css_put(&blkcg
->css
);
229 struct blkcg_gq
*blkg_lookup_create(struct blkcg
*blkcg
,
230 struct request_queue
*q
)
233 * This could be the first entry point of blkcg implementation and
234 * we shouldn't allow anything to go through for a bypassing queue.
236 if (unlikely(blk_queue_bypass(q
)))
237 return ERR_PTR(blk_queue_dead(q
) ? -EINVAL
: -EBUSY
);
238 return __blkg_lookup_create(blkcg
, q
);
240 EXPORT_SYMBOL_GPL(blkg_lookup_create
);
242 static void blkg_destroy(struct blkcg_gq
*blkg
)
244 struct request_queue
*q
= blkg
->q
;
245 struct blkcg
*blkcg
= blkg
->blkcg
;
247 lockdep_assert_held(q
->queue_lock
);
248 lockdep_assert_held(&blkcg
->lock
);
250 /* Something wrong if we are trying to remove same group twice */
251 WARN_ON_ONCE(list_empty(&blkg
->q_node
));
252 WARN_ON_ONCE(hlist_unhashed(&blkg
->blkcg_node
));
254 radix_tree_delete(&blkcg
->blkg_tree
, blkg
->q
->id
);
255 list_del_init(&blkg
->q_node
);
256 hlist_del_init_rcu(&blkg
->blkcg_node
);
259 * Both setting lookup hint to and clearing it from @blkg are done
260 * under queue_lock. If it's not pointing to @blkg now, it never
261 * will. Hint assignment itself can race safely.
263 if (rcu_dereference_raw(blkcg
->blkg_hint
) == blkg
)
264 rcu_assign_pointer(blkcg
->blkg_hint
, NULL
);
267 * Put the reference taken at the time of creation so that when all
268 * queues are gone, group can be destroyed.
274 * blkg_destroy_all - destroy all blkgs associated with a request_queue
275 * @q: request_queue of interest
277 * Destroy all blkgs associated with @q.
279 static void blkg_destroy_all(struct request_queue
*q
)
281 struct blkcg_gq
*blkg
, *n
;
283 lockdep_assert_held(q
->queue_lock
);
285 list_for_each_entry_safe(blkg
, n
, &q
->blkg_list
, q_node
) {
286 struct blkcg
*blkcg
= blkg
->blkcg
;
288 spin_lock(&blkcg
->lock
);
290 spin_unlock(&blkcg
->lock
);
294 static void blkg_rcu_free(struct rcu_head
*rcu_head
)
296 blkg_free(container_of(rcu_head
, struct blkcg_gq
, rcu_head
));
299 void __blkg_release(struct blkcg_gq
*blkg
)
301 /* release the extra blkcg reference this blkg has been holding */
302 css_put(&blkg
->blkcg
->css
);
305 * A group is freed in rcu manner. But having an rcu lock does not
306 * mean that one can access all the fields of blkg and assume these
307 * are valid. For example, don't try to follow throtl_data and
308 * request queue links.
310 * Having a reference to blkg under an rcu allows acess to only
311 * values local to groups like group stats and group rate limits
313 call_rcu(&blkg
->rcu_head
, blkg_rcu_free
);
315 EXPORT_SYMBOL_GPL(__blkg_release
);
317 static int blkcg_reset_stats(struct cgroup
*cgroup
, struct cftype
*cftype
,
320 struct blkcg
*blkcg
= cgroup_to_blkcg(cgroup
);
321 struct blkcg_gq
*blkg
;
322 struct hlist_node
*n
;
325 mutex_lock(&blkcg_pol_mutex
);
326 spin_lock_irq(&blkcg
->lock
);
329 * Note that stat reset is racy - it doesn't synchronize against
330 * stat updates. This is a debug feature which shouldn't exist
331 * anyway. If you get hit by a race, retry.
333 hlist_for_each_entry(blkg
, n
, &blkcg
->blkg_list
, blkcg_node
) {
334 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
335 struct blkcg_policy
*pol
= blkcg_policy
[i
];
337 if (blkcg_policy_enabled(blkg
->q
, pol
) &&
338 pol
->pd_reset_stats_fn
)
339 pol
->pd_reset_stats_fn(blkg
);
343 spin_unlock_irq(&blkcg
->lock
);
344 mutex_unlock(&blkcg_pol_mutex
);
348 static const char *blkg_dev_name(struct blkcg_gq
*blkg
)
350 /* some drivers (floppy) instantiate a queue w/o disk registered */
351 if (blkg
->q
->backing_dev_info
.dev
)
352 return dev_name(blkg
->q
->backing_dev_info
.dev
);
357 * blkcg_print_blkgs - helper for printing per-blkg data
358 * @sf: seq_file to print to
359 * @blkcg: blkcg of interest
360 * @prfill: fill function to print out a blkg
361 * @pol: policy in question
362 * @data: data to be passed to @prfill
363 * @show_total: to print out sum of prfill return values or not
365 * This function invokes @prfill on each blkg of @blkcg if pd for the
366 * policy specified by @pol exists. @prfill is invoked with @sf, the
367 * policy data and @data. If @show_total is %true, the sum of the return
368 * values from @prfill is printed with "Total" label at the end.
370 * This is to be used to construct print functions for
371 * cftype->read_seq_string method.
373 void blkcg_print_blkgs(struct seq_file
*sf
, struct blkcg
*blkcg
,
374 u64 (*prfill
)(struct seq_file
*,
375 struct blkg_policy_data
*, int),
376 const struct blkcg_policy
*pol
, int data
,
379 struct blkcg_gq
*blkg
;
380 struct hlist_node
*n
;
383 spin_lock_irq(&blkcg
->lock
);
384 hlist_for_each_entry(blkg
, n
, &blkcg
->blkg_list
, blkcg_node
)
385 if (blkcg_policy_enabled(blkg
->q
, pol
))
386 total
+= prfill(sf
, blkg
->pd
[pol
->plid
], data
);
387 spin_unlock_irq(&blkcg
->lock
);
390 seq_printf(sf
, "Total %llu\n", (unsigned long long)total
);
392 EXPORT_SYMBOL_GPL(blkcg_print_blkgs
);
395 * __blkg_prfill_u64 - prfill helper for a single u64 value
396 * @sf: seq_file to print to
397 * @pd: policy private data of interest
400 * Print @v to @sf for the device assocaited with @pd.
402 u64
__blkg_prfill_u64(struct seq_file
*sf
, struct blkg_policy_data
*pd
, u64 v
)
404 const char *dname
= blkg_dev_name(pd
->blkg
);
409 seq_printf(sf
, "%s %llu\n", dname
, (unsigned long long)v
);
412 EXPORT_SYMBOL_GPL(__blkg_prfill_u64
);
415 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
416 * @sf: seq_file to print to
417 * @pd: policy private data of interest
418 * @rwstat: rwstat to print
420 * Print @rwstat to @sf for the device assocaited with @pd.
422 u64
__blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
423 const struct blkg_rwstat
*rwstat
)
425 static const char *rwstr
[] = {
426 [BLKG_RWSTAT_READ
] = "Read",
427 [BLKG_RWSTAT_WRITE
] = "Write",
428 [BLKG_RWSTAT_SYNC
] = "Sync",
429 [BLKG_RWSTAT_ASYNC
] = "Async",
431 const char *dname
= blkg_dev_name(pd
->blkg
);
438 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
439 seq_printf(sf
, "%s %s %llu\n", dname
, rwstr
[i
],
440 (unsigned long long)rwstat
->cnt
[i
]);
442 v
= rwstat
->cnt
[BLKG_RWSTAT_READ
] + rwstat
->cnt
[BLKG_RWSTAT_WRITE
];
443 seq_printf(sf
, "%s Total %llu\n", dname
, (unsigned long long)v
);
448 * blkg_prfill_stat - prfill callback for blkg_stat
449 * @sf: seq_file to print to
450 * @pd: policy private data of interest
451 * @off: offset to the blkg_stat in @pd
453 * prfill callback for printing a blkg_stat.
455 u64
blkg_prfill_stat(struct seq_file
*sf
, struct blkg_policy_data
*pd
, int off
)
457 return __blkg_prfill_u64(sf
, pd
, blkg_stat_read((void *)pd
+ off
));
459 EXPORT_SYMBOL_GPL(blkg_prfill_stat
);
462 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
463 * @sf: seq_file to print to
464 * @pd: policy private data of interest
465 * @off: offset to the blkg_rwstat in @pd
467 * prfill callback for printing a blkg_rwstat.
469 u64
blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
472 struct blkg_rwstat rwstat
= blkg_rwstat_read((void *)pd
+ off
);
474 return __blkg_prfill_rwstat(sf
, pd
, &rwstat
);
476 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat
);
479 * blkg_conf_prep - parse and prepare for per-blkg config update
480 * @blkcg: target block cgroup
481 * @pol: target policy
482 * @input: input string
483 * @ctx: blkg_conf_ctx to be filled
485 * Parse per-blkg config update from @input and initialize @ctx with the
486 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
487 * value. This function returns with RCU read lock and queue lock held and
488 * must be paired with blkg_conf_finish().
490 int blkg_conf_prep(struct blkcg
*blkcg
, const struct blkcg_policy
*pol
,
491 const char *input
, struct blkg_conf_ctx
*ctx
)
492 __acquires(rcu
) __acquires(disk
->queue
->queue_lock
)
494 struct gendisk
*disk
;
495 struct blkcg_gq
*blkg
;
496 unsigned int major
, minor
;
497 unsigned long long v
;
500 if (sscanf(input
, "%u:%u %llu", &major
, &minor
, &v
) != 3)
503 disk
= get_gendisk(MKDEV(major
, minor
), &part
);
508 spin_lock_irq(disk
->queue
->queue_lock
);
510 if (blkcg_policy_enabled(disk
->queue
, pol
))
511 blkg
= blkg_lookup_create(blkcg
, disk
->queue
);
513 blkg
= ERR_PTR(-EINVAL
);
518 spin_unlock_irq(disk
->queue
->queue_lock
);
521 * If queue was bypassing, we should retry. Do so after a
522 * short msleep(). It isn't strictly necessary but queue
523 * can be bypassing for some time and it's always nice to
524 * avoid busy looping.
528 ret
= restart_syscall();
538 EXPORT_SYMBOL_GPL(blkg_conf_prep
);
541 * blkg_conf_finish - finish up per-blkg config update
542 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
544 * Finish up after per-blkg config update. This function must be paired
545 * with blkg_conf_prep().
547 void blkg_conf_finish(struct blkg_conf_ctx
*ctx
)
548 __releases(ctx
->disk
->queue
->queue_lock
) __releases(rcu
)
550 spin_unlock_irq(ctx
->disk
->queue
->queue_lock
);
554 EXPORT_SYMBOL_GPL(blkg_conf_finish
);
556 struct cftype blkcg_files
[] = {
558 .name
= "reset_stats",
559 .write_u64
= blkcg_reset_stats
,
565 * blkcg_pre_destroy - cgroup pre_destroy callback
566 * @cgroup: cgroup of interest
568 * This function is called when @cgroup is about to go away and responsible
569 * for shooting down all blkgs associated with @cgroup. blkgs should be
570 * removed while holding both q and blkcg locks. As blkcg lock is nested
571 * inside q lock, this function performs reverse double lock dancing.
573 * This is the blkcg counterpart of ioc_release_fn().
575 static int blkcg_pre_destroy(struct cgroup
*cgroup
)
577 struct blkcg
*blkcg
= cgroup_to_blkcg(cgroup
);
579 spin_lock_irq(&blkcg
->lock
);
581 while (!hlist_empty(&blkcg
->blkg_list
)) {
582 struct blkcg_gq
*blkg
= hlist_entry(blkcg
->blkg_list
.first
,
583 struct blkcg_gq
, blkcg_node
);
584 struct request_queue
*q
= blkg
->q
;
586 if (spin_trylock(q
->queue_lock
)) {
588 spin_unlock(q
->queue_lock
);
590 spin_unlock_irq(&blkcg
->lock
);
592 spin_lock_irq(&blkcg
->lock
);
596 spin_unlock_irq(&blkcg
->lock
);
600 static void blkcg_destroy(struct cgroup
*cgroup
)
602 struct blkcg
*blkcg
= cgroup_to_blkcg(cgroup
);
604 if (blkcg
!= &blkcg_root
)
608 static struct cgroup_subsys_state
*blkcg_create(struct cgroup
*cgroup
)
610 static atomic64_t id_seq
= ATOMIC64_INIT(0);
612 struct cgroup
*parent
= cgroup
->parent
;
619 blkcg
= kzalloc(sizeof(*blkcg
), GFP_KERNEL
);
621 return ERR_PTR(-ENOMEM
);
623 blkcg
->cfq_weight
= CFQ_WEIGHT_DEFAULT
;
624 blkcg
->id
= atomic64_inc_return(&id_seq
); /* root is 0, start from 1 */
626 spin_lock_init(&blkcg
->lock
);
627 INIT_RADIX_TREE(&blkcg
->blkg_tree
, GFP_ATOMIC
);
628 INIT_HLIST_HEAD(&blkcg
->blkg_list
);
634 * blkcg_init_queue - initialize blkcg part of request queue
635 * @q: request_queue to initialize
637 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
638 * part of new request_queue @q.
641 * 0 on success, -errno on failure.
643 int blkcg_init_queue(struct request_queue
*q
)
647 return blk_throtl_init(q
);
651 * blkcg_drain_queue - drain blkcg part of request_queue
652 * @q: request_queue to drain
654 * Called from blk_drain_queue(). Responsible for draining blkcg part.
656 void blkcg_drain_queue(struct request_queue
*q
)
658 lockdep_assert_held(q
->queue_lock
);
664 * blkcg_exit_queue - exit and release blkcg part of request_queue
665 * @q: request_queue being released
667 * Called from blk_release_queue(). Responsible for exiting blkcg part.
669 void blkcg_exit_queue(struct request_queue
*q
)
671 spin_lock_irq(q
->queue_lock
);
673 spin_unlock_irq(q
->queue_lock
);
679 * We cannot support shared io contexts, as we have no mean to support
680 * two tasks with the same ioc in two different groups without major rework
681 * of the main cic data structures. For now we allow a task to change
682 * its cgroup only if it's the only owner of its ioc.
684 static int blkcg_can_attach(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
)
686 struct task_struct
*task
;
687 struct io_context
*ioc
;
690 /* task_lock() is needed to avoid races with exit_io_context() */
691 cgroup_taskset_for_each(task
, cgrp
, tset
) {
693 ioc
= task
->io_context
;
694 if (ioc
&& atomic_read(&ioc
->nr_tasks
) > 1)
703 struct cgroup_subsys blkio_subsys
= {
705 .create
= blkcg_create
,
706 .can_attach
= blkcg_can_attach
,
707 .pre_destroy
= blkcg_pre_destroy
,
708 .destroy
= blkcg_destroy
,
709 .subsys_id
= blkio_subsys_id
,
710 .base_cftypes
= blkcg_files
,
711 .module
= THIS_MODULE
,
713 EXPORT_SYMBOL_GPL(blkio_subsys
);
716 * blkcg_activate_policy - activate a blkcg policy on a request_queue
717 * @q: request_queue of interest
718 * @pol: blkcg policy to activate
720 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
721 * bypass mode to populate its blkgs with policy_data for @pol.
723 * Activation happens with @q bypassed, so nobody would be accessing blkgs
724 * from IO path. Update of each blkg is protected by both queue and blkcg
725 * locks so that holding either lock and testing blkcg_policy_enabled() is
726 * always enough for dereferencing policy data.
728 * The caller is responsible for synchronizing [de]activations and policy
729 * [un]registerations. Returns 0 on success, -errno on failure.
731 int blkcg_activate_policy(struct request_queue
*q
,
732 const struct blkcg_policy
*pol
)
735 struct blkcg_gq
*blkg
;
736 struct blkg_policy_data
*pd
, *n
;
739 if (blkcg_policy_enabled(q
, pol
))
742 blk_queue_bypass_start(q
);
744 /* make sure the root blkg exists and count the existing blkgs */
745 spin_lock_irq(q
->queue_lock
);
748 blkg
= __blkg_lookup_create(&blkcg_root
, q
);
757 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
)
760 spin_unlock_irq(q
->queue_lock
);
762 /* allocate policy_data for all existing blkgs */
764 pd
= kzalloc_node(pol
->pd_size
, GFP_KERNEL
, q
->node
);
769 list_add_tail(&pd
->alloc_node
, &pds
);
773 * Install the allocated pds. With @q bypassing, no new blkg
774 * should have been created while the queue lock was dropped.
776 spin_lock_irq(q
->queue_lock
);
778 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
) {
779 if (WARN_ON(list_empty(&pds
))) {
780 /* umm... this shouldn't happen, just abort */
784 pd
= list_first_entry(&pds
, struct blkg_policy_data
, alloc_node
);
785 list_del_init(&pd
->alloc_node
);
787 /* grab blkcg lock too while installing @pd on @blkg */
788 spin_lock(&blkg
->blkcg
->lock
);
790 blkg
->pd
[pol
->plid
] = pd
;
792 pol
->pd_init_fn(blkg
);
794 spin_unlock(&blkg
->blkcg
->lock
);
797 __set_bit(pol
->plid
, q
->blkcg_pols
);
800 spin_unlock_irq(q
->queue_lock
);
802 blk_queue_bypass_end(q
);
803 list_for_each_entry_safe(pd
, n
, &pds
, alloc_node
)
807 EXPORT_SYMBOL_GPL(blkcg_activate_policy
);
810 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
811 * @q: request_queue of interest
812 * @pol: blkcg policy to deactivate
814 * Deactivate @pol on @q. Follows the same synchronization rules as
815 * blkcg_activate_policy().
817 void blkcg_deactivate_policy(struct request_queue
*q
,
818 const struct blkcg_policy
*pol
)
820 struct blkcg_gq
*blkg
;
822 if (!blkcg_policy_enabled(q
, pol
))
825 blk_queue_bypass_start(q
);
826 spin_lock_irq(q
->queue_lock
);
828 __clear_bit(pol
->plid
, q
->blkcg_pols
);
830 /* if no policy is left, no need for blkgs - shoot them down */
831 if (bitmap_empty(q
->blkcg_pols
, BLKCG_MAX_POLS
))
834 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
) {
835 /* grab blkcg lock too while removing @pd from @blkg */
836 spin_lock(&blkg
->blkcg
->lock
);
839 pol
->pd_exit_fn(blkg
);
841 kfree(blkg
->pd
[pol
->plid
]);
842 blkg
->pd
[pol
->plid
] = NULL
;
844 spin_unlock(&blkg
->blkcg
->lock
);
847 spin_unlock_irq(q
->queue_lock
);
848 blk_queue_bypass_end(q
);
850 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy
);
853 * blkcg_policy_register - register a blkcg policy
854 * @pol: blkcg policy to register
856 * Register @pol with blkcg core. Might sleep and @pol may be modified on
857 * successful registration. Returns 0 on success and -errno on failure.
859 int blkcg_policy_register(struct blkcg_policy
*pol
)
863 if (WARN_ON(pol
->pd_size
< sizeof(struct blkg_policy_data
)))
866 mutex_lock(&blkcg_pol_mutex
);
868 /* find an empty slot */
870 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++)
871 if (!blkcg_policy
[i
])
873 if (i
>= BLKCG_MAX_POLS
)
876 /* register and update blkgs */
878 blkcg_policy
[i
] = pol
;
880 /* everything is in place, add intf files for the new policy */
882 WARN_ON(cgroup_add_cftypes(&blkio_subsys
, pol
->cftypes
));
885 mutex_unlock(&blkcg_pol_mutex
);
888 EXPORT_SYMBOL_GPL(blkcg_policy_register
);
891 * blkcg_policy_unregister - unregister a blkcg policy
892 * @pol: blkcg policy to unregister
894 * Undo blkcg_policy_register(@pol). Might sleep.
896 void blkcg_policy_unregister(struct blkcg_policy
*pol
)
898 mutex_lock(&blkcg_pol_mutex
);
900 if (WARN_ON(blkcg_policy
[pol
->plid
] != pol
))
903 /* kill the intf files first */
905 cgroup_rm_cftypes(&blkio_subsys
, pol
->cftypes
);
907 /* unregister and update blkgs */
908 blkcg_policy
[pol
->plid
] = NULL
;
910 mutex_unlock(&blkcg_pol_mutex
);
912 EXPORT_SYMBOL_GPL(blkcg_policy_unregister
);