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
);
66 blk_exit_rl(&blkg
->rl
);
71 * blkg_alloc - allocate a blkg
72 * @blkcg: block cgroup the new blkg is associated with
73 * @q: request_queue the new blkg is associated with
74 * @gfp_mask: allocation mask to use
76 * Allocate a new blkg assocating @blkcg and @q.
78 static struct blkcg_gq
*blkg_alloc(struct blkcg
*blkcg
, struct request_queue
*q
,
81 struct blkcg_gq
*blkg
;
84 /* alloc and init base part */
85 blkg
= kzalloc_node(sizeof(*blkg
), gfp_mask
, q
->node
);
90 INIT_LIST_HEAD(&blkg
->q_node
);
94 /* root blkg uses @q->root_rl, init rl only for !root blkgs */
95 if (blkcg
!= &blkcg_root
) {
96 if (blk_init_rl(&blkg
->rl
, q
, gfp_mask
))
101 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
102 struct blkcg_policy
*pol
= blkcg_policy
[i
];
103 struct blkg_policy_data
*pd
;
105 if (!blkcg_policy_enabled(q
, pol
))
108 /* alloc per-policy data and attach it to blkg */
109 pd
= kzalloc_node(pol
->pd_size
, gfp_mask
, q
->node
);
116 /* invoke per-policy init */
118 pol
->pd_init_fn(blkg
);
128 static struct blkcg_gq
*__blkg_lookup(struct blkcg
*blkcg
,
129 struct request_queue
*q
, bool update_hint
)
131 struct blkcg_gq
*blkg
;
133 blkg
= rcu_dereference(blkcg
->blkg_hint
);
134 if (blkg
&& blkg
->q
== q
)
138 * Hint didn't match. Look up from the radix tree. Note that the
139 * hint can only be updated under queue_lock as otherwise @blkg
140 * could have already been removed from blkg_tree. The caller is
141 * responsible for grabbing queue_lock if @update_hint.
143 blkg
= radix_tree_lookup(&blkcg
->blkg_tree
, q
->id
);
144 if (blkg
&& blkg
->q
== q
) {
146 lockdep_assert_held(q
->queue_lock
);
147 rcu_assign_pointer(blkcg
->blkg_hint
, blkg
);
156 * blkg_lookup - lookup blkg for the specified blkcg - q pair
157 * @blkcg: blkcg of interest
158 * @q: request_queue of interest
160 * Lookup blkg for the @blkcg - @q pair. This function should be called
161 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
162 * - see blk_queue_bypass_start() for details.
164 struct blkcg_gq
*blkg_lookup(struct blkcg
*blkcg
, struct request_queue
*q
)
166 WARN_ON_ONCE(!rcu_read_lock_held());
168 if (unlikely(blk_queue_bypass(q
)))
170 return __blkg_lookup(blkcg
, q
, false);
172 EXPORT_SYMBOL_GPL(blkg_lookup
);
175 * If @new_blkg is %NULL, this function tries to allocate a new one as
176 * necessary using %GFP_ATOMIC. @new_blkg is always consumed on return.
178 static struct blkcg_gq
*blkg_create(struct blkcg
*blkcg
,
179 struct request_queue
*q
,
180 struct blkcg_gq
*new_blkg
)
182 struct blkcg_gq
*blkg
;
185 WARN_ON_ONCE(!rcu_read_lock_held());
186 lockdep_assert_held(q
->queue_lock
);
188 /* blkg holds a reference to blkcg */
189 if (!css_tryget(&blkcg
->css
)) {
196 new_blkg
= blkg_alloc(blkcg
, q
, GFP_ATOMIC
);
197 if (unlikely(!new_blkg
)) {
204 /* link parent and insert */
205 if (blkcg_parent(blkcg
)) {
206 blkg
->parent
= __blkg_lookup(blkcg_parent(blkcg
), q
, false);
207 if (WARN_ON_ONCE(!blkg
->parent
)) {
208 blkg
= ERR_PTR(-EINVAL
);
211 blkg_get(blkg
->parent
);
214 spin_lock(&blkcg
->lock
);
215 ret
= radix_tree_insert(&blkcg
->blkg_tree
, q
->id
, blkg
);
217 hlist_add_head_rcu(&blkg
->blkcg_node
, &blkcg
->blkg_list
);
218 list_add(&blkg
->q_node
, &q
->blkg_list
);
220 spin_unlock(&blkcg
->lock
);
225 /* @blkg failed fully initialized, use the usual release path */
230 css_put(&blkcg
->css
);
237 * blkg_lookup_create - lookup blkg, try to create one if not there
238 * @blkcg: blkcg of interest
239 * @q: request_queue of interest
241 * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
242 * create one. blkg creation is performed recursively from blkcg_root such
243 * that all non-root blkg's have access to the parent blkg. This function
244 * should be called under RCU read lock and @q->queue_lock.
246 * Returns pointer to the looked up or created blkg on success, ERR_PTR()
247 * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not
248 * dead and bypassing, returns ERR_PTR(-EBUSY).
250 struct blkcg_gq
*blkg_lookup_create(struct blkcg
*blkcg
,
251 struct request_queue
*q
)
253 struct blkcg_gq
*blkg
;
255 WARN_ON_ONCE(!rcu_read_lock_held());
256 lockdep_assert_held(q
->queue_lock
);
259 * This could be the first entry point of blkcg implementation and
260 * we shouldn't allow anything to go through for a bypassing queue.
262 if (unlikely(blk_queue_bypass(q
)))
263 return ERR_PTR(blk_queue_dying(q
) ? -EINVAL
: -EBUSY
);
265 blkg
= __blkg_lookup(blkcg
, q
, true);
270 * Create blkgs walking down from blkcg_root to @blkcg, so that all
271 * non-root blkgs have access to their parents.
274 struct blkcg
*pos
= blkcg
;
275 struct blkcg
*parent
= blkcg_parent(blkcg
);
277 while (parent
&& !__blkg_lookup(parent
, q
, false)) {
279 parent
= blkcg_parent(parent
);
282 blkg
= blkg_create(pos
, q
, NULL
);
283 if (pos
== blkcg
|| IS_ERR(blkg
))
287 EXPORT_SYMBOL_GPL(blkg_lookup_create
);
289 static void blkg_destroy(struct blkcg_gq
*blkg
)
291 struct blkcg
*blkcg
= blkg
->blkcg
;
293 lockdep_assert_held(blkg
->q
->queue_lock
);
294 lockdep_assert_held(&blkcg
->lock
);
296 /* Something wrong if we are trying to remove same group twice */
297 WARN_ON_ONCE(list_empty(&blkg
->q_node
));
298 WARN_ON_ONCE(hlist_unhashed(&blkg
->blkcg_node
));
300 radix_tree_delete(&blkcg
->blkg_tree
, blkg
->q
->id
);
301 list_del_init(&blkg
->q_node
);
302 hlist_del_init_rcu(&blkg
->blkcg_node
);
305 * Both setting lookup hint to and clearing it from @blkg are done
306 * under queue_lock. If it's not pointing to @blkg now, it never
307 * will. Hint assignment itself can race safely.
309 if (rcu_dereference_raw(blkcg
->blkg_hint
) == blkg
)
310 rcu_assign_pointer(blkcg
->blkg_hint
, NULL
);
313 * Put the reference taken at the time of creation so that when all
314 * queues are gone, group can be destroyed.
320 * blkg_destroy_all - destroy all blkgs associated with a request_queue
321 * @q: request_queue of interest
323 * Destroy all blkgs associated with @q.
325 static void blkg_destroy_all(struct request_queue
*q
)
327 struct blkcg_gq
*blkg
, *n
;
329 lockdep_assert_held(q
->queue_lock
);
331 list_for_each_entry_safe(blkg
, n
, &q
->blkg_list
, q_node
) {
332 struct blkcg
*blkcg
= blkg
->blkcg
;
334 spin_lock(&blkcg
->lock
);
336 spin_unlock(&blkcg
->lock
);
340 * root blkg is destroyed. Just clear the pointer since
341 * root_rl does not take reference on root blkg.
344 q
->root_rl
.blkg
= NULL
;
347 static void blkg_rcu_free(struct rcu_head
*rcu_head
)
349 blkg_free(container_of(rcu_head
, struct blkcg_gq
, rcu_head
));
352 void __blkg_release(struct blkcg_gq
*blkg
)
354 /* release the blkcg and parent blkg refs this blkg has been holding */
355 css_put(&blkg
->blkcg
->css
);
357 blkg_put(blkg
->parent
);
360 * A group is freed in rcu manner. But having an rcu lock does not
361 * mean that one can access all the fields of blkg and assume these
362 * are valid. For example, don't try to follow throtl_data and
363 * request queue links.
365 * Having a reference to blkg under an rcu allows acess to only
366 * values local to groups like group stats and group rate limits
368 call_rcu(&blkg
->rcu_head
, blkg_rcu_free
);
370 EXPORT_SYMBOL_GPL(__blkg_release
);
373 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
374 * because the root blkg uses @q->root_rl instead of its own rl.
376 struct request_list
*__blk_queue_next_rl(struct request_list
*rl
,
377 struct request_queue
*q
)
379 struct list_head
*ent
;
380 struct blkcg_gq
*blkg
;
383 * Determine the current blkg list_head. The first entry is
384 * root_rl which is off @q->blkg_list and mapped to the head.
386 if (rl
== &q
->root_rl
) {
388 /* There are no more block groups, hence no request lists */
392 blkg
= container_of(rl
, struct blkcg_gq
, rl
);
396 /* walk to the next list_head, skip root blkcg */
398 if (ent
== &q
->root_blkg
->q_node
)
400 if (ent
== &q
->blkg_list
)
403 blkg
= container_of(ent
, struct blkcg_gq
, q_node
);
407 static int blkcg_reset_stats(struct cgroup
*cgroup
, struct cftype
*cftype
,
410 struct blkcg
*blkcg
= cgroup_to_blkcg(cgroup
);
411 struct blkcg_gq
*blkg
;
412 struct hlist_node
*n
;
415 mutex_lock(&blkcg_pol_mutex
);
416 spin_lock_irq(&blkcg
->lock
);
419 * Note that stat reset is racy - it doesn't synchronize against
420 * stat updates. This is a debug feature which shouldn't exist
421 * anyway. If you get hit by a race, retry.
423 hlist_for_each_entry(blkg
, n
, &blkcg
->blkg_list
, blkcg_node
) {
424 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
425 struct blkcg_policy
*pol
= blkcg_policy
[i
];
427 if (blkcg_policy_enabled(blkg
->q
, pol
) &&
428 pol
->pd_reset_stats_fn
)
429 pol
->pd_reset_stats_fn(blkg
);
433 spin_unlock_irq(&blkcg
->lock
);
434 mutex_unlock(&blkcg_pol_mutex
);
438 static const char *blkg_dev_name(struct blkcg_gq
*blkg
)
440 /* some drivers (floppy) instantiate a queue w/o disk registered */
441 if (blkg
->q
->backing_dev_info
.dev
)
442 return dev_name(blkg
->q
->backing_dev_info
.dev
);
447 * blkcg_print_blkgs - helper for printing per-blkg data
448 * @sf: seq_file to print to
449 * @blkcg: blkcg of interest
450 * @prfill: fill function to print out a blkg
451 * @pol: policy in question
452 * @data: data to be passed to @prfill
453 * @show_total: to print out sum of prfill return values or not
455 * This function invokes @prfill on each blkg of @blkcg if pd for the
456 * policy specified by @pol exists. @prfill is invoked with @sf, the
457 * policy data and @data. If @show_total is %true, the sum of the return
458 * values from @prfill is printed with "Total" label at the end.
460 * This is to be used to construct print functions for
461 * cftype->read_seq_string method.
463 void blkcg_print_blkgs(struct seq_file
*sf
, struct blkcg
*blkcg
,
464 u64 (*prfill
)(struct seq_file
*,
465 struct blkg_policy_data
*, int),
466 const struct blkcg_policy
*pol
, int data
,
469 struct blkcg_gq
*blkg
;
470 struct hlist_node
*n
;
473 spin_lock_irq(&blkcg
->lock
);
474 hlist_for_each_entry(blkg
, n
, &blkcg
->blkg_list
, blkcg_node
)
475 if (blkcg_policy_enabled(blkg
->q
, pol
))
476 total
+= prfill(sf
, blkg
->pd
[pol
->plid
], data
);
477 spin_unlock_irq(&blkcg
->lock
);
480 seq_printf(sf
, "Total %llu\n", (unsigned long long)total
);
482 EXPORT_SYMBOL_GPL(blkcg_print_blkgs
);
485 * __blkg_prfill_u64 - prfill helper for a single u64 value
486 * @sf: seq_file to print to
487 * @pd: policy private data of interest
490 * Print @v to @sf for the device assocaited with @pd.
492 u64
__blkg_prfill_u64(struct seq_file
*sf
, struct blkg_policy_data
*pd
, u64 v
)
494 const char *dname
= blkg_dev_name(pd
->blkg
);
499 seq_printf(sf
, "%s %llu\n", dname
, (unsigned long long)v
);
502 EXPORT_SYMBOL_GPL(__blkg_prfill_u64
);
505 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
506 * @sf: seq_file to print to
507 * @pd: policy private data of interest
508 * @rwstat: rwstat to print
510 * Print @rwstat to @sf for the device assocaited with @pd.
512 u64
__blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
513 const struct blkg_rwstat
*rwstat
)
515 static const char *rwstr
[] = {
516 [BLKG_RWSTAT_READ
] = "Read",
517 [BLKG_RWSTAT_WRITE
] = "Write",
518 [BLKG_RWSTAT_SYNC
] = "Sync",
519 [BLKG_RWSTAT_ASYNC
] = "Async",
521 const char *dname
= blkg_dev_name(pd
->blkg
);
528 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
529 seq_printf(sf
, "%s %s %llu\n", dname
, rwstr
[i
],
530 (unsigned long long)rwstat
->cnt
[i
]);
532 v
= rwstat
->cnt
[BLKG_RWSTAT_READ
] + rwstat
->cnt
[BLKG_RWSTAT_WRITE
];
533 seq_printf(sf
, "%s Total %llu\n", dname
, (unsigned long long)v
);
538 * blkg_prfill_stat - prfill callback for blkg_stat
539 * @sf: seq_file to print to
540 * @pd: policy private data of interest
541 * @off: offset to the blkg_stat in @pd
543 * prfill callback for printing a blkg_stat.
545 u64
blkg_prfill_stat(struct seq_file
*sf
, struct blkg_policy_data
*pd
, int off
)
547 return __blkg_prfill_u64(sf
, pd
, blkg_stat_read((void *)pd
+ off
));
549 EXPORT_SYMBOL_GPL(blkg_prfill_stat
);
552 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
553 * @sf: seq_file to print to
554 * @pd: policy private data of interest
555 * @off: offset to the blkg_rwstat in @pd
557 * prfill callback for printing a blkg_rwstat.
559 u64
blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
562 struct blkg_rwstat rwstat
= blkg_rwstat_read((void *)pd
+ off
);
564 return __blkg_prfill_rwstat(sf
, pd
, &rwstat
);
566 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat
);
569 * blkg_conf_prep - parse and prepare for per-blkg config update
570 * @blkcg: target block cgroup
571 * @pol: target policy
572 * @input: input string
573 * @ctx: blkg_conf_ctx to be filled
575 * Parse per-blkg config update from @input and initialize @ctx with the
576 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
577 * value. This function returns with RCU read lock and queue lock held and
578 * must be paired with blkg_conf_finish().
580 int blkg_conf_prep(struct blkcg
*blkcg
, const struct blkcg_policy
*pol
,
581 const char *input
, struct blkg_conf_ctx
*ctx
)
582 __acquires(rcu
) __acquires(disk
->queue
->queue_lock
)
584 struct gendisk
*disk
;
585 struct blkcg_gq
*blkg
;
586 unsigned int major
, minor
;
587 unsigned long long v
;
590 if (sscanf(input
, "%u:%u %llu", &major
, &minor
, &v
) != 3)
593 disk
= get_gendisk(MKDEV(major
, minor
), &part
);
598 spin_lock_irq(disk
->queue
->queue_lock
);
600 if (blkcg_policy_enabled(disk
->queue
, pol
))
601 blkg
= blkg_lookup_create(blkcg
, disk
->queue
);
603 blkg
= ERR_PTR(-EINVAL
);
608 spin_unlock_irq(disk
->queue
->queue_lock
);
611 * If queue was bypassing, we should retry. Do so after a
612 * short msleep(). It isn't strictly necessary but queue
613 * can be bypassing for some time and it's always nice to
614 * avoid busy looping.
618 ret
= restart_syscall();
628 EXPORT_SYMBOL_GPL(blkg_conf_prep
);
631 * blkg_conf_finish - finish up per-blkg config update
632 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
634 * Finish up after per-blkg config update. This function must be paired
635 * with blkg_conf_prep().
637 void blkg_conf_finish(struct blkg_conf_ctx
*ctx
)
638 __releases(ctx
->disk
->queue
->queue_lock
) __releases(rcu
)
640 spin_unlock_irq(ctx
->disk
->queue
->queue_lock
);
644 EXPORT_SYMBOL_GPL(blkg_conf_finish
);
646 struct cftype blkcg_files
[] = {
648 .name
= "reset_stats",
649 .write_u64
= blkcg_reset_stats
,
655 * blkcg_css_offline - cgroup css_offline callback
656 * @cgroup: cgroup of interest
658 * This function is called when @cgroup is about to go away and responsible
659 * for shooting down all blkgs associated with @cgroup. blkgs should be
660 * removed while holding both q and blkcg locks. As blkcg lock is nested
661 * inside q lock, this function performs reverse double lock dancing.
663 * This is the blkcg counterpart of ioc_release_fn().
665 static void blkcg_css_offline(struct cgroup
*cgroup
)
667 struct blkcg
*blkcg
= cgroup_to_blkcg(cgroup
);
669 spin_lock_irq(&blkcg
->lock
);
671 while (!hlist_empty(&blkcg
->blkg_list
)) {
672 struct blkcg_gq
*blkg
= hlist_entry(blkcg
->blkg_list
.first
,
673 struct blkcg_gq
, blkcg_node
);
674 struct request_queue
*q
= blkg
->q
;
676 if (spin_trylock(q
->queue_lock
)) {
678 spin_unlock(q
->queue_lock
);
680 spin_unlock_irq(&blkcg
->lock
);
682 spin_lock_irq(&blkcg
->lock
);
686 spin_unlock_irq(&blkcg
->lock
);
689 static void blkcg_css_free(struct cgroup
*cgroup
)
691 struct blkcg
*blkcg
= cgroup_to_blkcg(cgroup
);
693 if (blkcg
!= &blkcg_root
)
697 static struct cgroup_subsys_state
*blkcg_css_alloc(struct cgroup
*cgroup
)
699 static atomic64_t id_seq
= ATOMIC64_INIT(0);
701 struct cgroup
*parent
= cgroup
->parent
;
708 blkcg
= kzalloc(sizeof(*blkcg
), GFP_KERNEL
);
710 return ERR_PTR(-ENOMEM
);
712 blkcg
->cfq_weight
= CFQ_WEIGHT_DEFAULT
;
713 blkcg
->id
= atomic64_inc_return(&id_seq
); /* root is 0, start from 1 */
715 spin_lock_init(&blkcg
->lock
);
716 INIT_RADIX_TREE(&blkcg
->blkg_tree
, GFP_ATOMIC
);
717 INIT_HLIST_HEAD(&blkcg
->blkg_list
);
723 * blkcg_init_queue - initialize blkcg part of request queue
724 * @q: request_queue to initialize
726 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
727 * part of new request_queue @q.
730 * 0 on success, -errno on failure.
732 int blkcg_init_queue(struct request_queue
*q
)
736 return blk_throtl_init(q
);
740 * blkcg_drain_queue - drain blkcg part of request_queue
741 * @q: request_queue to drain
743 * Called from blk_drain_queue(). Responsible for draining blkcg part.
745 void blkcg_drain_queue(struct request_queue
*q
)
747 lockdep_assert_held(q
->queue_lock
);
753 * blkcg_exit_queue - exit and release blkcg part of request_queue
754 * @q: request_queue being released
756 * Called from blk_release_queue(). Responsible for exiting blkcg part.
758 void blkcg_exit_queue(struct request_queue
*q
)
760 spin_lock_irq(q
->queue_lock
);
762 spin_unlock_irq(q
->queue_lock
);
768 * We cannot support shared io contexts, as we have no mean to support
769 * two tasks with the same ioc in two different groups without major rework
770 * of the main cic data structures. For now we allow a task to change
771 * its cgroup only if it's the only owner of its ioc.
773 static int blkcg_can_attach(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
)
775 struct task_struct
*task
;
776 struct io_context
*ioc
;
779 /* task_lock() is needed to avoid races with exit_io_context() */
780 cgroup_taskset_for_each(task
, cgrp
, tset
) {
782 ioc
= task
->io_context
;
783 if (ioc
&& atomic_read(&ioc
->nr_tasks
) > 1)
792 struct cgroup_subsys blkio_subsys
= {
794 .css_alloc
= blkcg_css_alloc
,
795 .css_offline
= blkcg_css_offline
,
796 .css_free
= blkcg_css_free
,
797 .can_attach
= blkcg_can_attach
,
798 .subsys_id
= blkio_subsys_id
,
799 .base_cftypes
= blkcg_files
,
800 .module
= THIS_MODULE
,
803 * blkio subsystem is utterly broken in terms of hierarchy support.
804 * It treats all cgroups equally regardless of where they're
805 * located in the hierarchy - all cgroups are treated as if they're
806 * right below the root. Fix it and remove the following.
808 .broken_hierarchy
= true,
810 EXPORT_SYMBOL_GPL(blkio_subsys
);
813 * blkcg_activate_policy - activate a blkcg policy on a request_queue
814 * @q: request_queue of interest
815 * @pol: blkcg policy to activate
817 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
818 * bypass mode to populate its blkgs with policy_data for @pol.
820 * Activation happens with @q bypassed, so nobody would be accessing blkgs
821 * from IO path. Update of each blkg is protected by both queue and blkcg
822 * locks so that holding either lock and testing blkcg_policy_enabled() is
823 * always enough for dereferencing policy data.
825 * The caller is responsible for synchronizing [de]activations and policy
826 * [un]registerations. Returns 0 on success, -errno on failure.
828 int blkcg_activate_policy(struct request_queue
*q
,
829 const struct blkcg_policy
*pol
)
832 struct blkcg_gq
*blkg
, *new_blkg
;
833 struct blkg_policy_data
*pd
, *n
;
837 if (blkcg_policy_enabled(q
, pol
))
840 /* preallocations for root blkg */
841 new_blkg
= blkg_alloc(&blkcg_root
, q
, GFP_KERNEL
);
845 preloaded
= !radix_tree_preload(GFP_KERNEL
);
847 blk_queue_bypass_start(q
);
850 * Make sure the root blkg exists and count the existing blkgs. As
851 * @q is bypassing at this point, blkg_lookup_create() can't be
852 * used. Open code it.
854 spin_lock_irq(q
->queue_lock
);
857 blkg
= __blkg_lookup(&blkcg_root
, q
, false);
861 blkg
= blkg_create(&blkcg_root
, q
, new_blkg
);
865 radix_tree_preload_end();
872 q
->root_rl
.blkg
= blkg
;
874 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
)
877 spin_unlock_irq(q
->queue_lock
);
879 /* allocate policy_data for all existing blkgs */
881 pd
= kzalloc_node(pol
->pd_size
, GFP_KERNEL
, q
->node
);
886 list_add_tail(&pd
->alloc_node
, &pds
);
890 * Install the allocated pds. With @q bypassing, no new blkg
891 * should have been created while the queue lock was dropped.
893 spin_lock_irq(q
->queue_lock
);
895 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
) {
896 if (WARN_ON(list_empty(&pds
))) {
897 /* umm... this shouldn't happen, just abort */
901 pd
= list_first_entry(&pds
, struct blkg_policy_data
, alloc_node
);
902 list_del_init(&pd
->alloc_node
);
904 /* grab blkcg lock too while installing @pd on @blkg */
905 spin_lock(&blkg
->blkcg
->lock
);
907 blkg
->pd
[pol
->plid
] = pd
;
909 pol
->pd_init_fn(blkg
);
911 spin_unlock(&blkg
->blkcg
->lock
);
914 __set_bit(pol
->plid
, q
->blkcg_pols
);
917 spin_unlock_irq(q
->queue_lock
);
919 blk_queue_bypass_end(q
);
920 list_for_each_entry_safe(pd
, n
, &pds
, alloc_node
)
924 EXPORT_SYMBOL_GPL(blkcg_activate_policy
);
927 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
928 * @q: request_queue of interest
929 * @pol: blkcg policy to deactivate
931 * Deactivate @pol on @q. Follows the same synchronization rules as
932 * blkcg_activate_policy().
934 void blkcg_deactivate_policy(struct request_queue
*q
,
935 const struct blkcg_policy
*pol
)
937 struct blkcg_gq
*blkg
;
939 if (!blkcg_policy_enabled(q
, pol
))
942 blk_queue_bypass_start(q
);
943 spin_lock_irq(q
->queue_lock
);
945 __clear_bit(pol
->plid
, q
->blkcg_pols
);
947 /* if no policy is left, no need for blkgs - shoot them down */
948 if (bitmap_empty(q
->blkcg_pols
, BLKCG_MAX_POLS
))
951 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
) {
952 /* grab blkcg lock too while removing @pd from @blkg */
953 spin_lock(&blkg
->blkcg
->lock
);
956 pol
->pd_exit_fn(blkg
);
958 kfree(blkg
->pd
[pol
->plid
]);
959 blkg
->pd
[pol
->plid
] = NULL
;
961 spin_unlock(&blkg
->blkcg
->lock
);
964 spin_unlock_irq(q
->queue_lock
);
965 blk_queue_bypass_end(q
);
967 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy
);
970 * blkcg_policy_register - register a blkcg policy
971 * @pol: blkcg policy to register
973 * Register @pol with blkcg core. Might sleep and @pol may be modified on
974 * successful registration. Returns 0 on success and -errno on failure.
976 int blkcg_policy_register(struct blkcg_policy
*pol
)
980 if (WARN_ON(pol
->pd_size
< sizeof(struct blkg_policy_data
)))
983 mutex_lock(&blkcg_pol_mutex
);
985 /* find an empty slot */
987 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++)
988 if (!blkcg_policy
[i
])
990 if (i
>= BLKCG_MAX_POLS
)
993 /* register and update blkgs */
995 blkcg_policy
[i
] = pol
;
997 /* everything is in place, add intf files for the new policy */
999 WARN_ON(cgroup_add_cftypes(&blkio_subsys
, pol
->cftypes
));
1002 mutex_unlock(&blkcg_pol_mutex
);
1005 EXPORT_SYMBOL_GPL(blkcg_policy_register
);
1008 * blkcg_policy_unregister - unregister a blkcg policy
1009 * @pol: blkcg policy to unregister
1011 * Undo blkcg_policy_register(@pol). Might sleep.
1013 void blkcg_policy_unregister(struct blkcg_policy
*pol
)
1015 mutex_lock(&blkcg_pol_mutex
);
1017 if (WARN_ON(blkcg_policy
[pol
->plid
] != pol
))
1020 /* kill the intf files first */
1022 cgroup_rm_cftypes(&blkio_subsys
, pol
->cftypes
);
1024 /* unregister and update blkgs */
1025 blkcg_policy
[pol
->plid
] = NULL
;
1027 mutex_unlock(&blkcg_pol_mutex
);
1029 EXPORT_SYMBOL_GPL(blkcg_policy_unregister
);