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 .cfq_leaf_weight
= 2 * CFQ_WEIGHT_DEFAULT
, };
31 EXPORT_SYMBOL_GPL(blkcg_root
);
33 static struct blkcg_policy
*blkcg_policy
[BLKCG_MAX_POLS
];
35 static bool blkcg_policy_enabled(struct request_queue
*q
,
36 const struct blkcg_policy
*pol
)
38 return pol
&& test_bit(pol
->plid
, q
->blkcg_pols
);
42 * blkg_free - free a blkg
45 * Free @blkg which may be partially allocated.
47 static void blkg_free(struct blkcg_gq
*blkg
)
54 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
55 struct blkcg_policy
*pol
= blkcg_policy
[i
];
56 struct blkg_policy_data
*pd
= blkg
->pd
[i
];
61 if (pol
&& pol
->pd_exit_fn
)
62 pol
->pd_exit_fn(blkg
);
67 blk_exit_rl(&blkg
->rl
);
72 * blkg_alloc - allocate a blkg
73 * @blkcg: block cgroup the new blkg is associated with
74 * @q: request_queue the new blkg is associated with
75 * @gfp_mask: allocation mask to use
77 * Allocate a new blkg assocating @blkcg and @q.
79 static struct blkcg_gq
*blkg_alloc(struct blkcg
*blkcg
, struct request_queue
*q
,
82 struct blkcg_gq
*blkg
;
85 /* alloc and init base part */
86 blkg
= kzalloc_node(sizeof(*blkg
), gfp_mask
, q
->node
);
91 INIT_LIST_HEAD(&blkg
->q_node
);
95 /* root blkg uses @q->root_rl, init rl only for !root blkgs */
96 if (blkcg
!= &blkcg_root
) {
97 if (blk_init_rl(&blkg
->rl
, q
, gfp_mask
))
102 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
103 struct blkcg_policy
*pol
= blkcg_policy
[i
];
104 struct blkg_policy_data
*pd
;
106 if (!blkcg_policy_enabled(q
, pol
))
109 /* alloc per-policy data and attach it to blkg */
110 pd
= kzalloc_node(pol
->pd_size
, gfp_mask
, q
->node
);
117 /* invoke per-policy init */
119 pol
->pd_init_fn(blkg
);
129 static struct blkcg_gq
*__blkg_lookup(struct blkcg
*blkcg
,
130 struct request_queue
*q
, bool update_hint
)
132 struct blkcg_gq
*blkg
;
134 blkg
= rcu_dereference(blkcg
->blkg_hint
);
135 if (blkg
&& blkg
->q
== q
)
139 * Hint didn't match. Look up from the radix tree. Note that the
140 * hint can only be updated under queue_lock as otherwise @blkg
141 * could have already been removed from blkg_tree. The caller is
142 * responsible for grabbing queue_lock if @update_hint.
144 blkg
= radix_tree_lookup(&blkcg
->blkg_tree
, q
->id
);
145 if (blkg
&& blkg
->q
== q
) {
147 lockdep_assert_held(q
->queue_lock
);
148 rcu_assign_pointer(blkcg
->blkg_hint
, blkg
);
157 * blkg_lookup - lookup blkg for the specified blkcg - q pair
158 * @blkcg: blkcg of interest
159 * @q: request_queue of interest
161 * Lookup blkg for the @blkcg - @q pair. This function should be called
162 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
163 * - see blk_queue_bypass_start() for details.
165 struct blkcg_gq
*blkg_lookup(struct blkcg
*blkcg
, struct request_queue
*q
)
167 WARN_ON_ONCE(!rcu_read_lock_held());
169 if (unlikely(blk_queue_bypass(q
)))
171 return __blkg_lookup(blkcg
, q
, false);
173 EXPORT_SYMBOL_GPL(blkg_lookup
);
176 * If @new_blkg is %NULL, this function tries to allocate a new one as
177 * necessary using %GFP_ATOMIC. @new_blkg is always consumed on return.
179 static struct blkcg_gq
*blkg_create(struct blkcg
*blkcg
,
180 struct request_queue
*q
,
181 struct blkcg_gq
*new_blkg
)
183 struct blkcg_gq
*blkg
;
186 WARN_ON_ONCE(!rcu_read_lock_held());
187 lockdep_assert_held(q
->queue_lock
);
189 /* blkg holds a reference to blkcg */
190 if (!css_tryget(&blkcg
->css
)) {
197 new_blkg
= blkg_alloc(blkcg
, q
, GFP_ATOMIC
);
198 if (unlikely(!new_blkg
)) {
205 /* link parent and insert */
206 if (blkcg_parent(blkcg
)) {
207 blkg
->parent
= __blkg_lookup(blkcg_parent(blkcg
), q
, false);
208 if (WARN_ON_ONCE(!blkg
->parent
)) {
209 blkg
= ERR_PTR(-EINVAL
);
212 blkg_get(blkg
->parent
);
215 spin_lock(&blkcg
->lock
);
216 ret
= radix_tree_insert(&blkcg
->blkg_tree
, q
->id
, blkg
);
218 hlist_add_head_rcu(&blkg
->blkcg_node
, &blkcg
->blkg_list
);
219 list_add(&blkg
->q_node
, &q
->blkg_list
);
221 spin_unlock(&blkcg
->lock
);
226 /* @blkg failed fully initialized, use the usual release path */
231 css_put(&blkcg
->css
);
238 * blkg_lookup_create - lookup blkg, try to create one if not there
239 * @blkcg: blkcg of interest
240 * @q: request_queue of interest
242 * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
243 * create one. blkg creation is performed recursively from blkcg_root such
244 * that all non-root blkg's have access to the parent blkg. This function
245 * should be called under RCU read lock and @q->queue_lock.
247 * Returns pointer to the looked up or created blkg on success, ERR_PTR()
248 * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not
249 * dead and bypassing, returns ERR_PTR(-EBUSY).
251 struct blkcg_gq
*blkg_lookup_create(struct blkcg
*blkcg
,
252 struct request_queue
*q
)
254 struct blkcg_gq
*blkg
;
256 WARN_ON_ONCE(!rcu_read_lock_held());
257 lockdep_assert_held(q
->queue_lock
);
260 * This could be the first entry point of blkcg implementation and
261 * we shouldn't allow anything to go through for a bypassing queue.
263 if (unlikely(blk_queue_bypass(q
)))
264 return ERR_PTR(blk_queue_dying(q
) ? -EINVAL
: -EBUSY
);
266 blkg
= __blkg_lookup(blkcg
, q
, true);
271 * Create blkgs walking down from blkcg_root to @blkcg, so that all
272 * non-root blkgs have access to their parents.
275 struct blkcg
*pos
= blkcg
;
276 struct blkcg
*parent
= blkcg_parent(blkcg
);
278 while (parent
&& !__blkg_lookup(parent
, q
, false)) {
280 parent
= blkcg_parent(parent
);
283 blkg
= blkg_create(pos
, q
, NULL
);
284 if (pos
== blkcg
|| IS_ERR(blkg
))
288 EXPORT_SYMBOL_GPL(blkg_lookup_create
);
290 static void blkg_destroy(struct blkcg_gq
*blkg
)
292 struct blkcg
*blkcg
= blkg
->blkcg
;
294 lockdep_assert_held(blkg
->q
->queue_lock
);
295 lockdep_assert_held(&blkcg
->lock
);
297 /* Something wrong if we are trying to remove same group twice */
298 WARN_ON_ONCE(list_empty(&blkg
->q_node
));
299 WARN_ON_ONCE(hlist_unhashed(&blkg
->blkcg_node
));
301 radix_tree_delete(&blkcg
->blkg_tree
, blkg
->q
->id
);
302 list_del_init(&blkg
->q_node
);
303 hlist_del_init_rcu(&blkg
->blkcg_node
);
306 * Both setting lookup hint to and clearing it from @blkg are done
307 * under queue_lock. If it's not pointing to @blkg now, it never
308 * will. Hint assignment itself can race safely.
310 if (rcu_dereference_raw(blkcg
->blkg_hint
) == blkg
)
311 rcu_assign_pointer(blkcg
->blkg_hint
, NULL
);
314 * Put the reference taken at the time of creation so that when all
315 * queues are gone, group can be destroyed.
321 * blkg_destroy_all - destroy all blkgs associated with a request_queue
322 * @q: request_queue of interest
324 * Destroy all blkgs associated with @q.
326 static void blkg_destroy_all(struct request_queue
*q
)
328 struct blkcg_gq
*blkg
, *n
;
330 lockdep_assert_held(q
->queue_lock
);
332 list_for_each_entry_safe(blkg
, n
, &q
->blkg_list
, q_node
) {
333 struct blkcg
*blkcg
= blkg
->blkcg
;
335 spin_lock(&blkcg
->lock
);
337 spin_unlock(&blkcg
->lock
);
341 * root blkg is destroyed. Just clear the pointer since
342 * root_rl does not take reference on root blkg.
345 q
->root_rl
.blkg
= NULL
;
348 static void blkg_rcu_free(struct rcu_head
*rcu_head
)
350 blkg_free(container_of(rcu_head
, struct blkcg_gq
, rcu_head
));
353 void __blkg_release(struct blkcg_gq
*blkg
)
355 /* release the blkcg and parent blkg refs this blkg has been holding */
356 css_put(&blkg
->blkcg
->css
);
358 blkg_put(blkg
->parent
);
361 * A group is freed in rcu manner. But having an rcu lock does not
362 * mean that one can access all the fields of blkg and assume these
363 * are valid. For example, don't try to follow throtl_data and
364 * request queue links.
366 * Having a reference to blkg under an rcu allows acess to only
367 * values local to groups like group stats and group rate limits
369 call_rcu(&blkg
->rcu_head
, blkg_rcu_free
);
371 EXPORT_SYMBOL_GPL(__blkg_release
);
374 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
375 * because the root blkg uses @q->root_rl instead of its own rl.
377 struct request_list
*__blk_queue_next_rl(struct request_list
*rl
,
378 struct request_queue
*q
)
380 struct list_head
*ent
;
381 struct blkcg_gq
*blkg
;
384 * Determine the current blkg list_head. The first entry is
385 * root_rl which is off @q->blkg_list and mapped to the head.
387 if (rl
== &q
->root_rl
) {
389 /* There are no more block groups, hence no request lists */
393 blkg
= container_of(rl
, struct blkcg_gq
, rl
);
397 /* walk to the next list_head, skip root blkcg */
399 if (ent
== &q
->root_blkg
->q_node
)
401 if (ent
== &q
->blkg_list
)
404 blkg
= container_of(ent
, struct blkcg_gq
, q_node
);
408 static int blkcg_reset_stats(struct cgroup
*cgroup
, struct cftype
*cftype
,
411 struct blkcg
*blkcg
= cgroup_to_blkcg(cgroup
);
412 struct blkcg_gq
*blkg
;
413 struct hlist_node
*n
;
416 mutex_lock(&blkcg_pol_mutex
);
417 spin_lock_irq(&blkcg
->lock
);
420 * Note that stat reset is racy - it doesn't synchronize against
421 * stat updates. This is a debug feature which shouldn't exist
422 * anyway. If you get hit by a race, retry.
424 hlist_for_each_entry(blkg
, n
, &blkcg
->blkg_list
, blkcg_node
) {
425 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
426 struct blkcg_policy
*pol
= blkcg_policy
[i
];
428 if (blkcg_policy_enabled(blkg
->q
, pol
) &&
429 pol
->pd_reset_stats_fn
)
430 pol
->pd_reset_stats_fn(blkg
);
434 spin_unlock_irq(&blkcg
->lock
);
435 mutex_unlock(&blkcg_pol_mutex
);
439 static const char *blkg_dev_name(struct blkcg_gq
*blkg
)
441 /* some drivers (floppy) instantiate a queue w/o disk registered */
442 if (blkg
->q
->backing_dev_info
.dev
)
443 return dev_name(blkg
->q
->backing_dev_info
.dev
);
448 * blkcg_print_blkgs - helper for printing per-blkg data
449 * @sf: seq_file to print to
450 * @blkcg: blkcg of interest
451 * @prfill: fill function to print out a blkg
452 * @pol: policy in question
453 * @data: data to be passed to @prfill
454 * @show_total: to print out sum of prfill return values or not
456 * This function invokes @prfill on each blkg of @blkcg if pd for the
457 * policy specified by @pol exists. @prfill is invoked with @sf, the
458 * policy data and @data. If @show_total is %true, the sum of the return
459 * values from @prfill is printed with "Total" label at the end.
461 * This is to be used to construct print functions for
462 * cftype->read_seq_string method.
464 void blkcg_print_blkgs(struct seq_file
*sf
, struct blkcg
*blkcg
,
465 u64 (*prfill
)(struct seq_file
*,
466 struct blkg_policy_data
*, int),
467 const struct blkcg_policy
*pol
, int data
,
470 struct blkcg_gq
*blkg
;
471 struct hlist_node
*n
;
474 spin_lock_irq(&blkcg
->lock
);
475 hlist_for_each_entry(blkg
, n
, &blkcg
->blkg_list
, blkcg_node
)
476 if (blkcg_policy_enabled(blkg
->q
, pol
))
477 total
+= prfill(sf
, blkg
->pd
[pol
->plid
], data
);
478 spin_unlock_irq(&blkcg
->lock
);
481 seq_printf(sf
, "Total %llu\n", (unsigned long long)total
);
483 EXPORT_SYMBOL_GPL(blkcg_print_blkgs
);
486 * __blkg_prfill_u64 - prfill helper for a single u64 value
487 * @sf: seq_file to print to
488 * @pd: policy private data of interest
491 * Print @v to @sf for the device assocaited with @pd.
493 u64
__blkg_prfill_u64(struct seq_file
*sf
, struct blkg_policy_data
*pd
, u64 v
)
495 const char *dname
= blkg_dev_name(pd
->blkg
);
500 seq_printf(sf
, "%s %llu\n", dname
, (unsigned long long)v
);
503 EXPORT_SYMBOL_GPL(__blkg_prfill_u64
);
506 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
507 * @sf: seq_file to print to
508 * @pd: policy private data of interest
509 * @rwstat: rwstat to print
511 * Print @rwstat to @sf for the device assocaited with @pd.
513 u64
__blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
514 const struct blkg_rwstat
*rwstat
)
516 static const char *rwstr
[] = {
517 [BLKG_RWSTAT_READ
] = "Read",
518 [BLKG_RWSTAT_WRITE
] = "Write",
519 [BLKG_RWSTAT_SYNC
] = "Sync",
520 [BLKG_RWSTAT_ASYNC
] = "Async",
522 const char *dname
= blkg_dev_name(pd
->blkg
);
529 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
530 seq_printf(sf
, "%s %s %llu\n", dname
, rwstr
[i
],
531 (unsigned long long)rwstat
->cnt
[i
]);
533 v
= rwstat
->cnt
[BLKG_RWSTAT_READ
] + rwstat
->cnt
[BLKG_RWSTAT_WRITE
];
534 seq_printf(sf
, "%s Total %llu\n", dname
, (unsigned long long)v
);
539 * blkg_prfill_stat - prfill callback for blkg_stat
540 * @sf: seq_file to print to
541 * @pd: policy private data of interest
542 * @off: offset to the blkg_stat in @pd
544 * prfill callback for printing a blkg_stat.
546 u64
blkg_prfill_stat(struct seq_file
*sf
, struct blkg_policy_data
*pd
, int off
)
548 return __blkg_prfill_u64(sf
, pd
, blkg_stat_read((void *)pd
+ off
));
550 EXPORT_SYMBOL_GPL(blkg_prfill_stat
);
553 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
554 * @sf: seq_file to print to
555 * @pd: policy private data of interest
556 * @off: offset to the blkg_rwstat in @pd
558 * prfill callback for printing a blkg_rwstat.
560 u64
blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
563 struct blkg_rwstat rwstat
= blkg_rwstat_read((void *)pd
+ off
);
565 return __blkg_prfill_rwstat(sf
, pd
, &rwstat
);
567 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat
);
570 * blkg_conf_prep - parse and prepare for per-blkg config update
571 * @blkcg: target block cgroup
572 * @pol: target policy
573 * @input: input string
574 * @ctx: blkg_conf_ctx to be filled
576 * Parse per-blkg config update from @input and initialize @ctx with the
577 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
578 * value. This function returns with RCU read lock and queue lock held and
579 * must be paired with blkg_conf_finish().
581 int blkg_conf_prep(struct blkcg
*blkcg
, const struct blkcg_policy
*pol
,
582 const char *input
, struct blkg_conf_ctx
*ctx
)
583 __acquires(rcu
) __acquires(disk
->queue
->queue_lock
)
585 struct gendisk
*disk
;
586 struct blkcg_gq
*blkg
;
587 unsigned int major
, minor
;
588 unsigned long long v
;
591 if (sscanf(input
, "%u:%u %llu", &major
, &minor
, &v
) != 3)
594 disk
= get_gendisk(MKDEV(major
, minor
), &part
);
599 spin_lock_irq(disk
->queue
->queue_lock
);
601 if (blkcg_policy_enabled(disk
->queue
, pol
))
602 blkg
= blkg_lookup_create(blkcg
, disk
->queue
);
604 blkg
= ERR_PTR(-EINVAL
);
609 spin_unlock_irq(disk
->queue
->queue_lock
);
612 * If queue was bypassing, we should retry. Do so after a
613 * short msleep(). It isn't strictly necessary but queue
614 * can be bypassing for some time and it's always nice to
615 * avoid busy looping.
619 ret
= restart_syscall();
629 EXPORT_SYMBOL_GPL(blkg_conf_prep
);
632 * blkg_conf_finish - finish up per-blkg config update
633 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
635 * Finish up after per-blkg config update. This function must be paired
636 * with blkg_conf_prep().
638 void blkg_conf_finish(struct blkg_conf_ctx
*ctx
)
639 __releases(ctx
->disk
->queue
->queue_lock
) __releases(rcu
)
641 spin_unlock_irq(ctx
->disk
->queue
->queue_lock
);
645 EXPORT_SYMBOL_GPL(blkg_conf_finish
);
647 struct cftype blkcg_files
[] = {
649 .name
= "reset_stats",
650 .write_u64
= blkcg_reset_stats
,
656 * blkcg_css_offline - cgroup css_offline callback
657 * @cgroup: cgroup of interest
659 * This function is called when @cgroup is about to go away and responsible
660 * for shooting down all blkgs associated with @cgroup. blkgs should be
661 * removed while holding both q and blkcg locks. As blkcg lock is nested
662 * inside q lock, this function performs reverse double lock dancing.
664 * This is the blkcg counterpart of ioc_release_fn().
666 static void blkcg_css_offline(struct cgroup
*cgroup
)
668 struct blkcg
*blkcg
= cgroup_to_blkcg(cgroup
);
670 spin_lock_irq(&blkcg
->lock
);
672 while (!hlist_empty(&blkcg
->blkg_list
)) {
673 struct blkcg_gq
*blkg
= hlist_entry(blkcg
->blkg_list
.first
,
674 struct blkcg_gq
, blkcg_node
);
675 struct request_queue
*q
= blkg
->q
;
677 if (spin_trylock(q
->queue_lock
)) {
679 spin_unlock(q
->queue_lock
);
681 spin_unlock_irq(&blkcg
->lock
);
683 spin_lock_irq(&blkcg
->lock
);
687 spin_unlock_irq(&blkcg
->lock
);
690 static void blkcg_css_free(struct cgroup
*cgroup
)
692 struct blkcg
*blkcg
= cgroup_to_blkcg(cgroup
);
694 if (blkcg
!= &blkcg_root
)
698 static struct cgroup_subsys_state
*blkcg_css_alloc(struct cgroup
*cgroup
)
700 static atomic64_t id_seq
= ATOMIC64_INIT(0);
702 struct cgroup
*parent
= cgroup
->parent
;
709 blkcg
= kzalloc(sizeof(*blkcg
), GFP_KERNEL
);
711 return ERR_PTR(-ENOMEM
);
713 blkcg
->cfq_weight
= CFQ_WEIGHT_DEFAULT
;
714 blkcg
->cfq_leaf_weight
= CFQ_WEIGHT_DEFAULT
;
715 blkcg
->id
= atomic64_inc_return(&id_seq
); /* root is 0, start from 1 */
717 spin_lock_init(&blkcg
->lock
);
718 INIT_RADIX_TREE(&blkcg
->blkg_tree
, GFP_ATOMIC
);
719 INIT_HLIST_HEAD(&blkcg
->blkg_list
);
725 * blkcg_init_queue - initialize blkcg part of request queue
726 * @q: request_queue to initialize
728 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
729 * part of new request_queue @q.
732 * 0 on success, -errno on failure.
734 int blkcg_init_queue(struct request_queue
*q
)
738 return blk_throtl_init(q
);
742 * blkcg_drain_queue - drain blkcg part of request_queue
743 * @q: request_queue to drain
745 * Called from blk_drain_queue(). Responsible for draining blkcg part.
747 void blkcg_drain_queue(struct request_queue
*q
)
749 lockdep_assert_held(q
->queue_lock
);
755 * blkcg_exit_queue - exit and release blkcg part of request_queue
756 * @q: request_queue being released
758 * Called from blk_release_queue(). Responsible for exiting blkcg part.
760 void blkcg_exit_queue(struct request_queue
*q
)
762 spin_lock_irq(q
->queue_lock
);
764 spin_unlock_irq(q
->queue_lock
);
770 * We cannot support shared io contexts, as we have no mean to support
771 * two tasks with the same ioc in two different groups without major rework
772 * of the main cic data structures. For now we allow a task to change
773 * its cgroup only if it's the only owner of its ioc.
775 static int blkcg_can_attach(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
)
777 struct task_struct
*task
;
778 struct io_context
*ioc
;
781 /* task_lock() is needed to avoid races with exit_io_context() */
782 cgroup_taskset_for_each(task
, cgrp
, tset
) {
784 ioc
= task
->io_context
;
785 if (ioc
&& atomic_read(&ioc
->nr_tasks
) > 1)
794 struct cgroup_subsys blkio_subsys
= {
796 .css_alloc
= blkcg_css_alloc
,
797 .css_offline
= blkcg_css_offline
,
798 .css_free
= blkcg_css_free
,
799 .can_attach
= blkcg_can_attach
,
800 .subsys_id
= blkio_subsys_id
,
801 .base_cftypes
= blkcg_files
,
802 .module
= THIS_MODULE
,
805 * blkio subsystem is utterly broken in terms of hierarchy support.
806 * It treats all cgroups equally regardless of where they're
807 * located in the hierarchy - all cgroups are treated as if they're
808 * right below the root. Fix it and remove the following.
810 .broken_hierarchy
= true,
812 EXPORT_SYMBOL_GPL(blkio_subsys
);
815 * blkcg_activate_policy - activate a blkcg policy on a request_queue
816 * @q: request_queue of interest
817 * @pol: blkcg policy to activate
819 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
820 * bypass mode to populate its blkgs with policy_data for @pol.
822 * Activation happens with @q bypassed, so nobody would be accessing blkgs
823 * from IO path. Update of each blkg is protected by both queue and blkcg
824 * locks so that holding either lock and testing blkcg_policy_enabled() is
825 * always enough for dereferencing policy data.
827 * The caller is responsible for synchronizing [de]activations and policy
828 * [un]registerations. Returns 0 on success, -errno on failure.
830 int blkcg_activate_policy(struct request_queue
*q
,
831 const struct blkcg_policy
*pol
)
834 struct blkcg_gq
*blkg
, *new_blkg
;
835 struct blkg_policy_data
*pd
, *n
;
839 if (blkcg_policy_enabled(q
, pol
))
842 /* preallocations for root blkg */
843 new_blkg
= blkg_alloc(&blkcg_root
, q
, GFP_KERNEL
);
847 preloaded
= !radix_tree_preload(GFP_KERNEL
);
849 blk_queue_bypass_start(q
);
852 * Make sure the root blkg exists and count the existing blkgs. As
853 * @q is bypassing at this point, blkg_lookup_create() can't be
854 * used. Open code it.
856 spin_lock_irq(q
->queue_lock
);
859 blkg
= __blkg_lookup(&blkcg_root
, q
, false);
863 blkg
= blkg_create(&blkcg_root
, q
, new_blkg
);
867 radix_tree_preload_end();
874 q
->root_rl
.blkg
= blkg
;
876 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
)
879 spin_unlock_irq(q
->queue_lock
);
881 /* allocate policy_data for all existing blkgs */
883 pd
= kzalloc_node(pol
->pd_size
, GFP_KERNEL
, q
->node
);
888 list_add_tail(&pd
->alloc_node
, &pds
);
892 * Install the allocated pds. With @q bypassing, no new blkg
893 * should have been created while the queue lock was dropped.
895 spin_lock_irq(q
->queue_lock
);
897 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
) {
898 if (WARN_ON(list_empty(&pds
))) {
899 /* umm... this shouldn't happen, just abort */
903 pd
= list_first_entry(&pds
, struct blkg_policy_data
, alloc_node
);
904 list_del_init(&pd
->alloc_node
);
906 /* grab blkcg lock too while installing @pd on @blkg */
907 spin_lock(&blkg
->blkcg
->lock
);
909 blkg
->pd
[pol
->plid
] = pd
;
911 pol
->pd_init_fn(blkg
);
913 spin_unlock(&blkg
->blkcg
->lock
);
916 __set_bit(pol
->plid
, q
->blkcg_pols
);
919 spin_unlock_irq(q
->queue_lock
);
921 blk_queue_bypass_end(q
);
922 list_for_each_entry_safe(pd
, n
, &pds
, alloc_node
)
926 EXPORT_SYMBOL_GPL(blkcg_activate_policy
);
929 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
930 * @q: request_queue of interest
931 * @pol: blkcg policy to deactivate
933 * Deactivate @pol on @q. Follows the same synchronization rules as
934 * blkcg_activate_policy().
936 void blkcg_deactivate_policy(struct request_queue
*q
,
937 const struct blkcg_policy
*pol
)
939 struct blkcg_gq
*blkg
;
941 if (!blkcg_policy_enabled(q
, pol
))
944 blk_queue_bypass_start(q
);
945 spin_lock_irq(q
->queue_lock
);
947 __clear_bit(pol
->plid
, q
->blkcg_pols
);
949 /* if no policy is left, no need for blkgs - shoot them down */
950 if (bitmap_empty(q
->blkcg_pols
, BLKCG_MAX_POLS
))
953 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
) {
954 /* grab blkcg lock too while removing @pd from @blkg */
955 spin_lock(&blkg
->blkcg
->lock
);
958 pol
->pd_exit_fn(blkg
);
960 kfree(blkg
->pd
[pol
->plid
]);
961 blkg
->pd
[pol
->plid
] = NULL
;
963 spin_unlock(&blkg
->blkcg
->lock
);
966 spin_unlock_irq(q
->queue_lock
);
967 blk_queue_bypass_end(q
);
969 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy
);
972 * blkcg_policy_register - register a blkcg policy
973 * @pol: blkcg policy to register
975 * Register @pol with blkcg core. Might sleep and @pol may be modified on
976 * successful registration. Returns 0 on success and -errno on failure.
978 int blkcg_policy_register(struct blkcg_policy
*pol
)
982 if (WARN_ON(pol
->pd_size
< sizeof(struct blkg_policy_data
)))
985 mutex_lock(&blkcg_pol_mutex
);
987 /* find an empty slot */
989 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++)
990 if (!blkcg_policy
[i
])
992 if (i
>= BLKCG_MAX_POLS
)
995 /* register and update blkgs */
997 blkcg_policy
[i
] = pol
;
999 /* everything is in place, add intf files for the new policy */
1001 WARN_ON(cgroup_add_cftypes(&blkio_subsys
, pol
->cftypes
));
1004 mutex_unlock(&blkcg_pol_mutex
);
1007 EXPORT_SYMBOL_GPL(blkcg_policy_register
);
1010 * blkcg_policy_unregister - unregister a blkcg policy
1011 * @pol: blkcg policy to unregister
1013 * Undo blkcg_policy_register(@pol). Might sleep.
1015 void blkcg_policy_unregister(struct blkcg_policy
*pol
)
1017 mutex_lock(&blkcg_pol_mutex
);
1019 if (WARN_ON(blkcg_policy
[pol
->plid
] != pol
))
1022 /* kill the intf files first */
1024 cgroup_rm_cftypes(&blkio_subsys
, pol
->cftypes
);
1026 /* unregister and update blkgs */
1027 blkcg_policy
[pol
->plid
] = NULL
;
1029 mutex_unlock(&blkcg_pol_mutex
);
1031 EXPORT_SYMBOL_GPL(blkcg_policy_unregister
);