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
);
118 /* invoke per-policy init */
120 pol
->pd_init_fn(blkg
);
130 static struct blkcg_gq
*__blkg_lookup(struct blkcg
*blkcg
,
131 struct request_queue
*q
, bool update_hint
)
133 struct blkcg_gq
*blkg
;
135 blkg
= rcu_dereference(blkcg
->blkg_hint
);
136 if (blkg
&& blkg
->q
== q
)
140 * Hint didn't match. Look up from the radix tree. Note that the
141 * hint can only be updated under queue_lock as otherwise @blkg
142 * could have already been removed from blkg_tree. The caller is
143 * responsible for grabbing queue_lock if @update_hint.
145 blkg
= radix_tree_lookup(&blkcg
->blkg_tree
, q
->id
);
146 if (blkg
&& blkg
->q
== q
) {
148 lockdep_assert_held(q
->queue_lock
);
149 rcu_assign_pointer(blkcg
->blkg_hint
, blkg
);
158 * blkg_lookup - lookup blkg for the specified blkcg - q pair
159 * @blkcg: blkcg of interest
160 * @q: request_queue of interest
162 * Lookup blkg for the @blkcg - @q pair. This function should be called
163 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
164 * - see blk_queue_bypass_start() for details.
166 struct blkcg_gq
*blkg_lookup(struct blkcg
*blkcg
, struct request_queue
*q
)
168 WARN_ON_ONCE(!rcu_read_lock_held());
170 if (unlikely(blk_queue_bypass(q
)))
172 return __blkg_lookup(blkcg
, q
, false);
174 EXPORT_SYMBOL_GPL(blkg_lookup
);
177 * If @new_blkg is %NULL, this function tries to allocate a new one as
178 * necessary using %GFP_ATOMIC. @new_blkg is always consumed on return.
180 static struct blkcg_gq
*blkg_create(struct blkcg
*blkcg
,
181 struct request_queue
*q
,
182 struct blkcg_gq
*new_blkg
)
184 struct blkcg_gq
*blkg
;
187 WARN_ON_ONCE(!rcu_read_lock_held());
188 lockdep_assert_held(q
->queue_lock
);
190 /* blkg holds a reference to blkcg */
191 if (!css_tryget(&blkcg
->css
)) {
198 new_blkg
= blkg_alloc(blkcg
, q
, GFP_ATOMIC
);
199 if (unlikely(!new_blkg
)) {
206 /* link parent and insert */
207 if (blkcg_parent(blkcg
)) {
208 blkg
->parent
= __blkg_lookup(blkcg_parent(blkcg
), q
, false);
209 if (WARN_ON_ONCE(!blkg
->parent
)) {
210 blkg
= ERR_PTR(-EINVAL
);
213 blkg_get(blkg
->parent
);
216 spin_lock(&blkcg
->lock
);
217 ret
= radix_tree_insert(&blkcg
->blkg_tree
, q
->id
, blkg
);
219 hlist_add_head_rcu(&blkg
->blkcg_node
, &blkcg
->blkg_list
);
220 list_add(&blkg
->q_node
, &q
->blkg_list
);
222 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
223 struct blkcg_policy
*pol
= blkcg_policy
[i
];
225 if (blkg
->pd
[i
] && pol
->pd_online_fn
)
226 pol
->pd_online_fn(blkg
);
230 spin_unlock(&blkcg
->lock
);
235 /* @blkg failed fully initialized, use the usual release path */
240 css_put(&blkcg
->css
);
247 * blkg_lookup_create - lookup blkg, try to create one if not there
248 * @blkcg: blkcg of interest
249 * @q: request_queue of interest
251 * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
252 * create one. blkg creation is performed recursively from blkcg_root such
253 * that all non-root blkg's have access to the parent blkg. This function
254 * should be called under RCU read lock and @q->queue_lock.
256 * Returns pointer to the looked up or created blkg on success, ERR_PTR()
257 * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not
258 * dead and bypassing, returns ERR_PTR(-EBUSY).
260 struct blkcg_gq
*blkg_lookup_create(struct blkcg
*blkcg
,
261 struct request_queue
*q
)
263 struct blkcg_gq
*blkg
;
265 WARN_ON_ONCE(!rcu_read_lock_held());
266 lockdep_assert_held(q
->queue_lock
);
269 * This could be the first entry point of blkcg implementation and
270 * we shouldn't allow anything to go through for a bypassing queue.
272 if (unlikely(blk_queue_bypass(q
)))
273 return ERR_PTR(blk_queue_dying(q
) ? -EINVAL
: -EBUSY
);
275 blkg
= __blkg_lookup(blkcg
, q
, true);
280 * Create blkgs walking down from blkcg_root to @blkcg, so that all
281 * non-root blkgs have access to their parents.
284 struct blkcg
*pos
= blkcg
;
285 struct blkcg
*parent
= blkcg_parent(blkcg
);
287 while (parent
&& !__blkg_lookup(parent
, q
, false)) {
289 parent
= blkcg_parent(parent
);
292 blkg
= blkg_create(pos
, q
, NULL
);
293 if (pos
== blkcg
|| IS_ERR(blkg
))
297 EXPORT_SYMBOL_GPL(blkg_lookup_create
);
299 static void blkg_destroy(struct blkcg_gq
*blkg
)
301 struct blkcg
*blkcg
= blkg
->blkcg
;
304 lockdep_assert_held(blkg
->q
->queue_lock
);
305 lockdep_assert_held(&blkcg
->lock
);
307 /* Something wrong if we are trying to remove same group twice */
308 WARN_ON_ONCE(list_empty(&blkg
->q_node
));
309 WARN_ON_ONCE(hlist_unhashed(&blkg
->blkcg_node
));
311 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
312 struct blkcg_policy
*pol
= blkcg_policy
[i
];
314 if (blkg
->pd
[i
] && pol
->pd_offline_fn
)
315 pol
->pd_offline_fn(blkg
);
317 blkg
->online
= false;
319 radix_tree_delete(&blkcg
->blkg_tree
, blkg
->q
->id
);
320 list_del_init(&blkg
->q_node
);
321 hlist_del_init_rcu(&blkg
->blkcg_node
);
324 * Both setting lookup hint to and clearing it from @blkg are done
325 * under queue_lock. If it's not pointing to @blkg now, it never
326 * will. Hint assignment itself can race safely.
328 if (rcu_dereference_raw(blkcg
->blkg_hint
) == blkg
)
329 rcu_assign_pointer(blkcg
->blkg_hint
, NULL
);
332 * Put the reference taken at the time of creation so that when all
333 * queues are gone, group can be destroyed.
339 * blkg_destroy_all - destroy all blkgs associated with a request_queue
340 * @q: request_queue of interest
342 * Destroy all blkgs associated with @q.
344 static void blkg_destroy_all(struct request_queue
*q
)
346 struct blkcg_gq
*blkg
, *n
;
348 lockdep_assert_held(q
->queue_lock
);
350 list_for_each_entry_safe(blkg
, n
, &q
->blkg_list
, q_node
) {
351 struct blkcg
*blkcg
= blkg
->blkcg
;
353 spin_lock(&blkcg
->lock
);
355 spin_unlock(&blkcg
->lock
);
359 * root blkg is destroyed. Just clear the pointer since
360 * root_rl does not take reference on root blkg.
363 q
->root_rl
.blkg
= NULL
;
366 static void blkg_rcu_free(struct rcu_head
*rcu_head
)
368 blkg_free(container_of(rcu_head
, struct blkcg_gq
, rcu_head
));
371 void __blkg_release(struct blkcg_gq
*blkg
)
373 /* release the blkcg and parent blkg refs this blkg has been holding */
374 css_put(&blkg
->blkcg
->css
);
376 blkg_put(blkg
->parent
);
379 * A group is freed in rcu manner. But having an rcu lock does not
380 * mean that one can access all the fields of blkg and assume these
381 * are valid. For example, don't try to follow throtl_data and
382 * request queue links.
384 * Having a reference to blkg under an rcu allows acess to only
385 * values local to groups like group stats and group rate limits
387 call_rcu(&blkg
->rcu_head
, blkg_rcu_free
);
389 EXPORT_SYMBOL_GPL(__blkg_release
);
392 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
393 * because the root blkg uses @q->root_rl instead of its own rl.
395 struct request_list
*__blk_queue_next_rl(struct request_list
*rl
,
396 struct request_queue
*q
)
398 struct list_head
*ent
;
399 struct blkcg_gq
*blkg
;
402 * Determine the current blkg list_head. The first entry is
403 * root_rl which is off @q->blkg_list and mapped to the head.
405 if (rl
== &q
->root_rl
) {
407 /* There are no more block groups, hence no request lists */
411 blkg
= container_of(rl
, struct blkcg_gq
, rl
);
415 /* walk to the next list_head, skip root blkcg */
417 if (ent
== &q
->root_blkg
->q_node
)
419 if (ent
== &q
->blkg_list
)
422 blkg
= container_of(ent
, struct blkcg_gq
, q_node
);
426 static int blkcg_reset_stats(struct cgroup
*cgroup
, struct cftype
*cftype
,
429 struct blkcg
*blkcg
= cgroup_to_blkcg(cgroup
);
430 struct blkcg_gq
*blkg
;
431 struct hlist_node
*n
;
434 mutex_lock(&blkcg_pol_mutex
);
435 spin_lock_irq(&blkcg
->lock
);
438 * Note that stat reset is racy - it doesn't synchronize against
439 * stat updates. This is a debug feature which shouldn't exist
440 * anyway. If you get hit by a race, retry.
442 hlist_for_each_entry(blkg
, n
, &blkcg
->blkg_list
, blkcg_node
) {
443 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
444 struct blkcg_policy
*pol
= blkcg_policy
[i
];
446 if (blkcg_policy_enabled(blkg
->q
, pol
) &&
447 pol
->pd_reset_stats_fn
)
448 pol
->pd_reset_stats_fn(blkg
);
452 spin_unlock_irq(&blkcg
->lock
);
453 mutex_unlock(&blkcg_pol_mutex
);
457 static const char *blkg_dev_name(struct blkcg_gq
*blkg
)
459 /* some drivers (floppy) instantiate a queue w/o disk registered */
460 if (blkg
->q
->backing_dev_info
.dev
)
461 return dev_name(blkg
->q
->backing_dev_info
.dev
);
466 * blkcg_print_blkgs - helper for printing per-blkg data
467 * @sf: seq_file to print to
468 * @blkcg: blkcg of interest
469 * @prfill: fill function to print out a blkg
470 * @pol: policy in question
471 * @data: data to be passed to @prfill
472 * @show_total: to print out sum of prfill return values or not
474 * This function invokes @prfill on each blkg of @blkcg if pd for the
475 * policy specified by @pol exists. @prfill is invoked with @sf, the
476 * policy data and @data. If @show_total is %true, the sum of the return
477 * values from @prfill is printed with "Total" label at the end.
479 * This is to be used to construct print functions for
480 * cftype->read_seq_string method.
482 void blkcg_print_blkgs(struct seq_file
*sf
, struct blkcg
*blkcg
,
483 u64 (*prfill
)(struct seq_file
*,
484 struct blkg_policy_data
*, int),
485 const struct blkcg_policy
*pol
, int data
,
488 struct blkcg_gq
*blkg
;
489 struct hlist_node
*n
;
492 spin_lock_irq(&blkcg
->lock
);
493 hlist_for_each_entry(blkg
, n
, &blkcg
->blkg_list
, blkcg_node
)
494 if (blkcg_policy_enabled(blkg
->q
, pol
))
495 total
+= prfill(sf
, blkg
->pd
[pol
->plid
], data
);
496 spin_unlock_irq(&blkcg
->lock
);
499 seq_printf(sf
, "Total %llu\n", (unsigned long long)total
);
501 EXPORT_SYMBOL_GPL(blkcg_print_blkgs
);
504 * __blkg_prfill_u64 - prfill helper for a single u64 value
505 * @sf: seq_file to print to
506 * @pd: policy private data of interest
509 * Print @v to @sf for the device assocaited with @pd.
511 u64
__blkg_prfill_u64(struct seq_file
*sf
, struct blkg_policy_data
*pd
, u64 v
)
513 const char *dname
= blkg_dev_name(pd
->blkg
);
518 seq_printf(sf
, "%s %llu\n", dname
, (unsigned long long)v
);
521 EXPORT_SYMBOL_GPL(__blkg_prfill_u64
);
524 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
525 * @sf: seq_file to print to
526 * @pd: policy private data of interest
527 * @rwstat: rwstat to print
529 * Print @rwstat to @sf for the device assocaited with @pd.
531 u64
__blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
532 const struct blkg_rwstat
*rwstat
)
534 static const char *rwstr
[] = {
535 [BLKG_RWSTAT_READ
] = "Read",
536 [BLKG_RWSTAT_WRITE
] = "Write",
537 [BLKG_RWSTAT_SYNC
] = "Sync",
538 [BLKG_RWSTAT_ASYNC
] = "Async",
540 const char *dname
= blkg_dev_name(pd
->blkg
);
547 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
548 seq_printf(sf
, "%s %s %llu\n", dname
, rwstr
[i
],
549 (unsigned long long)rwstat
->cnt
[i
]);
551 v
= rwstat
->cnt
[BLKG_RWSTAT_READ
] + rwstat
->cnt
[BLKG_RWSTAT_WRITE
];
552 seq_printf(sf
, "%s Total %llu\n", dname
, (unsigned long long)v
);
557 * blkg_prfill_stat - prfill callback for blkg_stat
558 * @sf: seq_file to print to
559 * @pd: policy private data of interest
560 * @off: offset to the blkg_stat in @pd
562 * prfill callback for printing a blkg_stat.
564 u64
blkg_prfill_stat(struct seq_file
*sf
, struct blkg_policy_data
*pd
, int off
)
566 return __blkg_prfill_u64(sf
, pd
, blkg_stat_read((void *)pd
+ off
));
568 EXPORT_SYMBOL_GPL(blkg_prfill_stat
);
571 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
572 * @sf: seq_file to print to
573 * @pd: policy private data of interest
574 * @off: offset to the blkg_rwstat in @pd
576 * prfill callback for printing a blkg_rwstat.
578 u64
blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
581 struct blkg_rwstat rwstat
= blkg_rwstat_read((void *)pd
+ off
);
583 return __blkg_prfill_rwstat(sf
, pd
, &rwstat
);
585 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat
);
588 * blkg_conf_prep - parse and prepare for per-blkg config update
589 * @blkcg: target block cgroup
590 * @pol: target policy
591 * @input: input string
592 * @ctx: blkg_conf_ctx to be filled
594 * Parse per-blkg config update from @input and initialize @ctx with the
595 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
596 * value. This function returns with RCU read lock and queue lock held and
597 * must be paired with blkg_conf_finish().
599 int blkg_conf_prep(struct blkcg
*blkcg
, const struct blkcg_policy
*pol
,
600 const char *input
, struct blkg_conf_ctx
*ctx
)
601 __acquires(rcu
) __acquires(disk
->queue
->queue_lock
)
603 struct gendisk
*disk
;
604 struct blkcg_gq
*blkg
;
605 unsigned int major
, minor
;
606 unsigned long long v
;
609 if (sscanf(input
, "%u:%u %llu", &major
, &minor
, &v
) != 3)
612 disk
= get_gendisk(MKDEV(major
, minor
), &part
);
617 spin_lock_irq(disk
->queue
->queue_lock
);
619 if (blkcg_policy_enabled(disk
->queue
, pol
))
620 blkg
= blkg_lookup_create(blkcg
, disk
->queue
);
622 blkg
= ERR_PTR(-EINVAL
);
627 spin_unlock_irq(disk
->queue
->queue_lock
);
630 * If queue was bypassing, we should retry. Do so after a
631 * short msleep(). It isn't strictly necessary but queue
632 * can be bypassing for some time and it's always nice to
633 * avoid busy looping.
637 ret
= restart_syscall();
647 EXPORT_SYMBOL_GPL(blkg_conf_prep
);
650 * blkg_conf_finish - finish up per-blkg config update
651 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
653 * Finish up after per-blkg config update. This function must be paired
654 * with blkg_conf_prep().
656 void blkg_conf_finish(struct blkg_conf_ctx
*ctx
)
657 __releases(ctx
->disk
->queue
->queue_lock
) __releases(rcu
)
659 spin_unlock_irq(ctx
->disk
->queue
->queue_lock
);
663 EXPORT_SYMBOL_GPL(blkg_conf_finish
);
665 struct cftype blkcg_files
[] = {
667 .name
= "reset_stats",
668 .write_u64
= blkcg_reset_stats
,
674 * blkcg_css_offline - cgroup css_offline callback
675 * @cgroup: cgroup of interest
677 * This function is called when @cgroup is about to go away and responsible
678 * for shooting down all blkgs associated with @cgroup. blkgs should be
679 * removed while holding both q and blkcg locks. As blkcg lock is nested
680 * inside q lock, this function performs reverse double lock dancing.
682 * This is the blkcg counterpart of ioc_release_fn().
684 static void blkcg_css_offline(struct cgroup
*cgroup
)
686 struct blkcg
*blkcg
= cgroup_to_blkcg(cgroup
);
688 spin_lock_irq(&blkcg
->lock
);
690 while (!hlist_empty(&blkcg
->blkg_list
)) {
691 struct blkcg_gq
*blkg
= hlist_entry(blkcg
->blkg_list
.first
,
692 struct blkcg_gq
, blkcg_node
);
693 struct request_queue
*q
= blkg
->q
;
695 if (spin_trylock(q
->queue_lock
)) {
697 spin_unlock(q
->queue_lock
);
699 spin_unlock_irq(&blkcg
->lock
);
701 spin_lock_irq(&blkcg
->lock
);
705 spin_unlock_irq(&blkcg
->lock
);
708 static void blkcg_css_free(struct cgroup
*cgroup
)
710 struct blkcg
*blkcg
= cgroup_to_blkcg(cgroup
);
712 if (blkcg
!= &blkcg_root
)
716 static struct cgroup_subsys_state
*blkcg_css_alloc(struct cgroup
*cgroup
)
718 static atomic64_t id_seq
= ATOMIC64_INIT(0);
720 struct cgroup
*parent
= cgroup
->parent
;
727 blkcg
= kzalloc(sizeof(*blkcg
), GFP_KERNEL
);
729 return ERR_PTR(-ENOMEM
);
731 blkcg
->cfq_weight
= CFQ_WEIGHT_DEFAULT
;
732 blkcg
->cfq_leaf_weight
= CFQ_WEIGHT_DEFAULT
;
733 blkcg
->id
= atomic64_inc_return(&id_seq
); /* root is 0, start from 1 */
735 spin_lock_init(&blkcg
->lock
);
736 INIT_RADIX_TREE(&blkcg
->blkg_tree
, GFP_ATOMIC
);
737 INIT_HLIST_HEAD(&blkcg
->blkg_list
);
743 * blkcg_init_queue - initialize blkcg part of request queue
744 * @q: request_queue to initialize
746 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
747 * part of new request_queue @q.
750 * 0 on success, -errno on failure.
752 int blkcg_init_queue(struct request_queue
*q
)
756 return blk_throtl_init(q
);
760 * blkcg_drain_queue - drain blkcg part of request_queue
761 * @q: request_queue to drain
763 * Called from blk_drain_queue(). Responsible for draining blkcg part.
765 void blkcg_drain_queue(struct request_queue
*q
)
767 lockdep_assert_held(q
->queue_lock
);
773 * blkcg_exit_queue - exit and release blkcg part of request_queue
774 * @q: request_queue being released
776 * Called from blk_release_queue(). Responsible for exiting blkcg part.
778 void blkcg_exit_queue(struct request_queue
*q
)
780 spin_lock_irq(q
->queue_lock
);
782 spin_unlock_irq(q
->queue_lock
);
788 * We cannot support shared io contexts, as we have no mean to support
789 * two tasks with the same ioc in two different groups without major rework
790 * of the main cic data structures. For now we allow a task to change
791 * its cgroup only if it's the only owner of its ioc.
793 static int blkcg_can_attach(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
)
795 struct task_struct
*task
;
796 struct io_context
*ioc
;
799 /* task_lock() is needed to avoid races with exit_io_context() */
800 cgroup_taskset_for_each(task
, cgrp
, tset
) {
802 ioc
= task
->io_context
;
803 if (ioc
&& atomic_read(&ioc
->nr_tasks
) > 1)
812 struct cgroup_subsys blkio_subsys
= {
814 .css_alloc
= blkcg_css_alloc
,
815 .css_offline
= blkcg_css_offline
,
816 .css_free
= blkcg_css_free
,
817 .can_attach
= blkcg_can_attach
,
818 .subsys_id
= blkio_subsys_id
,
819 .base_cftypes
= blkcg_files
,
820 .module
= THIS_MODULE
,
823 * blkio subsystem is utterly broken in terms of hierarchy support.
824 * It treats all cgroups equally regardless of where they're
825 * located in the hierarchy - all cgroups are treated as if they're
826 * right below the root. Fix it and remove the following.
828 .broken_hierarchy
= true,
830 EXPORT_SYMBOL_GPL(blkio_subsys
);
833 * blkcg_activate_policy - activate a blkcg policy on a request_queue
834 * @q: request_queue of interest
835 * @pol: blkcg policy to activate
837 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
838 * bypass mode to populate its blkgs with policy_data for @pol.
840 * Activation happens with @q bypassed, so nobody would be accessing blkgs
841 * from IO path. Update of each blkg is protected by both queue and blkcg
842 * locks so that holding either lock and testing blkcg_policy_enabled() is
843 * always enough for dereferencing policy data.
845 * The caller is responsible for synchronizing [de]activations and policy
846 * [un]registerations. Returns 0 on success, -errno on failure.
848 int blkcg_activate_policy(struct request_queue
*q
,
849 const struct blkcg_policy
*pol
)
852 struct blkcg_gq
*blkg
, *new_blkg
;
853 struct blkg_policy_data
*pd
, *n
;
857 if (blkcg_policy_enabled(q
, pol
))
860 /* preallocations for root blkg */
861 new_blkg
= blkg_alloc(&blkcg_root
, q
, GFP_KERNEL
);
865 preloaded
= !radix_tree_preload(GFP_KERNEL
);
867 blk_queue_bypass_start(q
);
870 * Make sure the root blkg exists and count the existing blkgs. As
871 * @q is bypassing at this point, blkg_lookup_create() can't be
872 * used. Open code it.
874 spin_lock_irq(q
->queue_lock
);
877 blkg
= __blkg_lookup(&blkcg_root
, q
, false);
881 blkg
= blkg_create(&blkcg_root
, q
, new_blkg
);
885 radix_tree_preload_end();
892 q
->root_rl
.blkg
= blkg
;
894 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
)
897 spin_unlock_irq(q
->queue_lock
);
899 /* allocate policy_data for all existing blkgs */
901 pd
= kzalloc_node(pol
->pd_size
, GFP_KERNEL
, q
->node
);
906 list_add_tail(&pd
->alloc_node
, &pds
);
910 * Install the allocated pds. With @q bypassing, no new blkg
911 * should have been created while the queue lock was dropped.
913 spin_lock_irq(q
->queue_lock
);
915 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
) {
916 if (WARN_ON(list_empty(&pds
))) {
917 /* umm... this shouldn't happen, just abort */
921 pd
= list_first_entry(&pds
, struct blkg_policy_data
, alloc_node
);
922 list_del_init(&pd
->alloc_node
);
924 /* grab blkcg lock too while installing @pd on @blkg */
925 spin_lock(&blkg
->blkcg
->lock
);
927 blkg
->pd
[pol
->plid
] = pd
;
929 pd
->plid
= pol
->plid
;
930 pol
->pd_init_fn(blkg
);
932 spin_unlock(&blkg
->blkcg
->lock
);
935 __set_bit(pol
->plid
, q
->blkcg_pols
);
938 spin_unlock_irq(q
->queue_lock
);
940 blk_queue_bypass_end(q
);
941 list_for_each_entry_safe(pd
, n
, &pds
, alloc_node
)
945 EXPORT_SYMBOL_GPL(blkcg_activate_policy
);
948 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
949 * @q: request_queue of interest
950 * @pol: blkcg policy to deactivate
952 * Deactivate @pol on @q. Follows the same synchronization rules as
953 * blkcg_activate_policy().
955 void blkcg_deactivate_policy(struct request_queue
*q
,
956 const struct blkcg_policy
*pol
)
958 struct blkcg_gq
*blkg
;
960 if (!blkcg_policy_enabled(q
, pol
))
963 blk_queue_bypass_start(q
);
964 spin_lock_irq(q
->queue_lock
);
966 __clear_bit(pol
->plid
, q
->blkcg_pols
);
968 /* if no policy is left, no need for blkgs - shoot them down */
969 if (bitmap_empty(q
->blkcg_pols
, BLKCG_MAX_POLS
))
972 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
) {
973 /* grab blkcg lock too while removing @pd from @blkg */
974 spin_lock(&blkg
->blkcg
->lock
);
976 if (pol
->pd_offline_fn
)
977 pol
->pd_offline_fn(blkg
);
979 pol
->pd_exit_fn(blkg
);
981 kfree(blkg
->pd
[pol
->plid
]);
982 blkg
->pd
[pol
->plid
] = NULL
;
984 spin_unlock(&blkg
->blkcg
->lock
);
987 spin_unlock_irq(q
->queue_lock
);
988 blk_queue_bypass_end(q
);
990 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy
);
993 * blkcg_policy_register - register a blkcg policy
994 * @pol: blkcg policy to register
996 * Register @pol with blkcg core. Might sleep and @pol may be modified on
997 * successful registration. Returns 0 on success and -errno on failure.
999 int blkcg_policy_register(struct blkcg_policy
*pol
)
1003 if (WARN_ON(pol
->pd_size
< sizeof(struct blkg_policy_data
)))
1006 mutex_lock(&blkcg_pol_mutex
);
1008 /* find an empty slot */
1010 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++)
1011 if (!blkcg_policy
[i
])
1013 if (i
>= BLKCG_MAX_POLS
)
1016 /* register and update blkgs */
1018 blkcg_policy
[i
] = pol
;
1020 /* everything is in place, add intf files for the new policy */
1022 WARN_ON(cgroup_add_cftypes(&blkio_subsys
, pol
->cftypes
));
1025 mutex_unlock(&blkcg_pol_mutex
);
1028 EXPORT_SYMBOL_GPL(blkcg_policy_register
);
1031 * blkcg_policy_unregister - unregister a blkcg policy
1032 * @pol: blkcg policy to unregister
1034 * Undo blkcg_policy_register(@pol). Might sleep.
1036 void blkcg_policy_unregister(struct blkcg_policy
*pol
)
1038 mutex_lock(&blkcg_pol_mutex
);
1040 if (WARN_ON(blkcg_policy
[pol
->plid
] != pol
))
1043 /* kill the intf files first */
1045 cgroup_rm_cftypes(&blkio_subsys
, pol
->cftypes
);
1047 /* unregister and update blkgs */
1048 blkcg_policy
[pol
->plid
] = NULL
;
1050 mutex_unlock(&blkcg_pol_mutex
);
1052 EXPORT_SYMBOL_GPL(blkcg_policy_unregister
);