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
);
131 * __blkg_lookup - internal version of blkg_lookup()
132 * @blkcg: blkcg of interest
133 * @q: request_queue of interest
134 * @update_hint: whether to update lookup hint with the result or not
136 * This is internal version and shouldn't be used by policy
137 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
138 * @q's bypass state. If @update_hint is %true, the caller should be
139 * holding @q->queue_lock and lookup hint is updated on success.
141 struct blkcg_gq
*__blkg_lookup(struct blkcg
*blkcg
, struct request_queue
*q
,
144 struct blkcg_gq
*blkg
;
146 blkg
= rcu_dereference(blkcg
->blkg_hint
);
147 if (blkg
&& blkg
->q
== q
)
151 * Hint didn't match. Look up from the radix tree. Note that the
152 * hint can only be updated under queue_lock as otherwise @blkg
153 * could have already been removed from blkg_tree. The caller is
154 * responsible for grabbing queue_lock if @update_hint.
156 blkg
= radix_tree_lookup(&blkcg
->blkg_tree
, q
->id
);
157 if (blkg
&& blkg
->q
== q
) {
159 lockdep_assert_held(q
->queue_lock
);
160 rcu_assign_pointer(blkcg
->blkg_hint
, blkg
);
169 * blkg_lookup - lookup blkg for the specified blkcg - q pair
170 * @blkcg: blkcg of interest
171 * @q: request_queue of interest
173 * Lookup blkg for the @blkcg - @q pair. This function should be called
174 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
175 * - see blk_queue_bypass_start() for details.
177 struct blkcg_gq
*blkg_lookup(struct blkcg
*blkcg
, struct request_queue
*q
)
179 WARN_ON_ONCE(!rcu_read_lock_held());
181 if (unlikely(blk_queue_bypass(q
)))
183 return __blkg_lookup(blkcg
, q
, false);
185 EXPORT_SYMBOL_GPL(blkg_lookup
);
188 * If @new_blkg is %NULL, this function tries to allocate a new one as
189 * necessary using %GFP_ATOMIC. @new_blkg is always consumed on return.
191 static struct blkcg_gq
*blkg_create(struct blkcg
*blkcg
,
192 struct request_queue
*q
,
193 struct blkcg_gq
*new_blkg
)
195 struct blkcg_gq
*blkg
;
198 WARN_ON_ONCE(!rcu_read_lock_held());
199 lockdep_assert_held(q
->queue_lock
);
201 /* blkg holds a reference to blkcg */
202 if (!css_tryget(&blkcg
->css
)) {
209 new_blkg
= blkg_alloc(blkcg
, q
, GFP_ATOMIC
);
210 if (unlikely(!new_blkg
)) {
217 /* link parent and insert */
218 if (blkcg_parent(blkcg
)) {
219 blkg
->parent
= __blkg_lookup(blkcg_parent(blkcg
), q
, false);
220 if (WARN_ON_ONCE(!blkg
->parent
)) {
224 blkg_get(blkg
->parent
);
227 spin_lock(&blkcg
->lock
);
228 ret
= radix_tree_insert(&blkcg
->blkg_tree
, q
->id
, blkg
);
230 hlist_add_head_rcu(&blkg
->blkcg_node
, &blkcg
->blkg_list
);
231 list_add(&blkg
->q_node
, &q
->blkg_list
);
233 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
234 struct blkcg_policy
*pol
= blkcg_policy
[i
];
236 if (blkg
->pd
[i
] && pol
->pd_online_fn
)
237 pol
->pd_online_fn(blkg
);
241 spin_unlock(&blkcg
->lock
);
246 /* @blkg failed fully initialized, use the usual release path */
251 css_put(&blkcg
->css
);
258 * blkg_lookup_create - lookup blkg, try to create one if not there
259 * @blkcg: blkcg of interest
260 * @q: request_queue of interest
262 * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
263 * create one. blkg creation is performed recursively from blkcg_root such
264 * that all non-root blkg's have access to the parent blkg. This function
265 * should be called under RCU read lock and @q->queue_lock.
267 * Returns pointer to the looked up or created blkg on success, ERR_PTR()
268 * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not
269 * dead and bypassing, returns ERR_PTR(-EBUSY).
271 struct blkcg_gq
*blkg_lookup_create(struct blkcg
*blkcg
,
272 struct request_queue
*q
)
274 struct blkcg_gq
*blkg
;
276 WARN_ON_ONCE(!rcu_read_lock_held());
277 lockdep_assert_held(q
->queue_lock
);
280 * This could be the first entry point of blkcg implementation and
281 * we shouldn't allow anything to go through for a bypassing queue.
283 if (unlikely(blk_queue_bypass(q
)))
284 return ERR_PTR(blk_queue_dying(q
) ? -EINVAL
: -EBUSY
);
286 blkg
= __blkg_lookup(blkcg
, q
, true);
291 * Create blkgs walking down from blkcg_root to @blkcg, so that all
292 * non-root blkgs have access to their parents.
295 struct blkcg
*pos
= blkcg
;
296 struct blkcg
*parent
= blkcg_parent(blkcg
);
298 while (parent
&& !__blkg_lookup(parent
, q
, false)) {
300 parent
= blkcg_parent(parent
);
303 blkg
= blkg_create(pos
, q
, NULL
);
304 if (pos
== blkcg
|| IS_ERR(blkg
))
308 EXPORT_SYMBOL_GPL(blkg_lookup_create
);
310 static void blkg_destroy(struct blkcg_gq
*blkg
)
312 struct blkcg
*blkcg
= blkg
->blkcg
;
315 lockdep_assert_held(blkg
->q
->queue_lock
);
316 lockdep_assert_held(&blkcg
->lock
);
318 /* Something wrong if we are trying to remove same group twice */
319 WARN_ON_ONCE(list_empty(&blkg
->q_node
));
320 WARN_ON_ONCE(hlist_unhashed(&blkg
->blkcg_node
));
322 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
323 struct blkcg_policy
*pol
= blkcg_policy
[i
];
325 if (blkg
->pd
[i
] && pol
->pd_offline_fn
)
326 pol
->pd_offline_fn(blkg
);
328 blkg
->online
= false;
330 radix_tree_delete(&blkcg
->blkg_tree
, blkg
->q
->id
);
331 list_del_init(&blkg
->q_node
);
332 hlist_del_init_rcu(&blkg
->blkcg_node
);
335 * Both setting lookup hint to and clearing it from @blkg are done
336 * under queue_lock. If it's not pointing to @blkg now, it never
337 * will. Hint assignment itself can race safely.
339 if (rcu_dereference_raw(blkcg
->blkg_hint
) == blkg
)
340 rcu_assign_pointer(blkcg
->blkg_hint
, NULL
);
343 * Put the reference taken at the time of creation so that when all
344 * queues are gone, group can be destroyed.
350 * blkg_destroy_all - destroy all blkgs associated with a request_queue
351 * @q: request_queue of interest
353 * Destroy all blkgs associated with @q.
355 static void blkg_destroy_all(struct request_queue
*q
)
357 struct blkcg_gq
*blkg
, *n
;
359 lockdep_assert_held(q
->queue_lock
);
361 list_for_each_entry_safe(blkg
, n
, &q
->blkg_list
, q_node
) {
362 struct blkcg
*blkcg
= blkg
->blkcg
;
364 spin_lock(&blkcg
->lock
);
366 spin_unlock(&blkcg
->lock
);
370 * root blkg is destroyed. Just clear the pointer since
371 * root_rl does not take reference on root blkg.
374 q
->root_rl
.blkg
= NULL
;
377 static void blkg_rcu_free(struct rcu_head
*rcu_head
)
379 blkg_free(container_of(rcu_head
, struct blkcg_gq
, rcu_head
));
382 void __blkg_release(struct blkcg_gq
*blkg
)
384 /* release the blkcg and parent blkg refs this blkg has been holding */
385 css_put(&blkg
->blkcg
->css
);
387 blkg_put(blkg
->parent
);
390 * A group is freed in rcu manner. But having an rcu lock does not
391 * mean that one can access all the fields of blkg and assume these
392 * are valid. For example, don't try to follow throtl_data and
393 * request queue links.
395 * Having a reference to blkg under an rcu allows acess to only
396 * values local to groups like group stats and group rate limits
398 call_rcu(&blkg
->rcu_head
, blkg_rcu_free
);
400 EXPORT_SYMBOL_GPL(__blkg_release
);
403 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
404 * because the root blkg uses @q->root_rl instead of its own rl.
406 struct request_list
*__blk_queue_next_rl(struct request_list
*rl
,
407 struct request_queue
*q
)
409 struct list_head
*ent
;
410 struct blkcg_gq
*blkg
;
413 * Determine the current blkg list_head. The first entry is
414 * root_rl which is off @q->blkg_list and mapped to the head.
416 if (rl
== &q
->root_rl
) {
418 /* There are no more block groups, hence no request lists */
422 blkg
= container_of(rl
, struct blkcg_gq
, rl
);
426 /* walk to the next list_head, skip root blkcg */
428 if (ent
== &q
->root_blkg
->q_node
)
430 if (ent
== &q
->blkg_list
)
433 blkg
= container_of(ent
, struct blkcg_gq
, q_node
);
437 static int blkcg_reset_stats(struct cgroup
*cgroup
, struct cftype
*cftype
,
440 struct blkcg
*blkcg
= cgroup_to_blkcg(cgroup
);
441 struct blkcg_gq
*blkg
;
444 mutex_lock(&blkcg_pol_mutex
);
445 spin_lock_irq(&blkcg
->lock
);
448 * Note that stat reset is racy - it doesn't synchronize against
449 * stat updates. This is a debug feature which shouldn't exist
450 * anyway. If you get hit by a race, retry.
452 hlist_for_each_entry(blkg
, &blkcg
->blkg_list
, blkcg_node
) {
453 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
454 struct blkcg_policy
*pol
= blkcg_policy
[i
];
456 if (blkcg_policy_enabled(blkg
->q
, pol
) &&
457 pol
->pd_reset_stats_fn
)
458 pol
->pd_reset_stats_fn(blkg
);
462 spin_unlock_irq(&blkcg
->lock
);
463 mutex_unlock(&blkcg_pol_mutex
);
467 static const char *blkg_dev_name(struct blkcg_gq
*blkg
)
469 /* some drivers (floppy) instantiate a queue w/o disk registered */
470 if (blkg
->q
->backing_dev_info
.dev
)
471 return dev_name(blkg
->q
->backing_dev_info
.dev
);
476 * blkcg_print_blkgs - helper for printing per-blkg data
477 * @sf: seq_file to print to
478 * @blkcg: blkcg of interest
479 * @prfill: fill function to print out a blkg
480 * @pol: policy in question
481 * @data: data to be passed to @prfill
482 * @show_total: to print out sum of prfill return values or not
484 * This function invokes @prfill on each blkg of @blkcg if pd for the
485 * policy specified by @pol exists. @prfill is invoked with @sf, the
486 * policy data and @data and the matching queue lock held. If @show_total
487 * is %true, the sum of the return values from @prfill is printed with
488 * "Total" label at the end.
490 * This is to be used to construct print functions for
491 * cftype->read_seq_string method.
493 void blkcg_print_blkgs(struct seq_file
*sf
, struct blkcg
*blkcg
,
494 u64 (*prfill
)(struct seq_file
*,
495 struct blkg_policy_data
*, int),
496 const struct blkcg_policy
*pol
, int data
,
499 struct blkcg_gq
*blkg
;
503 hlist_for_each_entry_rcu(blkg
, &blkcg
->blkg_list
, blkcg_node
) {
504 spin_lock_irq(blkg
->q
->queue_lock
);
505 if (blkcg_policy_enabled(blkg
->q
, pol
))
506 total
+= prfill(sf
, blkg
->pd
[pol
->plid
], data
);
507 spin_unlock_irq(blkg
->q
->queue_lock
);
512 seq_printf(sf
, "Total %llu\n", (unsigned long long)total
);
514 EXPORT_SYMBOL_GPL(blkcg_print_blkgs
);
517 * __blkg_prfill_u64 - prfill helper for a single u64 value
518 * @sf: seq_file to print to
519 * @pd: policy private data of interest
522 * Print @v to @sf for the device assocaited with @pd.
524 u64
__blkg_prfill_u64(struct seq_file
*sf
, struct blkg_policy_data
*pd
, u64 v
)
526 const char *dname
= blkg_dev_name(pd
->blkg
);
531 seq_printf(sf
, "%s %llu\n", dname
, (unsigned long long)v
);
534 EXPORT_SYMBOL_GPL(__blkg_prfill_u64
);
537 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
538 * @sf: seq_file to print to
539 * @pd: policy private data of interest
540 * @rwstat: rwstat to print
542 * Print @rwstat to @sf for the device assocaited with @pd.
544 u64
__blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
545 const struct blkg_rwstat
*rwstat
)
547 static const char *rwstr
[] = {
548 [BLKG_RWSTAT_READ
] = "Read",
549 [BLKG_RWSTAT_WRITE
] = "Write",
550 [BLKG_RWSTAT_SYNC
] = "Sync",
551 [BLKG_RWSTAT_ASYNC
] = "Async",
553 const char *dname
= blkg_dev_name(pd
->blkg
);
560 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
561 seq_printf(sf
, "%s %s %llu\n", dname
, rwstr
[i
],
562 (unsigned long long)rwstat
->cnt
[i
]);
564 v
= rwstat
->cnt
[BLKG_RWSTAT_READ
] + rwstat
->cnt
[BLKG_RWSTAT_WRITE
];
565 seq_printf(sf
, "%s Total %llu\n", dname
, (unsigned long long)v
);
568 EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat
);
571 * blkg_prfill_stat - prfill callback for blkg_stat
572 * @sf: seq_file to print to
573 * @pd: policy private data of interest
574 * @off: offset to the blkg_stat in @pd
576 * prfill callback for printing a blkg_stat.
578 u64
blkg_prfill_stat(struct seq_file
*sf
, struct blkg_policy_data
*pd
, int off
)
580 return __blkg_prfill_u64(sf
, pd
, blkg_stat_read((void *)pd
+ off
));
582 EXPORT_SYMBOL_GPL(blkg_prfill_stat
);
585 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
586 * @sf: seq_file to print to
587 * @pd: policy private data of interest
588 * @off: offset to the blkg_rwstat in @pd
590 * prfill callback for printing a blkg_rwstat.
592 u64
blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
595 struct blkg_rwstat rwstat
= blkg_rwstat_read((void *)pd
+ off
);
597 return __blkg_prfill_rwstat(sf
, pd
, &rwstat
);
599 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat
);
602 * blkg_stat_recursive_sum - collect hierarchical blkg_stat
603 * @pd: policy private data of interest
604 * @off: offset to the blkg_stat in @pd
606 * Collect the blkg_stat specified by @off from @pd and all its online
607 * descendants and return the sum. The caller must be holding the queue
608 * lock for online tests.
610 u64
blkg_stat_recursive_sum(struct blkg_policy_data
*pd
, int off
)
612 struct blkcg_policy
*pol
= blkcg_policy
[pd
->plid
];
613 struct blkcg_gq
*pos_blkg
;
614 struct cgroup
*pos_cgrp
;
617 lockdep_assert_held(pd
->blkg
->q
->queue_lock
);
619 sum
= blkg_stat_read((void *)pd
+ off
);
622 blkg_for_each_descendant_pre(pos_blkg
, pos_cgrp
, pd_to_blkg(pd
)) {
623 struct blkg_policy_data
*pos_pd
= blkg_to_pd(pos_blkg
, pol
);
624 struct blkg_stat
*stat
= (void *)pos_pd
+ off
;
626 if (pos_blkg
->online
)
627 sum
+= blkg_stat_read(stat
);
633 EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum
);
636 * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat
637 * @pd: policy private data of interest
638 * @off: offset to the blkg_stat in @pd
640 * Collect the blkg_rwstat specified by @off from @pd and all its online
641 * descendants and return the sum. The caller must be holding the queue
642 * lock for online tests.
644 struct blkg_rwstat
blkg_rwstat_recursive_sum(struct blkg_policy_data
*pd
,
647 struct blkcg_policy
*pol
= blkcg_policy
[pd
->plid
];
648 struct blkcg_gq
*pos_blkg
;
649 struct cgroup
*pos_cgrp
;
650 struct blkg_rwstat sum
;
653 lockdep_assert_held(pd
->blkg
->q
->queue_lock
);
655 sum
= blkg_rwstat_read((void *)pd
+ off
);
658 blkg_for_each_descendant_pre(pos_blkg
, pos_cgrp
, pd_to_blkg(pd
)) {
659 struct blkg_policy_data
*pos_pd
= blkg_to_pd(pos_blkg
, pol
);
660 struct blkg_rwstat
*rwstat
= (void *)pos_pd
+ off
;
661 struct blkg_rwstat tmp
;
663 if (!pos_blkg
->online
)
666 tmp
= blkg_rwstat_read(rwstat
);
668 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
669 sum
.cnt
[i
] += tmp
.cnt
[i
];
675 EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum
);
678 * blkg_conf_prep - parse and prepare for per-blkg config update
679 * @blkcg: target block cgroup
680 * @pol: target policy
681 * @input: input string
682 * @ctx: blkg_conf_ctx to be filled
684 * Parse per-blkg config update from @input and initialize @ctx with the
685 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
686 * value. This function returns with RCU read lock and queue lock held and
687 * must be paired with blkg_conf_finish().
689 int blkg_conf_prep(struct blkcg
*blkcg
, const struct blkcg_policy
*pol
,
690 const char *input
, struct blkg_conf_ctx
*ctx
)
691 __acquires(rcu
) __acquires(disk
->queue
->queue_lock
)
693 struct gendisk
*disk
;
694 struct blkcg_gq
*blkg
;
695 unsigned int major
, minor
;
696 unsigned long long v
;
699 if (sscanf(input
, "%u:%u %llu", &major
, &minor
, &v
) != 3)
702 disk
= get_gendisk(MKDEV(major
, minor
), &part
);
707 spin_lock_irq(disk
->queue
->queue_lock
);
709 if (blkcg_policy_enabled(disk
->queue
, pol
))
710 blkg
= blkg_lookup_create(blkcg
, disk
->queue
);
712 blkg
= ERR_PTR(-EINVAL
);
717 spin_unlock_irq(disk
->queue
->queue_lock
);
720 * If queue was bypassing, we should retry. Do so after a
721 * short msleep(). It isn't strictly necessary but queue
722 * can be bypassing for some time and it's always nice to
723 * avoid busy looping.
727 ret
= restart_syscall();
737 EXPORT_SYMBOL_GPL(blkg_conf_prep
);
740 * blkg_conf_finish - finish up per-blkg config update
741 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
743 * Finish up after per-blkg config update. This function must be paired
744 * with blkg_conf_prep().
746 void blkg_conf_finish(struct blkg_conf_ctx
*ctx
)
747 __releases(ctx
->disk
->queue
->queue_lock
) __releases(rcu
)
749 spin_unlock_irq(ctx
->disk
->queue
->queue_lock
);
753 EXPORT_SYMBOL_GPL(blkg_conf_finish
);
755 struct cftype blkcg_files
[] = {
757 .name
= "reset_stats",
758 .write_u64
= blkcg_reset_stats
,
764 * blkcg_css_offline - cgroup css_offline callback
765 * @cgroup: cgroup of interest
767 * This function is called when @cgroup is about to go away and responsible
768 * for shooting down all blkgs associated with @cgroup. blkgs should be
769 * removed while holding both q and blkcg locks. As blkcg lock is nested
770 * inside q lock, this function performs reverse double lock dancing.
772 * This is the blkcg counterpart of ioc_release_fn().
774 static void blkcg_css_offline(struct cgroup
*cgroup
)
776 struct blkcg
*blkcg
= cgroup_to_blkcg(cgroup
);
778 spin_lock_irq(&blkcg
->lock
);
780 while (!hlist_empty(&blkcg
->blkg_list
)) {
781 struct blkcg_gq
*blkg
= hlist_entry(blkcg
->blkg_list
.first
,
782 struct blkcg_gq
, blkcg_node
);
783 struct request_queue
*q
= blkg
->q
;
785 if (spin_trylock(q
->queue_lock
)) {
787 spin_unlock(q
->queue_lock
);
789 spin_unlock_irq(&blkcg
->lock
);
791 spin_lock_irq(&blkcg
->lock
);
795 spin_unlock_irq(&blkcg
->lock
);
798 static void blkcg_css_free(struct cgroup
*cgroup
)
800 struct blkcg
*blkcg
= cgroup_to_blkcg(cgroup
);
802 if (blkcg
!= &blkcg_root
)
806 static struct cgroup_subsys_state
*blkcg_css_alloc(struct cgroup
*cgroup
)
808 static atomic64_t id_seq
= ATOMIC64_INIT(0);
810 struct cgroup
*parent
= cgroup
->parent
;
817 blkcg
= kzalloc(sizeof(*blkcg
), GFP_KERNEL
);
819 return ERR_PTR(-ENOMEM
);
821 blkcg
->cfq_weight
= CFQ_WEIGHT_DEFAULT
;
822 blkcg
->cfq_leaf_weight
= CFQ_WEIGHT_DEFAULT
;
823 blkcg
->id
= atomic64_inc_return(&id_seq
); /* root is 0, start from 1 */
825 spin_lock_init(&blkcg
->lock
);
826 INIT_RADIX_TREE(&blkcg
->blkg_tree
, GFP_ATOMIC
);
827 INIT_HLIST_HEAD(&blkcg
->blkg_list
);
833 * blkcg_init_queue - initialize blkcg part of request queue
834 * @q: request_queue to initialize
836 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
837 * part of new request_queue @q.
840 * 0 on success, -errno on failure.
842 int blkcg_init_queue(struct request_queue
*q
)
846 return blk_throtl_init(q
);
850 * blkcg_drain_queue - drain blkcg part of request_queue
851 * @q: request_queue to drain
853 * Called from blk_drain_queue(). Responsible for draining blkcg part.
855 void blkcg_drain_queue(struct request_queue
*q
)
857 lockdep_assert_held(q
->queue_lock
);
863 * blkcg_exit_queue - exit and release blkcg part of request_queue
864 * @q: request_queue being released
866 * Called from blk_release_queue(). Responsible for exiting blkcg part.
868 void blkcg_exit_queue(struct request_queue
*q
)
870 spin_lock_irq(q
->queue_lock
);
872 spin_unlock_irq(q
->queue_lock
);
878 * We cannot support shared io contexts, as we have no mean to support
879 * two tasks with the same ioc in two different groups without major rework
880 * of the main cic data structures. For now we allow a task to change
881 * its cgroup only if it's the only owner of its ioc.
883 static int blkcg_can_attach(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
)
885 struct task_struct
*task
;
886 struct io_context
*ioc
;
889 /* task_lock() is needed to avoid races with exit_io_context() */
890 cgroup_taskset_for_each(task
, cgrp
, tset
) {
892 ioc
= task
->io_context
;
893 if (ioc
&& atomic_read(&ioc
->nr_tasks
) > 1)
902 struct cgroup_subsys blkio_subsys
= {
904 .css_alloc
= blkcg_css_alloc
,
905 .css_offline
= blkcg_css_offline
,
906 .css_free
= blkcg_css_free
,
907 .can_attach
= blkcg_can_attach
,
908 .subsys_id
= blkio_subsys_id
,
909 .base_cftypes
= blkcg_files
,
910 .module
= THIS_MODULE
,
913 * blkio subsystem is utterly broken in terms of hierarchy support.
914 * It treats all cgroups equally regardless of where they're
915 * located in the hierarchy - all cgroups are treated as if they're
916 * right below the root. Fix it and remove the following.
918 .broken_hierarchy
= true,
920 EXPORT_SYMBOL_GPL(blkio_subsys
);
923 * blkcg_activate_policy - activate a blkcg policy on a request_queue
924 * @q: request_queue of interest
925 * @pol: blkcg policy to activate
927 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
928 * bypass mode to populate its blkgs with policy_data for @pol.
930 * Activation happens with @q bypassed, so nobody would be accessing blkgs
931 * from IO path. Update of each blkg is protected by both queue and blkcg
932 * locks so that holding either lock and testing blkcg_policy_enabled() is
933 * always enough for dereferencing policy data.
935 * The caller is responsible for synchronizing [de]activations and policy
936 * [un]registerations. Returns 0 on success, -errno on failure.
938 int blkcg_activate_policy(struct request_queue
*q
,
939 const struct blkcg_policy
*pol
)
942 struct blkcg_gq
*blkg
, *new_blkg
;
943 struct blkg_policy_data
*pd
, *n
;
947 if (blkcg_policy_enabled(q
, pol
))
950 /* preallocations for root blkg */
951 new_blkg
= blkg_alloc(&blkcg_root
, q
, GFP_KERNEL
);
955 blk_queue_bypass_start(q
);
957 preloaded
= !radix_tree_preload(GFP_KERNEL
);
960 * Make sure the root blkg exists and count the existing blkgs. As
961 * @q is bypassing at this point, blkg_lookup_create() can't be
962 * used. Open code it.
964 spin_lock_irq(q
->queue_lock
);
967 blkg
= __blkg_lookup(&blkcg_root
, q
, false);
971 blkg
= blkg_create(&blkcg_root
, q
, new_blkg
);
975 radix_tree_preload_end();
982 q
->root_rl
.blkg
= blkg
;
984 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
)
987 spin_unlock_irq(q
->queue_lock
);
989 /* allocate policy_data for all existing blkgs */
991 pd
= kzalloc_node(pol
->pd_size
, GFP_KERNEL
, q
->node
);
996 list_add_tail(&pd
->alloc_node
, &pds
);
1000 * Install the allocated pds. With @q bypassing, no new blkg
1001 * should have been created while the queue lock was dropped.
1003 spin_lock_irq(q
->queue_lock
);
1005 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
) {
1006 if (WARN_ON(list_empty(&pds
))) {
1007 /* umm... this shouldn't happen, just abort */
1011 pd
= list_first_entry(&pds
, struct blkg_policy_data
, alloc_node
);
1012 list_del_init(&pd
->alloc_node
);
1014 /* grab blkcg lock too while installing @pd on @blkg */
1015 spin_lock(&blkg
->blkcg
->lock
);
1017 blkg
->pd
[pol
->plid
] = pd
;
1019 pd
->plid
= pol
->plid
;
1020 pol
->pd_init_fn(blkg
);
1022 spin_unlock(&blkg
->blkcg
->lock
);
1025 __set_bit(pol
->plid
, q
->blkcg_pols
);
1028 spin_unlock_irq(q
->queue_lock
);
1030 blk_queue_bypass_end(q
);
1031 list_for_each_entry_safe(pd
, n
, &pds
, alloc_node
)
1035 EXPORT_SYMBOL_GPL(blkcg_activate_policy
);
1038 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
1039 * @q: request_queue of interest
1040 * @pol: blkcg policy to deactivate
1042 * Deactivate @pol on @q. Follows the same synchronization rules as
1043 * blkcg_activate_policy().
1045 void blkcg_deactivate_policy(struct request_queue
*q
,
1046 const struct blkcg_policy
*pol
)
1048 struct blkcg_gq
*blkg
;
1050 if (!blkcg_policy_enabled(q
, pol
))
1053 blk_queue_bypass_start(q
);
1054 spin_lock_irq(q
->queue_lock
);
1056 __clear_bit(pol
->plid
, q
->blkcg_pols
);
1058 /* if no policy is left, no need for blkgs - shoot them down */
1059 if (bitmap_empty(q
->blkcg_pols
, BLKCG_MAX_POLS
))
1060 blkg_destroy_all(q
);
1062 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
) {
1063 /* grab blkcg lock too while removing @pd from @blkg */
1064 spin_lock(&blkg
->blkcg
->lock
);
1066 if (pol
->pd_offline_fn
)
1067 pol
->pd_offline_fn(blkg
);
1068 if (pol
->pd_exit_fn
)
1069 pol
->pd_exit_fn(blkg
);
1071 kfree(blkg
->pd
[pol
->plid
]);
1072 blkg
->pd
[pol
->plid
] = NULL
;
1074 spin_unlock(&blkg
->blkcg
->lock
);
1077 spin_unlock_irq(q
->queue_lock
);
1078 blk_queue_bypass_end(q
);
1080 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy
);
1083 * blkcg_policy_register - register a blkcg policy
1084 * @pol: blkcg policy to register
1086 * Register @pol with blkcg core. Might sleep and @pol may be modified on
1087 * successful registration. Returns 0 on success and -errno on failure.
1089 int blkcg_policy_register(struct blkcg_policy
*pol
)
1093 if (WARN_ON(pol
->pd_size
< sizeof(struct blkg_policy_data
)))
1096 mutex_lock(&blkcg_pol_mutex
);
1098 /* find an empty slot */
1100 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++)
1101 if (!blkcg_policy
[i
])
1103 if (i
>= BLKCG_MAX_POLS
)
1106 /* register and update blkgs */
1108 blkcg_policy
[i
] = pol
;
1110 /* everything is in place, add intf files for the new policy */
1112 WARN_ON(cgroup_add_cftypes(&blkio_subsys
, pol
->cftypes
));
1115 mutex_unlock(&blkcg_pol_mutex
);
1118 EXPORT_SYMBOL_GPL(blkcg_policy_register
);
1121 * blkcg_policy_unregister - unregister a blkcg policy
1122 * @pol: blkcg policy to unregister
1124 * Undo blkcg_policy_register(@pol). Might sleep.
1126 void blkcg_policy_unregister(struct blkcg_policy
*pol
)
1128 mutex_lock(&blkcg_pol_mutex
);
1130 if (WARN_ON(blkcg_policy
[pol
->plid
] != pol
))
1133 /* kill the intf files first */
1135 cgroup_rm_cftypes(&blkio_subsys
, pol
->cftypes
);
1137 /* unregister and update blkgs */
1138 blkcg_policy
[pol
->plid
] = NULL
;
1140 mutex_unlock(&blkcg_pol_mutex
);
1142 EXPORT_SYMBOL_GPL(blkcg_policy_unregister
);