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 struct blkcg_gq
*__blkg_lookup(struct blkcg
*blkcg
,
36 struct request_queue
*q
, bool update_hint
);
39 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
40 * @d_blkg: loop cursor pointing to the current descendant
41 * @pos_cgrp: used for iteration
42 * @p_blkg: target blkg to walk descendants of
44 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
45 * read locked. If called under either blkcg or queue lock, the iteration
46 * is guaranteed to include all and only online blkgs. The caller may
47 * update @pos_cgrp by calling cgroup_rightmost_descendant() to skip
50 #define blkg_for_each_descendant_pre(d_blkg, pos_cgrp, p_blkg) \
51 cgroup_for_each_descendant_pre((pos_cgrp), (p_blkg)->blkcg->css.cgroup) \
52 if (((d_blkg) = __blkg_lookup(cgroup_to_blkcg(pos_cgrp), \
55 static bool blkcg_policy_enabled(struct request_queue
*q
,
56 const struct blkcg_policy
*pol
)
58 return pol
&& test_bit(pol
->plid
, q
->blkcg_pols
);
62 * blkg_free - free a blkg
65 * Free @blkg which may be partially allocated.
67 static void blkg_free(struct blkcg_gq
*blkg
)
74 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
75 struct blkcg_policy
*pol
= blkcg_policy
[i
];
76 struct blkg_policy_data
*pd
= blkg
->pd
[i
];
81 if (pol
&& pol
->pd_exit_fn
)
82 pol
->pd_exit_fn(blkg
);
87 blk_exit_rl(&blkg
->rl
);
92 * blkg_alloc - allocate a blkg
93 * @blkcg: block cgroup the new blkg is associated with
94 * @q: request_queue the new blkg is associated with
95 * @gfp_mask: allocation mask to use
97 * Allocate a new blkg assocating @blkcg and @q.
99 static struct blkcg_gq
*blkg_alloc(struct blkcg
*blkcg
, struct request_queue
*q
,
102 struct blkcg_gq
*blkg
;
105 /* alloc and init base part */
106 blkg
= kzalloc_node(sizeof(*blkg
), gfp_mask
, q
->node
);
111 INIT_LIST_HEAD(&blkg
->q_node
);
115 /* root blkg uses @q->root_rl, init rl only for !root blkgs */
116 if (blkcg
!= &blkcg_root
) {
117 if (blk_init_rl(&blkg
->rl
, q
, gfp_mask
))
119 blkg
->rl
.blkg
= blkg
;
122 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
123 struct blkcg_policy
*pol
= blkcg_policy
[i
];
124 struct blkg_policy_data
*pd
;
126 if (!blkcg_policy_enabled(q
, pol
))
129 /* alloc per-policy data and attach it to blkg */
130 pd
= kzalloc_node(pol
->pd_size
, gfp_mask
, q
->node
);
138 /* invoke per-policy init */
140 pol
->pd_init_fn(blkg
);
151 * __blkg_lookup - internal version of blkg_lookup()
152 * @blkcg: blkcg of interest
153 * @q: request_queue of interest
154 * @update_hint: whether to update lookup hint with the result or not
156 * This is internal version and shouldn't be used by policy
157 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
158 * @q's bypass state. If @update_hint is %true, the caller should be
159 * holding @q->queue_lock and lookup hint is updated on success.
161 static struct blkcg_gq
*__blkg_lookup(struct blkcg
*blkcg
,
162 struct request_queue
*q
, bool update_hint
)
164 struct blkcg_gq
*blkg
;
166 blkg
= rcu_dereference(blkcg
->blkg_hint
);
167 if (blkg
&& blkg
->q
== q
)
171 * Hint didn't match. Look up from the radix tree. Note that the
172 * hint can only be updated under queue_lock as otherwise @blkg
173 * could have already been removed from blkg_tree. The caller is
174 * responsible for grabbing queue_lock if @update_hint.
176 blkg
= radix_tree_lookup(&blkcg
->blkg_tree
, q
->id
);
177 if (blkg
&& blkg
->q
== q
) {
179 lockdep_assert_held(q
->queue_lock
);
180 rcu_assign_pointer(blkcg
->blkg_hint
, blkg
);
189 * blkg_lookup - lookup blkg for the specified blkcg - q pair
190 * @blkcg: blkcg of interest
191 * @q: request_queue of interest
193 * Lookup blkg for the @blkcg - @q pair. This function should be called
194 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
195 * - see blk_queue_bypass_start() for details.
197 struct blkcg_gq
*blkg_lookup(struct blkcg
*blkcg
, struct request_queue
*q
)
199 WARN_ON_ONCE(!rcu_read_lock_held());
201 if (unlikely(blk_queue_bypass(q
)))
203 return __blkg_lookup(blkcg
, q
, false);
205 EXPORT_SYMBOL_GPL(blkg_lookup
);
208 * If @new_blkg is %NULL, this function tries to allocate a new one as
209 * necessary using %GFP_ATOMIC. @new_blkg is always consumed on return.
211 static struct blkcg_gq
*blkg_create(struct blkcg
*blkcg
,
212 struct request_queue
*q
,
213 struct blkcg_gq
*new_blkg
)
215 struct blkcg_gq
*blkg
;
218 WARN_ON_ONCE(!rcu_read_lock_held());
219 lockdep_assert_held(q
->queue_lock
);
221 /* blkg holds a reference to blkcg */
222 if (!css_tryget(&blkcg
->css
)) {
229 new_blkg
= blkg_alloc(blkcg
, q
, GFP_ATOMIC
);
230 if (unlikely(!new_blkg
)) {
237 /* link parent and insert */
238 if (blkcg_parent(blkcg
)) {
239 blkg
->parent
= __blkg_lookup(blkcg_parent(blkcg
), q
, false);
240 if (WARN_ON_ONCE(!blkg
->parent
)) {
241 blkg
= ERR_PTR(-EINVAL
);
244 blkg_get(blkg
->parent
);
247 spin_lock(&blkcg
->lock
);
248 ret
= radix_tree_insert(&blkcg
->blkg_tree
, q
->id
, blkg
);
250 hlist_add_head_rcu(&blkg
->blkcg_node
, &blkcg
->blkg_list
);
251 list_add(&blkg
->q_node
, &q
->blkg_list
);
253 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
254 struct blkcg_policy
*pol
= blkcg_policy
[i
];
256 if (blkg
->pd
[i
] && pol
->pd_online_fn
)
257 pol
->pd_online_fn(blkg
);
261 spin_unlock(&blkcg
->lock
);
266 /* @blkg failed fully initialized, use the usual release path */
271 css_put(&blkcg
->css
);
278 * blkg_lookup_create - lookup blkg, try to create one if not there
279 * @blkcg: blkcg of interest
280 * @q: request_queue of interest
282 * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
283 * create one. blkg creation is performed recursively from blkcg_root such
284 * that all non-root blkg's have access to the parent blkg. This function
285 * should be called under RCU read lock and @q->queue_lock.
287 * Returns pointer to the looked up or created blkg on success, ERR_PTR()
288 * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not
289 * dead and bypassing, returns ERR_PTR(-EBUSY).
291 struct blkcg_gq
*blkg_lookup_create(struct blkcg
*blkcg
,
292 struct request_queue
*q
)
294 struct blkcg_gq
*blkg
;
296 WARN_ON_ONCE(!rcu_read_lock_held());
297 lockdep_assert_held(q
->queue_lock
);
300 * This could be the first entry point of blkcg implementation and
301 * we shouldn't allow anything to go through for a bypassing queue.
303 if (unlikely(blk_queue_bypass(q
)))
304 return ERR_PTR(blk_queue_dying(q
) ? -EINVAL
: -EBUSY
);
306 blkg
= __blkg_lookup(blkcg
, q
, true);
311 * Create blkgs walking down from blkcg_root to @blkcg, so that all
312 * non-root blkgs have access to their parents.
315 struct blkcg
*pos
= blkcg
;
316 struct blkcg
*parent
= blkcg_parent(blkcg
);
318 while (parent
&& !__blkg_lookup(parent
, q
, false)) {
320 parent
= blkcg_parent(parent
);
323 blkg
= blkg_create(pos
, q
, NULL
);
324 if (pos
== blkcg
|| IS_ERR(blkg
))
328 EXPORT_SYMBOL_GPL(blkg_lookup_create
);
330 static void blkg_destroy(struct blkcg_gq
*blkg
)
332 struct blkcg
*blkcg
= blkg
->blkcg
;
335 lockdep_assert_held(blkg
->q
->queue_lock
);
336 lockdep_assert_held(&blkcg
->lock
);
338 /* Something wrong if we are trying to remove same group twice */
339 WARN_ON_ONCE(list_empty(&blkg
->q_node
));
340 WARN_ON_ONCE(hlist_unhashed(&blkg
->blkcg_node
));
342 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
343 struct blkcg_policy
*pol
= blkcg_policy
[i
];
345 if (blkg
->pd
[i
] && pol
->pd_offline_fn
)
346 pol
->pd_offline_fn(blkg
);
348 blkg
->online
= false;
350 radix_tree_delete(&blkcg
->blkg_tree
, blkg
->q
->id
);
351 list_del_init(&blkg
->q_node
);
352 hlist_del_init_rcu(&blkg
->blkcg_node
);
355 * Both setting lookup hint to and clearing it from @blkg are done
356 * under queue_lock. If it's not pointing to @blkg now, it never
357 * will. Hint assignment itself can race safely.
359 if (rcu_dereference_raw(blkcg
->blkg_hint
) == blkg
)
360 rcu_assign_pointer(blkcg
->blkg_hint
, NULL
);
363 * Put the reference taken at the time of creation so that when all
364 * queues are gone, group can be destroyed.
370 * blkg_destroy_all - destroy all blkgs associated with a request_queue
371 * @q: request_queue of interest
373 * Destroy all blkgs associated with @q.
375 static void blkg_destroy_all(struct request_queue
*q
)
377 struct blkcg_gq
*blkg
, *n
;
379 lockdep_assert_held(q
->queue_lock
);
381 list_for_each_entry_safe(blkg
, n
, &q
->blkg_list
, q_node
) {
382 struct blkcg
*blkcg
= blkg
->blkcg
;
384 spin_lock(&blkcg
->lock
);
386 spin_unlock(&blkcg
->lock
);
390 * root blkg is destroyed. Just clear the pointer since
391 * root_rl does not take reference on root blkg.
394 q
->root_rl
.blkg
= NULL
;
397 static void blkg_rcu_free(struct rcu_head
*rcu_head
)
399 blkg_free(container_of(rcu_head
, struct blkcg_gq
, rcu_head
));
402 void __blkg_release(struct blkcg_gq
*blkg
)
404 /* release the blkcg and parent blkg refs this blkg has been holding */
405 css_put(&blkg
->blkcg
->css
);
407 blkg_put(blkg
->parent
);
410 * A group is freed in rcu manner. But having an rcu lock does not
411 * mean that one can access all the fields of blkg and assume these
412 * are valid. For example, don't try to follow throtl_data and
413 * request queue links.
415 * Having a reference to blkg under an rcu allows acess to only
416 * values local to groups like group stats and group rate limits
418 call_rcu(&blkg
->rcu_head
, blkg_rcu_free
);
420 EXPORT_SYMBOL_GPL(__blkg_release
);
423 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
424 * because the root blkg uses @q->root_rl instead of its own rl.
426 struct request_list
*__blk_queue_next_rl(struct request_list
*rl
,
427 struct request_queue
*q
)
429 struct list_head
*ent
;
430 struct blkcg_gq
*blkg
;
433 * Determine the current blkg list_head. The first entry is
434 * root_rl which is off @q->blkg_list and mapped to the head.
436 if (rl
== &q
->root_rl
) {
438 /* There are no more block groups, hence no request lists */
442 blkg
= container_of(rl
, struct blkcg_gq
, rl
);
446 /* walk to the next list_head, skip root blkcg */
448 if (ent
== &q
->root_blkg
->q_node
)
450 if (ent
== &q
->blkg_list
)
453 blkg
= container_of(ent
, struct blkcg_gq
, q_node
);
457 static int blkcg_reset_stats(struct cgroup
*cgroup
, struct cftype
*cftype
,
460 struct blkcg
*blkcg
= cgroup_to_blkcg(cgroup
);
461 struct blkcg_gq
*blkg
;
462 struct hlist_node
*n
;
465 mutex_lock(&blkcg_pol_mutex
);
466 spin_lock_irq(&blkcg
->lock
);
469 * Note that stat reset is racy - it doesn't synchronize against
470 * stat updates. This is a debug feature which shouldn't exist
471 * anyway. If you get hit by a race, retry.
473 hlist_for_each_entry(blkg
, n
, &blkcg
->blkg_list
, blkcg_node
) {
474 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
475 struct blkcg_policy
*pol
= blkcg_policy
[i
];
477 if (blkcg_policy_enabled(blkg
->q
, pol
) &&
478 pol
->pd_reset_stats_fn
)
479 pol
->pd_reset_stats_fn(blkg
);
483 spin_unlock_irq(&blkcg
->lock
);
484 mutex_unlock(&blkcg_pol_mutex
);
488 static const char *blkg_dev_name(struct blkcg_gq
*blkg
)
490 /* some drivers (floppy) instantiate a queue w/o disk registered */
491 if (blkg
->q
->backing_dev_info
.dev
)
492 return dev_name(blkg
->q
->backing_dev_info
.dev
);
497 * blkcg_print_blkgs - helper for printing per-blkg data
498 * @sf: seq_file to print to
499 * @blkcg: blkcg of interest
500 * @prfill: fill function to print out a blkg
501 * @pol: policy in question
502 * @data: data to be passed to @prfill
503 * @show_total: to print out sum of prfill return values or not
505 * This function invokes @prfill on each blkg of @blkcg if pd for the
506 * policy specified by @pol exists. @prfill is invoked with @sf, the
507 * policy data and @data and the matching queue lock held. If @show_total
508 * is %true, the sum of the return values from @prfill is printed with
509 * "Total" label at the end.
511 * This is to be used to construct print functions for
512 * cftype->read_seq_string method.
514 void blkcg_print_blkgs(struct seq_file
*sf
, struct blkcg
*blkcg
,
515 u64 (*prfill
)(struct seq_file
*,
516 struct blkg_policy_data
*, int),
517 const struct blkcg_policy
*pol
, int data
,
520 struct blkcg_gq
*blkg
;
521 struct hlist_node
*n
;
525 hlist_for_each_entry_rcu(blkg
, n
, &blkcg
->blkg_list
, blkcg_node
) {
526 spin_lock_irq(blkg
->q
->queue_lock
);
527 if (blkcg_policy_enabled(blkg
->q
, pol
))
528 total
+= prfill(sf
, blkg
->pd
[pol
->plid
], data
);
529 spin_unlock_irq(blkg
->q
->queue_lock
);
534 seq_printf(sf
, "Total %llu\n", (unsigned long long)total
);
536 EXPORT_SYMBOL_GPL(blkcg_print_blkgs
);
539 * __blkg_prfill_u64 - prfill helper for a single u64 value
540 * @sf: seq_file to print to
541 * @pd: policy private data of interest
544 * Print @v to @sf for the device assocaited with @pd.
546 u64
__blkg_prfill_u64(struct seq_file
*sf
, struct blkg_policy_data
*pd
, u64 v
)
548 const char *dname
= blkg_dev_name(pd
->blkg
);
553 seq_printf(sf
, "%s %llu\n", dname
, (unsigned long long)v
);
556 EXPORT_SYMBOL_GPL(__blkg_prfill_u64
);
559 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
560 * @sf: seq_file to print to
561 * @pd: policy private data of interest
562 * @rwstat: rwstat to print
564 * Print @rwstat to @sf for the device assocaited with @pd.
566 u64
__blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
567 const struct blkg_rwstat
*rwstat
)
569 static const char *rwstr
[] = {
570 [BLKG_RWSTAT_READ
] = "Read",
571 [BLKG_RWSTAT_WRITE
] = "Write",
572 [BLKG_RWSTAT_SYNC
] = "Sync",
573 [BLKG_RWSTAT_ASYNC
] = "Async",
575 const char *dname
= blkg_dev_name(pd
->blkg
);
582 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
583 seq_printf(sf
, "%s %s %llu\n", dname
, rwstr
[i
],
584 (unsigned long long)rwstat
->cnt
[i
]);
586 v
= rwstat
->cnt
[BLKG_RWSTAT_READ
] + rwstat
->cnt
[BLKG_RWSTAT_WRITE
];
587 seq_printf(sf
, "%s Total %llu\n", dname
, (unsigned long long)v
);
590 EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat
);
593 * blkg_prfill_stat - prfill callback for blkg_stat
594 * @sf: seq_file to print to
595 * @pd: policy private data of interest
596 * @off: offset to the blkg_stat in @pd
598 * prfill callback for printing a blkg_stat.
600 u64
blkg_prfill_stat(struct seq_file
*sf
, struct blkg_policy_data
*pd
, int off
)
602 return __blkg_prfill_u64(sf
, pd
, blkg_stat_read((void *)pd
+ off
));
604 EXPORT_SYMBOL_GPL(blkg_prfill_stat
);
607 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
608 * @sf: seq_file to print to
609 * @pd: policy private data of interest
610 * @off: offset to the blkg_rwstat in @pd
612 * prfill callback for printing a blkg_rwstat.
614 u64
blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
617 struct blkg_rwstat rwstat
= blkg_rwstat_read((void *)pd
+ off
);
619 return __blkg_prfill_rwstat(sf
, pd
, &rwstat
);
621 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat
);
624 * blkg_stat_recursive_sum - collect hierarchical blkg_stat
625 * @pd: policy private data of interest
626 * @off: offset to the blkg_stat in @pd
628 * Collect the blkg_stat specified by @off from @pd and all its online
629 * descendants and return the sum. The caller must be holding the queue
630 * lock for online tests.
632 u64
blkg_stat_recursive_sum(struct blkg_policy_data
*pd
, int off
)
634 struct blkcg_policy
*pol
= blkcg_policy
[pd
->plid
];
635 struct blkcg_gq
*pos_blkg
;
636 struct cgroup
*pos_cgrp
;
639 lockdep_assert_held(pd
->blkg
->q
->queue_lock
);
641 sum
= blkg_stat_read((void *)pd
+ off
);
644 blkg_for_each_descendant_pre(pos_blkg
, pos_cgrp
, pd_to_blkg(pd
)) {
645 struct blkg_policy_data
*pos_pd
= blkg_to_pd(pos_blkg
, pol
);
646 struct blkg_stat
*stat
= (void *)pos_pd
+ off
;
648 if (pos_blkg
->online
)
649 sum
+= blkg_stat_read(stat
);
655 EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum
);
658 * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat
659 * @pd: policy private data of interest
660 * @off: offset to the blkg_stat in @pd
662 * Collect the blkg_rwstat specified by @off from @pd and all its online
663 * descendants and return the sum. The caller must be holding the queue
664 * lock for online tests.
666 struct blkg_rwstat
blkg_rwstat_recursive_sum(struct blkg_policy_data
*pd
,
669 struct blkcg_policy
*pol
= blkcg_policy
[pd
->plid
];
670 struct blkcg_gq
*pos_blkg
;
671 struct cgroup
*pos_cgrp
;
672 struct blkg_rwstat sum
;
675 lockdep_assert_held(pd
->blkg
->q
->queue_lock
);
677 sum
= blkg_rwstat_read((void *)pd
+ off
);
680 blkg_for_each_descendant_pre(pos_blkg
, pos_cgrp
, pd_to_blkg(pd
)) {
681 struct blkg_policy_data
*pos_pd
= blkg_to_pd(pos_blkg
, pol
);
682 struct blkg_rwstat
*rwstat
= (void *)pos_pd
+ off
;
683 struct blkg_rwstat tmp
;
685 if (!pos_blkg
->online
)
688 tmp
= blkg_rwstat_read(rwstat
);
690 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
691 sum
.cnt
[i
] += tmp
.cnt
[i
];
697 EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum
);
700 * blkg_conf_prep - parse and prepare for per-blkg config update
701 * @blkcg: target block cgroup
702 * @pol: target policy
703 * @input: input string
704 * @ctx: blkg_conf_ctx to be filled
706 * Parse per-blkg config update from @input and initialize @ctx with the
707 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
708 * value. This function returns with RCU read lock and queue lock held and
709 * must be paired with blkg_conf_finish().
711 int blkg_conf_prep(struct blkcg
*blkcg
, const struct blkcg_policy
*pol
,
712 const char *input
, struct blkg_conf_ctx
*ctx
)
713 __acquires(rcu
) __acquires(disk
->queue
->queue_lock
)
715 struct gendisk
*disk
;
716 struct blkcg_gq
*blkg
;
717 unsigned int major
, minor
;
718 unsigned long long v
;
721 if (sscanf(input
, "%u:%u %llu", &major
, &minor
, &v
) != 3)
724 disk
= get_gendisk(MKDEV(major
, minor
), &part
);
729 spin_lock_irq(disk
->queue
->queue_lock
);
731 if (blkcg_policy_enabled(disk
->queue
, pol
))
732 blkg
= blkg_lookup_create(blkcg
, disk
->queue
);
734 blkg
= ERR_PTR(-EINVAL
);
739 spin_unlock_irq(disk
->queue
->queue_lock
);
742 * If queue was bypassing, we should retry. Do so after a
743 * short msleep(). It isn't strictly necessary but queue
744 * can be bypassing for some time and it's always nice to
745 * avoid busy looping.
749 ret
= restart_syscall();
759 EXPORT_SYMBOL_GPL(blkg_conf_prep
);
762 * blkg_conf_finish - finish up per-blkg config update
763 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
765 * Finish up after per-blkg config update. This function must be paired
766 * with blkg_conf_prep().
768 void blkg_conf_finish(struct blkg_conf_ctx
*ctx
)
769 __releases(ctx
->disk
->queue
->queue_lock
) __releases(rcu
)
771 spin_unlock_irq(ctx
->disk
->queue
->queue_lock
);
775 EXPORT_SYMBOL_GPL(blkg_conf_finish
);
777 struct cftype blkcg_files
[] = {
779 .name
= "reset_stats",
780 .write_u64
= blkcg_reset_stats
,
786 * blkcg_css_offline - cgroup css_offline callback
787 * @cgroup: cgroup of interest
789 * This function is called when @cgroup is about to go away and responsible
790 * for shooting down all blkgs associated with @cgroup. blkgs should be
791 * removed while holding both q and blkcg locks. As blkcg lock is nested
792 * inside q lock, this function performs reverse double lock dancing.
794 * This is the blkcg counterpart of ioc_release_fn().
796 static void blkcg_css_offline(struct cgroup
*cgroup
)
798 struct blkcg
*blkcg
= cgroup_to_blkcg(cgroup
);
800 spin_lock_irq(&blkcg
->lock
);
802 while (!hlist_empty(&blkcg
->blkg_list
)) {
803 struct blkcg_gq
*blkg
= hlist_entry(blkcg
->blkg_list
.first
,
804 struct blkcg_gq
, blkcg_node
);
805 struct request_queue
*q
= blkg
->q
;
807 if (spin_trylock(q
->queue_lock
)) {
809 spin_unlock(q
->queue_lock
);
811 spin_unlock_irq(&blkcg
->lock
);
813 spin_lock_irq(&blkcg
->lock
);
817 spin_unlock_irq(&blkcg
->lock
);
820 static void blkcg_css_free(struct cgroup
*cgroup
)
822 struct blkcg
*blkcg
= cgroup_to_blkcg(cgroup
);
824 if (blkcg
!= &blkcg_root
)
828 static struct cgroup_subsys_state
*blkcg_css_alloc(struct cgroup
*cgroup
)
830 static atomic64_t id_seq
= ATOMIC64_INIT(0);
832 struct cgroup
*parent
= cgroup
->parent
;
839 blkcg
= kzalloc(sizeof(*blkcg
), GFP_KERNEL
);
841 return ERR_PTR(-ENOMEM
);
843 blkcg
->cfq_weight
= CFQ_WEIGHT_DEFAULT
;
844 blkcg
->cfq_leaf_weight
= CFQ_WEIGHT_DEFAULT
;
845 blkcg
->id
= atomic64_inc_return(&id_seq
); /* root is 0, start from 1 */
847 spin_lock_init(&blkcg
->lock
);
848 INIT_RADIX_TREE(&blkcg
->blkg_tree
, GFP_ATOMIC
);
849 INIT_HLIST_HEAD(&blkcg
->blkg_list
);
855 * blkcg_init_queue - initialize blkcg part of request queue
856 * @q: request_queue to initialize
858 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
859 * part of new request_queue @q.
862 * 0 on success, -errno on failure.
864 int blkcg_init_queue(struct request_queue
*q
)
868 return blk_throtl_init(q
);
872 * blkcg_drain_queue - drain blkcg part of request_queue
873 * @q: request_queue to drain
875 * Called from blk_drain_queue(). Responsible for draining blkcg part.
877 void blkcg_drain_queue(struct request_queue
*q
)
879 lockdep_assert_held(q
->queue_lock
);
885 * blkcg_exit_queue - exit and release blkcg part of request_queue
886 * @q: request_queue being released
888 * Called from blk_release_queue(). Responsible for exiting blkcg part.
890 void blkcg_exit_queue(struct request_queue
*q
)
892 spin_lock_irq(q
->queue_lock
);
894 spin_unlock_irq(q
->queue_lock
);
900 * We cannot support shared io contexts, as we have no mean to support
901 * two tasks with the same ioc in two different groups without major rework
902 * of the main cic data structures. For now we allow a task to change
903 * its cgroup only if it's the only owner of its ioc.
905 static int blkcg_can_attach(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
)
907 struct task_struct
*task
;
908 struct io_context
*ioc
;
911 /* task_lock() is needed to avoid races with exit_io_context() */
912 cgroup_taskset_for_each(task
, cgrp
, tset
) {
914 ioc
= task
->io_context
;
915 if (ioc
&& atomic_read(&ioc
->nr_tasks
) > 1)
924 struct cgroup_subsys blkio_subsys
= {
926 .css_alloc
= blkcg_css_alloc
,
927 .css_offline
= blkcg_css_offline
,
928 .css_free
= blkcg_css_free
,
929 .can_attach
= blkcg_can_attach
,
930 .subsys_id
= blkio_subsys_id
,
931 .base_cftypes
= blkcg_files
,
932 .module
= THIS_MODULE
,
935 * blkio subsystem is utterly broken in terms of hierarchy support.
936 * It treats all cgroups equally regardless of where they're
937 * located in the hierarchy - all cgroups are treated as if they're
938 * right below the root. Fix it and remove the following.
940 .broken_hierarchy
= true,
942 EXPORT_SYMBOL_GPL(blkio_subsys
);
945 * blkcg_activate_policy - activate a blkcg policy on a request_queue
946 * @q: request_queue of interest
947 * @pol: blkcg policy to activate
949 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
950 * bypass mode to populate its blkgs with policy_data for @pol.
952 * Activation happens with @q bypassed, so nobody would be accessing blkgs
953 * from IO path. Update of each blkg is protected by both queue and blkcg
954 * locks so that holding either lock and testing blkcg_policy_enabled() is
955 * always enough for dereferencing policy data.
957 * The caller is responsible for synchronizing [de]activations and policy
958 * [un]registerations. Returns 0 on success, -errno on failure.
960 int blkcg_activate_policy(struct request_queue
*q
,
961 const struct blkcg_policy
*pol
)
964 struct blkcg_gq
*blkg
, *new_blkg
;
965 struct blkg_policy_data
*pd
, *n
;
969 if (blkcg_policy_enabled(q
, pol
))
972 /* preallocations for root blkg */
973 new_blkg
= blkg_alloc(&blkcg_root
, q
, GFP_KERNEL
);
977 preloaded
= !radix_tree_preload(GFP_KERNEL
);
979 blk_queue_bypass_start(q
);
982 * Make sure the root blkg exists and count the existing blkgs. As
983 * @q is bypassing at this point, blkg_lookup_create() can't be
984 * used. Open code it.
986 spin_lock_irq(q
->queue_lock
);
989 blkg
= __blkg_lookup(&blkcg_root
, q
, false);
993 blkg
= blkg_create(&blkcg_root
, q
, new_blkg
);
997 radix_tree_preload_end();
1000 ret
= PTR_ERR(blkg
);
1003 q
->root_blkg
= blkg
;
1004 q
->root_rl
.blkg
= blkg
;
1006 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
)
1009 spin_unlock_irq(q
->queue_lock
);
1011 /* allocate policy_data for all existing blkgs */
1013 pd
= kzalloc_node(pol
->pd_size
, GFP_KERNEL
, q
->node
);
1018 list_add_tail(&pd
->alloc_node
, &pds
);
1022 * Install the allocated pds. With @q bypassing, no new blkg
1023 * should have been created while the queue lock was dropped.
1025 spin_lock_irq(q
->queue_lock
);
1027 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
) {
1028 if (WARN_ON(list_empty(&pds
))) {
1029 /* umm... this shouldn't happen, just abort */
1033 pd
= list_first_entry(&pds
, struct blkg_policy_data
, alloc_node
);
1034 list_del_init(&pd
->alloc_node
);
1036 /* grab blkcg lock too while installing @pd on @blkg */
1037 spin_lock(&blkg
->blkcg
->lock
);
1039 blkg
->pd
[pol
->plid
] = pd
;
1041 pd
->plid
= pol
->plid
;
1042 pol
->pd_init_fn(blkg
);
1044 spin_unlock(&blkg
->blkcg
->lock
);
1047 __set_bit(pol
->plid
, q
->blkcg_pols
);
1050 spin_unlock_irq(q
->queue_lock
);
1052 blk_queue_bypass_end(q
);
1053 list_for_each_entry_safe(pd
, n
, &pds
, alloc_node
)
1057 EXPORT_SYMBOL_GPL(blkcg_activate_policy
);
1060 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
1061 * @q: request_queue of interest
1062 * @pol: blkcg policy to deactivate
1064 * Deactivate @pol on @q. Follows the same synchronization rules as
1065 * blkcg_activate_policy().
1067 void blkcg_deactivate_policy(struct request_queue
*q
,
1068 const struct blkcg_policy
*pol
)
1070 struct blkcg_gq
*blkg
;
1072 if (!blkcg_policy_enabled(q
, pol
))
1075 blk_queue_bypass_start(q
);
1076 spin_lock_irq(q
->queue_lock
);
1078 __clear_bit(pol
->plid
, q
->blkcg_pols
);
1080 /* if no policy is left, no need for blkgs - shoot them down */
1081 if (bitmap_empty(q
->blkcg_pols
, BLKCG_MAX_POLS
))
1082 blkg_destroy_all(q
);
1084 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
) {
1085 /* grab blkcg lock too while removing @pd from @blkg */
1086 spin_lock(&blkg
->blkcg
->lock
);
1088 if (pol
->pd_offline_fn
)
1089 pol
->pd_offline_fn(blkg
);
1090 if (pol
->pd_exit_fn
)
1091 pol
->pd_exit_fn(blkg
);
1093 kfree(blkg
->pd
[pol
->plid
]);
1094 blkg
->pd
[pol
->plid
] = NULL
;
1096 spin_unlock(&blkg
->blkcg
->lock
);
1099 spin_unlock_irq(q
->queue_lock
);
1100 blk_queue_bypass_end(q
);
1102 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy
);
1105 * blkcg_policy_register - register a blkcg policy
1106 * @pol: blkcg policy to register
1108 * Register @pol with blkcg core. Might sleep and @pol may be modified on
1109 * successful registration. Returns 0 on success and -errno on failure.
1111 int blkcg_policy_register(struct blkcg_policy
*pol
)
1115 if (WARN_ON(pol
->pd_size
< sizeof(struct blkg_policy_data
)))
1118 mutex_lock(&blkcg_pol_mutex
);
1120 /* find an empty slot */
1122 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++)
1123 if (!blkcg_policy
[i
])
1125 if (i
>= BLKCG_MAX_POLS
)
1128 /* register and update blkgs */
1130 blkcg_policy
[i
] = pol
;
1132 /* everything is in place, add intf files for the new policy */
1134 WARN_ON(cgroup_add_cftypes(&blkio_subsys
, pol
->cftypes
));
1137 mutex_unlock(&blkcg_pol_mutex
);
1140 EXPORT_SYMBOL_GPL(blkcg_policy_register
);
1143 * blkcg_policy_unregister - unregister a blkcg policy
1144 * @pol: blkcg policy to unregister
1146 * Undo blkcg_policy_register(@pol). Might sleep.
1148 void blkcg_policy_unregister(struct blkcg_policy
*pol
)
1150 mutex_lock(&blkcg_pol_mutex
);
1152 if (WARN_ON(blkcg_policy
[pol
->plid
] != pol
))
1155 /* kill the intf files first */
1157 cgroup_rm_cftypes(&blkio_subsys
, pol
->cftypes
);
1159 /* unregister and update blkgs */
1160 blkcg_policy
[pol
->plid
] = NULL
;
1162 mutex_unlock(&blkcg_pol_mutex
);
1164 EXPORT_SYMBOL_GPL(blkcg_policy_unregister
);