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 * For policy-specific per-blkcg data:
14 * Copyright (C) 2015 Paolo Valente <paolo.valente@unimore.it>
15 * Arianna Avanzini <avanzini.arianna@gmail.com>
17 #include <linux/ioprio.h>
18 #include <linux/kdev_t.h>
19 #include <linux/module.h>
20 #include <linux/err.h>
21 #include <linux/blkdev.h>
22 #include <linux/backing-dev.h>
23 #include <linux/slab.h>
24 #include <linux/genhd.h>
25 #include <linux/delay.h>
26 #include <linux/atomic.h>
27 #include <linux/ctype.h>
28 #include <linux/blk-cgroup.h>
31 #define MAX_KEY_LEN 100
34 * blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation.
35 * blkcg_pol_register_mutex nests outside of it and synchronizes entire
36 * policy [un]register operations including cgroup file additions /
37 * removals. Putting cgroup file registration outside blkcg_pol_mutex
38 * allows grabbing it from cgroup callbacks.
40 static DEFINE_MUTEX(blkcg_pol_register_mutex
);
41 static DEFINE_MUTEX(blkcg_pol_mutex
);
43 struct blkcg blkcg_root
;
44 EXPORT_SYMBOL_GPL(blkcg_root
);
46 struct cgroup_subsys_state
* const blkcg_root_css
= &blkcg_root
.css
;
48 static struct blkcg_policy
*blkcg_policy
[BLKCG_MAX_POLS
];
50 static LIST_HEAD(all_blkcgs
); /* protected by blkcg_pol_mutex */
52 static bool blkcg_policy_enabled(struct request_queue
*q
,
53 const struct blkcg_policy
*pol
)
55 return pol
&& test_bit(pol
->plid
, q
->blkcg_pols
);
59 * blkg_free - free a blkg
62 * Free @blkg which may be partially allocated.
64 static void blkg_free(struct blkcg_gq
*blkg
)
71 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++)
73 blkcg_policy
[i
]->pd_free_fn(blkg
->pd
[i
]);
75 if (blkg
->blkcg
!= &blkcg_root
)
76 blk_exit_rl(&blkg
->rl
);
78 blkg_rwstat_exit(&blkg
->stat_ios
);
79 blkg_rwstat_exit(&blkg
->stat_bytes
);
84 * blkg_alloc - allocate a blkg
85 * @blkcg: block cgroup the new blkg is associated with
86 * @q: request_queue the new blkg is associated with
87 * @gfp_mask: allocation mask to use
89 * Allocate a new blkg assocating @blkcg and @q.
91 static struct blkcg_gq
*blkg_alloc(struct blkcg
*blkcg
, struct request_queue
*q
,
94 struct blkcg_gq
*blkg
;
97 /* alloc and init base part */
98 blkg
= kzalloc_node(sizeof(*blkg
), gfp_mask
, q
->node
);
102 if (blkg_rwstat_init(&blkg
->stat_bytes
, gfp_mask
) ||
103 blkg_rwstat_init(&blkg
->stat_ios
, gfp_mask
))
107 INIT_LIST_HEAD(&blkg
->q_node
);
109 atomic_set(&blkg
->refcnt
, 1);
111 /* root blkg uses @q->root_rl, init rl only for !root blkgs */
112 if (blkcg
!= &blkcg_root
) {
113 if (blk_init_rl(&blkg
->rl
, q
, gfp_mask
))
115 blkg
->rl
.blkg
= blkg
;
118 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
119 struct blkcg_policy
*pol
= blkcg_policy
[i
];
120 struct blkg_policy_data
*pd
;
122 if (!blkcg_policy_enabled(q
, pol
))
125 /* alloc per-policy data and attach it to blkg */
126 pd
= pol
->pd_alloc_fn(gfp_mask
, q
->node
);
142 struct blkcg_gq
*blkg_lookup_slowpath(struct blkcg
*blkcg
,
143 struct request_queue
*q
, bool update_hint
)
145 struct blkcg_gq
*blkg
;
148 * Hint didn't match. Look up from the radix tree. Note that the
149 * hint can only be updated under queue_lock as otherwise @blkg
150 * could have already been removed from blkg_tree. The caller is
151 * responsible for grabbing queue_lock if @update_hint.
153 blkg
= radix_tree_lookup(&blkcg
->blkg_tree
, q
->id
);
154 if (blkg
&& blkg
->q
== q
) {
156 lockdep_assert_held(q
->queue_lock
);
157 rcu_assign_pointer(blkcg
->blkg_hint
, blkg
);
164 EXPORT_SYMBOL_GPL(blkg_lookup_slowpath
);
167 * If @new_blkg is %NULL, this function tries to allocate a new one as
168 * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return.
170 static struct blkcg_gq
*blkg_create(struct blkcg
*blkcg
,
171 struct request_queue
*q
,
172 struct blkcg_gq
*new_blkg
)
174 struct blkcg_gq
*blkg
;
175 struct bdi_writeback_congested
*wb_congested
;
178 WARN_ON_ONCE(!rcu_read_lock_held());
179 lockdep_assert_held(q
->queue_lock
);
181 /* blkg holds a reference to blkcg */
182 if (!css_tryget_online(&blkcg
->css
)) {
187 wb_congested
= wb_congested_get_create(&q
->backing_dev_info
,
188 blkcg
->css
.id
, GFP_NOWAIT
);
196 new_blkg
= blkg_alloc(blkcg
, q
, GFP_NOWAIT
);
197 if (unlikely(!new_blkg
)) {
199 goto err_put_congested
;
203 blkg
->wb_congested
= wb_congested
;
206 if (blkcg_parent(blkcg
)) {
207 blkg
->parent
= __blkg_lookup(blkcg_parent(blkcg
), q
, false);
208 if (WARN_ON_ONCE(!blkg
->parent
)) {
210 goto err_put_congested
;
212 blkg_get(blkg
->parent
);
215 /* invoke per-policy init */
216 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
217 struct blkcg_policy
*pol
= blkcg_policy
[i
];
219 if (blkg
->pd
[i
] && pol
->pd_init_fn
)
220 pol
->pd_init_fn(blkg
->pd
[i
]);
224 spin_lock(&blkcg
->lock
);
225 ret
= radix_tree_insert(&blkcg
->blkg_tree
, q
->id
, blkg
);
227 hlist_add_head_rcu(&blkg
->blkcg_node
, &blkcg
->blkg_list
);
228 list_add(&blkg
->q_node
, &q
->blkg_list
);
230 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
231 struct blkcg_policy
*pol
= blkcg_policy
[i
];
233 if (blkg
->pd
[i
] && pol
->pd_online_fn
)
234 pol
->pd_online_fn(blkg
->pd
[i
]);
238 spin_unlock(&blkcg
->lock
);
243 /* @blkg failed fully initialized, use the usual release path */
248 wb_congested_put(wb_congested
);
250 css_put(&blkcg
->css
);
257 * blkg_lookup_create - lookup blkg, try to create one if not there
258 * @blkcg: blkcg of interest
259 * @q: request_queue of interest
261 * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
262 * create one. blkg creation is performed recursively from blkcg_root such
263 * that all non-root blkg's have access to the parent blkg. This function
264 * should be called under RCU read lock and @q->queue_lock.
266 * Returns pointer to the looked up or created blkg on success, ERR_PTR()
267 * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not
268 * dead and bypassing, returns ERR_PTR(-EBUSY).
270 struct blkcg_gq
*blkg_lookup_create(struct blkcg
*blkcg
,
271 struct request_queue
*q
)
273 struct blkcg_gq
*blkg
;
275 WARN_ON_ONCE(!rcu_read_lock_held());
276 lockdep_assert_held(q
->queue_lock
);
279 * This could be the first entry point of blkcg implementation and
280 * we shouldn't allow anything to go through for a bypassing queue.
282 if (unlikely(blk_queue_bypass(q
)))
283 return ERR_PTR(blk_queue_dying(q
) ? -ENODEV
: -EBUSY
);
285 blkg
= __blkg_lookup(blkcg
, q
, true);
290 * Create blkgs walking down from blkcg_root to @blkcg, so that all
291 * non-root blkgs have access to their parents.
294 struct blkcg
*pos
= blkcg
;
295 struct blkcg
*parent
= blkcg_parent(blkcg
);
297 while (parent
&& !__blkg_lookup(parent
, q
, false)) {
299 parent
= blkcg_parent(parent
);
302 blkg
= blkg_create(pos
, q
, NULL
);
303 if (pos
== blkcg
|| IS_ERR(blkg
))
308 static void blkg_destroy(struct blkcg_gq
*blkg
)
310 struct blkcg
*blkcg
= blkg
->blkcg
;
311 struct blkcg_gq
*parent
= blkg
->parent
;
314 lockdep_assert_held(blkg
->q
->queue_lock
);
315 lockdep_assert_held(&blkcg
->lock
);
317 /* Something wrong if we are trying to remove same group twice */
318 WARN_ON_ONCE(list_empty(&blkg
->q_node
));
319 WARN_ON_ONCE(hlist_unhashed(&blkg
->blkcg_node
));
321 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
322 struct blkcg_policy
*pol
= blkcg_policy
[i
];
324 if (blkg
->pd
[i
] && pol
->pd_offline_fn
)
325 pol
->pd_offline_fn(blkg
->pd
[i
]);
329 blkg_rwstat_add_aux(&parent
->stat_bytes
, &blkg
->stat_bytes
);
330 blkg_rwstat_add_aux(&parent
->stat_ios
, &blkg
->stat_ios
);
333 blkg
->online
= false;
335 radix_tree_delete(&blkcg
->blkg_tree
, blkg
->q
->id
);
336 list_del_init(&blkg
->q_node
);
337 hlist_del_init_rcu(&blkg
->blkcg_node
);
340 * Both setting lookup hint to and clearing it from @blkg are done
341 * under queue_lock. If it's not pointing to @blkg now, it never
342 * will. Hint assignment itself can race safely.
344 if (rcu_access_pointer(blkcg
->blkg_hint
) == blkg
)
345 rcu_assign_pointer(blkcg
->blkg_hint
, NULL
);
348 * Put the reference taken at the time of creation so that when all
349 * queues are gone, group can be destroyed.
355 * blkg_destroy_all - destroy all blkgs associated with a request_queue
356 * @q: request_queue of interest
358 * Destroy all blkgs associated with @q.
360 static void blkg_destroy_all(struct request_queue
*q
)
362 struct blkcg_gq
*blkg
, *n
;
364 lockdep_assert_held(q
->queue_lock
);
366 list_for_each_entry_safe(blkg
, n
, &q
->blkg_list
, q_node
) {
367 struct blkcg
*blkcg
= blkg
->blkcg
;
369 spin_lock(&blkcg
->lock
);
371 spin_unlock(&blkcg
->lock
);
376 * A group is RCU protected, but having an rcu lock does not mean that one
377 * can access all the fields of blkg and assume these are valid. For
378 * example, don't try to follow throtl_data and request queue links.
380 * Having a reference to blkg under an rcu allows accesses to only values
381 * local to groups like group stats and group rate limits.
383 void __blkg_release_rcu(struct rcu_head
*rcu_head
)
385 struct blkcg_gq
*blkg
= container_of(rcu_head
, struct blkcg_gq
, rcu_head
);
387 /* release the blkcg and parent blkg refs this blkg has been holding */
388 css_put(&blkg
->blkcg
->css
);
390 blkg_put(blkg
->parent
);
392 wb_congested_put(blkg
->wb_congested
);
396 EXPORT_SYMBOL_GPL(__blkg_release_rcu
);
399 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
400 * because the root blkg uses @q->root_rl instead of its own rl.
402 struct request_list
*__blk_queue_next_rl(struct request_list
*rl
,
403 struct request_queue
*q
)
405 struct list_head
*ent
;
406 struct blkcg_gq
*blkg
;
409 * Determine the current blkg list_head. The first entry is
410 * root_rl which is off @q->blkg_list and mapped to the head.
412 if (rl
== &q
->root_rl
) {
414 /* There are no more block groups, hence no request lists */
418 blkg
= container_of(rl
, struct blkcg_gq
, rl
);
422 /* walk to the next list_head, skip root blkcg */
424 if (ent
== &q
->root_blkg
->q_node
)
426 if (ent
== &q
->blkg_list
)
429 blkg
= container_of(ent
, struct blkcg_gq
, q_node
);
433 static int blkcg_reset_stats(struct cgroup_subsys_state
*css
,
434 struct cftype
*cftype
, u64 val
)
436 struct blkcg
*blkcg
= css_to_blkcg(css
);
437 struct blkcg_gq
*blkg
;
440 mutex_lock(&blkcg_pol_mutex
);
441 spin_lock_irq(&blkcg
->lock
);
444 * Note that stat reset is racy - it doesn't synchronize against
445 * stat updates. This is a debug feature which shouldn't exist
446 * anyway. If you get hit by a race, retry.
448 hlist_for_each_entry(blkg
, &blkcg
->blkg_list
, blkcg_node
) {
449 blkg_rwstat_reset(&blkg
->stat_bytes
);
450 blkg_rwstat_reset(&blkg
->stat_ios
);
452 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
453 struct blkcg_policy
*pol
= blkcg_policy
[i
];
455 if (blkg
->pd
[i
] && pol
->pd_reset_stats_fn
)
456 pol
->pd_reset_stats_fn(blkg
->pd
[i
]);
460 spin_unlock_irq(&blkcg
->lock
);
461 mutex_unlock(&blkcg_pol_mutex
);
465 const char *blkg_dev_name(struct blkcg_gq
*blkg
)
467 /* some drivers (floppy) instantiate a queue w/o disk registered */
468 if (blkg
->q
->backing_dev_info
.dev
)
469 return dev_name(blkg
->q
->backing_dev_info
.dev
);
472 EXPORT_SYMBOL_GPL(blkg_dev_name
);
475 * blkcg_print_blkgs - helper for printing per-blkg data
476 * @sf: seq_file to print to
477 * @blkcg: blkcg of interest
478 * @prfill: fill function to print out a blkg
479 * @pol: policy in question
480 * @data: data to be passed to @prfill
481 * @show_total: to print out sum of prfill return values or not
483 * This function invokes @prfill on each blkg of @blkcg if pd for the
484 * policy specified by @pol exists. @prfill is invoked with @sf, the
485 * policy data and @data and the matching queue lock held. If @show_total
486 * is %true, the sum of the return values from @prfill is printed with
487 * "Total" label at the end.
489 * This is to be used to construct print functions for
490 * cftype->read_seq_string method.
492 void blkcg_print_blkgs(struct seq_file
*sf
, struct blkcg
*blkcg
,
493 u64 (*prfill
)(struct seq_file
*,
494 struct blkg_policy_data
*, int),
495 const struct blkcg_policy
*pol
, int data
,
498 struct blkcg_gq
*blkg
;
502 hlist_for_each_entry_rcu(blkg
, &blkcg
->blkg_list
, blkcg_node
) {
503 spin_lock_irq(blkg
->q
->queue_lock
);
504 if (blkcg_policy_enabled(blkg
->q
, pol
))
505 total
+= prfill(sf
, blkg
->pd
[pol
->plid
], data
);
506 spin_unlock_irq(blkg
->q
->queue_lock
);
511 seq_printf(sf
, "Total %llu\n", (unsigned long long)total
);
513 EXPORT_SYMBOL_GPL(blkcg_print_blkgs
);
516 * __blkg_prfill_u64 - prfill helper for a single u64 value
517 * @sf: seq_file to print to
518 * @pd: policy private data of interest
521 * Print @v to @sf for the device assocaited with @pd.
523 u64
__blkg_prfill_u64(struct seq_file
*sf
, struct blkg_policy_data
*pd
, u64 v
)
525 const char *dname
= blkg_dev_name(pd
->blkg
);
530 seq_printf(sf
, "%s %llu\n", dname
, (unsigned long long)v
);
533 EXPORT_SYMBOL_GPL(__blkg_prfill_u64
);
536 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
537 * @sf: seq_file to print to
538 * @pd: policy private data of interest
539 * @rwstat: rwstat to print
541 * Print @rwstat to @sf for the device assocaited with @pd.
543 u64
__blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
544 const struct blkg_rwstat
*rwstat
)
546 static const char *rwstr
[] = {
547 [BLKG_RWSTAT_READ
] = "Read",
548 [BLKG_RWSTAT_WRITE
] = "Write",
549 [BLKG_RWSTAT_SYNC
] = "Sync",
550 [BLKG_RWSTAT_ASYNC
] = "Async",
552 const char *dname
= blkg_dev_name(pd
->blkg
);
559 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
560 seq_printf(sf
, "%s %s %llu\n", dname
, rwstr
[i
],
561 (unsigned long long)atomic64_read(&rwstat
->aux_cnt
[i
]));
563 v
= atomic64_read(&rwstat
->aux_cnt
[BLKG_RWSTAT_READ
]) +
564 atomic64_read(&rwstat
->aux_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
);
601 static u64
blkg_prfill_rwstat_field(struct seq_file
*sf
,
602 struct blkg_policy_data
*pd
, int off
)
604 struct blkg_rwstat rwstat
= blkg_rwstat_read((void *)pd
->blkg
+ off
);
606 return __blkg_prfill_rwstat(sf
, pd
, &rwstat
);
610 * blkg_print_stat_bytes - seq_show callback for blkg->stat_bytes
611 * @sf: seq_file to print to
614 * To be used as cftype->seq_show to print blkg->stat_bytes.
615 * cftype->private must be set to the blkcg_policy.
617 int blkg_print_stat_bytes(struct seq_file
*sf
, void *v
)
619 blkcg_print_blkgs(sf
, css_to_blkcg(seq_css(sf
)),
620 blkg_prfill_rwstat_field
, (void *)seq_cft(sf
)->private,
621 offsetof(struct blkcg_gq
, stat_bytes
), true);
624 EXPORT_SYMBOL_GPL(blkg_print_stat_bytes
);
627 * blkg_print_stat_bytes - seq_show callback for blkg->stat_ios
628 * @sf: seq_file to print to
631 * To be used as cftype->seq_show to print blkg->stat_ios. cftype->private
632 * must be set to the blkcg_policy.
634 int blkg_print_stat_ios(struct seq_file
*sf
, void *v
)
636 blkcg_print_blkgs(sf
, css_to_blkcg(seq_css(sf
)),
637 blkg_prfill_rwstat_field
, (void *)seq_cft(sf
)->private,
638 offsetof(struct blkcg_gq
, stat_ios
), true);
641 EXPORT_SYMBOL_GPL(blkg_print_stat_ios
);
643 static u64
blkg_prfill_rwstat_field_recursive(struct seq_file
*sf
,
644 struct blkg_policy_data
*pd
,
647 struct blkg_rwstat rwstat
= blkg_rwstat_recursive_sum(pd
->blkg
,
649 return __blkg_prfill_rwstat(sf
, pd
, &rwstat
);
653 * blkg_print_stat_bytes_recursive - recursive version of blkg_print_stat_bytes
654 * @sf: seq_file to print to
657 int blkg_print_stat_bytes_recursive(struct seq_file
*sf
, void *v
)
659 blkcg_print_blkgs(sf
, css_to_blkcg(seq_css(sf
)),
660 blkg_prfill_rwstat_field_recursive
,
661 (void *)seq_cft(sf
)->private,
662 offsetof(struct blkcg_gq
, stat_bytes
), true);
665 EXPORT_SYMBOL_GPL(blkg_print_stat_bytes_recursive
);
668 * blkg_print_stat_ios_recursive - recursive version of blkg_print_stat_ios
669 * @sf: seq_file to print to
672 int blkg_print_stat_ios_recursive(struct seq_file
*sf
, void *v
)
674 blkcg_print_blkgs(sf
, css_to_blkcg(seq_css(sf
)),
675 blkg_prfill_rwstat_field_recursive
,
676 (void *)seq_cft(sf
)->private,
677 offsetof(struct blkcg_gq
, stat_ios
), true);
680 EXPORT_SYMBOL_GPL(blkg_print_stat_ios_recursive
);
683 * blkg_stat_recursive_sum - collect hierarchical blkg_stat
684 * @blkg: blkg of interest
685 * @pol: blkcg_policy which contains the blkg_stat
686 * @off: offset to the blkg_stat in blkg_policy_data or @blkg
688 * Collect the blkg_stat specified by @blkg, @pol and @off and all its
689 * online descendants and their aux counts. The caller must be holding the
690 * queue lock for online tests.
692 * If @pol is NULL, blkg_stat is at @off bytes into @blkg; otherwise, it is
693 * at @off bytes into @blkg's blkg_policy_data of the policy.
695 u64
blkg_stat_recursive_sum(struct blkcg_gq
*blkg
,
696 struct blkcg_policy
*pol
, int off
)
698 struct blkcg_gq
*pos_blkg
;
699 struct cgroup_subsys_state
*pos_css
;
702 lockdep_assert_held(blkg
->q
->queue_lock
);
705 blkg_for_each_descendant_pre(pos_blkg
, pos_css
, blkg
) {
706 struct blkg_stat
*stat
;
708 if (!pos_blkg
->online
)
712 stat
= (void *)blkg_to_pd(pos_blkg
, pol
) + off
;
714 stat
= (void *)blkg
+ off
;
716 sum
+= blkg_stat_read(stat
) + atomic64_read(&stat
->aux_cnt
);
722 EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum
);
725 * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat
726 * @blkg: blkg of interest
727 * @pol: blkcg_policy which contains the blkg_rwstat
728 * @off: offset to the blkg_rwstat in blkg_policy_data or @blkg
730 * Collect the blkg_rwstat specified by @blkg, @pol and @off and all its
731 * online descendants and their aux counts. The caller must be holding the
732 * queue lock for online tests.
734 * If @pol is NULL, blkg_rwstat is at @off bytes into @blkg; otherwise, it
735 * is at @off bytes into @blkg's blkg_policy_data of the policy.
737 struct blkg_rwstat
blkg_rwstat_recursive_sum(struct blkcg_gq
*blkg
,
738 struct blkcg_policy
*pol
, int off
)
740 struct blkcg_gq
*pos_blkg
;
741 struct cgroup_subsys_state
*pos_css
;
742 struct blkg_rwstat sum
= { };
745 lockdep_assert_held(blkg
->q
->queue_lock
);
748 blkg_for_each_descendant_pre(pos_blkg
, pos_css
, blkg
) {
749 struct blkg_rwstat
*rwstat
;
751 if (!pos_blkg
->online
)
755 rwstat
= (void *)blkg_to_pd(pos_blkg
, pol
) + off
;
757 rwstat
= (void *)pos_blkg
+ off
;
759 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
760 atomic64_add(atomic64_read(&rwstat
->aux_cnt
[i
]) +
761 percpu_counter_sum_positive(&rwstat
->cpu_cnt
[i
]),
768 EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum
);
771 * blkg_conf_prep - parse and prepare for per-blkg config update
772 * @blkcg: target block cgroup
773 * @pol: target policy
774 * @input: input string
775 * @ctx: blkg_conf_ctx to be filled
777 * Parse per-blkg config update from @input and initialize @ctx with the
778 * result. @ctx->blkg points to the blkg to be updated and @ctx->body the
779 * part of @input following MAJ:MIN. This function returns with RCU read
780 * lock and queue lock held and must be paired with blkg_conf_finish().
782 int blkg_conf_prep(struct blkcg
*blkcg
, const struct blkcg_policy
*pol
,
783 char *input
, struct blkg_conf_ctx
*ctx
)
784 __acquires(rcu
) __acquires(disk
->queue
->queue_lock
)
786 struct gendisk
*disk
;
787 struct blkcg_gq
*blkg
;
788 unsigned int major
, minor
;
789 int key_len
, part
, ret
;
792 if (sscanf(input
, "%u:%u%n", &major
, &minor
, &key_len
) != 2)
795 body
= input
+ key_len
;
798 body
= skip_spaces(body
);
800 disk
= get_gendisk(MKDEV(major
, minor
), &part
);
809 spin_lock_irq(disk
->queue
->queue_lock
);
811 if (blkcg_policy_enabled(disk
->queue
, pol
))
812 blkg
= blkg_lookup_create(blkcg
, disk
->queue
);
814 blkg
= ERR_PTR(-EOPNOTSUPP
);
819 spin_unlock_irq(disk
->queue
->queue_lock
);
822 * If queue was bypassing, we should retry. Do so after a
823 * short msleep(). It isn't strictly necessary but queue
824 * can be bypassing for some time and it's always nice to
825 * avoid busy looping.
829 ret
= restart_syscall();
839 EXPORT_SYMBOL_GPL(blkg_conf_prep
);
842 * blkg_conf_finish - finish up per-blkg config update
843 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
845 * Finish up after per-blkg config update. This function must be paired
846 * with blkg_conf_prep().
848 void blkg_conf_finish(struct blkg_conf_ctx
*ctx
)
849 __releases(ctx
->disk
->queue
->queue_lock
) __releases(rcu
)
851 spin_unlock_irq(ctx
->disk
->queue
->queue_lock
);
855 EXPORT_SYMBOL_GPL(blkg_conf_finish
);
857 static int blkcg_print_stat(struct seq_file
*sf
, void *v
)
859 struct blkcg
*blkcg
= css_to_blkcg(seq_css(sf
));
860 struct blkcg_gq
*blkg
;
864 hlist_for_each_entry_rcu(blkg
, &blkcg
->blkg_list
, blkcg_node
) {
866 struct blkg_rwstat rwstat
;
867 u64 rbytes
, wbytes
, rios
, wios
;
869 dname
= blkg_dev_name(blkg
);
873 spin_lock_irq(blkg
->q
->queue_lock
);
875 rwstat
= blkg_rwstat_recursive_sum(blkg
, NULL
,
876 offsetof(struct blkcg_gq
, stat_bytes
));
877 rbytes
= atomic64_read(&rwstat
.aux_cnt
[BLKG_RWSTAT_READ
]);
878 wbytes
= atomic64_read(&rwstat
.aux_cnt
[BLKG_RWSTAT_WRITE
]);
880 rwstat
= blkg_rwstat_recursive_sum(blkg
, NULL
,
881 offsetof(struct blkcg_gq
, stat_ios
));
882 rios
= atomic64_read(&rwstat
.aux_cnt
[BLKG_RWSTAT_READ
]);
883 wios
= atomic64_read(&rwstat
.aux_cnt
[BLKG_RWSTAT_WRITE
]);
885 spin_unlock_irq(blkg
->q
->queue_lock
);
887 if (rbytes
|| wbytes
|| rios
|| wios
)
888 seq_printf(sf
, "%s rbytes=%llu wbytes=%llu rios=%llu wios=%llu\n",
889 dname
, rbytes
, wbytes
, rios
, wios
);
896 struct cftype blkcg_files
[] = {
899 .seq_show
= blkcg_print_stat
,
904 struct cftype blkcg_legacy_files
[] = {
906 .name
= "reset_stats",
907 .write_u64
= blkcg_reset_stats
,
913 * blkcg_css_offline - cgroup css_offline callback
914 * @css: css of interest
916 * This function is called when @css is about to go away and responsible
917 * for shooting down all blkgs associated with @css. blkgs should be
918 * removed while holding both q and blkcg locks. As blkcg lock is nested
919 * inside q lock, this function performs reverse double lock dancing.
921 * This is the blkcg counterpart of ioc_release_fn().
923 static void blkcg_css_offline(struct cgroup_subsys_state
*css
)
925 struct blkcg
*blkcg
= css_to_blkcg(css
);
927 spin_lock_irq(&blkcg
->lock
);
929 while (!hlist_empty(&blkcg
->blkg_list
)) {
930 struct blkcg_gq
*blkg
= hlist_entry(blkcg
->blkg_list
.first
,
931 struct blkcg_gq
, blkcg_node
);
932 struct request_queue
*q
= blkg
->q
;
934 if (spin_trylock(q
->queue_lock
)) {
936 spin_unlock(q
->queue_lock
);
938 spin_unlock_irq(&blkcg
->lock
);
940 spin_lock_irq(&blkcg
->lock
);
944 spin_unlock_irq(&blkcg
->lock
);
946 wb_blkcg_offline(blkcg
);
949 static void blkcg_css_free(struct cgroup_subsys_state
*css
)
951 struct blkcg
*blkcg
= css_to_blkcg(css
);
954 mutex_lock(&blkcg_pol_mutex
);
956 list_del(&blkcg
->all_blkcgs_node
);
958 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++)
960 blkcg_policy
[i
]->cpd_free_fn(blkcg
->cpd
[i
]);
962 mutex_unlock(&blkcg_pol_mutex
);
967 static struct cgroup_subsys_state
*
968 blkcg_css_alloc(struct cgroup_subsys_state
*parent_css
)
971 struct cgroup_subsys_state
*ret
;
974 mutex_lock(&blkcg_pol_mutex
);
979 blkcg
= kzalloc(sizeof(*blkcg
), GFP_KERNEL
);
981 ret
= ERR_PTR(-ENOMEM
);
986 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
987 struct blkcg_policy
*pol
= blkcg_policy
[i
];
988 struct blkcg_policy_data
*cpd
;
991 * If the policy hasn't been attached yet, wait for it
992 * to be attached before doing anything else. Otherwise,
993 * check if the policy requires any specific per-cgroup
994 * data: if it does, allocate and initialize it.
996 if (!pol
|| !pol
->cpd_alloc_fn
)
999 cpd
= pol
->cpd_alloc_fn(GFP_KERNEL
);
1001 ret
= ERR_PTR(-ENOMEM
);
1004 blkcg
->cpd
[i
] = cpd
;
1007 if (pol
->cpd_init_fn
)
1008 pol
->cpd_init_fn(cpd
);
1011 spin_lock_init(&blkcg
->lock
);
1012 INIT_RADIX_TREE(&blkcg
->blkg_tree
, GFP_NOWAIT
);
1013 INIT_HLIST_HEAD(&blkcg
->blkg_list
);
1014 #ifdef CONFIG_CGROUP_WRITEBACK
1015 INIT_LIST_HEAD(&blkcg
->cgwb_list
);
1017 list_add_tail(&blkcg
->all_blkcgs_node
, &all_blkcgs
);
1019 mutex_unlock(&blkcg_pol_mutex
);
1023 for (i
--; i
>= 0; i
--)
1025 blkcg_policy
[i
]->cpd_free_fn(blkcg
->cpd
[i
]);
1028 mutex_unlock(&blkcg_pol_mutex
);
1033 * blkcg_init_queue - initialize blkcg part of request queue
1034 * @q: request_queue to initialize
1036 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
1037 * part of new request_queue @q.
1040 * 0 on success, -errno on failure.
1042 int blkcg_init_queue(struct request_queue
*q
)
1044 struct blkcg_gq
*new_blkg
, *blkg
;
1048 new_blkg
= blkg_alloc(&blkcg_root
, q
, GFP_KERNEL
);
1052 preloaded
= !radix_tree_preload(GFP_KERNEL
);
1055 * Make sure the root blkg exists and count the existing blkgs. As
1056 * @q is bypassing at this point, blkg_lookup_create() can't be
1057 * used. Open code insertion.
1060 spin_lock_irq(q
->queue_lock
);
1061 blkg
= blkg_create(&blkcg_root
, q
, new_blkg
);
1062 spin_unlock_irq(q
->queue_lock
);
1066 radix_tree_preload_end();
1069 blkg_free(new_blkg
);
1070 return PTR_ERR(blkg
);
1073 q
->root_blkg
= blkg
;
1074 q
->root_rl
.blkg
= blkg
;
1076 ret
= blk_throtl_init(q
);
1078 spin_lock_irq(q
->queue_lock
);
1079 blkg_destroy_all(q
);
1080 spin_unlock_irq(q
->queue_lock
);
1086 * blkcg_drain_queue - drain blkcg part of request_queue
1087 * @q: request_queue to drain
1089 * Called from blk_drain_queue(). Responsible for draining blkcg part.
1091 void blkcg_drain_queue(struct request_queue
*q
)
1093 lockdep_assert_held(q
->queue_lock
);
1096 * @q could be exiting and already have destroyed all blkgs as
1097 * indicated by NULL root_blkg. If so, don't confuse policies.
1102 blk_throtl_drain(q
);
1106 * blkcg_exit_queue - exit and release blkcg part of request_queue
1107 * @q: request_queue being released
1109 * Called from blk_release_queue(). Responsible for exiting blkcg part.
1111 void blkcg_exit_queue(struct request_queue
*q
)
1113 spin_lock_irq(q
->queue_lock
);
1114 blkg_destroy_all(q
);
1115 spin_unlock_irq(q
->queue_lock
);
1121 * We cannot support shared io contexts, as we have no mean to support
1122 * two tasks with the same ioc in two different groups without major rework
1123 * of the main cic data structures. For now we allow a task to change
1124 * its cgroup only if it's the only owner of its ioc.
1126 static int blkcg_can_attach(struct cgroup_subsys_state
*css
,
1127 struct cgroup_taskset
*tset
)
1129 struct task_struct
*task
;
1130 struct io_context
*ioc
;
1133 /* task_lock() is needed to avoid races with exit_io_context() */
1134 cgroup_taskset_for_each(task
, tset
) {
1136 ioc
= task
->io_context
;
1137 if (ioc
&& atomic_read(&ioc
->nr_tasks
) > 1)
1146 static void blkcg_bind(struct cgroup_subsys_state
*root_css
)
1150 mutex_lock(&blkcg_pol_mutex
);
1152 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
1153 struct blkcg_policy
*pol
= blkcg_policy
[i
];
1154 struct blkcg
*blkcg
;
1156 if (!pol
|| !pol
->cpd_bind_fn
)
1159 list_for_each_entry(blkcg
, &all_blkcgs
, all_blkcgs_node
)
1160 if (blkcg
->cpd
[pol
->plid
])
1161 pol
->cpd_bind_fn(blkcg
->cpd
[pol
->plid
]);
1163 mutex_unlock(&blkcg_pol_mutex
);
1166 struct cgroup_subsys io_cgrp_subsys
= {
1167 .css_alloc
= blkcg_css_alloc
,
1168 .css_offline
= blkcg_css_offline
,
1169 .css_free
= blkcg_css_free
,
1170 .can_attach
= blkcg_can_attach
,
1172 .dfl_cftypes
= blkcg_files
,
1173 .legacy_cftypes
= blkcg_legacy_files
,
1174 .legacy_name
= "blkio",
1177 * This ensures that, if available, memcg is automatically enabled
1178 * together on the default hierarchy so that the owner cgroup can
1179 * be retrieved from writeback pages.
1181 .depends_on
= 1 << memory_cgrp_id
,
1184 EXPORT_SYMBOL_GPL(io_cgrp_subsys
);
1187 * blkcg_activate_policy - activate a blkcg policy on a request_queue
1188 * @q: request_queue of interest
1189 * @pol: blkcg policy to activate
1191 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
1192 * bypass mode to populate its blkgs with policy_data for @pol.
1194 * Activation happens with @q bypassed, so nobody would be accessing blkgs
1195 * from IO path. Update of each blkg is protected by both queue and blkcg
1196 * locks so that holding either lock and testing blkcg_policy_enabled() is
1197 * always enough for dereferencing policy data.
1199 * The caller is responsible for synchronizing [de]activations and policy
1200 * [un]registerations. Returns 0 on success, -errno on failure.
1202 int blkcg_activate_policy(struct request_queue
*q
,
1203 const struct blkcg_policy
*pol
)
1205 struct blkg_policy_data
*pd_prealloc
= NULL
;
1206 struct blkcg_gq
*blkg
;
1209 if (blkcg_policy_enabled(q
, pol
))
1212 blk_queue_bypass_start(q
);
1215 pd_prealloc
= pol
->pd_alloc_fn(GFP_KERNEL
, q
->node
);
1218 goto out_bypass_end
;
1222 spin_lock_irq(q
->queue_lock
);
1224 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
) {
1225 struct blkg_policy_data
*pd
;
1227 if (blkg
->pd
[pol
->plid
])
1230 pd
= pol
->pd_alloc_fn(GFP_NOWAIT
, q
->node
);
1232 swap(pd
, pd_prealloc
);
1234 spin_unlock_irq(q
->queue_lock
);
1238 blkg
->pd
[pol
->plid
] = pd
;
1240 pd
->plid
= pol
->plid
;
1241 if (pol
->pd_init_fn
)
1242 pol
->pd_init_fn(pd
);
1245 __set_bit(pol
->plid
, q
->blkcg_pols
);
1248 spin_unlock_irq(q
->queue_lock
);
1250 blk_queue_bypass_end(q
);
1252 pol
->pd_free_fn(pd_prealloc
);
1255 EXPORT_SYMBOL_GPL(blkcg_activate_policy
);
1258 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
1259 * @q: request_queue of interest
1260 * @pol: blkcg policy to deactivate
1262 * Deactivate @pol on @q. Follows the same synchronization rules as
1263 * blkcg_activate_policy().
1265 void blkcg_deactivate_policy(struct request_queue
*q
,
1266 const struct blkcg_policy
*pol
)
1268 struct blkcg_gq
*blkg
;
1270 if (!blkcg_policy_enabled(q
, pol
))
1273 blk_queue_bypass_start(q
);
1274 spin_lock_irq(q
->queue_lock
);
1276 __clear_bit(pol
->plid
, q
->blkcg_pols
);
1278 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
) {
1279 /* grab blkcg lock too while removing @pd from @blkg */
1280 spin_lock(&blkg
->blkcg
->lock
);
1282 if (blkg
->pd
[pol
->plid
]) {
1283 if (pol
->pd_offline_fn
)
1284 pol
->pd_offline_fn(blkg
->pd
[pol
->plid
]);
1285 pol
->pd_free_fn(blkg
->pd
[pol
->plid
]);
1286 blkg
->pd
[pol
->plid
] = NULL
;
1289 spin_unlock(&blkg
->blkcg
->lock
);
1292 spin_unlock_irq(q
->queue_lock
);
1293 blk_queue_bypass_end(q
);
1295 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy
);
1298 * blkcg_policy_register - register a blkcg policy
1299 * @pol: blkcg policy to register
1301 * Register @pol with blkcg core. Might sleep and @pol may be modified on
1302 * successful registration. Returns 0 on success and -errno on failure.
1304 int blkcg_policy_register(struct blkcg_policy
*pol
)
1306 struct blkcg
*blkcg
;
1309 mutex_lock(&blkcg_pol_register_mutex
);
1310 mutex_lock(&blkcg_pol_mutex
);
1312 /* find an empty slot */
1314 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++)
1315 if (!blkcg_policy
[i
])
1317 if (i
>= BLKCG_MAX_POLS
)
1322 blkcg_policy
[pol
->plid
] = pol
;
1324 /* allocate and install cpd's */
1325 if (pol
->cpd_alloc_fn
) {
1326 list_for_each_entry(blkcg
, &all_blkcgs
, all_blkcgs_node
) {
1327 struct blkcg_policy_data
*cpd
;
1329 cpd
= pol
->cpd_alloc_fn(GFP_KERNEL
);
1331 mutex_unlock(&blkcg_pol_mutex
);
1335 blkcg
->cpd
[pol
->plid
] = cpd
;
1337 cpd
->plid
= pol
->plid
;
1338 pol
->cpd_init_fn(cpd
);
1342 mutex_unlock(&blkcg_pol_mutex
);
1344 /* everything is in place, add intf files for the new policy */
1345 if (pol
->dfl_cftypes
)
1346 WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys
,
1348 if (pol
->legacy_cftypes
)
1349 WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys
,
1350 pol
->legacy_cftypes
));
1351 mutex_unlock(&blkcg_pol_register_mutex
);
1355 if (pol
->cpd_alloc_fn
) {
1356 list_for_each_entry(blkcg
, &all_blkcgs
, all_blkcgs_node
) {
1357 if (blkcg
->cpd
[pol
->plid
]) {
1358 pol
->cpd_free_fn(blkcg
->cpd
[pol
->plid
]);
1359 blkcg
->cpd
[pol
->plid
] = NULL
;
1363 blkcg_policy
[pol
->plid
] = NULL
;
1365 mutex_unlock(&blkcg_pol_mutex
);
1366 mutex_unlock(&blkcg_pol_register_mutex
);
1369 EXPORT_SYMBOL_GPL(blkcg_policy_register
);
1372 * blkcg_policy_unregister - unregister a blkcg policy
1373 * @pol: blkcg policy to unregister
1375 * Undo blkcg_policy_register(@pol). Might sleep.
1377 void blkcg_policy_unregister(struct blkcg_policy
*pol
)
1379 struct blkcg
*blkcg
;
1381 mutex_lock(&blkcg_pol_register_mutex
);
1383 if (WARN_ON(blkcg_policy
[pol
->plid
] != pol
))
1386 /* kill the intf files first */
1387 if (pol
->dfl_cftypes
)
1388 cgroup_rm_cftypes(pol
->dfl_cftypes
);
1389 if (pol
->legacy_cftypes
)
1390 cgroup_rm_cftypes(pol
->legacy_cftypes
);
1392 /* remove cpds and unregister */
1393 mutex_lock(&blkcg_pol_mutex
);
1395 if (pol
->cpd_alloc_fn
) {
1396 list_for_each_entry(blkcg
, &all_blkcgs
, all_blkcgs_node
) {
1397 if (blkcg
->cpd
[pol
->plid
]) {
1398 pol
->cpd_free_fn(blkcg
->cpd
[pol
->plid
]);
1399 blkcg
->cpd
[pol
->plid
] = NULL
;
1403 blkcg_policy
[pol
->plid
] = NULL
;
1405 mutex_unlock(&blkcg_pol_mutex
);
1407 mutex_unlock(&blkcg_pol_register_mutex
);
1409 EXPORT_SYMBOL_GPL(blkcg_policy_unregister
);