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/blk-cgroup.h>
30 #define MAX_KEY_LEN 100
33 * blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation.
34 * blkcg_pol_register_mutex nests outside of it and synchronizes entire
35 * policy [un]register operations including cgroup file additions /
36 * removals. Putting cgroup file registration outside blkcg_pol_mutex
37 * allows grabbing it from cgroup callbacks.
39 static DEFINE_MUTEX(blkcg_pol_register_mutex
);
40 static DEFINE_MUTEX(blkcg_pol_mutex
);
42 struct blkcg blkcg_root
;
43 EXPORT_SYMBOL_GPL(blkcg_root
);
45 struct cgroup_subsys_state
* const blkcg_root_css
= &blkcg_root
.css
;
47 static struct blkcg_policy
*blkcg_policy
[BLKCG_MAX_POLS
];
49 static LIST_HEAD(all_blkcgs
); /* protected by blkcg_pol_mutex */
51 static bool blkcg_policy_enabled(struct request_queue
*q
,
52 const struct blkcg_policy
*pol
)
54 return pol
&& test_bit(pol
->plid
, q
->blkcg_pols
);
58 * blkg_free - free a blkg
61 * Free @blkg which may be partially allocated.
63 static void blkg_free(struct blkcg_gq
*blkg
)
70 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++)
72 blkcg_policy
[i
]->pd_free_fn(blkg
->pd
[i
]);
74 if (blkg
->blkcg
!= &blkcg_root
)
75 blk_exit_rl(&blkg
->rl
);
80 * blkg_alloc - allocate a blkg
81 * @blkcg: block cgroup the new blkg is associated with
82 * @q: request_queue the new blkg is associated with
83 * @gfp_mask: allocation mask to use
85 * Allocate a new blkg assocating @blkcg and @q.
87 static struct blkcg_gq
*blkg_alloc(struct blkcg
*blkcg
, struct request_queue
*q
,
90 struct blkcg_gq
*blkg
;
93 /* alloc and init base part */
94 blkg
= kzalloc_node(sizeof(*blkg
), gfp_mask
, q
->node
);
99 INIT_LIST_HEAD(&blkg
->q_node
);
101 atomic_set(&blkg
->refcnt
, 1);
103 /* root blkg uses @q->root_rl, init rl only for !root blkgs */
104 if (blkcg
!= &blkcg_root
) {
105 if (blk_init_rl(&blkg
->rl
, q
, gfp_mask
))
107 blkg
->rl
.blkg
= blkg
;
110 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
111 struct blkcg_policy
*pol
= blkcg_policy
[i
];
112 struct blkg_policy_data
*pd
;
114 if (!blkcg_policy_enabled(q
, pol
))
117 /* alloc per-policy data and attach it to blkg */
118 pd
= pol
->pd_alloc_fn(gfp_mask
, q
->node
);
135 * __blkg_lookup - internal version of blkg_lookup()
136 * @blkcg: blkcg of interest
137 * @q: request_queue of interest
138 * @update_hint: whether to update lookup hint with the result or not
140 * This is internal version and shouldn't be used by policy
141 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
142 * @q's bypass state. If @update_hint is %true, the caller should be
143 * holding @q->queue_lock and lookup hint is updated on success.
145 struct blkcg_gq
*__blkg_lookup(struct blkcg
*blkcg
, struct request_queue
*q
,
148 struct blkcg_gq
*blkg
;
150 blkg
= rcu_dereference(blkcg
->blkg_hint
);
151 if (blkg
&& blkg
->q
== q
)
155 * Hint didn't match. Look up from the radix tree. Note that the
156 * hint can only be updated under queue_lock as otherwise @blkg
157 * could have already been removed from blkg_tree. The caller is
158 * responsible for grabbing queue_lock if @update_hint.
160 blkg
= radix_tree_lookup(&blkcg
->blkg_tree
, q
->id
);
161 if (blkg
&& blkg
->q
== q
) {
163 lockdep_assert_held(q
->queue_lock
);
164 rcu_assign_pointer(blkcg
->blkg_hint
, blkg
);
173 * blkg_lookup - lookup blkg for the specified blkcg - q pair
174 * @blkcg: blkcg of interest
175 * @q: request_queue of interest
177 * Lookup blkg for the @blkcg - @q pair. This function should be called
178 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
179 * - see blk_queue_bypass_start() for details.
181 struct blkcg_gq
*blkg_lookup(struct blkcg
*blkcg
, struct request_queue
*q
)
183 WARN_ON_ONCE(!rcu_read_lock_held());
185 if (unlikely(blk_queue_bypass(q
)))
187 return __blkg_lookup(blkcg
, q
, false);
189 EXPORT_SYMBOL_GPL(blkg_lookup
);
192 * If @new_blkg is %NULL, this function tries to allocate a new one as
193 * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return.
195 static struct blkcg_gq
*blkg_create(struct blkcg
*blkcg
,
196 struct request_queue
*q
,
197 struct blkcg_gq
*new_blkg
)
199 struct blkcg_gq
*blkg
;
200 struct bdi_writeback_congested
*wb_congested
;
203 WARN_ON_ONCE(!rcu_read_lock_held());
204 lockdep_assert_held(q
->queue_lock
);
206 /* blkg holds a reference to blkcg */
207 if (!css_tryget_online(&blkcg
->css
)) {
212 wb_congested
= wb_congested_get_create(&q
->backing_dev_info
,
213 blkcg
->css
.id
, GFP_NOWAIT
);
221 new_blkg
= blkg_alloc(blkcg
, q
, GFP_NOWAIT
);
222 if (unlikely(!new_blkg
)) {
224 goto err_put_congested
;
228 blkg
->wb_congested
= wb_congested
;
231 if (blkcg_parent(blkcg
)) {
232 blkg
->parent
= __blkg_lookup(blkcg_parent(blkcg
), q
, false);
233 if (WARN_ON_ONCE(!blkg
->parent
)) {
235 goto err_put_congested
;
237 blkg_get(blkg
->parent
);
240 /* invoke per-policy init */
241 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
242 struct blkcg_policy
*pol
= blkcg_policy
[i
];
244 if (blkg
->pd
[i
] && pol
->pd_init_fn
)
245 pol
->pd_init_fn(blkg
->pd
[i
]);
249 spin_lock(&blkcg
->lock
);
250 ret
= radix_tree_insert(&blkcg
->blkg_tree
, q
->id
, blkg
);
252 hlist_add_head_rcu(&blkg
->blkcg_node
, &blkcg
->blkg_list
);
253 list_add(&blkg
->q_node
, &q
->blkg_list
);
255 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
256 struct blkcg_policy
*pol
= blkcg_policy
[i
];
258 if (blkg
->pd
[i
] && pol
->pd_online_fn
)
259 pol
->pd_online_fn(blkg
->pd
[i
]);
263 spin_unlock(&blkcg
->lock
);
268 /* @blkg failed fully initialized, use the usual release path */
273 wb_congested_put(wb_congested
);
275 css_put(&blkcg
->css
);
282 * blkg_lookup_create - lookup blkg, try to create one if not there
283 * @blkcg: blkcg of interest
284 * @q: request_queue of interest
286 * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
287 * create one. blkg creation is performed recursively from blkcg_root such
288 * that all non-root blkg's have access to the parent blkg. This function
289 * should be called under RCU read lock and @q->queue_lock.
291 * Returns pointer to the looked up or created blkg on success, ERR_PTR()
292 * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not
293 * dead and bypassing, returns ERR_PTR(-EBUSY).
295 struct blkcg_gq
*blkg_lookup_create(struct blkcg
*blkcg
,
296 struct request_queue
*q
)
298 struct blkcg_gq
*blkg
;
300 WARN_ON_ONCE(!rcu_read_lock_held());
301 lockdep_assert_held(q
->queue_lock
);
304 * This could be the first entry point of blkcg implementation and
305 * we shouldn't allow anything to go through for a bypassing queue.
307 if (unlikely(blk_queue_bypass(q
)))
308 return ERR_PTR(blk_queue_dying(q
) ? -EINVAL
: -EBUSY
);
310 blkg
= __blkg_lookup(blkcg
, q
, true);
315 * Create blkgs walking down from blkcg_root to @blkcg, so that all
316 * non-root blkgs have access to their parents.
319 struct blkcg
*pos
= blkcg
;
320 struct blkcg
*parent
= blkcg_parent(blkcg
);
322 while (parent
&& !__blkg_lookup(parent
, q
, false)) {
324 parent
= blkcg_parent(parent
);
327 blkg
= blkg_create(pos
, q
, NULL
);
328 if (pos
== blkcg
|| IS_ERR(blkg
))
332 EXPORT_SYMBOL_GPL(blkg_lookup_create
);
334 static void blkg_destroy(struct blkcg_gq
*blkg
)
336 struct blkcg
*blkcg
= blkg
->blkcg
;
339 lockdep_assert_held(blkg
->q
->queue_lock
);
340 lockdep_assert_held(&blkcg
->lock
);
342 /* Something wrong if we are trying to remove same group twice */
343 WARN_ON_ONCE(list_empty(&blkg
->q_node
));
344 WARN_ON_ONCE(hlist_unhashed(&blkg
->blkcg_node
));
346 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
347 struct blkcg_policy
*pol
= blkcg_policy
[i
];
349 if (blkg
->pd
[i
] && pol
->pd_offline_fn
)
350 pol
->pd_offline_fn(blkg
->pd
[i
]);
352 blkg
->online
= false;
354 radix_tree_delete(&blkcg
->blkg_tree
, blkg
->q
->id
);
355 list_del_init(&blkg
->q_node
);
356 hlist_del_init_rcu(&blkg
->blkcg_node
);
359 * Both setting lookup hint to and clearing it from @blkg are done
360 * under queue_lock. If it's not pointing to @blkg now, it never
361 * will. Hint assignment itself can race safely.
363 if (rcu_access_pointer(blkcg
->blkg_hint
) == blkg
)
364 rcu_assign_pointer(blkcg
->blkg_hint
, NULL
);
367 * Put the reference taken at the time of creation so that when all
368 * queues are gone, group can be destroyed.
374 * blkg_destroy_all - destroy all blkgs associated with a request_queue
375 * @q: request_queue of interest
377 * Destroy all blkgs associated with @q.
379 static void blkg_destroy_all(struct request_queue
*q
)
381 struct blkcg_gq
*blkg
, *n
;
383 lockdep_assert_held(q
->queue_lock
);
385 list_for_each_entry_safe(blkg
, n
, &q
->blkg_list
, q_node
) {
386 struct blkcg
*blkcg
= blkg
->blkcg
;
388 spin_lock(&blkcg
->lock
);
390 spin_unlock(&blkcg
->lock
);
395 * A group is RCU protected, but having an rcu lock does not mean that one
396 * can access all the fields of blkg and assume these are valid. For
397 * example, don't try to follow throtl_data and request queue links.
399 * Having a reference to blkg under an rcu allows accesses to only values
400 * local to groups like group stats and group rate limits.
402 void __blkg_release_rcu(struct rcu_head
*rcu_head
)
404 struct blkcg_gq
*blkg
= container_of(rcu_head
, struct blkcg_gq
, rcu_head
);
406 /* release the blkcg and parent blkg refs this blkg has been holding */
407 css_put(&blkg
->blkcg
->css
);
409 blkg_put(blkg
->parent
);
411 wb_congested_put(blkg
->wb_congested
);
415 EXPORT_SYMBOL_GPL(__blkg_release_rcu
);
418 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
419 * because the root blkg uses @q->root_rl instead of its own rl.
421 struct request_list
*__blk_queue_next_rl(struct request_list
*rl
,
422 struct request_queue
*q
)
424 struct list_head
*ent
;
425 struct blkcg_gq
*blkg
;
428 * Determine the current blkg list_head. The first entry is
429 * root_rl which is off @q->blkg_list and mapped to the head.
431 if (rl
== &q
->root_rl
) {
433 /* There are no more block groups, hence no request lists */
437 blkg
= container_of(rl
, struct blkcg_gq
, rl
);
441 /* walk to the next list_head, skip root blkcg */
443 if (ent
== &q
->root_blkg
->q_node
)
445 if (ent
== &q
->blkg_list
)
448 blkg
= container_of(ent
, struct blkcg_gq
, q_node
);
452 static int blkcg_reset_stats(struct cgroup_subsys_state
*css
,
453 struct cftype
*cftype
, u64 val
)
455 struct blkcg
*blkcg
= css_to_blkcg(css
);
456 struct blkcg_gq
*blkg
;
459 mutex_lock(&blkcg_pol_mutex
);
460 spin_lock_irq(&blkcg
->lock
);
463 * Note that stat reset is racy - it doesn't synchronize against
464 * stat updates. This is a debug feature which shouldn't exist
465 * anyway. If you get hit by a race, retry.
467 hlist_for_each_entry(blkg
, &blkcg
->blkg_list
, blkcg_node
) {
468 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
469 struct blkcg_policy
*pol
= blkcg_policy
[i
];
471 if (blkg
->pd
[i
] && pol
->pd_reset_stats_fn
)
472 pol
->pd_reset_stats_fn(blkg
->pd
[i
]);
476 spin_unlock_irq(&blkcg
->lock
);
477 mutex_unlock(&blkcg_pol_mutex
);
481 static const char *blkg_dev_name(struct blkcg_gq
*blkg
)
483 /* some drivers (floppy) instantiate a queue w/o disk registered */
484 if (blkg
->q
->backing_dev_info
.dev
)
485 return dev_name(blkg
->q
->backing_dev_info
.dev
);
490 * blkcg_print_blkgs - helper for printing per-blkg data
491 * @sf: seq_file to print to
492 * @blkcg: blkcg of interest
493 * @prfill: fill function to print out a blkg
494 * @pol: policy in question
495 * @data: data to be passed to @prfill
496 * @show_total: to print out sum of prfill return values or not
498 * This function invokes @prfill on each blkg of @blkcg if pd for the
499 * policy specified by @pol exists. @prfill is invoked with @sf, the
500 * policy data and @data and the matching queue lock held. If @show_total
501 * is %true, the sum of the return values from @prfill is printed with
502 * "Total" label at the end.
504 * This is to be used to construct print functions for
505 * cftype->read_seq_string method.
507 void blkcg_print_blkgs(struct seq_file
*sf
, struct blkcg
*blkcg
,
508 u64 (*prfill
)(struct seq_file
*,
509 struct blkg_policy_data
*, int),
510 const struct blkcg_policy
*pol
, int data
,
513 struct blkcg_gq
*blkg
;
517 hlist_for_each_entry_rcu(blkg
, &blkcg
->blkg_list
, blkcg_node
) {
518 spin_lock_irq(blkg
->q
->queue_lock
);
519 if (blkcg_policy_enabled(blkg
->q
, pol
))
520 total
+= prfill(sf
, blkg
->pd
[pol
->plid
], data
);
521 spin_unlock_irq(blkg
->q
->queue_lock
);
526 seq_printf(sf
, "Total %llu\n", (unsigned long long)total
);
528 EXPORT_SYMBOL_GPL(blkcg_print_blkgs
);
531 * __blkg_prfill_u64 - prfill helper for a single u64 value
532 * @sf: seq_file to print to
533 * @pd: policy private data of interest
536 * Print @v to @sf for the device assocaited with @pd.
538 u64
__blkg_prfill_u64(struct seq_file
*sf
, struct blkg_policy_data
*pd
, u64 v
)
540 const char *dname
= blkg_dev_name(pd
->blkg
);
545 seq_printf(sf
, "%s %llu\n", dname
, (unsigned long long)v
);
548 EXPORT_SYMBOL_GPL(__blkg_prfill_u64
);
551 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
552 * @sf: seq_file to print to
553 * @pd: policy private data of interest
554 * @rwstat: rwstat to print
556 * Print @rwstat to @sf for the device assocaited with @pd.
558 u64
__blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
559 const struct blkg_rwstat
*rwstat
)
561 static const char *rwstr
[] = {
562 [BLKG_RWSTAT_READ
] = "Read",
563 [BLKG_RWSTAT_WRITE
] = "Write",
564 [BLKG_RWSTAT_SYNC
] = "Sync",
565 [BLKG_RWSTAT_ASYNC
] = "Async",
567 const char *dname
= blkg_dev_name(pd
->blkg
);
574 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
575 seq_printf(sf
, "%s %s %llu\n", dname
, rwstr
[i
],
576 (unsigned long long)rwstat
->cnt
[i
]);
578 v
= rwstat
->cnt
[BLKG_RWSTAT_READ
] + rwstat
->cnt
[BLKG_RWSTAT_WRITE
];
579 seq_printf(sf
, "%s Total %llu\n", dname
, (unsigned long long)v
);
582 EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat
);
585 * blkg_prfill_stat - prfill callback for blkg_stat
586 * @sf: seq_file to print to
587 * @pd: policy private data of interest
588 * @off: offset to the blkg_stat in @pd
590 * prfill callback for printing a blkg_stat.
592 u64
blkg_prfill_stat(struct seq_file
*sf
, struct blkg_policy_data
*pd
, int off
)
594 return __blkg_prfill_u64(sf
, pd
, blkg_stat_read((void *)pd
+ off
));
596 EXPORT_SYMBOL_GPL(blkg_prfill_stat
);
599 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
600 * @sf: seq_file to print to
601 * @pd: policy private data of interest
602 * @off: offset to the blkg_rwstat in @pd
604 * prfill callback for printing a blkg_rwstat.
606 u64
blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
609 struct blkg_rwstat rwstat
= blkg_rwstat_read((void *)pd
+ off
);
611 return __blkg_prfill_rwstat(sf
, pd
, &rwstat
);
613 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat
);
616 * blkg_stat_recursive_sum - collect hierarchical blkg_stat
617 * @pd: policy private data of interest
618 * @off: offset to the blkg_stat in @pd
620 * Collect the blkg_stat specified by @off from @pd and all its online
621 * descendants and return the sum. The caller must be holding the queue
622 * lock for online tests.
624 u64
blkg_stat_recursive_sum(struct blkg_policy_data
*pd
, int off
)
626 struct blkcg_policy
*pol
= blkcg_policy
[pd
->plid
];
627 struct blkcg_gq
*pos_blkg
;
628 struct cgroup_subsys_state
*pos_css
;
631 lockdep_assert_held(pd
->blkg
->q
->queue_lock
);
634 blkg_for_each_descendant_pre(pos_blkg
, pos_css
, pd_to_blkg(pd
)) {
635 struct blkg_policy_data
*pos_pd
= blkg_to_pd(pos_blkg
, pol
);
636 struct blkg_stat
*stat
= (void *)pos_pd
+ off
;
638 if (pos_blkg
->online
)
639 sum
+= blkg_stat_read(stat
);
645 EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum
);
648 * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat
649 * @pd: policy private data of interest
650 * @off: offset to the blkg_stat in @pd
652 * Collect the blkg_rwstat specified by @off from @pd and all its online
653 * descendants and return the sum. The caller must be holding the queue
654 * lock for online tests.
656 struct blkg_rwstat
blkg_rwstat_recursive_sum(struct blkg_policy_data
*pd
,
659 struct blkcg_policy
*pol
= blkcg_policy
[pd
->plid
];
660 struct blkcg_gq
*pos_blkg
;
661 struct cgroup_subsys_state
*pos_css
;
662 struct blkg_rwstat sum
= { };
665 lockdep_assert_held(pd
->blkg
->q
->queue_lock
);
668 blkg_for_each_descendant_pre(pos_blkg
, pos_css
, pd_to_blkg(pd
)) {
669 struct blkg_policy_data
*pos_pd
= blkg_to_pd(pos_blkg
, pol
);
670 struct blkg_rwstat
*rwstat
= (void *)pos_pd
+ off
;
671 struct blkg_rwstat tmp
;
673 if (!pos_blkg
->online
)
676 tmp
= blkg_rwstat_read(rwstat
);
678 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
679 sum
.cnt
[i
] += tmp
.cnt
[i
];
685 EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum
);
688 * blkg_conf_prep - parse and prepare for per-blkg config update
689 * @blkcg: target block cgroup
690 * @pol: target policy
691 * @input: input string
692 * @ctx: blkg_conf_ctx to be filled
694 * Parse per-blkg config update from @input and initialize @ctx with the
695 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
696 * value. This function returns with RCU read lock and queue lock held and
697 * must be paired with blkg_conf_finish().
699 int blkg_conf_prep(struct blkcg
*blkcg
, const struct blkcg_policy
*pol
,
700 const char *input
, struct blkg_conf_ctx
*ctx
)
701 __acquires(rcu
) __acquires(disk
->queue
->queue_lock
)
703 struct gendisk
*disk
;
704 struct blkcg_gq
*blkg
;
705 unsigned int major
, minor
;
706 unsigned long long v
;
709 if (sscanf(input
, "%u:%u %llu", &major
, &minor
, &v
) != 3)
712 disk
= get_gendisk(MKDEV(major
, minor
), &part
);
721 spin_lock_irq(disk
->queue
->queue_lock
);
723 if (blkcg_policy_enabled(disk
->queue
, pol
))
724 blkg
= blkg_lookup_create(blkcg
, disk
->queue
);
726 blkg
= ERR_PTR(-EINVAL
);
731 spin_unlock_irq(disk
->queue
->queue_lock
);
734 * If queue was bypassing, we should retry. Do so after a
735 * short msleep(). It isn't strictly necessary but queue
736 * can be bypassing for some time and it's always nice to
737 * avoid busy looping.
741 ret
= restart_syscall();
751 EXPORT_SYMBOL_GPL(blkg_conf_prep
);
754 * blkg_conf_finish - finish up per-blkg config update
755 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
757 * Finish up after per-blkg config update. This function must be paired
758 * with blkg_conf_prep().
760 void blkg_conf_finish(struct blkg_conf_ctx
*ctx
)
761 __releases(ctx
->disk
->queue
->queue_lock
) __releases(rcu
)
763 spin_unlock_irq(ctx
->disk
->queue
->queue_lock
);
767 EXPORT_SYMBOL_GPL(blkg_conf_finish
);
769 struct cftype blkcg_files
[] = {
771 .name
= "reset_stats",
772 .write_u64
= blkcg_reset_stats
,
778 * blkcg_css_offline - cgroup css_offline callback
779 * @css: css of interest
781 * This function is called when @css is about to go away and responsible
782 * for shooting down all blkgs associated with @css. blkgs should be
783 * removed while holding both q and blkcg locks. As blkcg lock is nested
784 * inside q lock, this function performs reverse double lock dancing.
786 * This is the blkcg counterpart of ioc_release_fn().
788 static void blkcg_css_offline(struct cgroup_subsys_state
*css
)
790 struct blkcg
*blkcg
= css_to_blkcg(css
);
792 spin_lock_irq(&blkcg
->lock
);
794 while (!hlist_empty(&blkcg
->blkg_list
)) {
795 struct blkcg_gq
*blkg
= hlist_entry(blkcg
->blkg_list
.first
,
796 struct blkcg_gq
, blkcg_node
);
797 struct request_queue
*q
= blkg
->q
;
799 if (spin_trylock(q
->queue_lock
)) {
801 spin_unlock(q
->queue_lock
);
803 spin_unlock_irq(&blkcg
->lock
);
805 spin_lock_irq(&blkcg
->lock
);
809 spin_unlock_irq(&blkcg
->lock
);
811 wb_blkcg_offline(blkcg
);
814 static void blkcg_css_free(struct cgroup_subsys_state
*css
)
816 struct blkcg
*blkcg
= css_to_blkcg(css
);
819 mutex_lock(&blkcg_pol_mutex
);
820 list_del(&blkcg
->all_blkcgs_node
);
821 mutex_unlock(&blkcg_pol_mutex
);
823 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++)
828 static struct cgroup_subsys_state
*
829 blkcg_css_alloc(struct cgroup_subsys_state
*parent_css
)
832 struct cgroup_subsys_state
*ret
;
835 mutex_lock(&blkcg_pol_mutex
);
840 blkcg
= kzalloc(sizeof(*blkcg
), GFP_KERNEL
);
842 ret
= ERR_PTR(-ENOMEM
);
847 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++) {
848 struct blkcg_policy
*pol
= blkcg_policy
[i
];
849 struct blkcg_policy_data
*cpd
;
852 * If the policy hasn't been attached yet, wait for it
853 * to be attached before doing anything else. Otherwise,
854 * check if the policy requires any specific per-cgroup
855 * data: if it does, allocate and initialize it.
857 if (!pol
|| !pol
->cpd_size
)
860 BUG_ON(blkcg
->pd
[i
]);
861 cpd
= kzalloc(pol
->cpd_size
, GFP_KERNEL
);
863 ret
= ERR_PTR(-ENOMEM
);
868 pol
->cpd_init_fn(blkcg
);
871 spin_lock_init(&blkcg
->lock
);
872 INIT_RADIX_TREE(&blkcg
->blkg_tree
, GFP_NOWAIT
);
873 INIT_HLIST_HEAD(&blkcg
->blkg_list
);
874 #ifdef CONFIG_CGROUP_WRITEBACK
875 INIT_LIST_HEAD(&blkcg
->cgwb_list
);
877 list_add_tail(&blkcg
->all_blkcgs_node
, &all_blkcgs
);
879 mutex_unlock(&blkcg_pol_mutex
);
883 for (i
--; i
>= 0; i
--)
887 mutex_unlock(&blkcg_pol_mutex
);
892 * blkcg_init_queue - initialize blkcg part of request queue
893 * @q: request_queue to initialize
895 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
896 * part of new request_queue @q.
899 * 0 on success, -errno on failure.
901 int blkcg_init_queue(struct request_queue
*q
)
903 struct blkcg_gq
*new_blkg
, *blkg
;
907 new_blkg
= blkg_alloc(&blkcg_root
, q
, GFP_KERNEL
);
911 preloaded
= !radix_tree_preload(GFP_KERNEL
);
914 * Make sure the root blkg exists and count the existing blkgs. As
915 * @q is bypassing at this point, blkg_lookup_create() can't be
916 * used. Open code insertion.
919 spin_lock_irq(q
->queue_lock
);
920 blkg
= blkg_create(&blkcg_root
, q
, new_blkg
);
921 spin_unlock_irq(q
->queue_lock
);
925 radix_tree_preload_end();
929 return PTR_ERR(blkg
);
933 q
->root_rl
.blkg
= blkg
;
935 ret
= blk_throtl_init(q
);
937 spin_lock_irq(q
->queue_lock
);
939 spin_unlock_irq(q
->queue_lock
);
945 * blkcg_drain_queue - drain blkcg part of request_queue
946 * @q: request_queue to drain
948 * Called from blk_drain_queue(). Responsible for draining blkcg part.
950 void blkcg_drain_queue(struct request_queue
*q
)
952 lockdep_assert_held(q
->queue_lock
);
955 * @q could be exiting and already have destroyed all blkgs as
956 * indicated by NULL root_blkg. If so, don't confuse policies.
965 * blkcg_exit_queue - exit and release blkcg part of request_queue
966 * @q: request_queue being released
968 * Called from blk_release_queue(). Responsible for exiting blkcg part.
970 void blkcg_exit_queue(struct request_queue
*q
)
972 spin_lock_irq(q
->queue_lock
);
974 spin_unlock_irq(q
->queue_lock
);
980 * We cannot support shared io contexts, as we have no mean to support
981 * two tasks with the same ioc in two different groups without major rework
982 * of the main cic data structures. For now we allow a task to change
983 * its cgroup only if it's the only owner of its ioc.
985 static int blkcg_can_attach(struct cgroup_subsys_state
*css
,
986 struct cgroup_taskset
*tset
)
988 struct task_struct
*task
;
989 struct io_context
*ioc
;
992 /* task_lock() is needed to avoid races with exit_io_context() */
993 cgroup_taskset_for_each(task
, tset
) {
995 ioc
= task
->io_context
;
996 if (ioc
&& atomic_read(&ioc
->nr_tasks
) > 1)
1005 struct cgroup_subsys blkio_cgrp_subsys
= {
1006 .css_alloc
= blkcg_css_alloc
,
1007 .css_offline
= blkcg_css_offline
,
1008 .css_free
= blkcg_css_free
,
1009 .can_attach
= blkcg_can_attach
,
1010 .legacy_cftypes
= blkcg_files
,
1013 * This ensures that, if available, memcg is automatically enabled
1014 * together on the default hierarchy so that the owner cgroup can
1015 * be retrieved from writeback pages.
1017 .depends_on
= 1 << memory_cgrp_id
,
1020 EXPORT_SYMBOL_GPL(blkio_cgrp_subsys
);
1023 * blkcg_activate_policy - activate a blkcg policy on a request_queue
1024 * @q: request_queue of interest
1025 * @pol: blkcg policy to activate
1027 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
1028 * bypass mode to populate its blkgs with policy_data for @pol.
1030 * Activation happens with @q bypassed, so nobody would be accessing blkgs
1031 * from IO path. Update of each blkg is protected by both queue and blkcg
1032 * locks so that holding either lock and testing blkcg_policy_enabled() is
1033 * always enough for dereferencing policy data.
1035 * The caller is responsible for synchronizing [de]activations and policy
1036 * [un]registerations. Returns 0 on success, -errno on failure.
1038 int blkcg_activate_policy(struct request_queue
*q
,
1039 const struct blkcg_policy
*pol
)
1041 struct blkg_policy_data
*pd_prealloc
= NULL
;
1042 struct blkcg_gq
*blkg
;
1045 if (blkcg_policy_enabled(q
, pol
))
1048 blk_queue_bypass_start(q
);
1051 pd_prealloc
= pol
->pd_alloc_fn(GFP_KERNEL
, q
->node
);
1054 goto out_bypass_end
;
1058 spin_lock_irq(q
->queue_lock
);
1060 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
) {
1061 struct blkg_policy_data
*pd
;
1063 if (blkg
->pd
[pol
->plid
])
1066 pd
= pol
->pd_alloc_fn(GFP_NOWAIT
, q
->node
);
1068 swap(pd
, pd_prealloc
);
1070 spin_unlock_irq(q
->queue_lock
);
1074 blkg
->pd
[pol
->plid
] = pd
;
1076 pd
->plid
= pol
->plid
;
1077 if (pol
->pd_init_fn
)
1078 pol
->pd_init_fn(pd
);
1081 __set_bit(pol
->plid
, q
->blkcg_pols
);
1084 spin_unlock_irq(q
->queue_lock
);
1086 blk_queue_bypass_end(q
);
1088 pol
->pd_free_fn(pd_prealloc
);
1091 EXPORT_SYMBOL_GPL(blkcg_activate_policy
);
1094 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
1095 * @q: request_queue of interest
1096 * @pol: blkcg policy to deactivate
1098 * Deactivate @pol on @q. Follows the same synchronization rules as
1099 * blkcg_activate_policy().
1101 void blkcg_deactivate_policy(struct request_queue
*q
,
1102 const struct blkcg_policy
*pol
)
1104 struct blkcg_gq
*blkg
;
1106 if (!blkcg_policy_enabled(q
, pol
))
1109 blk_queue_bypass_start(q
);
1110 spin_lock_irq(q
->queue_lock
);
1112 __clear_bit(pol
->plid
, q
->blkcg_pols
);
1114 list_for_each_entry(blkg
, &q
->blkg_list
, q_node
) {
1115 /* grab blkcg lock too while removing @pd from @blkg */
1116 spin_lock(&blkg
->blkcg
->lock
);
1118 if (blkg
->pd
[pol
->plid
]) {
1119 if (pol
->pd_offline_fn
)
1120 pol
->pd_offline_fn(blkg
->pd
[pol
->plid
]);
1121 pol
->pd_free_fn(blkg
->pd
[pol
->plid
]);
1122 blkg
->pd
[pol
->plid
] = NULL
;
1125 spin_unlock(&blkg
->blkcg
->lock
);
1128 spin_unlock_irq(q
->queue_lock
);
1129 blk_queue_bypass_end(q
);
1131 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy
);
1134 * blkcg_policy_register - register a blkcg policy
1135 * @pol: blkcg policy to register
1137 * Register @pol with blkcg core. Might sleep and @pol may be modified on
1138 * successful registration. Returns 0 on success and -errno on failure.
1140 int blkcg_policy_register(struct blkcg_policy
*pol
)
1142 struct blkcg
*blkcg
;
1145 mutex_lock(&blkcg_pol_register_mutex
);
1146 mutex_lock(&blkcg_pol_mutex
);
1148 /* find an empty slot */
1150 for (i
= 0; i
< BLKCG_MAX_POLS
; i
++)
1151 if (!blkcg_policy
[i
])
1153 if (i
>= BLKCG_MAX_POLS
)
1158 blkcg_policy
[pol
->plid
] = pol
;
1160 /* allocate and install cpd's */
1161 if (pol
->cpd_size
) {
1162 list_for_each_entry(blkcg
, &all_blkcgs
, all_blkcgs_node
) {
1163 struct blkcg_policy_data
*cpd
;
1165 cpd
= kzalloc(pol
->cpd_size
, GFP_KERNEL
);
1167 mutex_unlock(&blkcg_pol_mutex
);
1171 blkcg
->pd
[pol
->plid
] = cpd
;
1172 cpd
->plid
= pol
->plid
;
1173 pol
->cpd_init_fn(blkcg
);
1177 mutex_unlock(&blkcg_pol_mutex
);
1179 /* everything is in place, add intf files for the new policy */
1181 WARN_ON(cgroup_add_legacy_cftypes(&blkio_cgrp_subsys
,
1183 mutex_unlock(&blkcg_pol_register_mutex
);
1187 if (pol
->cpd_size
) {
1188 list_for_each_entry(blkcg
, &all_blkcgs
, all_blkcgs_node
) {
1189 kfree(blkcg
->pd
[pol
->plid
]);
1190 blkcg
->pd
[pol
->plid
] = NULL
;
1193 blkcg_policy
[pol
->plid
] = NULL
;
1195 mutex_unlock(&blkcg_pol_mutex
);
1196 mutex_unlock(&blkcg_pol_register_mutex
);
1199 EXPORT_SYMBOL_GPL(blkcg_policy_register
);
1202 * blkcg_policy_unregister - unregister a blkcg policy
1203 * @pol: blkcg policy to unregister
1205 * Undo blkcg_policy_register(@pol). Might sleep.
1207 void blkcg_policy_unregister(struct blkcg_policy
*pol
)
1209 struct blkcg
*blkcg
;
1211 mutex_lock(&blkcg_pol_register_mutex
);
1213 if (WARN_ON(blkcg_policy
[pol
->plid
] != pol
))
1216 /* kill the intf files first */
1218 cgroup_rm_cftypes(pol
->cftypes
);
1220 /* remove cpds and unregister */
1221 mutex_lock(&blkcg_pol_mutex
);
1223 if (pol
->cpd_size
) {
1224 list_for_each_entry(blkcg
, &all_blkcgs
, all_blkcgs_node
) {
1225 kfree(blkcg
->pd
[pol
->plid
]);
1226 blkcg
->pd
[pol
->plid
] = NULL
;
1229 blkcg_policy
[pol
->plid
] = NULL
;
1231 mutex_unlock(&blkcg_pol_mutex
);
1233 mutex_unlock(&blkcg_pol_register_mutex
);
1235 EXPORT_SYMBOL_GPL(blkcg_policy_unregister
);