1 /* SPDX-License-Identifier: GPL-2.0 */
5 * Common Block IO controller cgroup interface
7 * Based on ideas and code from CFQ, CFS and BFQ:
8 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
10 * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
11 * Paolo Valente <paolo.valente@unimore.it>
13 * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
14 * Nauman Rafique <nauman@google.com>
17 #include <linux/cgroup.h>
18 #include <linux/percpu_counter.h>
19 #include <linux/seq_file.h>
20 #include <linux/radix-tree.h>
21 #include <linux/blkdev.h>
22 #include <linux/atomic.h>
23 #include <linux/kthread.h>
26 /* percpu_counter batch for blkg_[rw]stats, per-cpu drift doesn't matter */
27 #define BLKG_STAT_CPU_BATCH (INT_MAX / 2)
29 /* Max limits for throttle policy */
30 #define THROTL_IOPS_MAX UINT_MAX
32 #ifdef CONFIG_BLK_CGROUP
34 enum blkg_rwstat_type
{
42 BLKG_RWSTAT_TOTAL
= BLKG_RWSTAT_NR
,
48 struct cgroup_subsys_state css
;
51 struct radix_tree_root blkg_tree
;
52 struct blkcg_gq __rcu
*blkg_hint
;
53 struct hlist_head blkg_list
;
55 struct blkcg_policy_data
*cpd
[BLKCG_MAX_POLS
];
57 struct list_head all_blkcgs_node
;
58 #ifdef CONFIG_CGROUP_WRITEBACK
59 struct list_head cgwb_list
;
60 refcount_t cgwb_refcnt
;
65 * blkg_[rw]stat->aux_cnt is excluded for local stats but included for
66 * recursive. Used to carry stats of dead children.
69 struct percpu_counter cpu_cnt
[BLKG_RWSTAT_NR
];
70 atomic64_t aux_cnt
[BLKG_RWSTAT_NR
];
73 struct blkg_rwstat_sample
{
74 u64 cnt
[BLKG_RWSTAT_NR
];
78 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
79 * request_queue (q). This is used by blkcg policies which need to track
80 * information per blkcg - q pair.
82 * There can be multiple active blkcg policies and each blkg:policy pair is
83 * represented by a blkg_policy_data which is allocated and freed by each
84 * policy's pd_alloc/free_fn() methods. A policy can allocate private data
85 * area by allocating larger data structure which embeds blkg_policy_data
88 struct blkg_policy_data
{
89 /* the blkg and policy id this per-policy data belongs to */
90 struct blkcg_gq
*blkg
;
95 * Policies that need to keep per-blkcg data which is independent from any
96 * request_queue associated to it should implement cpd_alloc/free_fn()
97 * methods. A policy can allocate private data area by allocating larger
98 * data structure which embeds blkcg_policy_data at the beginning.
99 * cpd_init() is invoked to let each policy handle per-blkcg data.
101 struct blkcg_policy_data
{
102 /* the blkcg and policy id this per-policy data belongs to */
107 /* association between a blk cgroup and a request queue */
109 /* Pointer to the associated request_queue */
110 struct request_queue
*q
;
111 struct list_head q_node
;
112 struct hlist_node blkcg_node
;
116 * Each blkg gets congested separately and the congestion state is
117 * propagated to the matching bdi_writeback_congested.
119 struct bdi_writeback_congested
*wb_congested
;
121 /* all non-root blkcg_gq's are guaranteed to have access to parent */
122 struct blkcg_gq
*parent
;
124 /* reference count */
125 struct percpu_ref refcnt
;
127 /* is this blkg online? protected by both blkcg and q locks */
130 struct blkg_rwstat stat_bytes
;
131 struct blkg_rwstat stat_ios
;
133 struct blkg_policy_data
*pd
[BLKCG_MAX_POLS
];
135 spinlock_t async_bio_lock
;
136 struct bio_list async_bios
;
137 struct work_struct async_bio_work
;
140 atomic64_t delay_nsec
;
141 atomic64_t delay_start
;
145 struct rcu_head rcu_head
;
148 typedef struct blkcg_policy_data
*(blkcg_pol_alloc_cpd_fn
)(gfp_t gfp
);
149 typedef void (blkcg_pol_init_cpd_fn
)(struct blkcg_policy_data
*cpd
);
150 typedef void (blkcg_pol_free_cpd_fn
)(struct blkcg_policy_data
*cpd
);
151 typedef void (blkcg_pol_bind_cpd_fn
)(struct blkcg_policy_data
*cpd
);
152 typedef struct blkg_policy_data
*(blkcg_pol_alloc_pd_fn
)(gfp_t gfp
, int node
);
153 typedef void (blkcg_pol_init_pd_fn
)(struct blkg_policy_data
*pd
);
154 typedef void (blkcg_pol_online_pd_fn
)(struct blkg_policy_data
*pd
);
155 typedef void (blkcg_pol_offline_pd_fn
)(struct blkg_policy_data
*pd
);
156 typedef void (blkcg_pol_free_pd_fn
)(struct blkg_policy_data
*pd
);
157 typedef void (blkcg_pol_reset_pd_stats_fn
)(struct blkg_policy_data
*pd
);
158 typedef size_t (blkcg_pol_stat_pd_fn
)(struct blkg_policy_data
*pd
, char *buf
,
161 struct blkcg_policy
{
163 /* cgroup files for the policy */
164 struct cftype
*dfl_cftypes
;
165 struct cftype
*legacy_cftypes
;
168 blkcg_pol_alloc_cpd_fn
*cpd_alloc_fn
;
169 blkcg_pol_init_cpd_fn
*cpd_init_fn
;
170 blkcg_pol_free_cpd_fn
*cpd_free_fn
;
171 blkcg_pol_bind_cpd_fn
*cpd_bind_fn
;
173 blkcg_pol_alloc_pd_fn
*pd_alloc_fn
;
174 blkcg_pol_init_pd_fn
*pd_init_fn
;
175 blkcg_pol_online_pd_fn
*pd_online_fn
;
176 blkcg_pol_offline_pd_fn
*pd_offline_fn
;
177 blkcg_pol_free_pd_fn
*pd_free_fn
;
178 blkcg_pol_reset_pd_stats_fn
*pd_reset_stats_fn
;
179 blkcg_pol_stat_pd_fn
*pd_stat_fn
;
182 extern struct blkcg blkcg_root
;
183 extern struct cgroup_subsys_state
* const blkcg_root_css
;
184 extern bool blkcg_debug_stats
;
186 struct blkcg_gq
*blkg_lookup_slowpath(struct blkcg
*blkcg
,
187 struct request_queue
*q
, bool update_hint
);
188 struct blkcg_gq
*__blkg_lookup_create(struct blkcg
*blkcg
,
189 struct request_queue
*q
);
190 struct blkcg_gq
*blkg_lookup_create(struct blkcg
*blkcg
,
191 struct request_queue
*q
);
192 int blkcg_init_queue(struct request_queue
*q
);
193 void blkcg_drain_queue(struct request_queue
*q
);
194 void blkcg_exit_queue(struct request_queue
*q
);
196 /* Blkio controller policy registration */
197 int blkcg_policy_register(struct blkcg_policy
*pol
);
198 void blkcg_policy_unregister(struct blkcg_policy
*pol
);
199 int blkcg_activate_policy(struct request_queue
*q
,
200 const struct blkcg_policy
*pol
);
201 void blkcg_deactivate_policy(struct request_queue
*q
,
202 const struct blkcg_policy
*pol
);
204 static inline u64
blkg_rwstat_read_counter(struct blkg_rwstat
*rwstat
,
207 return atomic64_read(&rwstat
->aux_cnt
[idx
]) +
208 percpu_counter_sum_positive(&rwstat
->cpu_cnt
[idx
]);
211 const char *blkg_dev_name(struct blkcg_gq
*blkg
);
212 void blkcg_print_blkgs(struct seq_file
*sf
, struct blkcg
*blkcg
,
213 u64 (*prfill
)(struct seq_file
*,
214 struct blkg_policy_data
*, int),
215 const struct blkcg_policy
*pol
, int data
,
217 u64
__blkg_prfill_u64(struct seq_file
*sf
, struct blkg_policy_data
*pd
, u64 v
);
218 u64
__blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
219 const struct blkg_rwstat_sample
*rwstat
);
220 u64
blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
222 int blkg_print_stat_bytes(struct seq_file
*sf
, void *v
);
223 int blkg_print_stat_ios(struct seq_file
*sf
, void *v
);
224 int blkg_print_stat_bytes_recursive(struct seq_file
*sf
, void *v
);
225 int blkg_print_stat_ios_recursive(struct seq_file
*sf
, void *v
);
227 void blkg_rwstat_recursive_sum(struct blkcg_gq
*blkg
, struct blkcg_policy
*pol
,
228 int off
, struct blkg_rwstat_sample
*sum
);
230 struct blkg_conf_ctx
{
231 struct gendisk
*disk
;
232 struct blkcg_gq
*blkg
;
236 int blkg_conf_prep(struct blkcg
*blkcg
, const struct blkcg_policy
*pol
,
237 char *input
, struct blkg_conf_ctx
*ctx
);
238 void blkg_conf_finish(struct blkg_conf_ctx
*ctx
);
241 * blkcg_css - find the current css
243 * Find the css associated with either the kthread or the current task.
244 * This may return a dying css, so it is up to the caller to use tryget logic
245 * to confirm it is alive and well.
247 static inline struct cgroup_subsys_state
*blkcg_css(void)
249 struct cgroup_subsys_state
*css
;
251 css
= kthread_blkcg();
254 return task_css(current
, io_cgrp_id
);
257 static inline struct blkcg
*css_to_blkcg(struct cgroup_subsys_state
*css
)
259 return css
? container_of(css
, struct blkcg
, css
) : NULL
;
263 * __bio_blkcg - internal, inconsistent version to get blkcg
266 * This function is inconsistent and consequently is dangerous to use. The
267 * first part of the function returns a blkcg where a reference is owned by the
268 * bio. This means it does not need to be rcu protected as it cannot go away
269 * with the bio owning a reference to it. However, the latter potentially gets
270 * it from task_css(). This can race against task migration and the cgroup
271 * dying. It is also semantically different as it must be called rcu protected
272 * and is susceptible to failure when trying to get a reference to it.
273 * Therefore, it is not ok to assume that *_get() will always succeed on the
274 * blkcg returned here.
276 static inline struct blkcg
*__bio_blkcg(struct bio
*bio
)
278 if (bio
&& bio
->bi_blkg
)
279 return bio
->bi_blkg
->blkcg
;
280 return css_to_blkcg(blkcg_css());
284 * bio_blkcg - grab the blkcg associated with a bio
287 * This returns the blkcg associated with a bio, %NULL if not associated.
288 * Callers are expected to either handle %NULL or know association has been
289 * done prior to calling this.
291 static inline struct blkcg
*bio_blkcg(struct bio
*bio
)
293 if (bio
&& bio
->bi_blkg
)
294 return bio
->bi_blkg
->blkcg
;
298 static inline bool blk_cgroup_congested(void)
300 struct cgroup_subsys_state
*css
;
304 css
= kthread_blkcg();
306 css
= task_css(current
, io_cgrp_id
);
308 if (atomic_read(&css
->cgroup
->congestion_count
)) {
319 * bio_issue_as_root_blkg - see if this bio needs to be issued as root blkg
320 * @return: true if this bio needs to be submitted with the root blkg context.
322 * In order to avoid priority inversions we sometimes need to issue a bio as if
323 * it were attached to the root blkg, and then backcharge to the actual owning
324 * blkg. The idea is we do bio_blkcg() to look up the actual context for the
325 * bio and attach the appropriate blkg to the bio. Then we call this helper and
326 * if it is true run with the root blkg for that queue and then do any
327 * backcharging to the originating cgroup once the io is complete.
329 static inline bool bio_issue_as_root_blkg(struct bio
*bio
)
331 return (bio
->bi_opf
& (REQ_META
| REQ_SWAP
)) != 0;
335 * blkcg_parent - get the parent of a blkcg
336 * @blkcg: blkcg of interest
338 * Return the parent blkcg of @blkcg. Can be called anytime.
340 static inline struct blkcg
*blkcg_parent(struct blkcg
*blkcg
)
342 return css_to_blkcg(blkcg
->css
.parent
);
346 * __blkg_lookup - internal version of blkg_lookup()
347 * @blkcg: blkcg of interest
348 * @q: request_queue of interest
349 * @update_hint: whether to update lookup hint with the result or not
351 * This is internal version and shouldn't be used by policy
352 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
353 * @q's bypass state. If @update_hint is %true, the caller should be
354 * holding @q->queue_lock and lookup hint is updated on success.
356 static inline struct blkcg_gq
*__blkg_lookup(struct blkcg
*blkcg
,
357 struct request_queue
*q
,
360 struct blkcg_gq
*blkg
;
362 if (blkcg
== &blkcg_root
)
365 blkg
= rcu_dereference(blkcg
->blkg_hint
);
366 if (blkg
&& blkg
->q
== q
)
369 return blkg_lookup_slowpath(blkcg
, q
, update_hint
);
373 * blkg_lookup - lookup blkg for the specified blkcg - q pair
374 * @blkcg: blkcg of interest
375 * @q: request_queue of interest
377 * Lookup blkg for the @blkcg - @q pair. This function should be called
378 * under RCU read loc.
380 static inline struct blkcg_gq
*blkg_lookup(struct blkcg
*blkcg
,
381 struct request_queue
*q
)
383 WARN_ON_ONCE(!rcu_read_lock_held());
384 return __blkg_lookup(blkcg
, q
, false);
388 * blk_queue_root_blkg - return blkg for the (blkcg_root, @q) pair
389 * @q: request_queue of interest
391 * Lookup blkg for @q at the root level. See also blkg_lookup().
393 static inline struct blkcg_gq
*blk_queue_root_blkg(struct request_queue
*q
)
399 * blkg_to_pdata - get policy private data
400 * @blkg: blkg of interest
401 * @pol: policy of interest
403 * Return pointer to private data associated with the @blkg-@pol pair.
405 static inline struct blkg_policy_data
*blkg_to_pd(struct blkcg_gq
*blkg
,
406 struct blkcg_policy
*pol
)
408 return blkg
? blkg
->pd
[pol
->plid
] : NULL
;
411 static inline struct blkcg_policy_data
*blkcg_to_cpd(struct blkcg
*blkcg
,
412 struct blkcg_policy
*pol
)
414 return blkcg
? blkcg
->cpd
[pol
->plid
] : NULL
;
418 * pdata_to_blkg - get blkg associated with policy private data
419 * @pd: policy private data of interest
421 * @pd is policy private data. Determine the blkg it's associated with.
423 static inline struct blkcg_gq
*pd_to_blkg(struct blkg_policy_data
*pd
)
425 return pd
? pd
->blkg
: NULL
;
428 static inline struct blkcg
*cpd_to_blkcg(struct blkcg_policy_data
*cpd
)
430 return cpd
? cpd
->blkcg
: NULL
;
433 extern void blkcg_destroy_blkgs(struct blkcg
*blkcg
);
435 #ifdef CONFIG_CGROUP_WRITEBACK
438 * blkcg_cgwb_get - get a reference for blkcg->cgwb_list
439 * @blkcg: blkcg of interest
441 * This is used to track the number of active wb's related to a blkcg.
443 static inline void blkcg_cgwb_get(struct blkcg
*blkcg
)
445 refcount_inc(&blkcg
->cgwb_refcnt
);
449 * blkcg_cgwb_put - put a reference for @blkcg->cgwb_list
450 * @blkcg: blkcg of interest
452 * This is used to track the number of active wb's related to a blkcg.
453 * When this count goes to zero, all active wb has finished so the
454 * blkcg can continue destruction by calling blkcg_destroy_blkgs().
455 * This work may occur in cgwb_release_workfn() on the cgwb_release
458 static inline void blkcg_cgwb_put(struct blkcg
*blkcg
)
460 if (refcount_dec_and_test(&blkcg
->cgwb_refcnt
))
461 blkcg_destroy_blkgs(blkcg
);
466 static inline void blkcg_cgwb_get(struct blkcg
*blkcg
) { }
468 static inline void blkcg_cgwb_put(struct blkcg
*blkcg
)
470 /* wb isn't being accounted, so trigger destruction right away */
471 blkcg_destroy_blkgs(blkcg
);
477 * blkg_path - format cgroup path of blkg
478 * @blkg: blkg of interest
479 * @buf: target buffer
480 * @buflen: target buffer length
482 * Format the path of the cgroup of @blkg into @buf.
484 static inline int blkg_path(struct blkcg_gq
*blkg
, char *buf
, int buflen
)
486 return cgroup_path(blkg
->blkcg
->css
.cgroup
, buf
, buflen
);
490 * blkg_get - get a blkg reference
493 * The caller should be holding an existing reference.
495 static inline void blkg_get(struct blkcg_gq
*blkg
)
497 percpu_ref_get(&blkg
->refcnt
);
501 * blkg_tryget - try and get a blkg reference
504 * This is for use when doing an RCU lookup of the blkg. We may be in the midst
505 * of freeing this blkg, so we can only use it if the refcnt is not zero.
507 static inline bool blkg_tryget(struct blkcg_gq
*blkg
)
509 return blkg
&& percpu_ref_tryget(&blkg
->refcnt
);
513 * blkg_tryget_closest - try and get a blkg ref on the closet blkg
516 * This needs to be called rcu protected. As the failure mode here is to walk
517 * up the blkg tree, this ensure that the blkg->parent pointers are always
518 * valid. This returns the blkg that it ended up taking a reference on or %NULL
519 * if no reference was taken.
521 static inline struct blkcg_gq
*blkg_tryget_closest(struct blkcg_gq
*blkg
)
523 struct blkcg_gq
*ret_blkg
= NULL
;
525 WARN_ON_ONCE(!rcu_read_lock_held());
528 if (blkg_tryget(blkg
)) {
539 * blkg_put - put a blkg reference
542 static inline void blkg_put(struct blkcg_gq
*blkg
)
544 percpu_ref_put(&blkg
->refcnt
);
548 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
549 * @d_blkg: loop cursor pointing to the current descendant
550 * @pos_css: used for iteration
551 * @p_blkg: target blkg to walk descendants of
553 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
554 * read locked. If called under either blkcg or queue lock, the iteration
555 * is guaranteed to include all and only online blkgs. The caller may
556 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
557 * @p_blkg is included in the iteration and the first node to be visited.
559 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
560 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
561 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
562 (p_blkg)->q, false)))
565 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
566 * @d_blkg: loop cursor pointing to the current descendant
567 * @pos_css: used for iteration
568 * @p_blkg: target blkg to walk descendants of
570 * Similar to blkg_for_each_descendant_pre() but performs post-order
571 * traversal instead. Synchronization rules are the same. @p_blkg is
572 * included in the iteration and the last node to be visited.
574 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
575 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
576 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
577 (p_blkg)->q, false)))
579 static inline int blkg_rwstat_init(struct blkg_rwstat
*rwstat
, gfp_t gfp
)
583 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++) {
584 ret
= percpu_counter_init(&rwstat
->cpu_cnt
[i
], 0, gfp
);
587 percpu_counter_destroy(&rwstat
->cpu_cnt
[i
]);
590 atomic64_set(&rwstat
->aux_cnt
[i
], 0);
595 static inline void blkg_rwstat_exit(struct blkg_rwstat
*rwstat
)
599 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
600 percpu_counter_destroy(&rwstat
->cpu_cnt
[i
]);
604 * blkg_rwstat_add - add a value to a blkg_rwstat
605 * @rwstat: target blkg_rwstat
606 * @op: REQ_OP and flags
609 * Add @val to @rwstat. The counters are chosen according to @rw. The
610 * caller is responsible for synchronizing calls to this function.
612 static inline void blkg_rwstat_add(struct blkg_rwstat
*rwstat
,
613 unsigned int op
, uint64_t val
)
615 struct percpu_counter
*cnt
;
617 if (op_is_discard(op
))
618 cnt
= &rwstat
->cpu_cnt
[BLKG_RWSTAT_DISCARD
];
619 else if (op_is_write(op
))
620 cnt
= &rwstat
->cpu_cnt
[BLKG_RWSTAT_WRITE
];
622 cnt
= &rwstat
->cpu_cnt
[BLKG_RWSTAT_READ
];
624 percpu_counter_add_batch(cnt
, val
, BLKG_STAT_CPU_BATCH
);
627 cnt
= &rwstat
->cpu_cnt
[BLKG_RWSTAT_SYNC
];
629 cnt
= &rwstat
->cpu_cnt
[BLKG_RWSTAT_ASYNC
];
631 percpu_counter_add_batch(cnt
, val
, BLKG_STAT_CPU_BATCH
);
635 * blkg_rwstat_read - read the current values of a blkg_rwstat
636 * @rwstat: blkg_rwstat to read
638 * Read the current snapshot of @rwstat and return it in the aux counts.
640 static inline void blkg_rwstat_read(struct blkg_rwstat
*rwstat
,
641 struct blkg_rwstat_sample
*result
)
645 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
647 percpu_counter_sum_positive(&rwstat
->cpu_cnt
[i
]);
651 * blkg_rwstat_total - read the total count of a blkg_rwstat
652 * @rwstat: blkg_rwstat to read
654 * Return the total count of @rwstat regardless of the IO direction. This
655 * function can be called without synchronization and takes care of u64
658 static inline uint64_t blkg_rwstat_total(struct blkg_rwstat
*rwstat
)
660 struct blkg_rwstat_sample tmp
= { };
662 blkg_rwstat_read(rwstat
, &tmp
);
663 return tmp
.cnt
[BLKG_RWSTAT_READ
] + tmp
.cnt
[BLKG_RWSTAT_WRITE
];
667 * blkg_rwstat_reset - reset a blkg_rwstat
668 * @rwstat: blkg_rwstat to reset
670 static inline void blkg_rwstat_reset(struct blkg_rwstat
*rwstat
)
674 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++) {
675 percpu_counter_set(&rwstat
->cpu_cnt
[i
], 0);
676 atomic64_set(&rwstat
->aux_cnt
[i
], 0);
681 * blkg_rwstat_add_aux - add a blkg_rwstat into another's aux count
682 * @to: the destination blkg_rwstat
685 * Add @from's count including the aux one to @to's aux count.
687 static inline void blkg_rwstat_add_aux(struct blkg_rwstat
*to
,
688 struct blkg_rwstat
*from
)
690 u64 sum
[BLKG_RWSTAT_NR
];
693 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
694 sum
[i
] = percpu_counter_sum_positive(&from
->cpu_cnt
[i
]);
696 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
697 atomic64_add(sum
[i
] + atomic64_read(&from
->aux_cnt
[i
]),
701 #ifdef CONFIG_BLK_DEV_THROTTLING
702 extern bool blk_throtl_bio(struct request_queue
*q
, struct blkcg_gq
*blkg
,
705 static inline bool blk_throtl_bio(struct request_queue
*q
, struct blkcg_gq
*blkg
,
706 struct bio
*bio
) { return false; }
709 bool __blkcg_punt_bio_submit(struct bio
*bio
);
711 static inline bool blkcg_punt_bio_submit(struct bio
*bio
)
713 if (bio
->bi_opf
& REQ_CGROUP_PUNT
)
714 return __blkcg_punt_bio_submit(bio
);
719 static inline void blkcg_bio_issue_init(struct bio
*bio
)
721 bio_issue_init(&bio
->bi_issue
, bio_sectors(bio
));
724 static inline bool blkcg_bio_issue_check(struct request_queue
*q
,
727 struct blkcg_gq
*blkg
;
733 char b
[BDEVNAME_SIZE
];
736 "no blkg associated for bio on block-device: %s\n",
737 bio_devname(bio
, b
));
738 bio_associate_blkg(bio
);
743 throtl
= blk_throtl_bio(q
, blkg
, bio
);
747 * If the bio is flagged with BIO_QUEUE_ENTERED it means this
748 * is a split bio and we would have already accounted for the
751 if (!bio_flagged(bio
, BIO_QUEUE_ENTERED
))
752 blkg_rwstat_add(&blkg
->stat_bytes
, bio
->bi_opf
,
753 bio
->bi_iter
.bi_size
);
754 blkg_rwstat_add(&blkg
->stat_ios
, bio
->bi_opf
, 1);
757 blkcg_bio_issue_init(bio
);
763 static inline void blkcg_use_delay(struct blkcg_gq
*blkg
)
765 if (atomic_add_return(1, &blkg
->use_delay
) == 1)
766 atomic_inc(&blkg
->blkcg
->css
.cgroup
->congestion_count
);
769 static inline int blkcg_unuse_delay(struct blkcg_gq
*blkg
)
771 int old
= atomic_read(&blkg
->use_delay
);
777 * We do this song and dance because we can race with somebody else
778 * adding or removing delay. If we just did an atomic_dec we'd end up
779 * negative and we'd already be in trouble. We need to subtract 1 and
780 * then check to see if we were the last delay so we can drop the
781 * congestion count on the cgroup.
784 int cur
= atomic_cmpxchg(&blkg
->use_delay
, old
, old
- 1);
793 atomic_dec(&blkg
->blkcg
->css
.cgroup
->congestion_count
);
797 static inline void blkcg_clear_delay(struct blkcg_gq
*blkg
)
799 int old
= atomic_read(&blkg
->use_delay
);
802 /* We only want 1 person clearing the congestion count for this blkg. */
804 int cur
= atomic_cmpxchg(&blkg
->use_delay
, old
, 0);
806 atomic_dec(&blkg
->blkcg
->css
.cgroup
->congestion_count
);
813 void blkcg_add_delay(struct blkcg_gq
*blkg
, u64 now
, u64 delta
);
814 void blkcg_schedule_throttle(struct request_queue
*q
, bool use_memdelay
);
815 void blkcg_maybe_throttle_current(void);
816 #else /* CONFIG_BLK_CGROUP */
821 struct blkg_policy_data
{
824 struct blkcg_policy_data
{
830 struct blkcg_policy
{
833 #define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
835 static inline void blkcg_maybe_throttle_current(void) { }
836 static inline bool blk_cgroup_congested(void) { return false; }
840 static inline void blkcg_schedule_throttle(struct request_queue
*q
, bool use_memdelay
) { }
842 static inline struct blkcg_gq
*blkg_lookup(struct blkcg
*blkcg
, void *key
) { return NULL
; }
843 static inline struct blkcg_gq
*blk_queue_root_blkg(struct request_queue
*q
)
845 static inline int blkcg_init_queue(struct request_queue
*q
) { return 0; }
846 static inline void blkcg_drain_queue(struct request_queue
*q
) { }
847 static inline void blkcg_exit_queue(struct request_queue
*q
) { }
848 static inline int blkcg_policy_register(struct blkcg_policy
*pol
) { return 0; }
849 static inline void blkcg_policy_unregister(struct blkcg_policy
*pol
) { }
850 static inline int blkcg_activate_policy(struct request_queue
*q
,
851 const struct blkcg_policy
*pol
) { return 0; }
852 static inline void blkcg_deactivate_policy(struct request_queue
*q
,
853 const struct blkcg_policy
*pol
) { }
855 static inline struct blkcg
*__bio_blkcg(struct bio
*bio
) { return NULL
; }
856 static inline struct blkcg
*bio_blkcg(struct bio
*bio
) { return NULL
; }
858 static inline struct blkg_policy_data
*blkg_to_pd(struct blkcg_gq
*blkg
,
859 struct blkcg_policy
*pol
) { return NULL
; }
860 static inline struct blkcg_gq
*pd_to_blkg(struct blkg_policy_data
*pd
) { return NULL
; }
861 static inline char *blkg_path(struct blkcg_gq
*blkg
) { return NULL
; }
862 static inline void blkg_get(struct blkcg_gq
*blkg
) { }
863 static inline void blkg_put(struct blkcg_gq
*blkg
) { }
865 static inline bool blkcg_punt_bio_submit(struct bio
*bio
) { return false; }
866 static inline void blkcg_bio_issue_init(struct bio
*bio
) { }
867 static inline bool blkcg_bio_issue_check(struct request_queue
*q
,
868 struct bio
*bio
) { return true; }
870 #define blk_queue_for_each_rl(rl, q) \
871 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
873 #endif /* CONFIG_BLOCK */
874 #endif /* CONFIG_BLK_CGROUP */
875 #endif /* _BLK_CGROUP_H */