4 * Common Block IO controller cgroup interface
6 * Based on ideas and code from CFQ, CFS and BFQ:
7 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
9 * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
10 * Paolo Valente <paolo.valente@unimore.it>
12 * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
13 * Nauman Rafique <nauman@google.com>
16 #include <linux/cgroup.h>
17 #include <linux/u64_stats_sync.h>
18 #include <linux/seq_file.h>
19 #include <linux/radix-tree.h>
20 #include <linux/blkdev.h>
21 #include <linux/atomic.h>
23 /* Max limits for throttle policy */
24 #define THROTL_IOPS_MAX UINT_MAX
26 #ifdef CONFIG_BLK_CGROUP
28 enum blkg_rwstat_type
{
35 BLKG_RWSTAT_TOTAL
= BLKG_RWSTAT_NR
,
41 struct cgroup_subsys_state css
;
44 struct radix_tree_root blkg_tree
;
45 struct blkcg_gq
*blkg_hint
;
46 struct hlist_head blkg_list
;
48 struct blkcg_policy_data
*pd
[BLKCG_MAX_POLS
];
50 struct list_head all_blkcgs_node
;
51 #ifdef CONFIG_CGROUP_WRITEBACK
52 struct list_head cgwb_list
;
57 struct u64_stats_sync syncp
;
62 struct u64_stats_sync syncp
;
63 uint64_t cnt
[BLKG_RWSTAT_NR
];
67 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
68 * request_queue (q). This is used by blkcg policies which need to track
69 * information per blkcg - q pair.
71 * There can be multiple active blkcg policies and each blkg:policy pair is
72 * represented by a blkg_policy_data which is allocated and freed by each
73 * policy's pd_alloc/free_fn() methods. A policy can allocate private data
74 * area by allocating larger data structure which embeds blkg_policy_data
77 struct blkg_policy_data
{
78 /* the blkg and policy id this per-policy data belongs to */
79 struct blkcg_gq
*blkg
;
84 * Policies that need to keep per-blkcg data which is independent
85 * from any request_queue associated to it must specify its size
86 * with the cpd_size field of the blkcg_policy structure and
87 * embed a blkcg_policy_data in it. cpd_init() is invoked to let
88 * each policy handle per-blkcg data.
90 struct blkcg_policy_data
{
91 /* the policy id this per-policy data belongs to */
95 /* association between a blk cgroup and a request queue */
97 /* Pointer to the associated request_queue */
98 struct request_queue
*q
;
99 struct list_head q_node
;
100 struct hlist_node blkcg_node
;
104 * Each blkg gets congested separately and the congestion state is
105 * propagated to the matching bdi_writeback_congested.
107 struct bdi_writeback_congested
*wb_congested
;
109 /* all non-root blkcg_gq's are guaranteed to have access to parent */
110 struct blkcg_gq
*parent
;
112 /* request allocation list for this blkcg-q pair */
113 struct request_list rl
;
115 /* reference count */
118 /* is this blkg online? protected by both blkcg and q locks */
121 struct blkg_policy_data
*pd
[BLKCG_MAX_POLS
];
123 struct rcu_head rcu_head
;
126 typedef void (blkcg_pol_init_cpd_fn
)(const struct blkcg
*blkcg
);
127 typedef struct blkg_policy_data
*(blkcg_pol_alloc_pd_fn
)(gfp_t gfp
, int node
);
128 typedef void (blkcg_pol_init_pd_fn
)(struct blkg_policy_data
*pd
);
129 typedef void (blkcg_pol_online_pd_fn
)(struct blkg_policy_data
*pd
);
130 typedef void (blkcg_pol_offline_pd_fn
)(struct blkg_policy_data
*pd
);
131 typedef void (blkcg_pol_free_pd_fn
)(struct blkg_policy_data
*pd
);
132 typedef void (blkcg_pol_reset_pd_stats_fn
)(struct blkg_policy_data
*pd
);
134 struct blkcg_policy
{
136 /* policy specific per-blkcg data size */
138 /* cgroup files for the policy */
139 struct cftype
*cftypes
;
142 blkcg_pol_init_cpd_fn
*cpd_init_fn
;
143 blkcg_pol_alloc_pd_fn
*pd_alloc_fn
;
144 blkcg_pol_init_pd_fn
*pd_init_fn
;
145 blkcg_pol_online_pd_fn
*pd_online_fn
;
146 blkcg_pol_offline_pd_fn
*pd_offline_fn
;
147 blkcg_pol_free_pd_fn
*pd_free_fn
;
148 blkcg_pol_reset_pd_stats_fn
*pd_reset_stats_fn
;
151 extern struct blkcg blkcg_root
;
152 extern struct cgroup_subsys_state
* const blkcg_root_css
;
154 struct blkcg_gq
*blkg_lookup(struct blkcg
*blkcg
, struct request_queue
*q
);
155 struct blkcg_gq
*blkg_lookup_create(struct blkcg
*blkcg
,
156 struct request_queue
*q
);
157 int blkcg_init_queue(struct request_queue
*q
);
158 void blkcg_drain_queue(struct request_queue
*q
);
159 void blkcg_exit_queue(struct request_queue
*q
);
161 /* Blkio controller policy registration */
162 int blkcg_policy_register(struct blkcg_policy
*pol
);
163 void blkcg_policy_unregister(struct blkcg_policy
*pol
);
164 int blkcg_activate_policy(struct request_queue
*q
,
165 const struct blkcg_policy
*pol
);
166 void blkcg_deactivate_policy(struct request_queue
*q
,
167 const struct blkcg_policy
*pol
);
169 void blkcg_print_blkgs(struct seq_file
*sf
, struct blkcg
*blkcg
,
170 u64 (*prfill
)(struct seq_file
*,
171 struct blkg_policy_data
*, int),
172 const struct blkcg_policy
*pol
, int data
,
174 u64
__blkg_prfill_u64(struct seq_file
*sf
, struct blkg_policy_data
*pd
, u64 v
);
175 u64
__blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
176 const struct blkg_rwstat
*rwstat
);
177 u64
blkg_prfill_stat(struct seq_file
*sf
, struct blkg_policy_data
*pd
, int off
);
178 u64
blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
181 u64
blkg_stat_recursive_sum(struct blkg_policy_data
*pd
, int off
);
182 struct blkg_rwstat
blkg_rwstat_recursive_sum(struct blkg_policy_data
*pd
,
185 struct blkg_conf_ctx
{
186 struct gendisk
*disk
;
187 struct blkcg_gq
*blkg
;
191 int blkg_conf_prep(struct blkcg
*blkcg
, const struct blkcg_policy
*pol
,
192 const char *input
, struct blkg_conf_ctx
*ctx
);
193 void blkg_conf_finish(struct blkg_conf_ctx
*ctx
);
196 static inline struct blkcg
*css_to_blkcg(struct cgroup_subsys_state
*css
)
198 return css
? container_of(css
, struct blkcg
, css
) : NULL
;
201 static inline struct blkcg
*task_blkcg(struct task_struct
*tsk
)
203 return css_to_blkcg(task_css(tsk
, blkio_cgrp_id
));
206 static inline struct blkcg
*bio_blkcg(struct bio
*bio
)
208 if (bio
&& bio
->bi_css
)
209 return css_to_blkcg(bio
->bi_css
);
210 return task_blkcg(current
);
213 static inline struct cgroup_subsys_state
*
214 task_get_blkcg_css(struct task_struct
*task
)
216 return task_get_css(task
, blkio_cgrp_id
);
220 * blkcg_parent - get the parent of a blkcg
221 * @blkcg: blkcg of interest
223 * Return the parent blkcg of @blkcg. Can be called anytime.
225 static inline struct blkcg
*blkcg_parent(struct blkcg
*blkcg
)
227 return css_to_blkcg(blkcg
->css
.parent
);
231 * blkg_to_pdata - get policy private data
232 * @blkg: blkg of interest
233 * @pol: policy of interest
235 * Return pointer to private data associated with the @blkg-@pol pair.
237 static inline struct blkg_policy_data
*blkg_to_pd(struct blkcg_gq
*blkg
,
238 struct blkcg_policy
*pol
)
240 return blkg
? blkg
->pd
[pol
->plid
] : NULL
;
243 static inline struct blkcg_policy_data
*blkcg_to_cpd(struct blkcg
*blkcg
,
244 struct blkcg_policy
*pol
)
246 return blkcg
? blkcg
->pd
[pol
->plid
] : NULL
;
250 * pdata_to_blkg - get blkg associated with policy private data
251 * @pd: policy private data of interest
253 * @pd is policy private data. Determine the blkg it's associated with.
255 static inline struct blkcg_gq
*pd_to_blkg(struct blkg_policy_data
*pd
)
257 return pd
? pd
->blkg
: NULL
;
261 * blkg_path - format cgroup path of blkg
262 * @blkg: blkg of interest
263 * @buf: target buffer
264 * @buflen: target buffer length
266 * Format the path of the cgroup of @blkg into @buf.
268 static inline int blkg_path(struct blkcg_gq
*blkg
, char *buf
, int buflen
)
272 p
= cgroup_path(blkg
->blkcg
->css
.cgroup
, buf
, buflen
);
274 strncpy(buf
, "<unavailable>", buflen
);
275 return -ENAMETOOLONG
;
278 memmove(buf
, p
, buf
+ buflen
- p
);
283 * blkg_get - get a blkg reference
286 * The caller should be holding an existing reference.
288 static inline void blkg_get(struct blkcg_gq
*blkg
)
290 WARN_ON_ONCE(atomic_read(&blkg
->refcnt
) <= 0);
291 atomic_inc(&blkg
->refcnt
);
294 void __blkg_release_rcu(struct rcu_head
*rcu
);
297 * blkg_put - put a blkg reference
300 static inline void blkg_put(struct blkcg_gq
*blkg
)
302 WARN_ON_ONCE(atomic_read(&blkg
->refcnt
) <= 0);
303 if (atomic_dec_and_test(&blkg
->refcnt
))
304 call_rcu(&blkg
->rcu_head
, __blkg_release_rcu
);
307 struct blkcg_gq
*__blkg_lookup(struct blkcg
*blkcg
, struct request_queue
*q
,
311 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
312 * @d_blkg: loop cursor pointing to the current descendant
313 * @pos_css: used for iteration
314 * @p_blkg: target blkg to walk descendants of
316 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
317 * read locked. If called under either blkcg or queue lock, the iteration
318 * is guaranteed to include all and only online blkgs. The caller may
319 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
320 * @p_blkg is included in the iteration and the first node to be visited.
322 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
323 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
324 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
325 (p_blkg)->q, false)))
328 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
329 * @d_blkg: loop cursor pointing to the current descendant
330 * @pos_css: used for iteration
331 * @p_blkg: target blkg to walk descendants of
333 * Similar to blkg_for_each_descendant_pre() but performs post-order
334 * traversal instead. Synchronization rules are the same. @p_blkg is
335 * included in the iteration and the last node to be visited.
337 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
338 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
339 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
340 (p_blkg)->q, false)))
343 * blk_get_rl - get request_list to use
344 * @q: request_queue of interest
345 * @bio: bio which will be attached to the allocated request (may be %NULL)
347 * The caller wants to allocate a request from @q to use for @bio. Find
348 * the request_list to use and obtain a reference on it. Should be called
349 * under queue_lock. This function is guaranteed to return non-%NULL
352 static inline struct request_list
*blk_get_rl(struct request_queue
*q
,
356 struct blkcg_gq
*blkg
;
360 blkcg
= bio_blkcg(bio
);
362 /* bypass blkg lookup and use @q->root_rl directly for root */
363 if (blkcg
== &blkcg_root
)
367 * Try to use blkg->rl. blkg lookup may fail under memory pressure
368 * or if either the blkcg or queue is going away. Fall back to
369 * root_rl in such cases.
371 blkg
= blkg_lookup_create(blkcg
, q
);
372 if (unlikely(IS_ERR(blkg
)))
384 * blk_put_rl - put request_list
385 * @rl: request_list to put
387 * Put the reference acquired by blk_get_rl(). Should be called under
390 static inline void blk_put_rl(struct request_list
*rl
)
392 if (rl
->blkg
->blkcg
!= &blkcg_root
)
397 * blk_rq_set_rl - associate a request with a request_list
398 * @rq: request of interest
399 * @rl: target request_list
401 * Associate @rq with @rl so that accounting and freeing can know the
402 * request_list @rq came from.
404 static inline void blk_rq_set_rl(struct request
*rq
, struct request_list
*rl
)
410 * blk_rq_rl - return the request_list a request came from
411 * @rq: request of interest
413 * Return the request_list @rq is allocated from.
415 static inline struct request_list
*blk_rq_rl(struct request
*rq
)
420 struct request_list
*__blk_queue_next_rl(struct request_list
*rl
,
421 struct request_queue
*q
);
423 * blk_queue_for_each_rl - iterate through all request_lists of a request_queue
425 * Should be used under queue_lock.
427 #define blk_queue_for_each_rl(rl, q) \
428 for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
430 static inline void blkg_stat_init(struct blkg_stat
*stat
)
432 u64_stats_init(&stat
->syncp
);
436 * blkg_stat_add - add a value to a blkg_stat
437 * @stat: target blkg_stat
440 * Add @val to @stat. The caller is responsible for synchronizing calls to
443 static inline void blkg_stat_add(struct blkg_stat
*stat
, uint64_t val
)
445 u64_stats_update_begin(&stat
->syncp
);
447 u64_stats_update_end(&stat
->syncp
);
451 * blkg_stat_read - read the current value of a blkg_stat
452 * @stat: blkg_stat to read
454 * Read the current value of @stat. This function can be called without
455 * synchroniztion and takes care of u64 atomicity.
457 static inline uint64_t blkg_stat_read(struct blkg_stat
*stat
)
463 start
= u64_stats_fetch_begin_irq(&stat
->syncp
);
465 } while (u64_stats_fetch_retry_irq(&stat
->syncp
, start
));
471 * blkg_stat_reset - reset a blkg_stat
472 * @stat: blkg_stat to reset
474 static inline void blkg_stat_reset(struct blkg_stat
*stat
)
480 * blkg_stat_merge - merge a blkg_stat into another
481 * @to: the destination blkg_stat
484 * Add @from's count to @to.
486 static inline void blkg_stat_merge(struct blkg_stat
*to
, struct blkg_stat
*from
)
488 blkg_stat_add(to
, blkg_stat_read(from
));
491 static inline void blkg_rwstat_init(struct blkg_rwstat
*rwstat
)
493 u64_stats_init(&rwstat
->syncp
);
497 * blkg_rwstat_add - add a value to a blkg_rwstat
498 * @rwstat: target blkg_rwstat
499 * @rw: mask of REQ_{WRITE|SYNC}
502 * Add @val to @rwstat. The counters are chosen according to @rw. The
503 * caller is responsible for synchronizing calls to this function.
505 static inline void blkg_rwstat_add(struct blkg_rwstat
*rwstat
,
506 int rw
, uint64_t val
)
508 u64_stats_update_begin(&rwstat
->syncp
);
511 rwstat
->cnt
[BLKG_RWSTAT_WRITE
] += val
;
513 rwstat
->cnt
[BLKG_RWSTAT_READ
] += val
;
515 rwstat
->cnt
[BLKG_RWSTAT_SYNC
] += val
;
517 rwstat
->cnt
[BLKG_RWSTAT_ASYNC
] += val
;
519 u64_stats_update_end(&rwstat
->syncp
);
523 * blkg_rwstat_read - read the current values of a blkg_rwstat
524 * @rwstat: blkg_rwstat to read
526 * Read the current snapshot of @rwstat and return it as the return value.
527 * This function can be called without synchronization and takes care of
530 static inline struct blkg_rwstat
blkg_rwstat_read(struct blkg_rwstat
*rwstat
)
533 struct blkg_rwstat tmp
;
536 start
= u64_stats_fetch_begin_irq(&rwstat
->syncp
);
538 } while (u64_stats_fetch_retry_irq(&rwstat
->syncp
, start
));
544 * blkg_rwstat_total - read the total count of a blkg_rwstat
545 * @rwstat: blkg_rwstat to read
547 * Return the total count of @rwstat regardless of the IO direction. This
548 * function can be called without synchronization and takes care of u64
551 static inline uint64_t blkg_rwstat_total(struct blkg_rwstat
*rwstat
)
553 struct blkg_rwstat tmp
= blkg_rwstat_read(rwstat
);
555 return tmp
.cnt
[BLKG_RWSTAT_READ
] + tmp
.cnt
[BLKG_RWSTAT_WRITE
];
559 * blkg_rwstat_reset - reset a blkg_rwstat
560 * @rwstat: blkg_rwstat to reset
562 static inline void blkg_rwstat_reset(struct blkg_rwstat
*rwstat
)
564 memset(rwstat
->cnt
, 0, sizeof(rwstat
->cnt
));
568 * blkg_rwstat_merge - merge a blkg_rwstat into another
569 * @to: the destination blkg_rwstat
572 * Add @from's counts to @to.
574 static inline void blkg_rwstat_merge(struct blkg_rwstat
*to
,
575 struct blkg_rwstat
*from
)
577 struct blkg_rwstat v
= blkg_rwstat_read(from
);
580 u64_stats_update_begin(&to
->syncp
);
581 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
582 to
->cnt
[i
] += v
.cnt
[i
];
583 u64_stats_update_end(&to
->syncp
);
586 #else /* CONFIG_BLK_CGROUP */
591 struct blkg_policy_data
{
594 struct blkcg_policy_data
{
600 struct blkcg_policy
{
603 #define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
605 static inline struct cgroup_subsys_state
*
606 task_get_blkcg_css(struct task_struct
*task
)
613 static inline struct blkcg_gq
*blkg_lookup(struct blkcg
*blkcg
, void *key
) { return NULL
; }
614 static inline int blkcg_init_queue(struct request_queue
*q
) { return 0; }
615 static inline void blkcg_drain_queue(struct request_queue
*q
) { }
616 static inline void blkcg_exit_queue(struct request_queue
*q
) { }
617 static inline int blkcg_policy_register(struct blkcg_policy
*pol
) { return 0; }
618 static inline void blkcg_policy_unregister(struct blkcg_policy
*pol
) { }
619 static inline int blkcg_activate_policy(struct request_queue
*q
,
620 const struct blkcg_policy
*pol
) { return 0; }
621 static inline void blkcg_deactivate_policy(struct request_queue
*q
,
622 const struct blkcg_policy
*pol
) { }
624 static inline struct blkcg
*bio_blkcg(struct bio
*bio
) { return NULL
; }
626 static inline struct blkg_policy_data
*blkg_to_pd(struct blkcg_gq
*blkg
,
627 struct blkcg_policy
*pol
) { return NULL
; }
628 static inline struct blkcg_gq
*pd_to_blkg(struct blkg_policy_data
*pd
) { return NULL
; }
629 static inline char *blkg_path(struct blkcg_gq
*blkg
) { return NULL
; }
630 static inline void blkg_get(struct blkcg_gq
*blkg
) { }
631 static inline void blkg_put(struct blkcg_gq
*blkg
) { }
633 static inline struct request_list
*blk_get_rl(struct request_queue
*q
,
634 struct bio
*bio
) { return &q
->root_rl
; }
635 static inline void blk_put_rl(struct request_list
*rl
) { }
636 static inline void blk_rq_set_rl(struct request
*rq
, struct request_list
*rl
) { }
637 static inline struct request_list
*blk_rq_rl(struct request
*rq
) { return &rq
->q
->root_rl
; }
639 #define blk_queue_for_each_rl(rl, q) \
640 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
642 #endif /* CONFIG_BLOCK */
643 #endif /* CONFIG_BLK_CGROUP */
644 #endif /* _BLK_CGROUP_H */