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>
22 /* Max limits for throttle policy */
23 #define THROTL_IOPS_MAX UINT_MAX
25 /* CFQ specific, out here for blkcg->cfq_weight */
26 #define CFQ_WEIGHT_MIN 10
27 #define CFQ_WEIGHT_MAX 1000
28 #define CFQ_WEIGHT_DEFAULT 500
30 #ifdef CONFIG_BLK_CGROUP
32 enum blkg_rwstat_type
{
39 BLKG_RWSTAT_TOTAL
= BLKG_RWSTAT_NR
,
45 struct cgroup_subsys_state css
;
48 struct radix_tree_root blkg_tree
;
49 struct blkcg_gq
*blkg_hint
;
50 struct hlist_head blkg_list
;
52 /* for policies to test whether associated blkcg has changed */
55 /* TODO: per-policy storage in blkcg */
56 unsigned int cfq_weight
; /* belongs to cfq */
57 unsigned int cfq_leaf_weight
;
61 struct u64_stats_sync syncp
;
66 struct u64_stats_sync syncp
;
67 uint64_t cnt
[BLKG_RWSTAT_NR
];
71 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
72 * request_queue (q). This is used by blkcg policies which need to track
73 * information per blkcg - q pair.
75 * There can be multiple active blkcg policies and each has its private
76 * data on each blkg, the size of which is determined by
77 * blkcg_policy->pd_size. blkcg core allocates and frees such areas
78 * together with blkg and invokes pd_init/exit_fn() methods.
80 * Such private data must embed struct blkg_policy_data (pd) at the
81 * beginning and pd_size can't be smaller than pd.
83 struct blkg_policy_data
{
84 /* the blkg and policy id this per-policy data belongs to */
85 struct blkcg_gq
*blkg
;
88 /* used during policy activation */
89 struct list_head alloc_node
;
92 /* association between a blk cgroup and a request queue */
94 /* Pointer to the associated request_queue */
95 struct request_queue
*q
;
96 struct list_head q_node
;
97 struct hlist_node blkcg_node
;
100 /* all non-root blkcg_gq's are guaranteed to have access to parent */
101 struct blkcg_gq
*parent
;
103 /* request allocation list for this blkcg-q pair */
104 struct request_list rl
;
106 /* reference count */
109 /* is this blkg online? protected by both blkcg and q locks */
112 struct blkg_policy_data
*pd
[BLKCG_MAX_POLS
];
114 struct rcu_head rcu_head
;
117 typedef void (blkcg_pol_init_pd_fn
)(struct blkcg_gq
*blkg
);
118 typedef void (blkcg_pol_online_pd_fn
)(struct blkcg_gq
*blkg
);
119 typedef void (blkcg_pol_offline_pd_fn
)(struct blkcg_gq
*blkg
);
120 typedef void (blkcg_pol_exit_pd_fn
)(struct blkcg_gq
*blkg
);
121 typedef void (blkcg_pol_reset_pd_stats_fn
)(struct blkcg_gq
*blkg
);
123 struct blkcg_policy
{
125 /* policy specific private data size */
127 /* cgroup files for the policy */
128 struct cftype
*cftypes
;
131 blkcg_pol_init_pd_fn
*pd_init_fn
;
132 blkcg_pol_online_pd_fn
*pd_online_fn
;
133 blkcg_pol_offline_pd_fn
*pd_offline_fn
;
134 blkcg_pol_exit_pd_fn
*pd_exit_fn
;
135 blkcg_pol_reset_pd_stats_fn
*pd_reset_stats_fn
;
138 extern struct blkcg blkcg_root
;
140 struct blkcg_gq
*blkg_lookup(struct blkcg
*blkcg
, struct request_queue
*q
);
141 struct blkcg_gq
*blkg_lookup_create(struct blkcg
*blkcg
,
142 struct request_queue
*q
);
143 int blkcg_init_queue(struct request_queue
*q
);
144 void blkcg_drain_queue(struct request_queue
*q
);
145 void blkcg_exit_queue(struct request_queue
*q
);
147 /* Blkio controller policy registration */
148 int blkcg_policy_register(struct blkcg_policy
*pol
);
149 void blkcg_policy_unregister(struct blkcg_policy
*pol
);
150 int blkcg_activate_policy(struct request_queue
*q
,
151 const struct blkcg_policy
*pol
);
152 void blkcg_deactivate_policy(struct request_queue
*q
,
153 const struct blkcg_policy
*pol
);
155 void blkcg_print_blkgs(struct seq_file
*sf
, struct blkcg
*blkcg
,
156 u64 (*prfill
)(struct seq_file
*,
157 struct blkg_policy_data
*, int),
158 const struct blkcg_policy
*pol
, int data
,
160 u64
__blkg_prfill_u64(struct seq_file
*sf
, struct blkg_policy_data
*pd
, u64 v
);
161 u64
__blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
162 const struct blkg_rwstat
*rwstat
);
163 u64
blkg_prfill_stat(struct seq_file
*sf
, struct blkg_policy_data
*pd
, int off
);
164 u64
blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
167 u64
blkg_stat_recursive_sum(struct blkg_policy_data
*pd
, int off
);
168 struct blkg_rwstat
blkg_rwstat_recursive_sum(struct blkg_policy_data
*pd
,
171 struct blkg_conf_ctx
{
172 struct gendisk
*disk
;
173 struct blkcg_gq
*blkg
;
177 int blkg_conf_prep(struct blkcg
*blkcg
, const struct blkcg_policy
*pol
,
178 const char *input
, struct blkg_conf_ctx
*ctx
);
179 void blkg_conf_finish(struct blkg_conf_ctx
*ctx
);
182 static inline struct blkcg
*css_to_blkcg(struct cgroup_subsys_state
*css
)
184 return css
? container_of(css
, struct blkcg
, css
) : NULL
;
187 static inline struct blkcg
*task_blkcg(struct task_struct
*tsk
)
189 return css_to_blkcg(task_css(tsk
, blkio_cgrp_id
));
192 static inline struct blkcg
*bio_blkcg(struct bio
*bio
)
194 if (bio
&& bio
->bi_css
)
195 return css_to_blkcg(bio
->bi_css
);
196 return task_blkcg(current
);
200 * blkcg_parent - get the parent of a blkcg
201 * @blkcg: blkcg of interest
203 * Return the parent blkcg of @blkcg. Can be called anytime.
205 static inline struct blkcg
*blkcg_parent(struct blkcg
*blkcg
)
207 return css_to_blkcg(css_parent(&blkcg
->css
));
211 * blkg_to_pdata - get policy private data
212 * @blkg: blkg of interest
213 * @pol: policy of interest
215 * Return pointer to private data associated with the @blkg-@pol pair.
217 static inline struct blkg_policy_data
*blkg_to_pd(struct blkcg_gq
*blkg
,
218 struct blkcg_policy
*pol
)
220 return blkg
? blkg
->pd
[pol
->plid
] : NULL
;
224 * pdata_to_blkg - get blkg associated with policy private data
225 * @pd: policy private data of interest
227 * @pd is policy private data. Determine the blkg it's associated with.
229 static inline struct blkcg_gq
*pd_to_blkg(struct blkg_policy_data
*pd
)
231 return pd
? pd
->blkg
: NULL
;
235 * blkg_path - format cgroup path of blkg
236 * @blkg: blkg of interest
237 * @buf: target buffer
238 * @buflen: target buffer length
240 * Format the path of the cgroup of @blkg into @buf.
242 static inline int blkg_path(struct blkcg_gq
*blkg
, char *buf
, int buflen
)
246 p
= cgroup_path(blkg
->blkcg
->css
.cgroup
, buf
, buflen
);
248 strncpy(buf
, "<unavailable>", buflen
);
249 return -ENAMETOOLONG
;
252 memmove(buf
, p
, buf
+ buflen
- p
);
257 * blkg_get - get a blkg reference
260 * The caller should be holding queue_lock and an existing reference.
262 static inline void blkg_get(struct blkcg_gq
*blkg
)
264 lockdep_assert_held(blkg
->q
->queue_lock
);
265 WARN_ON_ONCE(!blkg
->refcnt
);
269 void __blkg_release_rcu(struct rcu_head
*rcu
);
272 * blkg_put - put a blkg reference
275 * The caller should be holding queue_lock.
277 static inline void blkg_put(struct blkcg_gq
*blkg
)
279 lockdep_assert_held(blkg
->q
->queue_lock
);
280 WARN_ON_ONCE(blkg
->refcnt
<= 0);
282 call_rcu(&blkg
->rcu_head
, __blkg_release_rcu
);
285 struct blkcg_gq
*__blkg_lookup(struct blkcg
*blkcg
, struct request_queue
*q
,
289 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
290 * @d_blkg: loop cursor pointing to the current descendant
291 * @pos_css: used for iteration
292 * @p_blkg: target blkg to walk descendants of
294 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
295 * read locked. If called under either blkcg or queue lock, the iteration
296 * is guaranteed to include all and only online blkgs. The caller may
297 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
298 * @p_blkg is included in the iteration and the first node to be visited.
300 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
301 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
302 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
303 (p_blkg)->q, false)))
306 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
307 * @d_blkg: loop cursor pointing to the current descendant
308 * @pos_css: used for iteration
309 * @p_blkg: target blkg to walk descendants of
311 * Similar to blkg_for_each_descendant_pre() but performs post-order
312 * traversal instead. Synchronization rules are the same. @p_blkg is
313 * included in the iteration and the last node to be visited.
315 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
316 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
317 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
318 (p_blkg)->q, false)))
321 * blk_get_rl - get request_list to use
322 * @q: request_queue of interest
323 * @bio: bio which will be attached to the allocated request (may be %NULL)
325 * The caller wants to allocate a request from @q to use for @bio. Find
326 * the request_list to use and obtain a reference on it. Should be called
327 * under queue_lock. This function is guaranteed to return non-%NULL
330 static inline struct request_list
*blk_get_rl(struct request_queue
*q
,
334 struct blkcg_gq
*blkg
;
338 blkcg
= bio_blkcg(bio
);
340 /* bypass blkg lookup and use @q->root_rl directly for root */
341 if (blkcg
== &blkcg_root
)
345 * Try to use blkg->rl. blkg lookup may fail under memory pressure
346 * or if either the blkcg or queue is going away. Fall back to
347 * root_rl in such cases.
349 blkg
= blkg_lookup_create(blkcg
, q
);
350 if (unlikely(IS_ERR(blkg
)))
362 * blk_put_rl - put request_list
363 * @rl: request_list to put
365 * Put the reference acquired by blk_get_rl(). Should be called under
368 static inline void blk_put_rl(struct request_list
*rl
)
370 /* root_rl may not have blkg set */
371 if (rl
->blkg
&& rl
->blkg
->blkcg
!= &blkcg_root
)
376 * blk_rq_set_rl - associate a request with a request_list
377 * @rq: request of interest
378 * @rl: target request_list
380 * Associate @rq with @rl so that accounting and freeing can know the
381 * request_list @rq came from.
383 static inline void blk_rq_set_rl(struct request
*rq
, struct request_list
*rl
)
389 * blk_rq_rl - return the request_list a request came from
390 * @rq: request of interest
392 * Return the request_list @rq is allocated from.
394 static inline struct request_list
*blk_rq_rl(struct request
*rq
)
399 struct request_list
*__blk_queue_next_rl(struct request_list
*rl
,
400 struct request_queue
*q
);
402 * blk_queue_for_each_rl - iterate through all request_lists of a request_queue
404 * Should be used under queue_lock.
406 #define blk_queue_for_each_rl(rl, q) \
407 for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
409 static inline void blkg_stat_init(struct blkg_stat
*stat
)
411 u64_stats_init(&stat
->syncp
);
415 * blkg_stat_add - add a value to a blkg_stat
416 * @stat: target blkg_stat
419 * Add @val to @stat. The caller is responsible for synchronizing calls to
422 static inline void blkg_stat_add(struct blkg_stat
*stat
, uint64_t val
)
424 u64_stats_update_begin(&stat
->syncp
);
426 u64_stats_update_end(&stat
->syncp
);
430 * blkg_stat_read - read the current value of a blkg_stat
431 * @stat: blkg_stat to read
433 * Read the current value of @stat. This function can be called without
434 * synchroniztion and takes care of u64 atomicity.
436 static inline uint64_t blkg_stat_read(struct blkg_stat
*stat
)
442 start
= u64_stats_fetch_begin_bh(&stat
->syncp
);
444 } while (u64_stats_fetch_retry_bh(&stat
->syncp
, start
));
450 * blkg_stat_reset - reset a blkg_stat
451 * @stat: blkg_stat to reset
453 static inline void blkg_stat_reset(struct blkg_stat
*stat
)
459 * blkg_stat_merge - merge a blkg_stat into another
460 * @to: the destination blkg_stat
463 * Add @from's count to @to.
465 static inline void blkg_stat_merge(struct blkg_stat
*to
, struct blkg_stat
*from
)
467 blkg_stat_add(to
, blkg_stat_read(from
));
470 static inline void blkg_rwstat_init(struct blkg_rwstat
*rwstat
)
472 u64_stats_init(&rwstat
->syncp
);
476 * blkg_rwstat_add - add a value to a blkg_rwstat
477 * @rwstat: target blkg_rwstat
478 * @rw: mask of REQ_{WRITE|SYNC}
481 * Add @val to @rwstat. The counters are chosen according to @rw. The
482 * caller is responsible for synchronizing calls to this function.
484 static inline void blkg_rwstat_add(struct blkg_rwstat
*rwstat
,
485 int rw
, uint64_t val
)
487 u64_stats_update_begin(&rwstat
->syncp
);
490 rwstat
->cnt
[BLKG_RWSTAT_WRITE
] += val
;
492 rwstat
->cnt
[BLKG_RWSTAT_READ
] += val
;
494 rwstat
->cnt
[BLKG_RWSTAT_SYNC
] += val
;
496 rwstat
->cnt
[BLKG_RWSTAT_ASYNC
] += val
;
498 u64_stats_update_end(&rwstat
->syncp
);
502 * blkg_rwstat_read - read the current values of a blkg_rwstat
503 * @rwstat: blkg_rwstat to read
505 * Read the current snapshot of @rwstat and return it as the return value.
506 * This function can be called without synchronization and takes care of
509 static inline struct blkg_rwstat
blkg_rwstat_read(struct blkg_rwstat
*rwstat
)
512 struct blkg_rwstat tmp
;
515 start
= u64_stats_fetch_begin_bh(&rwstat
->syncp
);
517 } while (u64_stats_fetch_retry_bh(&rwstat
->syncp
, start
));
523 * blkg_rwstat_total - read the total count of a blkg_rwstat
524 * @rwstat: blkg_rwstat to read
526 * Return the total count of @rwstat regardless of the IO direction. This
527 * function can be called without synchronization and takes care of u64
530 static inline uint64_t blkg_rwstat_total(struct blkg_rwstat
*rwstat
)
532 struct blkg_rwstat tmp
= blkg_rwstat_read(rwstat
);
534 return tmp
.cnt
[BLKG_RWSTAT_READ
] + tmp
.cnt
[BLKG_RWSTAT_WRITE
];
538 * blkg_rwstat_reset - reset a blkg_rwstat
539 * @rwstat: blkg_rwstat to reset
541 static inline void blkg_rwstat_reset(struct blkg_rwstat
*rwstat
)
543 memset(rwstat
->cnt
, 0, sizeof(rwstat
->cnt
));
547 * blkg_rwstat_merge - merge a blkg_rwstat into another
548 * @to: the destination blkg_rwstat
551 * Add @from's counts to @to.
553 static inline void blkg_rwstat_merge(struct blkg_rwstat
*to
,
554 struct blkg_rwstat
*from
)
556 struct blkg_rwstat v
= blkg_rwstat_read(from
);
559 u64_stats_update_begin(&to
->syncp
);
560 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
561 to
->cnt
[i
] += v
.cnt
[i
];
562 u64_stats_update_end(&to
->syncp
);
565 #else /* CONFIG_BLK_CGROUP */
570 struct blkg_policy_data
{
576 struct blkcg_policy
{
579 static inline struct blkcg_gq
*blkg_lookup(struct blkcg
*blkcg
, void *key
) { return NULL
; }
580 static inline int blkcg_init_queue(struct request_queue
*q
) { return 0; }
581 static inline void blkcg_drain_queue(struct request_queue
*q
) { }
582 static inline void blkcg_exit_queue(struct request_queue
*q
) { }
583 static inline int blkcg_policy_register(struct blkcg_policy
*pol
) { return 0; }
584 static inline void blkcg_policy_unregister(struct blkcg_policy
*pol
) { }
585 static inline int blkcg_activate_policy(struct request_queue
*q
,
586 const struct blkcg_policy
*pol
) { return 0; }
587 static inline void blkcg_deactivate_policy(struct request_queue
*q
,
588 const struct blkcg_policy
*pol
) { }
590 static inline struct blkcg
*bio_blkcg(struct bio
*bio
) { return NULL
; }
592 static inline struct blkg_policy_data
*blkg_to_pd(struct blkcg_gq
*blkg
,
593 struct blkcg_policy
*pol
) { return NULL
; }
594 static inline struct blkcg_gq
*pd_to_blkg(struct blkg_policy_data
*pd
) { return NULL
; }
595 static inline char *blkg_path(struct blkcg_gq
*blkg
) { return NULL
; }
596 static inline void blkg_get(struct blkcg_gq
*blkg
) { }
597 static inline void blkg_put(struct blkcg_gq
*blkg
) { }
599 static inline struct request_list
*blk_get_rl(struct request_queue
*q
,
600 struct bio
*bio
) { return &q
->root_rl
; }
601 static inline void blk_put_rl(struct request_list
*rl
) { }
602 static inline void blk_rq_set_rl(struct request
*rq
, struct request_list
*rl
) { }
603 static inline struct request_list
*blk_rq_rl(struct request
*rq
) { return &rq
->q
->root_rl
; }
605 #define blk_queue_for_each_rl(rl, q) \
606 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
608 #endif /* CONFIG_BLK_CGROUP */
609 #endif /* _BLK_CGROUP_H */