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 struct blkg_policy_data
*pd
[BLKCG_MAX_POLS
];
111 struct rcu_head rcu_head
;
114 typedef void (blkcg_pol_init_pd_fn
)(struct blkcg_gq
*blkg
);
115 typedef void (blkcg_pol_exit_pd_fn
)(struct blkcg_gq
*blkg
);
116 typedef void (blkcg_pol_reset_pd_stats_fn
)(struct blkcg_gq
*blkg
);
118 struct blkcg_policy
{
120 /* policy specific private data size */
122 /* cgroup files for the policy */
123 struct cftype
*cftypes
;
126 blkcg_pol_init_pd_fn
*pd_init_fn
;
127 blkcg_pol_exit_pd_fn
*pd_exit_fn
;
128 blkcg_pol_reset_pd_stats_fn
*pd_reset_stats_fn
;
131 extern struct blkcg blkcg_root
;
133 struct blkcg_gq
*blkg_lookup(struct blkcg
*blkcg
, struct request_queue
*q
);
134 struct blkcg_gq
*blkg_lookup_create(struct blkcg
*blkcg
,
135 struct request_queue
*q
);
136 int blkcg_init_queue(struct request_queue
*q
);
137 void blkcg_drain_queue(struct request_queue
*q
);
138 void blkcg_exit_queue(struct request_queue
*q
);
140 /* Blkio controller policy registration */
141 int blkcg_policy_register(struct blkcg_policy
*pol
);
142 void blkcg_policy_unregister(struct blkcg_policy
*pol
);
143 int blkcg_activate_policy(struct request_queue
*q
,
144 const struct blkcg_policy
*pol
);
145 void blkcg_deactivate_policy(struct request_queue
*q
,
146 const struct blkcg_policy
*pol
);
148 void blkcg_print_blkgs(struct seq_file
*sf
, struct blkcg
*blkcg
,
149 u64 (*prfill
)(struct seq_file
*,
150 struct blkg_policy_data
*, int),
151 const struct blkcg_policy
*pol
, int data
,
153 u64
__blkg_prfill_u64(struct seq_file
*sf
, struct blkg_policy_data
*pd
, u64 v
);
154 u64
__blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
155 const struct blkg_rwstat
*rwstat
);
156 u64
blkg_prfill_stat(struct seq_file
*sf
, struct blkg_policy_data
*pd
, int off
);
157 u64
blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
160 struct blkg_conf_ctx
{
161 struct gendisk
*disk
;
162 struct blkcg_gq
*blkg
;
166 int blkg_conf_prep(struct blkcg
*blkcg
, const struct blkcg_policy
*pol
,
167 const char *input
, struct blkg_conf_ctx
*ctx
);
168 void blkg_conf_finish(struct blkg_conf_ctx
*ctx
);
171 static inline struct blkcg
*cgroup_to_blkcg(struct cgroup
*cgroup
)
173 return container_of(cgroup_subsys_state(cgroup
, blkio_subsys_id
),
177 static inline struct blkcg
*task_blkcg(struct task_struct
*tsk
)
179 return container_of(task_subsys_state(tsk
, blkio_subsys_id
),
183 static inline struct blkcg
*bio_blkcg(struct bio
*bio
)
185 if (bio
&& bio
->bi_css
)
186 return container_of(bio
->bi_css
, struct blkcg
, css
);
187 return task_blkcg(current
);
191 * blkcg_parent - get the parent of a blkcg
192 * @blkcg: blkcg of interest
194 * Return the parent blkcg of @blkcg. Can be called anytime.
196 static inline struct blkcg
*blkcg_parent(struct blkcg
*blkcg
)
198 struct cgroup
*pcg
= blkcg
->css
.cgroup
->parent
;
200 return pcg
? cgroup_to_blkcg(pcg
) : NULL
;
204 * blkg_to_pdata - get policy private data
205 * @blkg: blkg of interest
206 * @pol: policy of interest
208 * Return pointer to private data associated with the @blkg-@pol pair.
210 static inline struct blkg_policy_data
*blkg_to_pd(struct blkcg_gq
*blkg
,
211 struct blkcg_policy
*pol
)
213 return blkg
? blkg
->pd
[pol
->plid
] : NULL
;
217 * pdata_to_blkg - get blkg associated with policy private data
218 * @pd: policy private data of interest
220 * @pd is policy private data. Determine the blkg it's associated with.
222 static inline struct blkcg_gq
*pd_to_blkg(struct blkg_policy_data
*pd
)
224 return pd
? pd
->blkg
: NULL
;
228 * blkg_path - format cgroup path of blkg
229 * @blkg: blkg of interest
230 * @buf: target buffer
231 * @buflen: target buffer length
233 * Format the path of the cgroup of @blkg into @buf.
235 static inline int blkg_path(struct blkcg_gq
*blkg
, char *buf
, int buflen
)
240 ret
= cgroup_path(blkg
->blkcg
->css
.cgroup
, buf
, buflen
);
243 strncpy(buf
, "<unavailable>", buflen
);
248 * blkg_get - get a blkg reference
251 * The caller should be holding queue_lock and an existing reference.
253 static inline void blkg_get(struct blkcg_gq
*blkg
)
255 lockdep_assert_held(blkg
->q
->queue_lock
);
256 WARN_ON_ONCE(!blkg
->refcnt
);
260 void __blkg_release(struct blkcg_gq
*blkg
);
263 * blkg_put - put a blkg reference
266 * The caller should be holding queue_lock.
268 static inline void blkg_put(struct blkcg_gq
*blkg
)
270 lockdep_assert_held(blkg
->q
->queue_lock
);
271 WARN_ON_ONCE(blkg
->refcnt
<= 0);
273 __blkg_release(blkg
);
277 * blk_get_rl - get request_list to use
278 * @q: request_queue of interest
279 * @bio: bio which will be attached to the allocated request (may be %NULL)
281 * The caller wants to allocate a request from @q to use for @bio. Find
282 * the request_list to use and obtain a reference on it. Should be called
283 * under queue_lock. This function is guaranteed to return non-%NULL
286 static inline struct request_list
*blk_get_rl(struct request_queue
*q
,
290 struct blkcg_gq
*blkg
;
294 blkcg
= bio_blkcg(bio
);
296 /* bypass blkg lookup and use @q->root_rl directly for root */
297 if (blkcg
== &blkcg_root
)
301 * Try to use blkg->rl. blkg lookup may fail under memory pressure
302 * or if either the blkcg or queue is going away. Fall back to
303 * root_rl in such cases.
305 blkg
= blkg_lookup_create(blkcg
, q
);
306 if (unlikely(IS_ERR(blkg
)))
318 * blk_put_rl - put request_list
319 * @rl: request_list to put
321 * Put the reference acquired by blk_get_rl(). Should be called under
324 static inline void blk_put_rl(struct request_list
*rl
)
326 /* root_rl may not have blkg set */
327 if (rl
->blkg
&& rl
->blkg
->blkcg
!= &blkcg_root
)
332 * blk_rq_set_rl - associate a request with a request_list
333 * @rq: request of interest
334 * @rl: target request_list
336 * Associate @rq with @rl so that accounting and freeing can know the
337 * request_list @rq came from.
339 static inline void blk_rq_set_rl(struct request
*rq
, struct request_list
*rl
)
345 * blk_rq_rl - return the request_list a request came from
346 * @rq: request of interest
348 * Return the request_list @rq is allocated from.
350 static inline struct request_list
*blk_rq_rl(struct request
*rq
)
355 struct request_list
*__blk_queue_next_rl(struct request_list
*rl
,
356 struct request_queue
*q
);
358 * blk_queue_for_each_rl - iterate through all request_lists of a request_queue
360 * Should be used under queue_lock.
362 #define blk_queue_for_each_rl(rl, q) \
363 for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
366 * blkg_stat_add - add a value to a blkg_stat
367 * @stat: target blkg_stat
370 * Add @val to @stat. The caller is responsible for synchronizing calls to
373 static inline void blkg_stat_add(struct blkg_stat
*stat
, uint64_t val
)
375 u64_stats_update_begin(&stat
->syncp
);
377 u64_stats_update_end(&stat
->syncp
);
381 * blkg_stat_read - read the current value of a blkg_stat
382 * @stat: blkg_stat to read
384 * Read the current value of @stat. This function can be called without
385 * synchroniztion and takes care of u64 atomicity.
387 static inline uint64_t blkg_stat_read(struct blkg_stat
*stat
)
393 start
= u64_stats_fetch_begin(&stat
->syncp
);
395 } while (u64_stats_fetch_retry(&stat
->syncp
, start
));
401 * blkg_stat_reset - reset a blkg_stat
402 * @stat: blkg_stat to reset
404 static inline void blkg_stat_reset(struct blkg_stat
*stat
)
410 * blkg_rwstat_add - add a value to a blkg_rwstat
411 * @rwstat: target blkg_rwstat
412 * @rw: mask of REQ_{WRITE|SYNC}
415 * Add @val to @rwstat. The counters are chosen according to @rw. The
416 * caller is responsible for synchronizing calls to this function.
418 static inline void blkg_rwstat_add(struct blkg_rwstat
*rwstat
,
419 int rw
, uint64_t val
)
421 u64_stats_update_begin(&rwstat
->syncp
);
424 rwstat
->cnt
[BLKG_RWSTAT_WRITE
] += val
;
426 rwstat
->cnt
[BLKG_RWSTAT_READ
] += val
;
428 rwstat
->cnt
[BLKG_RWSTAT_SYNC
] += val
;
430 rwstat
->cnt
[BLKG_RWSTAT_ASYNC
] += val
;
432 u64_stats_update_end(&rwstat
->syncp
);
436 * blkg_rwstat_read - read the current values of a blkg_rwstat
437 * @rwstat: blkg_rwstat to read
439 * Read the current snapshot of @rwstat and return it as the return value.
440 * This function can be called without synchronization and takes care of
443 static inline struct blkg_rwstat
blkg_rwstat_read(struct blkg_rwstat
*rwstat
)
446 struct blkg_rwstat tmp
;
449 start
= u64_stats_fetch_begin(&rwstat
->syncp
);
451 } while (u64_stats_fetch_retry(&rwstat
->syncp
, start
));
457 * blkg_rwstat_sum - read the total count of a blkg_rwstat
458 * @rwstat: blkg_rwstat to read
460 * Return the total count of @rwstat regardless of the IO direction. This
461 * function can be called without synchronization and takes care of u64
464 static inline uint64_t blkg_rwstat_sum(struct blkg_rwstat
*rwstat
)
466 struct blkg_rwstat tmp
= blkg_rwstat_read(rwstat
);
468 return tmp
.cnt
[BLKG_RWSTAT_READ
] + tmp
.cnt
[BLKG_RWSTAT_WRITE
];
472 * blkg_rwstat_reset - reset a blkg_rwstat
473 * @rwstat: blkg_rwstat to reset
475 static inline void blkg_rwstat_reset(struct blkg_rwstat
*rwstat
)
477 memset(rwstat
->cnt
, 0, sizeof(rwstat
->cnt
));
480 #else /* CONFIG_BLK_CGROUP */
485 struct blkg_policy_data
{
491 struct blkcg_policy
{
494 static inline struct blkcg_gq
*blkg_lookup(struct blkcg
*blkcg
, void *key
) { return NULL
; }
495 static inline int blkcg_init_queue(struct request_queue
*q
) { return 0; }
496 static inline void blkcg_drain_queue(struct request_queue
*q
) { }
497 static inline void blkcg_exit_queue(struct request_queue
*q
) { }
498 static inline int blkcg_policy_register(struct blkcg_policy
*pol
) { return 0; }
499 static inline void blkcg_policy_unregister(struct blkcg_policy
*pol
) { }
500 static inline int blkcg_activate_policy(struct request_queue
*q
,
501 const struct blkcg_policy
*pol
) { return 0; }
502 static inline void blkcg_deactivate_policy(struct request_queue
*q
,
503 const struct blkcg_policy
*pol
) { }
505 static inline struct blkcg
*cgroup_to_blkcg(struct cgroup
*cgroup
) { return NULL
; }
506 static inline struct blkcg
*bio_blkcg(struct bio
*bio
) { return NULL
; }
508 static inline struct blkg_policy_data
*blkg_to_pd(struct blkcg_gq
*blkg
,
509 struct blkcg_policy
*pol
) { return NULL
; }
510 static inline struct blkcg_gq
*pd_to_blkg(struct blkg_policy_data
*pd
) { return NULL
; }
511 static inline char *blkg_path(struct blkcg_gq
*blkg
) { return NULL
; }
512 static inline void blkg_get(struct blkcg_gq
*blkg
) { }
513 static inline void blkg_put(struct blkcg_gq
*blkg
) { }
515 static inline struct request_list
*blk_get_rl(struct request_queue
*q
,
516 struct bio
*bio
) { return &q
->root_rl
; }
517 static inline void blk_put_rl(struct request_list
*rl
) { }
518 static inline void blk_rq_set_rl(struct request
*rq
, struct request_list
*rl
) { }
519 static inline struct request_list
*blk_rq_rl(struct request
*rq
) { return &rq
->q
->root_rl
; }
521 #define blk_queue_for_each_rl(rl, q) \
522 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
524 #endif /* CONFIG_BLK_CGROUP */
525 #endif /* _BLK_CGROUP_H */