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 #ifdef CONFIG_CGROUP_WRITEBACK
51 struct list_head cgwb_list
;
56 struct u64_stats_sync syncp
;
61 struct u64_stats_sync syncp
;
62 uint64_t cnt
[BLKG_RWSTAT_NR
];
66 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
67 * request_queue (q). This is used by blkcg policies which need to track
68 * information per blkcg - q pair.
70 * There can be multiple active blkcg policies and each has its private
71 * data on each blkg, the size of which is determined by
72 * blkcg_policy->pd_size. blkcg core allocates and frees such areas
73 * together with blkg and invokes pd_init/exit_fn() methods.
75 * Such private data must embed struct blkg_policy_data (pd) at the
76 * beginning and pd_size can't be smaller than pd.
78 struct blkg_policy_data
{
79 /* the blkg and policy id this per-policy data belongs to */
80 struct blkcg_gq
*blkg
;
83 /* used during policy activation */
84 struct list_head alloc_node
;
88 * Policies that need to keep per-blkcg data which is independent
89 * from any request_queue associated to it must specify its size
90 * with the cpd_size field of the blkcg_policy structure and
91 * embed a blkcg_policy_data in it. blkcg core allocates
92 * policy-specific per-blkcg structures lazily the first time
93 * they are actually needed, so it handles them together with
94 * blkgs. cpd_init() is invoked to let each policy handle
97 struct blkcg_policy_data
{
98 /* the policy id this per-policy data belongs to */
101 /* used during policy activation */
102 struct list_head alloc_node
;
105 /* association between a blk cgroup and a request queue */
107 /* Pointer to the associated request_queue */
108 struct request_queue
*q
;
109 struct list_head q_node
;
110 struct hlist_node blkcg_node
;
114 * Each blkg gets congested separately and the congestion state is
115 * propagated to the matching bdi_writeback_congested.
117 struct bdi_writeback_congested
*wb_congested
;
119 /* all non-root blkcg_gq's are guaranteed to have access to parent */
120 struct blkcg_gq
*parent
;
122 /* request allocation list for this blkcg-q pair */
123 struct request_list rl
;
125 /* reference count */
128 /* is this blkg online? protected by both blkcg and q locks */
131 struct blkg_policy_data
*pd
[BLKCG_MAX_POLS
];
133 struct rcu_head rcu_head
;
136 typedef void (blkcg_pol_init_cpd_fn
)(const struct blkcg
*blkcg
);
137 typedef void (blkcg_pol_init_pd_fn
)(struct blkcg_gq
*blkg
);
138 typedef void (blkcg_pol_online_pd_fn
)(struct blkcg_gq
*blkg
);
139 typedef void (blkcg_pol_offline_pd_fn
)(struct blkcg_gq
*blkg
);
140 typedef void (blkcg_pol_exit_pd_fn
)(struct blkcg_gq
*blkg
);
141 typedef void (blkcg_pol_reset_pd_stats_fn
)(struct blkcg_gq
*blkg
);
143 struct blkcg_policy
{
145 /* policy specific private data size */
147 /* policy specific per-blkcg data size */
149 /* cgroup files for the policy */
150 struct cftype
*cftypes
;
153 blkcg_pol_init_cpd_fn
*cpd_init_fn
;
154 blkcg_pol_init_pd_fn
*pd_init_fn
;
155 blkcg_pol_online_pd_fn
*pd_online_fn
;
156 blkcg_pol_offline_pd_fn
*pd_offline_fn
;
157 blkcg_pol_exit_pd_fn
*pd_exit_fn
;
158 blkcg_pol_reset_pd_stats_fn
*pd_reset_stats_fn
;
161 extern struct blkcg blkcg_root
;
162 extern struct cgroup_subsys_state
* const blkcg_root_css
;
164 struct blkcg_gq
*blkg_lookup(struct blkcg
*blkcg
, struct request_queue
*q
);
165 struct blkcg_gq
*blkg_lookup_create(struct blkcg
*blkcg
,
166 struct request_queue
*q
);
167 int blkcg_init_queue(struct request_queue
*q
);
168 void blkcg_drain_queue(struct request_queue
*q
);
169 void blkcg_exit_queue(struct request_queue
*q
);
171 /* Blkio controller policy registration */
172 int blkcg_policy_register(struct blkcg_policy
*pol
);
173 void blkcg_policy_unregister(struct blkcg_policy
*pol
);
174 int blkcg_activate_policy(struct request_queue
*q
,
175 const struct blkcg_policy
*pol
);
176 void blkcg_deactivate_policy(struct request_queue
*q
,
177 const struct blkcg_policy
*pol
);
179 void blkcg_print_blkgs(struct seq_file
*sf
, struct blkcg
*blkcg
,
180 u64 (*prfill
)(struct seq_file
*,
181 struct blkg_policy_data
*, int),
182 const struct blkcg_policy
*pol
, int data
,
184 u64
__blkg_prfill_u64(struct seq_file
*sf
, struct blkg_policy_data
*pd
, u64 v
);
185 u64
__blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
186 const struct blkg_rwstat
*rwstat
);
187 u64
blkg_prfill_stat(struct seq_file
*sf
, struct blkg_policy_data
*pd
, int off
);
188 u64
blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
191 u64
blkg_stat_recursive_sum(struct blkg_policy_data
*pd
, int off
);
192 struct blkg_rwstat
blkg_rwstat_recursive_sum(struct blkg_policy_data
*pd
,
195 struct blkg_conf_ctx
{
196 struct gendisk
*disk
;
197 struct blkcg_gq
*blkg
;
201 int blkg_conf_prep(struct blkcg
*blkcg
, const struct blkcg_policy
*pol
,
202 const char *input
, struct blkg_conf_ctx
*ctx
);
203 void blkg_conf_finish(struct blkg_conf_ctx
*ctx
);
206 static inline struct blkcg
*css_to_blkcg(struct cgroup_subsys_state
*css
)
208 return css
? container_of(css
, struct blkcg
, css
) : NULL
;
211 static inline struct blkcg
*task_blkcg(struct task_struct
*tsk
)
213 return css_to_blkcg(task_css(tsk
, blkio_cgrp_id
));
216 static inline struct blkcg
*bio_blkcg(struct bio
*bio
)
218 if (bio
&& bio
->bi_css
)
219 return css_to_blkcg(bio
->bi_css
);
220 return task_blkcg(current
);
223 static inline struct cgroup_subsys_state
*
224 task_get_blkcg_css(struct task_struct
*task
)
226 return task_get_css(task
, blkio_cgrp_id
);
230 * blkcg_parent - get the parent of a blkcg
231 * @blkcg: blkcg of interest
233 * Return the parent blkcg of @blkcg. Can be called anytime.
235 static inline struct blkcg
*blkcg_parent(struct blkcg
*blkcg
)
237 return css_to_blkcg(blkcg
->css
.parent
);
241 * blkg_to_pdata - get policy private data
242 * @blkg: blkg of interest
243 * @pol: policy of interest
245 * Return pointer to private data associated with the @blkg-@pol pair.
247 static inline struct blkg_policy_data
*blkg_to_pd(struct blkcg_gq
*blkg
,
248 struct blkcg_policy
*pol
)
250 return blkg
? blkg
->pd
[pol
->plid
] : NULL
;
253 static inline struct blkcg_policy_data
*blkcg_to_cpd(struct blkcg
*blkcg
,
254 struct blkcg_policy
*pol
)
256 return blkcg
? blkcg
->pd
[pol
->plid
] : NULL
;
260 * pdata_to_blkg - get blkg associated with policy private data
261 * @pd: policy private data of interest
263 * @pd is policy private data. Determine the blkg it's associated with.
265 static inline struct blkcg_gq
*pd_to_blkg(struct blkg_policy_data
*pd
)
267 return pd
? pd
->blkg
: NULL
;
271 * blkg_path - format cgroup path of blkg
272 * @blkg: blkg of interest
273 * @buf: target buffer
274 * @buflen: target buffer length
276 * Format the path of the cgroup of @blkg into @buf.
278 static inline int blkg_path(struct blkcg_gq
*blkg
, char *buf
, int buflen
)
282 p
= cgroup_path(blkg
->blkcg
->css
.cgroup
, buf
, buflen
);
284 strncpy(buf
, "<unavailable>", buflen
);
285 return -ENAMETOOLONG
;
288 memmove(buf
, p
, buf
+ buflen
- p
);
293 * blkg_get - get a blkg reference
296 * The caller should be holding an existing reference.
298 static inline void blkg_get(struct blkcg_gq
*blkg
)
300 WARN_ON_ONCE(atomic_read(&blkg
->refcnt
) <= 0);
301 atomic_inc(&blkg
->refcnt
);
304 void __blkg_release_rcu(struct rcu_head
*rcu
);
307 * blkg_put - put a blkg reference
310 static inline void blkg_put(struct blkcg_gq
*blkg
)
312 WARN_ON_ONCE(atomic_read(&blkg
->refcnt
) <= 0);
313 if (atomic_dec_and_test(&blkg
->refcnt
))
314 call_rcu(&blkg
->rcu_head
, __blkg_release_rcu
);
317 struct blkcg_gq
*__blkg_lookup(struct blkcg
*blkcg
, struct request_queue
*q
,
321 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
322 * @d_blkg: loop cursor pointing to the current descendant
323 * @pos_css: used for iteration
324 * @p_blkg: target blkg to walk descendants of
326 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
327 * read locked. If called under either blkcg or queue lock, the iteration
328 * is guaranteed to include all and only online blkgs. The caller may
329 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
330 * @p_blkg is included in the iteration and the first node to be visited.
332 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
333 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
334 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
335 (p_blkg)->q, false)))
338 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
339 * @d_blkg: loop cursor pointing to the current descendant
340 * @pos_css: used for iteration
341 * @p_blkg: target blkg to walk descendants of
343 * Similar to blkg_for_each_descendant_pre() but performs post-order
344 * traversal instead. Synchronization rules are the same. @p_blkg is
345 * included in the iteration and the last node to be visited.
347 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
348 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
349 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
350 (p_blkg)->q, false)))
353 * blk_get_rl - get request_list to use
354 * @q: request_queue of interest
355 * @bio: bio which will be attached to the allocated request (may be %NULL)
357 * The caller wants to allocate a request from @q to use for @bio. Find
358 * the request_list to use and obtain a reference on it. Should be called
359 * under queue_lock. This function is guaranteed to return non-%NULL
362 static inline struct request_list
*blk_get_rl(struct request_queue
*q
,
366 struct blkcg_gq
*blkg
;
370 blkcg
= bio_blkcg(bio
);
372 /* bypass blkg lookup and use @q->root_rl directly for root */
373 if (blkcg
== &blkcg_root
)
377 * Try to use blkg->rl. blkg lookup may fail under memory pressure
378 * or if either the blkcg or queue is going away. Fall back to
379 * root_rl in such cases.
381 blkg
= blkg_lookup_create(blkcg
, q
);
382 if (unlikely(IS_ERR(blkg
)))
394 * blk_put_rl - put request_list
395 * @rl: request_list to put
397 * Put the reference acquired by blk_get_rl(). Should be called under
400 static inline void blk_put_rl(struct request_list
*rl
)
402 /* root_rl may not have blkg set */
403 if (rl
->blkg
&& rl
->blkg
->blkcg
!= &blkcg_root
)
408 * blk_rq_set_rl - associate a request with a request_list
409 * @rq: request of interest
410 * @rl: target request_list
412 * Associate @rq with @rl so that accounting and freeing can know the
413 * request_list @rq came from.
415 static inline void blk_rq_set_rl(struct request
*rq
, struct request_list
*rl
)
421 * blk_rq_rl - return the request_list a request came from
422 * @rq: request of interest
424 * Return the request_list @rq is allocated from.
426 static inline struct request_list
*blk_rq_rl(struct request
*rq
)
431 struct request_list
*__blk_queue_next_rl(struct request_list
*rl
,
432 struct request_queue
*q
);
434 * blk_queue_for_each_rl - iterate through all request_lists of a request_queue
436 * Should be used under queue_lock.
438 #define blk_queue_for_each_rl(rl, q) \
439 for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
441 static inline void blkg_stat_init(struct blkg_stat
*stat
)
443 u64_stats_init(&stat
->syncp
);
447 * blkg_stat_add - add a value to a blkg_stat
448 * @stat: target blkg_stat
451 * Add @val to @stat. The caller is responsible for synchronizing calls to
454 static inline void blkg_stat_add(struct blkg_stat
*stat
, uint64_t val
)
456 u64_stats_update_begin(&stat
->syncp
);
458 u64_stats_update_end(&stat
->syncp
);
462 * blkg_stat_read - read the current value of a blkg_stat
463 * @stat: blkg_stat to read
465 * Read the current value of @stat. This function can be called without
466 * synchroniztion and takes care of u64 atomicity.
468 static inline uint64_t blkg_stat_read(struct blkg_stat
*stat
)
474 start
= u64_stats_fetch_begin_irq(&stat
->syncp
);
476 } while (u64_stats_fetch_retry_irq(&stat
->syncp
, start
));
482 * blkg_stat_reset - reset a blkg_stat
483 * @stat: blkg_stat to reset
485 static inline void blkg_stat_reset(struct blkg_stat
*stat
)
491 * blkg_stat_merge - merge a blkg_stat into another
492 * @to: the destination blkg_stat
495 * Add @from's count to @to.
497 static inline void blkg_stat_merge(struct blkg_stat
*to
, struct blkg_stat
*from
)
499 blkg_stat_add(to
, blkg_stat_read(from
));
502 static inline void blkg_rwstat_init(struct blkg_rwstat
*rwstat
)
504 u64_stats_init(&rwstat
->syncp
);
508 * blkg_rwstat_add - add a value to a blkg_rwstat
509 * @rwstat: target blkg_rwstat
510 * @rw: mask of REQ_{WRITE|SYNC}
513 * Add @val to @rwstat. The counters are chosen according to @rw. The
514 * caller is responsible for synchronizing calls to this function.
516 static inline void blkg_rwstat_add(struct blkg_rwstat
*rwstat
,
517 int rw
, uint64_t val
)
519 u64_stats_update_begin(&rwstat
->syncp
);
522 rwstat
->cnt
[BLKG_RWSTAT_WRITE
] += val
;
524 rwstat
->cnt
[BLKG_RWSTAT_READ
] += val
;
526 rwstat
->cnt
[BLKG_RWSTAT_SYNC
] += val
;
528 rwstat
->cnt
[BLKG_RWSTAT_ASYNC
] += val
;
530 u64_stats_update_end(&rwstat
->syncp
);
534 * blkg_rwstat_read - read the current values of a blkg_rwstat
535 * @rwstat: blkg_rwstat to read
537 * Read the current snapshot of @rwstat and return it as the return value.
538 * This function can be called without synchronization and takes care of
541 static inline struct blkg_rwstat
blkg_rwstat_read(struct blkg_rwstat
*rwstat
)
544 struct blkg_rwstat tmp
;
547 start
= u64_stats_fetch_begin_irq(&rwstat
->syncp
);
549 } while (u64_stats_fetch_retry_irq(&rwstat
->syncp
, start
));
555 * blkg_rwstat_total - read the total count of a blkg_rwstat
556 * @rwstat: blkg_rwstat to read
558 * Return the total count of @rwstat regardless of the IO direction. This
559 * function can be called without synchronization and takes care of u64
562 static inline uint64_t blkg_rwstat_total(struct blkg_rwstat
*rwstat
)
564 struct blkg_rwstat tmp
= blkg_rwstat_read(rwstat
);
566 return tmp
.cnt
[BLKG_RWSTAT_READ
] + tmp
.cnt
[BLKG_RWSTAT_WRITE
];
570 * blkg_rwstat_reset - reset a blkg_rwstat
571 * @rwstat: blkg_rwstat to reset
573 static inline void blkg_rwstat_reset(struct blkg_rwstat
*rwstat
)
575 memset(rwstat
->cnt
, 0, sizeof(rwstat
->cnt
));
579 * blkg_rwstat_merge - merge a blkg_rwstat into another
580 * @to: the destination blkg_rwstat
583 * Add @from's counts to @to.
585 static inline void blkg_rwstat_merge(struct blkg_rwstat
*to
,
586 struct blkg_rwstat
*from
)
588 struct blkg_rwstat v
= blkg_rwstat_read(from
);
591 u64_stats_update_begin(&to
->syncp
);
592 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
593 to
->cnt
[i
] += v
.cnt
[i
];
594 u64_stats_update_end(&to
->syncp
);
597 #else /* CONFIG_BLK_CGROUP */
602 struct blkg_policy_data
{
605 struct blkcg_policy_data
{
611 struct blkcg_policy
{
614 #define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
616 static inline struct cgroup_subsys_state
*
617 task_get_blkcg_css(struct task_struct
*task
)
624 static inline struct blkcg_gq
*blkg_lookup(struct blkcg
*blkcg
, void *key
) { return NULL
; }
625 static inline int blkcg_init_queue(struct request_queue
*q
) { return 0; }
626 static inline void blkcg_drain_queue(struct request_queue
*q
) { }
627 static inline void blkcg_exit_queue(struct request_queue
*q
) { }
628 static inline int blkcg_policy_register(struct blkcg_policy
*pol
) { return 0; }
629 static inline void blkcg_policy_unregister(struct blkcg_policy
*pol
) { }
630 static inline int blkcg_activate_policy(struct request_queue
*q
,
631 const struct blkcg_policy
*pol
) { return 0; }
632 static inline void blkcg_deactivate_policy(struct request_queue
*q
,
633 const struct blkcg_policy
*pol
) { }
635 static inline struct blkcg
*bio_blkcg(struct bio
*bio
) { return NULL
; }
637 static inline struct blkg_policy_data
*blkg_to_pd(struct blkcg_gq
*blkg
,
638 struct blkcg_policy
*pol
) { return NULL
; }
639 static inline struct blkcg_gq
*pd_to_blkg(struct blkg_policy_data
*pd
) { return NULL
; }
640 static inline char *blkg_path(struct blkcg_gq
*blkg
) { return NULL
; }
641 static inline void blkg_get(struct blkcg_gq
*blkg
) { }
642 static inline void blkg_put(struct blkcg_gq
*blkg
) { }
644 static inline struct request_list
*blk_get_rl(struct request_queue
*q
,
645 struct bio
*bio
) { return &q
->root_rl
; }
646 static inline void blk_put_rl(struct request_list
*rl
) { }
647 static inline void blk_rq_set_rl(struct request
*rq
, struct request_list
*rl
) { }
648 static inline struct request_list
*blk_rq_rl(struct request
*rq
) { return &rq
->q
->root_rl
; }
650 #define blk_queue_for_each_rl(rl, q) \
651 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
653 #endif /* CONFIG_BLOCK */
654 #endif /* CONFIG_BLK_CGROUP */
655 #endif /* _BLK_CGROUP_H */