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 has its private
72 * data on each blkg, the size of which is determined by
73 * blkcg_policy->pd_size. blkcg core allocates and frees such areas
74 * together with blkg and invokes pd_init/exit_fn() methods.
76 * Such private data must embed struct blkg_policy_data (pd) at the
77 * beginning and pd_size can't be smaller than pd.
79 struct blkg_policy_data
{
80 /* the blkg and policy id this per-policy data belongs to */
81 struct blkcg_gq
*blkg
;
86 * Policies that need to keep per-blkcg data which is independent
87 * from any request_queue associated to it must specify its size
88 * with the cpd_size field of the blkcg_policy structure and
89 * embed a blkcg_policy_data in it. cpd_init() is invoked to let
90 * each policy handle per-blkcg data.
92 struct blkcg_policy_data
{
93 /* the policy id this per-policy data belongs to */
97 /* association between a blk cgroup and a request queue */
99 /* Pointer to the associated request_queue */
100 struct request_queue
*q
;
101 struct list_head q_node
;
102 struct hlist_node blkcg_node
;
106 * Each blkg gets congested separately and the congestion state is
107 * propagated to the matching bdi_writeback_congested.
109 struct bdi_writeback_congested
*wb_congested
;
111 /* all non-root blkcg_gq's are guaranteed to have access to parent */
112 struct blkcg_gq
*parent
;
114 /* request allocation list for this blkcg-q pair */
115 struct request_list rl
;
117 /* reference count */
120 /* is this blkg online? protected by both blkcg and q locks */
123 struct blkg_policy_data
*pd
[BLKCG_MAX_POLS
];
125 struct rcu_head rcu_head
;
128 typedef void (blkcg_pol_init_cpd_fn
)(const struct blkcg
*blkcg
);
129 typedef void (blkcg_pol_init_pd_fn
)(struct blkcg_gq
*blkg
);
130 typedef void (blkcg_pol_online_pd_fn
)(struct blkcg_gq
*blkg
);
131 typedef void (blkcg_pol_offline_pd_fn
)(struct blkcg_gq
*blkg
);
132 typedef void (blkcg_pol_exit_pd_fn
)(struct blkcg_gq
*blkg
);
133 typedef void (blkcg_pol_reset_pd_stats_fn
)(struct blkcg_gq
*blkg
);
135 struct blkcg_policy
{
137 /* policy specific private data size */
139 /* policy specific per-blkcg data size */
141 /* cgroup files for the policy */
142 struct cftype
*cftypes
;
145 blkcg_pol_init_cpd_fn
*cpd_init_fn
;
146 blkcg_pol_init_pd_fn
*pd_init_fn
;
147 blkcg_pol_online_pd_fn
*pd_online_fn
;
148 blkcg_pol_offline_pd_fn
*pd_offline_fn
;
149 blkcg_pol_exit_pd_fn
*pd_exit_fn
;
150 blkcg_pol_reset_pd_stats_fn
*pd_reset_stats_fn
;
153 extern struct blkcg blkcg_root
;
154 extern struct cgroup_subsys_state
* const blkcg_root_css
;
156 struct blkcg_gq
*blkg_lookup(struct blkcg
*blkcg
, struct request_queue
*q
);
157 struct blkcg_gq
*blkg_lookup_create(struct blkcg
*blkcg
,
158 struct request_queue
*q
);
159 int blkcg_init_queue(struct request_queue
*q
);
160 void blkcg_drain_queue(struct request_queue
*q
);
161 void blkcg_exit_queue(struct request_queue
*q
);
163 /* Blkio controller policy registration */
164 int blkcg_policy_register(struct blkcg_policy
*pol
);
165 void blkcg_policy_unregister(struct blkcg_policy
*pol
);
166 int blkcg_activate_policy(struct request_queue
*q
,
167 const struct blkcg_policy
*pol
);
168 void blkcg_deactivate_policy(struct request_queue
*q
,
169 const struct blkcg_policy
*pol
);
171 void blkcg_print_blkgs(struct seq_file
*sf
, struct blkcg
*blkcg
,
172 u64 (*prfill
)(struct seq_file
*,
173 struct blkg_policy_data
*, int),
174 const struct blkcg_policy
*pol
, int data
,
176 u64
__blkg_prfill_u64(struct seq_file
*sf
, struct blkg_policy_data
*pd
, u64 v
);
177 u64
__blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
178 const struct blkg_rwstat
*rwstat
);
179 u64
blkg_prfill_stat(struct seq_file
*sf
, struct blkg_policy_data
*pd
, int off
);
180 u64
blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
183 u64
blkg_stat_recursive_sum(struct blkg_policy_data
*pd
, int off
);
184 struct blkg_rwstat
blkg_rwstat_recursive_sum(struct blkg_policy_data
*pd
,
187 struct blkg_conf_ctx
{
188 struct gendisk
*disk
;
189 struct blkcg_gq
*blkg
;
193 int blkg_conf_prep(struct blkcg
*blkcg
, const struct blkcg_policy
*pol
,
194 const char *input
, struct blkg_conf_ctx
*ctx
);
195 void blkg_conf_finish(struct blkg_conf_ctx
*ctx
);
198 static inline struct blkcg
*css_to_blkcg(struct cgroup_subsys_state
*css
)
200 return css
? container_of(css
, struct blkcg
, css
) : NULL
;
203 static inline struct blkcg
*task_blkcg(struct task_struct
*tsk
)
205 return css_to_blkcg(task_css(tsk
, blkio_cgrp_id
));
208 static inline struct blkcg
*bio_blkcg(struct bio
*bio
)
210 if (bio
&& bio
->bi_css
)
211 return css_to_blkcg(bio
->bi_css
);
212 return task_blkcg(current
);
215 static inline struct cgroup_subsys_state
*
216 task_get_blkcg_css(struct task_struct
*task
)
218 return task_get_css(task
, blkio_cgrp_id
);
222 * blkcg_parent - get the parent of a blkcg
223 * @blkcg: blkcg of interest
225 * Return the parent blkcg of @blkcg. Can be called anytime.
227 static inline struct blkcg
*blkcg_parent(struct blkcg
*blkcg
)
229 return css_to_blkcg(blkcg
->css
.parent
);
233 * blkg_to_pdata - get policy private data
234 * @blkg: blkg of interest
235 * @pol: policy of interest
237 * Return pointer to private data associated with the @blkg-@pol pair.
239 static inline struct blkg_policy_data
*blkg_to_pd(struct blkcg_gq
*blkg
,
240 struct blkcg_policy
*pol
)
242 return blkg
? blkg
->pd
[pol
->plid
] : NULL
;
245 static inline struct blkcg_policy_data
*blkcg_to_cpd(struct blkcg
*blkcg
,
246 struct blkcg_policy
*pol
)
248 return blkcg
? blkcg
->pd
[pol
->plid
] : NULL
;
252 * pdata_to_blkg - get blkg associated with policy private data
253 * @pd: policy private data of interest
255 * @pd is policy private data. Determine the blkg it's associated with.
257 static inline struct blkcg_gq
*pd_to_blkg(struct blkg_policy_data
*pd
)
259 return pd
? pd
->blkg
: NULL
;
263 * blkg_path - format cgroup path of blkg
264 * @blkg: blkg of interest
265 * @buf: target buffer
266 * @buflen: target buffer length
268 * Format the path of the cgroup of @blkg into @buf.
270 static inline int blkg_path(struct blkcg_gq
*blkg
, char *buf
, int buflen
)
274 p
= cgroup_path(blkg
->blkcg
->css
.cgroup
, buf
, buflen
);
276 strncpy(buf
, "<unavailable>", buflen
);
277 return -ENAMETOOLONG
;
280 memmove(buf
, p
, buf
+ buflen
- p
);
285 * blkg_get - get a blkg reference
288 * The caller should be holding an existing reference.
290 static inline void blkg_get(struct blkcg_gq
*blkg
)
292 WARN_ON_ONCE(atomic_read(&blkg
->refcnt
) <= 0);
293 atomic_inc(&blkg
->refcnt
);
296 void __blkg_release_rcu(struct rcu_head
*rcu
);
299 * blkg_put - put a blkg reference
302 static inline void blkg_put(struct blkcg_gq
*blkg
)
304 WARN_ON_ONCE(atomic_read(&blkg
->refcnt
) <= 0);
305 if (atomic_dec_and_test(&blkg
->refcnt
))
306 call_rcu(&blkg
->rcu_head
, __blkg_release_rcu
);
309 struct blkcg_gq
*__blkg_lookup(struct blkcg
*blkcg
, struct request_queue
*q
,
313 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
314 * @d_blkg: loop cursor pointing to the current descendant
315 * @pos_css: used for iteration
316 * @p_blkg: target blkg to walk descendants of
318 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
319 * read locked. If called under either blkcg or queue lock, the iteration
320 * is guaranteed to include all and only online blkgs. The caller may
321 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
322 * @p_blkg is included in the iteration and the first node to be visited.
324 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
325 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
326 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
327 (p_blkg)->q, false)))
330 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
331 * @d_blkg: loop cursor pointing to the current descendant
332 * @pos_css: used for iteration
333 * @p_blkg: target blkg to walk descendants of
335 * Similar to blkg_for_each_descendant_pre() but performs post-order
336 * traversal instead. Synchronization rules are the same. @p_blkg is
337 * included in the iteration and the last node to be visited.
339 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
340 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
341 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
342 (p_blkg)->q, false)))
345 * blk_get_rl - get request_list to use
346 * @q: request_queue of interest
347 * @bio: bio which will be attached to the allocated request (may be %NULL)
349 * The caller wants to allocate a request from @q to use for @bio. Find
350 * the request_list to use and obtain a reference on it. Should be called
351 * under queue_lock. This function is guaranteed to return non-%NULL
354 static inline struct request_list
*blk_get_rl(struct request_queue
*q
,
358 struct blkcg_gq
*blkg
;
362 blkcg
= bio_blkcg(bio
);
364 /* bypass blkg lookup and use @q->root_rl directly for root */
365 if (blkcg
== &blkcg_root
)
369 * Try to use blkg->rl. blkg lookup may fail under memory pressure
370 * or if either the blkcg or queue is going away. Fall back to
371 * root_rl in such cases.
373 blkg
= blkg_lookup_create(blkcg
, q
);
374 if (unlikely(IS_ERR(blkg
)))
386 * blk_put_rl - put request_list
387 * @rl: request_list to put
389 * Put the reference acquired by blk_get_rl(). Should be called under
392 static inline void blk_put_rl(struct request_list
*rl
)
394 if (rl
->blkg
->blkcg
!= &blkcg_root
)
399 * blk_rq_set_rl - associate a request with a request_list
400 * @rq: request of interest
401 * @rl: target request_list
403 * Associate @rq with @rl so that accounting and freeing can know the
404 * request_list @rq came from.
406 static inline void blk_rq_set_rl(struct request
*rq
, struct request_list
*rl
)
412 * blk_rq_rl - return the request_list a request came from
413 * @rq: request of interest
415 * Return the request_list @rq is allocated from.
417 static inline struct request_list
*blk_rq_rl(struct request
*rq
)
422 struct request_list
*__blk_queue_next_rl(struct request_list
*rl
,
423 struct request_queue
*q
);
425 * blk_queue_for_each_rl - iterate through all request_lists of a request_queue
427 * Should be used under queue_lock.
429 #define blk_queue_for_each_rl(rl, q) \
430 for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
432 static inline void blkg_stat_init(struct blkg_stat
*stat
)
434 u64_stats_init(&stat
->syncp
);
438 * blkg_stat_add - add a value to a blkg_stat
439 * @stat: target blkg_stat
442 * Add @val to @stat. The caller is responsible for synchronizing calls to
445 static inline void blkg_stat_add(struct blkg_stat
*stat
, uint64_t val
)
447 u64_stats_update_begin(&stat
->syncp
);
449 u64_stats_update_end(&stat
->syncp
);
453 * blkg_stat_read - read the current value of a blkg_stat
454 * @stat: blkg_stat to read
456 * Read the current value of @stat. This function can be called without
457 * synchroniztion and takes care of u64 atomicity.
459 static inline uint64_t blkg_stat_read(struct blkg_stat
*stat
)
465 start
= u64_stats_fetch_begin_irq(&stat
->syncp
);
467 } while (u64_stats_fetch_retry_irq(&stat
->syncp
, start
));
473 * blkg_stat_reset - reset a blkg_stat
474 * @stat: blkg_stat to reset
476 static inline void blkg_stat_reset(struct blkg_stat
*stat
)
482 * blkg_stat_merge - merge a blkg_stat into another
483 * @to: the destination blkg_stat
486 * Add @from's count to @to.
488 static inline void blkg_stat_merge(struct blkg_stat
*to
, struct blkg_stat
*from
)
490 blkg_stat_add(to
, blkg_stat_read(from
));
493 static inline void blkg_rwstat_init(struct blkg_rwstat
*rwstat
)
495 u64_stats_init(&rwstat
->syncp
);
499 * blkg_rwstat_add - add a value to a blkg_rwstat
500 * @rwstat: target blkg_rwstat
501 * @rw: mask of REQ_{WRITE|SYNC}
504 * Add @val to @rwstat. The counters are chosen according to @rw. The
505 * caller is responsible for synchronizing calls to this function.
507 static inline void blkg_rwstat_add(struct blkg_rwstat
*rwstat
,
508 int rw
, uint64_t val
)
510 u64_stats_update_begin(&rwstat
->syncp
);
513 rwstat
->cnt
[BLKG_RWSTAT_WRITE
] += val
;
515 rwstat
->cnt
[BLKG_RWSTAT_READ
] += val
;
517 rwstat
->cnt
[BLKG_RWSTAT_SYNC
] += val
;
519 rwstat
->cnt
[BLKG_RWSTAT_ASYNC
] += val
;
521 u64_stats_update_end(&rwstat
->syncp
);
525 * blkg_rwstat_read - read the current values of a blkg_rwstat
526 * @rwstat: blkg_rwstat to read
528 * Read the current snapshot of @rwstat and return it as the return value.
529 * This function can be called without synchronization and takes care of
532 static inline struct blkg_rwstat
blkg_rwstat_read(struct blkg_rwstat
*rwstat
)
535 struct blkg_rwstat tmp
;
538 start
= u64_stats_fetch_begin_irq(&rwstat
->syncp
);
540 } while (u64_stats_fetch_retry_irq(&rwstat
->syncp
, start
));
546 * blkg_rwstat_total - read the total count of a blkg_rwstat
547 * @rwstat: blkg_rwstat to read
549 * Return the total count of @rwstat regardless of the IO direction. This
550 * function can be called without synchronization and takes care of u64
553 static inline uint64_t blkg_rwstat_total(struct blkg_rwstat
*rwstat
)
555 struct blkg_rwstat tmp
= blkg_rwstat_read(rwstat
);
557 return tmp
.cnt
[BLKG_RWSTAT_READ
] + tmp
.cnt
[BLKG_RWSTAT_WRITE
];
561 * blkg_rwstat_reset - reset a blkg_rwstat
562 * @rwstat: blkg_rwstat to reset
564 static inline void blkg_rwstat_reset(struct blkg_rwstat
*rwstat
)
566 memset(rwstat
->cnt
, 0, sizeof(rwstat
->cnt
));
570 * blkg_rwstat_merge - merge a blkg_rwstat into another
571 * @to: the destination blkg_rwstat
574 * Add @from's counts to @to.
576 static inline void blkg_rwstat_merge(struct blkg_rwstat
*to
,
577 struct blkg_rwstat
*from
)
579 struct blkg_rwstat v
= blkg_rwstat_read(from
);
582 u64_stats_update_begin(&to
->syncp
);
583 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
584 to
->cnt
[i
] += v
.cnt
[i
];
585 u64_stats_update_end(&to
->syncp
);
588 #else /* CONFIG_BLK_CGROUP */
593 struct blkg_policy_data
{
596 struct blkcg_policy_data
{
602 struct blkcg_policy
{
605 #define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
607 static inline struct cgroup_subsys_state
*
608 task_get_blkcg_css(struct task_struct
*task
)
615 static inline struct blkcg_gq
*blkg_lookup(struct blkcg
*blkcg
, void *key
) { return NULL
; }
616 static inline int blkcg_init_queue(struct request_queue
*q
) { return 0; }
617 static inline void blkcg_drain_queue(struct request_queue
*q
) { }
618 static inline void blkcg_exit_queue(struct request_queue
*q
) { }
619 static inline int blkcg_policy_register(struct blkcg_policy
*pol
) { return 0; }
620 static inline void blkcg_policy_unregister(struct blkcg_policy
*pol
) { }
621 static inline int blkcg_activate_policy(struct request_queue
*q
,
622 const struct blkcg_policy
*pol
) { return 0; }
623 static inline void blkcg_deactivate_policy(struct request_queue
*q
,
624 const struct blkcg_policy
*pol
) { }
626 static inline struct blkcg
*bio_blkcg(struct bio
*bio
) { return NULL
; }
628 static inline struct blkg_policy_data
*blkg_to_pd(struct blkcg_gq
*blkg
,
629 struct blkcg_policy
*pol
) { return NULL
; }
630 static inline struct blkcg_gq
*pd_to_blkg(struct blkg_policy_data
*pd
) { return NULL
; }
631 static inline char *blkg_path(struct blkcg_gq
*blkg
) { return NULL
; }
632 static inline void blkg_get(struct blkcg_gq
*blkg
) { }
633 static inline void blkg_put(struct blkcg_gq
*blkg
) { }
635 static inline struct request_list
*blk_get_rl(struct request_queue
*q
,
636 struct bio
*bio
) { return &q
->root_rl
; }
637 static inline void blk_put_rl(struct request_list
*rl
) { }
638 static inline void blk_rq_set_rl(struct request
*rq
, struct request_list
*rl
) { }
639 static inline struct request_list
*blk_rq_rl(struct request
*rq
) { return &rq
->q
->root_rl
; }
641 #define blk_queue_for_each_rl(rl, q) \
642 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
644 #endif /* CONFIG_BLOCK */
645 #endif /* CONFIG_BLK_CGROUP */
646 #endif /* _BLK_CGROUP_H */