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 /* CFQ specific, out here for blkcg->cfq_weight */
27 #define CFQ_WEIGHT_MIN 10
28 #define CFQ_WEIGHT_MAX 1000
29 #define CFQ_WEIGHT_DEFAULT 500
31 #ifdef CONFIG_BLK_CGROUP
33 enum blkg_rwstat_type {
40 BLKG_RWSTAT_TOTAL = BLKG_RWSTAT_NR,
46 struct cgroup_subsys_state css;
49 struct radix_tree_root blkg_tree;
50 struct blkcg_gq *blkg_hint;
51 struct hlist_head blkg_list;
53 /* TODO: per-policy storage in blkcg */
54 unsigned int cfq_weight; /* belongs to cfq */
55 unsigned int cfq_leaf_weight;
59 struct u64_stats_sync syncp;
64 struct u64_stats_sync syncp;
65 uint64_t cnt[BLKG_RWSTAT_NR];
69 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
70 * request_queue (q). This is used by blkcg policies which need to track
71 * information per blkcg - q pair.
73 * There can be multiple active blkcg policies and each has its private
74 * data on each blkg, the size of which is determined by
75 * blkcg_policy->pd_size. blkcg core allocates and frees such areas
76 * together with blkg and invokes pd_init/exit_fn() methods.
78 * Such private data must embed struct blkg_policy_data (pd) at the
79 * beginning and pd_size can't be smaller than pd.
81 struct blkg_policy_data {
82 /* the blkg and policy id this per-policy data belongs to */
83 struct blkcg_gq *blkg;
86 /* used during policy activation */
87 struct list_head alloc_node;
90 /* association between a blk cgroup and a request queue */
92 /* Pointer to the associated request_queue */
93 struct request_queue *q;
94 struct list_head q_node;
95 struct hlist_node blkcg_node;
98 /* all non-root blkcg_gq's are guaranteed to have access to parent */
99 struct blkcg_gq *parent;
101 /* request allocation list for this blkcg-q pair */
102 struct request_list rl;
104 /* reference count */
107 /* is this blkg online? protected by both blkcg and q locks */
110 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
112 struct rcu_head rcu_head;
115 typedef void (blkcg_pol_init_pd_fn)(struct blkcg_gq *blkg);
116 typedef void (blkcg_pol_online_pd_fn)(struct blkcg_gq *blkg);
117 typedef void (blkcg_pol_offline_pd_fn)(struct blkcg_gq *blkg);
118 typedef void (blkcg_pol_exit_pd_fn)(struct blkcg_gq *blkg);
119 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkcg_gq *blkg);
121 struct blkcg_policy {
123 /* policy specific private data size */
125 /* cgroup files for the policy */
126 struct cftype *cftypes;
129 blkcg_pol_init_pd_fn *pd_init_fn;
130 blkcg_pol_online_pd_fn *pd_online_fn;
131 blkcg_pol_offline_pd_fn *pd_offline_fn;
132 blkcg_pol_exit_pd_fn *pd_exit_fn;
133 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;
136 extern struct blkcg blkcg_root;
138 struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q);
139 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
140 struct request_queue *q);
141 int blkcg_init_queue(struct request_queue *q);
142 void blkcg_drain_queue(struct request_queue *q);
143 void blkcg_exit_queue(struct request_queue *q);
145 /* Blkio controller policy registration */
146 int blkcg_policy_register(struct blkcg_policy *pol);
147 void blkcg_policy_unregister(struct blkcg_policy *pol);
148 int blkcg_activate_policy(struct request_queue *q,
149 const struct blkcg_policy *pol);
150 void blkcg_deactivate_policy(struct request_queue *q,
151 const struct blkcg_policy *pol);
153 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
154 u64 (*prfill)(struct seq_file *,
155 struct blkg_policy_data *, int),
156 const struct blkcg_policy *pol, int data,
158 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
159 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
160 const struct blkg_rwstat *rwstat);
161 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off);
162 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
165 u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off);
166 struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd,
169 struct blkg_conf_ctx {
170 struct gendisk *disk;
171 struct blkcg_gq *blkg;
175 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
176 const char *input, struct blkg_conf_ctx *ctx);
177 void blkg_conf_finish(struct blkg_conf_ctx *ctx);
180 static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
182 return css ? container_of(css, struct blkcg, css) : NULL;
185 static inline struct blkcg *task_blkcg(struct task_struct *tsk)
187 return css_to_blkcg(task_css(tsk, blkio_cgrp_id));
190 static inline struct blkcg *bio_blkcg(struct bio *bio)
192 if (bio && bio->bi_css)
193 return css_to_blkcg(bio->bi_css);
194 return task_blkcg(current);
198 * blkcg_parent - get the parent of a blkcg
199 * @blkcg: blkcg of interest
201 * Return the parent blkcg of @blkcg. Can be called anytime.
203 static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
205 return css_to_blkcg(blkcg->css.parent);
209 * blkg_to_pdata - get policy private data
210 * @blkg: blkg of interest
211 * @pol: policy of interest
213 * Return pointer to private data associated with the @blkg-@pol pair.
215 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
216 struct blkcg_policy *pol)
218 return blkg ? blkg->pd[pol->plid] : NULL;
222 * pdata_to_blkg - get blkg associated with policy private data
223 * @pd: policy private data of interest
225 * @pd is policy private data. Determine the blkg it's associated with.
227 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
229 return pd ? pd->blkg : NULL;
233 * blkg_path - format cgroup path of blkg
234 * @blkg: blkg of interest
235 * @buf: target buffer
236 * @buflen: target buffer length
238 * Format the path of the cgroup of @blkg into @buf.
240 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
244 p = cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
246 strncpy(buf, "<unavailable>", buflen);
247 return -ENAMETOOLONG;
250 memmove(buf, p, buf + buflen - p);
255 * blkg_get - get a blkg reference
258 * The caller should be holding an existing reference.
260 static inline void blkg_get(struct blkcg_gq *blkg)
262 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
263 atomic_inc(&blkg->refcnt);
266 void __blkg_release_rcu(struct rcu_head *rcu);
269 * blkg_put - put a blkg reference
272 static inline void blkg_put(struct blkcg_gq *blkg)
274 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
275 if (atomic_dec_and_test(&blkg->refcnt))
276 call_rcu(&blkg->rcu_head, __blkg_release_rcu);
279 struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, struct request_queue *q,
283 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
284 * @d_blkg: loop cursor pointing to the current descendant
285 * @pos_css: used for iteration
286 * @p_blkg: target blkg to walk descendants of
288 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
289 * read locked. If called under either blkcg or queue lock, the iteration
290 * is guaranteed to include all and only online blkgs. The caller may
291 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
292 * @p_blkg is included in the iteration and the first node to be visited.
294 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
295 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
296 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
297 (p_blkg)->q, false)))
300 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
301 * @d_blkg: loop cursor pointing to the current descendant
302 * @pos_css: used for iteration
303 * @p_blkg: target blkg to walk descendants of
305 * Similar to blkg_for_each_descendant_pre() but performs post-order
306 * traversal instead. Synchronization rules are the same. @p_blkg is
307 * included in the iteration and the last node to be visited.
309 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
310 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
311 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
312 (p_blkg)->q, false)))
315 * blk_get_rl - get request_list to use
316 * @q: request_queue of interest
317 * @bio: bio which will be attached to the allocated request (may be %NULL)
319 * The caller wants to allocate a request from @q to use for @bio. Find
320 * the request_list to use and obtain a reference on it. Should be called
321 * under queue_lock. This function is guaranteed to return non-%NULL
324 static inline struct request_list *blk_get_rl(struct request_queue *q,
328 struct blkcg_gq *blkg;
332 blkcg = bio_blkcg(bio);
334 /* bypass blkg lookup and use @q->root_rl directly for root */
335 if (blkcg == &blkcg_root)
339 * Try to use blkg->rl. blkg lookup may fail under memory pressure
340 * or if either the blkcg or queue is going away. Fall back to
341 * root_rl in such cases.
343 blkg = blkg_lookup_create(blkcg, q);
344 if (unlikely(IS_ERR(blkg)))
356 * blk_put_rl - put request_list
357 * @rl: request_list to put
359 * Put the reference acquired by blk_get_rl(). Should be called under
362 static inline void blk_put_rl(struct request_list *rl)
364 /* root_rl may not have blkg set */
365 if (rl->blkg && rl->blkg->blkcg != &blkcg_root)
370 * blk_rq_set_rl - associate a request with a request_list
371 * @rq: request of interest
372 * @rl: target request_list
374 * Associate @rq with @rl so that accounting and freeing can know the
375 * request_list @rq came from.
377 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl)
383 * blk_rq_rl - return the request_list a request came from
384 * @rq: request of interest
386 * Return the request_list @rq is allocated from.
388 static inline struct request_list *blk_rq_rl(struct request *rq)
393 struct request_list *__blk_queue_next_rl(struct request_list *rl,
394 struct request_queue *q);
396 * blk_queue_for_each_rl - iterate through all request_lists of a request_queue
398 * Should be used under queue_lock.
400 #define blk_queue_for_each_rl(rl, q) \
401 for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
403 static inline void blkg_stat_init(struct blkg_stat *stat)
405 u64_stats_init(&stat->syncp);
409 * blkg_stat_add - add a value to a blkg_stat
410 * @stat: target blkg_stat
413 * Add @val to @stat. The caller is responsible for synchronizing calls to
416 static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
418 u64_stats_update_begin(&stat->syncp);
420 u64_stats_update_end(&stat->syncp);
424 * blkg_stat_read - read the current value of a blkg_stat
425 * @stat: blkg_stat to read
427 * Read the current value of @stat. This function can be called without
428 * synchroniztion and takes care of u64 atomicity.
430 static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
436 start = u64_stats_fetch_begin_irq(&stat->syncp);
438 } while (u64_stats_fetch_retry_irq(&stat->syncp, start));
444 * blkg_stat_reset - reset a blkg_stat
445 * @stat: blkg_stat to reset
447 static inline void blkg_stat_reset(struct blkg_stat *stat)
453 * blkg_stat_merge - merge a blkg_stat into another
454 * @to: the destination blkg_stat
457 * Add @from's count to @to.
459 static inline void blkg_stat_merge(struct blkg_stat *to, struct blkg_stat *from)
461 blkg_stat_add(to, blkg_stat_read(from));
464 static inline void blkg_rwstat_init(struct blkg_rwstat *rwstat)
466 u64_stats_init(&rwstat->syncp);
470 * blkg_rwstat_add - add a value to a blkg_rwstat
471 * @rwstat: target blkg_rwstat
472 * @rw: mask of REQ_{WRITE|SYNC}
475 * Add @val to @rwstat. The counters are chosen according to @rw. The
476 * caller is responsible for synchronizing calls to this function.
478 static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
479 int rw, uint64_t val)
481 u64_stats_update_begin(&rwstat->syncp);
484 rwstat->cnt[BLKG_RWSTAT_WRITE] += val;
486 rwstat->cnt[BLKG_RWSTAT_READ] += val;
488 rwstat->cnt[BLKG_RWSTAT_SYNC] += val;
490 rwstat->cnt[BLKG_RWSTAT_ASYNC] += val;
492 u64_stats_update_end(&rwstat->syncp);
496 * blkg_rwstat_read - read the current values of a blkg_rwstat
497 * @rwstat: blkg_rwstat to read
499 * Read the current snapshot of @rwstat and return it as the return value.
500 * This function can be called without synchronization and takes care of
503 static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
506 struct blkg_rwstat tmp;
509 start = u64_stats_fetch_begin_irq(&rwstat->syncp);
511 } while (u64_stats_fetch_retry_irq(&rwstat->syncp, start));
517 * blkg_rwstat_total - read the total count of a blkg_rwstat
518 * @rwstat: blkg_rwstat to read
520 * Return the total count of @rwstat regardless of the IO direction. This
521 * function can be called without synchronization and takes care of u64
524 static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat)
526 struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
528 return tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE];
532 * blkg_rwstat_reset - reset a blkg_rwstat
533 * @rwstat: blkg_rwstat to reset
535 static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
537 memset(rwstat->cnt, 0, sizeof(rwstat->cnt));
541 * blkg_rwstat_merge - merge a blkg_rwstat into another
542 * @to: the destination blkg_rwstat
545 * Add @from's counts to @to.
547 static inline void blkg_rwstat_merge(struct blkg_rwstat *to,
548 struct blkg_rwstat *from)
550 struct blkg_rwstat v = blkg_rwstat_read(from);
553 u64_stats_update_begin(&to->syncp);
554 for (i = 0; i < BLKG_RWSTAT_NR; i++)
555 to->cnt[i] += v.cnt[i];
556 u64_stats_update_end(&to->syncp);
559 #else /* CONFIG_BLK_CGROUP */
564 struct blkg_policy_data {
570 struct blkcg_policy {
573 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
574 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
575 static inline void blkcg_drain_queue(struct request_queue *q) { }
576 static inline void blkcg_exit_queue(struct request_queue *q) { }
577 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
578 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
579 static inline int blkcg_activate_policy(struct request_queue *q,
580 const struct blkcg_policy *pol) { return 0; }
581 static inline void blkcg_deactivate_policy(struct request_queue *q,
582 const struct blkcg_policy *pol) { }
584 static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
586 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
587 struct blkcg_policy *pol) { return NULL; }
588 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
589 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
590 static inline void blkg_get(struct blkcg_gq *blkg) { }
591 static inline void blkg_put(struct blkcg_gq *blkg) { }
593 static inline struct request_list *blk_get_rl(struct request_queue *q,
594 struct bio *bio) { return &q->root_rl; }
595 static inline void blk_put_rl(struct request_list *rl) { }
596 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) { }
597 static inline struct request_list *blk_rq_rl(struct request *rq) { return &rq->q->root_rl; }
599 #define blk_queue_for_each_rl(rl, q) \
600 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
602 #endif /* CONFIG_BLK_CGROUP */
603 #endif /* _BLK_CGROUP_H */