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;
57 #ifdef CONFIG_CGROUP_WRITEBACK
58 struct list_head cgwb_list;
63 struct u64_stats_sync syncp;
68 struct u64_stats_sync syncp;
69 uint64_t cnt[BLKG_RWSTAT_NR];
73 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
74 * request_queue (q). This is used by blkcg policies which need to track
75 * information per blkcg - q pair.
77 * There can be multiple active blkcg policies and each has its private
78 * data on each blkg, the size of which is determined by
79 * blkcg_policy->pd_size. blkcg core allocates and frees such areas
80 * together with blkg and invokes pd_init/exit_fn() methods.
82 * Such private data must embed struct blkg_policy_data (pd) at the
83 * beginning and pd_size can't be smaller than pd.
85 struct blkg_policy_data {
86 /* the blkg and policy id this per-policy data belongs to */
87 struct blkcg_gq *blkg;
90 /* used during policy activation */
91 struct list_head alloc_node;
94 /* association between a blk cgroup and a request queue */
96 /* Pointer to the associated request_queue */
97 struct request_queue *q;
98 struct list_head q_node;
99 struct hlist_node blkcg_node;
102 /* all non-root blkcg_gq's are guaranteed to have access to parent */
103 struct blkcg_gq *parent;
105 /* request allocation list for this blkcg-q pair */
106 struct request_list rl;
108 /* reference count */
111 /* is this blkg online? protected by both blkcg and q locks */
114 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
116 struct rcu_head rcu_head;
119 typedef void (blkcg_pol_init_pd_fn)(struct blkcg_gq *blkg);
120 typedef void (blkcg_pol_online_pd_fn)(struct blkcg_gq *blkg);
121 typedef void (blkcg_pol_offline_pd_fn)(struct blkcg_gq *blkg);
122 typedef void (blkcg_pol_exit_pd_fn)(struct blkcg_gq *blkg);
123 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkcg_gq *blkg);
125 struct blkcg_policy {
127 /* policy specific private data size */
129 /* cgroup files for the policy */
130 struct cftype *cftypes;
133 blkcg_pol_init_pd_fn *pd_init_fn;
134 blkcg_pol_online_pd_fn *pd_online_fn;
135 blkcg_pol_offline_pd_fn *pd_offline_fn;
136 blkcg_pol_exit_pd_fn *pd_exit_fn;
137 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;
140 extern struct blkcg blkcg_root;
141 extern struct cgroup_subsys_state * const blkcg_root_css;
143 struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q);
144 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
145 struct request_queue *q);
146 int blkcg_init_queue(struct request_queue *q);
147 void blkcg_drain_queue(struct request_queue *q);
148 void blkcg_exit_queue(struct request_queue *q);
150 /* Blkio controller policy registration */
151 int blkcg_policy_register(struct blkcg_policy *pol);
152 void blkcg_policy_unregister(struct blkcg_policy *pol);
153 int blkcg_activate_policy(struct request_queue *q,
154 const struct blkcg_policy *pol);
155 void blkcg_deactivate_policy(struct request_queue *q,
156 const struct blkcg_policy *pol);
158 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
159 u64 (*prfill)(struct seq_file *,
160 struct blkg_policy_data *, int),
161 const struct blkcg_policy *pol, int data,
163 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
164 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
165 const struct blkg_rwstat *rwstat);
166 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off);
167 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
170 u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off);
171 struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd,
174 struct blkg_conf_ctx {
175 struct gendisk *disk;
176 struct blkcg_gq *blkg;
180 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
181 const char *input, struct blkg_conf_ctx *ctx);
182 void blkg_conf_finish(struct blkg_conf_ctx *ctx);
185 static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
187 return css ? container_of(css, struct blkcg, css) : NULL;
190 static inline struct blkcg *task_blkcg(struct task_struct *tsk)
192 return css_to_blkcg(task_css(tsk, blkio_cgrp_id));
195 static inline struct blkcg *bio_blkcg(struct bio *bio)
197 if (bio && bio->bi_css)
198 return css_to_blkcg(bio->bi_css);
199 return task_blkcg(current);
202 static inline struct cgroup_subsys_state *
203 task_get_blkcg_css(struct task_struct *task)
205 return task_get_css(task, blkio_cgrp_id);
209 * blkcg_parent - get the parent of a blkcg
210 * @blkcg: blkcg of interest
212 * Return the parent blkcg of @blkcg. Can be called anytime.
214 static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
216 return css_to_blkcg(blkcg->css.parent);
220 * blkg_to_pdata - get policy private data
221 * @blkg: blkg of interest
222 * @pol: policy of interest
224 * Return pointer to private data associated with the @blkg-@pol pair.
226 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
227 struct blkcg_policy *pol)
229 return blkg ? blkg->pd[pol->plid] : NULL;
233 * pdata_to_blkg - get blkg associated with policy private data
234 * @pd: policy private data of interest
236 * @pd is policy private data. Determine the blkg it's associated with.
238 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
240 return pd ? pd->blkg : NULL;
244 * blkg_path - format cgroup path of blkg
245 * @blkg: blkg of interest
246 * @buf: target buffer
247 * @buflen: target buffer length
249 * Format the path of the cgroup of @blkg into @buf.
251 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
255 p = cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
257 strncpy(buf, "<unavailable>", buflen);
258 return -ENAMETOOLONG;
261 memmove(buf, p, buf + buflen - p);
266 * blkg_get - get a blkg reference
269 * The caller should be holding an existing reference.
271 static inline void blkg_get(struct blkcg_gq *blkg)
273 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
274 atomic_inc(&blkg->refcnt);
277 void __blkg_release_rcu(struct rcu_head *rcu);
280 * blkg_put - put a blkg reference
283 static inline void blkg_put(struct blkcg_gq *blkg)
285 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
286 if (atomic_dec_and_test(&blkg->refcnt))
287 call_rcu(&blkg->rcu_head, __blkg_release_rcu);
290 struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, struct request_queue *q,
294 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
295 * @d_blkg: loop cursor pointing to the current descendant
296 * @pos_css: used for iteration
297 * @p_blkg: target blkg to walk descendants of
299 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
300 * read locked. If called under either blkcg or queue lock, the iteration
301 * is guaranteed to include all and only online blkgs. The caller may
302 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
303 * @p_blkg is included in the iteration and the first node to be visited.
305 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
306 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
307 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
308 (p_blkg)->q, false)))
311 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
312 * @d_blkg: loop cursor pointing to the current descendant
313 * @pos_css: used for iteration
314 * @p_blkg: target blkg to walk descendants of
316 * Similar to blkg_for_each_descendant_pre() but performs post-order
317 * traversal instead. Synchronization rules are the same. @p_blkg is
318 * included in the iteration and the last node to be visited.
320 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
321 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
322 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
323 (p_blkg)->q, false)))
326 * blk_get_rl - get request_list to use
327 * @q: request_queue of interest
328 * @bio: bio which will be attached to the allocated request (may be %NULL)
330 * The caller wants to allocate a request from @q to use for @bio. Find
331 * the request_list to use and obtain a reference on it. Should be called
332 * under queue_lock. This function is guaranteed to return non-%NULL
335 static inline struct request_list *blk_get_rl(struct request_queue *q,
339 struct blkcg_gq *blkg;
343 blkcg = bio_blkcg(bio);
345 /* bypass blkg lookup and use @q->root_rl directly for root */
346 if (blkcg == &blkcg_root)
350 * Try to use blkg->rl. blkg lookup may fail under memory pressure
351 * or if either the blkcg or queue is going away. Fall back to
352 * root_rl in such cases.
354 blkg = blkg_lookup_create(blkcg, q);
355 if (unlikely(IS_ERR(blkg)))
367 * blk_put_rl - put request_list
368 * @rl: request_list to put
370 * Put the reference acquired by blk_get_rl(). Should be called under
373 static inline void blk_put_rl(struct request_list *rl)
375 /* root_rl may not have blkg set */
376 if (rl->blkg && rl->blkg->blkcg != &blkcg_root)
381 * blk_rq_set_rl - associate a request with a request_list
382 * @rq: request of interest
383 * @rl: target request_list
385 * Associate @rq with @rl so that accounting and freeing can know the
386 * request_list @rq came from.
388 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl)
394 * blk_rq_rl - return the request_list a request came from
395 * @rq: request of interest
397 * Return the request_list @rq is allocated from.
399 static inline struct request_list *blk_rq_rl(struct request *rq)
404 struct request_list *__blk_queue_next_rl(struct request_list *rl,
405 struct request_queue *q);
407 * blk_queue_for_each_rl - iterate through all request_lists of a request_queue
409 * Should be used under queue_lock.
411 #define blk_queue_for_each_rl(rl, q) \
412 for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
414 static inline void blkg_stat_init(struct blkg_stat *stat)
416 u64_stats_init(&stat->syncp);
420 * blkg_stat_add - add a value to a blkg_stat
421 * @stat: target blkg_stat
424 * Add @val to @stat. The caller is responsible for synchronizing calls to
427 static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
429 u64_stats_update_begin(&stat->syncp);
431 u64_stats_update_end(&stat->syncp);
435 * blkg_stat_read - read the current value of a blkg_stat
436 * @stat: blkg_stat to read
438 * Read the current value of @stat. This function can be called without
439 * synchroniztion and takes care of u64 atomicity.
441 static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
447 start = u64_stats_fetch_begin_irq(&stat->syncp);
449 } while (u64_stats_fetch_retry_irq(&stat->syncp, start));
455 * blkg_stat_reset - reset a blkg_stat
456 * @stat: blkg_stat to reset
458 static inline void blkg_stat_reset(struct blkg_stat *stat)
464 * blkg_stat_merge - merge a blkg_stat into another
465 * @to: the destination blkg_stat
468 * Add @from's count to @to.
470 static inline void blkg_stat_merge(struct blkg_stat *to, struct blkg_stat *from)
472 blkg_stat_add(to, blkg_stat_read(from));
475 static inline void blkg_rwstat_init(struct blkg_rwstat *rwstat)
477 u64_stats_init(&rwstat->syncp);
481 * blkg_rwstat_add - add a value to a blkg_rwstat
482 * @rwstat: target blkg_rwstat
483 * @rw: mask of REQ_{WRITE|SYNC}
486 * Add @val to @rwstat. The counters are chosen according to @rw. The
487 * caller is responsible for synchronizing calls to this function.
489 static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
490 int rw, uint64_t val)
492 u64_stats_update_begin(&rwstat->syncp);
495 rwstat->cnt[BLKG_RWSTAT_WRITE] += val;
497 rwstat->cnt[BLKG_RWSTAT_READ] += val;
499 rwstat->cnt[BLKG_RWSTAT_SYNC] += val;
501 rwstat->cnt[BLKG_RWSTAT_ASYNC] += val;
503 u64_stats_update_end(&rwstat->syncp);
507 * blkg_rwstat_read - read the current values of a blkg_rwstat
508 * @rwstat: blkg_rwstat to read
510 * Read the current snapshot of @rwstat and return it as the return value.
511 * This function can be called without synchronization and takes care of
514 static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
517 struct blkg_rwstat tmp;
520 start = u64_stats_fetch_begin_irq(&rwstat->syncp);
522 } while (u64_stats_fetch_retry_irq(&rwstat->syncp, start));
528 * blkg_rwstat_total - read the total count of a blkg_rwstat
529 * @rwstat: blkg_rwstat to read
531 * Return the total count of @rwstat regardless of the IO direction. This
532 * function can be called without synchronization and takes care of u64
535 static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat)
537 struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
539 return tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE];
543 * blkg_rwstat_reset - reset a blkg_rwstat
544 * @rwstat: blkg_rwstat to reset
546 static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
548 memset(rwstat->cnt, 0, sizeof(rwstat->cnt));
552 * blkg_rwstat_merge - merge a blkg_rwstat into another
553 * @to: the destination blkg_rwstat
556 * Add @from's counts to @to.
558 static inline void blkg_rwstat_merge(struct blkg_rwstat *to,
559 struct blkg_rwstat *from)
561 struct blkg_rwstat v = blkg_rwstat_read(from);
564 u64_stats_update_begin(&to->syncp);
565 for (i = 0; i < BLKG_RWSTAT_NR; i++)
566 to->cnt[i] += v.cnt[i];
567 u64_stats_update_end(&to->syncp);
570 #else /* CONFIG_BLK_CGROUP */
575 struct blkg_policy_data {
581 struct blkcg_policy {
584 #define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
586 static inline struct cgroup_subsys_state *
587 task_get_blkcg_css(struct task_struct *task)
594 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
595 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
596 static inline void blkcg_drain_queue(struct request_queue *q) { }
597 static inline void blkcg_exit_queue(struct request_queue *q) { }
598 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
599 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
600 static inline int blkcg_activate_policy(struct request_queue *q,
601 const struct blkcg_policy *pol) { return 0; }
602 static inline void blkcg_deactivate_policy(struct request_queue *q,
603 const struct blkcg_policy *pol) { }
605 static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
607 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
608 struct blkcg_policy *pol) { return NULL; }
609 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
610 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
611 static inline void blkg_get(struct blkcg_gq *blkg) { }
612 static inline void blkg_put(struct blkcg_gq *blkg) { }
614 static inline struct request_list *blk_get_rl(struct request_queue *q,
615 struct bio *bio) { return &q->root_rl; }
616 static inline void blk_put_rl(struct request_list *rl) { }
617 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) { }
618 static inline struct request_list *blk_rq_rl(struct request *rq) { return &rq->q->root_rl; }
620 #define blk_queue_for_each_rl(rl, q) \
621 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
623 #endif /* CONFIG_BLOCK */
624 #endif /* CONFIG_BLK_CGROUP */
625 #endif /* _BLK_CGROUP_H */