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/percpu_counter.h>
18 #include <linux/seq_file.h>
19 #include <linux/radix-tree.h>
20 #include <linux/blkdev.h>
21 #include <linux/atomic.h>
23 /* percpu_counter batch for blkg_[rw]stats, per-cpu drift doesn't matter */
24 #define BLKG_STAT_CPU_BATCH (INT_MAX / 2)
26 /* Max limits for throttle policy */
27 #define THROTL_IOPS_MAX UINT_MAX
29 #ifdef CONFIG_BLK_CGROUP
31 enum blkg_rwstat_type {
38 BLKG_RWSTAT_TOTAL = BLKG_RWSTAT_NR,
44 struct cgroup_subsys_state css;
47 struct radix_tree_root blkg_tree;
48 struct blkcg_gq *blkg_hint;
49 struct hlist_head blkg_list;
51 struct blkcg_policy_data *cpd[BLKCG_MAX_POLS];
53 struct list_head all_blkcgs_node;
54 #ifdef CONFIG_CGROUP_WRITEBACK
55 struct list_head cgwb_list;
60 * blkg_[rw]stat->aux_cnt is excluded for local stats but included for
61 * recursive. Used to carry stats of dead children, and, for blkg_rwstat,
62 * to carry result values from read and sum operations.
65 struct percpu_counter cpu_cnt;
70 struct percpu_counter cpu_cnt[BLKG_RWSTAT_NR];
71 atomic64_t aux_cnt[BLKG_RWSTAT_NR];
75 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
76 * request_queue (q). This is used by blkcg policies which need to track
77 * information per blkcg - q pair.
79 * There can be multiple active blkcg policies and each blkg:policy pair is
80 * represented by a blkg_policy_data which is allocated and freed by each
81 * policy's pd_alloc/free_fn() methods. A policy can allocate private data
82 * area by allocating larger data structure which embeds blkg_policy_data
85 struct blkg_policy_data {
86 /* the blkg and policy id this per-policy data belongs to */
87 struct blkcg_gq *blkg;
92 * Policies that need to keep per-blkcg data which is independent from any
93 * request_queue associated to it should implement cpd_alloc/free_fn()
94 * methods. A policy can allocate private data area by allocating larger
95 * data structure which embeds blkcg_policy_data at the beginning.
96 * cpd_init() is invoked to let each policy handle per-blkcg data.
98 struct blkcg_policy_data {
99 /* the blkcg and policy id this per-policy data belongs to */
104 /* association between a blk cgroup and a request queue */
106 /* Pointer to the associated request_queue */
107 struct request_queue *q;
108 struct list_head q_node;
109 struct hlist_node blkcg_node;
113 * Each blkg gets congested separately and the congestion state is
114 * propagated to the matching bdi_writeback_congested.
116 struct bdi_writeback_congested *wb_congested;
118 /* all non-root blkcg_gq's are guaranteed to have access to parent */
119 struct blkcg_gq *parent;
121 /* request allocation list for this blkcg-q pair */
122 struct request_list rl;
124 /* reference count */
127 /* is this blkg online? protected by both blkcg and q locks */
130 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
132 struct rcu_head rcu_head;
135 typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp);
136 typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd);
137 typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd);
138 typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(gfp_t gfp, int node);
139 typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd);
140 typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd);
141 typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd);
142 typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd);
143 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd);
145 struct blkcg_policy {
147 /* cgroup files for the policy */
148 struct cftype *cftypes;
151 blkcg_pol_alloc_cpd_fn *cpd_alloc_fn;
152 blkcg_pol_init_cpd_fn *cpd_init_fn;
153 blkcg_pol_free_cpd_fn *cpd_free_fn;
155 blkcg_pol_alloc_pd_fn *pd_alloc_fn;
156 blkcg_pol_init_pd_fn *pd_init_fn;
157 blkcg_pol_online_pd_fn *pd_online_fn;
158 blkcg_pol_offline_pd_fn *pd_offline_fn;
159 blkcg_pol_free_pd_fn *pd_free_fn;
160 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;
163 extern struct blkcg blkcg_root;
164 extern struct cgroup_subsys_state * const blkcg_root_css;
166 struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
167 struct request_queue *q, bool update_hint);
168 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
169 struct request_queue *q);
170 int blkcg_init_queue(struct request_queue *q);
171 void blkcg_drain_queue(struct request_queue *q);
172 void blkcg_exit_queue(struct request_queue *q);
174 /* Blkio controller policy registration */
175 int blkcg_policy_register(struct blkcg_policy *pol);
176 void blkcg_policy_unregister(struct blkcg_policy *pol);
177 int blkcg_activate_policy(struct request_queue *q,
178 const struct blkcg_policy *pol);
179 void blkcg_deactivate_policy(struct request_queue *q,
180 const struct blkcg_policy *pol);
182 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
183 u64 (*prfill)(struct seq_file *,
184 struct blkg_policy_data *, int),
185 const struct blkcg_policy *pol, int data,
187 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
188 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
189 const struct blkg_rwstat *rwstat);
190 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off);
191 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
194 u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg,
195 struct blkcg_policy *pol, int off);
196 struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg,
197 struct blkcg_policy *pol, int off);
199 struct blkg_conf_ctx {
200 struct gendisk *disk;
201 struct blkcg_gq *blkg;
205 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
206 const char *input, struct blkg_conf_ctx *ctx);
207 void blkg_conf_finish(struct blkg_conf_ctx *ctx);
210 static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
212 return css ? container_of(css, struct blkcg, css) : NULL;
215 static inline struct blkcg *task_blkcg(struct task_struct *tsk)
217 return css_to_blkcg(task_css(tsk, blkio_cgrp_id));
220 static inline struct blkcg *bio_blkcg(struct bio *bio)
222 if (bio && bio->bi_css)
223 return css_to_blkcg(bio->bi_css);
224 return task_blkcg(current);
227 static inline struct cgroup_subsys_state *
228 task_get_blkcg_css(struct task_struct *task)
230 return task_get_css(task, blkio_cgrp_id);
234 * blkcg_parent - get the parent of a blkcg
235 * @blkcg: blkcg of interest
237 * Return the parent blkcg of @blkcg. Can be called anytime.
239 static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
241 return css_to_blkcg(blkcg->css.parent);
245 * __blkg_lookup - internal version of blkg_lookup()
246 * @blkcg: blkcg of interest
247 * @q: request_queue of interest
248 * @update_hint: whether to update lookup hint with the result or not
250 * This is internal version and shouldn't be used by policy
251 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
252 * @q's bypass state. If @update_hint is %true, the caller should be
253 * holding @q->queue_lock and lookup hint is updated on success.
255 static inline struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
256 struct request_queue *q,
259 struct blkcg_gq *blkg;
261 if (blkcg == &blkcg_root)
264 blkg = rcu_dereference(blkcg->blkg_hint);
265 if (blkg && blkg->q == q)
268 return blkg_lookup_slowpath(blkcg, q, update_hint);
272 * blkg_lookup - lookup blkg for the specified blkcg - q pair
273 * @blkcg: blkcg of interest
274 * @q: request_queue of interest
276 * Lookup blkg for the @blkcg - @q pair. This function should be called
277 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
278 * - see blk_queue_bypass_start() for details.
280 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg,
281 struct request_queue *q)
283 WARN_ON_ONCE(!rcu_read_lock_held());
285 if (unlikely(blk_queue_bypass(q)))
287 return __blkg_lookup(blkcg, q, false);
291 * blkg_to_pdata - get policy private data
292 * @blkg: blkg of interest
293 * @pol: policy of interest
295 * Return pointer to private data associated with the @blkg-@pol pair.
297 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
298 struct blkcg_policy *pol)
300 return blkg ? blkg->pd[pol->plid] : NULL;
303 static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,
304 struct blkcg_policy *pol)
306 return blkcg ? blkcg->cpd[pol->plid] : NULL;
310 * pdata_to_blkg - get blkg associated with policy private data
311 * @pd: policy private data of interest
313 * @pd is policy private data. Determine the blkg it's associated with.
315 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
317 return pd ? pd->blkg : NULL;
320 static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd)
322 return cpd ? cpd->blkcg : NULL;
326 * blkg_path - format cgroup path of blkg
327 * @blkg: blkg of interest
328 * @buf: target buffer
329 * @buflen: target buffer length
331 * Format the path of the cgroup of @blkg into @buf.
333 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
337 p = cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
339 strncpy(buf, "<unavailable>", buflen);
340 return -ENAMETOOLONG;
343 memmove(buf, p, buf + buflen - p);
348 * blkg_get - get a blkg reference
351 * The caller should be holding an existing reference.
353 static inline void blkg_get(struct blkcg_gq *blkg)
355 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
356 atomic_inc(&blkg->refcnt);
359 void __blkg_release_rcu(struct rcu_head *rcu);
362 * blkg_put - put a blkg reference
365 static inline void blkg_put(struct blkcg_gq *blkg)
367 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
368 if (atomic_dec_and_test(&blkg->refcnt))
369 call_rcu(&blkg->rcu_head, __blkg_release_rcu);
373 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
374 * @d_blkg: loop cursor pointing to the current descendant
375 * @pos_css: used for iteration
376 * @p_blkg: target blkg to walk descendants of
378 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
379 * read locked. If called under either blkcg or queue lock, the iteration
380 * is guaranteed to include all and only online blkgs. The caller may
381 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
382 * @p_blkg is included in the iteration and the first node to be visited.
384 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
385 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
386 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
387 (p_blkg)->q, false)))
390 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
391 * @d_blkg: loop cursor pointing to the current descendant
392 * @pos_css: used for iteration
393 * @p_blkg: target blkg to walk descendants of
395 * Similar to blkg_for_each_descendant_pre() but performs post-order
396 * traversal instead. Synchronization rules are the same. @p_blkg is
397 * included in the iteration and the last node to be visited.
399 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
400 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
401 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
402 (p_blkg)->q, false)))
405 * blk_get_rl - get request_list to use
406 * @q: request_queue of interest
407 * @bio: bio which will be attached to the allocated request (may be %NULL)
409 * The caller wants to allocate a request from @q to use for @bio. Find
410 * the request_list to use and obtain a reference on it. Should be called
411 * under queue_lock. This function is guaranteed to return non-%NULL
414 static inline struct request_list *blk_get_rl(struct request_queue *q,
418 struct blkcg_gq *blkg;
422 blkcg = bio_blkcg(bio);
424 /* bypass blkg lookup and use @q->root_rl directly for root */
425 if (blkcg == &blkcg_root)
429 * Try to use blkg->rl. blkg lookup may fail under memory pressure
430 * or if either the blkcg or queue is going away. Fall back to
431 * root_rl in such cases.
433 blkg = blkg_lookup(blkcg, q);
446 * blk_put_rl - put request_list
447 * @rl: request_list to put
449 * Put the reference acquired by blk_get_rl(). Should be called under
452 static inline void blk_put_rl(struct request_list *rl)
454 if (rl->blkg->blkcg != &blkcg_root)
459 * blk_rq_set_rl - associate a request with a request_list
460 * @rq: request of interest
461 * @rl: target request_list
463 * Associate @rq with @rl so that accounting and freeing can know the
464 * request_list @rq came from.
466 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl)
472 * blk_rq_rl - return the request_list a request came from
473 * @rq: request of interest
475 * Return the request_list @rq is allocated from.
477 static inline struct request_list *blk_rq_rl(struct request *rq)
482 struct request_list *__blk_queue_next_rl(struct request_list *rl,
483 struct request_queue *q);
485 * blk_queue_for_each_rl - iterate through all request_lists of a request_queue
487 * Should be used under queue_lock.
489 #define blk_queue_for_each_rl(rl, q) \
490 for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
492 static inline int blkg_stat_init(struct blkg_stat *stat, gfp_t gfp)
496 ret = percpu_counter_init(&stat->cpu_cnt, 0, gfp);
500 atomic64_set(&stat->aux_cnt, 0);
504 static inline void blkg_stat_exit(struct blkg_stat *stat)
506 percpu_counter_destroy(&stat->cpu_cnt);
510 * blkg_stat_add - add a value to a blkg_stat
511 * @stat: target blkg_stat
514 * Add @val to @stat. The caller must ensure that IRQ on the same CPU
515 * don't re-enter this function for the same counter.
517 static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
519 __percpu_counter_add(&stat->cpu_cnt, val, BLKG_STAT_CPU_BATCH);
523 * blkg_stat_read - read the current value of a blkg_stat
524 * @stat: blkg_stat to read
526 static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
528 return percpu_counter_sum_positive(&stat->cpu_cnt);
532 * blkg_stat_reset - reset a blkg_stat
533 * @stat: blkg_stat to reset
535 static inline void blkg_stat_reset(struct blkg_stat *stat)
537 percpu_counter_set(&stat->cpu_cnt, 0);
538 atomic64_set(&stat->aux_cnt, 0);
542 * blkg_stat_add_aux - add a blkg_stat into another's aux count
543 * @to: the destination blkg_stat
546 * Add @from's count including the aux one to @to's aux count.
548 static inline void blkg_stat_add_aux(struct blkg_stat *to,
549 struct blkg_stat *from)
551 atomic64_add(blkg_stat_read(from) + atomic64_read(&from->aux_cnt),
555 static inline int blkg_rwstat_init(struct blkg_rwstat *rwstat, gfp_t gfp)
559 for (i = 0; i < BLKG_RWSTAT_NR; i++) {
560 ret = percpu_counter_init(&rwstat->cpu_cnt[i], 0, gfp);
563 percpu_counter_destroy(&rwstat->cpu_cnt[i]);
566 atomic64_set(&rwstat->aux_cnt[i], 0);
571 static inline void blkg_rwstat_exit(struct blkg_rwstat *rwstat)
575 for (i = 0; i < BLKG_RWSTAT_NR; i++)
576 percpu_counter_destroy(&rwstat->cpu_cnt[i]);
580 * blkg_rwstat_add - add a value to a blkg_rwstat
581 * @rwstat: target blkg_rwstat
582 * @rw: mask of REQ_{WRITE|SYNC}
585 * Add @val to @rwstat. The counters are chosen according to @rw. The
586 * caller is responsible for synchronizing calls to this function.
588 static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
589 int rw, uint64_t val)
591 struct percpu_counter *cnt;
594 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_WRITE];
596 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_READ];
598 __percpu_counter_add(cnt, val, BLKG_STAT_CPU_BATCH);
601 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_SYNC];
603 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_ASYNC];
605 __percpu_counter_add(cnt, val, BLKG_STAT_CPU_BATCH);
609 * blkg_rwstat_read - read the current values of a blkg_rwstat
610 * @rwstat: blkg_rwstat to read
612 * Read the current snapshot of @rwstat and return it in the aux counts.
614 static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
616 struct blkg_rwstat result;
619 for (i = 0; i < BLKG_RWSTAT_NR; i++)
620 atomic64_set(&result.aux_cnt[i],
621 percpu_counter_sum_positive(&rwstat->cpu_cnt[i]));
626 * blkg_rwstat_total - read the total count of a blkg_rwstat
627 * @rwstat: blkg_rwstat to read
629 * Return the total count of @rwstat regardless of the IO direction. This
630 * function can be called without synchronization and takes care of u64
633 static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat)
635 struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
637 return atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_READ]) +
638 atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_WRITE]);
642 * blkg_rwstat_reset - reset a blkg_rwstat
643 * @rwstat: blkg_rwstat to reset
645 static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
649 for (i = 0; i < BLKG_RWSTAT_NR; i++) {
650 percpu_counter_set(&rwstat->cpu_cnt[i], 0);
651 atomic64_set(&rwstat->aux_cnt[i], 0);
656 * blkg_rwstat_add_aux - add a blkg_rwstat into another's aux count
657 * @to: the destination blkg_rwstat
660 * Add @from's count including the aux one to @to's aux count.
662 static inline void blkg_rwstat_add_aux(struct blkg_rwstat *to,
663 struct blkg_rwstat *from)
665 struct blkg_rwstat v = blkg_rwstat_read(from);
668 for (i = 0; i < BLKG_RWSTAT_NR; i++)
669 atomic64_add(atomic64_read(&v.aux_cnt[i]) +
670 atomic64_read(&from->aux_cnt[i]),
674 #ifdef CONFIG_BLK_DEV_THROTTLING
675 extern bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg,
678 static inline bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg,
679 struct bio *bio) { return false; }
682 static inline bool blkcg_bio_issue_check(struct request_queue *q,
686 struct blkcg_gq *blkg;
690 blkcg = bio_blkcg(bio);
692 blkg = blkg_lookup(blkcg, q);
693 if (unlikely(!blkg)) {
694 spin_lock_irq(q->queue_lock);
695 blkg = blkg_lookup_create(blkcg, q);
698 spin_unlock_irq(q->queue_lock);
701 throtl = blk_throtl_bio(q, blkg, bio);
707 #else /* CONFIG_BLK_CGROUP */
712 struct blkg_policy_data {
715 struct blkcg_policy_data {
721 struct blkcg_policy {
724 #define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
726 static inline struct cgroup_subsys_state *
727 task_get_blkcg_css(struct task_struct *task)
734 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
735 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
736 static inline void blkcg_drain_queue(struct request_queue *q) { }
737 static inline void blkcg_exit_queue(struct request_queue *q) { }
738 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
739 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
740 static inline int blkcg_activate_policy(struct request_queue *q,
741 const struct blkcg_policy *pol) { return 0; }
742 static inline void blkcg_deactivate_policy(struct request_queue *q,
743 const struct blkcg_policy *pol) { }
745 static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
747 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
748 struct blkcg_policy *pol) { return NULL; }
749 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
750 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
751 static inline void blkg_get(struct blkcg_gq *blkg) { }
752 static inline void blkg_put(struct blkcg_gq *blkg) { }
754 static inline struct request_list *blk_get_rl(struct request_queue *q,
755 struct bio *bio) { return &q->root_rl; }
756 static inline void blk_put_rl(struct request_list *rl) { }
757 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) { }
758 static inline struct request_list *blk_rq_rl(struct request *rq) { return &rq->q->root_rl; }
760 static inline bool blkcg_bio_issue_check(struct request_queue *q,
761 struct bio *bio) { return true; }
763 #define blk_queue_for_each_rl(rl, q) \
764 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
766 #endif /* CONFIG_BLOCK */
767 #endif /* CONFIG_BLK_CGROUP */
768 #endif /* _BLK_CGROUP_H */