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 *cpd[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 blkg:policy pair is
72 * represented by a blkg_policy_data which is allocated and freed by each
73 * policy's pd_alloc/free_fn() methods. A policy can allocate private data
74 * area by allocating larger data structure which embeds blkg_policy_data
77 struct blkg_policy_data {
78 /* the blkg and policy id this per-policy data belongs to */
79 struct blkcg_gq *blkg;
84 * Policies that need to keep per-blkcg data which is independent from any
85 * request_queue associated to it should implement cpd_alloc/free_fn()
86 * methods. A policy can allocate private data area by allocating larger
87 * data structure which embeds blkcg_policy_data at the beginning.
88 * cpd_init() is invoked to let each policy handle per-blkcg data.
90 struct blkcg_policy_data {
91 /* the blkcg and policy id this per-policy data belongs to */
96 /* association between a blk cgroup and a request queue */
98 /* Pointer to the associated request_queue */
99 struct request_queue *q;
100 struct list_head q_node;
101 struct hlist_node blkcg_node;
105 * Each blkg gets congested separately and the congestion state is
106 * propagated to the matching bdi_writeback_congested.
108 struct bdi_writeback_congested *wb_congested;
110 /* all non-root blkcg_gq's are guaranteed to have access to parent */
111 struct blkcg_gq *parent;
113 /* request allocation list for this blkcg-q pair */
114 struct request_list rl;
116 /* reference count */
119 /* is this blkg online? protected by both blkcg and q locks */
122 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
124 struct rcu_head rcu_head;
127 typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp);
128 typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd);
129 typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd);
130 typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(gfp_t gfp, int node);
131 typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd);
132 typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd);
133 typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd);
134 typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd);
135 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd);
137 struct blkcg_policy {
139 /* cgroup files for the policy */
140 struct cftype *cftypes;
143 blkcg_pol_alloc_cpd_fn *cpd_alloc_fn;
144 blkcg_pol_init_cpd_fn *cpd_init_fn;
145 blkcg_pol_free_cpd_fn *cpd_free_fn;
147 blkcg_pol_alloc_pd_fn *pd_alloc_fn;
148 blkcg_pol_init_pd_fn *pd_init_fn;
149 blkcg_pol_online_pd_fn *pd_online_fn;
150 blkcg_pol_offline_pd_fn *pd_offline_fn;
151 blkcg_pol_free_pd_fn *pd_free_fn;
152 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;
155 extern struct blkcg blkcg_root;
156 extern struct cgroup_subsys_state * const blkcg_root_css;
158 struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q);
159 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
160 struct request_queue *q);
161 int blkcg_init_queue(struct request_queue *q);
162 void blkcg_drain_queue(struct request_queue *q);
163 void blkcg_exit_queue(struct request_queue *q);
165 /* Blkio controller policy registration */
166 int blkcg_policy_register(struct blkcg_policy *pol);
167 void blkcg_policy_unregister(struct blkcg_policy *pol);
168 int blkcg_activate_policy(struct request_queue *q,
169 const struct blkcg_policy *pol);
170 void blkcg_deactivate_policy(struct request_queue *q,
171 const struct blkcg_policy *pol);
173 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
174 u64 (*prfill)(struct seq_file *,
175 struct blkg_policy_data *, int),
176 const struct blkcg_policy *pol, int data,
178 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
179 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
180 const struct blkg_rwstat *rwstat);
181 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off);
182 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
185 u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off);
186 struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd,
189 struct blkg_conf_ctx {
190 struct gendisk *disk;
191 struct blkcg_gq *blkg;
195 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
196 const char *input, struct blkg_conf_ctx *ctx);
197 void blkg_conf_finish(struct blkg_conf_ctx *ctx);
200 static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
202 return css ? container_of(css, struct blkcg, css) : NULL;
205 static inline struct blkcg *task_blkcg(struct task_struct *tsk)
207 return css_to_blkcg(task_css(tsk, blkio_cgrp_id));
210 static inline struct blkcg *bio_blkcg(struct bio *bio)
212 if (bio && bio->bi_css)
213 return css_to_blkcg(bio->bi_css);
214 return task_blkcg(current);
217 static inline struct cgroup_subsys_state *
218 task_get_blkcg_css(struct task_struct *task)
220 return task_get_css(task, blkio_cgrp_id);
224 * blkcg_parent - get the parent of a blkcg
225 * @blkcg: blkcg of interest
227 * Return the parent blkcg of @blkcg. Can be called anytime.
229 static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
231 return css_to_blkcg(blkcg->css.parent);
235 * blkg_to_pdata - get policy private data
236 * @blkg: blkg of interest
237 * @pol: policy of interest
239 * Return pointer to private data associated with the @blkg-@pol pair.
241 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
242 struct blkcg_policy *pol)
244 return blkg ? blkg->pd[pol->plid] : NULL;
247 static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,
248 struct blkcg_policy *pol)
250 return blkcg ? blkcg->cpd[pol->plid] : NULL;
254 * pdata_to_blkg - get blkg associated with policy private data
255 * @pd: policy private data of interest
257 * @pd is policy private data. Determine the blkg it's associated with.
259 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
261 return pd ? pd->blkg : NULL;
264 static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd)
266 return cpd ? cpd->blkcg : NULL;
270 * blkg_path - format cgroup path of blkg
271 * @blkg: blkg of interest
272 * @buf: target buffer
273 * @buflen: target buffer length
275 * Format the path of the cgroup of @blkg into @buf.
277 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
281 p = cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
283 strncpy(buf, "<unavailable>", buflen);
284 return -ENAMETOOLONG;
287 memmove(buf, p, buf + buflen - p);
292 * blkg_get - get a blkg reference
295 * The caller should be holding an existing reference.
297 static inline void blkg_get(struct blkcg_gq *blkg)
299 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
300 atomic_inc(&blkg->refcnt);
303 void __blkg_release_rcu(struct rcu_head *rcu);
306 * blkg_put - put a blkg reference
309 static inline void blkg_put(struct blkcg_gq *blkg)
311 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
312 if (atomic_dec_and_test(&blkg->refcnt))
313 call_rcu(&blkg->rcu_head, __blkg_release_rcu);
316 struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, struct request_queue *q,
320 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
321 * @d_blkg: loop cursor pointing to the current descendant
322 * @pos_css: used for iteration
323 * @p_blkg: target blkg to walk descendants of
325 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
326 * read locked. If called under either blkcg or queue lock, the iteration
327 * is guaranteed to include all and only online blkgs. The caller may
328 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
329 * @p_blkg is included in the iteration and the first node to be visited.
331 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
332 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
333 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
334 (p_blkg)->q, false)))
337 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
338 * @d_blkg: loop cursor pointing to the current descendant
339 * @pos_css: used for iteration
340 * @p_blkg: target blkg to walk descendants of
342 * Similar to blkg_for_each_descendant_pre() but performs post-order
343 * traversal instead. Synchronization rules are the same. @p_blkg is
344 * included in the iteration and the last node to be visited.
346 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
347 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
348 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
349 (p_blkg)->q, false)))
352 * blk_get_rl - get request_list to use
353 * @q: request_queue of interest
354 * @bio: bio which will be attached to the allocated request (may be %NULL)
356 * The caller wants to allocate a request from @q to use for @bio. Find
357 * the request_list to use and obtain a reference on it. Should be called
358 * under queue_lock. This function is guaranteed to return non-%NULL
361 static inline struct request_list *blk_get_rl(struct request_queue *q,
365 struct blkcg_gq *blkg;
369 blkcg = bio_blkcg(bio);
371 /* bypass blkg lookup and use @q->root_rl directly for root */
372 if (blkcg == &blkcg_root)
376 * Try to use blkg->rl. blkg lookup may fail under memory pressure
377 * or if either the blkcg or queue is going away. Fall back to
378 * root_rl in such cases.
380 blkg = blkg_lookup_create(blkcg, q);
381 if (unlikely(IS_ERR(blkg)))
393 * blk_put_rl - put request_list
394 * @rl: request_list to put
396 * Put the reference acquired by blk_get_rl(). Should be called under
399 static inline void blk_put_rl(struct request_list *rl)
401 if (rl->blkg->blkcg != &blkcg_root)
406 * blk_rq_set_rl - associate a request with a request_list
407 * @rq: request of interest
408 * @rl: target request_list
410 * Associate @rq with @rl so that accounting and freeing can know the
411 * request_list @rq came from.
413 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl)
419 * blk_rq_rl - return the request_list a request came from
420 * @rq: request of interest
422 * Return the request_list @rq is allocated from.
424 static inline struct request_list *blk_rq_rl(struct request *rq)
429 struct request_list *__blk_queue_next_rl(struct request_list *rl,
430 struct request_queue *q);
432 * blk_queue_for_each_rl - iterate through all request_lists of a request_queue
434 * Should be used under queue_lock.
436 #define blk_queue_for_each_rl(rl, q) \
437 for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
439 static inline void blkg_stat_init(struct blkg_stat *stat)
441 u64_stats_init(&stat->syncp);
445 * blkg_stat_add - add a value to a blkg_stat
446 * @stat: target blkg_stat
449 * Add @val to @stat. The caller is responsible for synchronizing calls to
452 static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
454 u64_stats_update_begin(&stat->syncp);
456 u64_stats_update_end(&stat->syncp);
460 * blkg_stat_read - read the current value of a blkg_stat
461 * @stat: blkg_stat to read
463 * Read the current value of @stat. This function can be called without
464 * synchroniztion and takes care of u64 atomicity.
466 static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
472 start = u64_stats_fetch_begin_irq(&stat->syncp);
474 } while (u64_stats_fetch_retry_irq(&stat->syncp, start));
480 * blkg_stat_reset - reset a blkg_stat
481 * @stat: blkg_stat to reset
483 static inline void blkg_stat_reset(struct blkg_stat *stat)
489 * blkg_stat_merge - merge a blkg_stat into another
490 * @to: the destination blkg_stat
493 * Add @from's count to @to.
495 static inline void blkg_stat_merge(struct blkg_stat *to, struct blkg_stat *from)
497 blkg_stat_add(to, blkg_stat_read(from));
500 static inline void blkg_rwstat_init(struct blkg_rwstat *rwstat)
502 u64_stats_init(&rwstat->syncp);
506 * blkg_rwstat_add - add a value to a blkg_rwstat
507 * @rwstat: target blkg_rwstat
508 * @rw: mask of REQ_{WRITE|SYNC}
511 * Add @val to @rwstat. The counters are chosen according to @rw. The
512 * caller is responsible for synchronizing calls to this function.
514 static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
515 int rw, uint64_t val)
517 u64_stats_update_begin(&rwstat->syncp);
520 rwstat->cnt[BLKG_RWSTAT_WRITE] += val;
522 rwstat->cnt[BLKG_RWSTAT_READ] += val;
524 rwstat->cnt[BLKG_RWSTAT_SYNC] += val;
526 rwstat->cnt[BLKG_RWSTAT_ASYNC] += val;
528 u64_stats_update_end(&rwstat->syncp);
532 * blkg_rwstat_read - read the current values of a blkg_rwstat
533 * @rwstat: blkg_rwstat to read
535 * Read the current snapshot of @rwstat and return it as the return value.
536 * This function can be called without synchronization and takes care of
539 static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
542 struct blkg_rwstat tmp;
545 start = u64_stats_fetch_begin_irq(&rwstat->syncp);
547 } while (u64_stats_fetch_retry_irq(&rwstat->syncp, start));
553 * blkg_rwstat_total - read the total count of a blkg_rwstat
554 * @rwstat: blkg_rwstat to read
556 * Return the total count of @rwstat regardless of the IO direction. This
557 * function can be called without synchronization and takes care of u64
560 static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat)
562 struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
564 return tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE];
568 * blkg_rwstat_reset - reset a blkg_rwstat
569 * @rwstat: blkg_rwstat to reset
571 static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
573 memset(rwstat->cnt, 0, sizeof(rwstat->cnt));
577 * blkg_rwstat_merge - merge a blkg_rwstat into another
578 * @to: the destination blkg_rwstat
581 * Add @from's counts to @to.
583 static inline void blkg_rwstat_merge(struct blkg_rwstat *to,
584 struct blkg_rwstat *from)
586 struct blkg_rwstat v = blkg_rwstat_read(from);
589 u64_stats_update_begin(&to->syncp);
590 for (i = 0; i < BLKG_RWSTAT_NR; i++)
591 to->cnt[i] += v.cnt[i];
592 u64_stats_update_end(&to->syncp);
595 #else /* CONFIG_BLK_CGROUP */
600 struct blkg_policy_data {
603 struct blkcg_policy_data {
609 struct blkcg_policy {
612 #define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
614 static inline struct cgroup_subsys_state *
615 task_get_blkcg_css(struct task_struct *task)
622 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
623 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
624 static inline void blkcg_drain_queue(struct request_queue *q) { }
625 static inline void blkcg_exit_queue(struct request_queue *q) { }
626 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
627 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
628 static inline int blkcg_activate_policy(struct request_queue *q,
629 const struct blkcg_policy *pol) { return 0; }
630 static inline void blkcg_deactivate_policy(struct request_queue *q,
631 const struct blkcg_policy *pol) { }
633 static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
635 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
636 struct blkcg_policy *pol) { return NULL; }
637 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
638 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
639 static inline void blkg_get(struct blkcg_gq *blkg) { }
640 static inline void blkg_put(struct blkcg_gq *blkg) { }
642 static inline struct request_list *blk_get_rl(struct request_queue *q,
643 struct bio *bio) { return &q->root_rl; }
644 static inline void blk_put_rl(struct request_list *rl) { }
645 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) { }
646 static inline struct request_list *blk_rq_rl(struct request *rq) { return &rq->q->root_rl; }
648 #define blk_queue_for_each_rl(rl, q) \
649 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
651 #endif /* CONFIG_BLOCK */
652 #endif /* CONFIG_BLK_CGROUP */
653 #endif /* _BLK_CGROUP_H */