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 /* for policies to test whether associated blkcg has changed */
56 /* TODO: per-policy storage in blkcg */
57 unsigned int cfq_weight; /* belongs to cfq */
58 unsigned int cfq_leaf_weight;
62 struct u64_stats_sync syncp;
67 struct u64_stats_sync syncp;
68 uint64_t cnt[BLKG_RWSTAT_NR];
72 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
73 * request_queue (q). This is used by blkcg policies which need to track
74 * information per blkcg - q pair.
76 * There can be multiple active blkcg policies and each has its private
77 * data on each blkg, the size of which is determined by
78 * blkcg_policy->pd_size. blkcg core allocates and frees such areas
79 * together with blkg and invokes pd_init/exit_fn() methods.
81 * Such private data must embed struct blkg_policy_data (pd) at the
82 * beginning and pd_size can't be smaller than pd.
84 struct blkg_policy_data {
85 /* the blkg and policy id this per-policy data belongs to */
86 struct blkcg_gq *blkg;
89 /* used during policy activation */
90 struct list_head alloc_node;
93 /* association between a blk cgroup and a request queue */
95 /* Pointer to the associated request_queue */
96 struct request_queue *q;
97 struct list_head q_node;
98 struct hlist_node blkcg_node;
101 /* all non-root blkcg_gq's are guaranteed to have access to parent */
102 struct blkcg_gq *parent;
104 /* request allocation list for this blkcg-q pair */
105 struct request_list rl;
107 /* reference count */
110 /* is this blkg online? protected by both blkcg and q locks */
113 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
115 struct rcu_head rcu_head;
118 typedef void (blkcg_pol_init_pd_fn)(struct blkcg_gq *blkg);
119 typedef void (blkcg_pol_online_pd_fn)(struct blkcg_gq *blkg);
120 typedef void (blkcg_pol_offline_pd_fn)(struct blkcg_gq *blkg);
121 typedef void (blkcg_pol_exit_pd_fn)(struct blkcg_gq *blkg);
122 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkcg_gq *blkg);
124 struct blkcg_policy {
126 /* policy specific private data size */
128 /* cgroup files for the policy */
129 struct cftype *cftypes;
132 blkcg_pol_init_pd_fn *pd_init_fn;
133 blkcg_pol_online_pd_fn *pd_online_fn;
134 blkcg_pol_offline_pd_fn *pd_offline_fn;
135 blkcg_pol_exit_pd_fn *pd_exit_fn;
136 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;
139 extern struct blkcg blkcg_root;
141 struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q);
142 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
143 struct request_queue *q);
144 int blkcg_init_queue(struct request_queue *q);
145 void blkcg_drain_queue(struct request_queue *q);
146 void blkcg_exit_queue(struct request_queue *q);
148 /* Blkio controller policy registration */
149 int blkcg_policy_register(struct blkcg_policy *pol);
150 void blkcg_policy_unregister(struct blkcg_policy *pol);
151 int blkcg_activate_policy(struct request_queue *q,
152 const struct blkcg_policy *pol);
153 void blkcg_deactivate_policy(struct request_queue *q,
154 const struct blkcg_policy *pol);
156 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
157 u64 (*prfill)(struct seq_file *,
158 struct blkg_policy_data *, int),
159 const struct blkcg_policy *pol, int data,
161 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
162 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
163 const struct blkg_rwstat *rwstat);
164 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off);
165 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
168 u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off);
169 struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd,
172 struct blkg_conf_ctx {
173 struct gendisk *disk;
174 struct blkcg_gq *blkg;
178 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
179 const char *input, struct blkg_conf_ctx *ctx);
180 void blkg_conf_finish(struct blkg_conf_ctx *ctx);
183 static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
185 return css ? container_of(css, struct blkcg, css) : NULL;
188 static inline struct blkcg *task_blkcg(struct task_struct *tsk)
190 return css_to_blkcg(task_css(tsk, blkio_cgrp_id));
193 static inline struct blkcg *bio_blkcg(struct bio *bio)
195 if (bio && bio->bi_css)
196 return css_to_blkcg(bio->bi_css);
197 return task_blkcg(current);
201 * blkcg_parent - get the parent of a blkcg
202 * @blkcg: blkcg of interest
204 * Return the parent blkcg of @blkcg. Can be called anytime.
206 static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
208 return css_to_blkcg(blkcg->css.parent);
212 * blkg_to_pdata - get policy private data
213 * @blkg: blkg of interest
214 * @pol: policy of interest
216 * Return pointer to private data associated with the @blkg-@pol pair.
218 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
219 struct blkcg_policy *pol)
221 return blkg ? blkg->pd[pol->plid] : NULL;
225 * pdata_to_blkg - get blkg associated with policy private data
226 * @pd: policy private data of interest
228 * @pd is policy private data. Determine the blkg it's associated with.
230 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
232 return pd ? pd->blkg : NULL;
236 * blkg_path - format cgroup path of blkg
237 * @blkg: blkg of interest
238 * @buf: target buffer
239 * @buflen: target buffer length
241 * Format the path of the cgroup of @blkg into @buf.
243 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
247 p = cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
249 strncpy(buf, "<unavailable>", buflen);
250 return -ENAMETOOLONG;
253 memmove(buf, p, buf + buflen - p);
258 * blkg_get - get a blkg reference
261 * The caller should be holding an existing reference.
263 static inline void blkg_get(struct blkcg_gq *blkg)
265 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
266 atomic_inc(&blkg->refcnt);
269 void __blkg_release_rcu(struct rcu_head *rcu);
272 * blkg_put - put a blkg reference
275 static inline void blkg_put(struct blkcg_gq *blkg)
277 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
278 if (atomic_dec_and_test(&blkg->refcnt))
279 call_rcu(&blkg->rcu_head, __blkg_release_rcu);
282 struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, struct request_queue *q,
286 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
287 * @d_blkg: loop cursor pointing to the current descendant
288 * @pos_css: used for iteration
289 * @p_blkg: target blkg to walk descendants of
291 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
292 * read locked. If called under either blkcg or queue lock, the iteration
293 * is guaranteed to include all and only online blkgs. The caller may
294 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
295 * @p_blkg is included in the iteration and the first node to be visited.
297 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
298 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
299 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
300 (p_blkg)->q, false)))
303 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
304 * @d_blkg: loop cursor pointing to the current descendant
305 * @pos_css: used for iteration
306 * @p_blkg: target blkg to walk descendants of
308 * Similar to blkg_for_each_descendant_pre() but performs post-order
309 * traversal instead. Synchronization rules are the same. @p_blkg is
310 * included in the iteration and the last node to be visited.
312 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
313 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
314 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
315 (p_blkg)->q, false)))
318 * blk_get_rl - get request_list to use
319 * @q: request_queue of interest
320 * @bio: bio which will be attached to the allocated request (may be %NULL)
322 * The caller wants to allocate a request from @q to use for @bio. Find
323 * the request_list to use and obtain a reference on it. Should be called
324 * under queue_lock. This function is guaranteed to return non-%NULL
327 static inline struct request_list *blk_get_rl(struct request_queue *q,
331 struct blkcg_gq *blkg;
335 blkcg = bio_blkcg(bio);
337 /* bypass blkg lookup and use @q->root_rl directly for root */
338 if (blkcg == &blkcg_root)
342 * Try to use blkg->rl. blkg lookup may fail under memory pressure
343 * or if either the blkcg or queue is going away. Fall back to
344 * root_rl in such cases.
346 blkg = blkg_lookup_create(blkcg, q);
347 if (unlikely(IS_ERR(blkg)))
359 * blk_put_rl - put request_list
360 * @rl: request_list to put
362 * Put the reference acquired by blk_get_rl(). Should be called under
365 static inline void blk_put_rl(struct request_list *rl)
367 /* root_rl may not have blkg set */
368 if (rl->blkg && rl->blkg->blkcg != &blkcg_root)
373 * blk_rq_set_rl - associate a request with a request_list
374 * @rq: request of interest
375 * @rl: target request_list
377 * Associate @rq with @rl so that accounting and freeing can know the
378 * request_list @rq came from.
380 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl)
386 * blk_rq_rl - return the request_list a request came from
387 * @rq: request of interest
389 * Return the request_list @rq is allocated from.
391 static inline struct request_list *blk_rq_rl(struct request *rq)
396 struct request_list *__blk_queue_next_rl(struct request_list *rl,
397 struct request_queue *q);
399 * blk_queue_for_each_rl - iterate through all request_lists of a request_queue
401 * Should be used under queue_lock.
403 #define blk_queue_for_each_rl(rl, q) \
404 for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
406 static inline void blkg_stat_init(struct blkg_stat *stat)
408 u64_stats_init(&stat->syncp);
412 * blkg_stat_add - add a value to a blkg_stat
413 * @stat: target blkg_stat
416 * Add @val to @stat. The caller is responsible for synchronizing calls to
419 static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
421 u64_stats_update_begin(&stat->syncp);
423 u64_stats_update_end(&stat->syncp);
427 * blkg_stat_read - read the current value of a blkg_stat
428 * @stat: blkg_stat to read
430 * Read the current value of @stat. This function can be called without
431 * synchroniztion and takes care of u64 atomicity.
433 static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
439 start = u64_stats_fetch_begin_irq(&stat->syncp);
441 } while (u64_stats_fetch_retry_irq(&stat->syncp, start));
447 * blkg_stat_reset - reset a blkg_stat
448 * @stat: blkg_stat to reset
450 static inline void blkg_stat_reset(struct blkg_stat *stat)
456 * blkg_stat_merge - merge a blkg_stat into another
457 * @to: the destination blkg_stat
460 * Add @from's count to @to.
462 static inline void blkg_stat_merge(struct blkg_stat *to, struct blkg_stat *from)
464 blkg_stat_add(to, blkg_stat_read(from));
467 static inline void blkg_rwstat_init(struct blkg_rwstat *rwstat)
469 u64_stats_init(&rwstat->syncp);
473 * blkg_rwstat_add - add a value to a blkg_rwstat
474 * @rwstat: target blkg_rwstat
475 * @rw: mask of REQ_{WRITE|SYNC}
478 * Add @val to @rwstat. The counters are chosen according to @rw. The
479 * caller is responsible for synchronizing calls to this function.
481 static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
482 int rw, uint64_t val)
484 u64_stats_update_begin(&rwstat->syncp);
487 rwstat->cnt[BLKG_RWSTAT_WRITE] += val;
489 rwstat->cnt[BLKG_RWSTAT_READ] += val;
491 rwstat->cnt[BLKG_RWSTAT_SYNC] += val;
493 rwstat->cnt[BLKG_RWSTAT_ASYNC] += val;
495 u64_stats_update_end(&rwstat->syncp);
499 * blkg_rwstat_read - read the current values of a blkg_rwstat
500 * @rwstat: blkg_rwstat to read
502 * Read the current snapshot of @rwstat and return it as the return value.
503 * This function can be called without synchronization and takes care of
506 static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
509 struct blkg_rwstat tmp;
512 start = u64_stats_fetch_begin_irq(&rwstat->syncp);
514 } while (u64_stats_fetch_retry_irq(&rwstat->syncp, start));
520 * blkg_rwstat_total - read the total count of a blkg_rwstat
521 * @rwstat: blkg_rwstat to read
523 * Return the total count of @rwstat regardless of the IO direction. This
524 * function can be called without synchronization and takes care of u64
527 static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat)
529 struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
531 return tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE];
535 * blkg_rwstat_reset - reset a blkg_rwstat
536 * @rwstat: blkg_rwstat to reset
538 static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
540 memset(rwstat->cnt, 0, sizeof(rwstat->cnt));
544 * blkg_rwstat_merge - merge a blkg_rwstat into another
545 * @to: the destination blkg_rwstat
548 * Add @from's counts to @to.
550 static inline void blkg_rwstat_merge(struct blkg_rwstat *to,
551 struct blkg_rwstat *from)
553 struct blkg_rwstat v = blkg_rwstat_read(from);
556 u64_stats_update_begin(&to->syncp);
557 for (i = 0; i < BLKG_RWSTAT_NR; i++)
558 to->cnt[i] += v.cnt[i];
559 u64_stats_update_end(&to->syncp);
562 #else /* CONFIG_BLK_CGROUP */
567 struct blkg_policy_data {
573 struct blkcg_policy {
576 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
577 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
578 static inline void blkcg_drain_queue(struct request_queue *q) { }
579 static inline void blkcg_exit_queue(struct request_queue *q) { }
580 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
581 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
582 static inline int blkcg_activate_policy(struct request_queue *q,
583 const struct blkcg_policy *pol) { return 0; }
584 static inline void blkcg_deactivate_policy(struct request_queue *q,
585 const struct blkcg_policy *pol) { }
587 static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
589 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
590 struct blkcg_policy *pol) { return NULL; }
591 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
592 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
593 static inline void blkg_get(struct blkcg_gq *blkg) { }
594 static inline void blkg_put(struct blkcg_gq *blkg) { }
596 static inline struct request_list *blk_get_rl(struct request_queue *q,
597 struct bio *bio) { return &q->root_rl; }
598 static inline void blk_put_rl(struct request_list *rl) { }
599 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) { }
600 static inline struct request_list *blk_rq_rl(struct request *rq) { return &rq->q->root_rl; }
602 #define blk_queue_for_each_rl(rl, q) \
603 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
605 #endif /* CONFIG_BLK_CGROUP */
606 #endif /* _BLK_CGROUP_H */