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>
22 /* Max limits for throttle policy */
23 #define THROTL_IOPS_MAX UINT_MAX
25 /* CFQ specific, out here for blkcg->cfq_weight */
26 #define CFQ_WEIGHT_MIN 10
27 #define CFQ_WEIGHT_MAX 1000
28 #define CFQ_WEIGHT_DEFAULT 500
30 #ifdef CONFIG_BLK_CGROUP
32 enum blkg_rwstat_type {
39 BLKG_RWSTAT_TOTAL = BLKG_RWSTAT_NR,
45 struct cgroup_subsys_state css;
48 struct radix_tree_root blkg_tree;
49 struct blkcg_gq *blkg_hint;
50 struct hlist_head blkg_list;
52 /* for policies to test whether associated blkcg has changed */
55 /* TODO: per-policy storage in blkcg */
56 unsigned int cfq_weight; /* belongs to cfq */
57 unsigned int cfq_leaf_weight;
61 struct u64_stats_sync syncp;
66 struct u64_stats_sync syncp;
67 uint64_t cnt[BLKG_RWSTAT_NR];
71 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
72 * request_queue (q). This is used by blkcg policies which need to track
73 * information per blkcg - q pair.
75 * There can be multiple active blkcg policies and each has its private
76 * data on each blkg, the size of which is determined by
77 * blkcg_policy->pd_size. blkcg core allocates and frees such areas
78 * together with blkg and invokes pd_init/exit_fn() methods.
80 * Such private data must embed struct blkg_policy_data (pd) at the
81 * beginning and pd_size can't be smaller than pd.
83 struct blkg_policy_data {
84 /* the blkg and policy id this per-policy data belongs to */
85 struct blkcg_gq *blkg;
88 /* used during policy activation */
89 struct list_head alloc_node;
92 /* association between a blk cgroup and a request queue */
94 /* Pointer to the associated request_queue */
95 struct request_queue *q;
96 struct list_head q_node;
97 struct hlist_node blkcg_node;
100 /* all non-root blkcg_gq's are guaranteed to have access to parent */
101 struct blkcg_gq *parent;
103 /* request allocation list for this blkcg-q pair */
104 struct request_list rl;
106 /* reference count */
109 /* is this blkg online? protected by both blkcg and q locks */
112 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
114 struct rcu_head rcu_head;
117 typedef void (blkcg_pol_init_pd_fn)(struct blkcg_gq *blkg);
118 typedef void (blkcg_pol_online_pd_fn)(struct blkcg_gq *blkg);
119 typedef void (blkcg_pol_offline_pd_fn)(struct blkcg_gq *blkg);
120 typedef void (blkcg_pol_exit_pd_fn)(struct blkcg_gq *blkg);
121 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkcg_gq *blkg);
123 struct blkcg_policy {
125 /* policy specific private data size */
127 /* cgroup files for the policy */
128 struct cftype *cftypes;
131 blkcg_pol_init_pd_fn *pd_init_fn;
132 blkcg_pol_online_pd_fn *pd_online_fn;
133 blkcg_pol_offline_pd_fn *pd_offline_fn;
134 blkcg_pol_exit_pd_fn *pd_exit_fn;
135 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;
138 extern struct blkcg blkcg_root;
140 struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q);
141 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
142 struct request_queue *q);
143 int blkcg_init_queue(struct request_queue *q);
144 void blkcg_drain_queue(struct request_queue *q);
145 void blkcg_exit_queue(struct request_queue *q);
147 /* Blkio controller policy registration */
148 int blkcg_policy_register(struct blkcg_policy *pol);
149 void blkcg_policy_unregister(struct blkcg_policy *pol);
150 int blkcg_activate_policy(struct request_queue *q,
151 const struct blkcg_policy *pol);
152 void blkcg_deactivate_policy(struct request_queue *q,
153 const struct blkcg_policy *pol);
155 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
156 u64 (*prfill)(struct seq_file *,
157 struct blkg_policy_data *, int),
158 const struct blkcg_policy *pol, int data,
160 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
161 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
162 const struct blkg_rwstat *rwstat);
163 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off);
164 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
167 u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off);
168 struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd,
171 struct blkg_conf_ctx {
172 struct gendisk *disk;
173 struct blkcg_gq *blkg;
177 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
178 const char *input, struct blkg_conf_ctx *ctx);
179 void blkg_conf_finish(struct blkg_conf_ctx *ctx);
182 static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
184 return css ? container_of(css, struct blkcg, css) : NULL;
187 static inline struct blkcg *cgroup_to_blkcg(struct cgroup *cgroup)
189 return css_to_blkcg(cgroup_css(cgroup, blkio_subsys_id));
192 static inline struct blkcg *task_blkcg(struct task_struct *tsk)
194 return css_to_blkcg(task_css(tsk, blkio_subsys_id));
197 static inline struct blkcg *bio_blkcg(struct bio *bio)
199 if (bio && bio->bi_css)
200 return css_to_blkcg(bio->bi_css);
201 return task_blkcg(current);
205 * blkcg_parent - get the parent of a blkcg
206 * @blkcg: blkcg of interest
208 * Return the parent blkcg of @blkcg. Can be called anytime.
210 static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
212 struct cgroup *pcg = blkcg->css.cgroup->parent;
214 return pcg ? cgroup_to_blkcg(pcg) : NULL;
218 * blkg_to_pdata - get policy private data
219 * @blkg: blkg of interest
220 * @pol: policy of interest
222 * Return pointer to private data associated with the @blkg-@pol pair.
224 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
225 struct blkcg_policy *pol)
227 return blkg ? blkg->pd[pol->plid] : NULL;
231 * pdata_to_blkg - get blkg associated with policy private data
232 * @pd: policy private data of interest
234 * @pd is policy private data. Determine the blkg it's associated with.
236 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
238 return pd ? pd->blkg : NULL;
242 * blkg_path - format cgroup path of blkg
243 * @blkg: blkg of interest
244 * @buf: target buffer
245 * @buflen: target buffer length
247 * Format the path of the cgroup of @blkg into @buf.
249 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
253 ret = cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
255 strncpy(buf, "<unavailable>", buflen);
260 * blkg_get - get a blkg reference
263 * The caller should be holding queue_lock and an existing reference.
265 static inline void blkg_get(struct blkcg_gq *blkg)
267 lockdep_assert_held(blkg->q->queue_lock);
268 WARN_ON_ONCE(!blkg->refcnt);
272 void __blkg_release_rcu(struct rcu_head *rcu);
275 * blkg_put - put a blkg reference
278 * The caller should be holding queue_lock.
280 static inline void blkg_put(struct blkcg_gq *blkg)
282 lockdep_assert_held(blkg->q->queue_lock);
283 WARN_ON_ONCE(blkg->refcnt <= 0);
285 call_rcu(&blkg->rcu_head, __blkg_release_rcu);
288 struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, struct request_queue *q,
292 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
293 * @d_blkg: loop cursor pointing to the current descendant
294 * @pos_cgrp: used for iteration
295 * @p_blkg: target blkg to walk descendants of
297 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
298 * read locked. If called under either blkcg or queue lock, the iteration
299 * is guaranteed to include all and only online blkgs. The caller may
300 * update @pos_cgrp by calling cgroup_rightmost_descendant() to skip
303 #define blkg_for_each_descendant_pre(d_blkg, pos_cgrp, p_blkg) \
304 cgroup_for_each_descendant_pre((pos_cgrp), (p_blkg)->blkcg->css.cgroup) \
305 if (((d_blkg) = __blkg_lookup(cgroup_to_blkcg(pos_cgrp), \
306 (p_blkg)->q, false)))
309 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
310 * @d_blkg: loop cursor pointing to the current descendant
311 * @pos_cgrp: used for iteration
312 * @p_blkg: target blkg to walk descendants of
314 * Similar to blkg_for_each_descendant_pre() but performs post-order
315 * traversal instead. Synchronization rules are the same.
317 #define blkg_for_each_descendant_post(d_blkg, pos_cgrp, p_blkg) \
318 cgroup_for_each_descendant_post((pos_cgrp), (p_blkg)->blkcg->css.cgroup) \
319 if (((d_blkg) = __blkg_lookup(cgroup_to_blkcg(pos_cgrp), \
320 (p_blkg)->q, false)))
323 * blk_get_rl - get request_list to use
324 * @q: request_queue of interest
325 * @bio: bio which will be attached to the allocated request (may be %NULL)
327 * The caller wants to allocate a request from @q to use for @bio. Find
328 * the request_list to use and obtain a reference on it. Should be called
329 * under queue_lock. This function is guaranteed to return non-%NULL
332 static inline struct request_list *blk_get_rl(struct request_queue *q,
336 struct blkcg_gq *blkg;
340 blkcg = bio_blkcg(bio);
342 /* bypass blkg lookup and use @q->root_rl directly for root */
343 if (blkcg == &blkcg_root)
347 * Try to use blkg->rl. blkg lookup may fail under memory pressure
348 * or if either the blkcg or queue is going away. Fall back to
349 * root_rl in such cases.
351 blkg = blkg_lookup_create(blkcg, q);
352 if (unlikely(IS_ERR(blkg)))
364 * blk_put_rl - put request_list
365 * @rl: request_list to put
367 * Put the reference acquired by blk_get_rl(). Should be called under
370 static inline void blk_put_rl(struct request_list *rl)
372 /* root_rl may not have blkg set */
373 if (rl->blkg && rl->blkg->blkcg != &blkcg_root)
378 * blk_rq_set_rl - associate a request with a request_list
379 * @rq: request of interest
380 * @rl: target request_list
382 * Associate @rq with @rl so that accounting and freeing can know the
383 * request_list @rq came from.
385 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl)
391 * blk_rq_rl - return the request_list a request came from
392 * @rq: request of interest
394 * Return the request_list @rq is allocated from.
396 static inline struct request_list *blk_rq_rl(struct request *rq)
401 struct request_list *__blk_queue_next_rl(struct request_list *rl,
402 struct request_queue *q);
404 * blk_queue_for_each_rl - iterate through all request_lists of a request_queue
406 * Should be used under queue_lock.
408 #define blk_queue_for_each_rl(rl, q) \
409 for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
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(&stat->syncp);
441 } while (u64_stats_fetch_retry(&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));
468 * blkg_rwstat_add - add a value to a blkg_rwstat
469 * @rwstat: target blkg_rwstat
470 * @rw: mask of REQ_{WRITE|SYNC}
473 * Add @val to @rwstat. The counters are chosen according to @rw. The
474 * caller is responsible for synchronizing calls to this function.
476 static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
477 int rw, uint64_t val)
479 u64_stats_update_begin(&rwstat->syncp);
482 rwstat->cnt[BLKG_RWSTAT_WRITE] += val;
484 rwstat->cnt[BLKG_RWSTAT_READ] += val;
486 rwstat->cnt[BLKG_RWSTAT_SYNC] += val;
488 rwstat->cnt[BLKG_RWSTAT_ASYNC] += val;
490 u64_stats_update_end(&rwstat->syncp);
494 * blkg_rwstat_read - read the current values of a blkg_rwstat
495 * @rwstat: blkg_rwstat to read
497 * Read the current snapshot of @rwstat and return it as the return value.
498 * This function can be called without synchronization and takes care of
501 static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
504 struct blkg_rwstat tmp;
507 start = u64_stats_fetch_begin(&rwstat->syncp);
509 } while (u64_stats_fetch_retry(&rwstat->syncp, start));
515 * blkg_rwstat_total - read the total count of a blkg_rwstat
516 * @rwstat: blkg_rwstat to read
518 * Return the total count of @rwstat regardless of the IO direction. This
519 * function can be called without synchronization and takes care of u64
522 static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat)
524 struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
526 return tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE];
530 * blkg_rwstat_reset - reset a blkg_rwstat
531 * @rwstat: blkg_rwstat to reset
533 static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
535 memset(rwstat->cnt, 0, sizeof(rwstat->cnt));
539 * blkg_rwstat_merge - merge a blkg_rwstat into another
540 * @to: the destination blkg_rwstat
543 * Add @from's counts to @to.
545 static inline void blkg_rwstat_merge(struct blkg_rwstat *to,
546 struct blkg_rwstat *from)
548 struct blkg_rwstat v = blkg_rwstat_read(from);
551 u64_stats_update_begin(&to->syncp);
552 for (i = 0; i < BLKG_RWSTAT_NR; i++)
553 to->cnt[i] += v.cnt[i];
554 u64_stats_update_end(&to->syncp);
557 #else /* CONFIG_BLK_CGROUP */
562 struct blkg_policy_data {
568 struct blkcg_policy {
571 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
572 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
573 static inline void blkcg_drain_queue(struct request_queue *q) { }
574 static inline void blkcg_exit_queue(struct request_queue *q) { }
575 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
576 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
577 static inline int blkcg_activate_policy(struct request_queue *q,
578 const struct blkcg_policy *pol) { return 0; }
579 static inline void blkcg_deactivate_policy(struct request_queue *q,
580 const struct blkcg_policy *pol) { }
582 static inline struct blkcg *cgroup_to_blkcg(struct cgroup *cgroup) { return NULL; }
583 static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
585 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
586 struct blkcg_policy *pol) { return NULL; }
587 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
588 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
589 static inline void blkg_get(struct blkcg_gq *blkg) { }
590 static inline void blkg_put(struct blkcg_gq *blkg) { }
592 static inline struct request_list *blk_get_rl(struct request_queue *q,
593 struct bio *bio) { return &q->root_rl; }
594 static inline void blk_put_rl(struct request_list *rl) { }
595 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) { }
596 static inline struct request_list *blk_rq_rl(struct request *rq) { return &rq->q->root_rl; }
598 #define blk_queue_for_each_rl(rl, q) \
599 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
601 #endif /* CONFIG_BLK_CGROUP */
602 #endif /* _BLK_CGROUP_H */