]>
Commit | Line | Data |
---|---|---|
1 | #ifndef _BLK_CGROUP_H | |
2 | #define _BLK_CGROUP_H | |
3 | /* | |
4 | * Common Block IO controller cgroup interface | |
5 | * | |
6 | * Based on ideas and code from CFQ, CFS and BFQ: | |
7 | * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk> | |
8 | * | |
9 | * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it> | |
10 | * Paolo Valente <paolo.valente@unimore.it> | |
11 | * | |
12 | * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com> | |
13 | * Nauman Rafique <nauman@google.com> | |
14 | */ | |
15 | ||
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> | |
22 | ||
23 | /* percpu_counter batch for blkg_[rw]stats, per-cpu drift doesn't matter */ | |
24 | #define BLKG_STAT_CPU_BATCH (INT_MAX / 2) | |
25 | ||
26 | /* Max limits for throttle policy */ | |
27 | #define THROTL_IOPS_MAX UINT_MAX | |
28 | ||
29 | #ifdef CONFIG_BLK_CGROUP | |
30 | ||
31 | enum blkg_rwstat_type { | |
32 | BLKG_RWSTAT_READ, | |
33 | BLKG_RWSTAT_WRITE, | |
34 | BLKG_RWSTAT_SYNC, | |
35 | BLKG_RWSTAT_ASYNC, | |
36 | ||
37 | BLKG_RWSTAT_NR, | |
38 | BLKG_RWSTAT_TOTAL = BLKG_RWSTAT_NR, | |
39 | }; | |
40 | ||
41 | struct blkcg_gq; | |
42 | ||
43 | struct blkcg { | |
44 | struct cgroup_subsys_state css; | |
45 | spinlock_t lock; | |
46 | ||
47 | struct radix_tree_root blkg_tree; | |
48 | struct blkcg_gq *blkg_hint; | |
49 | struct hlist_head blkg_list; | |
50 | ||
51 | struct blkcg_policy_data *cpd[BLKCG_MAX_POLS]; | |
52 | ||
53 | struct list_head all_blkcgs_node; | |
54 | #ifdef CONFIG_CGROUP_WRITEBACK | |
55 | struct list_head cgwb_list; | |
56 | #endif | |
57 | }; | |
58 | ||
59 | /* | |
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. | |
63 | */ | |
64 | struct blkg_stat { | |
65 | struct percpu_counter cpu_cnt; | |
66 | atomic64_t aux_cnt; | |
67 | }; | |
68 | ||
69 | struct blkg_rwstat { | |
70 | struct percpu_counter cpu_cnt[BLKG_RWSTAT_NR]; | |
71 | atomic64_t aux_cnt[BLKG_RWSTAT_NR]; | |
72 | }; | |
73 | ||
74 | /* | |
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. | |
78 | * | |
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 | |
83 | * at the beginning. | |
84 | */ | |
85 | struct blkg_policy_data { | |
86 | /* the blkg and policy id this per-policy data belongs to */ | |
87 | struct blkcg_gq *blkg; | |
88 | int plid; | |
89 | }; | |
90 | ||
91 | /* | |
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. | |
97 | */ | |
98 | struct blkcg_policy_data { | |
99 | /* the blkcg and policy id this per-policy data belongs to */ | |
100 | struct blkcg *blkcg; | |
101 | int plid; | |
102 | }; | |
103 | ||
104 | /* association between a blk cgroup and a request queue */ | |
105 | struct blkcg_gq { | |
106 | /* Pointer to the associated request_queue */ | |
107 | struct request_queue *q; | |
108 | struct list_head q_node; | |
109 | struct hlist_node blkcg_node; | |
110 | struct blkcg *blkcg; | |
111 | ||
112 | /* | |
113 | * Each blkg gets congested separately and the congestion state is | |
114 | * propagated to the matching bdi_writeback_congested. | |
115 | */ | |
116 | struct bdi_writeback_congested *wb_congested; | |
117 | ||
118 | /* all non-root blkcg_gq's are guaranteed to have access to parent */ | |
119 | struct blkcg_gq *parent; | |
120 | ||
121 | /* request allocation list for this blkcg-q pair */ | |
122 | struct request_list rl; | |
123 | ||
124 | /* reference count */ | |
125 | atomic_t refcnt; | |
126 | ||
127 | /* is this blkg online? protected by both blkcg and q locks */ | |
128 | bool online; | |
129 | ||
130 | struct blkg_rwstat stat_bytes; | |
131 | struct blkg_rwstat stat_ios; | |
132 | ||
133 | struct blkg_policy_data *pd[BLKCG_MAX_POLS]; | |
134 | ||
135 | struct rcu_head rcu_head; | |
136 | }; | |
137 | ||
138 | typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp); | |
139 | typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd); | |
140 | typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd); | |
141 | typedef void (blkcg_pol_bind_cpd_fn)(struct blkcg_policy_data *cpd); | |
142 | typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(gfp_t gfp, int node); | |
143 | typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd); | |
144 | typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd); | |
145 | typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd); | |
146 | typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd); | |
147 | typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd); | |
148 | ||
149 | struct blkcg_policy { | |
150 | int plid; | |
151 | /* cgroup files for the policy */ | |
152 | struct cftype *dfl_cftypes; | |
153 | struct cftype *legacy_cftypes; | |
154 | ||
155 | /* operations */ | |
156 | blkcg_pol_alloc_cpd_fn *cpd_alloc_fn; | |
157 | blkcg_pol_init_cpd_fn *cpd_init_fn; | |
158 | blkcg_pol_free_cpd_fn *cpd_free_fn; | |
159 | blkcg_pol_bind_cpd_fn *cpd_bind_fn; | |
160 | ||
161 | blkcg_pol_alloc_pd_fn *pd_alloc_fn; | |
162 | blkcg_pol_init_pd_fn *pd_init_fn; | |
163 | blkcg_pol_online_pd_fn *pd_online_fn; | |
164 | blkcg_pol_offline_pd_fn *pd_offline_fn; | |
165 | blkcg_pol_free_pd_fn *pd_free_fn; | |
166 | blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn; | |
167 | }; | |
168 | ||
169 | extern struct blkcg blkcg_root; | |
170 | extern struct cgroup_subsys_state * const blkcg_root_css; | |
171 | ||
172 | struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg, | |
173 | struct request_queue *q, bool update_hint); | |
174 | struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, | |
175 | struct request_queue *q); | |
176 | int blkcg_init_queue(struct request_queue *q); | |
177 | void blkcg_drain_queue(struct request_queue *q); | |
178 | void blkcg_exit_queue(struct request_queue *q); | |
179 | ||
180 | /* Blkio controller policy registration */ | |
181 | int blkcg_policy_register(struct blkcg_policy *pol); | |
182 | void blkcg_policy_unregister(struct blkcg_policy *pol); | |
183 | int blkcg_activate_policy(struct request_queue *q, | |
184 | const struct blkcg_policy *pol); | |
185 | void blkcg_deactivate_policy(struct request_queue *q, | |
186 | const struct blkcg_policy *pol); | |
187 | ||
188 | const char *blkg_dev_name(struct blkcg_gq *blkg); | |
189 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, | |
190 | u64 (*prfill)(struct seq_file *, | |
191 | struct blkg_policy_data *, int), | |
192 | const struct blkcg_policy *pol, int data, | |
193 | bool show_total); | |
194 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v); | |
195 | u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, | |
196 | const struct blkg_rwstat *rwstat); | |
197 | u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off); | |
198 | u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, | |
199 | int off); | |
200 | int blkg_print_stat_bytes(struct seq_file *sf, void *v); | |
201 | int blkg_print_stat_ios(struct seq_file *sf, void *v); | |
202 | int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v); | |
203 | int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v); | |
204 | ||
205 | u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg, | |
206 | struct blkcg_policy *pol, int off); | |
207 | struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg, | |
208 | struct blkcg_policy *pol, int off); | |
209 | ||
210 | struct blkg_conf_ctx { | |
211 | struct gendisk *disk; | |
212 | struct blkcg_gq *blkg; | |
213 | char *body; | |
214 | }; | |
215 | ||
216 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, | |
217 | char *input, struct blkg_conf_ctx *ctx); | |
218 | void blkg_conf_finish(struct blkg_conf_ctx *ctx); | |
219 | ||
220 | ||
221 | static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css) | |
222 | { | |
223 | return css ? container_of(css, struct blkcg, css) : NULL; | |
224 | } | |
225 | ||
226 | static inline struct blkcg *task_blkcg(struct task_struct *tsk) | |
227 | { | |
228 | return css_to_blkcg(task_css(tsk, io_cgrp_id)); | |
229 | } | |
230 | ||
231 | static inline struct blkcg *bio_blkcg(struct bio *bio) | |
232 | { | |
233 | if (bio && bio->bi_css) | |
234 | return css_to_blkcg(bio->bi_css); | |
235 | return task_blkcg(current); | |
236 | } | |
237 | ||
238 | static inline struct cgroup_subsys_state * | |
239 | task_get_blkcg_css(struct task_struct *task) | |
240 | { | |
241 | return task_get_css(task, io_cgrp_id); | |
242 | } | |
243 | ||
244 | /** | |
245 | * blkcg_parent - get the parent of a blkcg | |
246 | * @blkcg: blkcg of interest | |
247 | * | |
248 | * Return the parent blkcg of @blkcg. Can be called anytime. | |
249 | */ | |
250 | static inline struct blkcg *blkcg_parent(struct blkcg *blkcg) | |
251 | { | |
252 | return css_to_blkcg(blkcg->css.parent); | |
253 | } | |
254 | ||
255 | /** | |
256 | * __blkg_lookup - internal version of blkg_lookup() | |
257 | * @blkcg: blkcg of interest | |
258 | * @q: request_queue of interest | |
259 | * @update_hint: whether to update lookup hint with the result or not | |
260 | * | |
261 | * This is internal version and shouldn't be used by policy | |
262 | * implementations. Looks up blkgs for the @blkcg - @q pair regardless of | |
263 | * @q's bypass state. If @update_hint is %true, the caller should be | |
264 | * holding @q->queue_lock and lookup hint is updated on success. | |
265 | */ | |
266 | static inline struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, | |
267 | struct request_queue *q, | |
268 | bool update_hint) | |
269 | { | |
270 | struct blkcg_gq *blkg; | |
271 | ||
272 | if (blkcg == &blkcg_root) | |
273 | return q->root_blkg; | |
274 | ||
275 | blkg = rcu_dereference(blkcg->blkg_hint); | |
276 | if (blkg && blkg->q == q) | |
277 | return blkg; | |
278 | ||
279 | return blkg_lookup_slowpath(blkcg, q, update_hint); | |
280 | } | |
281 | ||
282 | /** | |
283 | * blkg_lookup - lookup blkg for the specified blkcg - q pair | |
284 | * @blkcg: blkcg of interest | |
285 | * @q: request_queue of interest | |
286 | * | |
287 | * Lookup blkg for the @blkcg - @q pair. This function should be called | |
288 | * under RCU read lock and is guaranteed to return %NULL if @q is bypassing | |
289 | * - see blk_queue_bypass_start() for details. | |
290 | */ | |
291 | static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, | |
292 | struct request_queue *q) | |
293 | { | |
294 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
295 | ||
296 | if (unlikely(blk_queue_bypass(q))) | |
297 | return NULL; | |
298 | return __blkg_lookup(blkcg, q, false); | |
299 | } | |
300 | ||
301 | /** | |
302 | * blkg_to_pdata - get policy private data | |
303 | * @blkg: blkg of interest | |
304 | * @pol: policy of interest | |
305 | * | |
306 | * Return pointer to private data associated with the @blkg-@pol pair. | |
307 | */ | |
308 | static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg, | |
309 | struct blkcg_policy *pol) | |
310 | { | |
311 | return blkg ? blkg->pd[pol->plid] : NULL; | |
312 | } | |
313 | ||
314 | static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg, | |
315 | struct blkcg_policy *pol) | |
316 | { | |
317 | return blkcg ? blkcg->cpd[pol->plid] : NULL; | |
318 | } | |
319 | ||
320 | /** | |
321 | * pdata_to_blkg - get blkg associated with policy private data | |
322 | * @pd: policy private data of interest | |
323 | * | |
324 | * @pd is policy private data. Determine the blkg it's associated with. | |
325 | */ | |
326 | static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) | |
327 | { | |
328 | return pd ? pd->blkg : NULL; | |
329 | } | |
330 | ||
331 | static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd) | |
332 | { | |
333 | return cpd ? cpd->blkcg : NULL; | |
334 | } | |
335 | ||
336 | /** | |
337 | * blkg_path - format cgroup path of blkg | |
338 | * @blkg: blkg of interest | |
339 | * @buf: target buffer | |
340 | * @buflen: target buffer length | |
341 | * | |
342 | * Format the path of the cgroup of @blkg into @buf. | |
343 | */ | |
344 | static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen) | |
345 | { | |
346 | char *p; | |
347 | ||
348 | p = cgroup_path(blkg->blkcg->css.cgroup, buf, buflen); | |
349 | if (!p) { | |
350 | strncpy(buf, "<unavailable>", buflen); | |
351 | return -ENAMETOOLONG; | |
352 | } | |
353 | ||
354 | memmove(buf, p, buf + buflen - p); | |
355 | return 0; | |
356 | } | |
357 | ||
358 | /** | |
359 | * blkg_get - get a blkg reference | |
360 | * @blkg: blkg to get | |
361 | * | |
362 | * The caller should be holding an existing reference. | |
363 | */ | |
364 | static inline void blkg_get(struct blkcg_gq *blkg) | |
365 | { | |
366 | WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0); | |
367 | atomic_inc(&blkg->refcnt); | |
368 | } | |
369 | ||
370 | void __blkg_release_rcu(struct rcu_head *rcu); | |
371 | ||
372 | /** | |
373 | * blkg_put - put a blkg reference | |
374 | * @blkg: blkg to put | |
375 | */ | |
376 | static inline void blkg_put(struct blkcg_gq *blkg) | |
377 | { | |
378 | WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0); | |
379 | if (atomic_dec_and_test(&blkg->refcnt)) | |
380 | call_rcu(&blkg->rcu_head, __blkg_release_rcu); | |
381 | } | |
382 | ||
383 | /** | |
384 | * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants | |
385 | * @d_blkg: loop cursor pointing to the current descendant | |
386 | * @pos_css: used for iteration | |
387 | * @p_blkg: target blkg to walk descendants of | |
388 | * | |
389 | * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU | |
390 | * read locked. If called under either blkcg or queue lock, the iteration | |
391 | * is guaranteed to include all and only online blkgs. The caller may | |
392 | * update @pos_css by calling css_rightmost_descendant() to skip subtree. | |
393 | * @p_blkg is included in the iteration and the first node to be visited. | |
394 | */ | |
395 | #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \ | |
396 | css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \ | |
397 | if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \ | |
398 | (p_blkg)->q, false))) | |
399 | ||
400 | /** | |
401 | * blkg_for_each_descendant_post - post-order walk of a blkg's descendants | |
402 | * @d_blkg: loop cursor pointing to the current descendant | |
403 | * @pos_css: used for iteration | |
404 | * @p_blkg: target blkg to walk descendants of | |
405 | * | |
406 | * Similar to blkg_for_each_descendant_pre() but performs post-order | |
407 | * traversal instead. Synchronization rules are the same. @p_blkg is | |
408 | * included in the iteration and the last node to be visited. | |
409 | */ | |
410 | #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \ | |
411 | css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \ | |
412 | if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \ | |
413 | (p_blkg)->q, false))) | |
414 | ||
415 | /** | |
416 | * blk_get_rl - get request_list to use | |
417 | * @q: request_queue of interest | |
418 | * @bio: bio which will be attached to the allocated request (may be %NULL) | |
419 | * | |
420 | * The caller wants to allocate a request from @q to use for @bio. Find | |
421 | * the request_list to use and obtain a reference on it. Should be called | |
422 | * under queue_lock. This function is guaranteed to return non-%NULL | |
423 | * request_list. | |
424 | */ | |
425 | static inline struct request_list *blk_get_rl(struct request_queue *q, | |
426 | struct bio *bio) | |
427 | { | |
428 | struct blkcg *blkcg; | |
429 | struct blkcg_gq *blkg; | |
430 | ||
431 | rcu_read_lock(); | |
432 | ||
433 | blkcg = bio_blkcg(bio); | |
434 | ||
435 | /* bypass blkg lookup and use @q->root_rl directly for root */ | |
436 | if (blkcg == &blkcg_root) | |
437 | goto root_rl; | |
438 | ||
439 | /* | |
440 | * Try to use blkg->rl. blkg lookup may fail under memory pressure | |
441 | * or if either the blkcg or queue is going away. Fall back to | |
442 | * root_rl in such cases. | |
443 | */ | |
444 | blkg = blkg_lookup(blkcg, q); | |
445 | if (unlikely(!blkg)) | |
446 | goto root_rl; | |
447 | ||
448 | blkg_get(blkg); | |
449 | rcu_read_unlock(); | |
450 | return &blkg->rl; | |
451 | root_rl: | |
452 | rcu_read_unlock(); | |
453 | return &q->root_rl; | |
454 | } | |
455 | ||
456 | /** | |
457 | * blk_put_rl - put request_list | |
458 | * @rl: request_list to put | |
459 | * | |
460 | * Put the reference acquired by blk_get_rl(). Should be called under | |
461 | * queue_lock. | |
462 | */ | |
463 | static inline void blk_put_rl(struct request_list *rl) | |
464 | { | |
465 | if (rl->blkg->blkcg != &blkcg_root) | |
466 | blkg_put(rl->blkg); | |
467 | } | |
468 | ||
469 | /** | |
470 | * blk_rq_set_rl - associate a request with a request_list | |
471 | * @rq: request of interest | |
472 | * @rl: target request_list | |
473 | * | |
474 | * Associate @rq with @rl so that accounting and freeing can know the | |
475 | * request_list @rq came from. | |
476 | */ | |
477 | static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) | |
478 | { | |
479 | rq->rl = rl; | |
480 | } | |
481 | ||
482 | /** | |
483 | * blk_rq_rl - return the request_list a request came from | |
484 | * @rq: request of interest | |
485 | * | |
486 | * Return the request_list @rq is allocated from. | |
487 | */ | |
488 | static inline struct request_list *blk_rq_rl(struct request *rq) | |
489 | { | |
490 | return rq->rl; | |
491 | } | |
492 | ||
493 | struct request_list *__blk_queue_next_rl(struct request_list *rl, | |
494 | struct request_queue *q); | |
495 | /** | |
496 | * blk_queue_for_each_rl - iterate through all request_lists of a request_queue | |
497 | * | |
498 | * Should be used under queue_lock. | |
499 | */ | |
500 | #define blk_queue_for_each_rl(rl, q) \ | |
501 | for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q))) | |
502 | ||
503 | static inline int blkg_stat_init(struct blkg_stat *stat, gfp_t gfp) | |
504 | { | |
505 | int ret; | |
506 | ||
507 | ret = percpu_counter_init(&stat->cpu_cnt, 0, gfp); | |
508 | if (ret) | |
509 | return ret; | |
510 | ||
511 | atomic64_set(&stat->aux_cnt, 0); | |
512 | return 0; | |
513 | } | |
514 | ||
515 | static inline void blkg_stat_exit(struct blkg_stat *stat) | |
516 | { | |
517 | percpu_counter_destroy(&stat->cpu_cnt); | |
518 | } | |
519 | ||
520 | /** | |
521 | * blkg_stat_add - add a value to a blkg_stat | |
522 | * @stat: target blkg_stat | |
523 | * @val: value to add | |
524 | * | |
525 | * Add @val to @stat. The caller must ensure that IRQ on the same CPU | |
526 | * don't re-enter this function for the same counter. | |
527 | */ | |
528 | static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val) | |
529 | { | |
530 | __percpu_counter_add(&stat->cpu_cnt, val, BLKG_STAT_CPU_BATCH); | |
531 | } | |
532 | ||
533 | /** | |
534 | * blkg_stat_read - read the current value of a blkg_stat | |
535 | * @stat: blkg_stat to read | |
536 | */ | |
537 | static inline uint64_t blkg_stat_read(struct blkg_stat *stat) | |
538 | { | |
539 | return percpu_counter_sum_positive(&stat->cpu_cnt); | |
540 | } | |
541 | ||
542 | /** | |
543 | * blkg_stat_reset - reset a blkg_stat | |
544 | * @stat: blkg_stat to reset | |
545 | */ | |
546 | static inline void blkg_stat_reset(struct blkg_stat *stat) | |
547 | { | |
548 | percpu_counter_set(&stat->cpu_cnt, 0); | |
549 | atomic64_set(&stat->aux_cnt, 0); | |
550 | } | |
551 | ||
552 | /** | |
553 | * blkg_stat_add_aux - add a blkg_stat into another's aux count | |
554 | * @to: the destination blkg_stat | |
555 | * @from: the source | |
556 | * | |
557 | * Add @from's count including the aux one to @to's aux count. | |
558 | */ | |
559 | static inline void blkg_stat_add_aux(struct blkg_stat *to, | |
560 | struct blkg_stat *from) | |
561 | { | |
562 | atomic64_add(blkg_stat_read(from) + atomic64_read(&from->aux_cnt), | |
563 | &to->aux_cnt); | |
564 | } | |
565 | ||
566 | static inline int blkg_rwstat_init(struct blkg_rwstat *rwstat, gfp_t gfp) | |
567 | { | |
568 | int i, ret; | |
569 | ||
570 | for (i = 0; i < BLKG_RWSTAT_NR; i++) { | |
571 | ret = percpu_counter_init(&rwstat->cpu_cnt[i], 0, gfp); | |
572 | if (ret) { | |
573 | while (--i >= 0) | |
574 | percpu_counter_destroy(&rwstat->cpu_cnt[i]); | |
575 | return ret; | |
576 | } | |
577 | atomic64_set(&rwstat->aux_cnt[i], 0); | |
578 | } | |
579 | return 0; | |
580 | } | |
581 | ||
582 | static inline void blkg_rwstat_exit(struct blkg_rwstat *rwstat) | |
583 | { | |
584 | int i; | |
585 | ||
586 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
587 | percpu_counter_destroy(&rwstat->cpu_cnt[i]); | |
588 | } | |
589 | ||
590 | /** | |
591 | * blkg_rwstat_add - add a value to a blkg_rwstat | |
592 | * @rwstat: target blkg_rwstat | |
593 | * @rw: mask of REQ_{WRITE|SYNC} | |
594 | * @val: value to add | |
595 | * | |
596 | * Add @val to @rwstat. The counters are chosen according to @rw. The | |
597 | * caller is responsible for synchronizing calls to this function. | |
598 | */ | |
599 | static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat, | |
600 | int rw, uint64_t val) | |
601 | { | |
602 | struct percpu_counter *cnt; | |
603 | ||
604 | if (rw & REQ_WRITE) | |
605 | cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_WRITE]; | |
606 | else | |
607 | cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_READ]; | |
608 | ||
609 | __percpu_counter_add(cnt, val, BLKG_STAT_CPU_BATCH); | |
610 | ||
611 | if (rw & REQ_SYNC) | |
612 | cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_SYNC]; | |
613 | else | |
614 | cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_ASYNC]; | |
615 | ||
616 | __percpu_counter_add(cnt, val, BLKG_STAT_CPU_BATCH); | |
617 | } | |
618 | ||
619 | /** | |
620 | * blkg_rwstat_read - read the current values of a blkg_rwstat | |
621 | * @rwstat: blkg_rwstat to read | |
622 | * | |
623 | * Read the current snapshot of @rwstat and return it in the aux counts. | |
624 | */ | |
625 | static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat) | |
626 | { | |
627 | struct blkg_rwstat result; | |
628 | int i; | |
629 | ||
630 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
631 | atomic64_set(&result.aux_cnt[i], | |
632 | percpu_counter_sum_positive(&rwstat->cpu_cnt[i])); | |
633 | return result; | |
634 | } | |
635 | ||
636 | /** | |
637 | * blkg_rwstat_total - read the total count of a blkg_rwstat | |
638 | * @rwstat: blkg_rwstat to read | |
639 | * | |
640 | * Return the total count of @rwstat regardless of the IO direction. This | |
641 | * function can be called without synchronization and takes care of u64 | |
642 | * atomicity. | |
643 | */ | |
644 | static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat) | |
645 | { | |
646 | struct blkg_rwstat tmp = blkg_rwstat_read(rwstat); | |
647 | ||
648 | return atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_READ]) + | |
649 | atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_WRITE]); | |
650 | } | |
651 | ||
652 | /** | |
653 | * blkg_rwstat_reset - reset a blkg_rwstat | |
654 | * @rwstat: blkg_rwstat to reset | |
655 | */ | |
656 | static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat) | |
657 | { | |
658 | int i; | |
659 | ||
660 | for (i = 0; i < BLKG_RWSTAT_NR; i++) { | |
661 | percpu_counter_set(&rwstat->cpu_cnt[i], 0); | |
662 | atomic64_set(&rwstat->aux_cnt[i], 0); | |
663 | } | |
664 | } | |
665 | ||
666 | /** | |
667 | * blkg_rwstat_add_aux - add a blkg_rwstat into another's aux count | |
668 | * @to: the destination blkg_rwstat | |
669 | * @from: the source | |
670 | * | |
671 | * Add @from's count including the aux one to @to's aux count. | |
672 | */ | |
673 | static inline void blkg_rwstat_add_aux(struct blkg_rwstat *to, | |
674 | struct blkg_rwstat *from) | |
675 | { | |
676 | struct blkg_rwstat v = blkg_rwstat_read(from); | |
677 | int i; | |
678 | ||
679 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
680 | atomic64_add(atomic64_read(&v.aux_cnt[i]) + | |
681 | atomic64_read(&from->aux_cnt[i]), | |
682 | &to->aux_cnt[i]); | |
683 | } | |
684 | ||
685 | #ifdef CONFIG_BLK_DEV_THROTTLING | |
686 | extern bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg, | |
687 | struct bio *bio); | |
688 | #else | |
689 | static inline bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg, | |
690 | struct bio *bio) { return false; } | |
691 | #endif | |
692 | ||
693 | static inline bool blkcg_bio_issue_check(struct request_queue *q, | |
694 | struct bio *bio) | |
695 | { | |
696 | struct blkcg *blkcg; | |
697 | struct blkcg_gq *blkg; | |
698 | bool throtl = false; | |
699 | ||
700 | rcu_read_lock(); | |
701 | blkcg = bio_blkcg(bio); | |
702 | ||
703 | blkg = blkg_lookup(blkcg, q); | |
704 | if (unlikely(!blkg)) { | |
705 | spin_lock_irq(q->queue_lock); | |
706 | blkg = blkg_lookup_create(blkcg, q); | |
707 | if (IS_ERR(blkg)) | |
708 | blkg = NULL; | |
709 | spin_unlock_irq(q->queue_lock); | |
710 | } | |
711 | ||
712 | throtl = blk_throtl_bio(q, blkg, bio); | |
713 | ||
714 | if (!throtl) { | |
715 | blkg = blkg ?: q->root_blkg; | |
716 | blkg_rwstat_add(&blkg->stat_bytes, bio->bi_flags, | |
717 | bio->bi_iter.bi_size); | |
718 | blkg_rwstat_add(&blkg->stat_ios, bio->bi_flags, 1); | |
719 | } | |
720 | ||
721 | rcu_read_unlock(); | |
722 | return !throtl; | |
723 | } | |
724 | ||
725 | #else /* CONFIG_BLK_CGROUP */ | |
726 | ||
727 | struct blkcg { | |
728 | }; | |
729 | ||
730 | struct blkg_policy_data { | |
731 | }; | |
732 | ||
733 | struct blkcg_policy_data { | |
734 | }; | |
735 | ||
736 | struct blkcg_gq { | |
737 | }; | |
738 | ||
739 | struct blkcg_policy { | |
740 | }; | |
741 | ||
742 | #define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL)) | |
743 | ||
744 | static inline struct cgroup_subsys_state * | |
745 | task_get_blkcg_css(struct task_struct *task) | |
746 | { | |
747 | return NULL; | |
748 | } | |
749 | ||
750 | #ifdef CONFIG_BLOCK | |
751 | ||
752 | static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; } | |
753 | static inline int blkcg_init_queue(struct request_queue *q) { return 0; } | |
754 | static inline void blkcg_drain_queue(struct request_queue *q) { } | |
755 | static inline void blkcg_exit_queue(struct request_queue *q) { } | |
756 | static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; } | |
757 | static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { } | |
758 | static inline int blkcg_activate_policy(struct request_queue *q, | |
759 | const struct blkcg_policy *pol) { return 0; } | |
760 | static inline void blkcg_deactivate_policy(struct request_queue *q, | |
761 | const struct blkcg_policy *pol) { } | |
762 | ||
763 | static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; } | |
764 | ||
765 | static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg, | |
766 | struct blkcg_policy *pol) { return NULL; } | |
767 | static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; } | |
768 | static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; } | |
769 | static inline void blkg_get(struct blkcg_gq *blkg) { } | |
770 | static inline void blkg_put(struct blkcg_gq *blkg) { } | |
771 | ||
772 | static inline struct request_list *blk_get_rl(struct request_queue *q, | |
773 | struct bio *bio) { return &q->root_rl; } | |
774 | static inline void blk_put_rl(struct request_list *rl) { } | |
775 | static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) { } | |
776 | static inline struct request_list *blk_rq_rl(struct request *rq) { return &rq->q->root_rl; } | |
777 | ||
778 | static inline bool blkcg_bio_issue_check(struct request_queue *q, | |
779 | struct bio *bio) { return true; } | |
780 | ||
781 | #define blk_queue_for_each_rl(rl, q) \ | |
782 | for ((rl) = &(q)->root_rl; (rl); (rl) = NULL) | |
783 | ||
784 | #endif /* CONFIG_BLOCK */ | |
785 | #endif /* CONFIG_BLK_CGROUP */ | |
786 | #endif /* _BLK_CGROUP_H */ |