1#ifndef _BLK_CGROUP_H
2#define _BLK_CGROUP_H
3
4
5
6
7
8
9
10
11
12
13
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
24#define BLKG_STAT_CPU_BATCH (INT_MAX / 2)
25
26
27#define THROTL_IOPS_MAX UINT_MAX
28
29#ifdef CONFIG_BLK_CGROUP
30
31enum 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
41struct blkcg_gq;
42
43struct blkcg {
44 struct cgroup_subsys_state css;
45 spinlock_t lock;
46
47 struct radix_tree_root blkg_tree;
48 struct blkcg_gq __rcu *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
61
62
63
64struct blkg_stat {
65 struct percpu_counter cpu_cnt;
66 atomic64_t aux_cnt;
67};
68
69struct blkg_rwstat {
70 struct percpu_counter cpu_cnt[BLKG_RWSTAT_NR];
71 atomic64_t aux_cnt[BLKG_RWSTAT_NR];
72};
73
74
75
76
77
78
79
80
81
82
83
84
85struct blkg_policy_data {
86
87 struct blkcg_gq *blkg;
88 int plid;
89};
90
91
92
93
94
95
96
97
98struct blkcg_policy_data {
99
100 struct blkcg *blkcg;
101 int plid;
102};
103
104
105struct blkcg_gq {
106
107 struct request_queue *q;
108 struct list_head q_node;
109 struct hlist_node blkcg_node;
110 struct blkcg *blkcg;
111
112
113
114
115
116 struct bdi_writeback_congested *wb_congested;
117
118
119 struct blkcg_gq *parent;
120
121
122 struct request_list rl;
123
124
125 atomic_t refcnt;
126
127
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
138typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp);
139typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd);
140typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd);
141typedef void (blkcg_pol_bind_cpd_fn)(struct blkcg_policy_data *cpd);
142typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(gfp_t gfp, int node);
143typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd);
144typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd);
145typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd);
146typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd);
147typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd);
148
149struct blkcg_policy {
150 int plid;
151
152 struct cftype *dfl_cftypes;
153 struct cftype *legacy_cftypes;
154
155
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
169extern struct blkcg blkcg_root;
170extern struct cgroup_subsys_state * const blkcg_root_css;
171
172struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
173 struct request_queue *q, bool update_hint);
174struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
175 struct request_queue *q);
176int blkcg_init_queue(struct request_queue *q);
177void blkcg_drain_queue(struct request_queue *q);
178void blkcg_exit_queue(struct request_queue *q);
179
180
181int blkcg_policy_register(struct blkcg_policy *pol);
182void blkcg_policy_unregister(struct blkcg_policy *pol);
183int blkcg_activate_policy(struct request_queue *q,
184 const struct blkcg_policy *pol);
185void blkcg_deactivate_policy(struct request_queue *q,
186 const struct blkcg_policy *pol);
187
188const char *blkg_dev_name(struct blkcg_gq *blkg);
189void 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);
194u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
195u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
196 const struct blkg_rwstat *rwstat);
197u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off);
198u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
199 int off);
200int blkg_print_stat_bytes(struct seq_file *sf, void *v);
201int blkg_print_stat_ios(struct seq_file *sf, void *v);
202int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v);
203int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v);
204
205u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg,
206 struct blkcg_policy *pol, int off);
207struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg,
208 struct blkcg_policy *pol, int off);
209
210struct blkg_conf_ctx {
211 struct gendisk *disk;
212 struct blkcg_gq *blkg;
213 char *body;
214};
215
216int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
217 char *input, struct blkg_conf_ctx *ctx);
218void blkg_conf_finish(struct blkg_conf_ctx *ctx);
219
220
221static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
222{
223 return css ? container_of(css, struct blkcg, css) : NULL;
224}
225
226static inline struct blkcg *task_blkcg(struct task_struct *tsk)
227{
228 return css_to_blkcg(task_css(tsk, io_cgrp_id));
229}
230
231static 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
238static inline struct cgroup_subsys_state *
239task_get_blkcg_css(struct task_struct *task)
240{
241 return task_get_css(task, io_cgrp_id);
242}
243
244
245
246
247
248
249
250static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
251{
252 return css_to_blkcg(blkcg->css.parent);
253}
254
255
256
257
258
259
260
261
262
263
264
265
266static 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
284
285
286
287
288
289
290
291static 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
303
304
305
306
307
308static 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
314static 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
322
323
324
325
326static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
327{
328 return pd ? pd->blkg : NULL;
329}
330
331static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd)
332{
333 return cpd ? cpd->blkcg : NULL;
334}
335
336
337
338
339
340
341
342
343
344static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
345{
346 return cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
347}
348
349
350
351
352
353
354
355static inline void blkg_get(struct blkcg_gq *blkg)
356{
357 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
358 atomic_inc(&blkg->refcnt);
359}
360
361void __blkg_release_rcu(struct rcu_head *rcu);
362
363
364
365
366
367static inline void blkg_put(struct blkcg_gq *blkg)
368{
369 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
370 if (atomic_dec_and_test(&blkg->refcnt))
371 call_rcu(&blkg->rcu_head, __blkg_release_rcu);
372}
373
374
375
376
377
378
379
380
381
382
383
384
385
386#define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
387 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
388 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
389 (p_blkg)->q, false)))
390
391
392
393
394
395
396
397
398
399
400
401#define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
402 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
403 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
404 (p_blkg)->q, false)))
405
406
407
408
409
410
411
412
413
414
415
416static inline struct request_list *blk_get_rl(struct request_queue *q,
417 struct bio *bio)
418{
419 struct blkcg *blkcg;
420 struct blkcg_gq *blkg;
421
422 rcu_read_lock();
423
424 blkcg = bio_blkcg(bio);
425
426
427 if (blkcg == &blkcg_root)
428 goto root_rl;
429
430
431
432
433
434
435 blkg = blkg_lookup(blkcg, q);
436 if (unlikely(!blkg))
437 goto root_rl;
438
439 blkg_get(blkg);
440 rcu_read_unlock();
441 return &blkg->rl;
442root_rl:
443 rcu_read_unlock();
444 return &q->root_rl;
445}
446
447
448
449
450
451
452
453
454static inline void blk_put_rl(struct request_list *rl)
455{
456 if (rl->blkg->blkcg != &blkcg_root)
457 blkg_put(rl->blkg);
458}
459
460
461
462
463
464
465
466
467
468static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl)
469{
470 rq->rl = rl;
471}
472
473
474
475
476
477
478
479static inline struct request_list *blk_rq_rl(struct request *rq)
480{
481 return rq->rl;
482}
483
484struct request_list *__blk_queue_next_rl(struct request_list *rl,
485 struct request_queue *q);
486
487
488
489
490
491#define blk_queue_for_each_rl(rl, q) \
492 for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
493
494static inline int blkg_stat_init(struct blkg_stat *stat, gfp_t gfp)
495{
496 int ret;
497
498 ret = percpu_counter_init(&stat->cpu_cnt, 0, gfp);
499 if (ret)
500 return ret;
501
502 atomic64_set(&stat->aux_cnt, 0);
503 return 0;
504}
505
506static inline void blkg_stat_exit(struct blkg_stat *stat)
507{
508 percpu_counter_destroy(&stat->cpu_cnt);
509}
510
511
512
513
514
515
516
517
518
519static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
520{
521 percpu_counter_add_batch(&stat->cpu_cnt, val, BLKG_STAT_CPU_BATCH);
522}
523
524
525
526
527
528static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
529{
530 return percpu_counter_sum_positive(&stat->cpu_cnt);
531}
532
533
534
535
536
537static inline void blkg_stat_reset(struct blkg_stat *stat)
538{
539 percpu_counter_set(&stat->cpu_cnt, 0);
540 atomic64_set(&stat->aux_cnt, 0);
541}
542
543
544
545
546
547
548
549
550static inline void blkg_stat_add_aux(struct blkg_stat *to,
551 struct blkg_stat *from)
552{
553 atomic64_add(blkg_stat_read(from) + atomic64_read(&from->aux_cnt),
554 &to->aux_cnt);
555}
556
557static inline int blkg_rwstat_init(struct blkg_rwstat *rwstat, gfp_t gfp)
558{
559 int i, ret;
560
561 for (i = 0; i < BLKG_RWSTAT_NR; i++) {
562 ret = percpu_counter_init(&rwstat->cpu_cnt[i], 0, gfp);
563 if (ret) {
564 while (--i >= 0)
565 percpu_counter_destroy(&rwstat->cpu_cnt[i]);
566 return ret;
567 }
568 atomic64_set(&rwstat->aux_cnt[i], 0);
569 }
570 return 0;
571}
572
573static inline void blkg_rwstat_exit(struct blkg_rwstat *rwstat)
574{
575 int i;
576
577 for (i = 0; i < BLKG_RWSTAT_NR; i++)
578 percpu_counter_destroy(&rwstat->cpu_cnt[i]);
579}
580
581
582
583
584
585
586
587
588
589
590static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
591 unsigned int op, uint64_t val)
592{
593 struct percpu_counter *cnt;
594
595 if (op_is_write(op))
596 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_WRITE];
597 else
598 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_READ];
599
600 percpu_counter_add_batch(cnt, val, BLKG_STAT_CPU_BATCH);
601
602 if (op_is_sync(op))
603 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_SYNC];
604 else
605 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_ASYNC];
606
607 percpu_counter_add_batch(cnt, val, BLKG_STAT_CPU_BATCH);
608}
609
610
611
612
613
614
615
616static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
617{
618 struct blkg_rwstat result;
619 int i;
620
621 for (i = 0; i < BLKG_RWSTAT_NR; i++)
622 atomic64_set(&result.aux_cnt[i],
623 percpu_counter_sum_positive(&rwstat->cpu_cnt[i]));
624 return result;
625}
626
627
628
629
630
631
632
633
634
635static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat)
636{
637 struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
638
639 return atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_READ]) +
640 atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_WRITE]);
641}
642
643
644
645
646
647static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
648{
649 int i;
650
651 for (i = 0; i < BLKG_RWSTAT_NR; i++) {
652 percpu_counter_set(&rwstat->cpu_cnt[i], 0);
653 atomic64_set(&rwstat->aux_cnt[i], 0);
654 }
655}
656
657
658
659
660
661
662
663
664static inline void blkg_rwstat_add_aux(struct blkg_rwstat *to,
665 struct blkg_rwstat *from)
666{
667 struct blkg_rwstat v = blkg_rwstat_read(from);
668 int i;
669
670 for (i = 0; i < BLKG_RWSTAT_NR; i++)
671 atomic64_add(atomic64_read(&v.aux_cnt[i]) +
672 atomic64_read(&from->aux_cnt[i]),
673 &to->aux_cnt[i]);
674}
675
676#ifdef CONFIG_BLK_DEV_THROTTLING
677extern bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg,
678 struct bio *bio);
679#else
680static inline bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg,
681 struct bio *bio) { return false; }
682#endif
683
684static inline bool blkcg_bio_issue_check(struct request_queue *q,
685 struct bio *bio)
686{
687 struct blkcg *blkcg;
688 struct blkcg_gq *blkg;
689 bool throtl = false;
690
691 rcu_read_lock();
692 blkcg = bio_blkcg(bio);
693
694 blkg = blkg_lookup(blkcg, q);
695 if (unlikely(!blkg)) {
696 spin_lock_irq(q->queue_lock);
697 blkg = blkg_lookup_create(blkcg, q);
698 if (IS_ERR(blkg))
699 blkg = NULL;
700 spin_unlock_irq(q->queue_lock);
701 }
702
703 throtl = blk_throtl_bio(q, blkg, bio);
704
705 if (!throtl) {
706 blkg = blkg ?: q->root_blkg;
707 blkg_rwstat_add(&blkg->stat_bytes, bio->bi_opf,
708 bio->bi_iter.bi_size);
709 blkg_rwstat_add(&blkg->stat_ios, bio->bi_opf, 1);
710 }
711
712 rcu_read_unlock();
713 return !throtl;
714}
715
716#else
717
718struct blkcg {
719};
720
721struct blkg_policy_data {
722};
723
724struct blkcg_policy_data {
725};
726
727struct blkcg_gq {
728};
729
730struct blkcg_policy {
731};
732
733#define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
734
735static inline struct cgroup_subsys_state *
736task_get_blkcg_css(struct task_struct *task)
737{
738 return NULL;
739}
740
741#ifdef CONFIG_BLOCK
742
743static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
744static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
745static inline void blkcg_drain_queue(struct request_queue *q) { }
746static inline void blkcg_exit_queue(struct request_queue *q) { }
747static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
748static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
749static inline int blkcg_activate_policy(struct request_queue *q,
750 const struct blkcg_policy *pol) { return 0; }
751static inline void blkcg_deactivate_policy(struct request_queue *q,
752 const struct blkcg_policy *pol) { }
753
754static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
755
756static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
757 struct blkcg_policy *pol) { return NULL; }
758static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
759static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
760static inline void blkg_get(struct blkcg_gq *blkg) { }
761static inline void blkg_put(struct blkcg_gq *blkg) { }
762
763static inline struct request_list *blk_get_rl(struct request_queue *q,
764 struct bio *bio) { return &q->root_rl; }
765static inline void blk_put_rl(struct request_list *rl) { }
766static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) { }
767static inline struct request_list *blk_rq_rl(struct request *rq) { return &rq->q->root_rl; }
768
769static inline bool blkcg_bio_issue_check(struct request_queue *q,
770 struct bio *bio) { return true; }
771
772#define blk_queue_for_each_rl(rl, q) \
773 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
774
775#endif
776#endif
777#endif
778