1
2#ifndef _BLK_CGROUP_H
3#define _BLK_CGROUP_H
4
5
6
7
8
9
10
11
12
13
14
15
16
17#include <linux/cgroup.h>
18#include <linux/percpu_counter.h>
19#include <linux/seq_file.h>
20#include <linux/radix-tree.h>
21#include <linux/blkdev.h>
22#include <linux/atomic.h>
23
24
25#define BLKG_STAT_CPU_BATCH (INT_MAX / 2)
26
27
28#define THROTL_IOPS_MAX UINT_MAX
29
30#ifdef CONFIG_BLK_CGROUP
31
32enum blkg_rwstat_type {
33 BLKG_RWSTAT_READ,
34 BLKG_RWSTAT_WRITE,
35 BLKG_RWSTAT_SYNC,
36 BLKG_RWSTAT_ASYNC,
37
38 BLKG_RWSTAT_NR,
39 BLKG_RWSTAT_TOTAL = BLKG_RWSTAT_NR,
40};
41
42struct blkcg_gq;
43
44struct blkcg {
45 struct cgroup_subsys_state css;
46 spinlock_t lock;
47
48 struct radix_tree_root blkg_tree;
49 struct blkcg_gq __rcu *blkg_hint;
50 struct hlist_head blkg_list;
51
52 struct blkcg_policy_data *cpd[BLKCG_MAX_POLS];
53
54 struct list_head all_blkcgs_node;
55#ifdef CONFIG_CGROUP_WRITEBACK
56 struct list_head cgwb_list;
57#endif
58};
59
60
61
62
63
64
65struct blkg_stat {
66 struct percpu_counter cpu_cnt;
67 atomic64_t aux_cnt;
68};
69
70struct blkg_rwstat {
71 struct percpu_counter cpu_cnt[BLKG_RWSTAT_NR];
72 atomic64_t aux_cnt[BLKG_RWSTAT_NR];
73};
74
75
76
77
78
79
80
81
82
83
84
85
86struct blkg_policy_data {
87
88 struct blkcg_gq *blkg;
89 int plid;
90};
91
92
93
94
95
96
97
98
99struct blkcg_policy_data {
100
101 struct blkcg *blkcg;
102 int plid;
103};
104
105
106struct blkcg_gq {
107
108 struct request_queue *q;
109 struct list_head q_node;
110 struct hlist_node blkcg_node;
111 struct blkcg *blkcg;
112
113
114
115
116
117 struct bdi_writeback_congested *wb_congested;
118
119
120 struct blkcg_gq *parent;
121
122
123 struct request_list rl;
124
125
126 atomic_t refcnt;
127
128
129 bool online;
130
131 struct blkg_rwstat stat_bytes;
132 struct blkg_rwstat stat_ios;
133
134 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
135
136 struct rcu_head rcu_head;
137};
138
139typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp);
140typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd);
141typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd);
142typedef void (blkcg_pol_bind_cpd_fn)(struct blkcg_policy_data *cpd);
143typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(gfp_t gfp, int node);
144typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd);
145typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd);
146typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd);
147typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd);
148typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd);
149
150struct blkcg_policy {
151 int plid;
152
153 struct cftype *dfl_cftypes;
154 struct cftype *legacy_cftypes;
155
156
157 blkcg_pol_alloc_cpd_fn *cpd_alloc_fn;
158 blkcg_pol_init_cpd_fn *cpd_init_fn;
159 blkcg_pol_free_cpd_fn *cpd_free_fn;
160 blkcg_pol_bind_cpd_fn *cpd_bind_fn;
161
162 blkcg_pol_alloc_pd_fn *pd_alloc_fn;
163 blkcg_pol_init_pd_fn *pd_init_fn;
164 blkcg_pol_online_pd_fn *pd_online_fn;
165 blkcg_pol_offline_pd_fn *pd_offline_fn;
166 blkcg_pol_free_pd_fn *pd_free_fn;
167 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;
168};
169
170extern struct blkcg blkcg_root;
171extern struct cgroup_subsys_state * const blkcg_root_css;
172
173struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
174 struct request_queue *q, bool update_hint);
175struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
176 struct request_queue *q);
177int blkcg_init_queue(struct request_queue *q);
178void blkcg_drain_queue(struct request_queue *q);
179void blkcg_exit_queue(struct request_queue *q);
180
181
182int blkcg_policy_register(struct blkcg_policy *pol);
183void blkcg_policy_unregister(struct blkcg_policy *pol);
184int blkcg_activate_policy(struct request_queue *q,
185 const struct blkcg_policy *pol);
186void blkcg_deactivate_policy(struct request_queue *q,
187 const struct blkcg_policy *pol);
188
189const char *blkg_dev_name(struct blkcg_gq *blkg);
190void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
191 u64 (*prfill)(struct seq_file *,
192 struct blkg_policy_data *, int),
193 const struct blkcg_policy *pol, int data,
194 bool show_total);
195u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
196u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
197 const struct blkg_rwstat *rwstat);
198u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off);
199u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
200 int off);
201int blkg_print_stat_bytes(struct seq_file *sf, void *v);
202int blkg_print_stat_ios(struct seq_file *sf, void *v);
203int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v);
204int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v);
205
206u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg,
207 struct blkcg_policy *pol, int off);
208struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg,
209 struct blkcg_policy *pol, int off);
210
211struct blkg_conf_ctx {
212 struct gendisk *disk;
213 struct blkcg_gq *blkg;
214 char *body;
215};
216
217int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
218 char *input, struct blkg_conf_ctx *ctx);
219void blkg_conf_finish(struct blkg_conf_ctx *ctx);
220
221
222static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
223{
224 return css ? container_of(css, struct blkcg, css) : NULL;
225}
226
227static inline struct blkcg *task_blkcg(struct task_struct *tsk)
228{
229 return css_to_blkcg(task_css(tsk, io_cgrp_id));
230}
231
232static inline struct blkcg *bio_blkcg(struct bio *bio)
233{
234 if (bio && bio->bi_css)
235 return css_to_blkcg(bio->bi_css);
236 return task_blkcg(current);
237}
238
239static inline struct cgroup_subsys_state *
240task_get_blkcg_css(struct task_struct *task)
241{
242 return task_get_css(task, io_cgrp_id);
243}
244
245
246
247
248
249
250
251static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
252{
253 return css_to_blkcg(blkcg->css.parent);
254}
255
256
257
258
259
260
261
262
263
264
265
266
267static inline struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
268 struct request_queue *q,
269 bool update_hint)
270{
271 struct blkcg_gq *blkg;
272
273 if (blkcg == &blkcg_root)
274 return q->root_blkg;
275
276 blkg = rcu_dereference(blkcg->blkg_hint);
277 if (blkg && blkg->q == q)
278 return blkg;
279
280 return blkg_lookup_slowpath(blkcg, q, update_hint);
281}
282
283
284
285
286
287
288
289
290
291
292static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg,
293 struct request_queue *q)
294{
295 WARN_ON_ONCE(!rcu_read_lock_held());
296
297 if (unlikely(blk_queue_bypass(q)))
298 return NULL;
299 return __blkg_lookup(blkcg, q, false);
300}
301
302
303
304
305
306
307
308
309static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
310 struct blkcg_policy *pol)
311{
312 return blkg ? blkg->pd[pol->plid] : NULL;
313}
314
315static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,
316 struct blkcg_policy *pol)
317{
318 return blkcg ? blkcg->cpd[pol->plid] : NULL;
319}
320
321
322
323
324
325
326
327static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
328{
329 return pd ? pd->blkg : NULL;
330}
331
332static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd)
333{
334 return cpd ? cpd->blkcg : NULL;
335}
336
337
338
339
340
341
342
343
344
345static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
346{
347 return cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
348}
349
350
351
352
353
354
355
356static inline void blkg_get(struct blkcg_gq *blkg)
357{
358 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
359 atomic_inc(&blkg->refcnt);
360}
361
362void __blkg_release_rcu(struct rcu_head *rcu);
363
364
365
366
367
368static inline void blkg_put(struct blkcg_gq *blkg)
369{
370 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
371 if (atomic_dec_and_test(&blkg->refcnt))
372 call_rcu(&blkg->rcu_head, __blkg_release_rcu);
373}
374
375
376
377
378
379
380
381
382
383
384
385
386
387#define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
388 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
389 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
390 (p_blkg)->q, false)))
391
392
393
394
395
396
397
398
399
400
401
402#define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
403 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
404 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
405 (p_blkg)->q, false)))
406
407
408
409
410
411
412
413
414
415
416
417static inline struct request_list *blk_get_rl(struct request_queue *q,
418 struct bio *bio)
419{
420 struct blkcg *blkcg;
421 struct blkcg_gq *blkg;
422
423 rcu_read_lock();
424
425 blkcg = bio_blkcg(bio);
426
427
428 if (blkcg == &blkcg_root)
429 goto root_rl;
430
431
432
433
434
435
436 blkg = blkg_lookup(blkcg, q);
437 if (unlikely(!blkg))
438 goto root_rl;
439
440 blkg_get(blkg);
441 rcu_read_unlock();
442 return &blkg->rl;
443root_rl:
444 rcu_read_unlock();
445 return &q->root_rl;
446}
447
448
449
450
451
452
453
454
455static inline void blk_put_rl(struct request_list *rl)
456{
457 if (rl->blkg->blkcg != &blkcg_root)
458 blkg_put(rl->blkg);
459}
460
461
462
463
464
465
466
467
468
469static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl)
470{
471 rq->rl = rl;
472}
473
474
475
476
477
478
479
480static inline struct request_list *blk_rq_rl(struct request *rq)
481{
482 return rq->rl;
483}
484
485struct request_list *__blk_queue_next_rl(struct request_list *rl,
486 struct request_queue *q);
487
488
489
490
491
492#define blk_queue_for_each_rl(rl, q) \
493 for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
494
495static inline int blkg_stat_init(struct blkg_stat *stat, gfp_t gfp)
496{
497 int ret;
498
499 ret = percpu_counter_init(&stat->cpu_cnt, 0, gfp);
500 if (ret)
501 return ret;
502
503 atomic64_set(&stat->aux_cnt, 0);
504 return 0;
505}
506
507static inline void blkg_stat_exit(struct blkg_stat *stat)
508{
509 percpu_counter_destroy(&stat->cpu_cnt);
510}
511
512
513
514
515
516
517
518
519
520static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
521{
522 percpu_counter_add_batch(&stat->cpu_cnt, val, BLKG_STAT_CPU_BATCH);
523}
524
525
526
527
528
529static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
530{
531 return percpu_counter_sum_positive(&stat->cpu_cnt);
532}
533
534
535
536
537
538static inline void blkg_stat_reset(struct blkg_stat *stat)
539{
540 percpu_counter_set(&stat->cpu_cnt, 0);
541 atomic64_set(&stat->aux_cnt, 0);
542}
543
544
545
546
547
548
549
550
551static inline void blkg_stat_add_aux(struct blkg_stat *to,
552 struct blkg_stat *from)
553{
554 atomic64_add(blkg_stat_read(from) + atomic64_read(&from->aux_cnt),
555 &to->aux_cnt);
556}
557
558static inline int blkg_rwstat_init(struct blkg_rwstat *rwstat, gfp_t gfp)
559{
560 int i, ret;
561
562 for (i = 0; i < BLKG_RWSTAT_NR; i++) {
563 ret = percpu_counter_init(&rwstat->cpu_cnt[i], 0, gfp);
564 if (ret) {
565 while (--i >= 0)
566 percpu_counter_destroy(&rwstat->cpu_cnt[i]);
567 return ret;
568 }
569 atomic64_set(&rwstat->aux_cnt[i], 0);
570 }
571 return 0;
572}
573
574static inline void blkg_rwstat_exit(struct blkg_rwstat *rwstat)
575{
576 int i;
577
578 for (i = 0; i < BLKG_RWSTAT_NR; i++)
579 percpu_counter_destroy(&rwstat->cpu_cnt[i]);
580}
581
582
583
584
585
586
587
588
589
590
591static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
592 unsigned int op, uint64_t val)
593{
594 struct percpu_counter *cnt;
595
596 if (op_is_write(op))
597 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_WRITE];
598 else
599 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_READ];
600
601 percpu_counter_add_batch(cnt, val, BLKG_STAT_CPU_BATCH);
602
603 if (op_is_sync(op))
604 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_SYNC];
605 else
606 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_ASYNC];
607
608 percpu_counter_add_batch(cnt, val, BLKG_STAT_CPU_BATCH);
609}
610
611
612
613
614
615
616
617static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
618{
619 struct blkg_rwstat result;
620 int i;
621
622 for (i = 0; i < BLKG_RWSTAT_NR; i++)
623 atomic64_set(&result.aux_cnt[i],
624 percpu_counter_sum_positive(&rwstat->cpu_cnt[i]));
625 return result;
626}
627
628
629
630
631
632
633
634
635
636static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat)
637{
638 struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
639
640 return atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_READ]) +
641 atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_WRITE]);
642}
643
644
645
646
647
648static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
649{
650 int i;
651
652 for (i = 0; i < BLKG_RWSTAT_NR; i++) {
653 percpu_counter_set(&rwstat->cpu_cnt[i], 0);
654 atomic64_set(&rwstat->aux_cnt[i], 0);
655 }
656}
657
658
659
660
661
662
663
664
665static inline void blkg_rwstat_add_aux(struct blkg_rwstat *to,
666 struct blkg_rwstat *from)
667{
668 struct blkg_rwstat v = blkg_rwstat_read(from);
669 int i;
670
671 for (i = 0; i < BLKG_RWSTAT_NR; i++)
672 atomic64_add(atomic64_read(&v.aux_cnt[i]) +
673 atomic64_read(&from->aux_cnt[i]),
674 &to->aux_cnt[i]);
675}
676
677#ifdef CONFIG_BLK_DEV_THROTTLING
678extern bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg,
679 struct bio *bio);
680#else
681static inline bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg,
682 struct bio *bio) { return false; }
683#endif
684
685static inline bool blkcg_bio_issue_check(struct request_queue *q,
686 struct bio *bio)
687{
688 struct blkcg *blkcg;
689 struct blkcg_gq *blkg;
690 bool throtl = false;
691
692 rcu_read_lock();
693 blkcg = bio_blkcg(bio);
694
695
696 bio_associate_blkcg(bio, &blkcg->css);
697
698 blkg = blkg_lookup(blkcg, q);
699 if (unlikely(!blkg)) {
700 spin_lock_irq(q->queue_lock);
701 blkg = blkg_lookup_create(blkcg, q);
702 if (IS_ERR(blkg))
703 blkg = NULL;
704 spin_unlock_irq(q->queue_lock);
705 }
706
707 throtl = blk_throtl_bio(q, blkg, bio);
708
709 if (!throtl) {
710 blkg = blkg ?: q->root_blkg;
711 blkg_rwstat_add(&blkg->stat_bytes, bio->bi_opf,
712 bio->bi_iter.bi_size);
713 blkg_rwstat_add(&blkg->stat_ios, bio->bi_opf, 1);
714 }
715
716 rcu_read_unlock();
717 return !throtl;
718}
719
720#else
721
722struct blkcg {
723};
724
725struct blkg_policy_data {
726};
727
728struct blkcg_policy_data {
729};
730
731struct blkcg_gq {
732};
733
734struct blkcg_policy {
735};
736
737#define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
738
739static inline struct cgroup_subsys_state *
740task_get_blkcg_css(struct task_struct *task)
741{
742 return NULL;
743}
744
745#ifdef CONFIG_BLOCK
746
747static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
748static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
749static inline void blkcg_drain_queue(struct request_queue *q) { }
750static inline void blkcg_exit_queue(struct request_queue *q) { }
751static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
752static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
753static inline int blkcg_activate_policy(struct request_queue *q,
754 const struct blkcg_policy *pol) { return 0; }
755static inline void blkcg_deactivate_policy(struct request_queue *q,
756 const struct blkcg_policy *pol) { }
757
758static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
759
760static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
761 struct blkcg_policy *pol) { return NULL; }
762static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
763static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
764static inline void blkg_get(struct blkcg_gq *blkg) { }
765static inline void blkg_put(struct blkcg_gq *blkg) { }
766
767static inline struct request_list *blk_get_rl(struct request_queue *q,
768 struct bio *bio) { return &q->root_rl; }
769static inline void blk_put_rl(struct request_list *rl) { }
770static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) { }
771static inline struct request_list *blk_rq_rl(struct request *rq) { return &rq->q->root_rl; }
772
773static inline bool blkcg_bio_issue_check(struct request_queue *q,
774 struct bio *bio) { return true; }
775
776#define blk_queue_for_each_rl(rl, q) \
777 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
778
779#endif
780#endif
781#endif
782