1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17#include <linux/ioprio.h>
18#include <linux/kdev_t.h>
19#include <linux/module.h>
20#include <linux/sched/signal.h>
21#include <linux/err.h>
22#include <linux/blkdev.h>
23#include <linux/backing-dev.h>
24#include <linux/slab.h>
25#include <linux/genhd.h>
26#include <linux/delay.h>
27#include <linux/atomic.h>
28#include <linux/ctype.h>
29#include <linux/blk-cgroup.h>
30#include "blk.h"
31
32#define MAX_KEY_LEN 100
33
34
35
36
37
38
39
40
41static DEFINE_MUTEX(blkcg_pol_register_mutex);
42static DEFINE_MUTEX(blkcg_pol_mutex);
43
44struct blkcg blkcg_root;
45EXPORT_SYMBOL_GPL(blkcg_root);
46
47struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css;
48
49static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS];
50
51static LIST_HEAD(all_blkcgs);
52
53static bool blkcg_policy_enabled(struct request_queue *q,
54 const struct blkcg_policy *pol)
55{
56 return pol && test_bit(pol->plid, q->blkcg_pols);
57}
58
59
60
61
62
63
64
65static void blkg_free(struct blkcg_gq *blkg)
66{
67 int i;
68
69 if (!blkg)
70 return;
71
72 for (i = 0; i < BLKCG_MAX_POLS; i++)
73 if (blkg->pd[i])
74 blkcg_policy[i]->pd_free_fn(blkg->pd[i]);
75
76 if (blkg->blkcg != &blkcg_root)
77 blk_exit_rl(blkg->q, &blkg->rl);
78
79 blkg_rwstat_exit(&blkg->stat_ios);
80 blkg_rwstat_exit(&blkg->stat_bytes);
81 kfree(blkg);
82}
83
84
85
86
87
88
89
90
91
92static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q,
93 gfp_t gfp_mask)
94{
95 struct blkcg_gq *blkg;
96 int i;
97
98
99 blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node);
100 if (!blkg)
101 return NULL;
102
103 if (blkg_rwstat_init(&blkg->stat_bytes, gfp_mask) ||
104 blkg_rwstat_init(&blkg->stat_ios, gfp_mask))
105 goto err_free;
106
107 blkg->q = q;
108 INIT_LIST_HEAD(&blkg->q_node);
109 blkg->blkcg = blkcg;
110 atomic_set(&blkg->refcnt, 1);
111
112
113 if (blkcg != &blkcg_root) {
114 if (blk_init_rl(&blkg->rl, q, gfp_mask))
115 goto err_free;
116 blkg->rl.blkg = blkg;
117 }
118
119 for (i = 0; i < BLKCG_MAX_POLS; i++) {
120 struct blkcg_policy *pol = blkcg_policy[i];
121 struct blkg_policy_data *pd;
122
123 if (!blkcg_policy_enabled(q, pol))
124 continue;
125
126
127 pd = pol->pd_alloc_fn(gfp_mask, q->node);
128 if (!pd)
129 goto err_free;
130
131 blkg->pd[i] = pd;
132 pd->blkg = blkg;
133 pd->plid = i;
134 }
135
136 return blkg;
137
138err_free:
139 blkg_free(blkg);
140 return NULL;
141}
142
143struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
144 struct request_queue *q, bool update_hint)
145{
146 struct blkcg_gq *blkg;
147
148
149
150
151
152
153
154 blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
155 if (blkg && blkg->q == q) {
156 if (update_hint) {
157 lockdep_assert_held(q->queue_lock);
158 rcu_assign_pointer(blkcg->blkg_hint, blkg);
159 }
160 return blkg;
161 }
162
163 return NULL;
164}
165EXPORT_SYMBOL_GPL(blkg_lookup_slowpath);
166
167
168
169
170
171static struct blkcg_gq *blkg_create(struct blkcg *blkcg,
172 struct request_queue *q,
173 struct blkcg_gq *new_blkg)
174{
175 struct blkcg_gq *blkg;
176 struct bdi_writeback_congested *wb_congested;
177 int i, ret;
178
179 WARN_ON_ONCE(!rcu_read_lock_held());
180 lockdep_assert_held(q->queue_lock);
181
182
183 if (!css_tryget_online(&blkcg->css)) {
184 ret = -ENODEV;
185 goto err_free_blkg;
186 }
187
188 wb_congested = wb_congested_get_create(q->backing_dev_info,
189 blkcg->css.id,
190 GFP_NOWAIT | __GFP_NOWARN);
191 if (!wb_congested) {
192 ret = -ENOMEM;
193 goto err_put_css;
194 }
195
196
197 if (!new_blkg) {
198 new_blkg = blkg_alloc(blkcg, q, GFP_NOWAIT | __GFP_NOWARN);
199 if (unlikely(!new_blkg)) {
200 ret = -ENOMEM;
201 goto err_put_congested;
202 }
203 }
204 blkg = new_blkg;
205 blkg->wb_congested = wb_congested;
206
207
208 if (blkcg_parent(blkcg)) {
209 blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false);
210 if (WARN_ON_ONCE(!blkg->parent)) {
211 ret = -ENODEV;
212 goto err_put_congested;
213 }
214 blkg_get(blkg->parent);
215 }
216
217
218 for (i = 0; i < BLKCG_MAX_POLS; i++) {
219 struct blkcg_policy *pol = blkcg_policy[i];
220
221 if (blkg->pd[i] && pol->pd_init_fn)
222 pol->pd_init_fn(blkg->pd[i]);
223 }
224
225
226 spin_lock(&blkcg->lock);
227 ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg);
228 if (likely(!ret)) {
229 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
230 list_add(&blkg->q_node, &q->blkg_list);
231
232 for (i = 0; i < BLKCG_MAX_POLS; i++) {
233 struct blkcg_policy *pol = blkcg_policy[i];
234
235 if (blkg->pd[i] && pol->pd_online_fn)
236 pol->pd_online_fn(blkg->pd[i]);
237 }
238 }
239 blkg->online = true;
240 spin_unlock(&blkcg->lock);
241
242 if (!ret)
243 return blkg;
244
245
246 blkg_put(blkg);
247 return ERR_PTR(ret);
248
249err_put_congested:
250 wb_congested_put(wb_congested);
251err_put_css:
252 css_put(&blkcg->css);
253err_free_blkg:
254 blkg_free(new_blkg);
255 return ERR_PTR(ret);
256}
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
273 struct request_queue *q)
274{
275 struct blkcg_gq *blkg;
276
277 WARN_ON_ONCE(!rcu_read_lock_held());
278 lockdep_assert_held(q->queue_lock);
279
280
281
282
283
284 if (unlikely(blk_queue_bypass(q)))
285 return ERR_PTR(blk_queue_dying(q) ? -ENODEV : -EBUSY);
286
287 blkg = __blkg_lookup(blkcg, q, true);
288 if (blkg)
289 return blkg;
290
291
292
293
294
295 while (true) {
296 struct blkcg *pos = blkcg;
297 struct blkcg *parent = blkcg_parent(blkcg);
298
299 while (parent && !__blkg_lookup(parent, q, false)) {
300 pos = parent;
301 parent = blkcg_parent(parent);
302 }
303
304 blkg = blkg_create(pos, q, NULL);
305 if (pos == blkcg || IS_ERR(blkg))
306 return blkg;
307 }
308}
309
310static void blkg_pd_offline(struct blkcg_gq *blkg)
311{
312 int i;
313
314 lockdep_assert_held(blkg->q->queue_lock);
315 lockdep_assert_held(&blkg->blkcg->lock);
316
317 for (i = 0; i < BLKCG_MAX_POLS; i++) {
318 struct blkcg_policy *pol = blkcg_policy[i];
319
320 if (blkg->pd[i] && !blkg->pd[i]->offline &&
321 pol->pd_offline_fn) {
322 pol->pd_offline_fn(blkg->pd[i]);
323 blkg->pd[i]->offline = true;
324 }
325 }
326}
327
328static void blkg_destroy(struct blkcg_gq *blkg)
329{
330 struct blkcg *blkcg = blkg->blkcg;
331 struct blkcg_gq *parent = blkg->parent;
332
333 lockdep_assert_held(blkg->q->queue_lock);
334 lockdep_assert_held(&blkcg->lock);
335
336
337 WARN_ON_ONCE(list_empty(&blkg->q_node));
338 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
339
340 if (parent) {
341 blkg_rwstat_add_aux(&parent->stat_bytes, &blkg->stat_bytes);
342 blkg_rwstat_add_aux(&parent->stat_ios, &blkg->stat_ios);
343 }
344
345 blkg->online = false;
346
347 radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
348 list_del_init(&blkg->q_node);
349 hlist_del_init_rcu(&blkg->blkcg_node);
350
351
352
353
354
355
356 if (rcu_access_pointer(blkcg->blkg_hint) == blkg)
357 rcu_assign_pointer(blkcg->blkg_hint, NULL);
358
359
360
361
362
363 blkg_put(blkg);
364}
365
366
367
368
369
370
371
372static void blkg_destroy_all(struct request_queue *q)
373{
374 struct blkcg_gq *blkg, *n;
375
376 lockdep_assert_held(q->queue_lock);
377
378 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
379 struct blkcg *blkcg = blkg->blkcg;
380
381 spin_lock(&blkcg->lock);
382 blkg_pd_offline(blkg);
383 blkg_destroy(blkg);
384 spin_unlock(&blkcg->lock);
385 }
386
387 q->root_blkg = NULL;
388 q->root_rl.blkg = NULL;
389}
390
391
392
393
394
395
396
397
398
399void __blkg_release_rcu(struct rcu_head *rcu_head)
400{
401 struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head);
402
403
404 css_put(&blkg->blkcg->css);
405 if (blkg->parent)
406 blkg_put(blkg->parent);
407
408 wb_congested_put(blkg->wb_congested);
409
410 blkg_free(blkg);
411}
412EXPORT_SYMBOL_GPL(__blkg_release_rcu);
413
414
415
416
417
418struct request_list *__blk_queue_next_rl(struct request_list *rl,
419 struct request_queue *q)
420{
421 struct list_head *ent;
422 struct blkcg_gq *blkg;
423
424
425
426
427
428 if (rl == &q->root_rl) {
429 ent = &q->blkg_list;
430
431 if (list_empty(ent))
432 return NULL;
433 } else {
434 blkg = container_of(rl, struct blkcg_gq, rl);
435 ent = &blkg->q_node;
436 }
437
438
439 ent = ent->next;
440 if (ent == &q->root_blkg->q_node)
441 ent = ent->next;
442 if (ent == &q->blkg_list)
443 return NULL;
444
445 blkg = container_of(ent, struct blkcg_gq, q_node);
446 return &blkg->rl;
447}
448
449static int blkcg_reset_stats(struct cgroup_subsys_state *css,
450 struct cftype *cftype, u64 val)
451{
452 struct blkcg *blkcg = css_to_blkcg(css);
453 struct blkcg_gq *blkg;
454 int i;
455
456 mutex_lock(&blkcg_pol_mutex);
457 spin_lock_irq(&blkcg->lock);
458
459
460
461
462
463
464 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
465 blkg_rwstat_reset(&blkg->stat_bytes);
466 blkg_rwstat_reset(&blkg->stat_ios);
467
468 for (i = 0; i < BLKCG_MAX_POLS; i++) {
469 struct blkcg_policy *pol = blkcg_policy[i];
470
471 if (blkg->pd[i] && pol->pd_reset_stats_fn)
472 pol->pd_reset_stats_fn(blkg->pd[i]);
473 }
474 }
475
476 spin_unlock_irq(&blkcg->lock);
477 mutex_unlock(&blkcg_pol_mutex);
478 return 0;
479}
480
481const char *blkg_dev_name(struct blkcg_gq *blkg)
482{
483
484 if (blkg->q->backing_dev_info->dev)
485 return dev_name(blkg->q->backing_dev_info->dev);
486 return NULL;
487}
488EXPORT_SYMBOL_GPL(blkg_dev_name);
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
509 u64 (*prfill)(struct seq_file *,
510 struct blkg_policy_data *, int),
511 const struct blkcg_policy *pol, int data,
512 bool show_total)
513{
514 struct blkcg_gq *blkg;
515 u64 total = 0;
516
517 rcu_read_lock();
518 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
519 spin_lock_irq(blkg->q->queue_lock);
520 if (blkcg_policy_enabled(blkg->q, pol))
521 total += prfill(sf, blkg->pd[pol->plid], data);
522 spin_unlock_irq(blkg->q->queue_lock);
523 }
524 rcu_read_unlock();
525
526 if (show_total)
527 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
528}
529EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
530
531
532
533
534
535
536
537
538
539u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
540{
541 const char *dname = blkg_dev_name(pd->blkg);
542
543 if (!dname)
544 return 0;
545
546 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
547 return v;
548}
549EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
550
551
552
553
554
555
556
557
558
559u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
560 const struct blkg_rwstat *rwstat)
561{
562 static const char *rwstr[] = {
563 [BLKG_RWSTAT_READ] = "Read",
564 [BLKG_RWSTAT_WRITE] = "Write",
565 [BLKG_RWSTAT_SYNC] = "Sync",
566 [BLKG_RWSTAT_ASYNC] = "Async",
567 };
568 const char *dname = blkg_dev_name(pd->blkg);
569 u64 v;
570 int i;
571
572 if (!dname)
573 return 0;
574
575 for (i = 0; i < BLKG_RWSTAT_NR; i++)
576 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
577 (unsigned long long)atomic64_read(&rwstat->aux_cnt[i]));
578
579 v = atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_READ]) +
580 atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_WRITE]);
581 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
582 return v;
583}
584EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat);
585
586
587
588
589
590
591
592
593
594u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off)
595{
596 return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off));
597}
598EXPORT_SYMBOL_GPL(blkg_prfill_stat);
599
600
601
602
603
604
605
606
607
608u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
609 int off)
610{
611 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off);
612
613 return __blkg_prfill_rwstat(sf, pd, &rwstat);
614}
615EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
616
617static u64 blkg_prfill_rwstat_field(struct seq_file *sf,
618 struct blkg_policy_data *pd, int off)
619{
620 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd->blkg + off);
621
622 return __blkg_prfill_rwstat(sf, pd, &rwstat);
623}
624
625
626
627
628
629
630
631
632
633int blkg_print_stat_bytes(struct seq_file *sf, void *v)
634{
635 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
636 blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private,
637 offsetof(struct blkcg_gq, stat_bytes), true);
638 return 0;
639}
640EXPORT_SYMBOL_GPL(blkg_print_stat_bytes);
641
642
643
644
645
646
647
648
649
650int blkg_print_stat_ios(struct seq_file *sf, void *v)
651{
652 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
653 blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private,
654 offsetof(struct blkcg_gq, stat_ios), true);
655 return 0;
656}
657EXPORT_SYMBOL_GPL(blkg_print_stat_ios);
658
659static u64 blkg_prfill_rwstat_field_recursive(struct seq_file *sf,
660 struct blkg_policy_data *pd,
661 int off)
662{
663 struct blkg_rwstat rwstat = blkg_rwstat_recursive_sum(pd->blkg,
664 NULL, off);
665 return __blkg_prfill_rwstat(sf, pd, &rwstat);
666}
667
668
669
670
671
672
673int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v)
674{
675 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
676 blkg_prfill_rwstat_field_recursive,
677 (void *)seq_cft(sf)->private,
678 offsetof(struct blkcg_gq, stat_bytes), true);
679 return 0;
680}
681EXPORT_SYMBOL_GPL(blkg_print_stat_bytes_recursive);
682
683
684
685
686
687
688int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v)
689{
690 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
691 blkg_prfill_rwstat_field_recursive,
692 (void *)seq_cft(sf)->private,
693 offsetof(struct blkcg_gq, stat_ios), true);
694 return 0;
695}
696EXPORT_SYMBOL_GPL(blkg_print_stat_ios_recursive);
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg,
712 struct blkcg_policy *pol, int off)
713{
714 struct blkcg_gq *pos_blkg;
715 struct cgroup_subsys_state *pos_css;
716 u64 sum = 0;
717
718 lockdep_assert_held(blkg->q->queue_lock);
719
720 rcu_read_lock();
721 blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) {
722 struct blkg_stat *stat;
723
724 if (!pos_blkg->online)
725 continue;
726
727 if (pol)
728 stat = (void *)blkg_to_pd(pos_blkg, pol) + off;
729 else
730 stat = (void *)blkg + off;
731
732 sum += blkg_stat_read(stat) + atomic64_read(&stat->aux_cnt);
733 }
734 rcu_read_unlock();
735
736 return sum;
737}
738EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum);
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg,
754 struct blkcg_policy *pol, int off)
755{
756 struct blkcg_gq *pos_blkg;
757 struct cgroup_subsys_state *pos_css;
758 struct blkg_rwstat sum = { };
759 int i;
760
761 lockdep_assert_held(blkg->q->queue_lock);
762
763 rcu_read_lock();
764 blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) {
765 struct blkg_rwstat *rwstat;
766
767 if (!pos_blkg->online)
768 continue;
769
770 if (pol)
771 rwstat = (void *)blkg_to_pd(pos_blkg, pol) + off;
772 else
773 rwstat = (void *)pos_blkg + off;
774
775 for (i = 0; i < BLKG_RWSTAT_NR; i++)
776 atomic64_add(atomic64_read(&rwstat->aux_cnt[i]) +
777 percpu_counter_sum_positive(&rwstat->cpu_cnt[i]),
778 &sum.aux_cnt[i]);
779 }
780 rcu_read_unlock();
781
782 return sum;
783}
784EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum);
785
786
787static struct blkcg_gq *blkg_lookup_check(struct blkcg *blkcg,
788 const struct blkcg_policy *pol,
789 struct request_queue *q)
790{
791 WARN_ON_ONCE(!rcu_read_lock_held());
792 lockdep_assert_held(q->queue_lock);
793
794 if (!blkcg_policy_enabled(q, pol))
795 return ERR_PTR(-EOPNOTSUPP);
796
797
798
799
800
801 if (unlikely(blk_queue_bypass(q)))
802 return ERR_PTR(blk_queue_dying(q) ? -ENODEV : -EBUSY);
803
804 return __blkg_lookup(blkcg, q, true );
805}
806
807
808
809
810
811
812
813
814
815
816
817
818
819int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
820 char *input, struct blkg_conf_ctx *ctx)
821 __acquires(rcu) __acquires(disk->queue->queue_lock)
822{
823 struct gendisk *disk;
824 struct request_queue *q;
825 struct blkcg_gq *blkg;
826 unsigned int major, minor;
827 int key_len, part, ret;
828 char *body;
829
830 if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2)
831 return -EINVAL;
832
833 body = input + key_len;
834 if (!isspace(*body))
835 return -EINVAL;
836 body = skip_spaces(body);
837
838 disk = get_gendisk(MKDEV(major, minor), &part);
839 if (!disk)
840 return -ENODEV;
841 if (part) {
842 ret = -ENODEV;
843 goto fail;
844 }
845
846 q = disk->queue;
847
848 rcu_read_lock();
849 spin_lock_irq(q->queue_lock);
850
851 blkg = blkg_lookup_check(blkcg, pol, q);
852 if (IS_ERR(blkg)) {
853 ret = PTR_ERR(blkg);
854 goto fail_unlock;
855 }
856
857 if (blkg)
858 goto success;
859
860
861
862
863
864 while (true) {
865 struct blkcg *pos = blkcg;
866 struct blkcg *parent;
867 struct blkcg_gq *new_blkg;
868
869 parent = blkcg_parent(blkcg);
870 while (parent && !__blkg_lookup(parent, q, false)) {
871 pos = parent;
872 parent = blkcg_parent(parent);
873 }
874
875
876 spin_unlock_irq(q->queue_lock);
877 rcu_read_unlock();
878
879 new_blkg = blkg_alloc(pos, q, GFP_KERNEL);
880 if (unlikely(!new_blkg)) {
881 ret = -ENOMEM;
882 goto fail;
883 }
884
885 rcu_read_lock();
886 spin_lock_irq(q->queue_lock);
887
888 blkg = blkg_lookup_check(pos, pol, q);
889 if (IS_ERR(blkg)) {
890 ret = PTR_ERR(blkg);
891 goto fail_unlock;
892 }
893
894 if (blkg) {
895 blkg_free(new_blkg);
896 } else {
897 blkg = blkg_create(pos, q, new_blkg);
898 if (unlikely(IS_ERR(blkg))) {
899 ret = PTR_ERR(blkg);
900 goto fail_unlock;
901 }
902 }
903
904 if (pos == blkcg)
905 goto success;
906 }
907success:
908 ctx->disk = disk;
909 ctx->blkg = blkg;
910 ctx->body = body;
911 return 0;
912
913fail_unlock:
914 spin_unlock_irq(q->queue_lock);
915 rcu_read_unlock();
916fail:
917 put_disk_and_module(disk);
918
919
920
921
922
923
924 if (ret == -EBUSY) {
925 msleep(10);
926 ret = restart_syscall();
927 }
928 return ret;
929}
930EXPORT_SYMBOL_GPL(blkg_conf_prep);
931
932
933
934
935
936
937
938
939void blkg_conf_finish(struct blkg_conf_ctx *ctx)
940 __releases(ctx->disk->queue->queue_lock) __releases(rcu)
941{
942 spin_unlock_irq(ctx->disk->queue->queue_lock);
943 rcu_read_unlock();
944 put_disk_and_module(ctx->disk);
945}
946EXPORT_SYMBOL_GPL(blkg_conf_finish);
947
948static int blkcg_print_stat(struct seq_file *sf, void *v)
949{
950 struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
951 struct blkcg_gq *blkg;
952
953 rcu_read_lock();
954
955 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
956 const char *dname;
957 struct blkg_rwstat rwstat;
958 u64 rbytes, wbytes, rios, wios;
959
960 dname = blkg_dev_name(blkg);
961 if (!dname)
962 continue;
963
964 spin_lock_irq(blkg->q->queue_lock);
965
966 rwstat = blkg_rwstat_recursive_sum(blkg, NULL,
967 offsetof(struct blkcg_gq, stat_bytes));
968 rbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]);
969 wbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]);
970
971 rwstat = blkg_rwstat_recursive_sum(blkg, NULL,
972 offsetof(struct blkcg_gq, stat_ios));
973 rios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]);
974 wios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]);
975
976 spin_unlock_irq(blkg->q->queue_lock);
977
978 if (rbytes || wbytes || rios || wios)
979 seq_printf(sf, "%s rbytes=%llu wbytes=%llu rios=%llu wios=%llu\n",
980 dname, rbytes, wbytes, rios, wios);
981 }
982
983 rcu_read_unlock();
984 return 0;
985}
986
987static struct cftype blkcg_files[] = {
988 {
989 .name = "stat",
990 .flags = CFTYPE_NOT_ON_ROOT,
991 .seq_show = blkcg_print_stat,
992 },
993 { }
994};
995
996static struct cftype blkcg_legacy_files[] = {
997 {
998 .name = "reset_stats",
999 .write_u64 = blkcg_reset_stats,
1000 },
1001 { }
1002};
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016static void blkcg_css_offline(struct cgroup_subsys_state *css)
1017{
1018 struct blkcg *blkcg = css_to_blkcg(css);
1019 struct blkcg_gq *blkg;
1020
1021 spin_lock_irq(&blkcg->lock);
1022
1023 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
1024 struct request_queue *q = blkg->q;
1025
1026 if (spin_trylock(q->queue_lock)) {
1027 blkg_pd_offline(blkg);
1028 spin_unlock(q->queue_lock);
1029 } else {
1030 spin_unlock_irq(&blkcg->lock);
1031 cpu_relax();
1032 spin_lock_irq(&blkcg->lock);
1033 }
1034 }
1035
1036 spin_unlock_irq(&blkcg->lock);
1037
1038 wb_blkcg_offline(blkcg);
1039}
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050static void blkcg_destroy_all_blkgs(struct blkcg *blkcg)
1051{
1052 spin_lock_irq(&blkcg->lock);
1053 while (!hlist_empty(&blkcg->blkg_list)) {
1054 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
1055 struct blkcg_gq,
1056 blkcg_node);
1057 struct request_queue *q = blkg->q;
1058
1059 if (spin_trylock(q->queue_lock)) {
1060 blkg_destroy(blkg);
1061 spin_unlock(q->queue_lock);
1062 } else {
1063 spin_unlock_irq(&blkcg->lock);
1064 cpu_relax();
1065 spin_lock_irq(&blkcg->lock);
1066 }
1067 }
1068 spin_unlock_irq(&blkcg->lock);
1069}
1070
1071static void blkcg_css_free(struct cgroup_subsys_state *css)
1072{
1073 struct blkcg *blkcg = css_to_blkcg(css);
1074 int i;
1075
1076 blkcg_destroy_all_blkgs(blkcg);
1077
1078 mutex_lock(&blkcg_pol_mutex);
1079
1080 list_del(&blkcg->all_blkcgs_node);
1081
1082 for (i = 0; i < BLKCG_MAX_POLS; i++)
1083 if (blkcg->cpd[i])
1084 blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]);
1085
1086 mutex_unlock(&blkcg_pol_mutex);
1087
1088 kfree(blkcg);
1089}
1090
1091static struct cgroup_subsys_state *
1092blkcg_css_alloc(struct cgroup_subsys_state *parent_css)
1093{
1094 struct blkcg *blkcg;
1095 struct cgroup_subsys_state *ret;
1096 int i;
1097
1098 mutex_lock(&blkcg_pol_mutex);
1099
1100 if (!parent_css) {
1101 blkcg = &blkcg_root;
1102 } else {
1103 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
1104 if (!blkcg) {
1105 ret = ERR_PTR(-ENOMEM);
1106 goto unlock;
1107 }
1108 }
1109
1110 for (i = 0; i < BLKCG_MAX_POLS ; i++) {
1111 struct blkcg_policy *pol = blkcg_policy[i];
1112 struct blkcg_policy_data *cpd;
1113
1114
1115
1116
1117
1118
1119
1120 if (!pol || !pol->cpd_alloc_fn)
1121 continue;
1122
1123 cpd = pol->cpd_alloc_fn(GFP_KERNEL);
1124 if (!cpd) {
1125 ret = ERR_PTR(-ENOMEM);
1126 goto free_pd_blkcg;
1127 }
1128 blkcg->cpd[i] = cpd;
1129 cpd->blkcg = blkcg;
1130 cpd->plid = i;
1131 if (pol->cpd_init_fn)
1132 pol->cpd_init_fn(cpd);
1133 }
1134
1135 spin_lock_init(&blkcg->lock);
1136 INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN);
1137 INIT_HLIST_HEAD(&blkcg->blkg_list);
1138#ifdef CONFIG_CGROUP_WRITEBACK
1139 INIT_LIST_HEAD(&blkcg->cgwb_list);
1140#endif
1141 list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs);
1142
1143 mutex_unlock(&blkcg_pol_mutex);
1144 return &blkcg->css;
1145
1146free_pd_blkcg:
1147 for (i--; i >= 0; i--)
1148 if (blkcg->cpd[i])
1149 blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]);
1150
1151 if (blkcg != &blkcg_root)
1152 kfree(blkcg);
1153unlock:
1154 mutex_unlock(&blkcg_pol_mutex);
1155 return ret;
1156}
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168int blkcg_init_queue(struct request_queue *q)
1169{
1170 struct blkcg_gq *new_blkg, *blkg;
1171 bool preloaded;
1172 int ret;
1173
1174 new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
1175 if (!new_blkg)
1176 return -ENOMEM;
1177
1178 preloaded = !radix_tree_preload(GFP_KERNEL);
1179
1180
1181 rcu_read_lock();
1182 spin_lock_irq(q->queue_lock);
1183 blkg = blkg_create(&blkcg_root, q, new_blkg);
1184 if (IS_ERR(blkg))
1185 goto err_unlock;
1186 q->root_blkg = blkg;
1187 q->root_rl.blkg = blkg;
1188 spin_unlock_irq(q->queue_lock);
1189 rcu_read_unlock();
1190
1191 if (preloaded)
1192 radix_tree_preload_end();
1193
1194 ret = blk_throtl_init(q);
1195 if (ret) {
1196 spin_lock_irq(q->queue_lock);
1197 blkg_destroy_all(q);
1198 spin_unlock_irq(q->queue_lock);
1199 }
1200 return ret;
1201
1202err_unlock:
1203 spin_unlock_irq(q->queue_lock);
1204 rcu_read_unlock();
1205 if (preloaded)
1206 radix_tree_preload_end();
1207 return PTR_ERR(blkg);
1208}
1209
1210
1211
1212
1213
1214
1215
1216void blkcg_drain_queue(struct request_queue *q)
1217{
1218 lockdep_assert_held(q->queue_lock);
1219
1220
1221
1222
1223
1224 if (!q->root_blkg)
1225 return;
1226
1227 blk_throtl_drain(q);
1228}
1229
1230
1231
1232
1233
1234
1235
1236void blkcg_exit_queue(struct request_queue *q)
1237{
1238 spin_lock_irq(q->queue_lock);
1239 blkg_destroy_all(q);
1240 spin_unlock_irq(q->queue_lock);
1241
1242 blk_throtl_exit(q);
1243}
1244
1245
1246
1247
1248
1249
1250
1251static int blkcg_can_attach(struct cgroup_taskset *tset)
1252{
1253 struct task_struct *task;
1254 struct cgroup_subsys_state *dst_css;
1255 struct io_context *ioc;
1256 int ret = 0;
1257
1258
1259 cgroup_taskset_for_each(task, dst_css, tset) {
1260 task_lock(task);
1261 ioc = task->io_context;
1262 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
1263 ret = -EINVAL;
1264 task_unlock(task);
1265 if (ret)
1266 break;
1267 }
1268 return ret;
1269}
1270
1271static void blkcg_bind(struct cgroup_subsys_state *root_css)
1272{
1273 int i;
1274
1275 mutex_lock(&blkcg_pol_mutex);
1276
1277 for (i = 0; i < BLKCG_MAX_POLS; i++) {
1278 struct blkcg_policy *pol = blkcg_policy[i];
1279 struct blkcg *blkcg;
1280
1281 if (!pol || !pol->cpd_bind_fn)
1282 continue;
1283
1284 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node)
1285 if (blkcg->cpd[pol->plid])
1286 pol->cpd_bind_fn(blkcg->cpd[pol->plid]);
1287 }
1288 mutex_unlock(&blkcg_pol_mutex);
1289}
1290
1291struct cgroup_subsys io_cgrp_subsys = {
1292 .css_alloc = blkcg_css_alloc,
1293 .css_offline = blkcg_css_offline,
1294 .css_free = blkcg_css_free,
1295 .can_attach = blkcg_can_attach,
1296 .bind = blkcg_bind,
1297 .dfl_cftypes = blkcg_files,
1298 .legacy_cftypes = blkcg_legacy_files,
1299 .legacy_name = "blkio",
1300#ifdef CONFIG_MEMCG
1301
1302
1303
1304
1305
1306 .depends_on = 1 << memory_cgrp_id,
1307#endif
1308};
1309EXPORT_SYMBOL_GPL(io_cgrp_subsys);
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327int blkcg_activate_policy(struct request_queue *q,
1328 const struct blkcg_policy *pol)
1329{
1330 struct blkg_policy_data *pd_prealloc = NULL;
1331 struct blkcg_gq *blkg;
1332 int ret;
1333
1334 if (blkcg_policy_enabled(q, pol))
1335 return 0;
1336
1337 if (q->mq_ops)
1338 blk_mq_freeze_queue(q);
1339 else
1340 blk_queue_bypass_start(q);
1341pd_prealloc:
1342 if (!pd_prealloc) {
1343 pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q->node);
1344 if (!pd_prealloc) {
1345 ret = -ENOMEM;
1346 goto out_bypass_end;
1347 }
1348 }
1349
1350 spin_lock_irq(q->queue_lock);
1351
1352 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1353 struct blkg_policy_data *pd;
1354
1355 if (blkg->pd[pol->plid])
1356 continue;
1357
1358 pd = pol->pd_alloc_fn(GFP_NOWAIT | __GFP_NOWARN, q->node);
1359 if (!pd)
1360 swap(pd, pd_prealloc);
1361 if (!pd) {
1362 spin_unlock_irq(q->queue_lock);
1363 goto pd_prealloc;
1364 }
1365
1366 blkg->pd[pol->plid] = pd;
1367 pd->blkg = blkg;
1368 pd->plid = pol->plid;
1369 if (pol->pd_init_fn)
1370 pol->pd_init_fn(pd);
1371 }
1372
1373 __set_bit(pol->plid, q->blkcg_pols);
1374 ret = 0;
1375
1376 spin_unlock_irq(q->queue_lock);
1377out_bypass_end:
1378 if (q->mq_ops)
1379 blk_mq_unfreeze_queue(q);
1380 else
1381 blk_queue_bypass_end(q);
1382 if (pd_prealloc)
1383 pol->pd_free_fn(pd_prealloc);
1384 return ret;
1385}
1386EXPORT_SYMBOL_GPL(blkcg_activate_policy);
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396void blkcg_deactivate_policy(struct request_queue *q,
1397 const struct blkcg_policy *pol)
1398{
1399 struct blkcg_gq *blkg;
1400
1401 if (!blkcg_policy_enabled(q, pol))
1402 return;
1403
1404 if (q->mq_ops)
1405 blk_mq_freeze_queue(q);
1406 else
1407 blk_queue_bypass_start(q);
1408
1409 spin_lock_irq(q->queue_lock);
1410
1411 __clear_bit(pol->plid, q->blkcg_pols);
1412
1413 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1414 if (blkg->pd[pol->plid]) {
1415 if (!blkg->pd[pol->plid]->offline &&
1416 pol->pd_offline_fn) {
1417 pol->pd_offline_fn(blkg->pd[pol->plid]);
1418 blkg->pd[pol->plid]->offline = true;
1419 }
1420 pol->pd_free_fn(blkg->pd[pol->plid]);
1421 blkg->pd[pol->plid] = NULL;
1422 }
1423 }
1424
1425 spin_unlock_irq(q->queue_lock);
1426
1427 if (q->mq_ops)
1428 blk_mq_unfreeze_queue(q);
1429 else
1430 blk_queue_bypass_end(q);
1431}
1432EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
1433
1434
1435
1436
1437
1438
1439
1440
1441int blkcg_policy_register(struct blkcg_policy *pol)
1442{
1443 struct blkcg *blkcg;
1444 int i, ret;
1445
1446 mutex_lock(&blkcg_pol_register_mutex);
1447 mutex_lock(&blkcg_pol_mutex);
1448
1449
1450 ret = -ENOSPC;
1451 for (i = 0; i < BLKCG_MAX_POLS; i++)
1452 if (!blkcg_policy[i])
1453 break;
1454 if (i >= BLKCG_MAX_POLS)
1455 goto err_unlock;
1456
1457
1458 if ((!pol->cpd_alloc_fn ^ !pol->cpd_free_fn) ||
1459 (!pol->pd_alloc_fn ^ !pol->pd_free_fn))
1460 goto err_unlock;
1461
1462
1463 pol->plid = i;
1464 blkcg_policy[pol->plid] = pol;
1465
1466
1467 if (pol->cpd_alloc_fn) {
1468 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1469 struct blkcg_policy_data *cpd;
1470
1471 cpd = pol->cpd_alloc_fn(GFP_KERNEL);
1472 if (!cpd)
1473 goto err_free_cpds;
1474
1475 blkcg->cpd[pol->plid] = cpd;
1476 cpd->blkcg = blkcg;
1477 cpd->plid = pol->plid;
1478 pol->cpd_init_fn(cpd);
1479 }
1480 }
1481
1482 mutex_unlock(&blkcg_pol_mutex);
1483
1484
1485 if (pol->dfl_cftypes)
1486 WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys,
1487 pol->dfl_cftypes));
1488 if (pol->legacy_cftypes)
1489 WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys,
1490 pol->legacy_cftypes));
1491 mutex_unlock(&blkcg_pol_register_mutex);
1492 return 0;
1493
1494err_free_cpds:
1495 if (pol->cpd_free_fn) {
1496 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1497 if (blkcg->cpd[pol->plid]) {
1498 pol->cpd_free_fn(blkcg->cpd[pol->plid]);
1499 blkcg->cpd[pol->plid] = NULL;
1500 }
1501 }
1502 }
1503 blkcg_policy[pol->plid] = NULL;
1504err_unlock:
1505 mutex_unlock(&blkcg_pol_mutex);
1506 mutex_unlock(&blkcg_pol_register_mutex);
1507 return ret;
1508}
1509EXPORT_SYMBOL_GPL(blkcg_policy_register);
1510
1511
1512
1513
1514
1515
1516
1517void blkcg_policy_unregister(struct blkcg_policy *pol)
1518{
1519 struct blkcg *blkcg;
1520
1521 mutex_lock(&blkcg_pol_register_mutex);
1522
1523 if (WARN_ON(blkcg_policy[pol->plid] != pol))
1524 goto out_unlock;
1525
1526
1527 if (pol->dfl_cftypes)
1528 cgroup_rm_cftypes(pol->dfl_cftypes);
1529 if (pol->legacy_cftypes)
1530 cgroup_rm_cftypes(pol->legacy_cftypes);
1531
1532
1533 mutex_lock(&blkcg_pol_mutex);
1534
1535 if (pol->cpd_free_fn) {
1536 list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1537 if (blkcg->cpd[pol->plid]) {
1538 pol->cpd_free_fn(blkcg->cpd[pol->plid]);
1539 blkcg->cpd[pol->plid] = NULL;
1540 }
1541 }
1542 }
1543 blkcg_policy[pol->plid] = NULL;
1544
1545 mutex_unlock(&blkcg_pol_mutex);
1546out_unlock:
1547 mutex_unlock(&blkcg_pol_register_mutex);
1548}
1549EXPORT_SYMBOL_GPL(blkcg_policy_unregister);
1550