2 * bcache sysfs interfaces
4 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
5 * Copyright 2012 Google, Inc.
12 #include "writeback.h"
14 #include <linux/blkdev.h>
15 #include <linux/sort.h>
17 static const char * const cache_replacement_policies[] = {
24 static const char * const error_actions[] = {
30 write_attribute(attach);
31 write_attribute(detach);
32 write_attribute(unregister);
33 write_attribute(stop);
34 write_attribute(clear_stats);
35 write_attribute(trigger_gc);
36 write_attribute(prune_cache);
37 write_attribute(flash_vol_create);
39 read_attribute(bucket_size);
40 read_attribute(block_size);
41 read_attribute(nbuckets);
42 read_attribute(tree_depth);
43 read_attribute(root_usage_percent);
44 read_attribute(priority_stats);
45 read_attribute(btree_cache_size);
46 read_attribute(btree_cache_max_chain);
47 read_attribute(cache_available_percent);
48 read_attribute(written);
49 read_attribute(btree_written);
50 read_attribute(metadata_written);
51 read_attribute(active_journal_entries);
53 sysfs_time_stats_attribute(btree_gc, sec, ms);
54 sysfs_time_stats_attribute(btree_split, sec, us);
55 sysfs_time_stats_attribute(btree_sort, ms, us);
56 sysfs_time_stats_attribute(btree_read, ms, us);
57 sysfs_time_stats_attribute(try_harder, ms, us);
59 read_attribute(btree_nodes);
60 read_attribute(btree_used_percent);
61 read_attribute(average_key_size);
62 read_attribute(dirty_data);
63 read_attribute(bset_tree_stats);
65 read_attribute(state);
66 read_attribute(cache_read_races);
67 read_attribute(writeback_keys_done);
68 read_attribute(writeback_keys_failed);
69 read_attribute(io_errors);
70 read_attribute(congested);
71 rw_attribute(congested_read_threshold_us);
72 rw_attribute(congested_write_threshold_us);
74 rw_attribute(sequential_cutoff);
75 rw_attribute(data_csum);
76 rw_attribute(cache_mode);
77 rw_attribute(writeback_metadata);
78 rw_attribute(writeback_running);
79 rw_attribute(writeback_percent);
80 rw_attribute(writeback_delay);
81 rw_attribute(writeback_rate);
83 rw_attribute(writeback_rate_update_seconds);
84 rw_attribute(writeback_rate_d_term);
85 rw_attribute(writeback_rate_p_term_inverse);
86 read_attribute(writeback_rate_debug);
88 read_attribute(stripe_size);
89 read_attribute(partial_stripes_expensive);
91 rw_attribute(synchronous);
92 rw_attribute(journal_delay_ms);
93 rw_attribute(discard);
94 rw_attribute(running);
96 rw_attribute(readahead);
98 rw_attribute(io_error_limit);
99 rw_attribute(io_error_halflife);
100 rw_attribute(verify);
101 rw_attribute(bypass_torture_test);
102 rw_attribute(key_merging_disabled);
103 rw_attribute(gc_always_rewrite);
104 rw_attribute(expensive_debug_checks);
105 rw_attribute(cache_replacement_policy);
106 rw_attribute(btree_shrinker_disabled);
107 rw_attribute(copy_gc_enabled);
110 SHOW(__bch_cached_dev)
112 struct cached_dev *dc = container_of(kobj, struct cached_dev,
114 const char *states[] = { "no cache", "clean", "dirty", "inconsistent" };
116 #define var(stat) (dc->stat)
118 if (attr == &sysfs_cache_mode)
119 return bch_snprint_string_list(buf, PAGE_SIZE,
121 BDEV_CACHE_MODE(&dc->sb));
123 sysfs_printf(data_csum, "%i", dc->disk.data_csum);
124 var_printf(verify, "%i");
125 var_printf(bypass_torture_test, "%i");
126 var_printf(writeback_metadata, "%i");
127 var_printf(writeback_running, "%i");
128 var_print(writeback_delay);
129 var_print(writeback_percent);
130 sysfs_hprint(writeback_rate, dc->writeback_rate.rate << 9);
132 var_print(writeback_rate_update_seconds);
133 var_print(writeback_rate_d_term);
134 var_print(writeback_rate_p_term_inverse);
136 if (attr == &sysfs_writeback_rate_debug) {
140 char proportional[20];
145 bch_hprint(rate, dc->writeback_rate.rate << 9);
146 bch_hprint(dirty, bcache_dev_sectors_dirty(&dc->disk) << 9);
147 bch_hprint(target, dc->writeback_rate_target << 9);
148 bch_hprint(proportional,dc->writeback_rate_proportional << 9);
149 bch_hprint(derivative, dc->writeback_rate_derivative << 9);
150 bch_hprint(change, dc->writeback_rate_change << 9);
152 next_io = div64_s64(dc->writeback_rate.next - local_clock(),
159 "proportional:\t%s\n"
161 "change:\t\t%s/sec\n"
162 "next io:\t%llims\n",
163 rate, dirty, target, proportional,
164 derivative, change, next_io);
167 sysfs_hprint(dirty_data,
168 bcache_dev_sectors_dirty(&dc->disk) << 9);
170 sysfs_hprint(stripe_size, dc->disk.stripe_size << 9);
171 var_printf(partial_stripes_expensive, "%u");
173 var_hprint(sequential_cutoff);
174 var_hprint(readahead);
176 sysfs_print(running, atomic_read(&dc->running));
177 sysfs_print(state, states[BDEV_STATE(&dc->sb)]);
179 if (attr == &sysfs_label) {
180 memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
181 buf[SB_LABEL_SIZE + 1] = '\0';
189 SHOW_LOCKED(bch_cached_dev)
193 struct cached_dev *dc = container_of(kobj, struct cached_dev,
197 struct kobj_uevent_env *env;
199 #define d_strtoul(var) sysfs_strtoul(var, dc->var)
200 #define d_strtoul_nonzero(var) sysfs_strtoul_clamp(var, dc->var, 1, INT_MAX)
201 #define d_strtoi_h(var) sysfs_hatoi(var, dc->var)
203 sysfs_strtoul(data_csum, dc->disk.data_csum);
205 d_strtoul(bypass_torture_test);
206 d_strtoul(writeback_metadata);
207 d_strtoul(writeback_running);
208 d_strtoul(writeback_delay);
210 sysfs_strtoul_clamp(writeback_percent, dc->writeback_percent, 0, 40);
212 sysfs_strtoul_clamp(writeback_rate,
213 dc->writeback_rate.rate, 1, INT_MAX);
215 d_strtoul_nonzero(writeback_rate_update_seconds);
216 d_strtoul(writeback_rate_d_term);
217 d_strtoul_nonzero(writeback_rate_p_term_inverse);
219 d_strtoi_h(sequential_cutoff);
220 d_strtoi_h(readahead);
222 if (attr == &sysfs_clear_stats)
223 bch_cache_accounting_clear(&dc->accounting);
225 if (attr == &sysfs_running &&
226 strtoul_or_return(buf))
227 bch_cached_dev_run(dc);
229 if (attr == &sysfs_cache_mode) {
230 ssize_t v = bch_read_string_list(buf, bch_cache_modes + 1);
235 if ((unsigned) v != BDEV_CACHE_MODE(&dc->sb)) {
236 SET_BDEV_CACHE_MODE(&dc->sb, v);
237 bch_write_bdev_super(dc, NULL);
241 if (attr == &sysfs_label) {
242 if (size > SB_LABEL_SIZE)
244 memcpy(dc->sb.label, buf, size);
245 if (size < SB_LABEL_SIZE)
246 dc->sb.label[size] = '\0';
247 if (size && dc->sb.label[size - 1] == '\n')
248 dc->sb.label[size - 1] = '\0';
249 bch_write_bdev_super(dc, NULL);
251 memcpy(dc->disk.c->uuids[dc->disk.id].label,
253 bch_uuid_write(dc->disk.c);
255 env = kzalloc(sizeof(struct kobj_uevent_env), GFP_KERNEL);
258 add_uevent_var(env, "DRIVER=bcache");
259 add_uevent_var(env, "CACHED_UUID=%pU", dc->sb.uuid),
260 add_uevent_var(env, "CACHED_LABEL=%s", buf);
262 &disk_to_dev(dc->disk.disk)->kobj, KOBJ_CHANGE, env->envp);
266 if (attr == &sysfs_attach) {
267 if (bch_parse_uuid(buf, dc->sb.set_uuid) < 16)
270 list_for_each_entry(c, &bch_cache_sets, list) {
271 v = bch_cached_dev_attach(dc, c);
276 pr_err("Can't attach %s: cache set not found", buf);
280 if (attr == &sysfs_detach && dc->disk.c)
281 bch_cached_dev_detach(dc);
283 if (attr == &sysfs_stop)
284 bcache_device_stop(&dc->disk);
289 STORE(bch_cached_dev)
291 struct cached_dev *dc = container_of(kobj, struct cached_dev,
294 mutex_lock(&bch_register_lock);
295 size = __cached_dev_store(kobj, attr, buf, size);
297 if (attr == &sysfs_writeback_running)
298 bch_writeback_queue(dc);
300 if (attr == &sysfs_writeback_percent)
301 schedule_delayed_work(&dc->writeback_rate_update,
302 dc->writeback_rate_update_seconds * HZ);
304 mutex_unlock(&bch_register_lock);
308 static struct attribute *bch_cached_dev_files[] = {
316 &sysfs_writeback_metadata,
317 &sysfs_writeback_running,
318 &sysfs_writeback_delay,
319 &sysfs_writeback_percent,
320 &sysfs_writeback_rate,
321 &sysfs_writeback_rate_update_seconds,
322 &sysfs_writeback_rate_d_term,
323 &sysfs_writeback_rate_p_term_inverse,
324 &sysfs_writeback_rate_debug,
327 &sysfs_partial_stripes_expensive,
328 &sysfs_sequential_cutoff,
334 #ifdef CONFIG_BCACHE_DEBUG
336 &sysfs_bypass_torture_test,
340 KTYPE(bch_cached_dev);
344 struct bcache_device *d = container_of(kobj, struct bcache_device,
346 struct uuid_entry *u = &d->c->uuids[d->id];
348 sysfs_printf(data_csum, "%i", d->data_csum);
349 sysfs_hprint(size, u->sectors << 9);
351 if (attr == &sysfs_label) {
352 memcpy(buf, u->label, SB_LABEL_SIZE);
353 buf[SB_LABEL_SIZE + 1] = '\0';
361 STORE(__bch_flash_dev)
363 struct bcache_device *d = container_of(kobj, struct bcache_device,
365 struct uuid_entry *u = &d->c->uuids[d->id];
367 sysfs_strtoul(data_csum, d->data_csum);
369 if (attr == &sysfs_size) {
371 strtoi_h_or_return(buf, v);
374 bch_uuid_write(d->c);
375 set_capacity(d->disk, u->sectors);
378 if (attr == &sysfs_label) {
379 memcpy(u->label, buf, SB_LABEL_SIZE);
380 bch_uuid_write(d->c);
383 if (attr == &sysfs_unregister) {
384 set_bit(BCACHE_DEV_DETACHING, &d->flags);
385 bcache_device_stop(d);
390 STORE_LOCKED(bch_flash_dev)
392 static struct attribute *bch_flash_dev_files[] = {
401 KTYPE(bch_flash_dev);
403 SHOW(__bch_cache_set)
405 unsigned root_usage(struct cache_set *c)
410 struct btree_iter iter;
418 rw_lock(false, b, b->level);
419 } while (b != c->root);
421 for_each_key_filter(b, k, &iter, bch_ptr_bad)
422 bytes += bkey_bytes(k);
426 return (bytes * 100) / btree_bytes(c);
429 size_t cache_size(struct cache_set *c)
434 mutex_lock(&c->bucket_lock);
435 list_for_each_entry(b, &c->btree_cache, list)
436 ret += 1 << (b->keys.page_order + PAGE_SHIFT);
438 mutex_unlock(&c->bucket_lock);
442 unsigned cache_max_chain(struct cache_set *c)
445 struct hlist_head *h;
447 mutex_lock(&c->bucket_lock);
449 for (h = c->bucket_hash;
450 h < c->bucket_hash + (1 << BUCKET_HASH_BITS);
453 struct hlist_node *p;
461 mutex_unlock(&c->bucket_lock);
465 unsigned btree_used(struct cache_set *c)
467 return div64_u64(c->gc_stats.key_bytes * 100,
468 (c->gc_stats.nodes ?: 1) * btree_bytes(c));
471 unsigned average_key_size(struct cache_set *c)
473 return c->gc_stats.nkeys
474 ? div64_u64(c->gc_stats.data, c->gc_stats.nkeys)
478 struct cache_set *c = container_of(kobj, struct cache_set, kobj);
480 sysfs_print(synchronous, CACHE_SYNC(&c->sb));
481 sysfs_print(journal_delay_ms, c->journal_delay_ms);
482 sysfs_hprint(bucket_size, bucket_bytes(c));
483 sysfs_hprint(block_size, block_bytes(c));
484 sysfs_print(tree_depth, c->root->level);
485 sysfs_print(root_usage_percent, root_usage(c));
487 sysfs_hprint(btree_cache_size, cache_size(c));
488 sysfs_print(btree_cache_max_chain, cache_max_chain(c));
489 sysfs_print(cache_available_percent, 100 - c->gc_stats.in_use);
491 sysfs_print_time_stats(&c->btree_gc_time, btree_gc, sec, ms);
492 sysfs_print_time_stats(&c->btree_split_time, btree_split, sec, us);
493 sysfs_print_time_stats(&c->sort.time, btree_sort, ms, us);
494 sysfs_print_time_stats(&c->btree_read_time, btree_read, ms, us);
495 sysfs_print_time_stats(&c->try_harder_time, try_harder, ms, us);
497 sysfs_print(btree_used_percent, btree_used(c));
498 sysfs_print(btree_nodes, c->gc_stats.nodes);
499 sysfs_hprint(average_key_size, average_key_size(c));
501 sysfs_print(cache_read_races,
502 atomic_long_read(&c->cache_read_races));
504 sysfs_print(writeback_keys_done,
505 atomic_long_read(&c->writeback_keys_done));
506 sysfs_print(writeback_keys_failed,
507 atomic_long_read(&c->writeback_keys_failed));
509 if (attr == &sysfs_errors)
510 return bch_snprint_string_list(buf, PAGE_SIZE, error_actions,
513 /* See count_io_errors for why 88 */
514 sysfs_print(io_error_halflife, c->error_decay * 88);
515 sysfs_print(io_error_limit, c->error_limit >> IO_ERROR_SHIFT);
517 sysfs_hprint(congested,
518 ((uint64_t) bch_get_congested(c)) << 9);
519 sysfs_print(congested_read_threshold_us,
520 c->congested_read_threshold_us);
521 sysfs_print(congested_write_threshold_us,
522 c->congested_write_threshold_us);
524 sysfs_print(active_journal_entries, fifo_used(&c->journal.pin));
525 sysfs_printf(verify, "%i", c->verify);
526 sysfs_printf(key_merging_disabled, "%i", c->key_merging_disabled);
527 sysfs_printf(expensive_debug_checks,
528 "%i", c->expensive_debug_checks);
529 sysfs_printf(gc_always_rewrite, "%i", c->gc_always_rewrite);
530 sysfs_printf(btree_shrinker_disabled, "%i", c->shrinker_disabled);
531 sysfs_printf(copy_gc_enabled, "%i", c->copy_gc_enabled);
533 if (attr == &sysfs_bset_tree_stats)
534 return bch_bset_print_stats(c, buf);
538 SHOW_LOCKED(bch_cache_set)
540 STORE(__bch_cache_set)
542 struct cache_set *c = container_of(kobj, struct cache_set, kobj);
544 if (attr == &sysfs_unregister)
545 bch_cache_set_unregister(c);
547 if (attr == &sysfs_stop)
548 bch_cache_set_stop(c);
550 if (attr == &sysfs_synchronous) {
551 bool sync = strtoul_or_return(buf);
553 if (sync != CACHE_SYNC(&c->sb)) {
554 SET_CACHE_SYNC(&c->sb, sync);
555 bcache_write_super(c);
559 if (attr == &sysfs_flash_vol_create) {
562 strtoi_h_or_return(buf, v);
564 r = bch_flash_dev_create(c, v);
569 if (attr == &sysfs_clear_stats) {
570 atomic_long_set(&c->writeback_keys_done, 0);
571 atomic_long_set(&c->writeback_keys_failed, 0);
573 memset(&c->gc_stats, 0, sizeof(struct gc_stat));
574 bch_cache_accounting_clear(&c->accounting);
577 if (attr == &sysfs_trigger_gc)
580 if (attr == &sysfs_prune_cache) {
581 struct shrink_control sc;
582 sc.gfp_mask = GFP_KERNEL;
583 sc.nr_to_scan = strtoul_or_return(buf);
584 c->shrink.scan_objects(&c->shrink, &sc);
587 sysfs_strtoul(congested_read_threshold_us,
588 c->congested_read_threshold_us);
589 sysfs_strtoul(congested_write_threshold_us,
590 c->congested_write_threshold_us);
592 if (attr == &sysfs_errors) {
593 ssize_t v = bch_read_string_list(buf, error_actions);
601 if (attr == &sysfs_io_error_limit)
602 c->error_limit = strtoul_or_return(buf) << IO_ERROR_SHIFT;
604 /* See count_io_errors() for why 88 */
605 if (attr == &sysfs_io_error_halflife)
606 c->error_decay = strtoul_or_return(buf) / 88;
608 sysfs_strtoul(journal_delay_ms, c->journal_delay_ms);
609 sysfs_strtoul(verify, c->verify);
610 sysfs_strtoul(key_merging_disabled, c->key_merging_disabled);
611 sysfs_strtoul(expensive_debug_checks, c->expensive_debug_checks);
612 sysfs_strtoul(gc_always_rewrite, c->gc_always_rewrite);
613 sysfs_strtoul(btree_shrinker_disabled, c->shrinker_disabled);
614 sysfs_strtoul(copy_gc_enabled, c->copy_gc_enabled);
618 STORE_LOCKED(bch_cache_set)
620 SHOW(bch_cache_set_internal)
622 struct cache_set *c = container_of(kobj, struct cache_set, internal);
623 return bch_cache_set_show(&c->kobj, attr, buf);
626 STORE(bch_cache_set_internal)
628 struct cache_set *c = container_of(kobj, struct cache_set, internal);
629 return bch_cache_set_store(&c->kobj, attr, buf, size);
632 static void bch_cache_set_internal_release(struct kobject *k)
636 static struct attribute *bch_cache_set_files[] = {
640 &sysfs_journal_delay_ms,
641 &sysfs_flash_vol_create,
646 &sysfs_root_usage_percent,
647 &sysfs_btree_cache_size,
648 &sysfs_cache_available_percent,
650 &sysfs_average_key_size,
653 &sysfs_io_error_limit,
654 &sysfs_io_error_halflife,
656 &sysfs_congested_read_threshold_us,
657 &sysfs_congested_write_threshold_us,
661 KTYPE(bch_cache_set);
663 static struct attribute *bch_cache_set_internal_files[] = {
664 &sysfs_active_journal_entries,
666 sysfs_time_stats_attribute_list(btree_gc, sec, ms)
667 sysfs_time_stats_attribute_list(btree_split, sec, us)
668 sysfs_time_stats_attribute_list(btree_sort, ms, us)
669 sysfs_time_stats_attribute_list(btree_read, ms, us)
670 sysfs_time_stats_attribute_list(try_harder, ms, us)
673 &sysfs_btree_used_percent,
674 &sysfs_btree_cache_max_chain,
676 &sysfs_bset_tree_stats,
677 &sysfs_cache_read_races,
678 &sysfs_writeback_keys_done,
679 &sysfs_writeback_keys_failed,
683 #ifdef CONFIG_BCACHE_DEBUG
685 &sysfs_key_merging_disabled,
686 &sysfs_expensive_debug_checks,
688 &sysfs_gc_always_rewrite,
689 &sysfs_btree_shrinker_disabled,
690 &sysfs_copy_gc_enabled,
693 KTYPE(bch_cache_set_internal);
697 struct cache *ca = container_of(kobj, struct cache, kobj);
699 sysfs_hprint(bucket_size, bucket_bytes(ca));
700 sysfs_hprint(block_size, block_bytes(ca));
701 sysfs_print(nbuckets, ca->sb.nbuckets);
702 sysfs_print(discard, ca->discard);
703 sysfs_hprint(written, atomic_long_read(&ca->sectors_written) << 9);
704 sysfs_hprint(btree_written,
705 atomic_long_read(&ca->btree_sectors_written) << 9);
706 sysfs_hprint(metadata_written,
707 (atomic_long_read(&ca->meta_sectors_written) +
708 atomic_long_read(&ca->btree_sectors_written)) << 9);
710 sysfs_print(io_errors,
711 atomic_read(&ca->io_errors) >> IO_ERROR_SHIFT);
713 if (attr == &sysfs_cache_replacement_policy)
714 return bch_snprint_string_list(buf, PAGE_SIZE,
715 cache_replacement_policies,
716 CACHE_REPLACEMENT(&ca->sb));
718 if (attr == &sysfs_priority_stats) {
719 int cmp(const void *l, const void *r)
720 { return *((uint16_t *) r) - *((uint16_t *) l); }
722 size_t n = ca->sb.nbuckets, i, unused, btree;
724 /* Compute 31 quantiles */
725 uint16_t q[31], *p, *cached;
728 cached = p = vmalloc(ca->sb.nbuckets * sizeof(uint16_t));
732 mutex_lock(&ca->set->bucket_lock);
733 for (i = ca->sb.first_bucket; i < n; i++)
734 p[i] = ca->buckets[i].prio;
735 mutex_unlock(&ca->set->bucket_lock);
737 sort(p, n, sizeof(uint16_t), cmp, NULL);
743 unused = ca->sb.nbuckets - n;
745 while (cached < p + n &&
746 *cached == BTREE_PRIO)
752 for (i = 0; i < n; i++)
753 sum += INITIAL_PRIO - cached[i];
758 for (i = 0; i < ARRAY_SIZE(q); i++)
759 q[i] = INITIAL_PRIO - cached[n * (i + 1) /
760 (ARRAY_SIZE(q) + 1)];
764 ret = scnprintf(buf, PAGE_SIZE,
768 "Sectors per Q: %zu\n"
770 unused * 100 / (size_t) ca->sb.nbuckets,
771 btree * 100 / (size_t) ca->sb.nbuckets, sum,
772 n * ca->sb.bucket_size / (ARRAY_SIZE(q) + 1));
774 for (i = 0; i < ARRAY_SIZE(q); i++)
775 ret += scnprintf(buf + ret, PAGE_SIZE - ret,
779 ret += scnprintf(buf + ret, PAGE_SIZE - ret, "]\n");
786 SHOW_LOCKED(bch_cache)
790 struct cache *ca = container_of(kobj, struct cache, kobj);
792 if (attr == &sysfs_discard) {
793 bool v = strtoul_or_return(buf);
795 if (blk_queue_discard(bdev_get_queue(ca->bdev)))
798 if (v != CACHE_DISCARD(&ca->sb)) {
799 SET_CACHE_DISCARD(&ca->sb, v);
800 bcache_write_super(ca->set);
804 if (attr == &sysfs_cache_replacement_policy) {
805 ssize_t v = bch_read_string_list(buf, cache_replacement_policies);
810 if ((unsigned) v != CACHE_REPLACEMENT(&ca->sb)) {
811 mutex_lock(&ca->set->bucket_lock);
812 SET_CACHE_REPLACEMENT(&ca->sb, v);
813 mutex_unlock(&ca->set->bucket_lock);
815 bcache_write_super(ca->set);
819 if (attr == &sysfs_clear_stats) {
820 atomic_long_set(&ca->sectors_written, 0);
821 atomic_long_set(&ca->btree_sectors_written, 0);
822 atomic_long_set(&ca->meta_sectors_written, 0);
823 atomic_set(&ca->io_count, 0);
824 atomic_set(&ca->io_errors, 0);
829 STORE_LOCKED(bch_cache)
831 static struct attribute *bch_cache_files[] = {
835 &sysfs_priority_stats,
838 &sysfs_btree_written,
839 &sysfs_metadata_written,
842 &sysfs_cache_replacement_policy,