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 write_attribute(attach);
25 write_attribute(detach);
26 write_attribute(unregister);
27 write_attribute(stop);
28 write_attribute(clear_stats);
29 write_attribute(trigger_gc);
30 write_attribute(prune_cache);
31 write_attribute(flash_vol_create);
33 read_attribute(bucket_size);
34 read_attribute(block_size);
35 read_attribute(nbuckets);
36 read_attribute(tree_depth);
37 read_attribute(root_usage_percent);
38 read_attribute(priority_stats);
39 read_attribute(btree_cache_size);
40 read_attribute(btree_cache_max_chain);
41 read_attribute(cache_available_percent);
42 read_attribute(written);
43 read_attribute(btree_written);
44 read_attribute(metadata_written);
45 read_attribute(active_journal_entries);
47 sysfs_time_stats_attribute(btree_gc, sec, ms);
48 sysfs_time_stats_attribute(btree_split, sec, us);
49 sysfs_time_stats_attribute(btree_sort, ms, us);
50 sysfs_time_stats_attribute(btree_read, ms, us);
51 sysfs_time_stats_attribute(try_harder, ms, us);
53 read_attribute(btree_nodes);
54 read_attribute(btree_used_percent);
55 read_attribute(average_key_size);
56 read_attribute(dirty_data);
57 read_attribute(bset_tree_stats);
59 read_attribute(state);
60 read_attribute(cache_read_races);
61 read_attribute(writeback_keys_done);
62 read_attribute(writeback_keys_failed);
63 read_attribute(io_errors);
64 read_attribute(congested);
65 rw_attribute(congested_read_threshold_us);
66 rw_attribute(congested_write_threshold_us);
68 rw_attribute(sequential_cutoff);
69 rw_attribute(sequential_merge);
70 rw_attribute(data_csum);
71 rw_attribute(cache_mode);
72 rw_attribute(writeback_metadata);
73 rw_attribute(writeback_running);
74 rw_attribute(writeback_percent);
75 rw_attribute(writeback_delay);
76 rw_attribute(writeback_rate);
78 rw_attribute(writeback_rate_update_seconds);
79 rw_attribute(writeback_rate_d_term);
80 rw_attribute(writeback_rate_p_term_inverse);
81 rw_attribute(writeback_rate_d_smooth);
82 read_attribute(writeback_rate_debug);
84 read_attribute(stripe_size);
85 read_attribute(partial_stripes_expensive);
87 rw_attribute(synchronous);
88 rw_attribute(journal_delay_ms);
89 rw_attribute(discard);
90 rw_attribute(running);
92 rw_attribute(readahead);
93 rw_attribute(io_error_limit);
94 rw_attribute(io_error_halflife);
96 rw_attribute(key_merging_disabled);
97 rw_attribute(gc_always_rewrite);
98 rw_attribute(freelist_percent);
99 rw_attribute(cache_replacement_policy);
100 rw_attribute(btree_shrinker_disabled);
101 rw_attribute(copy_gc_enabled);
104 SHOW(__bch_cached_dev)
106 struct cached_dev *dc = container_of(kobj, struct cached_dev,
108 const char *states[] = { "no cache", "clean", "dirty", "inconsistent" };
110 #define var(stat) (dc->stat)
112 if (attr == &sysfs_cache_mode)
113 return bch_snprint_string_list(buf, PAGE_SIZE,
115 BDEV_CACHE_MODE(&dc->sb));
117 sysfs_printf(data_csum, "%i", dc->disk.data_csum);
118 var_printf(verify, "%i");
119 var_printf(writeback_metadata, "%i");
120 var_printf(writeback_running, "%i");
121 var_print(writeback_delay);
122 var_print(writeback_percent);
123 sysfs_print(writeback_rate, dc->writeback_rate.rate);
125 var_print(writeback_rate_update_seconds);
126 var_print(writeback_rate_d_term);
127 var_print(writeback_rate_p_term_inverse);
128 var_print(writeback_rate_d_smooth);
130 if (attr == &sysfs_writeback_rate_debug) {
135 bcache_dev_sectors_dirty(&dc->disk) << 9);
136 bch_hprint(derivative, dc->writeback_rate_derivative << 9);
137 bch_hprint(target, dc->writeback_rate_target << 9);
145 dc->writeback_rate.rate,
146 dc->writeback_rate_change,
147 dirty, derivative, target);
150 sysfs_hprint(dirty_data,
151 bcache_dev_sectors_dirty(&dc->disk) << 9);
153 sysfs_hprint(stripe_size, (1 << dc->disk.stripe_size_bits) << 9);
154 var_printf(partial_stripes_expensive, "%u");
156 var_printf(sequential_merge, "%i");
157 var_hprint(sequential_cutoff);
158 var_hprint(readahead);
160 sysfs_print(running, atomic_read(&dc->running));
161 sysfs_print(state, states[BDEV_STATE(&dc->sb)]);
163 if (attr == &sysfs_label) {
164 memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
165 buf[SB_LABEL_SIZE + 1] = '\0';
173 SHOW_LOCKED(bch_cached_dev)
177 struct cached_dev *dc = container_of(kobj, struct cached_dev,
181 struct kobj_uevent_env *env;
183 #define d_strtoul(var) sysfs_strtoul(var, dc->var)
184 #define d_strtoi_h(var) sysfs_hatoi(var, dc->var)
186 sysfs_strtoul(data_csum, dc->disk.data_csum);
188 d_strtoul(writeback_metadata);
189 d_strtoul(writeback_running);
190 d_strtoul(writeback_delay);
191 sysfs_strtoul_clamp(writeback_rate,
192 dc->writeback_rate.rate, 1, 1000000);
193 sysfs_strtoul_clamp(writeback_percent, dc->writeback_percent, 0, 40);
195 d_strtoul(writeback_rate_update_seconds);
196 d_strtoul(writeback_rate_d_term);
197 d_strtoul(writeback_rate_p_term_inverse);
198 sysfs_strtoul_clamp(writeback_rate_p_term_inverse,
199 dc->writeback_rate_p_term_inverse, 1, INT_MAX);
200 d_strtoul(writeback_rate_d_smooth);
202 d_strtoul(sequential_merge);
203 d_strtoi_h(sequential_cutoff);
204 d_strtoi_h(readahead);
206 if (attr == &sysfs_clear_stats)
207 bch_cache_accounting_clear(&dc->accounting);
209 if (attr == &sysfs_running &&
210 strtoul_or_return(buf))
211 bch_cached_dev_run(dc);
213 if (attr == &sysfs_cache_mode) {
214 ssize_t v = bch_read_string_list(buf, bch_cache_modes + 1);
219 if ((unsigned) v != BDEV_CACHE_MODE(&dc->sb)) {
220 SET_BDEV_CACHE_MODE(&dc->sb, v);
221 bch_write_bdev_super(dc, NULL);
225 if (attr == &sysfs_label) {
226 if (size > SB_LABEL_SIZE)
228 memcpy(dc->sb.label, buf, size);
229 if (size < SB_LABEL_SIZE)
230 dc->sb.label[size] = '\0';
231 if (size && dc->sb.label[size - 1] == '\n')
232 dc->sb.label[size - 1] = '\0';
233 bch_write_bdev_super(dc, NULL);
235 memcpy(dc->disk.c->uuids[dc->disk.id].label,
237 bch_uuid_write(dc->disk.c);
239 env = kzalloc(sizeof(struct kobj_uevent_env), GFP_KERNEL);
242 add_uevent_var(env, "DRIVER=bcache");
243 add_uevent_var(env, "CACHED_UUID=%pU", dc->sb.uuid),
244 add_uevent_var(env, "CACHED_LABEL=%s", buf);
246 &disk_to_dev(dc->disk.disk)->kobj, KOBJ_CHANGE, env->envp);
250 if (attr == &sysfs_attach) {
251 if (bch_parse_uuid(buf, dc->sb.set_uuid) < 16)
254 list_for_each_entry(c, &bch_cache_sets, list) {
255 v = bch_cached_dev_attach(dc, c);
260 pr_err("Can't attach %s: cache set not found", buf);
264 if (attr == &sysfs_detach && dc->disk.c)
265 bch_cached_dev_detach(dc);
267 if (attr == &sysfs_stop)
268 bcache_device_stop(&dc->disk);
273 STORE(bch_cached_dev)
275 struct cached_dev *dc = container_of(kobj, struct cached_dev,
278 mutex_lock(&bch_register_lock);
279 size = __cached_dev_store(kobj, attr, buf, size);
281 if (attr == &sysfs_writeback_running)
282 bch_writeback_queue(dc);
284 if (attr == &sysfs_writeback_percent)
285 schedule_delayed_work(&dc->writeback_rate_update,
286 dc->writeback_rate_update_seconds * HZ);
288 mutex_unlock(&bch_register_lock);
292 static struct attribute *bch_cached_dev_files[] = {
300 &sysfs_writeback_metadata,
301 &sysfs_writeback_running,
302 &sysfs_writeback_delay,
303 &sysfs_writeback_percent,
304 &sysfs_writeback_rate,
305 &sysfs_writeback_rate_update_seconds,
306 &sysfs_writeback_rate_d_term,
307 &sysfs_writeback_rate_p_term_inverse,
308 &sysfs_writeback_rate_d_smooth,
309 &sysfs_writeback_rate_debug,
312 &sysfs_partial_stripes_expensive,
313 &sysfs_sequential_cutoff,
314 &sysfs_sequential_merge,
320 #ifdef CONFIG_BCACHE_DEBUG
325 KTYPE(bch_cached_dev);
329 struct bcache_device *d = container_of(kobj, struct bcache_device,
331 struct uuid_entry *u = &d->c->uuids[d->id];
333 sysfs_printf(data_csum, "%i", d->data_csum);
334 sysfs_hprint(size, u->sectors << 9);
336 if (attr == &sysfs_label) {
337 memcpy(buf, u->label, SB_LABEL_SIZE);
338 buf[SB_LABEL_SIZE + 1] = '\0';
346 STORE(__bch_flash_dev)
348 struct bcache_device *d = container_of(kobj, struct bcache_device,
350 struct uuid_entry *u = &d->c->uuids[d->id];
352 sysfs_strtoul(data_csum, d->data_csum);
354 if (attr == &sysfs_size) {
356 strtoi_h_or_return(buf, v);
359 bch_uuid_write(d->c);
360 set_capacity(d->disk, u->sectors);
363 if (attr == &sysfs_label) {
364 memcpy(u->label, buf, SB_LABEL_SIZE);
365 bch_uuid_write(d->c);
368 if (attr == &sysfs_unregister) {
369 atomic_set(&d->detaching, 1);
370 bcache_device_stop(d);
375 STORE_LOCKED(bch_flash_dev)
377 static struct attribute *bch_flash_dev_files[] = {
386 KTYPE(bch_flash_dev);
388 SHOW(__bch_cache_set)
390 unsigned root_usage(struct cache_set *c)
395 struct btree_iter iter;
403 rw_lock(false, b, b->level);
404 } while (b != c->root);
406 for_each_key_filter(b, k, &iter, bch_ptr_bad)
407 bytes += bkey_bytes(k);
411 return (bytes * 100) / btree_bytes(c);
414 size_t cache_size(struct cache_set *c)
419 mutex_lock(&c->bucket_lock);
420 list_for_each_entry(b, &c->btree_cache, list)
421 ret += 1 << (b->page_order + PAGE_SHIFT);
423 mutex_unlock(&c->bucket_lock);
427 unsigned cache_max_chain(struct cache_set *c)
430 struct hlist_head *h;
432 mutex_lock(&c->bucket_lock);
434 for (h = c->bucket_hash;
435 h < c->bucket_hash + (1 << BUCKET_HASH_BITS);
438 struct hlist_node *p;
446 mutex_unlock(&c->bucket_lock);
450 unsigned btree_used(struct cache_set *c)
452 return div64_u64(c->gc_stats.key_bytes * 100,
453 (c->gc_stats.nodes ?: 1) * btree_bytes(c));
456 unsigned average_key_size(struct cache_set *c)
458 return c->gc_stats.nkeys
459 ? div64_u64(c->gc_stats.data, c->gc_stats.nkeys)
463 struct cache_set *c = container_of(kobj, struct cache_set, kobj);
465 sysfs_print(synchronous, CACHE_SYNC(&c->sb));
466 sysfs_print(journal_delay_ms, c->journal_delay_ms);
467 sysfs_hprint(bucket_size, bucket_bytes(c));
468 sysfs_hprint(block_size, block_bytes(c));
469 sysfs_print(tree_depth, c->root->level);
470 sysfs_print(root_usage_percent, root_usage(c));
472 sysfs_hprint(btree_cache_size, cache_size(c));
473 sysfs_print(btree_cache_max_chain, cache_max_chain(c));
474 sysfs_print(cache_available_percent, 100 - c->gc_stats.in_use);
476 sysfs_print_time_stats(&c->btree_gc_time, btree_gc, sec, ms);
477 sysfs_print_time_stats(&c->btree_split_time, btree_split, sec, us);
478 sysfs_print_time_stats(&c->sort_time, btree_sort, ms, us);
479 sysfs_print_time_stats(&c->btree_read_time, btree_read, ms, us);
480 sysfs_print_time_stats(&c->try_harder_time, try_harder, ms, us);
482 sysfs_print(btree_used_percent, btree_used(c));
483 sysfs_print(btree_nodes, c->gc_stats.nodes);
484 sysfs_hprint(dirty_data, c->gc_stats.dirty);
485 sysfs_hprint(average_key_size, average_key_size(c));
487 sysfs_print(cache_read_races,
488 atomic_long_read(&c->cache_read_races));
490 sysfs_print(writeback_keys_done,
491 atomic_long_read(&c->writeback_keys_done));
492 sysfs_print(writeback_keys_failed,
493 atomic_long_read(&c->writeback_keys_failed));
495 /* See count_io_errors for why 88 */
496 sysfs_print(io_error_halflife, c->error_decay * 88);
497 sysfs_print(io_error_limit, c->error_limit >> IO_ERROR_SHIFT);
499 sysfs_hprint(congested,
500 ((uint64_t) bch_get_congested(c)) << 9);
501 sysfs_print(congested_read_threshold_us,
502 c->congested_read_threshold_us);
503 sysfs_print(congested_write_threshold_us,
504 c->congested_write_threshold_us);
506 sysfs_print(active_journal_entries, fifo_used(&c->journal.pin));
507 sysfs_printf(verify, "%i", c->verify);
508 sysfs_printf(key_merging_disabled, "%i", c->key_merging_disabled);
509 sysfs_printf(gc_always_rewrite, "%i", c->gc_always_rewrite);
510 sysfs_printf(btree_shrinker_disabled, "%i", c->shrinker_disabled);
511 sysfs_printf(copy_gc_enabled, "%i", c->copy_gc_enabled);
513 if (attr == &sysfs_bset_tree_stats)
514 return bch_bset_print_stats(c, buf);
518 SHOW_LOCKED(bch_cache_set)
520 STORE(__bch_cache_set)
522 struct cache_set *c = container_of(kobj, struct cache_set, kobj);
524 if (attr == &sysfs_unregister)
525 bch_cache_set_unregister(c);
527 if (attr == &sysfs_stop)
528 bch_cache_set_stop(c);
530 if (attr == &sysfs_synchronous) {
531 bool sync = strtoul_or_return(buf);
533 if (sync != CACHE_SYNC(&c->sb)) {
534 SET_CACHE_SYNC(&c->sb, sync);
535 bcache_write_super(c);
539 if (attr == &sysfs_flash_vol_create) {
542 strtoi_h_or_return(buf, v);
544 r = bch_flash_dev_create(c, v);
549 if (attr == &sysfs_clear_stats) {
550 atomic_long_set(&c->writeback_keys_done, 0);
551 atomic_long_set(&c->writeback_keys_failed, 0);
553 memset(&c->gc_stats, 0, sizeof(struct gc_stat));
554 bch_cache_accounting_clear(&c->accounting);
557 if (attr == &sysfs_trigger_gc)
560 if (attr == &sysfs_prune_cache) {
561 struct shrink_control sc;
562 sc.gfp_mask = GFP_KERNEL;
563 sc.nr_to_scan = strtoul_or_return(buf);
564 c->shrink.scan_objects(&c->shrink, &sc);
567 sysfs_strtoul(congested_read_threshold_us,
568 c->congested_read_threshold_us);
569 sysfs_strtoul(congested_write_threshold_us,
570 c->congested_write_threshold_us);
572 if (attr == &sysfs_io_error_limit)
573 c->error_limit = strtoul_or_return(buf) << IO_ERROR_SHIFT;
575 /* See count_io_errors() for why 88 */
576 if (attr == &sysfs_io_error_halflife)
577 c->error_decay = strtoul_or_return(buf) / 88;
579 sysfs_strtoul(journal_delay_ms, c->journal_delay_ms);
580 sysfs_strtoul(verify, c->verify);
581 sysfs_strtoul(key_merging_disabled, c->key_merging_disabled);
582 sysfs_strtoul(gc_always_rewrite, c->gc_always_rewrite);
583 sysfs_strtoul(btree_shrinker_disabled, c->shrinker_disabled);
584 sysfs_strtoul(copy_gc_enabled, c->copy_gc_enabled);
588 STORE_LOCKED(bch_cache_set)
590 SHOW(bch_cache_set_internal)
592 struct cache_set *c = container_of(kobj, struct cache_set, internal);
593 return bch_cache_set_show(&c->kobj, attr, buf);
596 STORE(bch_cache_set_internal)
598 struct cache_set *c = container_of(kobj, struct cache_set, internal);
599 return bch_cache_set_store(&c->kobj, attr, buf, size);
602 static void bch_cache_set_internal_release(struct kobject *k)
606 static struct attribute *bch_cache_set_files[] = {
610 &sysfs_journal_delay_ms,
611 &sysfs_flash_vol_create,
616 &sysfs_root_usage_percent,
617 &sysfs_btree_cache_size,
618 &sysfs_cache_available_percent,
620 &sysfs_average_key_size,
623 &sysfs_io_error_limit,
624 &sysfs_io_error_halflife,
626 &sysfs_congested_read_threshold_us,
627 &sysfs_congested_write_threshold_us,
631 KTYPE(bch_cache_set);
633 static struct attribute *bch_cache_set_internal_files[] = {
634 &sysfs_active_journal_entries,
636 sysfs_time_stats_attribute_list(btree_gc, sec, ms)
637 sysfs_time_stats_attribute_list(btree_split, sec, us)
638 sysfs_time_stats_attribute_list(btree_sort, ms, us)
639 sysfs_time_stats_attribute_list(btree_read, ms, us)
640 sysfs_time_stats_attribute_list(try_harder, ms, us)
643 &sysfs_btree_used_percent,
644 &sysfs_btree_cache_max_chain,
646 &sysfs_bset_tree_stats,
647 &sysfs_cache_read_races,
648 &sysfs_writeback_keys_done,
649 &sysfs_writeback_keys_failed,
653 #ifdef CONFIG_BCACHE_DEBUG
655 &sysfs_key_merging_disabled,
657 &sysfs_gc_always_rewrite,
658 &sysfs_btree_shrinker_disabled,
659 &sysfs_copy_gc_enabled,
662 KTYPE(bch_cache_set_internal);
666 struct cache *ca = container_of(kobj, struct cache, kobj);
668 sysfs_hprint(bucket_size, bucket_bytes(ca));
669 sysfs_hprint(block_size, block_bytes(ca));
670 sysfs_print(nbuckets, ca->sb.nbuckets);
671 sysfs_print(discard, ca->discard);
672 sysfs_hprint(written, atomic_long_read(&ca->sectors_written) << 9);
673 sysfs_hprint(btree_written,
674 atomic_long_read(&ca->btree_sectors_written) << 9);
675 sysfs_hprint(metadata_written,
676 (atomic_long_read(&ca->meta_sectors_written) +
677 atomic_long_read(&ca->btree_sectors_written)) << 9);
679 sysfs_print(io_errors,
680 atomic_read(&ca->io_errors) >> IO_ERROR_SHIFT);
682 sysfs_print(freelist_percent, ca->free.size * 100 /
683 ((size_t) ca->sb.nbuckets));
685 if (attr == &sysfs_cache_replacement_policy)
686 return bch_snprint_string_list(buf, PAGE_SIZE,
687 cache_replacement_policies,
688 CACHE_REPLACEMENT(&ca->sb));
690 if (attr == &sysfs_priority_stats) {
691 int cmp(const void *l, const void *r)
692 { return *((uint16_t *) r) - *((uint16_t *) l); }
694 size_t n = ca->sb.nbuckets, i, unused, btree;
696 /* Compute 31 quantiles */
697 uint16_t q[31], *p, *cached;
700 cached = p = vmalloc(ca->sb.nbuckets * sizeof(uint16_t));
704 mutex_lock(&ca->set->bucket_lock);
705 for (i = ca->sb.first_bucket; i < n; i++)
706 p[i] = ca->buckets[i].prio;
707 mutex_unlock(&ca->set->bucket_lock);
709 sort(p, n, sizeof(uint16_t), cmp, NULL);
715 unused = ca->sb.nbuckets - n;
717 while (cached < p + n &&
718 *cached == BTREE_PRIO)
724 for (i = 0; i < n; i++)
725 sum += INITIAL_PRIO - cached[i];
730 for (i = 0; i < ARRAY_SIZE(q); i++)
731 q[i] = INITIAL_PRIO - cached[n * (i + 1) /
732 (ARRAY_SIZE(q) + 1)];
736 ret = scnprintf(buf, PAGE_SIZE,
740 "Sectors per Q: %zu\n"
742 unused * 100 / (size_t) ca->sb.nbuckets,
743 btree * 100 / (size_t) ca->sb.nbuckets, sum,
744 n * ca->sb.bucket_size / (ARRAY_SIZE(q) + 1));
746 for (i = 0; i < ARRAY_SIZE(q); i++)
747 ret += scnprintf(buf + ret, PAGE_SIZE - ret,
751 ret += scnprintf(buf + ret, PAGE_SIZE - ret, "]\n");
758 SHOW_LOCKED(bch_cache)
762 struct cache *ca = container_of(kobj, struct cache, kobj);
764 if (attr == &sysfs_discard) {
765 bool v = strtoul_or_return(buf);
767 if (blk_queue_discard(bdev_get_queue(ca->bdev)))
770 if (v != CACHE_DISCARD(&ca->sb)) {
771 SET_CACHE_DISCARD(&ca->sb, v);
772 bcache_write_super(ca->set);
776 if (attr == &sysfs_cache_replacement_policy) {
777 ssize_t v = bch_read_string_list(buf, cache_replacement_policies);
782 if ((unsigned) v != CACHE_REPLACEMENT(&ca->sb)) {
783 mutex_lock(&ca->set->bucket_lock);
784 SET_CACHE_REPLACEMENT(&ca->sb, v);
785 mutex_unlock(&ca->set->bucket_lock);
787 bcache_write_super(ca->set);
791 if (attr == &sysfs_freelist_percent) {
792 DECLARE_FIFO(long, free);
794 size_t p = strtoul_or_return(buf);
797 ((size_t) ca->sb.nbuckets * p) / 100,
798 roundup_pow_of_two(ca->sb.nbuckets) >> 9,
799 ca->sb.nbuckets / 2);
801 if (!init_fifo_exact(&free, p, GFP_KERNEL))
804 mutex_lock(&ca->set->bucket_lock);
806 fifo_move(&free, &ca->free);
807 fifo_swap(&free, &ca->free);
809 mutex_unlock(&ca->set->bucket_lock);
811 while (fifo_pop(&free, i))
812 atomic_dec(&ca->buckets[i].pin);
817 if (attr == &sysfs_clear_stats) {
818 atomic_long_set(&ca->sectors_written, 0);
819 atomic_long_set(&ca->btree_sectors_written, 0);
820 atomic_long_set(&ca->meta_sectors_written, 0);
821 atomic_set(&ca->io_count, 0);
822 atomic_set(&ca->io_errors, 0);
827 STORE_LOCKED(bch_cache)
829 static struct attribute *bch_cache_files[] = {
833 &sysfs_priority_stats,
836 &sysfs_btree_written,
837 &sysfs_metadata_written,
840 &sysfs_freelist_percent,
841 &sysfs_cache_replacement_policy,