xref: /openbmc/linux/drivers/md/bcache/sysfs.c (revision eb3fcf00)
1 /*
2  * bcache sysfs interfaces
3  *
4  * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
5  * Copyright 2012 Google, Inc.
6  */
7 
8 #include "bcache.h"
9 #include "sysfs.h"
10 #include "btree.h"
11 #include "request.h"
12 #include "writeback.h"
13 
14 #include <linux/blkdev.h>
15 #include <linux/sort.h>
16 
17 static const char * const cache_replacement_policies[] = {
18 	"lru",
19 	"fifo",
20 	"random",
21 	NULL
22 };
23 
24 static const char * const error_actions[] = {
25 	"unregister",
26 	"panic",
27 	NULL
28 };
29 
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);
38 
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);
52 
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 
58 read_attribute(btree_nodes);
59 read_attribute(btree_used_percent);
60 read_attribute(average_key_size);
61 read_attribute(dirty_data);
62 read_attribute(bset_tree_stats);
63 
64 read_attribute(state);
65 read_attribute(cache_read_races);
66 read_attribute(writeback_keys_done);
67 read_attribute(writeback_keys_failed);
68 read_attribute(io_errors);
69 read_attribute(congested);
70 rw_attribute(congested_read_threshold_us);
71 rw_attribute(congested_write_threshold_us);
72 
73 rw_attribute(sequential_cutoff);
74 rw_attribute(data_csum);
75 rw_attribute(cache_mode);
76 rw_attribute(writeback_metadata);
77 rw_attribute(writeback_running);
78 rw_attribute(writeback_percent);
79 rw_attribute(writeback_delay);
80 rw_attribute(writeback_rate);
81 
82 rw_attribute(writeback_rate_update_seconds);
83 rw_attribute(writeback_rate_d_term);
84 rw_attribute(writeback_rate_p_term_inverse);
85 read_attribute(writeback_rate_debug);
86 
87 read_attribute(stripe_size);
88 read_attribute(partial_stripes_expensive);
89 
90 rw_attribute(synchronous);
91 rw_attribute(journal_delay_ms);
92 rw_attribute(discard);
93 rw_attribute(running);
94 rw_attribute(label);
95 rw_attribute(readahead);
96 rw_attribute(errors);
97 rw_attribute(io_error_limit);
98 rw_attribute(io_error_halflife);
99 rw_attribute(verify);
100 rw_attribute(bypass_torture_test);
101 rw_attribute(key_merging_disabled);
102 rw_attribute(gc_always_rewrite);
103 rw_attribute(expensive_debug_checks);
104 rw_attribute(cache_replacement_policy);
105 rw_attribute(btree_shrinker_disabled);
106 rw_attribute(copy_gc_enabled);
107 rw_attribute(size);
108 
109 SHOW(__bch_cached_dev)
110 {
111 	struct cached_dev *dc = container_of(kobj, struct cached_dev,
112 					     disk.kobj);
113 	const char *states[] = { "no cache", "clean", "dirty", "inconsistent" };
114 
115 #define var(stat)		(dc->stat)
116 
117 	if (attr == &sysfs_cache_mode)
118 		return bch_snprint_string_list(buf, PAGE_SIZE,
119 					       bch_cache_modes + 1,
120 					       BDEV_CACHE_MODE(&dc->sb));
121 
122 	sysfs_printf(data_csum,		"%i", dc->disk.data_csum);
123 	var_printf(verify,		"%i");
124 	var_printf(bypass_torture_test,	"%i");
125 	var_printf(writeback_metadata,	"%i");
126 	var_printf(writeback_running,	"%i");
127 	var_print(writeback_delay);
128 	var_print(writeback_percent);
129 	sysfs_hprint(writeback_rate,	dc->writeback_rate.rate << 9);
130 
131 	var_print(writeback_rate_update_seconds);
132 	var_print(writeback_rate_d_term);
133 	var_print(writeback_rate_p_term_inverse);
134 
135 	if (attr == &sysfs_writeback_rate_debug) {
136 		char rate[20];
137 		char dirty[20];
138 		char target[20];
139 		char proportional[20];
140 		char derivative[20];
141 		char change[20];
142 		s64 next_io;
143 
144 		bch_hprint(rate,	dc->writeback_rate.rate << 9);
145 		bch_hprint(dirty,	bcache_dev_sectors_dirty(&dc->disk) << 9);
146 		bch_hprint(target,	dc->writeback_rate_target << 9);
147 		bch_hprint(proportional,dc->writeback_rate_proportional << 9);
148 		bch_hprint(derivative,	dc->writeback_rate_derivative << 9);
149 		bch_hprint(change,	dc->writeback_rate_change << 9);
150 
151 		next_io = div64_s64(dc->writeback_rate.next - local_clock(),
152 				    NSEC_PER_MSEC);
153 
154 		return sprintf(buf,
155 			       "rate:\t\t%s/sec\n"
156 			       "dirty:\t\t%s\n"
157 			       "target:\t\t%s\n"
158 			       "proportional:\t%s\n"
159 			       "derivative:\t%s\n"
160 			       "change:\t\t%s/sec\n"
161 			       "next io:\t%llims\n",
162 			       rate, dirty, target, proportional,
163 			       derivative, change, next_io);
164 	}
165 
166 	sysfs_hprint(dirty_data,
167 		     bcache_dev_sectors_dirty(&dc->disk) << 9);
168 
169 	sysfs_hprint(stripe_size,	dc->disk.stripe_size << 9);
170 	var_printf(partial_stripes_expensive,	"%u");
171 
172 	var_hprint(sequential_cutoff);
173 	var_hprint(readahead);
174 
175 	sysfs_print(running,		atomic_read(&dc->running));
176 	sysfs_print(state,		states[BDEV_STATE(&dc->sb)]);
177 
178 	if (attr == &sysfs_label) {
179 		memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
180 		buf[SB_LABEL_SIZE + 1] = '\0';
181 		strcat(buf, "\n");
182 		return strlen(buf);
183 	}
184 
185 #undef var
186 	return 0;
187 }
188 SHOW_LOCKED(bch_cached_dev)
189 
190 STORE(__cached_dev)
191 {
192 	struct cached_dev *dc = container_of(kobj, struct cached_dev,
193 					     disk.kobj);
194 	unsigned v = size;
195 	struct cache_set *c;
196 	struct kobj_uevent_env *env;
197 
198 #define d_strtoul(var)		sysfs_strtoul(var, dc->var)
199 #define d_strtoul_nonzero(var)	sysfs_strtoul_clamp(var, dc->var, 1, INT_MAX)
200 #define d_strtoi_h(var)		sysfs_hatoi(var, dc->var)
201 
202 	sysfs_strtoul(data_csum,	dc->disk.data_csum);
203 	d_strtoul(verify);
204 	d_strtoul(bypass_torture_test);
205 	d_strtoul(writeback_metadata);
206 	d_strtoul(writeback_running);
207 	d_strtoul(writeback_delay);
208 
209 	sysfs_strtoul_clamp(writeback_percent, dc->writeback_percent, 0, 40);
210 
211 	sysfs_strtoul_clamp(writeback_rate,
212 			    dc->writeback_rate.rate, 1, INT_MAX);
213 
214 	d_strtoul_nonzero(writeback_rate_update_seconds);
215 	d_strtoul(writeback_rate_d_term);
216 	d_strtoul_nonzero(writeback_rate_p_term_inverse);
217 
218 	d_strtoi_h(sequential_cutoff);
219 	d_strtoi_h(readahead);
220 
221 	if (attr == &sysfs_clear_stats)
222 		bch_cache_accounting_clear(&dc->accounting);
223 
224 	if (attr == &sysfs_running &&
225 	    strtoul_or_return(buf))
226 		bch_cached_dev_run(dc);
227 
228 	if (attr == &sysfs_cache_mode) {
229 		ssize_t v = bch_read_string_list(buf, bch_cache_modes + 1);
230 
231 		if (v < 0)
232 			return v;
233 
234 		if ((unsigned) v != BDEV_CACHE_MODE(&dc->sb)) {
235 			SET_BDEV_CACHE_MODE(&dc->sb, v);
236 			bch_write_bdev_super(dc, NULL);
237 		}
238 	}
239 
240 	if (attr == &sysfs_label) {
241 		if (size > SB_LABEL_SIZE)
242 			return -EINVAL;
243 		memcpy(dc->sb.label, buf, size);
244 		if (size < SB_LABEL_SIZE)
245 			dc->sb.label[size] = '\0';
246 		if (size && dc->sb.label[size - 1] == '\n')
247 			dc->sb.label[size - 1] = '\0';
248 		bch_write_bdev_super(dc, NULL);
249 		if (dc->disk.c) {
250 			memcpy(dc->disk.c->uuids[dc->disk.id].label,
251 			       buf, SB_LABEL_SIZE);
252 			bch_uuid_write(dc->disk.c);
253 		}
254 		env = kzalloc(sizeof(struct kobj_uevent_env), GFP_KERNEL);
255 		if (!env)
256 			return -ENOMEM;
257 		add_uevent_var(env, "DRIVER=bcache");
258 		add_uevent_var(env, "CACHED_UUID=%pU", dc->sb.uuid),
259 		add_uevent_var(env, "CACHED_LABEL=%s", buf);
260 		kobject_uevent_env(
261 			&disk_to_dev(dc->disk.disk)->kobj, KOBJ_CHANGE, env->envp);
262 		kfree(env);
263 	}
264 
265 	if (attr == &sysfs_attach) {
266 		if (bch_parse_uuid(buf, dc->sb.set_uuid) < 16)
267 			return -EINVAL;
268 
269 		list_for_each_entry(c, &bch_cache_sets, list) {
270 			v = bch_cached_dev_attach(dc, c);
271 			if (!v)
272 				return size;
273 		}
274 
275 		pr_err("Can't attach %s: cache set not found", buf);
276 		size = v;
277 	}
278 
279 	if (attr == &sysfs_detach && dc->disk.c)
280 		bch_cached_dev_detach(dc);
281 
282 	if (attr == &sysfs_stop)
283 		bcache_device_stop(&dc->disk);
284 
285 	return size;
286 }
287 
288 STORE(bch_cached_dev)
289 {
290 	struct cached_dev *dc = container_of(kobj, struct cached_dev,
291 					     disk.kobj);
292 
293 	mutex_lock(&bch_register_lock);
294 	size = __cached_dev_store(kobj, attr, buf, size);
295 
296 	if (attr == &sysfs_writeback_running)
297 		bch_writeback_queue(dc);
298 
299 	if (attr == &sysfs_writeback_percent)
300 		schedule_delayed_work(&dc->writeback_rate_update,
301 				      dc->writeback_rate_update_seconds * HZ);
302 
303 	mutex_unlock(&bch_register_lock);
304 	return size;
305 }
306 
307 static struct attribute *bch_cached_dev_files[] = {
308 	&sysfs_attach,
309 	&sysfs_detach,
310 	&sysfs_stop,
311 #if 0
312 	&sysfs_data_csum,
313 #endif
314 	&sysfs_cache_mode,
315 	&sysfs_writeback_metadata,
316 	&sysfs_writeback_running,
317 	&sysfs_writeback_delay,
318 	&sysfs_writeback_percent,
319 	&sysfs_writeback_rate,
320 	&sysfs_writeback_rate_update_seconds,
321 	&sysfs_writeback_rate_d_term,
322 	&sysfs_writeback_rate_p_term_inverse,
323 	&sysfs_writeback_rate_debug,
324 	&sysfs_dirty_data,
325 	&sysfs_stripe_size,
326 	&sysfs_partial_stripes_expensive,
327 	&sysfs_sequential_cutoff,
328 	&sysfs_clear_stats,
329 	&sysfs_running,
330 	&sysfs_state,
331 	&sysfs_label,
332 	&sysfs_readahead,
333 #ifdef CONFIG_BCACHE_DEBUG
334 	&sysfs_verify,
335 	&sysfs_bypass_torture_test,
336 #endif
337 	NULL
338 };
339 KTYPE(bch_cached_dev);
340 
341 SHOW(bch_flash_dev)
342 {
343 	struct bcache_device *d = container_of(kobj, struct bcache_device,
344 					       kobj);
345 	struct uuid_entry *u = &d->c->uuids[d->id];
346 
347 	sysfs_printf(data_csum,	"%i", d->data_csum);
348 	sysfs_hprint(size,	u->sectors << 9);
349 
350 	if (attr == &sysfs_label) {
351 		memcpy(buf, u->label, SB_LABEL_SIZE);
352 		buf[SB_LABEL_SIZE + 1] = '\0';
353 		strcat(buf, "\n");
354 		return strlen(buf);
355 	}
356 
357 	return 0;
358 }
359 
360 STORE(__bch_flash_dev)
361 {
362 	struct bcache_device *d = container_of(kobj, struct bcache_device,
363 					       kobj);
364 	struct uuid_entry *u = &d->c->uuids[d->id];
365 
366 	sysfs_strtoul(data_csum,	d->data_csum);
367 
368 	if (attr == &sysfs_size) {
369 		uint64_t v;
370 		strtoi_h_or_return(buf, v);
371 
372 		u->sectors = v >> 9;
373 		bch_uuid_write(d->c);
374 		set_capacity(d->disk, u->sectors);
375 	}
376 
377 	if (attr == &sysfs_label) {
378 		memcpy(u->label, buf, SB_LABEL_SIZE);
379 		bch_uuid_write(d->c);
380 	}
381 
382 	if (attr == &sysfs_unregister) {
383 		set_bit(BCACHE_DEV_DETACHING, &d->flags);
384 		bcache_device_stop(d);
385 	}
386 
387 	return size;
388 }
389 STORE_LOCKED(bch_flash_dev)
390 
391 static struct attribute *bch_flash_dev_files[] = {
392 	&sysfs_unregister,
393 #if 0
394 	&sysfs_data_csum,
395 #endif
396 	&sysfs_label,
397 	&sysfs_size,
398 	NULL
399 };
400 KTYPE(bch_flash_dev);
401 
402 struct bset_stats_op {
403 	struct btree_op op;
404 	size_t nodes;
405 	struct bset_stats stats;
406 };
407 
408 static int bch_btree_bset_stats(struct btree_op *b_op, struct btree *b)
409 {
410 	struct bset_stats_op *op = container_of(b_op, struct bset_stats_op, op);
411 
412 	op->nodes++;
413 	bch_btree_keys_stats(&b->keys, &op->stats);
414 
415 	return MAP_CONTINUE;
416 }
417 
418 static int bch_bset_print_stats(struct cache_set *c, char *buf)
419 {
420 	struct bset_stats_op op;
421 	int ret;
422 
423 	memset(&op, 0, sizeof(op));
424 	bch_btree_op_init(&op.op, -1);
425 
426 	ret = bch_btree_map_nodes(&op.op, c, &ZERO_KEY, bch_btree_bset_stats);
427 	if (ret < 0)
428 		return ret;
429 
430 	return snprintf(buf, PAGE_SIZE,
431 			"btree nodes:		%zu\n"
432 			"written sets:		%zu\n"
433 			"unwritten sets:		%zu\n"
434 			"written key bytes:	%zu\n"
435 			"unwritten key bytes:	%zu\n"
436 			"floats:			%zu\n"
437 			"failed:			%zu\n",
438 			op.nodes,
439 			op.stats.sets_written, op.stats.sets_unwritten,
440 			op.stats.bytes_written, op.stats.bytes_unwritten,
441 			op.stats.floats, op.stats.failed);
442 }
443 
444 static unsigned bch_root_usage(struct cache_set *c)
445 {
446 	unsigned bytes = 0;
447 	struct bkey *k;
448 	struct btree *b;
449 	struct btree_iter iter;
450 
451 	goto lock_root;
452 
453 	do {
454 		rw_unlock(false, b);
455 lock_root:
456 		b = c->root;
457 		rw_lock(false, b, b->level);
458 	} while (b != c->root);
459 
460 	for_each_key_filter(&b->keys, k, &iter, bch_ptr_bad)
461 		bytes += bkey_bytes(k);
462 
463 	rw_unlock(false, b);
464 
465 	return (bytes * 100) / btree_bytes(c);
466 }
467 
468 static size_t bch_cache_size(struct cache_set *c)
469 {
470 	size_t ret = 0;
471 	struct btree *b;
472 
473 	mutex_lock(&c->bucket_lock);
474 	list_for_each_entry(b, &c->btree_cache, list)
475 		ret += 1 << (b->keys.page_order + PAGE_SHIFT);
476 
477 	mutex_unlock(&c->bucket_lock);
478 	return ret;
479 }
480 
481 static unsigned bch_cache_max_chain(struct cache_set *c)
482 {
483 	unsigned ret = 0;
484 	struct hlist_head *h;
485 
486 	mutex_lock(&c->bucket_lock);
487 
488 	for (h = c->bucket_hash;
489 	     h < c->bucket_hash + (1 << BUCKET_HASH_BITS);
490 	     h++) {
491 		unsigned i = 0;
492 		struct hlist_node *p;
493 
494 		hlist_for_each(p, h)
495 			i++;
496 
497 		ret = max(ret, i);
498 	}
499 
500 	mutex_unlock(&c->bucket_lock);
501 	return ret;
502 }
503 
504 static unsigned bch_btree_used(struct cache_set *c)
505 {
506 	return div64_u64(c->gc_stats.key_bytes * 100,
507 			 (c->gc_stats.nodes ?: 1) * btree_bytes(c));
508 }
509 
510 static unsigned bch_average_key_size(struct cache_set *c)
511 {
512 	return c->gc_stats.nkeys
513 		? div64_u64(c->gc_stats.data, c->gc_stats.nkeys)
514 		: 0;
515 }
516 
517 SHOW(__bch_cache_set)
518 {
519 	struct cache_set *c = container_of(kobj, struct cache_set, kobj);
520 
521 	sysfs_print(synchronous,		CACHE_SYNC(&c->sb));
522 	sysfs_print(journal_delay_ms,		c->journal_delay_ms);
523 	sysfs_hprint(bucket_size,		bucket_bytes(c));
524 	sysfs_hprint(block_size,		block_bytes(c));
525 	sysfs_print(tree_depth,			c->root->level);
526 	sysfs_print(root_usage_percent,		bch_root_usage(c));
527 
528 	sysfs_hprint(btree_cache_size,		bch_cache_size(c));
529 	sysfs_print(btree_cache_max_chain,	bch_cache_max_chain(c));
530 	sysfs_print(cache_available_percent,	100 - c->gc_stats.in_use);
531 
532 	sysfs_print_time_stats(&c->btree_gc_time,	btree_gc, sec, ms);
533 	sysfs_print_time_stats(&c->btree_split_time,	btree_split, sec, us);
534 	sysfs_print_time_stats(&c->sort.time,		btree_sort, ms, us);
535 	sysfs_print_time_stats(&c->btree_read_time,	btree_read, ms, us);
536 
537 	sysfs_print(btree_used_percent,	bch_btree_used(c));
538 	sysfs_print(btree_nodes,	c->gc_stats.nodes);
539 	sysfs_hprint(average_key_size,	bch_average_key_size(c));
540 
541 	sysfs_print(cache_read_races,
542 		    atomic_long_read(&c->cache_read_races));
543 
544 	sysfs_print(writeback_keys_done,
545 		    atomic_long_read(&c->writeback_keys_done));
546 	sysfs_print(writeback_keys_failed,
547 		    atomic_long_read(&c->writeback_keys_failed));
548 
549 	if (attr == &sysfs_errors)
550 		return bch_snprint_string_list(buf, PAGE_SIZE, error_actions,
551 					       c->on_error);
552 
553 	/* See count_io_errors for why 88 */
554 	sysfs_print(io_error_halflife,	c->error_decay * 88);
555 	sysfs_print(io_error_limit,	c->error_limit >> IO_ERROR_SHIFT);
556 
557 	sysfs_hprint(congested,
558 		     ((uint64_t) bch_get_congested(c)) << 9);
559 	sysfs_print(congested_read_threshold_us,
560 		    c->congested_read_threshold_us);
561 	sysfs_print(congested_write_threshold_us,
562 		    c->congested_write_threshold_us);
563 
564 	sysfs_print(active_journal_entries,	fifo_used(&c->journal.pin));
565 	sysfs_printf(verify,			"%i", c->verify);
566 	sysfs_printf(key_merging_disabled,	"%i", c->key_merging_disabled);
567 	sysfs_printf(expensive_debug_checks,
568 		     "%i", c->expensive_debug_checks);
569 	sysfs_printf(gc_always_rewrite,		"%i", c->gc_always_rewrite);
570 	sysfs_printf(btree_shrinker_disabled,	"%i", c->shrinker_disabled);
571 	sysfs_printf(copy_gc_enabled,		"%i", c->copy_gc_enabled);
572 
573 	if (attr == &sysfs_bset_tree_stats)
574 		return bch_bset_print_stats(c, buf);
575 
576 	return 0;
577 }
578 SHOW_LOCKED(bch_cache_set)
579 
580 STORE(__bch_cache_set)
581 {
582 	struct cache_set *c = container_of(kobj, struct cache_set, kobj);
583 
584 	if (attr == &sysfs_unregister)
585 		bch_cache_set_unregister(c);
586 
587 	if (attr == &sysfs_stop)
588 		bch_cache_set_stop(c);
589 
590 	if (attr == &sysfs_synchronous) {
591 		bool sync = strtoul_or_return(buf);
592 
593 		if (sync != CACHE_SYNC(&c->sb)) {
594 			SET_CACHE_SYNC(&c->sb, sync);
595 			bcache_write_super(c);
596 		}
597 	}
598 
599 	if (attr == &sysfs_flash_vol_create) {
600 		int r;
601 		uint64_t v;
602 		strtoi_h_or_return(buf, v);
603 
604 		r = bch_flash_dev_create(c, v);
605 		if (r)
606 			return r;
607 	}
608 
609 	if (attr == &sysfs_clear_stats) {
610 		atomic_long_set(&c->writeback_keys_done,	0);
611 		atomic_long_set(&c->writeback_keys_failed,	0);
612 
613 		memset(&c->gc_stats, 0, sizeof(struct gc_stat));
614 		bch_cache_accounting_clear(&c->accounting);
615 	}
616 
617 	if (attr == &sysfs_trigger_gc)
618 		wake_up_gc(c);
619 
620 	if (attr == &sysfs_prune_cache) {
621 		struct shrink_control sc;
622 		sc.gfp_mask = GFP_KERNEL;
623 		sc.nr_to_scan = strtoul_or_return(buf);
624 		c->shrink.scan_objects(&c->shrink, &sc);
625 	}
626 
627 	sysfs_strtoul(congested_read_threshold_us,
628 		      c->congested_read_threshold_us);
629 	sysfs_strtoul(congested_write_threshold_us,
630 		      c->congested_write_threshold_us);
631 
632 	if (attr == &sysfs_errors) {
633 		ssize_t v = bch_read_string_list(buf, error_actions);
634 
635 		if (v < 0)
636 			return v;
637 
638 		c->on_error = v;
639 	}
640 
641 	if (attr == &sysfs_io_error_limit)
642 		c->error_limit = strtoul_or_return(buf) << IO_ERROR_SHIFT;
643 
644 	/* See count_io_errors() for why 88 */
645 	if (attr == &sysfs_io_error_halflife)
646 		c->error_decay = strtoul_or_return(buf) / 88;
647 
648 	sysfs_strtoul(journal_delay_ms,		c->journal_delay_ms);
649 	sysfs_strtoul(verify,			c->verify);
650 	sysfs_strtoul(key_merging_disabled,	c->key_merging_disabled);
651 	sysfs_strtoul(expensive_debug_checks,	c->expensive_debug_checks);
652 	sysfs_strtoul(gc_always_rewrite,	c->gc_always_rewrite);
653 	sysfs_strtoul(btree_shrinker_disabled,	c->shrinker_disabled);
654 	sysfs_strtoul(copy_gc_enabled,		c->copy_gc_enabled);
655 
656 	return size;
657 }
658 STORE_LOCKED(bch_cache_set)
659 
660 SHOW(bch_cache_set_internal)
661 {
662 	struct cache_set *c = container_of(kobj, struct cache_set, internal);
663 	return bch_cache_set_show(&c->kobj, attr, buf);
664 }
665 
666 STORE(bch_cache_set_internal)
667 {
668 	struct cache_set *c = container_of(kobj, struct cache_set, internal);
669 	return bch_cache_set_store(&c->kobj, attr, buf, size);
670 }
671 
672 static void bch_cache_set_internal_release(struct kobject *k)
673 {
674 }
675 
676 static struct attribute *bch_cache_set_files[] = {
677 	&sysfs_unregister,
678 	&sysfs_stop,
679 	&sysfs_synchronous,
680 	&sysfs_journal_delay_ms,
681 	&sysfs_flash_vol_create,
682 
683 	&sysfs_bucket_size,
684 	&sysfs_block_size,
685 	&sysfs_tree_depth,
686 	&sysfs_root_usage_percent,
687 	&sysfs_btree_cache_size,
688 	&sysfs_cache_available_percent,
689 
690 	&sysfs_average_key_size,
691 
692 	&sysfs_errors,
693 	&sysfs_io_error_limit,
694 	&sysfs_io_error_halflife,
695 	&sysfs_congested,
696 	&sysfs_congested_read_threshold_us,
697 	&sysfs_congested_write_threshold_us,
698 	&sysfs_clear_stats,
699 	NULL
700 };
701 KTYPE(bch_cache_set);
702 
703 static struct attribute *bch_cache_set_internal_files[] = {
704 	&sysfs_active_journal_entries,
705 
706 	sysfs_time_stats_attribute_list(btree_gc, sec, ms)
707 	sysfs_time_stats_attribute_list(btree_split, sec, us)
708 	sysfs_time_stats_attribute_list(btree_sort, ms, us)
709 	sysfs_time_stats_attribute_list(btree_read, ms, us)
710 
711 	&sysfs_btree_nodes,
712 	&sysfs_btree_used_percent,
713 	&sysfs_btree_cache_max_chain,
714 
715 	&sysfs_bset_tree_stats,
716 	&sysfs_cache_read_races,
717 	&sysfs_writeback_keys_done,
718 	&sysfs_writeback_keys_failed,
719 
720 	&sysfs_trigger_gc,
721 	&sysfs_prune_cache,
722 #ifdef CONFIG_BCACHE_DEBUG
723 	&sysfs_verify,
724 	&sysfs_key_merging_disabled,
725 	&sysfs_expensive_debug_checks,
726 #endif
727 	&sysfs_gc_always_rewrite,
728 	&sysfs_btree_shrinker_disabled,
729 	&sysfs_copy_gc_enabled,
730 	NULL
731 };
732 KTYPE(bch_cache_set_internal);
733 
734 SHOW(__bch_cache)
735 {
736 	struct cache *ca = container_of(kobj, struct cache, kobj);
737 
738 	sysfs_hprint(bucket_size,	bucket_bytes(ca));
739 	sysfs_hprint(block_size,	block_bytes(ca));
740 	sysfs_print(nbuckets,		ca->sb.nbuckets);
741 	sysfs_print(discard,		ca->discard);
742 	sysfs_hprint(written, atomic_long_read(&ca->sectors_written) << 9);
743 	sysfs_hprint(btree_written,
744 		     atomic_long_read(&ca->btree_sectors_written) << 9);
745 	sysfs_hprint(metadata_written,
746 		     (atomic_long_read(&ca->meta_sectors_written) +
747 		      atomic_long_read(&ca->btree_sectors_written)) << 9);
748 
749 	sysfs_print(io_errors,
750 		    atomic_read(&ca->io_errors) >> IO_ERROR_SHIFT);
751 
752 	if (attr == &sysfs_cache_replacement_policy)
753 		return bch_snprint_string_list(buf, PAGE_SIZE,
754 					       cache_replacement_policies,
755 					       CACHE_REPLACEMENT(&ca->sb));
756 
757 	if (attr == &sysfs_priority_stats) {
758 		int cmp(const void *l, const void *r)
759 		{	return *((uint16_t *) r) - *((uint16_t *) l); }
760 
761 		struct bucket *b;
762 		size_t n = ca->sb.nbuckets, i;
763 		size_t unused = 0, available = 0, dirty = 0, meta = 0;
764 		uint64_t sum = 0;
765 		/* Compute 31 quantiles */
766 		uint16_t q[31], *p, *cached;
767 		ssize_t ret;
768 
769 		cached = p = vmalloc(ca->sb.nbuckets * sizeof(uint16_t));
770 		if (!p)
771 			return -ENOMEM;
772 
773 		mutex_lock(&ca->set->bucket_lock);
774 		for_each_bucket(b, ca) {
775 			if (!GC_SECTORS_USED(b))
776 				unused++;
777 			if (GC_MARK(b) == GC_MARK_RECLAIMABLE)
778 				available++;
779 			if (GC_MARK(b) == GC_MARK_DIRTY)
780 				dirty++;
781 			if (GC_MARK(b) == GC_MARK_METADATA)
782 				meta++;
783 		}
784 
785 		for (i = ca->sb.first_bucket; i < n; i++)
786 			p[i] = ca->buckets[i].prio;
787 		mutex_unlock(&ca->set->bucket_lock);
788 
789 		sort(p, n, sizeof(uint16_t), cmp, NULL);
790 
791 		while (n &&
792 		       !cached[n - 1])
793 			--n;
794 
795 		unused = ca->sb.nbuckets - n;
796 
797 		while (cached < p + n &&
798 		       *cached == BTREE_PRIO)
799 			cached++, n--;
800 
801 		for (i = 0; i < n; i++)
802 			sum += INITIAL_PRIO - cached[i];
803 
804 		if (n)
805 			do_div(sum, n);
806 
807 		for (i = 0; i < ARRAY_SIZE(q); i++)
808 			q[i] = INITIAL_PRIO - cached[n * (i + 1) /
809 				(ARRAY_SIZE(q) + 1)];
810 
811 		vfree(p);
812 
813 		ret = scnprintf(buf, PAGE_SIZE,
814 				"Unused:		%zu%%\n"
815 				"Clean:		%zu%%\n"
816 				"Dirty:		%zu%%\n"
817 				"Metadata:	%zu%%\n"
818 				"Average:	%llu\n"
819 				"Sectors per Q:	%zu\n"
820 				"Quantiles:	[",
821 				unused * 100 / (size_t) ca->sb.nbuckets,
822 				available * 100 / (size_t) ca->sb.nbuckets,
823 				dirty * 100 / (size_t) ca->sb.nbuckets,
824 				meta * 100 / (size_t) ca->sb.nbuckets, sum,
825 				n * ca->sb.bucket_size / (ARRAY_SIZE(q) + 1));
826 
827 		for (i = 0; i < ARRAY_SIZE(q); i++)
828 			ret += scnprintf(buf + ret, PAGE_SIZE - ret,
829 					 "%u ", q[i]);
830 		ret--;
831 
832 		ret += scnprintf(buf + ret, PAGE_SIZE - ret, "]\n");
833 
834 		return ret;
835 	}
836 
837 	return 0;
838 }
839 SHOW_LOCKED(bch_cache)
840 
841 STORE(__bch_cache)
842 {
843 	struct cache *ca = container_of(kobj, struct cache, kobj);
844 
845 	if (attr == &sysfs_discard) {
846 		bool v = strtoul_or_return(buf);
847 
848 		if (blk_queue_discard(bdev_get_queue(ca->bdev)))
849 			ca->discard = v;
850 
851 		if (v != CACHE_DISCARD(&ca->sb)) {
852 			SET_CACHE_DISCARD(&ca->sb, v);
853 			bcache_write_super(ca->set);
854 		}
855 	}
856 
857 	if (attr == &sysfs_cache_replacement_policy) {
858 		ssize_t v = bch_read_string_list(buf, cache_replacement_policies);
859 
860 		if (v < 0)
861 			return v;
862 
863 		if ((unsigned) v != CACHE_REPLACEMENT(&ca->sb)) {
864 			mutex_lock(&ca->set->bucket_lock);
865 			SET_CACHE_REPLACEMENT(&ca->sb, v);
866 			mutex_unlock(&ca->set->bucket_lock);
867 
868 			bcache_write_super(ca->set);
869 		}
870 	}
871 
872 	if (attr == &sysfs_clear_stats) {
873 		atomic_long_set(&ca->sectors_written, 0);
874 		atomic_long_set(&ca->btree_sectors_written, 0);
875 		atomic_long_set(&ca->meta_sectors_written, 0);
876 		atomic_set(&ca->io_count, 0);
877 		atomic_set(&ca->io_errors, 0);
878 	}
879 
880 	return size;
881 }
882 STORE_LOCKED(bch_cache)
883 
884 static struct attribute *bch_cache_files[] = {
885 	&sysfs_bucket_size,
886 	&sysfs_block_size,
887 	&sysfs_nbuckets,
888 	&sysfs_priority_stats,
889 	&sysfs_discard,
890 	&sysfs_written,
891 	&sysfs_btree_written,
892 	&sysfs_metadata_written,
893 	&sysfs_io_errors,
894 	&sysfs_clear_stats,
895 	&sysfs_cache_replacement_policy,
896 	NULL
897 };
898 KTYPE(bch_cache);
899