xref: /openbmc/linux/drivers/md/bcache/sysfs.c (revision 4cff79e9)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * bcache sysfs interfaces
4  *
5  * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
6  * Copyright 2012 Google, Inc.
7  */
8 
9 #include "bcache.h"
10 #include "sysfs.h"
11 #include "btree.h"
12 #include "request.h"
13 #include "writeback.h"
14 
15 #include <linux/blkdev.h>
16 #include <linux/sort.h>
17 #include <linux/sched/clock.h>
18 
19 static const char * const cache_replacement_policies[] = {
20 	"lru",
21 	"fifo",
22 	"random",
23 	NULL
24 };
25 
26 static const char * const error_actions[] = {
27 	"unregister",
28 	"panic",
29 	NULL
30 };
31 
32 write_attribute(attach);
33 write_attribute(detach);
34 write_attribute(unregister);
35 write_attribute(stop);
36 write_attribute(clear_stats);
37 write_attribute(trigger_gc);
38 write_attribute(prune_cache);
39 write_attribute(flash_vol_create);
40 
41 read_attribute(bucket_size);
42 read_attribute(block_size);
43 read_attribute(nbuckets);
44 read_attribute(tree_depth);
45 read_attribute(root_usage_percent);
46 read_attribute(priority_stats);
47 read_attribute(btree_cache_size);
48 read_attribute(btree_cache_max_chain);
49 read_attribute(cache_available_percent);
50 read_attribute(written);
51 read_attribute(btree_written);
52 read_attribute(metadata_written);
53 read_attribute(active_journal_entries);
54 
55 sysfs_time_stats_attribute(btree_gc,	sec, ms);
56 sysfs_time_stats_attribute(btree_split, sec, us);
57 sysfs_time_stats_attribute(btree_sort,	ms,  us);
58 sysfs_time_stats_attribute(btree_read,	ms,  us);
59 
60 read_attribute(btree_nodes);
61 read_attribute(btree_used_percent);
62 read_attribute(average_key_size);
63 read_attribute(dirty_data);
64 read_attribute(bset_tree_stats);
65 
66 read_attribute(state);
67 read_attribute(cache_read_races);
68 read_attribute(reclaim);
69 read_attribute(flush_write);
70 read_attribute(retry_flush_write);
71 read_attribute(writeback_keys_done);
72 read_attribute(writeback_keys_failed);
73 read_attribute(io_errors);
74 read_attribute(congested);
75 rw_attribute(congested_read_threshold_us);
76 rw_attribute(congested_write_threshold_us);
77 
78 rw_attribute(sequential_cutoff);
79 rw_attribute(data_csum);
80 rw_attribute(cache_mode);
81 rw_attribute(stop_when_cache_set_failed);
82 rw_attribute(writeback_metadata);
83 rw_attribute(writeback_running);
84 rw_attribute(writeback_percent);
85 rw_attribute(writeback_delay);
86 rw_attribute(writeback_rate);
87 
88 rw_attribute(writeback_rate_update_seconds);
89 rw_attribute(writeback_rate_i_term_inverse);
90 rw_attribute(writeback_rate_p_term_inverse);
91 rw_attribute(writeback_rate_minimum);
92 read_attribute(writeback_rate_debug);
93 
94 read_attribute(stripe_size);
95 read_attribute(partial_stripes_expensive);
96 
97 rw_attribute(synchronous);
98 rw_attribute(journal_delay_ms);
99 rw_attribute(io_disable);
100 rw_attribute(discard);
101 rw_attribute(running);
102 rw_attribute(label);
103 rw_attribute(readahead);
104 rw_attribute(errors);
105 rw_attribute(io_error_limit);
106 rw_attribute(io_error_halflife);
107 rw_attribute(verify);
108 rw_attribute(bypass_torture_test);
109 rw_attribute(key_merging_disabled);
110 rw_attribute(gc_always_rewrite);
111 rw_attribute(expensive_debug_checks);
112 rw_attribute(cache_replacement_policy);
113 rw_attribute(btree_shrinker_disabled);
114 rw_attribute(copy_gc_enabled);
115 rw_attribute(size);
116 
117 SHOW(__bch_cached_dev)
118 {
119 	struct cached_dev *dc = container_of(kobj, struct cached_dev,
120 					     disk.kobj);
121 	const char *states[] = { "no cache", "clean", "dirty", "inconsistent" };
122 
123 #define var(stat)		(dc->stat)
124 
125 	if (attr == &sysfs_cache_mode)
126 		return bch_snprint_string_list(buf, PAGE_SIZE,
127 					       bch_cache_modes + 1,
128 					       BDEV_CACHE_MODE(&dc->sb));
129 
130 	if (attr == &sysfs_stop_when_cache_set_failed)
131 		return bch_snprint_string_list(buf, PAGE_SIZE,
132 					       bch_stop_on_failure_modes + 1,
133 					       dc->stop_when_cache_set_failed);
134 
135 
136 	sysfs_printf(data_csum,		"%i", dc->disk.data_csum);
137 	var_printf(verify,		"%i");
138 	var_printf(bypass_torture_test,	"%i");
139 	var_printf(writeback_metadata,	"%i");
140 	var_printf(writeback_running,	"%i");
141 	var_print(writeback_delay);
142 	var_print(writeback_percent);
143 	sysfs_hprint(writeback_rate,	dc->writeback_rate.rate << 9);
144 	sysfs_hprint(io_errors,		atomic_read(&dc->io_errors));
145 	sysfs_printf(io_error_limit,	"%i", dc->error_limit);
146 	sysfs_printf(io_disable,	"%i", dc->io_disable);
147 	var_print(writeback_rate_update_seconds);
148 	var_print(writeback_rate_i_term_inverse);
149 	var_print(writeback_rate_p_term_inverse);
150 	var_print(writeback_rate_minimum);
151 
152 	if (attr == &sysfs_writeback_rate_debug) {
153 		char rate[20];
154 		char dirty[20];
155 		char target[20];
156 		char proportional[20];
157 		char integral[20];
158 		char change[20];
159 		s64 next_io;
160 
161 		bch_hprint(rate,	dc->writeback_rate.rate << 9);
162 		bch_hprint(dirty,	bcache_dev_sectors_dirty(&dc->disk) << 9);
163 		bch_hprint(target,	dc->writeback_rate_target << 9);
164 		bch_hprint(proportional,dc->writeback_rate_proportional << 9);
165 		bch_hprint(integral,	dc->writeback_rate_integral_scaled << 9);
166 		bch_hprint(change,	dc->writeback_rate_change << 9);
167 
168 		next_io = div64_s64(dc->writeback_rate.next - local_clock(),
169 				    NSEC_PER_MSEC);
170 
171 		return sprintf(buf,
172 			       "rate:\t\t%s/sec\n"
173 			       "dirty:\t\t%s\n"
174 			       "target:\t\t%s\n"
175 			       "proportional:\t%s\n"
176 			       "integral:\t%s\n"
177 			       "change:\t\t%s/sec\n"
178 			       "next io:\t%llims\n",
179 			       rate, dirty, target, proportional,
180 			       integral, change, next_io);
181 	}
182 
183 	sysfs_hprint(dirty_data,
184 		     bcache_dev_sectors_dirty(&dc->disk) << 9);
185 
186 	sysfs_hprint(stripe_size,	 ((uint64_t)dc->disk.stripe_size) << 9);
187 	var_printf(partial_stripes_expensive,	"%u");
188 
189 	var_hprint(sequential_cutoff);
190 	var_hprint(readahead);
191 
192 	sysfs_print(running,		atomic_read(&dc->running));
193 	sysfs_print(state,		states[BDEV_STATE(&dc->sb)]);
194 
195 	if (attr == &sysfs_label) {
196 		memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
197 		buf[SB_LABEL_SIZE + 1] = '\0';
198 		strcat(buf, "\n");
199 		return strlen(buf);
200 	}
201 
202 #undef var
203 	return 0;
204 }
205 SHOW_LOCKED(bch_cached_dev)
206 
207 STORE(__cached_dev)
208 {
209 	struct cached_dev *dc = container_of(kobj, struct cached_dev,
210 					     disk.kobj);
211 	ssize_t v;
212 	struct cache_set *c;
213 	struct kobj_uevent_env *env;
214 
215 #define d_strtoul(var)		sysfs_strtoul(var, dc->var)
216 #define d_strtoul_nonzero(var)	sysfs_strtoul_clamp(var, dc->var, 1, INT_MAX)
217 #define d_strtoi_h(var)		sysfs_hatoi(var, dc->var)
218 
219 	sysfs_strtoul(data_csum,	dc->disk.data_csum);
220 	d_strtoul(verify);
221 	d_strtoul(bypass_torture_test);
222 	d_strtoul(writeback_metadata);
223 	d_strtoul(writeback_running);
224 	d_strtoul(writeback_delay);
225 
226 	sysfs_strtoul_clamp(writeback_percent, dc->writeback_percent, 0, 40);
227 
228 	sysfs_strtoul_clamp(writeback_rate,
229 			    dc->writeback_rate.rate, 1, INT_MAX);
230 
231 	sysfs_strtoul_clamp(writeback_rate_update_seconds,
232 			    dc->writeback_rate_update_seconds,
233 			    1, WRITEBACK_RATE_UPDATE_SECS_MAX);
234 	d_strtoul(writeback_rate_i_term_inverse);
235 	d_strtoul_nonzero(writeback_rate_p_term_inverse);
236 
237 	sysfs_strtoul_clamp(io_error_limit, dc->error_limit, 0, INT_MAX);
238 
239 	if (attr == &sysfs_io_disable) {
240 		int v = strtoul_or_return(buf);
241 
242 		dc->io_disable = v ? 1 : 0;
243 	}
244 
245 	d_strtoi_h(sequential_cutoff);
246 	d_strtoi_h(readahead);
247 
248 	if (attr == &sysfs_clear_stats)
249 		bch_cache_accounting_clear(&dc->accounting);
250 
251 	if (attr == &sysfs_running &&
252 	    strtoul_or_return(buf))
253 		bch_cached_dev_run(dc);
254 
255 	if (attr == &sysfs_cache_mode) {
256 		v = bch_read_string_list(buf, bch_cache_modes + 1);
257 
258 		if (v < 0)
259 			return v;
260 
261 		if ((unsigned) v != BDEV_CACHE_MODE(&dc->sb)) {
262 			SET_BDEV_CACHE_MODE(&dc->sb, v);
263 			bch_write_bdev_super(dc, NULL);
264 		}
265 	}
266 
267 	if (attr == &sysfs_stop_when_cache_set_failed) {
268 		v = bch_read_string_list(buf, bch_stop_on_failure_modes + 1);
269 
270 		if (v < 0)
271 			return v;
272 
273 		dc->stop_when_cache_set_failed = v;
274 	}
275 
276 	if (attr == &sysfs_label) {
277 		if (size > SB_LABEL_SIZE)
278 			return -EINVAL;
279 		memcpy(dc->sb.label, buf, size);
280 		if (size < SB_LABEL_SIZE)
281 			dc->sb.label[size] = '\0';
282 		if (size && dc->sb.label[size - 1] == '\n')
283 			dc->sb.label[size - 1] = '\0';
284 		bch_write_bdev_super(dc, NULL);
285 		if (dc->disk.c) {
286 			memcpy(dc->disk.c->uuids[dc->disk.id].label,
287 			       buf, SB_LABEL_SIZE);
288 			bch_uuid_write(dc->disk.c);
289 		}
290 		env = kzalloc(sizeof(struct kobj_uevent_env), GFP_KERNEL);
291 		if (!env)
292 			return -ENOMEM;
293 		add_uevent_var(env, "DRIVER=bcache");
294 		add_uevent_var(env, "CACHED_UUID=%pU", dc->sb.uuid),
295 		add_uevent_var(env, "CACHED_LABEL=%s", buf);
296 		kobject_uevent_env(
297 			&disk_to_dev(dc->disk.disk)->kobj, KOBJ_CHANGE, env->envp);
298 		kfree(env);
299 	}
300 
301 	if (attr == &sysfs_attach) {
302 		uint8_t		set_uuid[16];
303 
304 		if (bch_parse_uuid(buf, set_uuid) < 16)
305 			return -EINVAL;
306 
307 		v = -ENOENT;
308 		list_for_each_entry(c, &bch_cache_sets, list) {
309 			v = bch_cached_dev_attach(dc, c, set_uuid);
310 			if (!v)
311 				return size;
312 		}
313 
314 		pr_err("Can't attach %s: cache set not found", buf);
315 		return v;
316 	}
317 
318 	if (attr == &sysfs_detach && dc->disk.c)
319 		bch_cached_dev_detach(dc);
320 
321 	if (attr == &sysfs_stop)
322 		bcache_device_stop(&dc->disk);
323 
324 	return size;
325 }
326 
327 STORE(bch_cached_dev)
328 {
329 	struct cached_dev *dc = container_of(kobj, struct cached_dev,
330 					     disk.kobj);
331 
332 	mutex_lock(&bch_register_lock);
333 	size = __cached_dev_store(kobj, attr, buf, size);
334 
335 	if (attr == &sysfs_writeback_running)
336 		bch_writeback_queue(dc);
337 
338 	if (attr == &sysfs_writeback_percent)
339 		if (!test_and_set_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags))
340 			schedule_delayed_work(&dc->writeback_rate_update,
341 				      dc->writeback_rate_update_seconds * HZ);
342 
343 	mutex_unlock(&bch_register_lock);
344 	return size;
345 }
346 
347 static struct attribute *bch_cached_dev_files[] = {
348 	&sysfs_attach,
349 	&sysfs_detach,
350 	&sysfs_stop,
351 #if 0
352 	&sysfs_data_csum,
353 #endif
354 	&sysfs_cache_mode,
355 	&sysfs_stop_when_cache_set_failed,
356 	&sysfs_writeback_metadata,
357 	&sysfs_writeback_running,
358 	&sysfs_writeback_delay,
359 	&sysfs_writeback_percent,
360 	&sysfs_writeback_rate,
361 	&sysfs_writeback_rate_update_seconds,
362 	&sysfs_writeback_rate_i_term_inverse,
363 	&sysfs_writeback_rate_p_term_inverse,
364 	&sysfs_writeback_rate_debug,
365 	&sysfs_errors,
366 	&sysfs_io_error_limit,
367 	&sysfs_io_disable,
368 	&sysfs_dirty_data,
369 	&sysfs_stripe_size,
370 	&sysfs_partial_stripes_expensive,
371 	&sysfs_sequential_cutoff,
372 	&sysfs_clear_stats,
373 	&sysfs_running,
374 	&sysfs_state,
375 	&sysfs_label,
376 	&sysfs_readahead,
377 #ifdef CONFIG_BCACHE_DEBUG
378 	&sysfs_verify,
379 	&sysfs_bypass_torture_test,
380 #endif
381 	NULL
382 };
383 KTYPE(bch_cached_dev);
384 
385 SHOW(bch_flash_dev)
386 {
387 	struct bcache_device *d = container_of(kobj, struct bcache_device,
388 					       kobj);
389 	struct uuid_entry *u = &d->c->uuids[d->id];
390 
391 	sysfs_printf(data_csum,	"%i", d->data_csum);
392 	sysfs_hprint(size,	u->sectors << 9);
393 
394 	if (attr == &sysfs_label) {
395 		memcpy(buf, u->label, SB_LABEL_SIZE);
396 		buf[SB_LABEL_SIZE + 1] = '\0';
397 		strcat(buf, "\n");
398 		return strlen(buf);
399 	}
400 
401 	return 0;
402 }
403 
404 STORE(__bch_flash_dev)
405 {
406 	struct bcache_device *d = container_of(kobj, struct bcache_device,
407 					       kobj);
408 	struct uuid_entry *u = &d->c->uuids[d->id];
409 
410 	sysfs_strtoul(data_csum,	d->data_csum);
411 
412 	if (attr == &sysfs_size) {
413 		uint64_t v;
414 		strtoi_h_or_return(buf, v);
415 
416 		u->sectors = v >> 9;
417 		bch_uuid_write(d->c);
418 		set_capacity(d->disk, u->sectors);
419 	}
420 
421 	if (attr == &sysfs_label) {
422 		memcpy(u->label, buf, SB_LABEL_SIZE);
423 		bch_uuid_write(d->c);
424 	}
425 
426 	if (attr == &sysfs_unregister) {
427 		set_bit(BCACHE_DEV_DETACHING, &d->flags);
428 		bcache_device_stop(d);
429 	}
430 
431 	return size;
432 }
433 STORE_LOCKED(bch_flash_dev)
434 
435 static struct attribute *bch_flash_dev_files[] = {
436 	&sysfs_unregister,
437 #if 0
438 	&sysfs_data_csum,
439 #endif
440 	&sysfs_label,
441 	&sysfs_size,
442 	NULL
443 };
444 KTYPE(bch_flash_dev);
445 
446 struct bset_stats_op {
447 	struct btree_op op;
448 	size_t nodes;
449 	struct bset_stats stats;
450 };
451 
452 static int bch_btree_bset_stats(struct btree_op *b_op, struct btree *b)
453 {
454 	struct bset_stats_op *op = container_of(b_op, struct bset_stats_op, op);
455 
456 	op->nodes++;
457 	bch_btree_keys_stats(&b->keys, &op->stats);
458 
459 	return MAP_CONTINUE;
460 }
461 
462 static int bch_bset_print_stats(struct cache_set *c, char *buf)
463 {
464 	struct bset_stats_op op;
465 	int ret;
466 
467 	memset(&op, 0, sizeof(op));
468 	bch_btree_op_init(&op.op, -1);
469 
470 	ret = bch_btree_map_nodes(&op.op, c, &ZERO_KEY, bch_btree_bset_stats);
471 	if (ret < 0)
472 		return ret;
473 
474 	return snprintf(buf, PAGE_SIZE,
475 			"btree nodes:		%zu\n"
476 			"written sets:		%zu\n"
477 			"unwritten sets:		%zu\n"
478 			"written key bytes:	%zu\n"
479 			"unwritten key bytes:	%zu\n"
480 			"floats:			%zu\n"
481 			"failed:			%zu\n",
482 			op.nodes,
483 			op.stats.sets_written, op.stats.sets_unwritten,
484 			op.stats.bytes_written, op.stats.bytes_unwritten,
485 			op.stats.floats, op.stats.failed);
486 }
487 
488 static unsigned bch_root_usage(struct cache_set *c)
489 {
490 	unsigned bytes = 0;
491 	struct bkey *k;
492 	struct btree *b;
493 	struct btree_iter iter;
494 
495 	goto lock_root;
496 
497 	do {
498 		rw_unlock(false, b);
499 lock_root:
500 		b = c->root;
501 		rw_lock(false, b, b->level);
502 	} while (b != c->root);
503 
504 	for_each_key_filter(&b->keys, k, &iter, bch_ptr_bad)
505 		bytes += bkey_bytes(k);
506 
507 	rw_unlock(false, b);
508 
509 	return (bytes * 100) / btree_bytes(c);
510 }
511 
512 static size_t bch_cache_size(struct cache_set *c)
513 {
514 	size_t ret = 0;
515 	struct btree *b;
516 
517 	mutex_lock(&c->bucket_lock);
518 	list_for_each_entry(b, &c->btree_cache, list)
519 		ret += 1 << (b->keys.page_order + PAGE_SHIFT);
520 
521 	mutex_unlock(&c->bucket_lock);
522 	return ret;
523 }
524 
525 static unsigned bch_cache_max_chain(struct cache_set *c)
526 {
527 	unsigned ret = 0;
528 	struct hlist_head *h;
529 
530 	mutex_lock(&c->bucket_lock);
531 
532 	for (h = c->bucket_hash;
533 	     h < c->bucket_hash + (1 << BUCKET_HASH_BITS);
534 	     h++) {
535 		unsigned i = 0;
536 		struct hlist_node *p;
537 
538 		hlist_for_each(p, h)
539 			i++;
540 
541 		ret = max(ret, i);
542 	}
543 
544 	mutex_unlock(&c->bucket_lock);
545 	return ret;
546 }
547 
548 static unsigned bch_btree_used(struct cache_set *c)
549 {
550 	return div64_u64(c->gc_stats.key_bytes * 100,
551 			 (c->gc_stats.nodes ?: 1) * btree_bytes(c));
552 }
553 
554 static unsigned bch_average_key_size(struct cache_set *c)
555 {
556 	return c->gc_stats.nkeys
557 		? div64_u64(c->gc_stats.data, c->gc_stats.nkeys)
558 		: 0;
559 }
560 
561 SHOW(__bch_cache_set)
562 {
563 	struct cache_set *c = container_of(kobj, struct cache_set, kobj);
564 
565 	sysfs_print(synchronous,		CACHE_SYNC(&c->sb));
566 	sysfs_print(journal_delay_ms,		c->journal_delay_ms);
567 	sysfs_hprint(bucket_size,		bucket_bytes(c));
568 	sysfs_hprint(block_size,		block_bytes(c));
569 	sysfs_print(tree_depth,			c->root->level);
570 	sysfs_print(root_usage_percent,		bch_root_usage(c));
571 
572 	sysfs_hprint(btree_cache_size,		bch_cache_size(c));
573 	sysfs_print(btree_cache_max_chain,	bch_cache_max_chain(c));
574 	sysfs_print(cache_available_percent,	100 - c->gc_stats.in_use);
575 
576 	sysfs_print_time_stats(&c->btree_gc_time,	btree_gc, sec, ms);
577 	sysfs_print_time_stats(&c->btree_split_time,	btree_split, sec, us);
578 	sysfs_print_time_stats(&c->sort.time,		btree_sort, ms, us);
579 	sysfs_print_time_stats(&c->btree_read_time,	btree_read, ms, us);
580 
581 	sysfs_print(btree_used_percent,	bch_btree_used(c));
582 	sysfs_print(btree_nodes,	c->gc_stats.nodes);
583 	sysfs_hprint(average_key_size,	bch_average_key_size(c));
584 
585 	sysfs_print(cache_read_races,
586 		    atomic_long_read(&c->cache_read_races));
587 
588 	sysfs_print(reclaim,
589 		    atomic_long_read(&c->reclaim));
590 
591 	sysfs_print(flush_write,
592 		    atomic_long_read(&c->flush_write));
593 
594 	sysfs_print(retry_flush_write,
595 		    atomic_long_read(&c->retry_flush_write));
596 
597 	sysfs_print(writeback_keys_done,
598 		    atomic_long_read(&c->writeback_keys_done));
599 	sysfs_print(writeback_keys_failed,
600 		    atomic_long_read(&c->writeback_keys_failed));
601 
602 	if (attr == &sysfs_errors)
603 		return bch_snprint_string_list(buf, PAGE_SIZE, error_actions,
604 					       c->on_error);
605 
606 	/* See count_io_errors for why 88 */
607 	sysfs_print(io_error_halflife,	c->error_decay * 88);
608 	sysfs_print(io_error_limit,	c->error_limit);
609 
610 	sysfs_hprint(congested,
611 		     ((uint64_t) bch_get_congested(c)) << 9);
612 	sysfs_print(congested_read_threshold_us,
613 		    c->congested_read_threshold_us);
614 	sysfs_print(congested_write_threshold_us,
615 		    c->congested_write_threshold_us);
616 
617 	sysfs_print(active_journal_entries,	fifo_used(&c->journal.pin));
618 	sysfs_printf(verify,			"%i", c->verify);
619 	sysfs_printf(key_merging_disabled,	"%i", c->key_merging_disabled);
620 	sysfs_printf(expensive_debug_checks,
621 		     "%i", c->expensive_debug_checks);
622 	sysfs_printf(gc_always_rewrite,		"%i", c->gc_always_rewrite);
623 	sysfs_printf(btree_shrinker_disabled,	"%i", c->shrinker_disabled);
624 	sysfs_printf(copy_gc_enabled,		"%i", c->copy_gc_enabled);
625 	sysfs_printf(io_disable,		"%i",
626 		     test_bit(CACHE_SET_IO_DISABLE, &c->flags));
627 
628 	if (attr == &sysfs_bset_tree_stats)
629 		return bch_bset_print_stats(c, buf);
630 
631 	return 0;
632 }
633 SHOW_LOCKED(bch_cache_set)
634 
635 STORE(__bch_cache_set)
636 {
637 	struct cache_set *c = container_of(kobj, struct cache_set, kobj);
638 
639 	if (attr == &sysfs_unregister)
640 		bch_cache_set_unregister(c);
641 
642 	if (attr == &sysfs_stop)
643 		bch_cache_set_stop(c);
644 
645 	if (attr == &sysfs_synchronous) {
646 		bool sync = strtoul_or_return(buf);
647 
648 		if (sync != CACHE_SYNC(&c->sb)) {
649 			SET_CACHE_SYNC(&c->sb, sync);
650 			bcache_write_super(c);
651 		}
652 	}
653 
654 	if (attr == &sysfs_flash_vol_create) {
655 		int r;
656 		uint64_t v;
657 		strtoi_h_or_return(buf, v);
658 
659 		r = bch_flash_dev_create(c, v);
660 		if (r)
661 			return r;
662 	}
663 
664 	if (attr == &sysfs_clear_stats) {
665 		atomic_long_set(&c->writeback_keys_done,	0);
666 		atomic_long_set(&c->writeback_keys_failed,	0);
667 
668 		memset(&c->gc_stats, 0, sizeof(struct gc_stat));
669 		bch_cache_accounting_clear(&c->accounting);
670 	}
671 
672 	if (attr == &sysfs_trigger_gc) {
673 		/*
674 		 * Garbage collection thread only works when sectors_to_gc < 0,
675 		 * when users write to sysfs entry trigger_gc, most of time
676 		 * they want to forcibly triger gargage collection. Here -1 is
677 		 * set to c->sectors_to_gc, to make gc_should_run() give a
678 		 * chance to permit gc thread to run. "give a chance" means
679 		 * before going into gc_should_run(), there is still chance
680 		 * that c->sectors_to_gc being set to other positive value. So
681 		 * writing sysfs entry trigger_gc won't always make sure gc
682 		 * thread takes effect.
683 		 */
684 		atomic_set(&c->sectors_to_gc, -1);
685 		wake_up_gc(c);
686 	}
687 
688 	if (attr == &sysfs_prune_cache) {
689 		struct shrink_control sc;
690 		sc.gfp_mask = GFP_KERNEL;
691 		sc.nr_to_scan = strtoul_or_return(buf);
692 		c->shrink.scan_objects(&c->shrink, &sc);
693 	}
694 
695 	sysfs_strtoul(congested_read_threshold_us,
696 		      c->congested_read_threshold_us);
697 	sysfs_strtoul(congested_write_threshold_us,
698 		      c->congested_write_threshold_us);
699 
700 	if (attr == &sysfs_errors) {
701 		ssize_t v = bch_read_string_list(buf, error_actions);
702 
703 		if (v < 0)
704 			return v;
705 
706 		c->on_error = v;
707 	}
708 
709 	if (attr == &sysfs_io_error_limit)
710 		c->error_limit = strtoul_or_return(buf);
711 
712 	/* See count_io_errors() for why 88 */
713 	if (attr == &sysfs_io_error_halflife)
714 		c->error_decay = strtoul_or_return(buf) / 88;
715 
716 	if (attr == &sysfs_io_disable) {
717 		int v = strtoul_or_return(buf);
718 
719 		if (v) {
720 			if (test_and_set_bit(CACHE_SET_IO_DISABLE,
721 					     &c->flags))
722 				pr_warn("CACHE_SET_IO_DISABLE already set");
723 		} else {
724 			if (!test_and_clear_bit(CACHE_SET_IO_DISABLE,
725 						&c->flags))
726 				pr_warn("CACHE_SET_IO_DISABLE already cleared");
727 		}
728 	}
729 
730 	sysfs_strtoul(journal_delay_ms,		c->journal_delay_ms);
731 	sysfs_strtoul(verify,			c->verify);
732 	sysfs_strtoul(key_merging_disabled,	c->key_merging_disabled);
733 	sysfs_strtoul(expensive_debug_checks,	c->expensive_debug_checks);
734 	sysfs_strtoul(gc_always_rewrite,	c->gc_always_rewrite);
735 	sysfs_strtoul(btree_shrinker_disabled,	c->shrinker_disabled);
736 	sysfs_strtoul(copy_gc_enabled,		c->copy_gc_enabled);
737 
738 	return size;
739 }
740 STORE_LOCKED(bch_cache_set)
741 
742 SHOW(bch_cache_set_internal)
743 {
744 	struct cache_set *c = container_of(kobj, struct cache_set, internal);
745 	return bch_cache_set_show(&c->kobj, attr, buf);
746 }
747 
748 STORE(bch_cache_set_internal)
749 {
750 	struct cache_set *c = container_of(kobj, struct cache_set, internal);
751 	return bch_cache_set_store(&c->kobj, attr, buf, size);
752 }
753 
754 static void bch_cache_set_internal_release(struct kobject *k)
755 {
756 }
757 
758 static struct attribute *bch_cache_set_files[] = {
759 	&sysfs_unregister,
760 	&sysfs_stop,
761 	&sysfs_synchronous,
762 	&sysfs_journal_delay_ms,
763 	&sysfs_flash_vol_create,
764 
765 	&sysfs_bucket_size,
766 	&sysfs_block_size,
767 	&sysfs_tree_depth,
768 	&sysfs_root_usage_percent,
769 	&sysfs_btree_cache_size,
770 	&sysfs_cache_available_percent,
771 
772 	&sysfs_average_key_size,
773 
774 	&sysfs_errors,
775 	&sysfs_io_error_limit,
776 	&sysfs_io_error_halflife,
777 	&sysfs_congested,
778 	&sysfs_congested_read_threshold_us,
779 	&sysfs_congested_write_threshold_us,
780 	&sysfs_clear_stats,
781 	NULL
782 };
783 KTYPE(bch_cache_set);
784 
785 static struct attribute *bch_cache_set_internal_files[] = {
786 	&sysfs_active_journal_entries,
787 
788 	sysfs_time_stats_attribute_list(btree_gc, sec, ms)
789 	sysfs_time_stats_attribute_list(btree_split, sec, us)
790 	sysfs_time_stats_attribute_list(btree_sort, ms, us)
791 	sysfs_time_stats_attribute_list(btree_read, ms, us)
792 
793 	&sysfs_btree_nodes,
794 	&sysfs_btree_used_percent,
795 	&sysfs_btree_cache_max_chain,
796 
797 	&sysfs_bset_tree_stats,
798 	&sysfs_cache_read_races,
799 	&sysfs_reclaim,
800 	&sysfs_flush_write,
801 	&sysfs_retry_flush_write,
802 	&sysfs_writeback_keys_done,
803 	&sysfs_writeback_keys_failed,
804 
805 	&sysfs_trigger_gc,
806 	&sysfs_prune_cache,
807 #ifdef CONFIG_BCACHE_DEBUG
808 	&sysfs_verify,
809 	&sysfs_key_merging_disabled,
810 	&sysfs_expensive_debug_checks,
811 #endif
812 	&sysfs_gc_always_rewrite,
813 	&sysfs_btree_shrinker_disabled,
814 	&sysfs_copy_gc_enabled,
815 	&sysfs_io_disable,
816 	NULL
817 };
818 KTYPE(bch_cache_set_internal);
819 
820 static int __bch_cache_cmp(const void *l, const void *r)
821 {
822 	return *((uint16_t *)r) - *((uint16_t *)l);
823 }
824 
825 SHOW(__bch_cache)
826 {
827 	struct cache *ca = container_of(kobj, struct cache, kobj);
828 
829 	sysfs_hprint(bucket_size,	bucket_bytes(ca));
830 	sysfs_hprint(block_size,	block_bytes(ca));
831 	sysfs_print(nbuckets,		ca->sb.nbuckets);
832 	sysfs_print(discard,		ca->discard);
833 	sysfs_hprint(written, atomic_long_read(&ca->sectors_written) << 9);
834 	sysfs_hprint(btree_written,
835 		     atomic_long_read(&ca->btree_sectors_written) << 9);
836 	sysfs_hprint(metadata_written,
837 		     (atomic_long_read(&ca->meta_sectors_written) +
838 		      atomic_long_read(&ca->btree_sectors_written)) << 9);
839 
840 	sysfs_print(io_errors,
841 		    atomic_read(&ca->io_errors) >> IO_ERROR_SHIFT);
842 
843 	if (attr == &sysfs_cache_replacement_policy)
844 		return bch_snprint_string_list(buf, PAGE_SIZE,
845 					       cache_replacement_policies,
846 					       CACHE_REPLACEMENT(&ca->sb));
847 
848 	if (attr == &sysfs_priority_stats) {
849 		struct bucket *b;
850 		size_t n = ca->sb.nbuckets, i;
851 		size_t unused = 0, available = 0, dirty = 0, meta = 0;
852 		uint64_t sum = 0;
853 		/* Compute 31 quantiles */
854 		uint16_t q[31], *p, *cached;
855 		ssize_t ret;
856 
857 		cached = p = vmalloc(ca->sb.nbuckets * sizeof(uint16_t));
858 		if (!p)
859 			return -ENOMEM;
860 
861 		mutex_lock(&ca->set->bucket_lock);
862 		for_each_bucket(b, ca) {
863 			if (!GC_SECTORS_USED(b))
864 				unused++;
865 			if (GC_MARK(b) == GC_MARK_RECLAIMABLE)
866 				available++;
867 			if (GC_MARK(b) == GC_MARK_DIRTY)
868 				dirty++;
869 			if (GC_MARK(b) == GC_MARK_METADATA)
870 				meta++;
871 		}
872 
873 		for (i = ca->sb.first_bucket; i < n; i++)
874 			p[i] = ca->buckets[i].prio;
875 		mutex_unlock(&ca->set->bucket_lock);
876 
877 		sort(p, n, sizeof(uint16_t), __bch_cache_cmp, NULL);
878 
879 		while (n &&
880 		       !cached[n - 1])
881 			--n;
882 
883 		unused = ca->sb.nbuckets - n;
884 
885 		while (cached < p + n &&
886 		       *cached == BTREE_PRIO)
887 			cached++, n--;
888 
889 		for (i = 0; i < n; i++)
890 			sum += INITIAL_PRIO - cached[i];
891 
892 		if (n)
893 			do_div(sum, n);
894 
895 		for (i = 0; i < ARRAY_SIZE(q); i++)
896 			q[i] = INITIAL_PRIO - cached[n * (i + 1) /
897 				(ARRAY_SIZE(q) + 1)];
898 
899 		vfree(p);
900 
901 		ret = scnprintf(buf, PAGE_SIZE,
902 				"Unused:		%zu%%\n"
903 				"Clean:		%zu%%\n"
904 				"Dirty:		%zu%%\n"
905 				"Metadata:	%zu%%\n"
906 				"Average:	%llu\n"
907 				"Sectors per Q:	%zu\n"
908 				"Quantiles:	[",
909 				unused * 100 / (size_t) ca->sb.nbuckets,
910 				available * 100 / (size_t) ca->sb.nbuckets,
911 				dirty * 100 / (size_t) ca->sb.nbuckets,
912 				meta * 100 / (size_t) ca->sb.nbuckets, sum,
913 				n * ca->sb.bucket_size / (ARRAY_SIZE(q) + 1));
914 
915 		for (i = 0; i < ARRAY_SIZE(q); i++)
916 			ret += scnprintf(buf + ret, PAGE_SIZE - ret,
917 					 "%u ", q[i]);
918 		ret--;
919 
920 		ret += scnprintf(buf + ret, PAGE_SIZE - ret, "]\n");
921 
922 		return ret;
923 	}
924 
925 	return 0;
926 }
927 SHOW_LOCKED(bch_cache)
928 
929 STORE(__bch_cache)
930 {
931 	struct cache *ca = container_of(kobj, struct cache, kobj);
932 
933 	if (attr == &sysfs_discard) {
934 		bool v = strtoul_or_return(buf);
935 
936 		if (blk_queue_discard(bdev_get_queue(ca->bdev)))
937 			ca->discard = v;
938 
939 		if (v != CACHE_DISCARD(&ca->sb)) {
940 			SET_CACHE_DISCARD(&ca->sb, v);
941 			bcache_write_super(ca->set);
942 		}
943 	}
944 
945 	if (attr == &sysfs_cache_replacement_policy) {
946 		ssize_t v = bch_read_string_list(buf, cache_replacement_policies);
947 
948 		if (v < 0)
949 			return v;
950 
951 		if ((unsigned) v != CACHE_REPLACEMENT(&ca->sb)) {
952 			mutex_lock(&ca->set->bucket_lock);
953 			SET_CACHE_REPLACEMENT(&ca->sb, v);
954 			mutex_unlock(&ca->set->bucket_lock);
955 
956 			bcache_write_super(ca->set);
957 		}
958 	}
959 
960 	if (attr == &sysfs_clear_stats) {
961 		atomic_long_set(&ca->sectors_written, 0);
962 		atomic_long_set(&ca->btree_sectors_written, 0);
963 		atomic_long_set(&ca->meta_sectors_written, 0);
964 		atomic_set(&ca->io_count, 0);
965 		atomic_set(&ca->io_errors, 0);
966 	}
967 
968 	return size;
969 }
970 STORE_LOCKED(bch_cache)
971 
972 static struct attribute *bch_cache_files[] = {
973 	&sysfs_bucket_size,
974 	&sysfs_block_size,
975 	&sysfs_nbuckets,
976 	&sysfs_priority_stats,
977 	&sysfs_discard,
978 	&sysfs_written,
979 	&sysfs_btree_written,
980 	&sysfs_metadata_written,
981 	&sysfs_io_errors,
982 	&sysfs_clear_stats,
983 	&sysfs_cache_replacement_policy,
984 	NULL
985 };
986 KTYPE(bch_cache);
987