xref: /openbmc/linux/drivers/md/bcache/sysfs.c (revision 04b3c795)
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 extern bool bcache_is_reboot;
20 
21 /* Default is 0 ("writethrough") */
22 static const char * const bch_cache_modes[] = {
23 	"writethrough",
24 	"writeback",
25 	"writearound",
26 	"none",
27 	NULL
28 };
29 
30 static const char * const bch_reada_cache_policies[] = {
31 	"all",
32 	"meta-only",
33 	NULL
34 };
35 
36 /* Default is 0 ("auto") */
37 static const char * const bch_stop_on_failure_modes[] = {
38 	"auto",
39 	"always",
40 	NULL
41 };
42 
43 static const char * const cache_replacement_policies[] = {
44 	"lru",
45 	"fifo",
46 	"random",
47 	NULL
48 };
49 
50 static const char * const error_actions[] = {
51 	"unregister",
52 	"panic",
53 	NULL
54 };
55 
56 write_attribute(attach);
57 write_attribute(detach);
58 write_attribute(unregister);
59 write_attribute(stop);
60 write_attribute(clear_stats);
61 write_attribute(trigger_gc);
62 write_attribute(prune_cache);
63 write_attribute(flash_vol_create);
64 
65 read_attribute(bucket_size);
66 read_attribute(block_size);
67 read_attribute(nbuckets);
68 read_attribute(tree_depth);
69 read_attribute(root_usage_percent);
70 read_attribute(priority_stats);
71 read_attribute(btree_cache_size);
72 read_attribute(btree_cache_max_chain);
73 read_attribute(cache_available_percent);
74 read_attribute(written);
75 read_attribute(btree_written);
76 read_attribute(metadata_written);
77 read_attribute(active_journal_entries);
78 read_attribute(backing_dev_name);
79 read_attribute(backing_dev_uuid);
80 
81 sysfs_time_stats_attribute(btree_gc,	sec, ms);
82 sysfs_time_stats_attribute(btree_split, sec, us);
83 sysfs_time_stats_attribute(btree_sort,	ms,  us);
84 sysfs_time_stats_attribute(btree_read,	ms,  us);
85 
86 read_attribute(btree_nodes);
87 read_attribute(btree_used_percent);
88 read_attribute(average_key_size);
89 read_attribute(dirty_data);
90 read_attribute(bset_tree_stats);
91 
92 read_attribute(state);
93 read_attribute(cache_read_races);
94 read_attribute(reclaim);
95 read_attribute(reclaimed_journal_buckets);
96 read_attribute(flush_write);
97 read_attribute(writeback_keys_done);
98 read_attribute(writeback_keys_failed);
99 read_attribute(io_errors);
100 read_attribute(congested);
101 read_attribute(cutoff_writeback);
102 read_attribute(cutoff_writeback_sync);
103 rw_attribute(congested_read_threshold_us);
104 rw_attribute(congested_write_threshold_us);
105 
106 rw_attribute(sequential_cutoff);
107 rw_attribute(data_csum);
108 rw_attribute(cache_mode);
109 rw_attribute(readahead_cache_policy);
110 rw_attribute(stop_when_cache_set_failed);
111 rw_attribute(writeback_metadata);
112 rw_attribute(writeback_running);
113 rw_attribute(writeback_percent);
114 rw_attribute(writeback_delay);
115 rw_attribute(writeback_rate);
116 
117 rw_attribute(writeback_rate_update_seconds);
118 rw_attribute(writeback_rate_i_term_inverse);
119 rw_attribute(writeback_rate_p_term_inverse);
120 rw_attribute(writeback_rate_minimum);
121 read_attribute(writeback_rate_debug);
122 
123 read_attribute(stripe_size);
124 read_attribute(partial_stripes_expensive);
125 
126 rw_attribute(synchronous);
127 rw_attribute(journal_delay_ms);
128 rw_attribute(io_disable);
129 rw_attribute(discard);
130 rw_attribute(running);
131 rw_attribute(label);
132 rw_attribute(readahead);
133 rw_attribute(errors);
134 rw_attribute(io_error_limit);
135 rw_attribute(io_error_halflife);
136 rw_attribute(verify);
137 rw_attribute(bypass_torture_test);
138 rw_attribute(key_merging_disabled);
139 rw_attribute(gc_always_rewrite);
140 rw_attribute(expensive_debug_checks);
141 rw_attribute(cache_replacement_policy);
142 rw_attribute(btree_shrinker_disabled);
143 rw_attribute(copy_gc_enabled);
144 rw_attribute(idle_max_writeback_rate);
145 rw_attribute(gc_after_writeback);
146 rw_attribute(size);
147 
148 static ssize_t bch_snprint_string_list(char *buf,
149 				       size_t size,
150 				       const char * const list[],
151 				       size_t selected)
152 {
153 	char *out = buf;
154 	size_t i;
155 
156 	for (i = 0; list[i]; i++)
157 		out += scnprintf(out, buf + size - out,
158 				i == selected ? "[%s] " : "%s ", list[i]);
159 
160 	out[-1] = '\n';
161 	return out - buf;
162 }
163 
164 SHOW(__bch_cached_dev)
165 {
166 	struct cached_dev *dc = container_of(kobj, struct cached_dev,
167 					     disk.kobj);
168 	char const *states[] = { "no cache", "clean", "dirty", "inconsistent" };
169 	int wb = dc->writeback_running;
170 
171 #define var(stat)		(dc->stat)
172 
173 	if (attr == &sysfs_cache_mode)
174 		return bch_snprint_string_list(buf, PAGE_SIZE,
175 					       bch_cache_modes,
176 					       BDEV_CACHE_MODE(&dc->sb));
177 
178 	if (attr == &sysfs_readahead_cache_policy)
179 		return bch_snprint_string_list(buf, PAGE_SIZE,
180 					      bch_reada_cache_policies,
181 					      dc->cache_readahead_policy);
182 
183 	if (attr == &sysfs_stop_when_cache_set_failed)
184 		return bch_snprint_string_list(buf, PAGE_SIZE,
185 					       bch_stop_on_failure_modes,
186 					       dc->stop_when_cache_set_failed);
187 
188 
189 	sysfs_printf(data_csum,		"%i", dc->disk.data_csum);
190 	var_printf(verify,		"%i");
191 	var_printf(bypass_torture_test,	"%i");
192 	var_printf(writeback_metadata,	"%i");
193 	var_printf(writeback_running,	"%i");
194 	var_print(writeback_delay);
195 	var_print(writeback_percent);
196 	sysfs_hprint(writeback_rate,
197 		     wb ? atomic_long_read(&dc->writeback_rate.rate) << 9 : 0);
198 	sysfs_printf(io_errors,		"%i", atomic_read(&dc->io_errors));
199 	sysfs_printf(io_error_limit,	"%i", dc->error_limit);
200 	sysfs_printf(io_disable,	"%i", dc->io_disable);
201 	var_print(writeback_rate_update_seconds);
202 	var_print(writeback_rate_i_term_inverse);
203 	var_print(writeback_rate_p_term_inverse);
204 	var_print(writeback_rate_minimum);
205 
206 	if (attr == &sysfs_writeback_rate_debug) {
207 		char rate[20];
208 		char dirty[20];
209 		char target[20];
210 		char proportional[20];
211 		char integral[20];
212 		char change[20];
213 		s64 next_io;
214 
215 		/*
216 		 * Except for dirty and target, other values should
217 		 * be 0 if writeback is not running.
218 		 */
219 		bch_hprint(rate,
220 			   wb ? atomic_long_read(&dc->writeback_rate.rate) << 9
221 			      : 0);
222 		bch_hprint(dirty, bcache_dev_sectors_dirty(&dc->disk) << 9);
223 		bch_hprint(target, dc->writeback_rate_target << 9);
224 		bch_hprint(proportional,
225 			   wb ? dc->writeback_rate_proportional << 9 : 0);
226 		bch_hprint(integral,
227 			   wb ? dc->writeback_rate_integral_scaled << 9 : 0);
228 		bch_hprint(change, wb ? dc->writeback_rate_change << 9 : 0);
229 		next_io = wb ? div64_s64(dc->writeback_rate.next-local_clock(),
230 					 NSEC_PER_MSEC) : 0;
231 
232 		return sprintf(buf,
233 			       "rate:\t\t%s/sec\n"
234 			       "dirty:\t\t%s\n"
235 			       "target:\t\t%s\n"
236 			       "proportional:\t%s\n"
237 			       "integral:\t%s\n"
238 			       "change:\t\t%s/sec\n"
239 			       "next io:\t%llims\n",
240 			       rate, dirty, target, proportional,
241 			       integral, change, next_io);
242 	}
243 
244 	sysfs_hprint(dirty_data,
245 		     bcache_dev_sectors_dirty(&dc->disk) << 9);
246 
247 	sysfs_hprint(stripe_size,	 ((uint64_t)dc->disk.stripe_size) << 9);
248 	var_printf(partial_stripes_expensive,	"%u");
249 
250 	var_hprint(sequential_cutoff);
251 	var_hprint(readahead);
252 
253 	sysfs_print(running,		atomic_read(&dc->running));
254 	sysfs_print(state,		states[BDEV_STATE(&dc->sb)]);
255 
256 	if (attr == &sysfs_label) {
257 		memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
258 		buf[SB_LABEL_SIZE + 1] = '\0';
259 		strcat(buf, "\n");
260 		return strlen(buf);
261 	}
262 
263 	if (attr == &sysfs_backing_dev_name) {
264 		snprintf(buf, BDEVNAME_SIZE + 1, "%s", dc->backing_dev_name);
265 		strcat(buf, "\n");
266 		return strlen(buf);
267 	}
268 
269 	if (attr == &sysfs_backing_dev_uuid) {
270 		/* convert binary uuid into 36-byte string plus '\0' */
271 		snprintf(buf, 36+1, "%pU", dc->sb.uuid);
272 		strcat(buf, "\n");
273 		return strlen(buf);
274 	}
275 
276 #undef var
277 	return 0;
278 }
279 SHOW_LOCKED(bch_cached_dev)
280 
281 STORE(__cached_dev)
282 {
283 	struct cached_dev *dc = container_of(kobj, struct cached_dev,
284 					     disk.kobj);
285 	ssize_t v;
286 	struct cache_set *c;
287 	struct kobj_uevent_env *env;
288 
289 	/* no user space access if system is rebooting */
290 	if (bcache_is_reboot)
291 		return -EBUSY;
292 
293 #define d_strtoul(var)		sysfs_strtoul(var, dc->var)
294 #define d_strtoul_nonzero(var)	sysfs_strtoul_clamp(var, dc->var, 1, INT_MAX)
295 #define d_strtoi_h(var)		sysfs_hatoi(var, dc->var)
296 
297 	sysfs_strtoul(data_csum,	dc->disk.data_csum);
298 	d_strtoul(verify);
299 	sysfs_strtoul_bool(bypass_torture_test, dc->bypass_torture_test);
300 	sysfs_strtoul_bool(writeback_metadata, dc->writeback_metadata);
301 	sysfs_strtoul_bool(writeback_running, dc->writeback_running);
302 	sysfs_strtoul_clamp(writeback_delay, dc->writeback_delay, 0, UINT_MAX);
303 
304 	sysfs_strtoul_clamp(writeback_percent, dc->writeback_percent,
305 			    0, bch_cutoff_writeback);
306 
307 	if (attr == &sysfs_writeback_rate) {
308 		ssize_t ret;
309 		long int v = atomic_long_read(&dc->writeback_rate.rate);
310 
311 		ret = strtoul_safe_clamp(buf, v, 1, INT_MAX);
312 
313 		if (!ret) {
314 			atomic_long_set(&dc->writeback_rate.rate, v);
315 			ret = size;
316 		}
317 
318 		return ret;
319 	}
320 
321 	sysfs_strtoul_clamp(writeback_rate_update_seconds,
322 			    dc->writeback_rate_update_seconds,
323 			    1, WRITEBACK_RATE_UPDATE_SECS_MAX);
324 	sysfs_strtoul_clamp(writeback_rate_i_term_inverse,
325 			    dc->writeback_rate_i_term_inverse,
326 			    1, UINT_MAX);
327 	sysfs_strtoul_clamp(writeback_rate_p_term_inverse,
328 			    dc->writeback_rate_p_term_inverse,
329 			    1, UINT_MAX);
330 	sysfs_strtoul_clamp(writeback_rate_minimum,
331 			    dc->writeback_rate_minimum,
332 			    1, UINT_MAX);
333 
334 	sysfs_strtoul_clamp(io_error_limit, dc->error_limit, 0, INT_MAX);
335 
336 	if (attr == &sysfs_io_disable) {
337 		int v = strtoul_or_return(buf);
338 
339 		dc->io_disable = v ? 1 : 0;
340 	}
341 
342 	sysfs_strtoul_clamp(sequential_cutoff,
343 			    dc->sequential_cutoff,
344 			    0, UINT_MAX);
345 	d_strtoi_h(readahead);
346 
347 	if (attr == &sysfs_clear_stats)
348 		bch_cache_accounting_clear(&dc->accounting);
349 
350 	if (attr == &sysfs_running &&
351 	    strtoul_or_return(buf)) {
352 		v = bch_cached_dev_run(dc);
353 		if (v)
354 			return v;
355 	}
356 
357 	if (attr == &sysfs_cache_mode) {
358 		v = __sysfs_match_string(bch_cache_modes, -1, buf);
359 		if (v < 0)
360 			return v;
361 
362 		if ((unsigned int) v != BDEV_CACHE_MODE(&dc->sb)) {
363 			SET_BDEV_CACHE_MODE(&dc->sb, v);
364 			bch_write_bdev_super(dc, NULL);
365 		}
366 	}
367 
368 	if (attr == &sysfs_readahead_cache_policy) {
369 		v = __sysfs_match_string(bch_reada_cache_policies, -1, buf);
370 		if (v < 0)
371 			return v;
372 
373 		if ((unsigned int) v != dc->cache_readahead_policy)
374 			dc->cache_readahead_policy = v;
375 	}
376 
377 	if (attr == &sysfs_stop_when_cache_set_failed) {
378 		v = __sysfs_match_string(bch_stop_on_failure_modes, -1, buf);
379 		if (v < 0)
380 			return v;
381 
382 		dc->stop_when_cache_set_failed = v;
383 	}
384 
385 	if (attr == &sysfs_label) {
386 		if (size > SB_LABEL_SIZE)
387 			return -EINVAL;
388 		memcpy(dc->sb.label, buf, size);
389 		if (size < SB_LABEL_SIZE)
390 			dc->sb.label[size] = '\0';
391 		if (size && dc->sb.label[size - 1] == '\n')
392 			dc->sb.label[size - 1] = '\0';
393 		bch_write_bdev_super(dc, NULL);
394 		if (dc->disk.c) {
395 			memcpy(dc->disk.c->uuids[dc->disk.id].label,
396 			       buf, SB_LABEL_SIZE);
397 			bch_uuid_write(dc->disk.c);
398 		}
399 		env = kzalloc(sizeof(struct kobj_uevent_env), GFP_KERNEL);
400 		if (!env)
401 			return -ENOMEM;
402 		add_uevent_var(env, "DRIVER=bcache");
403 		add_uevent_var(env, "CACHED_UUID=%pU", dc->sb.uuid),
404 		add_uevent_var(env, "CACHED_LABEL=%s", buf);
405 		kobject_uevent_env(&disk_to_dev(dc->disk.disk)->kobj,
406 				   KOBJ_CHANGE,
407 				   env->envp);
408 		kfree(env);
409 	}
410 
411 	if (attr == &sysfs_attach) {
412 		uint8_t		set_uuid[16];
413 
414 		if (bch_parse_uuid(buf, set_uuid) < 16)
415 			return -EINVAL;
416 
417 		v = -ENOENT;
418 		list_for_each_entry(c, &bch_cache_sets, list) {
419 			v = bch_cached_dev_attach(dc, c, set_uuid);
420 			if (!v)
421 				return size;
422 		}
423 		if (v == -ENOENT)
424 			pr_err("Can't attach %s: cache set not found\n", buf);
425 		return v;
426 	}
427 
428 	if (attr == &sysfs_detach && dc->disk.c)
429 		bch_cached_dev_detach(dc);
430 
431 	if (attr == &sysfs_stop)
432 		bcache_device_stop(&dc->disk);
433 
434 	return size;
435 }
436 
437 STORE(bch_cached_dev)
438 {
439 	struct cached_dev *dc = container_of(kobj, struct cached_dev,
440 					     disk.kobj);
441 
442 	/* no user space access if system is rebooting */
443 	if (bcache_is_reboot)
444 		return -EBUSY;
445 
446 	mutex_lock(&bch_register_lock);
447 	size = __cached_dev_store(kobj, attr, buf, size);
448 
449 	if (attr == &sysfs_writeback_running) {
450 		/* dc->writeback_running changed in __cached_dev_store() */
451 		if (IS_ERR_OR_NULL(dc->writeback_thread)) {
452 			/*
453 			 * reject setting it to 1 via sysfs if writeback
454 			 * kthread is not created yet.
455 			 */
456 			if (dc->writeback_running) {
457 				dc->writeback_running = false;
458 				pr_err("%s: failed to run non-existent writeback thread\n",
459 						dc->disk.disk->disk_name);
460 			}
461 		} else
462 			/*
463 			 * writeback kthread will check if dc->writeback_running
464 			 * is true or false.
465 			 */
466 			bch_writeback_queue(dc);
467 	}
468 
469 	/*
470 	 * Only set BCACHE_DEV_WB_RUNNING when cached device attached to
471 	 * a cache set, otherwise it doesn't make sense.
472 	 */
473 	if (attr == &sysfs_writeback_percent)
474 		if ((dc->disk.c != NULL) &&
475 		    (!test_and_set_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags)))
476 			schedule_delayed_work(&dc->writeback_rate_update,
477 				      dc->writeback_rate_update_seconds * HZ);
478 
479 	mutex_unlock(&bch_register_lock);
480 	return size;
481 }
482 
483 static struct attribute *bch_cached_dev_files[] = {
484 	&sysfs_attach,
485 	&sysfs_detach,
486 	&sysfs_stop,
487 #if 0
488 	&sysfs_data_csum,
489 #endif
490 	&sysfs_cache_mode,
491 	&sysfs_readahead_cache_policy,
492 	&sysfs_stop_when_cache_set_failed,
493 	&sysfs_writeback_metadata,
494 	&sysfs_writeback_running,
495 	&sysfs_writeback_delay,
496 	&sysfs_writeback_percent,
497 	&sysfs_writeback_rate,
498 	&sysfs_writeback_rate_update_seconds,
499 	&sysfs_writeback_rate_i_term_inverse,
500 	&sysfs_writeback_rate_p_term_inverse,
501 	&sysfs_writeback_rate_minimum,
502 	&sysfs_writeback_rate_debug,
503 	&sysfs_io_errors,
504 	&sysfs_io_error_limit,
505 	&sysfs_io_disable,
506 	&sysfs_dirty_data,
507 	&sysfs_stripe_size,
508 	&sysfs_partial_stripes_expensive,
509 	&sysfs_sequential_cutoff,
510 	&sysfs_clear_stats,
511 	&sysfs_running,
512 	&sysfs_state,
513 	&sysfs_label,
514 	&sysfs_readahead,
515 #ifdef CONFIG_BCACHE_DEBUG
516 	&sysfs_verify,
517 	&sysfs_bypass_torture_test,
518 #endif
519 	&sysfs_backing_dev_name,
520 	&sysfs_backing_dev_uuid,
521 	NULL
522 };
523 KTYPE(bch_cached_dev);
524 
525 SHOW(bch_flash_dev)
526 {
527 	struct bcache_device *d = container_of(kobj, struct bcache_device,
528 					       kobj);
529 	struct uuid_entry *u = &d->c->uuids[d->id];
530 
531 	sysfs_printf(data_csum,	"%i", d->data_csum);
532 	sysfs_hprint(size,	u->sectors << 9);
533 
534 	if (attr == &sysfs_label) {
535 		memcpy(buf, u->label, SB_LABEL_SIZE);
536 		buf[SB_LABEL_SIZE + 1] = '\0';
537 		strcat(buf, "\n");
538 		return strlen(buf);
539 	}
540 
541 	return 0;
542 }
543 
544 STORE(__bch_flash_dev)
545 {
546 	struct bcache_device *d = container_of(kobj, struct bcache_device,
547 					       kobj);
548 	struct uuid_entry *u = &d->c->uuids[d->id];
549 
550 	/* no user space access if system is rebooting */
551 	if (bcache_is_reboot)
552 		return -EBUSY;
553 
554 	sysfs_strtoul(data_csum,	d->data_csum);
555 
556 	if (attr == &sysfs_size) {
557 		uint64_t v;
558 
559 		strtoi_h_or_return(buf, v);
560 
561 		u->sectors = v >> 9;
562 		bch_uuid_write(d->c);
563 		set_capacity(d->disk, u->sectors);
564 	}
565 
566 	if (attr == &sysfs_label) {
567 		memcpy(u->label, buf, SB_LABEL_SIZE);
568 		bch_uuid_write(d->c);
569 	}
570 
571 	if (attr == &sysfs_unregister) {
572 		set_bit(BCACHE_DEV_DETACHING, &d->flags);
573 		bcache_device_stop(d);
574 	}
575 
576 	return size;
577 }
578 STORE_LOCKED(bch_flash_dev)
579 
580 static struct attribute *bch_flash_dev_files[] = {
581 	&sysfs_unregister,
582 #if 0
583 	&sysfs_data_csum,
584 #endif
585 	&sysfs_label,
586 	&sysfs_size,
587 	NULL
588 };
589 KTYPE(bch_flash_dev);
590 
591 struct bset_stats_op {
592 	struct btree_op op;
593 	size_t nodes;
594 	struct bset_stats stats;
595 };
596 
597 static int bch_btree_bset_stats(struct btree_op *b_op, struct btree *b)
598 {
599 	struct bset_stats_op *op = container_of(b_op, struct bset_stats_op, op);
600 
601 	op->nodes++;
602 	bch_btree_keys_stats(&b->keys, &op->stats);
603 
604 	return MAP_CONTINUE;
605 }
606 
607 static int bch_bset_print_stats(struct cache_set *c, char *buf)
608 {
609 	struct bset_stats_op op;
610 	int ret;
611 
612 	memset(&op, 0, sizeof(op));
613 	bch_btree_op_init(&op.op, -1);
614 
615 	ret = bch_btree_map_nodes(&op.op, c, &ZERO_KEY, bch_btree_bset_stats);
616 	if (ret < 0)
617 		return ret;
618 
619 	return snprintf(buf, PAGE_SIZE,
620 			"btree nodes:		%zu\n"
621 			"written sets:		%zu\n"
622 			"unwritten sets:		%zu\n"
623 			"written key bytes:	%zu\n"
624 			"unwritten key bytes:	%zu\n"
625 			"floats:			%zu\n"
626 			"failed:			%zu\n",
627 			op.nodes,
628 			op.stats.sets_written, op.stats.sets_unwritten,
629 			op.stats.bytes_written, op.stats.bytes_unwritten,
630 			op.stats.floats, op.stats.failed);
631 }
632 
633 static unsigned int bch_root_usage(struct cache_set *c)
634 {
635 	unsigned int bytes = 0;
636 	struct bkey *k;
637 	struct btree *b;
638 	struct btree_iter iter;
639 
640 	goto lock_root;
641 
642 	do {
643 		rw_unlock(false, b);
644 lock_root:
645 		b = c->root;
646 		rw_lock(false, b, b->level);
647 	} while (b != c->root);
648 
649 	for_each_key_filter(&b->keys, k, &iter, bch_ptr_bad)
650 		bytes += bkey_bytes(k);
651 
652 	rw_unlock(false, b);
653 
654 	return (bytes * 100) / btree_bytes(c);
655 }
656 
657 static size_t bch_cache_size(struct cache_set *c)
658 {
659 	size_t ret = 0;
660 	struct btree *b;
661 
662 	mutex_lock(&c->bucket_lock);
663 	list_for_each_entry(b, &c->btree_cache, list)
664 		ret += 1 << (b->keys.page_order + PAGE_SHIFT);
665 
666 	mutex_unlock(&c->bucket_lock);
667 	return ret;
668 }
669 
670 static unsigned int bch_cache_max_chain(struct cache_set *c)
671 {
672 	unsigned int ret = 0;
673 	struct hlist_head *h;
674 
675 	mutex_lock(&c->bucket_lock);
676 
677 	for (h = c->bucket_hash;
678 	     h < c->bucket_hash + (1 << BUCKET_HASH_BITS);
679 	     h++) {
680 		unsigned int i = 0;
681 		struct hlist_node *p;
682 
683 		hlist_for_each(p, h)
684 			i++;
685 
686 		ret = max(ret, i);
687 	}
688 
689 	mutex_unlock(&c->bucket_lock);
690 	return ret;
691 }
692 
693 static unsigned int bch_btree_used(struct cache_set *c)
694 {
695 	return div64_u64(c->gc_stats.key_bytes * 100,
696 			 (c->gc_stats.nodes ?: 1) * btree_bytes(c));
697 }
698 
699 static unsigned int bch_average_key_size(struct cache_set *c)
700 {
701 	return c->gc_stats.nkeys
702 		? div64_u64(c->gc_stats.data, c->gc_stats.nkeys)
703 		: 0;
704 }
705 
706 SHOW(__bch_cache_set)
707 {
708 	struct cache_set *c = container_of(kobj, struct cache_set, kobj);
709 
710 	sysfs_print(synchronous,		CACHE_SYNC(&c->sb));
711 	sysfs_print(journal_delay_ms,		c->journal_delay_ms);
712 	sysfs_hprint(bucket_size,		bucket_bytes(c));
713 	sysfs_hprint(block_size,		block_bytes(c));
714 	sysfs_print(tree_depth,			c->root->level);
715 	sysfs_print(root_usage_percent,		bch_root_usage(c));
716 
717 	sysfs_hprint(btree_cache_size,		bch_cache_size(c));
718 	sysfs_print(btree_cache_max_chain,	bch_cache_max_chain(c));
719 	sysfs_print(cache_available_percent,	100 - c->gc_stats.in_use);
720 
721 	sysfs_print_time_stats(&c->btree_gc_time,	btree_gc, sec, ms);
722 	sysfs_print_time_stats(&c->btree_split_time,	btree_split, sec, us);
723 	sysfs_print_time_stats(&c->sort.time,		btree_sort, ms, us);
724 	sysfs_print_time_stats(&c->btree_read_time,	btree_read, ms, us);
725 
726 	sysfs_print(btree_used_percent,	bch_btree_used(c));
727 	sysfs_print(btree_nodes,	c->gc_stats.nodes);
728 	sysfs_hprint(average_key_size,	bch_average_key_size(c));
729 
730 	sysfs_print(cache_read_races,
731 		    atomic_long_read(&c->cache_read_races));
732 
733 	sysfs_print(reclaim,
734 		    atomic_long_read(&c->reclaim));
735 
736 	sysfs_print(reclaimed_journal_buckets,
737 		    atomic_long_read(&c->reclaimed_journal_buckets));
738 
739 	sysfs_print(flush_write,
740 		    atomic_long_read(&c->flush_write));
741 
742 	sysfs_print(writeback_keys_done,
743 		    atomic_long_read(&c->writeback_keys_done));
744 	sysfs_print(writeback_keys_failed,
745 		    atomic_long_read(&c->writeback_keys_failed));
746 
747 	if (attr == &sysfs_errors)
748 		return bch_snprint_string_list(buf, PAGE_SIZE, error_actions,
749 					       c->on_error);
750 
751 	/* See count_io_errors for why 88 */
752 	sysfs_print(io_error_halflife,	c->error_decay * 88);
753 	sysfs_print(io_error_limit,	c->error_limit);
754 
755 	sysfs_hprint(congested,
756 		     ((uint64_t) bch_get_congested(c)) << 9);
757 	sysfs_print(congested_read_threshold_us,
758 		    c->congested_read_threshold_us);
759 	sysfs_print(congested_write_threshold_us,
760 		    c->congested_write_threshold_us);
761 
762 	sysfs_print(cutoff_writeback, bch_cutoff_writeback);
763 	sysfs_print(cutoff_writeback_sync, bch_cutoff_writeback_sync);
764 
765 	sysfs_print(active_journal_entries,	fifo_used(&c->journal.pin));
766 	sysfs_printf(verify,			"%i", c->verify);
767 	sysfs_printf(key_merging_disabled,	"%i", c->key_merging_disabled);
768 	sysfs_printf(expensive_debug_checks,
769 		     "%i", c->expensive_debug_checks);
770 	sysfs_printf(gc_always_rewrite,		"%i", c->gc_always_rewrite);
771 	sysfs_printf(btree_shrinker_disabled,	"%i", c->shrinker_disabled);
772 	sysfs_printf(copy_gc_enabled,		"%i", c->copy_gc_enabled);
773 	sysfs_printf(idle_max_writeback_rate,	"%i",
774 		     c->idle_max_writeback_rate_enabled);
775 	sysfs_printf(gc_after_writeback,	"%i", c->gc_after_writeback);
776 	sysfs_printf(io_disable,		"%i",
777 		     test_bit(CACHE_SET_IO_DISABLE, &c->flags));
778 
779 	if (attr == &sysfs_bset_tree_stats)
780 		return bch_bset_print_stats(c, buf);
781 
782 	return 0;
783 }
784 SHOW_LOCKED(bch_cache_set)
785 
786 STORE(__bch_cache_set)
787 {
788 	struct cache_set *c = container_of(kobj, struct cache_set, kobj);
789 	ssize_t v;
790 
791 	/* no user space access if system is rebooting */
792 	if (bcache_is_reboot)
793 		return -EBUSY;
794 
795 	if (attr == &sysfs_unregister)
796 		bch_cache_set_unregister(c);
797 
798 	if (attr == &sysfs_stop)
799 		bch_cache_set_stop(c);
800 
801 	if (attr == &sysfs_synchronous) {
802 		bool sync = strtoul_or_return(buf);
803 
804 		if (sync != CACHE_SYNC(&c->sb)) {
805 			SET_CACHE_SYNC(&c->sb, sync);
806 			bcache_write_super(c);
807 		}
808 	}
809 
810 	if (attr == &sysfs_flash_vol_create) {
811 		int r;
812 		uint64_t v;
813 
814 		strtoi_h_or_return(buf, v);
815 
816 		r = bch_flash_dev_create(c, v);
817 		if (r)
818 			return r;
819 	}
820 
821 	if (attr == &sysfs_clear_stats) {
822 		atomic_long_set(&c->writeback_keys_done,	0);
823 		atomic_long_set(&c->writeback_keys_failed,	0);
824 
825 		memset(&c->gc_stats, 0, sizeof(struct gc_stat));
826 		bch_cache_accounting_clear(&c->accounting);
827 	}
828 
829 	if (attr == &sysfs_trigger_gc)
830 		force_wake_up_gc(c);
831 
832 	if (attr == &sysfs_prune_cache) {
833 		struct shrink_control sc;
834 
835 		sc.gfp_mask = GFP_KERNEL;
836 		sc.nr_to_scan = strtoul_or_return(buf);
837 		c->shrink.scan_objects(&c->shrink, &sc);
838 	}
839 
840 	sysfs_strtoul_clamp(congested_read_threshold_us,
841 			    c->congested_read_threshold_us,
842 			    0, UINT_MAX);
843 	sysfs_strtoul_clamp(congested_write_threshold_us,
844 			    c->congested_write_threshold_us,
845 			    0, UINT_MAX);
846 
847 	if (attr == &sysfs_errors) {
848 		v = __sysfs_match_string(error_actions, -1, buf);
849 		if (v < 0)
850 			return v;
851 
852 		c->on_error = v;
853 	}
854 
855 	sysfs_strtoul_clamp(io_error_limit, c->error_limit, 0, UINT_MAX);
856 
857 	/* See count_io_errors() for why 88 */
858 	if (attr == &sysfs_io_error_halflife) {
859 		unsigned long v = 0;
860 		ssize_t ret;
861 
862 		ret = strtoul_safe_clamp(buf, v, 0, UINT_MAX);
863 		if (!ret) {
864 			c->error_decay = v / 88;
865 			return size;
866 		}
867 		return ret;
868 	}
869 
870 	if (attr == &sysfs_io_disable) {
871 		v = strtoul_or_return(buf);
872 		if (v) {
873 			if (test_and_set_bit(CACHE_SET_IO_DISABLE,
874 					     &c->flags))
875 				pr_warn("CACHE_SET_IO_DISABLE already set\n");
876 		} else {
877 			if (!test_and_clear_bit(CACHE_SET_IO_DISABLE,
878 						&c->flags))
879 				pr_warn("CACHE_SET_IO_DISABLE already cleared\n");
880 		}
881 	}
882 
883 	sysfs_strtoul_clamp(journal_delay_ms,
884 			    c->journal_delay_ms,
885 			    0, USHRT_MAX);
886 	sysfs_strtoul_bool(verify,		c->verify);
887 	sysfs_strtoul_bool(key_merging_disabled, c->key_merging_disabled);
888 	sysfs_strtoul(expensive_debug_checks,	c->expensive_debug_checks);
889 	sysfs_strtoul_bool(gc_always_rewrite,	c->gc_always_rewrite);
890 	sysfs_strtoul_bool(btree_shrinker_disabled, c->shrinker_disabled);
891 	sysfs_strtoul_bool(copy_gc_enabled,	c->copy_gc_enabled);
892 	sysfs_strtoul_bool(idle_max_writeback_rate,
893 			   c->idle_max_writeback_rate_enabled);
894 
895 	/*
896 	 * write gc_after_writeback here may overwrite an already set
897 	 * BCH_DO_AUTO_GC, it doesn't matter because this flag will be
898 	 * set in next chance.
899 	 */
900 	sysfs_strtoul_clamp(gc_after_writeback, c->gc_after_writeback, 0, 1);
901 
902 	return size;
903 }
904 STORE_LOCKED(bch_cache_set)
905 
906 SHOW(bch_cache_set_internal)
907 {
908 	struct cache_set *c = container_of(kobj, struct cache_set, internal);
909 
910 	return bch_cache_set_show(&c->kobj, attr, buf);
911 }
912 
913 STORE(bch_cache_set_internal)
914 {
915 	struct cache_set *c = container_of(kobj, struct cache_set, internal);
916 
917 	/* no user space access if system is rebooting */
918 	if (bcache_is_reboot)
919 		return -EBUSY;
920 
921 	return bch_cache_set_store(&c->kobj, attr, buf, size);
922 }
923 
924 static void bch_cache_set_internal_release(struct kobject *k)
925 {
926 }
927 
928 static struct attribute *bch_cache_set_files[] = {
929 	&sysfs_unregister,
930 	&sysfs_stop,
931 	&sysfs_synchronous,
932 	&sysfs_journal_delay_ms,
933 	&sysfs_flash_vol_create,
934 
935 	&sysfs_bucket_size,
936 	&sysfs_block_size,
937 	&sysfs_tree_depth,
938 	&sysfs_root_usage_percent,
939 	&sysfs_btree_cache_size,
940 	&sysfs_cache_available_percent,
941 
942 	&sysfs_average_key_size,
943 
944 	&sysfs_errors,
945 	&sysfs_io_error_limit,
946 	&sysfs_io_error_halflife,
947 	&sysfs_congested,
948 	&sysfs_congested_read_threshold_us,
949 	&sysfs_congested_write_threshold_us,
950 	&sysfs_clear_stats,
951 	NULL
952 };
953 KTYPE(bch_cache_set);
954 
955 static struct attribute *bch_cache_set_internal_files[] = {
956 	&sysfs_active_journal_entries,
957 
958 	sysfs_time_stats_attribute_list(btree_gc, sec, ms)
959 	sysfs_time_stats_attribute_list(btree_split, sec, us)
960 	sysfs_time_stats_attribute_list(btree_sort, ms, us)
961 	sysfs_time_stats_attribute_list(btree_read, ms, us)
962 
963 	&sysfs_btree_nodes,
964 	&sysfs_btree_used_percent,
965 	&sysfs_btree_cache_max_chain,
966 
967 	&sysfs_bset_tree_stats,
968 	&sysfs_cache_read_races,
969 	&sysfs_reclaim,
970 	&sysfs_reclaimed_journal_buckets,
971 	&sysfs_flush_write,
972 	&sysfs_writeback_keys_done,
973 	&sysfs_writeback_keys_failed,
974 
975 	&sysfs_trigger_gc,
976 	&sysfs_prune_cache,
977 #ifdef CONFIG_BCACHE_DEBUG
978 	&sysfs_verify,
979 	&sysfs_key_merging_disabled,
980 	&sysfs_expensive_debug_checks,
981 #endif
982 	&sysfs_gc_always_rewrite,
983 	&sysfs_btree_shrinker_disabled,
984 	&sysfs_copy_gc_enabled,
985 	&sysfs_idle_max_writeback_rate,
986 	&sysfs_gc_after_writeback,
987 	&sysfs_io_disable,
988 	&sysfs_cutoff_writeback,
989 	&sysfs_cutoff_writeback_sync,
990 	NULL
991 };
992 KTYPE(bch_cache_set_internal);
993 
994 static int __bch_cache_cmp(const void *l, const void *r)
995 {
996 	cond_resched();
997 	return *((uint16_t *)r) - *((uint16_t *)l);
998 }
999 
1000 SHOW(__bch_cache)
1001 {
1002 	struct cache *ca = container_of(kobj, struct cache, kobj);
1003 
1004 	sysfs_hprint(bucket_size,	bucket_bytes(ca));
1005 	sysfs_hprint(block_size,	block_bytes(ca));
1006 	sysfs_print(nbuckets,		ca->sb.nbuckets);
1007 	sysfs_print(discard,		ca->discard);
1008 	sysfs_hprint(written, atomic_long_read(&ca->sectors_written) << 9);
1009 	sysfs_hprint(btree_written,
1010 		     atomic_long_read(&ca->btree_sectors_written) << 9);
1011 	sysfs_hprint(metadata_written,
1012 		     (atomic_long_read(&ca->meta_sectors_written) +
1013 		      atomic_long_read(&ca->btree_sectors_written)) << 9);
1014 
1015 	sysfs_print(io_errors,
1016 		    atomic_read(&ca->io_errors) >> IO_ERROR_SHIFT);
1017 
1018 	if (attr == &sysfs_cache_replacement_policy)
1019 		return bch_snprint_string_list(buf, PAGE_SIZE,
1020 					       cache_replacement_policies,
1021 					       CACHE_REPLACEMENT(&ca->sb));
1022 
1023 	if (attr == &sysfs_priority_stats) {
1024 		struct bucket *b;
1025 		size_t n = ca->sb.nbuckets, i;
1026 		size_t unused = 0, available = 0, dirty = 0, meta = 0;
1027 		uint64_t sum = 0;
1028 		/* Compute 31 quantiles */
1029 		uint16_t q[31], *p, *cached;
1030 		ssize_t ret;
1031 
1032 		cached = p = vmalloc(array_size(sizeof(uint16_t),
1033 						ca->sb.nbuckets));
1034 		if (!p)
1035 			return -ENOMEM;
1036 
1037 		mutex_lock(&ca->set->bucket_lock);
1038 		for_each_bucket(b, ca) {
1039 			if (!GC_SECTORS_USED(b))
1040 				unused++;
1041 			if (GC_MARK(b) == GC_MARK_RECLAIMABLE)
1042 				available++;
1043 			if (GC_MARK(b) == GC_MARK_DIRTY)
1044 				dirty++;
1045 			if (GC_MARK(b) == GC_MARK_METADATA)
1046 				meta++;
1047 		}
1048 
1049 		for (i = ca->sb.first_bucket; i < n; i++)
1050 			p[i] = ca->buckets[i].prio;
1051 		mutex_unlock(&ca->set->bucket_lock);
1052 
1053 		sort(p, n, sizeof(uint16_t), __bch_cache_cmp, NULL);
1054 
1055 		while (n &&
1056 		       !cached[n - 1])
1057 			--n;
1058 
1059 		while (cached < p + n &&
1060 		       *cached == BTREE_PRIO)
1061 			cached++, n--;
1062 
1063 		for (i = 0; i < n; i++)
1064 			sum += INITIAL_PRIO - cached[i];
1065 
1066 		if (n)
1067 			do_div(sum, n);
1068 
1069 		for (i = 0; i < ARRAY_SIZE(q); i++)
1070 			q[i] = INITIAL_PRIO - cached[n * (i + 1) /
1071 				(ARRAY_SIZE(q) + 1)];
1072 
1073 		vfree(p);
1074 
1075 		ret = scnprintf(buf, PAGE_SIZE,
1076 				"Unused:		%zu%%\n"
1077 				"Clean:		%zu%%\n"
1078 				"Dirty:		%zu%%\n"
1079 				"Metadata:	%zu%%\n"
1080 				"Average:	%llu\n"
1081 				"Sectors per Q:	%zu\n"
1082 				"Quantiles:	[",
1083 				unused * 100 / (size_t) ca->sb.nbuckets,
1084 				available * 100 / (size_t) ca->sb.nbuckets,
1085 				dirty * 100 / (size_t) ca->sb.nbuckets,
1086 				meta * 100 / (size_t) ca->sb.nbuckets, sum,
1087 				n * ca->sb.bucket_size / (ARRAY_SIZE(q) + 1));
1088 
1089 		for (i = 0; i < ARRAY_SIZE(q); i++)
1090 			ret += scnprintf(buf + ret, PAGE_SIZE - ret,
1091 					 "%u ", q[i]);
1092 		ret--;
1093 
1094 		ret += scnprintf(buf + ret, PAGE_SIZE - ret, "]\n");
1095 
1096 		return ret;
1097 	}
1098 
1099 	return 0;
1100 }
1101 SHOW_LOCKED(bch_cache)
1102 
1103 STORE(__bch_cache)
1104 {
1105 	struct cache *ca = container_of(kobj, struct cache, kobj);
1106 	ssize_t v;
1107 
1108 	/* no user space access if system is rebooting */
1109 	if (bcache_is_reboot)
1110 		return -EBUSY;
1111 
1112 	if (attr == &sysfs_discard) {
1113 		bool v = strtoul_or_return(buf);
1114 
1115 		if (blk_queue_discard(bdev_get_queue(ca->bdev)))
1116 			ca->discard = v;
1117 
1118 		if (v != CACHE_DISCARD(&ca->sb)) {
1119 			SET_CACHE_DISCARD(&ca->sb, v);
1120 			bcache_write_super(ca->set);
1121 		}
1122 	}
1123 
1124 	if (attr == &sysfs_cache_replacement_policy) {
1125 		v = __sysfs_match_string(cache_replacement_policies, -1, buf);
1126 		if (v < 0)
1127 			return v;
1128 
1129 		if ((unsigned int) v != CACHE_REPLACEMENT(&ca->sb)) {
1130 			mutex_lock(&ca->set->bucket_lock);
1131 			SET_CACHE_REPLACEMENT(&ca->sb, v);
1132 			mutex_unlock(&ca->set->bucket_lock);
1133 
1134 			bcache_write_super(ca->set);
1135 		}
1136 	}
1137 
1138 	if (attr == &sysfs_clear_stats) {
1139 		atomic_long_set(&ca->sectors_written, 0);
1140 		atomic_long_set(&ca->btree_sectors_written, 0);
1141 		atomic_long_set(&ca->meta_sectors_written, 0);
1142 		atomic_set(&ca->io_count, 0);
1143 		atomic_set(&ca->io_errors, 0);
1144 	}
1145 
1146 	return size;
1147 }
1148 STORE_LOCKED(bch_cache)
1149 
1150 static struct attribute *bch_cache_files[] = {
1151 	&sysfs_bucket_size,
1152 	&sysfs_block_size,
1153 	&sysfs_nbuckets,
1154 	&sysfs_priority_stats,
1155 	&sysfs_discard,
1156 	&sysfs_written,
1157 	&sysfs_btree_written,
1158 	&sysfs_metadata_written,
1159 	&sysfs_io_errors,
1160 	&sysfs_clear_stats,
1161 	&sysfs_cache_replacement_policy,
1162 	NULL
1163 };
1164 KTYPE(bch_cache);
1165