xref: /openbmc/linux/drivers/md/bcache/sysfs.c (revision 2169e6da)
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 };
28 
29 /* Default is 0 ("auto") */
30 static const char * const bch_stop_on_failure_modes[] = {
31 	"auto",
32 	"always"
33 };
34 
35 static const char * const cache_replacement_policies[] = {
36 	"lru",
37 	"fifo",
38 	"random"
39 };
40 
41 static const char * const error_actions[] = {
42 	"unregister",
43 	"panic"
44 };
45 
46 write_attribute(attach);
47 write_attribute(detach);
48 write_attribute(unregister);
49 write_attribute(stop);
50 write_attribute(clear_stats);
51 write_attribute(trigger_gc);
52 write_attribute(prune_cache);
53 write_attribute(flash_vol_create);
54 
55 read_attribute(bucket_size);
56 read_attribute(block_size);
57 read_attribute(nbuckets);
58 read_attribute(tree_depth);
59 read_attribute(root_usage_percent);
60 read_attribute(priority_stats);
61 read_attribute(btree_cache_size);
62 read_attribute(btree_cache_max_chain);
63 read_attribute(cache_available_percent);
64 read_attribute(written);
65 read_attribute(btree_written);
66 read_attribute(metadata_written);
67 read_attribute(active_journal_entries);
68 read_attribute(backing_dev_name);
69 read_attribute(backing_dev_uuid);
70 
71 sysfs_time_stats_attribute(btree_gc,	sec, ms);
72 sysfs_time_stats_attribute(btree_split, sec, us);
73 sysfs_time_stats_attribute(btree_sort,	ms,  us);
74 sysfs_time_stats_attribute(btree_read,	ms,  us);
75 
76 read_attribute(btree_nodes);
77 read_attribute(btree_used_percent);
78 read_attribute(average_key_size);
79 read_attribute(dirty_data);
80 read_attribute(bset_tree_stats);
81 
82 read_attribute(state);
83 read_attribute(cache_read_races);
84 read_attribute(reclaim);
85 read_attribute(reclaimed_journal_buckets);
86 read_attribute(flush_write);
87 read_attribute(writeback_keys_done);
88 read_attribute(writeback_keys_failed);
89 read_attribute(io_errors);
90 read_attribute(congested);
91 read_attribute(cutoff_writeback);
92 read_attribute(cutoff_writeback_sync);
93 rw_attribute(congested_read_threshold_us);
94 rw_attribute(congested_write_threshold_us);
95 
96 rw_attribute(sequential_cutoff);
97 rw_attribute(data_csum);
98 rw_attribute(cache_mode);
99 rw_attribute(stop_when_cache_set_failed);
100 rw_attribute(writeback_metadata);
101 rw_attribute(writeback_running);
102 rw_attribute(writeback_percent);
103 rw_attribute(writeback_delay);
104 rw_attribute(writeback_rate);
105 
106 rw_attribute(writeback_rate_update_seconds);
107 rw_attribute(writeback_rate_i_term_inverse);
108 rw_attribute(writeback_rate_p_term_inverse);
109 rw_attribute(writeback_rate_minimum);
110 read_attribute(writeback_rate_debug);
111 
112 read_attribute(stripe_size);
113 read_attribute(partial_stripes_expensive);
114 
115 rw_attribute(synchronous);
116 rw_attribute(journal_delay_ms);
117 rw_attribute(io_disable);
118 rw_attribute(discard);
119 rw_attribute(running);
120 rw_attribute(label);
121 rw_attribute(readahead);
122 rw_attribute(errors);
123 rw_attribute(io_error_limit);
124 rw_attribute(io_error_halflife);
125 rw_attribute(verify);
126 rw_attribute(bypass_torture_test);
127 rw_attribute(key_merging_disabled);
128 rw_attribute(gc_always_rewrite);
129 rw_attribute(expensive_debug_checks);
130 rw_attribute(cache_replacement_policy);
131 rw_attribute(btree_shrinker_disabled);
132 rw_attribute(copy_gc_enabled);
133 rw_attribute(gc_after_writeback);
134 rw_attribute(size);
135 
136 static ssize_t bch_snprint_string_list(char *buf,
137 				       size_t size,
138 				       const char * const list[],
139 				       size_t selected)
140 {
141 	char *out = buf;
142 	size_t i;
143 
144 	for (i = 0; list[i]; i++)
145 		out += snprintf(out, buf + size - out,
146 				i == selected ? "[%s] " : "%s ", list[i]);
147 
148 	out[-1] = '\n';
149 	return out - buf;
150 }
151 
152 SHOW(__bch_cached_dev)
153 {
154 	struct cached_dev *dc = container_of(kobj, struct cached_dev,
155 					     disk.kobj);
156 	char const *states[] = { "no cache", "clean", "dirty", "inconsistent" };
157 	int wb = dc->writeback_running;
158 
159 #define var(stat)		(dc->stat)
160 
161 	if (attr == &sysfs_cache_mode)
162 		return bch_snprint_string_list(buf, PAGE_SIZE,
163 					       bch_cache_modes,
164 					       BDEV_CACHE_MODE(&dc->sb));
165 
166 	if (attr == &sysfs_stop_when_cache_set_failed)
167 		return bch_snprint_string_list(buf, PAGE_SIZE,
168 					       bch_stop_on_failure_modes,
169 					       dc->stop_when_cache_set_failed);
170 
171 
172 	sysfs_printf(data_csum,		"%i", dc->disk.data_csum);
173 	var_printf(verify,		"%i");
174 	var_printf(bypass_torture_test,	"%i");
175 	var_printf(writeback_metadata,	"%i");
176 	var_printf(writeback_running,	"%i");
177 	var_print(writeback_delay);
178 	var_print(writeback_percent);
179 	sysfs_hprint(writeback_rate,
180 		     wb ? atomic_long_read(&dc->writeback_rate.rate) << 9 : 0);
181 	sysfs_printf(io_errors,		"%i", atomic_read(&dc->io_errors));
182 	sysfs_printf(io_error_limit,	"%i", dc->error_limit);
183 	sysfs_printf(io_disable,	"%i", dc->io_disable);
184 	var_print(writeback_rate_update_seconds);
185 	var_print(writeback_rate_i_term_inverse);
186 	var_print(writeback_rate_p_term_inverse);
187 	var_print(writeback_rate_minimum);
188 
189 	if (attr == &sysfs_writeback_rate_debug) {
190 		char rate[20];
191 		char dirty[20];
192 		char target[20];
193 		char proportional[20];
194 		char integral[20];
195 		char change[20];
196 		s64 next_io;
197 
198 		/*
199 		 * Except for dirty and target, other values should
200 		 * be 0 if writeback is not running.
201 		 */
202 		bch_hprint(rate,
203 			   wb ? atomic_long_read(&dc->writeback_rate.rate) << 9
204 			      : 0);
205 		bch_hprint(dirty, bcache_dev_sectors_dirty(&dc->disk) << 9);
206 		bch_hprint(target, dc->writeback_rate_target << 9);
207 		bch_hprint(proportional,
208 			   wb ? dc->writeback_rate_proportional << 9 : 0);
209 		bch_hprint(integral,
210 			   wb ? dc->writeback_rate_integral_scaled << 9 : 0);
211 		bch_hprint(change, wb ? dc->writeback_rate_change << 9 : 0);
212 		next_io = wb ? div64_s64(dc->writeback_rate.next-local_clock(),
213 					 NSEC_PER_MSEC) : 0;
214 
215 		return sprintf(buf,
216 			       "rate:\t\t%s/sec\n"
217 			       "dirty:\t\t%s\n"
218 			       "target:\t\t%s\n"
219 			       "proportional:\t%s\n"
220 			       "integral:\t%s\n"
221 			       "change:\t\t%s/sec\n"
222 			       "next io:\t%llims\n",
223 			       rate, dirty, target, proportional,
224 			       integral, change, next_io);
225 	}
226 
227 	sysfs_hprint(dirty_data,
228 		     bcache_dev_sectors_dirty(&dc->disk) << 9);
229 
230 	sysfs_hprint(stripe_size,	 ((uint64_t)dc->disk.stripe_size) << 9);
231 	var_printf(partial_stripes_expensive,	"%u");
232 
233 	var_hprint(sequential_cutoff);
234 	var_hprint(readahead);
235 
236 	sysfs_print(running,		atomic_read(&dc->running));
237 	sysfs_print(state,		states[BDEV_STATE(&dc->sb)]);
238 
239 	if (attr == &sysfs_label) {
240 		memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
241 		buf[SB_LABEL_SIZE + 1] = '\0';
242 		strcat(buf, "\n");
243 		return strlen(buf);
244 	}
245 
246 	if (attr == &sysfs_backing_dev_name) {
247 		snprintf(buf, BDEVNAME_SIZE + 1, "%s", dc->backing_dev_name);
248 		strcat(buf, "\n");
249 		return strlen(buf);
250 	}
251 
252 	if (attr == &sysfs_backing_dev_uuid) {
253 		/* convert binary uuid into 36-byte string plus '\0' */
254 		snprintf(buf, 36+1, "%pU", dc->sb.uuid);
255 		strcat(buf, "\n");
256 		return strlen(buf);
257 	}
258 
259 #undef var
260 	return 0;
261 }
262 SHOW_LOCKED(bch_cached_dev)
263 
264 STORE(__cached_dev)
265 {
266 	struct cached_dev *dc = container_of(kobj, struct cached_dev,
267 					     disk.kobj);
268 	ssize_t v;
269 	struct cache_set *c;
270 	struct kobj_uevent_env *env;
271 
272 	/* no user space access if system is rebooting */
273 	if (bcache_is_reboot)
274 		return -EBUSY;
275 
276 #define d_strtoul(var)		sysfs_strtoul(var, dc->var)
277 #define d_strtoul_nonzero(var)	sysfs_strtoul_clamp(var, dc->var, 1, INT_MAX)
278 #define d_strtoi_h(var)		sysfs_hatoi(var, dc->var)
279 
280 	sysfs_strtoul(data_csum,	dc->disk.data_csum);
281 	d_strtoul(verify);
282 	sysfs_strtoul_bool(bypass_torture_test, dc->bypass_torture_test);
283 	sysfs_strtoul_bool(writeback_metadata, dc->writeback_metadata);
284 	sysfs_strtoul_bool(writeback_running, dc->writeback_running);
285 	sysfs_strtoul_clamp(writeback_delay, dc->writeback_delay, 0, UINT_MAX);
286 
287 	sysfs_strtoul_clamp(writeback_percent, dc->writeback_percent,
288 			    0, bch_cutoff_writeback);
289 
290 	if (attr == &sysfs_writeback_rate) {
291 		ssize_t ret;
292 		long int v = atomic_long_read(&dc->writeback_rate.rate);
293 
294 		ret = strtoul_safe_clamp(buf, v, 1, INT_MAX);
295 
296 		if (!ret) {
297 			atomic_long_set(&dc->writeback_rate.rate, v);
298 			ret = size;
299 		}
300 
301 		return ret;
302 	}
303 
304 	sysfs_strtoul_clamp(writeback_rate_update_seconds,
305 			    dc->writeback_rate_update_seconds,
306 			    1, WRITEBACK_RATE_UPDATE_SECS_MAX);
307 	sysfs_strtoul_clamp(writeback_rate_i_term_inverse,
308 			    dc->writeback_rate_i_term_inverse,
309 			    1, UINT_MAX);
310 	sysfs_strtoul_clamp(writeback_rate_p_term_inverse,
311 			    dc->writeback_rate_p_term_inverse,
312 			    1, UINT_MAX);
313 	sysfs_strtoul_clamp(writeback_rate_minimum,
314 			    dc->writeback_rate_minimum,
315 			    1, UINT_MAX);
316 
317 	sysfs_strtoul_clamp(io_error_limit, dc->error_limit, 0, INT_MAX);
318 
319 	if (attr == &sysfs_io_disable) {
320 		int v = strtoul_or_return(buf);
321 
322 		dc->io_disable = v ? 1 : 0;
323 	}
324 
325 	sysfs_strtoul_clamp(sequential_cutoff,
326 			    dc->sequential_cutoff,
327 			    0, UINT_MAX);
328 	d_strtoi_h(readahead);
329 
330 	if (attr == &sysfs_clear_stats)
331 		bch_cache_accounting_clear(&dc->accounting);
332 
333 	if (attr == &sysfs_running &&
334 	    strtoul_or_return(buf)) {
335 		v = bch_cached_dev_run(dc);
336 		if (v)
337 			return v;
338 	}
339 
340 	if (attr == &sysfs_cache_mode) {
341 		v = sysfs_match_string(bch_cache_modes, buf);
342 		if (v < 0)
343 			return v;
344 
345 		if ((unsigned int) v != BDEV_CACHE_MODE(&dc->sb)) {
346 			SET_BDEV_CACHE_MODE(&dc->sb, v);
347 			bch_write_bdev_super(dc, NULL);
348 		}
349 	}
350 
351 	if (attr == &sysfs_stop_when_cache_set_failed) {
352 		v = sysfs_match_string(bch_stop_on_failure_modes, buf);
353 		if (v < 0)
354 			return v;
355 
356 		dc->stop_when_cache_set_failed = v;
357 	}
358 
359 	if (attr == &sysfs_label) {
360 		if (size > SB_LABEL_SIZE)
361 			return -EINVAL;
362 		memcpy(dc->sb.label, buf, size);
363 		if (size < SB_LABEL_SIZE)
364 			dc->sb.label[size] = '\0';
365 		if (size && dc->sb.label[size - 1] == '\n')
366 			dc->sb.label[size - 1] = '\0';
367 		bch_write_bdev_super(dc, NULL);
368 		if (dc->disk.c) {
369 			memcpy(dc->disk.c->uuids[dc->disk.id].label,
370 			       buf, SB_LABEL_SIZE);
371 			bch_uuid_write(dc->disk.c);
372 		}
373 		env = kzalloc(sizeof(struct kobj_uevent_env), GFP_KERNEL);
374 		if (!env)
375 			return -ENOMEM;
376 		add_uevent_var(env, "DRIVER=bcache");
377 		add_uevent_var(env, "CACHED_UUID=%pU", dc->sb.uuid),
378 		add_uevent_var(env, "CACHED_LABEL=%s", buf);
379 		kobject_uevent_env(&disk_to_dev(dc->disk.disk)->kobj,
380 				   KOBJ_CHANGE,
381 				   env->envp);
382 		kfree(env);
383 	}
384 
385 	if (attr == &sysfs_attach) {
386 		uint8_t		set_uuid[16];
387 
388 		if (bch_parse_uuid(buf, set_uuid) < 16)
389 			return -EINVAL;
390 
391 		v = -ENOENT;
392 		list_for_each_entry(c, &bch_cache_sets, list) {
393 			v = bch_cached_dev_attach(dc, c, set_uuid);
394 			if (!v)
395 				return size;
396 		}
397 		if (v == -ENOENT)
398 			pr_err("Can't attach %s: cache set not found", buf);
399 		return v;
400 	}
401 
402 	if (attr == &sysfs_detach && dc->disk.c)
403 		bch_cached_dev_detach(dc);
404 
405 	if (attr == &sysfs_stop)
406 		bcache_device_stop(&dc->disk);
407 
408 	return size;
409 }
410 
411 STORE(bch_cached_dev)
412 {
413 	struct cached_dev *dc = container_of(kobj, struct cached_dev,
414 					     disk.kobj);
415 
416 	/* no user space access if system is rebooting */
417 	if (bcache_is_reboot)
418 		return -EBUSY;
419 
420 	mutex_lock(&bch_register_lock);
421 	size = __cached_dev_store(kobj, attr, buf, size);
422 
423 	if (attr == &sysfs_writeback_running) {
424 		/* dc->writeback_running changed in __cached_dev_store() */
425 		if (IS_ERR_OR_NULL(dc->writeback_thread)) {
426 			/*
427 			 * reject setting it to 1 via sysfs if writeback
428 			 * kthread is not created yet.
429 			 */
430 			if (dc->writeback_running) {
431 				dc->writeback_running = false;
432 				pr_err("%s: failed to run non-existent writeback thread",
433 						dc->disk.disk->disk_name);
434 			}
435 		} else
436 			/*
437 			 * writeback kthread will check if dc->writeback_running
438 			 * is true or false.
439 			 */
440 			bch_writeback_queue(dc);
441 	}
442 
443 	/*
444 	 * Only set BCACHE_DEV_WB_RUNNING when cached device attached to
445 	 * a cache set, otherwise it doesn't make sense.
446 	 */
447 	if (attr == &sysfs_writeback_percent)
448 		if ((dc->disk.c != NULL) &&
449 		    (!test_and_set_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags)))
450 			schedule_delayed_work(&dc->writeback_rate_update,
451 				      dc->writeback_rate_update_seconds * HZ);
452 
453 	mutex_unlock(&bch_register_lock);
454 	return size;
455 }
456 
457 static struct attribute *bch_cached_dev_files[] = {
458 	&sysfs_attach,
459 	&sysfs_detach,
460 	&sysfs_stop,
461 #if 0
462 	&sysfs_data_csum,
463 #endif
464 	&sysfs_cache_mode,
465 	&sysfs_stop_when_cache_set_failed,
466 	&sysfs_writeback_metadata,
467 	&sysfs_writeback_running,
468 	&sysfs_writeback_delay,
469 	&sysfs_writeback_percent,
470 	&sysfs_writeback_rate,
471 	&sysfs_writeback_rate_update_seconds,
472 	&sysfs_writeback_rate_i_term_inverse,
473 	&sysfs_writeback_rate_p_term_inverse,
474 	&sysfs_writeback_rate_minimum,
475 	&sysfs_writeback_rate_debug,
476 	&sysfs_io_errors,
477 	&sysfs_io_error_limit,
478 	&sysfs_io_disable,
479 	&sysfs_dirty_data,
480 	&sysfs_stripe_size,
481 	&sysfs_partial_stripes_expensive,
482 	&sysfs_sequential_cutoff,
483 	&sysfs_clear_stats,
484 	&sysfs_running,
485 	&sysfs_state,
486 	&sysfs_label,
487 	&sysfs_readahead,
488 #ifdef CONFIG_BCACHE_DEBUG
489 	&sysfs_verify,
490 	&sysfs_bypass_torture_test,
491 #endif
492 	&sysfs_backing_dev_name,
493 	&sysfs_backing_dev_uuid,
494 	NULL
495 };
496 KTYPE(bch_cached_dev);
497 
498 SHOW(bch_flash_dev)
499 {
500 	struct bcache_device *d = container_of(kobj, struct bcache_device,
501 					       kobj);
502 	struct uuid_entry *u = &d->c->uuids[d->id];
503 
504 	sysfs_printf(data_csum,	"%i", d->data_csum);
505 	sysfs_hprint(size,	u->sectors << 9);
506 
507 	if (attr == &sysfs_label) {
508 		memcpy(buf, u->label, SB_LABEL_SIZE);
509 		buf[SB_LABEL_SIZE + 1] = '\0';
510 		strcat(buf, "\n");
511 		return strlen(buf);
512 	}
513 
514 	return 0;
515 }
516 
517 STORE(__bch_flash_dev)
518 {
519 	struct bcache_device *d = container_of(kobj, struct bcache_device,
520 					       kobj);
521 	struct uuid_entry *u = &d->c->uuids[d->id];
522 
523 	/* no user space access if system is rebooting */
524 	if (bcache_is_reboot)
525 		return -EBUSY;
526 
527 	sysfs_strtoul(data_csum,	d->data_csum);
528 
529 	if (attr == &sysfs_size) {
530 		uint64_t v;
531 
532 		strtoi_h_or_return(buf, v);
533 
534 		u->sectors = v >> 9;
535 		bch_uuid_write(d->c);
536 		set_capacity(d->disk, u->sectors);
537 	}
538 
539 	if (attr == &sysfs_label) {
540 		memcpy(u->label, buf, SB_LABEL_SIZE);
541 		bch_uuid_write(d->c);
542 	}
543 
544 	if (attr == &sysfs_unregister) {
545 		set_bit(BCACHE_DEV_DETACHING, &d->flags);
546 		bcache_device_stop(d);
547 	}
548 
549 	return size;
550 }
551 STORE_LOCKED(bch_flash_dev)
552 
553 static struct attribute *bch_flash_dev_files[] = {
554 	&sysfs_unregister,
555 #if 0
556 	&sysfs_data_csum,
557 #endif
558 	&sysfs_label,
559 	&sysfs_size,
560 	NULL
561 };
562 KTYPE(bch_flash_dev);
563 
564 struct bset_stats_op {
565 	struct btree_op op;
566 	size_t nodes;
567 	struct bset_stats stats;
568 };
569 
570 static int bch_btree_bset_stats(struct btree_op *b_op, struct btree *b)
571 {
572 	struct bset_stats_op *op = container_of(b_op, struct bset_stats_op, op);
573 
574 	op->nodes++;
575 	bch_btree_keys_stats(&b->keys, &op->stats);
576 
577 	return MAP_CONTINUE;
578 }
579 
580 static int bch_bset_print_stats(struct cache_set *c, char *buf)
581 {
582 	struct bset_stats_op op;
583 	int ret;
584 
585 	memset(&op, 0, sizeof(op));
586 	bch_btree_op_init(&op.op, -1);
587 
588 	ret = bch_btree_map_nodes(&op.op, c, &ZERO_KEY, bch_btree_bset_stats);
589 	if (ret < 0)
590 		return ret;
591 
592 	return snprintf(buf, PAGE_SIZE,
593 			"btree nodes:		%zu\n"
594 			"written sets:		%zu\n"
595 			"unwritten sets:		%zu\n"
596 			"written key bytes:	%zu\n"
597 			"unwritten key bytes:	%zu\n"
598 			"floats:			%zu\n"
599 			"failed:			%zu\n",
600 			op.nodes,
601 			op.stats.sets_written, op.stats.sets_unwritten,
602 			op.stats.bytes_written, op.stats.bytes_unwritten,
603 			op.stats.floats, op.stats.failed);
604 }
605 
606 static unsigned int bch_root_usage(struct cache_set *c)
607 {
608 	unsigned int bytes = 0;
609 	struct bkey *k;
610 	struct btree *b;
611 	struct btree_iter iter;
612 
613 	goto lock_root;
614 
615 	do {
616 		rw_unlock(false, b);
617 lock_root:
618 		b = c->root;
619 		rw_lock(false, b, b->level);
620 	} while (b != c->root);
621 
622 	for_each_key_filter(&b->keys, k, &iter, bch_ptr_bad)
623 		bytes += bkey_bytes(k);
624 
625 	rw_unlock(false, b);
626 
627 	return (bytes * 100) / btree_bytes(c);
628 }
629 
630 static size_t bch_cache_size(struct cache_set *c)
631 {
632 	size_t ret = 0;
633 	struct btree *b;
634 
635 	mutex_lock(&c->bucket_lock);
636 	list_for_each_entry(b, &c->btree_cache, list)
637 		ret += 1 << (b->keys.page_order + PAGE_SHIFT);
638 
639 	mutex_unlock(&c->bucket_lock);
640 	return ret;
641 }
642 
643 static unsigned int bch_cache_max_chain(struct cache_set *c)
644 {
645 	unsigned int ret = 0;
646 	struct hlist_head *h;
647 
648 	mutex_lock(&c->bucket_lock);
649 
650 	for (h = c->bucket_hash;
651 	     h < c->bucket_hash + (1 << BUCKET_HASH_BITS);
652 	     h++) {
653 		unsigned int i = 0;
654 		struct hlist_node *p;
655 
656 		hlist_for_each(p, h)
657 			i++;
658 
659 		ret = max(ret, i);
660 	}
661 
662 	mutex_unlock(&c->bucket_lock);
663 	return ret;
664 }
665 
666 static unsigned int bch_btree_used(struct cache_set *c)
667 {
668 	return div64_u64(c->gc_stats.key_bytes * 100,
669 			 (c->gc_stats.nodes ?: 1) * btree_bytes(c));
670 }
671 
672 static unsigned int bch_average_key_size(struct cache_set *c)
673 {
674 	return c->gc_stats.nkeys
675 		? div64_u64(c->gc_stats.data, c->gc_stats.nkeys)
676 		: 0;
677 }
678 
679 SHOW(__bch_cache_set)
680 {
681 	struct cache_set *c = container_of(kobj, struct cache_set, kobj);
682 
683 	sysfs_print(synchronous,		CACHE_SYNC(&c->sb));
684 	sysfs_print(journal_delay_ms,		c->journal_delay_ms);
685 	sysfs_hprint(bucket_size,		bucket_bytes(c));
686 	sysfs_hprint(block_size,		block_bytes(c));
687 	sysfs_print(tree_depth,			c->root->level);
688 	sysfs_print(root_usage_percent,		bch_root_usage(c));
689 
690 	sysfs_hprint(btree_cache_size,		bch_cache_size(c));
691 	sysfs_print(btree_cache_max_chain,	bch_cache_max_chain(c));
692 	sysfs_print(cache_available_percent,	100 - c->gc_stats.in_use);
693 
694 	sysfs_print_time_stats(&c->btree_gc_time,	btree_gc, sec, ms);
695 	sysfs_print_time_stats(&c->btree_split_time,	btree_split, sec, us);
696 	sysfs_print_time_stats(&c->sort.time,		btree_sort, ms, us);
697 	sysfs_print_time_stats(&c->btree_read_time,	btree_read, ms, us);
698 
699 	sysfs_print(btree_used_percent,	bch_btree_used(c));
700 	sysfs_print(btree_nodes,	c->gc_stats.nodes);
701 	sysfs_hprint(average_key_size,	bch_average_key_size(c));
702 
703 	sysfs_print(cache_read_races,
704 		    atomic_long_read(&c->cache_read_races));
705 
706 	sysfs_print(reclaim,
707 		    atomic_long_read(&c->reclaim));
708 
709 	sysfs_print(reclaimed_journal_buckets,
710 		    atomic_long_read(&c->reclaimed_journal_buckets));
711 
712 	sysfs_print(flush_write,
713 		    atomic_long_read(&c->flush_write));
714 
715 	sysfs_print(writeback_keys_done,
716 		    atomic_long_read(&c->writeback_keys_done));
717 	sysfs_print(writeback_keys_failed,
718 		    atomic_long_read(&c->writeback_keys_failed));
719 
720 	if (attr == &sysfs_errors)
721 		return bch_snprint_string_list(buf, PAGE_SIZE, error_actions,
722 					       c->on_error);
723 
724 	/* See count_io_errors for why 88 */
725 	sysfs_print(io_error_halflife,	c->error_decay * 88);
726 	sysfs_print(io_error_limit,	c->error_limit);
727 
728 	sysfs_hprint(congested,
729 		     ((uint64_t) bch_get_congested(c)) << 9);
730 	sysfs_print(congested_read_threshold_us,
731 		    c->congested_read_threshold_us);
732 	sysfs_print(congested_write_threshold_us,
733 		    c->congested_write_threshold_us);
734 
735 	sysfs_print(cutoff_writeback, bch_cutoff_writeback);
736 	sysfs_print(cutoff_writeback_sync, bch_cutoff_writeback_sync);
737 
738 	sysfs_print(active_journal_entries,	fifo_used(&c->journal.pin));
739 	sysfs_printf(verify,			"%i", c->verify);
740 	sysfs_printf(key_merging_disabled,	"%i", c->key_merging_disabled);
741 	sysfs_printf(expensive_debug_checks,
742 		     "%i", c->expensive_debug_checks);
743 	sysfs_printf(gc_always_rewrite,		"%i", c->gc_always_rewrite);
744 	sysfs_printf(btree_shrinker_disabled,	"%i", c->shrinker_disabled);
745 	sysfs_printf(copy_gc_enabled,		"%i", c->copy_gc_enabled);
746 	sysfs_printf(gc_after_writeback,	"%i", c->gc_after_writeback);
747 	sysfs_printf(io_disable,		"%i",
748 		     test_bit(CACHE_SET_IO_DISABLE, &c->flags));
749 
750 	if (attr == &sysfs_bset_tree_stats)
751 		return bch_bset_print_stats(c, buf);
752 
753 	return 0;
754 }
755 SHOW_LOCKED(bch_cache_set)
756 
757 STORE(__bch_cache_set)
758 {
759 	struct cache_set *c = container_of(kobj, struct cache_set, kobj);
760 	ssize_t v;
761 
762 	/* no user space access if system is rebooting */
763 	if (bcache_is_reboot)
764 		return -EBUSY;
765 
766 	if (attr == &sysfs_unregister)
767 		bch_cache_set_unregister(c);
768 
769 	if (attr == &sysfs_stop)
770 		bch_cache_set_stop(c);
771 
772 	if (attr == &sysfs_synchronous) {
773 		bool sync = strtoul_or_return(buf);
774 
775 		if (sync != CACHE_SYNC(&c->sb)) {
776 			SET_CACHE_SYNC(&c->sb, sync);
777 			bcache_write_super(c);
778 		}
779 	}
780 
781 	if (attr == &sysfs_flash_vol_create) {
782 		int r;
783 		uint64_t v;
784 
785 		strtoi_h_or_return(buf, v);
786 
787 		r = bch_flash_dev_create(c, v);
788 		if (r)
789 			return r;
790 	}
791 
792 	if (attr == &sysfs_clear_stats) {
793 		atomic_long_set(&c->writeback_keys_done,	0);
794 		atomic_long_set(&c->writeback_keys_failed,	0);
795 
796 		memset(&c->gc_stats, 0, sizeof(struct gc_stat));
797 		bch_cache_accounting_clear(&c->accounting);
798 	}
799 
800 	if (attr == &sysfs_trigger_gc)
801 		force_wake_up_gc(c);
802 
803 	if (attr == &sysfs_prune_cache) {
804 		struct shrink_control sc;
805 
806 		sc.gfp_mask = GFP_KERNEL;
807 		sc.nr_to_scan = strtoul_or_return(buf);
808 		c->shrink.scan_objects(&c->shrink, &sc);
809 	}
810 
811 	sysfs_strtoul_clamp(congested_read_threshold_us,
812 			    c->congested_read_threshold_us,
813 			    0, UINT_MAX);
814 	sysfs_strtoul_clamp(congested_write_threshold_us,
815 			    c->congested_write_threshold_us,
816 			    0, UINT_MAX);
817 
818 	if (attr == &sysfs_errors) {
819 		v = sysfs_match_string(error_actions, buf);
820 		if (v < 0)
821 			return v;
822 
823 		c->on_error = v;
824 	}
825 
826 	sysfs_strtoul_clamp(io_error_limit, c->error_limit, 0, UINT_MAX);
827 
828 	/* See count_io_errors() for why 88 */
829 	if (attr == &sysfs_io_error_halflife) {
830 		unsigned long v = 0;
831 		ssize_t ret;
832 
833 		ret = strtoul_safe_clamp(buf, v, 0, UINT_MAX);
834 		if (!ret) {
835 			c->error_decay = v / 88;
836 			return size;
837 		}
838 		return ret;
839 	}
840 
841 	if (attr == &sysfs_io_disable) {
842 		v = strtoul_or_return(buf);
843 		if (v) {
844 			if (test_and_set_bit(CACHE_SET_IO_DISABLE,
845 					     &c->flags))
846 				pr_warn("CACHE_SET_IO_DISABLE already set");
847 		} else {
848 			if (!test_and_clear_bit(CACHE_SET_IO_DISABLE,
849 						&c->flags))
850 				pr_warn("CACHE_SET_IO_DISABLE already cleared");
851 		}
852 	}
853 
854 	sysfs_strtoul_clamp(journal_delay_ms,
855 			    c->journal_delay_ms,
856 			    0, USHRT_MAX);
857 	sysfs_strtoul_bool(verify,		c->verify);
858 	sysfs_strtoul_bool(key_merging_disabled, c->key_merging_disabled);
859 	sysfs_strtoul(expensive_debug_checks,	c->expensive_debug_checks);
860 	sysfs_strtoul_bool(gc_always_rewrite,	c->gc_always_rewrite);
861 	sysfs_strtoul_bool(btree_shrinker_disabled, c->shrinker_disabled);
862 	sysfs_strtoul_bool(copy_gc_enabled,	c->copy_gc_enabled);
863 	/*
864 	 * write gc_after_writeback here may overwrite an already set
865 	 * BCH_DO_AUTO_GC, it doesn't matter because this flag will be
866 	 * set in next chance.
867 	 */
868 	sysfs_strtoul_clamp(gc_after_writeback, c->gc_after_writeback, 0, 1);
869 
870 	return size;
871 }
872 STORE_LOCKED(bch_cache_set)
873 
874 SHOW(bch_cache_set_internal)
875 {
876 	struct cache_set *c = container_of(kobj, struct cache_set, internal);
877 
878 	return bch_cache_set_show(&c->kobj, attr, buf);
879 }
880 
881 STORE(bch_cache_set_internal)
882 {
883 	struct cache_set *c = container_of(kobj, struct cache_set, internal);
884 
885 	/* no user space access if system is rebooting */
886 	if (bcache_is_reboot)
887 		return -EBUSY;
888 
889 	return bch_cache_set_store(&c->kobj, attr, buf, size);
890 }
891 
892 static void bch_cache_set_internal_release(struct kobject *k)
893 {
894 }
895 
896 static struct attribute *bch_cache_set_files[] = {
897 	&sysfs_unregister,
898 	&sysfs_stop,
899 	&sysfs_synchronous,
900 	&sysfs_journal_delay_ms,
901 	&sysfs_flash_vol_create,
902 
903 	&sysfs_bucket_size,
904 	&sysfs_block_size,
905 	&sysfs_tree_depth,
906 	&sysfs_root_usage_percent,
907 	&sysfs_btree_cache_size,
908 	&sysfs_cache_available_percent,
909 
910 	&sysfs_average_key_size,
911 
912 	&sysfs_errors,
913 	&sysfs_io_error_limit,
914 	&sysfs_io_error_halflife,
915 	&sysfs_congested,
916 	&sysfs_congested_read_threshold_us,
917 	&sysfs_congested_write_threshold_us,
918 	&sysfs_clear_stats,
919 	NULL
920 };
921 KTYPE(bch_cache_set);
922 
923 static struct attribute *bch_cache_set_internal_files[] = {
924 	&sysfs_active_journal_entries,
925 
926 	sysfs_time_stats_attribute_list(btree_gc, sec, ms)
927 	sysfs_time_stats_attribute_list(btree_split, sec, us)
928 	sysfs_time_stats_attribute_list(btree_sort, ms, us)
929 	sysfs_time_stats_attribute_list(btree_read, ms, us)
930 
931 	&sysfs_btree_nodes,
932 	&sysfs_btree_used_percent,
933 	&sysfs_btree_cache_max_chain,
934 
935 	&sysfs_bset_tree_stats,
936 	&sysfs_cache_read_races,
937 	&sysfs_reclaim,
938 	&sysfs_reclaimed_journal_buckets,
939 	&sysfs_flush_write,
940 	&sysfs_writeback_keys_done,
941 	&sysfs_writeback_keys_failed,
942 
943 	&sysfs_trigger_gc,
944 	&sysfs_prune_cache,
945 #ifdef CONFIG_BCACHE_DEBUG
946 	&sysfs_verify,
947 	&sysfs_key_merging_disabled,
948 	&sysfs_expensive_debug_checks,
949 #endif
950 	&sysfs_gc_always_rewrite,
951 	&sysfs_btree_shrinker_disabled,
952 	&sysfs_copy_gc_enabled,
953 	&sysfs_gc_after_writeback,
954 	&sysfs_io_disable,
955 	&sysfs_cutoff_writeback,
956 	&sysfs_cutoff_writeback_sync,
957 	NULL
958 };
959 KTYPE(bch_cache_set_internal);
960 
961 static int __bch_cache_cmp(const void *l, const void *r)
962 {
963 	return *((uint16_t *)r) - *((uint16_t *)l);
964 }
965 
966 SHOW(__bch_cache)
967 {
968 	struct cache *ca = container_of(kobj, struct cache, kobj);
969 
970 	sysfs_hprint(bucket_size,	bucket_bytes(ca));
971 	sysfs_hprint(block_size,	block_bytes(ca));
972 	sysfs_print(nbuckets,		ca->sb.nbuckets);
973 	sysfs_print(discard,		ca->discard);
974 	sysfs_hprint(written, atomic_long_read(&ca->sectors_written) << 9);
975 	sysfs_hprint(btree_written,
976 		     atomic_long_read(&ca->btree_sectors_written) << 9);
977 	sysfs_hprint(metadata_written,
978 		     (atomic_long_read(&ca->meta_sectors_written) +
979 		      atomic_long_read(&ca->btree_sectors_written)) << 9);
980 
981 	sysfs_print(io_errors,
982 		    atomic_read(&ca->io_errors) >> IO_ERROR_SHIFT);
983 
984 	if (attr == &sysfs_cache_replacement_policy)
985 		return bch_snprint_string_list(buf, PAGE_SIZE,
986 					       cache_replacement_policies,
987 					       CACHE_REPLACEMENT(&ca->sb));
988 
989 	if (attr == &sysfs_priority_stats) {
990 		struct bucket *b;
991 		size_t n = ca->sb.nbuckets, i;
992 		size_t unused = 0, available = 0, dirty = 0, meta = 0;
993 		uint64_t sum = 0;
994 		/* Compute 31 quantiles */
995 		uint16_t q[31], *p, *cached;
996 		ssize_t ret;
997 
998 		cached = p = vmalloc(array_size(sizeof(uint16_t),
999 						ca->sb.nbuckets));
1000 		if (!p)
1001 			return -ENOMEM;
1002 
1003 		mutex_lock(&ca->set->bucket_lock);
1004 		for_each_bucket(b, ca) {
1005 			if (!GC_SECTORS_USED(b))
1006 				unused++;
1007 			if (GC_MARK(b) == GC_MARK_RECLAIMABLE)
1008 				available++;
1009 			if (GC_MARK(b) == GC_MARK_DIRTY)
1010 				dirty++;
1011 			if (GC_MARK(b) == GC_MARK_METADATA)
1012 				meta++;
1013 		}
1014 
1015 		for (i = ca->sb.first_bucket; i < n; i++)
1016 			p[i] = ca->buckets[i].prio;
1017 		mutex_unlock(&ca->set->bucket_lock);
1018 
1019 		sort(p, n, sizeof(uint16_t), __bch_cache_cmp, NULL);
1020 
1021 		while (n &&
1022 		       !cached[n - 1])
1023 			--n;
1024 
1025 		while (cached < p + n &&
1026 		       *cached == BTREE_PRIO)
1027 			cached++, n--;
1028 
1029 		for (i = 0; i < n; i++)
1030 			sum += INITIAL_PRIO - cached[i];
1031 
1032 		if (n)
1033 			do_div(sum, n);
1034 
1035 		for (i = 0; i < ARRAY_SIZE(q); i++)
1036 			q[i] = INITIAL_PRIO - cached[n * (i + 1) /
1037 				(ARRAY_SIZE(q) + 1)];
1038 
1039 		vfree(p);
1040 
1041 		ret = scnprintf(buf, PAGE_SIZE,
1042 				"Unused:		%zu%%\n"
1043 				"Clean:		%zu%%\n"
1044 				"Dirty:		%zu%%\n"
1045 				"Metadata:	%zu%%\n"
1046 				"Average:	%llu\n"
1047 				"Sectors per Q:	%zu\n"
1048 				"Quantiles:	[",
1049 				unused * 100 / (size_t) ca->sb.nbuckets,
1050 				available * 100 / (size_t) ca->sb.nbuckets,
1051 				dirty * 100 / (size_t) ca->sb.nbuckets,
1052 				meta * 100 / (size_t) ca->sb.nbuckets, sum,
1053 				n * ca->sb.bucket_size / (ARRAY_SIZE(q) + 1));
1054 
1055 		for (i = 0; i < ARRAY_SIZE(q); i++)
1056 			ret += scnprintf(buf + ret, PAGE_SIZE - ret,
1057 					 "%u ", q[i]);
1058 		ret--;
1059 
1060 		ret += scnprintf(buf + ret, PAGE_SIZE - ret, "]\n");
1061 
1062 		return ret;
1063 	}
1064 
1065 	return 0;
1066 }
1067 SHOW_LOCKED(bch_cache)
1068 
1069 STORE(__bch_cache)
1070 {
1071 	struct cache *ca = container_of(kobj, struct cache, kobj);
1072 	ssize_t v;
1073 
1074 	/* no user space access if system is rebooting */
1075 	if (bcache_is_reboot)
1076 		return -EBUSY;
1077 
1078 	if (attr == &sysfs_discard) {
1079 		bool v = strtoul_or_return(buf);
1080 
1081 		if (blk_queue_discard(bdev_get_queue(ca->bdev)))
1082 			ca->discard = v;
1083 
1084 		if (v != CACHE_DISCARD(&ca->sb)) {
1085 			SET_CACHE_DISCARD(&ca->sb, v);
1086 			bcache_write_super(ca->set);
1087 		}
1088 	}
1089 
1090 	if (attr == &sysfs_cache_replacement_policy) {
1091 		v = sysfs_match_string(cache_replacement_policies, buf);
1092 		if (v < 0)
1093 			return v;
1094 
1095 		if ((unsigned int) v != CACHE_REPLACEMENT(&ca->sb)) {
1096 			mutex_lock(&ca->set->bucket_lock);
1097 			SET_CACHE_REPLACEMENT(&ca->sb, v);
1098 			mutex_unlock(&ca->set->bucket_lock);
1099 
1100 			bcache_write_super(ca->set);
1101 		}
1102 	}
1103 
1104 	if (attr == &sysfs_clear_stats) {
1105 		atomic_long_set(&ca->sectors_written, 0);
1106 		atomic_long_set(&ca->btree_sectors_written, 0);
1107 		atomic_long_set(&ca->meta_sectors_written, 0);
1108 		atomic_set(&ca->io_count, 0);
1109 		atomic_set(&ca->io_errors, 0);
1110 	}
1111 
1112 	return size;
1113 }
1114 STORE_LOCKED(bch_cache)
1115 
1116 static struct attribute *bch_cache_files[] = {
1117 	&sysfs_bucket_size,
1118 	&sysfs_block_size,
1119 	&sysfs_nbuckets,
1120 	&sysfs_priority_stats,
1121 	&sysfs_discard,
1122 	&sysfs_written,
1123 	&sysfs_btree_written,
1124 	&sysfs_metadata_written,
1125 	&sysfs_io_errors,
1126 	&sysfs_clear_stats,
1127 	&sysfs_cache_replacement_policy,
1128 	NULL
1129 };
1130 KTYPE(bch_cache);
1131