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