drivers/md/bcache/sysfs.c

   1 /*
   2  * bcache sysfs interfaces
   3  *
   4  * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
   5  * Copyright 2012 Google, Inc.
   6  */
   7
   8 #include "bcache.h"
   9 #include "sysfs.h"
  10 #include "btree.h"
  11 #include "request.h"
  12 #include "writeback.h"
  13
  14 #include <linux/blkdev.h>
  15 #include <linux/sort.h>
  16
  17 static const char * const cache_replacement_policies[] = {
  18         "lru",
  19         "fifo",
  20         "random",
  21         NULL
  22 };
  23
  24 write_attribute(attach);
  25 write_attribute(detach);
  26 write_attribute(unregister);
  27 write_attribute(stop);
  28 write_attribute(clear_stats);
  29 write_attribute(trigger_gc);
  30 write_attribute(prune_cache);
  31 write_attribute(flash_vol_create);
  32
  33 read_attribute(bucket_size);
  34 read_attribute(block_size);
  35 read_attribute(nbuckets);
  36 read_attribute(tree_depth);
  37 read_attribute(root_usage_percent);
  38 read_attribute(priority_stats);
  39 read_attribute(btree_cache_size);
  40 read_attribute(btree_cache_max_chain);
  41 read_attribute(cache_available_percent);
  42 read_attribute(written);
  43 read_attribute(btree_written);
  44 read_attribute(metadata_written);
  45 read_attribute(active_journal_entries);
  46
  47 sysfs_time_stats_attribute(btree_gc,    sec, ms);
  48 sysfs_time_stats_attribute(btree_split, sec, us);
  49 sysfs_time_stats_attribute(btree_sort,  ms,  us);
  50 sysfs_time_stats_attribute(btree_read,  ms,  us);
  51 sysfs_time_stats_attribute(try_harder,  ms,  us);
  52
  53 read_attribute(btree_nodes);
  54 read_attribute(btree_used_percent);
  55 read_attribute(average_key_size);
  56 read_attribute(dirty_data);
  57 read_attribute(bset_tree_stats);
  58
  59 read_attribute(state);
  60 read_attribute(cache_read_races);
  61 read_attribute(writeback_keys_done);
  62 read_attribute(writeback_keys_failed);
  63 read_attribute(io_errors);
  64 read_attribute(congested);
  65 rw_attribute(congested_read_threshold_us);
  66 rw_attribute(congested_write_threshold_us);
  67
  68 rw_attribute(sequential_cutoff);
  69 rw_attribute(sequential_merge);
  70 rw_attribute(data_csum);
  71 rw_attribute(cache_mode);
  72 rw_attribute(writeback_metadata);
  73 rw_attribute(writeback_running);
  74 rw_attribute(writeback_percent);
  75 rw_attribute(writeback_delay);
  76 rw_attribute(writeback_rate);
  77
  78 rw_attribute(writeback_rate_update_seconds);
  79 rw_attribute(writeback_rate_d_term);
  80 rw_attribute(writeback_rate_p_term_inverse);
  81 rw_attribute(writeback_rate_d_smooth);
  82 read_attribute(writeback_rate_debug);
  83
  84 read_attribute(stripe_size);
  85 read_attribute(partial_stripes_expensive);
  86
  87 rw_attribute(synchronous);
  88 rw_attribute(journal_delay_ms);
  89 rw_attribute(discard);
  90 rw_attribute(running);
  91 rw_attribute(label);
  92 rw_attribute(readahead);
  93 rw_attribute(io_error_limit);
  94 rw_attribute(io_error_halflife);
  95 rw_attribute(verify);
  96 rw_attribute(key_merging_disabled);
  97 rw_attribute(gc_always_rewrite);
  98 rw_attribute(freelist_percent);
  99 rw_attribute(cache_replacement_policy);
 100 rw_attribute(btree_shrinker_disabled);
 101 rw_attribute(copy_gc_enabled);
 102 rw_attribute(size);
 103
 104 SHOW(__bch_cached_dev)
 105 {
 106         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 107                                              disk.kobj);
 108         const char *states[] = { "no cache", "clean", "dirty", "inconsistent" };
 109
 110 #define var(stat)               (dc->stat)
 111
 112         if (attr == &sysfs_cache_mode)
 113                 return bch_snprint_string_list(buf, PAGE_SIZE,
 114                                                bch_cache_modes + 1,
 115                                                BDEV_CACHE_MODE(&dc->sb));
 116
 117         sysfs_printf(data_csum,         "%i", dc->disk.data_csum);
 118         var_printf(verify,              "%i");
 119         var_printf(writeback_metadata,  "%i");
 120         var_printf(writeback_running,   "%i");
 121         var_print(writeback_delay);
 122         var_print(writeback_percent);
 123         sysfs_print(writeback_rate,     dc->writeback_rate.rate);
 124
 125         var_print(writeback_rate_update_seconds);
 126         var_print(writeback_rate_d_term);
 127         var_print(writeback_rate_p_term_inverse);
 128         var_print(writeback_rate_d_smooth);
 129
 130         if (attr == &sysfs_writeback_rate_debug) {
 131                 char dirty[20];
 132                 char derivative[20];
 133                 char target[20];
 134                 bch_hprint(dirty,
 135                            bcache_dev_sectors_dirty(&dc->disk) << 9);
 136                 bch_hprint(derivative,  dc->writeback_rate_derivative << 9);
 137                 bch_hprint(target,      dc->writeback_rate_target << 9);
 138
 139                 return sprintf(buf,
 140                                "rate:\t\t%u\n"
 141                                "change:\t\t%i\n"
 142                                "dirty:\t\t%s\n"
 143                                "derivative:\t%s\n"
 144                                "target:\t\t%s\n",
 145                                dc->writeback_rate.rate,
 146                                dc->writeback_rate_change,
 147                                dirty, derivative, target);
 148         }
 149
 150         sysfs_hprint(dirty_data,
 151                      bcache_dev_sectors_dirty(&dc->disk) << 9);
 152
 153         sysfs_hprint(stripe_size,       (1 << dc->disk.stripe_size_bits) << 9);
 154         var_printf(partial_stripes_expensive,   "%u");
 155
 156         var_printf(sequential_merge,    "%i");
 157         var_hprint(sequential_cutoff);
 158         var_hprint(readahead);
 159
 160         sysfs_print(running,            atomic_read(&dc->running));
 161         sysfs_print(state,              states[BDEV_STATE(&dc->sb)]);
 162
 163         if (attr == &sysfs_label) {
 164                 memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
 165                 buf[SB_LABEL_SIZE + 1] = '\0';
 166                 strcat(buf, "\n");
 167                 return strlen(buf);
 168         }
 169
 170 #undef var
 171         return 0;
 172 }
 173 SHOW_LOCKED(bch_cached_dev)
 174
 175 STORE(__cached_dev)
 176 {
 177         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 178                                              disk.kobj);
 179         unsigned v = size;
 180         struct cache_set *c;
 181         struct kobj_uevent_env *env;
 182
 183 #define d_strtoul(var)          sysfs_strtoul(var, dc->var)
 184 #define d_strtoi_h(var)         sysfs_hatoi(var, dc->var)
 185
 186         sysfs_strtoul(data_csum,        dc->disk.data_csum);
 187         d_strtoul(verify);
 188         d_strtoul(writeback_metadata);
 189         d_strtoul(writeback_running);
 190         d_strtoul(writeback_delay);
 191         sysfs_strtoul_clamp(writeback_rate,
 192                             dc->writeback_rate.rate, 1, 1000000);
 193         sysfs_strtoul_clamp(writeback_percent, dc->writeback_percent, 0, 40);
 194
 195         d_strtoul(writeback_rate_update_seconds);
 196         d_strtoul(writeback_rate_d_term);
 197         d_strtoul(writeback_rate_p_term_inverse);
 198         sysfs_strtoul_clamp(writeback_rate_p_term_inverse,
 199                             dc->writeback_rate_p_term_inverse, 1, INT_MAX);
 200         d_strtoul(writeback_rate_d_smooth);
 201
 202         d_strtoul(sequential_merge);
 203         d_strtoi_h(sequential_cutoff);
 204         d_strtoi_h(readahead);
 205
 206         if (attr == &sysfs_clear_stats)
 207                 bch_cache_accounting_clear(&dc->accounting);
 208
 209         if (attr == &sysfs_running &&
 210             strtoul_or_return(buf))
 211                 bch_cached_dev_run(dc);
 212
 213         if (attr == &sysfs_cache_mode) {
 214                 ssize_t v = bch_read_string_list(buf, bch_cache_modes + 1);
 215
 216                 if (v < 0)
 217                         return v;
 218
 219                 if ((unsigned) v != BDEV_CACHE_MODE(&dc->sb)) {
 220                         SET_BDEV_CACHE_MODE(&dc->sb, v);
 221                         bch_write_bdev_super(dc, NULL);
 222                 }
 223         }
 224
 225         if (attr == &sysfs_label) {
 226                 /* note: endlines are preserved */
 227                 memcpy(dc->sb.label, buf, SB_LABEL_SIZE);
 228                 bch_write_bdev_super(dc, NULL);
 229                 if (dc->disk.c) {
 230                         memcpy(dc->disk.c->uuids[dc->disk.id].label,
 231                                buf, SB_LABEL_SIZE);
 232                         bch_uuid_write(dc->disk.c);
 233                 }
 234                 env = kzalloc(sizeof(struct kobj_uevent_env), GFP_KERNEL);
 235                 if (!env)
 236                         return -ENOMEM;
 237                 add_uevent_var(env, "DRIVER=bcache");
 238                 add_uevent_var(env, "CACHED_UUID=%pU", dc->sb.uuid),
 239                 add_uevent_var(env, "CACHED_LABEL=%s", buf);
 240                 kobject_uevent_env(
 241                         &disk_to_dev(dc->disk.disk)->kobj, KOBJ_CHANGE, env->envp);
 242                 kfree(env);
 243         }
 244
 245         if (attr == &sysfs_attach) {
 246                 if (bch_parse_uuid(buf, dc->sb.set_uuid) < 16)
 247                         return -EINVAL;
 248
 249                 list_for_each_entry(c, &bch_cache_sets, list) {
 250                         v = bch_cached_dev_attach(dc, c);
 251                         if (!v)
 252                                 return size;
 253                 }
 254
 255                 pr_err("Can't attach %s: cache set not found", buf);
 256                 size = v;
 257         }
 258
 259         if (attr == &sysfs_detach && dc->disk.c)
 260                 bch_cached_dev_detach(dc);
 261
 262         if (attr == &sysfs_stop)
 263                 bcache_device_stop(&dc->disk);
 264
 265         return size;
 266 }
 267
 268 STORE(bch_cached_dev)
 269 {
 270         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 271                                              disk.kobj);
 272
 273         mutex_lock(&bch_register_lock);
 274         size = __cached_dev_store(kobj, attr, buf, size);
 275
 276         if (attr == &sysfs_writeback_running)
 277                 bch_writeback_queue(dc);
 278
 279         if (attr == &sysfs_writeback_percent)
 280                 schedule_delayed_work(&dc->writeback_rate_update,
 281                                       dc->writeback_rate_update_seconds * HZ);
 282
 283         mutex_unlock(&bch_register_lock);
 284         return size;
 285 }
 286
 287 static struct attribute *bch_cached_dev_files[] = {
 288         &sysfs_attach,
 289         &sysfs_detach,
 290         &sysfs_stop,
 291 #if 0
 292         &sysfs_data_csum,
 293 #endif
 294         &sysfs_cache_mode,
 295         &sysfs_writeback_metadata,
 296         &sysfs_writeback_running,
 297         &sysfs_writeback_delay,
 298         &sysfs_writeback_percent,
 299         &sysfs_writeback_rate,
 300         &sysfs_writeback_rate_update_seconds,
 301         &sysfs_writeback_rate_d_term,
 302         &sysfs_writeback_rate_p_term_inverse,
 303         &sysfs_writeback_rate_d_smooth,
 304         &sysfs_writeback_rate_debug,
 305         &sysfs_dirty_data,
 306         &sysfs_stripe_size,
 307         &sysfs_partial_stripes_expensive,
 308         &sysfs_sequential_cutoff,
 309         &sysfs_sequential_merge,
 310         &sysfs_clear_stats,
 311         &sysfs_running,
 312         &sysfs_state,
 313         &sysfs_label,
 314         &sysfs_readahead,
 315 #ifdef CONFIG_BCACHE_DEBUG
 316         &sysfs_verify,
 317 #endif
 318         NULL
 319 };
 320 KTYPE(bch_cached_dev);
 321
 322 SHOW(bch_flash_dev)
 323 {
 324         struct bcache_device *d = container_of(kobj, struct bcache_device,
 325                                                kobj);
 326         struct uuid_entry *u = &d->c->uuids[d->id];
 327
 328         sysfs_printf(data_csum, "%i", d->data_csum);
 329         sysfs_hprint(size,      u->sectors << 9);
 330
 331         if (attr == &sysfs_label) {
 332                 memcpy(buf, u->label, SB_LABEL_SIZE);
 333                 buf[SB_LABEL_SIZE + 1] = '\0';
 334                 strcat(buf, "\n");
 335                 return strlen(buf);
 336         }
 337
 338         return 0;
 339 }
 340
 341 STORE(__bch_flash_dev)
 342 {
 343         struct bcache_device *d = container_of(kobj, struct bcache_device,
 344                                                kobj);
 345         struct uuid_entry *u = &d->c->uuids[d->id];
 346
 347         sysfs_strtoul(data_csum,        d->data_csum);
 348
 349         if (attr == &sysfs_size) {
 350                 uint64_t v;
 351                 strtoi_h_or_return(buf, v);
 352
 353                 u->sectors = v >> 9;
 354                 bch_uuid_write(d->c);
 355                 set_capacity(d->disk, u->sectors);
 356         }
 357
 358         if (attr == &sysfs_label) {
 359                 memcpy(u->label, buf, SB_LABEL_SIZE);
 360                 bch_uuid_write(d->c);
 361         }
 362
 363         if (attr == &sysfs_unregister) {
 364                 atomic_set(&d->detaching, 1);
 365                 bcache_device_stop(d);
 366         }
 367
 368         return size;
 369 }
 370 STORE_LOCKED(bch_flash_dev)
 371
 372 static struct attribute *bch_flash_dev_files[] = {
 373         &sysfs_unregister,
 374 #if 0
 375         &sysfs_data_csum,
 376 #endif
 377         &sysfs_label,
 378         &sysfs_size,
 379         NULL
 380 };
 381 KTYPE(bch_flash_dev);
 382
 383 SHOW(__bch_cache_set)
 384 {
 385         unsigned root_usage(struct cache_set *c)
 386         {
 387                 unsigned bytes = 0;
 388                 struct bkey *k;
 389                 struct btree *b;
 390                 struct btree_iter iter;
 391
 392                 goto lock_root;
 393
 394                 do {
 395                         rw_unlock(false, b);
 396 lock_root:
 397                         b = c->root;
 398                         rw_lock(false, b, b->level);
 399                 } while (b != c->root);
 400
 401                 for_each_key_filter(b, k, &iter, bch_ptr_bad)
 402                         bytes += bkey_bytes(k);
 403
 404                 rw_unlock(false, b);
 405
 406                 return (bytes * 100) / btree_bytes(c);
 407         }
 408
 409         size_t cache_size(struct cache_set *c)
 410         {
 411                 size_t ret = 0;
 412                 struct btree *b;
 413
 414                 mutex_lock(&c->bucket_lock);
 415                 list_for_each_entry(b, &c->btree_cache, list)
 416                         ret += 1 << (b->page_order + PAGE_SHIFT);
 417
 418                 mutex_unlock(&c->bucket_lock);
 419                 return ret;
 420         }
 421
 422         unsigned cache_max_chain(struct cache_set *c)
 423         {
 424                 unsigned ret = 0;
 425                 struct hlist_head *h;
 426
 427                 mutex_lock(&c->bucket_lock);
 428
 429                 for (h = c->bucket_hash;
 430                      h < c->bucket_hash + (1 << BUCKET_HASH_BITS);
 431                      h++) {
 432                         unsigned i = 0;
 433                         struct hlist_node *p;
 434
 435                         hlist_for_each(p, h)
 436                                 i++;
 437
 438                         ret = max(ret, i);
 439                 }
 440
 441                 mutex_unlock(&c->bucket_lock);
 442                 return ret;
 443         }
 444
 445         unsigned btree_used(struct cache_set *c)
 446         {
 447                 return div64_u64(c->gc_stats.key_bytes * 100,
 448                                  (c->gc_stats.nodes ?: 1) * btree_bytes(c));
 449         }
 450
 451         unsigned average_key_size(struct cache_set *c)
 452         {
 453                 return c->gc_stats.nkeys
 454                         ? div64_u64(c->gc_stats.data, c->gc_stats.nkeys)
 455                         : 0;
 456         }
 457
 458         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 459
 460         sysfs_print(synchronous,                CACHE_SYNC(&c->sb));
 461         sysfs_print(journal_delay_ms,           c->journal_delay_ms);
 462         sysfs_hprint(bucket_size,               bucket_bytes(c));
 463         sysfs_hprint(block_size,                block_bytes(c));
 464         sysfs_print(tree_depth,                 c->root->level);
 465         sysfs_print(root_usage_percent,         root_usage(c));
 466
 467         sysfs_hprint(btree_cache_size,          cache_size(c));
 468         sysfs_print(btree_cache_max_chain,      cache_max_chain(c));
 469         sysfs_print(cache_available_percent,    100 - c->gc_stats.in_use);
 470
 471         sysfs_print_time_stats(&c->btree_gc_time,       btree_gc, sec, ms);
 472         sysfs_print_time_stats(&c->btree_split_time,    btree_split, sec, us);
 473         sysfs_print_time_stats(&c->sort_time,           btree_sort, ms, us);
 474         sysfs_print_time_stats(&c->btree_read_time,     btree_read, ms, us);
 475         sysfs_print_time_stats(&c->try_harder_time,     try_harder, ms, us);
 476
 477         sysfs_print(btree_used_percent, btree_used(c));
 478         sysfs_print(btree_nodes,        c->gc_stats.nodes);
 479         sysfs_hprint(dirty_data,        c->gc_stats.dirty);
 480         sysfs_hprint(average_key_size,  average_key_size(c));
 481
 482         sysfs_print(cache_read_races,
 483                     atomic_long_read(&c->cache_read_races));
 484
 485         sysfs_print(writeback_keys_done,
 486                     atomic_long_read(&c->writeback_keys_done));
 487         sysfs_print(writeback_keys_failed,
 488                     atomic_long_read(&c->writeback_keys_failed));
 489
 490         /* See count_io_errors for why 88 */
 491         sysfs_print(io_error_halflife,  c->error_decay * 88);
 492         sysfs_print(io_error_limit,     c->error_limit >> IO_ERROR_SHIFT);
 493
 494         sysfs_hprint(congested,
 495                      ((uint64_t) bch_get_congested(c)) << 9);
 496         sysfs_print(congested_read_threshold_us,
 497                     c->congested_read_threshold_us);
 498         sysfs_print(congested_write_threshold_us,
 499                     c->congested_write_threshold_us);
 500
 501         sysfs_print(active_journal_entries,     fifo_used(&c->journal.pin));
 502         sysfs_printf(verify,                    "%i", c->verify);
 503         sysfs_printf(key_merging_disabled,      "%i", c->key_merging_disabled);
 504         sysfs_printf(gc_always_rewrite,         "%i", c->gc_always_rewrite);
 505         sysfs_printf(btree_shrinker_disabled,   "%i", c->shrinker_disabled);
 506         sysfs_printf(copy_gc_enabled,           "%i", c->copy_gc_enabled);
 507
 508         if (attr == &sysfs_bset_tree_stats)
 509                 return bch_bset_print_stats(c, buf);
 510
 511         return 0;
 512 }
 513 SHOW_LOCKED(bch_cache_set)
 514
 515 STORE(__bch_cache_set)
 516 {
 517         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 518
 519         if (attr == &sysfs_unregister)
 520                 bch_cache_set_unregister(c);
 521
 522         if (attr == &sysfs_stop)
 523                 bch_cache_set_stop(c);
 524
 525         if (attr == &sysfs_synchronous) {
 526                 bool sync = strtoul_or_return(buf);
 527
 528                 if (sync != CACHE_SYNC(&c->sb)) {
 529                         SET_CACHE_SYNC(&c->sb, sync);
 530                         bcache_write_super(c);
 531                 }
 532         }
 533
 534         if (attr == &sysfs_flash_vol_create) {
 535                 int r;
 536                 uint64_t v;
 537                 strtoi_h_or_return(buf, v);
 538
 539                 r = bch_flash_dev_create(c, v);
 540                 if (r)
 541                         return r;
 542         }
 543
 544         if (attr == &sysfs_clear_stats) {
 545                 atomic_long_set(&c->writeback_keys_done,        0);
 546                 atomic_long_set(&c->writeback_keys_failed,      0);
 547
 548                 memset(&c->gc_stats, 0, sizeof(struct gc_stat));
 549                 bch_cache_accounting_clear(&c->accounting);
 550         }
 551
 552         if (attr == &sysfs_trigger_gc)
 553                 bch_queue_gc(c);
 554
 555         if (attr == &sysfs_prune_cache) {
 556                 struct shrink_control sc;
 557                 sc.gfp_mask = GFP_KERNEL;
 558                 sc.nr_to_scan = strtoul_or_return(buf);
 559                 c->shrink.shrink(&c->shrink, &sc);
 560         }
 561
 562         sysfs_strtoul(congested_read_threshold_us,
 563                       c->congested_read_threshold_us);
 564         sysfs_strtoul(congested_write_threshold_us,
 565                       c->congested_write_threshold_us);
 566
 567         if (attr == &sysfs_io_error_limit)
 568                 c->error_limit = strtoul_or_return(buf) << IO_ERROR_SHIFT;
 569
 570         /* See count_io_errors() for why 88 */
 571         if (attr == &sysfs_io_error_halflife)
 572                 c->error_decay = strtoul_or_return(buf) / 88;
 573
 574         sysfs_strtoul(journal_delay_ms,         c->journal_delay_ms);
 575         sysfs_strtoul(verify,                   c->verify);
 576         sysfs_strtoul(key_merging_disabled,     c->key_merging_disabled);
 577         sysfs_strtoul(gc_always_rewrite,        c->gc_always_rewrite);
 578         sysfs_strtoul(btree_shrinker_disabled,  c->shrinker_disabled);
 579         sysfs_strtoul(copy_gc_enabled,          c->copy_gc_enabled);
 580
 581         return size;
 582 }
 583 STORE_LOCKED(bch_cache_set)
 584
 585 SHOW(bch_cache_set_internal)
 586 {
 587         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 588         return bch_cache_set_show(&c->kobj, attr, buf);
 589 }
 590
 591 STORE(bch_cache_set_internal)
 592 {
 593         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 594         return bch_cache_set_store(&c->kobj, attr, buf, size);
 595 }
 596
 597 static void bch_cache_set_internal_release(struct kobject *k)
 598 {
 599 }
 600
 601 static struct attribute *bch_cache_set_files[] = {
 602         &sysfs_unregister,
 603         &sysfs_stop,
 604         &sysfs_synchronous,
 605         &sysfs_journal_delay_ms,
 606         &sysfs_flash_vol_create,
 607
 608         &sysfs_bucket_size,
 609         &sysfs_block_size,
 610         &sysfs_tree_depth,
 611         &sysfs_root_usage_percent,
 612         &sysfs_btree_cache_size,
 613         &sysfs_cache_available_percent,
 614
 615         &sysfs_average_key_size,
 616         &sysfs_dirty_data,
 617
 618         &sysfs_io_error_limit,
 619         &sysfs_io_error_halflife,
 620         &sysfs_congested,
 621         &sysfs_congested_read_threshold_us,
 622         &sysfs_congested_write_threshold_us,
 623         &sysfs_clear_stats,
 624         NULL
 625 };
 626 KTYPE(bch_cache_set);
 627
 628 static struct attribute *bch_cache_set_internal_files[] = {
 629         &sysfs_active_journal_entries,
 630
 631         sysfs_time_stats_attribute_list(btree_gc, sec, ms)
 632         sysfs_time_stats_attribute_list(btree_split, sec, us)
 633         sysfs_time_stats_attribute_list(btree_sort, ms, us)
 634         sysfs_time_stats_attribute_list(btree_read, ms, us)
 635         sysfs_time_stats_attribute_list(try_harder, ms, us)
 636
 637         &sysfs_btree_nodes,
 638         &sysfs_btree_used_percent,
 639         &sysfs_btree_cache_max_chain,
 640
 641         &sysfs_bset_tree_stats,
 642         &sysfs_cache_read_races,
 643         &sysfs_writeback_keys_done,
 644         &sysfs_writeback_keys_failed,
 645
 646         &sysfs_trigger_gc,
 647         &sysfs_prune_cache,
 648 #ifdef CONFIG_BCACHE_DEBUG
 649         &sysfs_verify,
 650         &sysfs_key_merging_disabled,
 651 #endif
 652         &sysfs_gc_always_rewrite,
 653         &sysfs_btree_shrinker_disabled,
 654         &sysfs_copy_gc_enabled,
 655         NULL
 656 };
 657 KTYPE(bch_cache_set_internal);
 658
 659 SHOW(__bch_cache)
 660 {
 661         struct cache *ca = container_of(kobj, struct cache, kobj);
 662
 663         sysfs_hprint(bucket_size,       bucket_bytes(ca));
 664         sysfs_hprint(block_size,        block_bytes(ca));
 665         sysfs_print(nbuckets,           ca->sb.nbuckets);
 666         sysfs_print(discard,            ca->discard);
 667         sysfs_hprint(written, atomic_long_read(&ca->sectors_written) << 9);
 668         sysfs_hprint(btree_written,
 669                      atomic_long_read(&ca->btree_sectors_written) << 9);
 670         sysfs_hprint(metadata_written,
 671                      (atomic_long_read(&ca->meta_sectors_written) +
 672                       atomic_long_read(&ca->btree_sectors_written)) << 9);
 673
 674         sysfs_print(io_errors,
 675                     atomic_read(&ca->io_errors) >> IO_ERROR_SHIFT);
 676
 677         sysfs_print(freelist_percent, ca->free.size * 100 /
 678                     ((size_t) ca->sb.nbuckets));
 679
 680         if (attr == &sysfs_cache_replacement_policy)
 681                 return bch_snprint_string_list(buf, PAGE_SIZE,
 682                                                cache_replacement_policies,
 683                                                CACHE_REPLACEMENT(&ca->sb));
 684
 685         if (attr == &sysfs_priority_stats) {
 686                 int cmp(const void *l, const void *r)
 687                 {       return *((uint16_t *) r) - *((uint16_t *) l); }
 688
 689                 size_t n = ca->sb.nbuckets, i, unused, btree;
 690                 uint64_t sum = 0;
 691                 /* Compute 31 quantiles */
 692                 uint16_t q[31], *p, *cached;
 693                 ssize_t ret;
 694
 695                 cached = p = vmalloc(ca->sb.nbuckets * sizeof(uint16_t));
 696                 if (!p)
 697                         return -ENOMEM;
 698
 699                 mutex_lock(&ca->set->bucket_lock);
 700                 for (i = ca->sb.first_bucket; i < n; i++)
 701                         p[i] = ca->buckets[i].prio;
 702                 mutex_unlock(&ca->set->bucket_lock);
 703
 704                 sort(p, n, sizeof(uint16_t), cmp, NULL);
 705
 706                 while (n &&
 707                        !cached[n - 1])
 708                         --n;
 709
 710                 unused = ca->sb.nbuckets - n;
 711
 712                 while (cached < p + n &&
 713                        *cached == BTREE_PRIO)
 714                         cached++;
 715
 716                 btree = cached - p;
 717                 n -= btree;
 718
 719                 for (i = 0; i < n; i++)
 720                         sum += INITIAL_PRIO - cached[i];
 721
 722                 if (n)
 723                         do_div(sum, n);
 724
 725                 for (i = 0; i < ARRAY_SIZE(q); i++)
 726                         q[i] = INITIAL_PRIO - cached[n * (i + 1) /
 727                                 (ARRAY_SIZE(q) + 1)];
 728
 729                 vfree(p);
 730
 731                 ret = scnprintf(buf, PAGE_SIZE,
 732                                 "Unused:                %zu%%\n"
 733                                 "Metadata:      %zu%%\n"
 734                                 "Average:       %llu\n"
 735                                 "Sectors per Q: %zu\n"
 736                                 "Quantiles:     [",
 737                                 unused * 100 / (size_t) ca->sb.nbuckets,
 738                                 btree * 100 / (size_t) ca->sb.nbuckets, sum,
 739                                 n * ca->sb.bucket_size / (ARRAY_SIZE(q) + 1));
 740
 741                 for (i = 0; i < ARRAY_SIZE(q); i++)
 742                         ret += scnprintf(buf + ret, PAGE_SIZE - ret,
 743                                          "%u ", q[i]);
 744                 ret--;
 745
 746                 ret += scnprintf(buf + ret, PAGE_SIZE - ret, "]\n");
 747
 748                 return ret;
 749         }
 750
 751         return 0;
 752 }
 753 SHOW_LOCKED(bch_cache)
 754
 755 STORE(__bch_cache)
 756 {
 757         struct cache *ca = container_of(kobj, struct cache, kobj);
 758
 759         if (attr == &sysfs_discard) {
 760                 bool v = strtoul_or_return(buf);
 761
 762                 if (blk_queue_discard(bdev_get_queue(ca->bdev)))
 763                         ca->discard = v;
 764
 765                 if (v != CACHE_DISCARD(&ca->sb)) {
 766                         SET_CACHE_DISCARD(&ca->sb, v);
 767                         bcache_write_super(ca->set);
 768                 }
 769         }
 770
 771         if (attr == &sysfs_cache_replacement_policy) {
 772                 ssize_t v = bch_read_string_list(buf, cache_replacement_policies);
 773
 774                 if (v < 0)
 775                         return v;
 776
 777                 if ((unsigned) v != CACHE_REPLACEMENT(&ca->sb)) {
 778                         mutex_lock(&ca->set->bucket_lock);
 779                         SET_CACHE_REPLACEMENT(&ca->sb, v);
 780                         mutex_unlock(&ca->set->bucket_lock);
 781
 782                         bcache_write_super(ca->set);
 783                 }
 784         }
 785
 786         if (attr == &sysfs_freelist_percent) {
 787                 DECLARE_FIFO(long, free);
 788                 long i;
 789                 size_t p = strtoul_or_return(buf);
 790
 791                 p = clamp_t(size_t,
 792                             ((size_t) ca->sb.nbuckets * p) / 100,
 793                             roundup_pow_of_two(ca->sb.nbuckets) >> 9,
 794                             ca->sb.nbuckets / 2);
 795
 796                 if (!init_fifo_exact(&free, p, GFP_KERNEL))
 797                         return -ENOMEM;
 798
 799                 mutex_lock(&ca->set->bucket_lock);
 800
 801                 fifo_move(&free, &ca->free);
 802                 fifo_swap(&free, &ca->free);
 803
 804                 mutex_unlock(&ca->set->bucket_lock);
 805
 806                 while (fifo_pop(&free, i))
 807                         atomic_dec(&ca->buckets[i].pin);
 808
 809                 free_fifo(&free);
 810         }
 811
 812         if (attr == &sysfs_clear_stats) {
 813                 atomic_long_set(&ca->sectors_written, 0);
 814                 atomic_long_set(&ca->btree_sectors_written, 0);
 815                 atomic_long_set(&ca->meta_sectors_written, 0);
 816                 atomic_set(&ca->io_count, 0);
 817                 atomic_set(&ca->io_errors, 0);
 818         }
 819
 820         return size;
 821 }
 822 STORE_LOCKED(bch_cache)
 823
 824 static struct attribute *bch_cache_files[] = {
 825         &sysfs_bucket_size,
 826         &sysfs_block_size,
 827         &sysfs_nbuckets,
 828         &sysfs_priority_stats,
 829         &sysfs_discard,
 830         &sysfs_written,
 831         &sysfs_btree_written,
 832         &sysfs_metadata_written,
 833         &sysfs_io_errors,
 834         &sysfs_clear_stats,
 835         &sysfs_freelist_percent,
 836         &sysfs_cache_replacement_policy,
 837         NULL
 838 };
 839 KTYPE(bch_cache);