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 static const char * const error_actions[] = {
  25         "unregister",
  26         "panic",
  27         NULL
  28 };
  29
  30 write_attribute(attach);
  31 write_attribute(detach);
  32 write_attribute(unregister);
  33 write_attribute(stop);
  34 write_attribute(clear_stats);
  35 write_attribute(trigger_gc);
  36 write_attribute(prune_cache);
  37 write_attribute(flash_vol_create);
  38
  39 read_attribute(bucket_size);
  40 read_attribute(block_size);
  41 read_attribute(nbuckets);
  42 read_attribute(tree_depth);
  43 read_attribute(root_usage_percent);
  44 read_attribute(priority_stats);
  45 read_attribute(btree_cache_size);
  46 read_attribute(btree_cache_max_chain);
  47 read_attribute(cache_available_percent);
  48 read_attribute(written);
  49 read_attribute(btree_written);
  50 read_attribute(metadata_written);
  51 read_attribute(active_journal_entries);
  52
  53 sysfs_time_stats_attribute(btree_gc,    sec, ms);
  54 sysfs_time_stats_attribute(btree_split, sec, us);
  55 sysfs_time_stats_attribute(btree_sort,  ms,  us);
  56 sysfs_time_stats_attribute(btree_read,  ms,  us);
  57 sysfs_time_stats_attribute(try_harder,  ms,  us);
  58
  59 read_attribute(btree_nodes);
  60 read_attribute(btree_used_percent);
  61 read_attribute(average_key_size);
  62 read_attribute(dirty_data);
  63 read_attribute(bset_tree_stats);
  64
  65 read_attribute(state);
  66 read_attribute(cache_read_races);
  67 read_attribute(writeback_keys_done);
  68 read_attribute(writeback_keys_failed);
  69 read_attribute(io_errors);
  70 read_attribute(congested);
  71 rw_attribute(congested_read_threshold_us);
  72 rw_attribute(congested_write_threshold_us);
  73
  74 rw_attribute(sequential_cutoff);
  75 rw_attribute(data_csum);
  76 rw_attribute(cache_mode);
  77 rw_attribute(writeback_metadata);
  78 rw_attribute(writeback_running);
  79 rw_attribute(writeback_percent);
  80 rw_attribute(writeback_delay);
  81 rw_attribute(writeback_rate);
  82
  83 rw_attribute(writeback_rate_update_seconds);
  84 rw_attribute(writeback_rate_d_term);
  85 rw_attribute(writeback_rate_p_term_inverse);
  86 rw_attribute(writeback_rate_d_smooth);
  87 read_attribute(writeback_rate_debug);
  88
  89 read_attribute(stripe_size);
  90 read_attribute(partial_stripes_expensive);
  91
  92 rw_attribute(synchronous);
  93 rw_attribute(journal_delay_ms);
  94 rw_attribute(discard);
  95 rw_attribute(running);
  96 rw_attribute(label);
  97 rw_attribute(readahead);
  98 rw_attribute(errors);
  99 rw_attribute(io_error_limit);
 100 rw_attribute(io_error_halflife);
 101 rw_attribute(verify);
 102 rw_attribute(bypass_torture_test);
 103 rw_attribute(key_merging_disabled);
 104 rw_attribute(gc_always_rewrite);
 105 rw_attribute(expensive_debug_checks);
 106 rw_attribute(freelist_percent);
 107 rw_attribute(cache_replacement_policy);
 108 rw_attribute(btree_shrinker_disabled);
 109 rw_attribute(copy_gc_enabled);
 110 rw_attribute(size);
 111
 112 SHOW(__bch_cached_dev)
 113 {
 114         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 115                                              disk.kobj);
 116         const char *states[] = { "no cache", "clean", "dirty", "inconsistent" };
 117
 118 #define var(stat)               (dc->stat)
 119
 120         if (attr == &sysfs_cache_mode)
 121                 return bch_snprint_string_list(buf, PAGE_SIZE,
 122                                                bch_cache_modes + 1,
 123                                                BDEV_CACHE_MODE(&dc->sb));
 124
 125         sysfs_printf(data_csum,         "%i", dc->disk.data_csum);
 126         var_printf(verify,              "%i");
 127         var_printf(bypass_torture_test, "%i");
 128         var_printf(writeback_metadata,  "%i");
 129         var_printf(writeback_running,   "%i");
 130         var_print(writeback_delay);
 131         var_print(writeback_percent);
 132         sysfs_print(writeback_rate,     dc->writeback_rate.rate);
 133
 134         var_print(writeback_rate_update_seconds);
 135         var_print(writeback_rate_d_term);
 136         var_print(writeback_rate_p_term_inverse);
 137         var_print(writeback_rate_d_smooth);
 138
 139         if (attr == &sysfs_writeback_rate_debug) {
 140                 char dirty[20];
 141                 char derivative[20];
 142                 char target[20];
 143                 bch_hprint(dirty,
 144                            bcache_dev_sectors_dirty(&dc->disk) << 9);
 145                 bch_hprint(derivative,  dc->writeback_rate_derivative << 9);
 146                 bch_hprint(target,      dc->writeback_rate_target << 9);
 147
 148                 return sprintf(buf,
 149                                "rate:\t\t%u\n"
 150                                "change:\t\t%i\n"
 151                                "dirty:\t\t%s\n"
 152                                "derivative:\t%s\n"
 153                                "target:\t\t%s\n",
 154                                dc->writeback_rate.rate,
 155                                dc->writeback_rate_change,
 156                                dirty, derivative, target);
 157         }
 158
 159         sysfs_hprint(dirty_data,
 160                      bcache_dev_sectors_dirty(&dc->disk) << 9);
 161
 162         sysfs_hprint(stripe_size,       dc->disk.stripe_size << 9);
 163         var_printf(partial_stripes_expensive,   "%u");
 164
 165         var_hprint(sequential_cutoff);
 166         var_hprint(readahead);
 167
 168         sysfs_print(running,            atomic_read(&dc->running));
 169         sysfs_print(state,              states[BDEV_STATE(&dc->sb)]);
 170
 171         if (attr == &sysfs_label) {
 172                 memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
 173                 buf[SB_LABEL_SIZE + 1] = '\0';
 174                 strcat(buf, "\n");
 175                 return strlen(buf);
 176         }
 177
 178 #undef var
 179         return 0;
 180 }
 181 SHOW_LOCKED(bch_cached_dev)
 182
 183 STORE(__cached_dev)
 184 {
 185         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 186                                              disk.kobj);
 187         unsigned v = size;
 188         struct cache_set *c;
 189         struct kobj_uevent_env *env;
 190
 191 #define d_strtoul(var)          sysfs_strtoul(var, dc->var)
 192 #define d_strtoi_h(var)         sysfs_hatoi(var, dc->var)
 193
 194         sysfs_strtoul(data_csum,        dc->disk.data_csum);
 195         d_strtoul(verify);
 196         d_strtoul(bypass_torture_test);
 197         d_strtoul(writeback_metadata);
 198         d_strtoul(writeback_running);
 199         d_strtoul(writeback_delay);
 200         sysfs_strtoul_clamp(writeback_rate,
 201                             dc->writeback_rate.rate, 1, 1000000);
 202         sysfs_strtoul_clamp(writeback_percent, dc->writeback_percent, 0, 40);
 203
 204         d_strtoul(writeback_rate_update_seconds);
 205         d_strtoul(writeback_rate_d_term);
 206         d_strtoul(writeback_rate_p_term_inverse);
 207         sysfs_strtoul_clamp(writeback_rate_p_term_inverse,
 208                             dc->writeback_rate_p_term_inverse, 1, INT_MAX);
 209         d_strtoul(writeback_rate_d_smooth);
 210
 211         d_strtoi_h(sequential_cutoff);
 212         d_strtoi_h(readahead);
 213
 214         if (attr == &sysfs_clear_stats)
 215                 bch_cache_accounting_clear(&dc->accounting);
 216
 217         if (attr == &sysfs_running &&
 218             strtoul_or_return(buf))
 219                 bch_cached_dev_run(dc);
 220
 221         if (attr == &sysfs_cache_mode) {
 222                 ssize_t v = bch_read_string_list(buf, bch_cache_modes + 1);
 223
 224                 if (v < 0)
 225                         return v;
 226
 227                 if ((unsigned) v != BDEV_CACHE_MODE(&dc->sb)) {
 228                         SET_BDEV_CACHE_MODE(&dc->sb, v);
 229                         bch_write_bdev_super(dc, NULL);
 230                 }
 231         }
 232
 233         if (attr == &sysfs_label) {
 234                 if (size > SB_LABEL_SIZE)
 235                         return -EINVAL;
 236                 memcpy(dc->sb.label, buf, size);
 237                 if (size < SB_LABEL_SIZE)
 238                         dc->sb.label[size] = '\0';
 239                 if (size && dc->sb.label[size - 1] == '\n')
 240                         dc->sb.label[size - 1] = '\0';
 241                 bch_write_bdev_super(dc, NULL);
 242                 if (dc->disk.c) {
 243                         memcpy(dc->disk.c->uuids[dc->disk.id].label,
 244                                buf, SB_LABEL_SIZE);
 245                         bch_uuid_write(dc->disk.c);
 246                 }
 247                 env = kzalloc(sizeof(struct kobj_uevent_env), GFP_KERNEL);
 248                 if (!env)
 249                         return -ENOMEM;
 250                 add_uevent_var(env, "DRIVER=bcache");
 251                 add_uevent_var(env, "CACHED_UUID=%pU", dc->sb.uuid),
 252                 add_uevent_var(env, "CACHED_LABEL=%s", buf);
 253                 kobject_uevent_env(
 254                         &disk_to_dev(dc->disk.disk)->kobj, KOBJ_CHANGE, env->envp);
 255                 kfree(env);
 256         }
 257
 258         if (attr == &sysfs_attach) {
 259                 if (bch_parse_uuid(buf, dc->sb.set_uuid) < 16)
 260                         return -EINVAL;
 261
 262                 list_for_each_entry(c, &bch_cache_sets, list) {
 263                         v = bch_cached_dev_attach(dc, c);
 264                         if (!v)
 265                                 return size;
 266                 }
 267
 268                 pr_err("Can't attach %s: cache set not found", buf);
 269                 size = v;
 270         }
 271
 272         if (attr == &sysfs_detach && dc->disk.c)
 273                 bch_cached_dev_detach(dc);
 274
 275         if (attr == &sysfs_stop)
 276                 bcache_device_stop(&dc->disk);
 277
 278         return size;
 279 }
 280
 281 STORE(bch_cached_dev)
 282 {
 283         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 284                                              disk.kobj);
 285
 286         mutex_lock(&bch_register_lock);
 287         size = __cached_dev_store(kobj, attr, buf, size);
 288
 289         if (attr == &sysfs_writeback_running)
 290                 bch_writeback_queue(dc);
 291
 292         if (attr == &sysfs_writeback_percent)
 293                 schedule_delayed_work(&dc->writeback_rate_update,
 294                                       dc->writeback_rate_update_seconds * HZ);
 295
 296         mutex_unlock(&bch_register_lock);
 297         return size;
 298 }
 299
 300 static struct attribute *bch_cached_dev_files[] = {
 301         &sysfs_attach,
 302         &sysfs_detach,
 303         &sysfs_stop,
 304 #if 0
 305         &sysfs_data_csum,
 306 #endif
 307         &sysfs_cache_mode,
 308         &sysfs_writeback_metadata,
 309         &sysfs_writeback_running,
 310         &sysfs_writeback_delay,
 311         &sysfs_writeback_percent,
 312         &sysfs_writeback_rate,
 313         &sysfs_writeback_rate_update_seconds,
 314         &sysfs_writeback_rate_d_term,
 315         &sysfs_writeback_rate_p_term_inverse,
 316         &sysfs_writeback_rate_d_smooth,
 317         &sysfs_writeback_rate_debug,
 318         &sysfs_dirty_data,
 319         &sysfs_stripe_size,
 320         &sysfs_partial_stripes_expensive,
 321         &sysfs_sequential_cutoff,
 322         &sysfs_clear_stats,
 323         &sysfs_running,
 324         &sysfs_state,
 325         &sysfs_label,
 326         &sysfs_readahead,
 327 #ifdef CONFIG_BCACHE_DEBUG
 328         &sysfs_verify,
 329         &sysfs_bypass_torture_test,
 330 #endif
 331         NULL
 332 };
 333 KTYPE(bch_cached_dev);
 334
 335 SHOW(bch_flash_dev)
 336 {
 337         struct bcache_device *d = container_of(kobj, struct bcache_device,
 338                                                kobj);
 339         struct uuid_entry *u = &d->c->uuids[d->id];
 340
 341         sysfs_printf(data_csum, "%i", d->data_csum);
 342         sysfs_hprint(size,      u->sectors << 9);
 343
 344         if (attr == &sysfs_label) {
 345                 memcpy(buf, u->label, SB_LABEL_SIZE);
 346                 buf[SB_LABEL_SIZE + 1] = '\0';
 347                 strcat(buf, "\n");
 348                 return strlen(buf);
 349         }
 350
 351         return 0;
 352 }
 353
 354 STORE(__bch_flash_dev)
 355 {
 356         struct bcache_device *d = container_of(kobj, struct bcache_device,
 357                                                kobj);
 358         struct uuid_entry *u = &d->c->uuids[d->id];
 359
 360         sysfs_strtoul(data_csum,        d->data_csum);
 361
 362         if (attr == &sysfs_size) {
 363                 uint64_t v;
 364                 strtoi_h_or_return(buf, v);
 365
 366                 u->sectors = v >> 9;
 367                 bch_uuid_write(d->c);
 368                 set_capacity(d->disk, u->sectors);
 369         }
 370
 371         if (attr == &sysfs_label) {
 372                 memcpy(u->label, buf, SB_LABEL_SIZE);
 373                 bch_uuid_write(d->c);
 374         }
 375
 376         if (attr == &sysfs_unregister) {
 377                 set_bit(BCACHE_DEV_DETACHING, &d->flags);
 378                 bcache_device_stop(d);
 379         }
 380
 381         return size;
 382 }
 383 STORE_LOCKED(bch_flash_dev)
 384
 385 static struct attribute *bch_flash_dev_files[] = {
 386         &sysfs_unregister,
 387 #if 0
 388         &sysfs_data_csum,
 389 #endif
 390         &sysfs_label,
 391         &sysfs_size,
 392         NULL
 393 };
 394 KTYPE(bch_flash_dev);
 395
 396 SHOW(__bch_cache_set)
 397 {
 398         unsigned root_usage(struct cache_set *c)
 399         {
 400                 unsigned bytes = 0;
 401                 struct bkey *k;
 402                 struct btree *b;
 403                 struct btree_iter iter;
 404
 405                 goto lock_root;
 406
 407                 do {
 408                         rw_unlock(false, b);
 409 lock_root:
 410                         b = c->root;
 411                         rw_lock(false, b, b->level);
 412                 } while (b != c->root);
 413
 414                 for_each_key_filter(b, k, &iter, bch_ptr_bad)
 415                         bytes += bkey_bytes(k);
 416
 417                 rw_unlock(false, b);
 418
 419                 return (bytes * 100) / btree_bytes(c);
 420         }
 421
 422         size_t cache_size(struct cache_set *c)
 423         {
 424                 size_t ret = 0;
 425                 struct btree *b;
 426
 427                 mutex_lock(&c->bucket_lock);
 428                 list_for_each_entry(b, &c->btree_cache, list)
 429                         ret += 1 << (b->page_order + PAGE_SHIFT);
 430
 431                 mutex_unlock(&c->bucket_lock);
 432                 return ret;
 433         }
 434
 435         unsigned cache_max_chain(struct cache_set *c)
 436         {
 437                 unsigned ret = 0;
 438                 struct hlist_head *h;
 439
 440                 mutex_lock(&c->bucket_lock);
 441
 442                 for (h = c->bucket_hash;
 443                      h < c->bucket_hash + (1 << BUCKET_HASH_BITS);
 444                      h++) {
 445                         unsigned i = 0;
 446                         struct hlist_node *p;
 447
 448                         hlist_for_each(p, h)
 449                                 i++;
 450
 451                         ret = max(ret, i);
 452                 }
 453
 454                 mutex_unlock(&c->bucket_lock);
 455                 return ret;
 456         }
 457
 458         unsigned btree_used(struct cache_set *c)
 459         {
 460                 return div64_u64(c->gc_stats.key_bytes * 100,
 461                                  (c->gc_stats.nodes ?: 1) * btree_bytes(c));
 462         }
 463
 464         unsigned average_key_size(struct cache_set *c)
 465         {
 466                 return c->gc_stats.nkeys
 467                         ? div64_u64(c->gc_stats.data, c->gc_stats.nkeys)
 468                         : 0;
 469         }
 470
 471         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 472
 473         sysfs_print(synchronous,                CACHE_SYNC(&c->sb));
 474         sysfs_print(journal_delay_ms,           c->journal_delay_ms);
 475         sysfs_hprint(bucket_size,               bucket_bytes(c));
 476         sysfs_hprint(block_size,                block_bytes(c));
 477         sysfs_print(tree_depth,                 c->root->level);
 478         sysfs_print(root_usage_percent,         root_usage(c));
 479
 480         sysfs_hprint(btree_cache_size,          cache_size(c));
 481         sysfs_print(btree_cache_max_chain,      cache_max_chain(c));
 482         sysfs_print(cache_available_percent,    100 - c->gc_stats.in_use);
 483
 484         sysfs_print_time_stats(&c->btree_gc_time,       btree_gc, sec, ms);
 485         sysfs_print_time_stats(&c->btree_split_time,    btree_split, sec, us);
 486         sysfs_print_time_stats(&c->sort_time,           btree_sort, ms, us);
 487         sysfs_print_time_stats(&c->btree_read_time,     btree_read, ms, us);
 488         sysfs_print_time_stats(&c->try_harder_time,     try_harder, ms, us);
 489
 490         sysfs_print(btree_used_percent, btree_used(c));
 491         sysfs_print(btree_nodes,        c->gc_stats.nodes);
 492         sysfs_hprint(average_key_size,  average_key_size(c));
 493
 494         sysfs_print(cache_read_races,
 495                     atomic_long_read(&c->cache_read_races));
 496
 497         sysfs_print(writeback_keys_done,
 498                     atomic_long_read(&c->writeback_keys_done));
 499         sysfs_print(writeback_keys_failed,
 500                     atomic_long_read(&c->writeback_keys_failed));
 501
 502         if (attr == &sysfs_errors)
 503                 return bch_snprint_string_list(buf, PAGE_SIZE, error_actions,
 504                                                c->on_error);
 505
 506         /* See count_io_errors for why 88 */
 507         sysfs_print(io_error_halflife,  c->error_decay * 88);
 508         sysfs_print(io_error_limit,     c->error_limit >> IO_ERROR_SHIFT);
 509
 510         sysfs_hprint(congested,
 511                      ((uint64_t) bch_get_congested(c)) << 9);
 512         sysfs_print(congested_read_threshold_us,
 513                     c->congested_read_threshold_us);
 514         sysfs_print(congested_write_threshold_us,
 515                     c->congested_write_threshold_us);
 516
 517         sysfs_print(active_journal_entries,     fifo_used(&c->journal.pin));
 518         sysfs_printf(verify,                    "%i", c->verify);
 519         sysfs_printf(key_merging_disabled,      "%i", c->key_merging_disabled);
 520         sysfs_printf(expensive_debug_checks,
 521                      "%i", c->expensive_debug_checks);
 522         sysfs_printf(gc_always_rewrite,         "%i", c->gc_always_rewrite);
 523         sysfs_printf(btree_shrinker_disabled,   "%i", c->shrinker_disabled);
 524         sysfs_printf(copy_gc_enabled,           "%i", c->copy_gc_enabled);
 525
 526         if (attr == &sysfs_bset_tree_stats)
 527                 return bch_bset_print_stats(c, buf);
 528
 529         return 0;
 530 }
 531 SHOW_LOCKED(bch_cache_set)
 532
 533 STORE(__bch_cache_set)
 534 {
 535         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 536
 537         if (attr == &sysfs_unregister)
 538                 bch_cache_set_unregister(c);
 539
 540         if (attr == &sysfs_stop)
 541                 bch_cache_set_stop(c);
 542
 543         if (attr == &sysfs_synchronous) {
 544                 bool sync = strtoul_or_return(buf);
 545
 546                 if (sync != CACHE_SYNC(&c->sb)) {
 547                         SET_CACHE_SYNC(&c->sb, sync);
 548                         bcache_write_super(c);
 549                 }
 550         }
 551
 552         if (attr == &sysfs_flash_vol_create) {
 553                 int r;
 554                 uint64_t v;
 555                 strtoi_h_or_return(buf, v);
 556
 557                 r = bch_flash_dev_create(c, v);
 558                 if (r)
 559                         return r;
 560         }
 561
 562         if (attr == &sysfs_clear_stats) {
 563                 atomic_long_set(&c->writeback_keys_done,        0);
 564                 atomic_long_set(&c->writeback_keys_failed,      0);
 565
 566                 memset(&c->gc_stats, 0, sizeof(struct gc_stat));
 567                 bch_cache_accounting_clear(&c->accounting);
 568         }
 569
 570         if (attr == &sysfs_trigger_gc)
 571                 wake_up_gc(c);
 572
 573         if (attr == &sysfs_prune_cache) {
 574                 struct shrink_control sc;
 575                 sc.gfp_mask = GFP_KERNEL;
 576                 sc.nr_to_scan = strtoul_or_return(buf);
 577                 c->shrink.scan_objects(&c->shrink, &sc);
 578         }
 579
 580         sysfs_strtoul(congested_read_threshold_us,
 581                       c->congested_read_threshold_us);
 582         sysfs_strtoul(congested_write_threshold_us,
 583                       c->congested_write_threshold_us);
 584
 585         if (attr == &sysfs_errors) {
 586                 ssize_t v = bch_read_string_list(buf, error_actions);
 587
 588                 if (v < 0)
 589                         return v;
 590
 591                 c->on_error = v;
 592         }
 593
 594         if (attr == &sysfs_io_error_limit)
 595                 c->error_limit = strtoul_or_return(buf) << IO_ERROR_SHIFT;
 596
 597         /* See count_io_errors() for why 88 */
 598         if (attr == &sysfs_io_error_halflife)
 599                 c->error_decay = strtoul_or_return(buf) / 88;
 600
 601         sysfs_strtoul(journal_delay_ms,         c->journal_delay_ms);
 602         sysfs_strtoul(verify,                   c->verify);
 603         sysfs_strtoul(key_merging_disabled,     c->key_merging_disabled);
 604         sysfs_strtoul(expensive_debug_checks,   c->expensive_debug_checks);
 605         sysfs_strtoul(gc_always_rewrite,        c->gc_always_rewrite);
 606         sysfs_strtoul(btree_shrinker_disabled,  c->shrinker_disabled);
 607         sysfs_strtoul(copy_gc_enabled,          c->copy_gc_enabled);
 608
 609         return size;
 610 }
 611 STORE_LOCKED(bch_cache_set)
 612
 613 SHOW(bch_cache_set_internal)
 614 {
 615         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 616         return bch_cache_set_show(&c->kobj, attr, buf);
 617 }
 618
 619 STORE(bch_cache_set_internal)
 620 {
 621         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 622         return bch_cache_set_store(&c->kobj, attr, buf, size);
 623 }
 624
 625 static void bch_cache_set_internal_release(struct kobject *k)
 626 {
 627 }
 628
 629 static struct attribute *bch_cache_set_files[] = {
 630         &sysfs_unregister,
 631         &sysfs_stop,
 632         &sysfs_synchronous,
 633         &sysfs_journal_delay_ms,
 634         &sysfs_flash_vol_create,
 635
 636         &sysfs_bucket_size,
 637         &sysfs_block_size,
 638         &sysfs_tree_depth,
 639         &sysfs_root_usage_percent,
 640         &sysfs_btree_cache_size,
 641         &sysfs_cache_available_percent,
 642
 643         &sysfs_average_key_size,
 644
 645         &sysfs_errors,
 646         &sysfs_io_error_limit,
 647         &sysfs_io_error_halflife,
 648         &sysfs_congested,
 649         &sysfs_congested_read_threshold_us,
 650         &sysfs_congested_write_threshold_us,
 651         &sysfs_clear_stats,
 652         NULL
 653 };
 654 KTYPE(bch_cache_set);
 655
 656 static struct attribute *bch_cache_set_internal_files[] = {
 657         &sysfs_active_journal_entries,
 658
 659         sysfs_time_stats_attribute_list(btree_gc, sec, ms)
 660         sysfs_time_stats_attribute_list(btree_split, sec, us)
 661         sysfs_time_stats_attribute_list(btree_sort, ms, us)
 662         sysfs_time_stats_attribute_list(btree_read, ms, us)
 663         sysfs_time_stats_attribute_list(try_harder, ms, us)
 664
 665         &sysfs_btree_nodes,
 666         &sysfs_btree_used_percent,
 667         &sysfs_btree_cache_max_chain,
 668
 669         &sysfs_bset_tree_stats,
 670         &sysfs_cache_read_races,
 671         &sysfs_writeback_keys_done,
 672         &sysfs_writeback_keys_failed,
 673
 674         &sysfs_trigger_gc,
 675         &sysfs_prune_cache,
 676 #ifdef CONFIG_BCACHE_DEBUG
 677         &sysfs_verify,
 678         &sysfs_key_merging_disabled,
 679         &sysfs_expensive_debug_checks,
 680 #endif
 681         &sysfs_gc_always_rewrite,
 682         &sysfs_btree_shrinker_disabled,
 683         &sysfs_copy_gc_enabled,
 684         NULL
 685 };
 686 KTYPE(bch_cache_set_internal);
 687
 688 SHOW(__bch_cache)
 689 {
 690         struct cache *ca = container_of(kobj, struct cache, kobj);
 691
 692         sysfs_hprint(bucket_size,       bucket_bytes(ca));
 693         sysfs_hprint(block_size,        block_bytes(ca));
 694         sysfs_print(nbuckets,           ca->sb.nbuckets);
 695         sysfs_print(discard,            ca->discard);
 696         sysfs_hprint(written, atomic_long_read(&ca->sectors_written) << 9);
 697         sysfs_hprint(btree_written,
 698                      atomic_long_read(&ca->btree_sectors_written) << 9);
 699         sysfs_hprint(metadata_written,
 700                      (atomic_long_read(&ca->meta_sectors_written) +
 701                       atomic_long_read(&ca->btree_sectors_written)) << 9);
 702
 703         sysfs_print(io_errors,
 704                     atomic_read(&ca->io_errors) >> IO_ERROR_SHIFT);
 705
 706         sysfs_print(freelist_percent, ca->free.size * 100 /
 707                     ((size_t) ca->sb.nbuckets));
 708
 709         if (attr == &sysfs_cache_replacement_policy)
 710                 return bch_snprint_string_list(buf, PAGE_SIZE,
 711                                                cache_replacement_policies,
 712                                                CACHE_REPLACEMENT(&ca->sb));
 713
 714         if (attr == &sysfs_priority_stats) {
 715                 int cmp(const void *l, const void *r)
 716                 {       return *((uint16_t *) r) - *((uint16_t *) l); }
 717
 718                 size_t n = ca->sb.nbuckets, i, unused, btree;
 719                 uint64_t sum = 0;
 720                 /* Compute 31 quantiles */
 721                 uint16_t q[31], *p, *cached;
 722                 ssize_t ret;
 723
 724                 cached = p = vmalloc(ca->sb.nbuckets * sizeof(uint16_t));
 725                 if (!p)
 726                         return -ENOMEM;
 727
 728                 mutex_lock(&ca->set->bucket_lock);
 729                 for (i = ca->sb.first_bucket; i < n; i++)
 730                         p[i] = ca->buckets[i].prio;
 731                 mutex_unlock(&ca->set->bucket_lock);
 732
 733                 sort(p, n, sizeof(uint16_t), cmp, NULL);
 734
 735                 while (n &&
 736                        !cached[n - 1])
 737                         --n;
 738
 739                 unused = ca->sb.nbuckets - n;
 740
 741                 while (cached < p + n &&
 742                        *cached == BTREE_PRIO)
 743                         cached++;
 744
 745                 btree = cached - p;
 746                 n -= btree;
 747
 748                 for (i = 0; i < n; i++)
 749                         sum += INITIAL_PRIO - cached[i];
 750
 751                 if (n)
 752                         do_div(sum, n);
 753
 754                 for (i = 0; i < ARRAY_SIZE(q); i++)
 755                         q[i] = INITIAL_PRIO - cached[n * (i + 1) /
 756                                 (ARRAY_SIZE(q) + 1)];
 757
 758                 vfree(p);
 759
 760                 ret = scnprintf(buf, PAGE_SIZE,
 761                                 "Unused:                %zu%%\n"
 762                                 "Metadata:      %zu%%\n"
 763                                 "Average:       %llu\n"
 764                                 "Sectors per Q: %zu\n"
 765                                 "Quantiles:     [",
 766                                 unused * 100 / (size_t) ca->sb.nbuckets,
 767                                 btree * 100 / (size_t) ca->sb.nbuckets, sum,
 768                                 n * ca->sb.bucket_size / (ARRAY_SIZE(q) + 1));
 769
 770                 for (i = 0; i < ARRAY_SIZE(q); i++)
 771                         ret += scnprintf(buf + ret, PAGE_SIZE - ret,
 772                                          "%u ", q[i]);
 773                 ret--;
 774
 775                 ret += scnprintf(buf + ret, PAGE_SIZE - ret, "]\n");
 776
 777                 return ret;
 778         }
 779
 780         return 0;
 781 }
 782 SHOW_LOCKED(bch_cache)
 783
 784 STORE(__bch_cache)
 785 {
 786         struct cache *ca = container_of(kobj, struct cache, kobj);
 787
 788         if (attr == &sysfs_discard) {
 789                 bool v = strtoul_or_return(buf);
 790
 791                 if (blk_queue_discard(bdev_get_queue(ca->bdev)))
 792                         ca->discard = v;
 793
 794                 if (v != CACHE_DISCARD(&ca->sb)) {
 795                         SET_CACHE_DISCARD(&ca->sb, v);
 796                         bcache_write_super(ca->set);
 797                 }
 798         }
 799
 800         if (attr == &sysfs_cache_replacement_policy) {
 801                 ssize_t v = bch_read_string_list(buf, cache_replacement_policies);
 802
 803                 if (v < 0)
 804                         return v;
 805
 806                 if ((unsigned) v != CACHE_REPLACEMENT(&ca->sb)) {
 807                         mutex_lock(&ca->set->bucket_lock);
 808                         SET_CACHE_REPLACEMENT(&ca->sb, v);
 809                         mutex_unlock(&ca->set->bucket_lock);
 810
 811                         bcache_write_super(ca->set);
 812                 }
 813         }
 814
 815         if (attr == &sysfs_freelist_percent) {
 816                 DECLARE_FIFO(long, free);
 817                 long i;
 818                 size_t p = strtoul_or_return(buf);
 819
 820                 p = clamp_t(size_t,
 821                             ((size_t) ca->sb.nbuckets * p) / 100,
 822                             roundup_pow_of_two(ca->sb.nbuckets) >> 9,
 823                             ca->sb.nbuckets / 2);
 824
 825                 if (!init_fifo_exact(&free, p, GFP_KERNEL))
 826                         return -ENOMEM;
 827
 828                 mutex_lock(&ca->set->bucket_lock);
 829
 830                 fifo_move(&free, &ca->free);
 831                 fifo_swap(&free, &ca->free);
 832
 833                 mutex_unlock(&ca->set->bucket_lock);
 834
 835                 while (fifo_pop(&free, i))
 836                         atomic_dec(&ca->buckets[i].pin);
 837
 838                 free_fifo(&free);
 839         }
 840
 841         if (attr == &sysfs_clear_stats) {
 842                 atomic_long_set(&ca->sectors_written, 0);
 843                 atomic_long_set(&ca->btree_sectors_written, 0);
 844                 atomic_long_set(&ca->meta_sectors_written, 0);
 845                 atomic_set(&ca->io_count, 0);
 846                 atomic_set(&ca->io_errors, 0);
 847         }
 848
 849         return size;
 850 }
 851 STORE_LOCKED(bch_cache)
 852
 853 static struct attribute *bch_cache_files[] = {
 854         &sysfs_bucket_size,
 855         &sysfs_block_size,
 856         &sysfs_nbuckets,
 857         &sysfs_priority_stats,
 858         &sysfs_discard,
 859         &sysfs_written,
 860         &sysfs_btree_written,
 861         &sysfs_metadata_written,
 862         &sysfs_io_errors,
 863         &sysfs_clear_stats,
 864         &sysfs_freelist_percent,
 865         &sysfs_cache_replacement_policy,
 866         NULL
 867 };
 868 KTYPE(bch_cache);