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 read_attribute(writeback_rate_debug);
  87
  88 read_attribute(stripe_size);
  89 read_attribute(partial_stripes_expensive);
  90
  91 rw_attribute(synchronous);
  92 rw_attribute(journal_delay_ms);
  93 rw_attribute(discard);
  94 rw_attribute(running);
  95 rw_attribute(label);
  96 rw_attribute(readahead);
  97 rw_attribute(errors);
  98 rw_attribute(io_error_limit);
  99 rw_attribute(io_error_halflife);
 100 rw_attribute(verify);
 101 rw_attribute(bypass_torture_test);
 102 rw_attribute(key_merging_disabled);
 103 rw_attribute(gc_always_rewrite);
 104 rw_attribute(expensive_debug_checks);
 105 rw_attribute(cache_replacement_policy);
 106 rw_attribute(btree_shrinker_disabled);
 107 rw_attribute(copy_gc_enabled);
 108 rw_attribute(size);
 109
 110 SHOW(__bch_cached_dev)
 111 {
 112         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 113                                              disk.kobj);
 114         const char *states[] = { "no cache", "clean", "dirty", "inconsistent" };
 115
 116 #define var(stat)               (dc->stat)
 117
 118         if (attr == &sysfs_cache_mode)
 119                 return bch_snprint_string_list(buf, PAGE_SIZE,
 120                                                bch_cache_modes + 1,
 121                                                BDEV_CACHE_MODE(&dc->sb));
 122
 123         sysfs_printf(data_csum,         "%i", dc->disk.data_csum);
 124         var_printf(verify,              "%i");
 125         var_printf(bypass_torture_test, "%i");
 126         var_printf(writeback_metadata,  "%i");
 127         var_printf(writeback_running,   "%i");
 128         var_print(writeback_delay);
 129         var_print(writeback_percent);
 130         sysfs_hprint(writeback_rate,    dc->writeback_rate.rate << 9);
 131
 132         var_print(writeback_rate_update_seconds);
 133         var_print(writeback_rate_d_term);
 134         var_print(writeback_rate_p_term_inverse);
 135
 136         if (attr == &sysfs_writeback_rate_debug) {
 137                 char rate[20];
 138                 char dirty[20];
 139                 char target[20];
 140                 char proportional[20];
 141                 char derivative[20];
 142                 char change[20];
 143                 s64 next_io;
 144
 145                 bch_hprint(rate,        dc->writeback_rate.rate << 9);
 146                 bch_hprint(dirty,       bcache_dev_sectors_dirty(&dc->disk) << 9);
 147                 bch_hprint(target,      dc->writeback_rate_target << 9);
 148                 bch_hprint(proportional,dc->writeback_rate_proportional << 9);
 149                 bch_hprint(derivative,  dc->writeback_rate_derivative << 9);
 150                 bch_hprint(change,      dc->writeback_rate_change << 9);
 151
 152                 next_io = div64_s64(dc->writeback_rate.next - local_clock(),
 153                                     NSEC_PER_MSEC);
 154
 155                 return sprintf(buf,
 156                                "rate:\t\t%s/sec\n"
 157                                "dirty:\t\t%s\n"
 158                                "target:\t\t%s\n"
 159                                "proportional:\t%s\n"
 160                                "derivative:\t%s\n"
 161                                "change:\t\t%s/sec\n"
 162                                "next io:\t%llims\n",
 163                                rate, dirty, target, proportional,
 164                                derivative, change, next_io);
 165         }
 166
 167         sysfs_hprint(dirty_data,
 168                      bcache_dev_sectors_dirty(&dc->disk) << 9);
 169
 170         sysfs_hprint(stripe_size,       dc->disk.stripe_size << 9);
 171         var_printf(partial_stripes_expensive,   "%u");
 172
 173         var_hprint(sequential_cutoff);
 174         var_hprint(readahead);
 175
 176         sysfs_print(running,            atomic_read(&dc->running));
 177         sysfs_print(state,              states[BDEV_STATE(&dc->sb)]);
 178
 179         if (attr == &sysfs_label) {
 180                 memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
 181                 buf[SB_LABEL_SIZE + 1] = '\0';
 182                 strcat(buf, "\n");
 183                 return strlen(buf);
 184         }
 185
 186 #undef var
 187         return 0;
 188 }
 189 SHOW_LOCKED(bch_cached_dev)
 190
 191 STORE(__cached_dev)
 192 {
 193         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 194                                              disk.kobj);
 195         unsigned v = size;
 196         struct cache_set *c;
 197         struct kobj_uevent_env *env;
 198
 199 #define d_strtoul(var)          sysfs_strtoul(var, dc->var)
 200 #define d_strtoul_nonzero(var)  sysfs_strtoul_clamp(var, dc->var, 1, INT_MAX)
 201 #define d_strtoi_h(var)         sysfs_hatoi(var, dc->var)
 202
 203         sysfs_strtoul(data_csum,        dc->disk.data_csum);
 204         d_strtoul(verify);
 205         d_strtoul(bypass_torture_test);
 206         d_strtoul(writeback_metadata);
 207         d_strtoul(writeback_running);
 208         d_strtoul(writeback_delay);
 209
 210         sysfs_strtoul_clamp(writeback_percent, dc->writeback_percent, 0, 40);
 211
 212         sysfs_strtoul_clamp(writeback_rate,
 213                             dc->writeback_rate.rate, 1, INT_MAX);
 214
 215         d_strtoul_nonzero(writeback_rate_update_seconds);
 216         d_strtoul(writeback_rate_d_term);
 217         d_strtoul_nonzero(writeback_rate_p_term_inverse);
 218
 219         d_strtoi_h(sequential_cutoff);
 220         d_strtoi_h(readahead);
 221
 222         if (attr == &sysfs_clear_stats)
 223                 bch_cache_accounting_clear(&dc->accounting);
 224
 225         if (attr == &sysfs_running &&
 226             strtoul_or_return(buf))
 227                 bch_cached_dev_run(dc);
 228
 229         if (attr == &sysfs_cache_mode) {
 230                 ssize_t v = bch_read_string_list(buf, bch_cache_modes + 1);
 231
 232                 if (v < 0)
 233                         return v;
 234
 235                 if ((unsigned) v != BDEV_CACHE_MODE(&dc->sb)) {
 236                         SET_BDEV_CACHE_MODE(&dc->sb, v);
 237                         bch_write_bdev_super(dc, NULL);
 238                 }
 239         }
 240
 241         if (attr == &sysfs_label) {
 242                 if (size > SB_LABEL_SIZE)
 243                         return -EINVAL;
 244                 memcpy(dc->sb.label, buf, size);
 245                 if (size < SB_LABEL_SIZE)
 246                         dc->sb.label[size] = '\0';
 247                 if (size && dc->sb.label[size - 1] == '\n')
 248                         dc->sb.label[size - 1] = '\0';
 249                 bch_write_bdev_super(dc, NULL);
 250                 if (dc->disk.c) {
 251                         memcpy(dc->disk.c->uuids[dc->disk.id].label,
 252                                buf, SB_LABEL_SIZE);
 253                         bch_uuid_write(dc->disk.c);
 254                 }
 255                 env = kzalloc(sizeof(struct kobj_uevent_env), GFP_KERNEL);
 256                 if (!env)
 257                         return -ENOMEM;
 258                 add_uevent_var(env, "DRIVER=bcache");
 259                 add_uevent_var(env, "CACHED_UUID=%pU", dc->sb.uuid),
 260                 add_uevent_var(env, "CACHED_LABEL=%s", buf);
 261                 kobject_uevent_env(
 262                         &disk_to_dev(dc->disk.disk)->kobj, KOBJ_CHANGE, env->envp);
 263                 kfree(env);
 264         }
 265
 266         if (attr == &sysfs_attach) {
 267                 if (bch_parse_uuid(buf, dc->sb.set_uuid) < 16)
 268                         return -EINVAL;
 269
 270                 list_for_each_entry(c, &bch_cache_sets, list) {
 271                         v = bch_cached_dev_attach(dc, c);
 272                         if (!v)
 273                                 return size;
 274                 }
 275
 276                 pr_err("Can't attach %s: cache set not found", buf);
 277                 size = v;
 278         }
 279
 280         if (attr == &sysfs_detach && dc->disk.c)
 281                 bch_cached_dev_detach(dc);
 282
 283         if (attr == &sysfs_stop)
 284                 bcache_device_stop(&dc->disk);
 285
 286         return size;
 287 }
 288
 289 STORE(bch_cached_dev)
 290 {
 291         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 292                                              disk.kobj);
 293
 294         mutex_lock(&bch_register_lock);
 295         size = __cached_dev_store(kobj, attr, buf, size);
 296
 297         if (attr == &sysfs_writeback_running)
 298                 bch_writeback_queue(dc);
 299
 300         if (attr == &sysfs_writeback_percent)
 301                 schedule_delayed_work(&dc->writeback_rate_update,
 302                                       dc->writeback_rate_update_seconds * HZ);
 303
 304         mutex_unlock(&bch_register_lock);
 305         return size;
 306 }
 307
 308 static struct attribute *bch_cached_dev_files[] = {
 309         &sysfs_attach,
 310         &sysfs_detach,
 311         &sysfs_stop,
 312 #if 0
 313         &sysfs_data_csum,
 314 #endif
 315         &sysfs_cache_mode,
 316         &sysfs_writeback_metadata,
 317         &sysfs_writeback_running,
 318         &sysfs_writeback_delay,
 319         &sysfs_writeback_percent,
 320         &sysfs_writeback_rate,
 321         &sysfs_writeback_rate_update_seconds,
 322         &sysfs_writeback_rate_d_term,
 323         &sysfs_writeback_rate_p_term_inverse,
 324         &sysfs_writeback_rate_debug,
 325         &sysfs_dirty_data,
 326         &sysfs_stripe_size,
 327         &sysfs_partial_stripes_expensive,
 328         &sysfs_sequential_cutoff,
 329         &sysfs_clear_stats,
 330         &sysfs_running,
 331         &sysfs_state,
 332         &sysfs_label,
 333         &sysfs_readahead,
 334 #ifdef CONFIG_BCACHE_DEBUG
 335         &sysfs_verify,
 336         &sysfs_bypass_torture_test,
 337 #endif
 338         NULL
 339 };
 340 KTYPE(bch_cached_dev);
 341
 342 SHOW(bch_flash_dev)
 343 {
 344         struct bcache_device *d = container_of(kobj, struct bcache_device,
 345                                                kobj);
 346         struct uuid_entry *u = &d->c->uuids[d->id];
 347
 348         sysfs_printf(data_csum, "%i", d->data_csum);
 349         sysfs_hprint(size,      u->sectors << 9);
 350
 351         if (attr == &sysfs_label) {
 352                 memcpy(buf, u->label, SB_LABEL_SIZE);
 353                 buf[SB_LABEL_SIZE + 1] = '\0';
 354                 strcat(buf, "\n");
 355                 return strlen(buf);
 356         }
 357
 358         return 0;
 359 }
 360
 361 STORE(__bch_flash_dev)
 362 {
 363         struct bcache_device *d = container_of(kobj, struct bcache_device,
 364                                                kobj);
 365         struct uuid_entry *u = &d->c->uuids[d->id];
 366
 367         sysfs_strtoul(data_csum,        d->data_csum);
 368
 369         if (attr == &sysfs_size) {
 370                 uint64_t v;
 371                 strtoi_h_or_return(buf, v);
 372
 373                 u->sectors = v >> 9;
 374                 bch_uuid_write(d->c);
 375                 set_capacity(d->disk, u->sectors);
 376         }
 377
 378         if (attr == &sysfs_label) {
 379                 memcpy(u->label, buf, SB_LABEL_SIZE);
 380                 bch_uuid_write(d->c);
 381         }
 382
 383         if (attr == &sysfs_unregister) {
 384                 set_bit(BCACHE_DEV_DETACHING, &d->flags);
 385                 bcache_device_stop(d);
 386         }
 387
 388         return size;
 389 }
 390 STORE_LOCKED(bch_flash_dev)
 391
 392 static struct attribute *bch_flash_dev_files[] = {
 393         &sysfs_unregister,
 394 #if 0
 395         &sysfs_data_csum,
 396 #endif
 397         &sysfs_label,
 398         &sysfs_size,
 399         NULL
 400 };
 401 KTYPE(bch_flash_dev);
 402
 403 struct bset_stats_op {
 404         struct btree_op op;
 405         size_t nodes;
 406         struct bset_stats stats;
 407 };
 408
 409 static int btree_bset_stats(struct btree_op *b_op, struct btree *b)
 410 {
 411         struct bset_stats_op *op = container_of(b_op, struct bset_stats_op, op);
 412
 413         op->nodes++;
 414         bch_btree_keys_stats(&b->keys, &op->stats);
 415
 416         return MAP_CONTINUE;
 417 }
 418
 419 static int bch_bset_print_stats(struct cache_set *c, char *buf)
 420 {
 421         struct bset_stats_op op;
 422         int ret;
 423
 424         memset(&op, 0, sizeof(op));
 425         bch_btree_op_init(&op.op, -1);
 426
 427         ret = bch_btree_map_nodes(&op.op, c, &ZERO_KEY, btree_bset_stats);
 428         if (ret < 0)
 429                 return ret;
 430
 431         return snprintf(buf, PAGE_SIZE,
 432                         "btree nodes:           %zu\n"
 433                         "written sets:          %zu\n"
 434                         "unwritten sets:                %zu\n"
 435                         "written key bytes:     %zu\n"
 436                         "unwritten key bytes:   %zu\n"
 437                         "floats:                        %zu\n"
 438                         "failed:                        %zu\n",
 439                         op.nodes,
 440                         op.stats.sets_written, op.stats.sets_unwritten,
 441                         op.stats.bytes_written, op.stats.bytes_unwritten,
 442                         op.stats.floats, op.stats.failed);
 443 }
 444
 445 SHOW(__bch_cache_set)
 446 {
 447         unsigned root_usage(struct cache_set *c)
 448         {
 449                 unsigned bytes = 0;
 450                 struct bkey *k;
 451                 struct btree *b;
 452                 struct btree_iter iter;
 453
 454                 goto lock_root;
 455
 456                 do {
 457                         rw_unlock(false, b);
 458 lock_root:
 459                         b = c->root;
 460                         rw_lock(false, b, b->level);
 461                 } while (b != c->root);
 462
 463                 for_each_key_filter(&b->keys, k, &iter, bch_ptr_bad)
 464                         bytes += bkey_bytes(k);
 465
 466                 rw_unlock(false, b);
 467
 468                 return (bytes * 100) / btree_bytes(c);
 469         }
 470
 471         size_t cache_size(struct cache_set *c)
 472         {
 473                 size_t ret = 0;
 474                 struct btree *b;
 475
 476                 mutex_lock(&c->bucket_lock);
 477                 list_for_each_entry(b, &c->btree_cache, list)
 478                         ret += 1 << (b->keys.page_order + PAGE_SHIFT);
 479
 480                 mutex_unlock(&c->bucket_lock);
 481                 return ret;
 482         }
 483
 484         unsigned cache_max_chain(struct cache_set *c)
 485         {
 486                 unsigned ret = 0;
 487                 struct hlist_head *h;
 488
 489                 mutex_lock(&c->bucket_lock);
 490
 491                 for (h = c->bucket_hash;
 492                      h < c->bucket_hash + (1 << BUCKET_HASH_BITS);
 493                      h++) {
 494                         unsigned i = 0;
 495                         struct hlist_node *p;
 496
 497                         hlist_for_each(p, h)
 498                                 i++;
 499
 500                         ret = max(ret, i);
 501                 }
 502
 503                 mutex_unlock(&c->bucket_lock);
 504                 return ret;
 505         }
 506
 507         unsigned btree_used(struct cache_set *c)
 508         {
 509                 return div64_u64(c->gc_stats.key_bytes * 100,
 510                                  (c->gc_stats.nodes ?: 1) * btree_bytes(c));
 511         }
 512
 513         unsigned average_key_size(struct cache_set *c)
 514         {
 515                 return c->gc_stats.nkeys
 516                         ? div64_u64(c->gc_stats.data, c->gc_stats.nkeys)
 517                         : 0;
 518         }
 519
 520         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 521
 522         sysfs_print(synchronous,                CACHE_SYNC(&c->sb));
 523         sysfs_print(journal_delay_ms,           c->journal_delay_ms);
 524         sysfs_hprint(bucket_size,               bucket_bytes(c));
 525         sysfs_hprint(block_size,                block_bytes(c));
 526         sysfs_print(tree_depth,                 c->root->level);
 527         sysfs_print(root_usage_percent,         root_usage(c));
 528
 529         sysfs_hprint(btree_cache_size,          cache_size(c));
 530         sysfs_print(btree_cache_max_chain,      cache_max_chain(c));
 531         sysfs_print(cache_available_percent,    100 - c->gc_stats.in_use);
 532
 533         sysfs_print_time_stats(&c->btree_gc_time,       btree_gc, sec, ms);
 534         sysfs_print_time_stats(&c->btree_split_time,    btree_split, sec, us);
 535         sysfs_print_time_stats(&c->sort.time,           btree_sort, ms, us);
 536         sysfs_print_time_stats(&c->btree_read_time,     btree_read, ms, us);
 537         sysfs_print_time_stats(&c->try_harder_time,     try_harder, ms, us);
 538
 539         sysfs_print(btree_used_percent, btree_used(c));
 540         sysfs_print(btree_nodes,        c->gc_stats.nodes);
 541         sysfs_hprint(average_key_size,  average_key_size(c));
 542
 543         sysfs_print(cache_read_races,
 544                     atomic_long_read(&c->cache_read_races));
 545
 546         sysfs_print(writeback_keys_done,
 547                     atomic_long_read(&c->writeback_keys_done));
 548         sysfs_print(writeback_keys_failed,
 549                     atomic_long_read(&c->writeback_keys_failed));
 550
 551         if (attr == &sysfs_errors)
 552                 return bch_snprint_string_list(buf, PAGE_SIZE, error_actions,
 553                                                c->on_error);
 554
 555         /* See count_io_errors for why 88 */
 556         sysfs_print(io_error_halflife,  c->error_decay * 88);
 557         sysfs_print(io_error_limit,     c->error_limit >> IO_ERROR_SHIFT);
 558
 559         sysfs_hprint(congested,
 560                      ((uint64_t) bch_get_congested(c)) << 9);
 561         sysfs_print(congested_read_threshold_us,
 562                     c->congested_read_threshold_us);
 563         sysfs_print(congested_write_threshold_us,
 564                     c->congested_write_threshold_us);
 565
 566         sysfs_print(active_journal_entries,     fifo_used(&c->journal.pin));
 567         sysfs_printf(verify,                    "%i", c->verify);
 568         sysfs_printf(key_merging_disabled,      "%i", c->key_merging_disabled);
 569         sysfs_printf(expensive_debug_checks,
 570                      "%i", c->expensive_debug_checks);
 571         sysfs_printf(gc_always_rewrite,         "%i", c->gc_always_rewrite);
 572         sysfs_printf(btree_shrinker_disabled,   "%i", c->shrinker_disabled);
 573         sysfs_printf(copy_gc_enabled,           "%i", c->copy_gc_enabled);
 574
 575         if (attr == &sysfs_bset_tree_stats)
 576                 return bch_bset_print_stats(c, buf);
 577
 578         return 0;
 579 }
 580 SHOW_LOCKED(bch_cache_set)
 581
 582 STORE(__bch_cache_set)
 583 {
 584         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 585
 586         if (attr == &sysfs_unregister)
 587                 bch_cache_set_unregister(c);
 588
 589         if (attr == &sysfs_stop)
 590                 bch_cache_set_stop(c);
 591
 592         if (attr == &sysfs_synchronous) {
 593                 bool sync = strtoul_or_return(buf);
 594
 595                 if (sync != CACHE_SYNC(&c->sb)) {
 596                         SET_CACHE_SYNC(&c->sb, sync);
 597                         bcache_write_super(c);
 598                 }
 599         }
 600
 601         if (attr == &sysfs_flash_vol_create) {
 602                 int r;
 603                 uint64_t v;
 604                 strtoi_h_or_return(buf, v);
 605
 606                 r = bch_flash_dev_create(c, v);
 607                 if (r)
 608                         return r;
 609         }
 610
 611         if (attr == &sysfs_clear_stats) {
 612                 atomic_long_set(&c->writeback_keys_done,        0);
 613                 atomic_long_set(&c->writeback_keys_failed,      0);
 614
 615                 memset(&c->gc_stats, 0, sizeof(struct gc_stat));
 616                 bch_cache_accounting_clear(&c->accounting);
 617         }
 618
 619         if (attr == &sysfs_trigger_gc)
 620                 wake_up_gc(c);
 621
 622         if (attr == &sysfs_prune_cache) {
 623                 struct shrink_control sc;
 624                 sc.gfp_mask = GFP_KERNEL;
 625                 sc.nr_to_scan = strtoul_or_return(buf);
 626                 c->shrink.scan_objects(&c->shrink, &sc);
 627         }
 628
 629         sysfs_strtoul(congested_read_threshold_us,
 630                       c->congested_read_threshold_us);
 631         sysfs_strtoul(congested_write_threshold_us,
 632                       c->congested_write_threshold_us);
 633
 634         if (attr == &sysfs_errors) {
 635                 ssize_t v = bch_read_string_list(buf, error_actions);
 636
 637                 if (v < 0)
 638                         return v;
 639
 640                 c->on_error = v;
 641         }
 642
 643         if (attr == &sysfs_io_error_limit)
 644                 c->error_limit = strtoul_or_return(buf) << IO_ERROR_SHIFT;
 645
 646         /* See count_io_errors() for why 88 */
 647         if (attr == &sysfs_io_error_halflife)
 648                 c->error_decay = strtoul_or_return(buf) / 88;
 649
 650         sysfs_strtoul(journal_delay_ms,         c->journal_delay_ms);
 651         sysfs_strtoul(verify,                   c->verify);
 652         sysfs_strtoul(key_merging_disabled,     c->key_merging_disabled);
 653         sysfs_strtoul(expensive_debug_checks,   c->expensive_debug_checks);
 654         sysfs_strtoul(gc_always_rewrite,        c->gc_always_rewrite);
 655         sysfs_strtoul(btree_shrinker_disabled,  c->shrinker_disabled);
 656         sysfs_strtoul(copy_gc_enabled,          c->copy_gc_enabled);
 657
 658         return size;
 659 }
 660 STORE_LOCKED(bch_cache_set)
 661
 662 SHOW(bch_cache_set_internal)
 663 {
 664         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 665         return bch_cache_set_show(&c->kobj, attr, buf);
 666 }
 667
 668 STORE(bch_cache_set_internal)
 669 {
 670         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 671         return bch_cache_set_store(&c->kobj, attr, buf, size);
 672 }
 673
 674 static void bch_cache_set_internal_release(struct kobject *k)
 675 {
 676 }
 677
 678 static struct attribute *bch_cache_set_files[] = {
 679         &sysfs_unregister,
 680         &sysfs_stop,
 681         &sysfs_synchronous,
 682         &sysfs_journal_delay_ms,
 683         &sysfs_flash_vol_create,
 684
 685         &sysfs_bucket_size,
 686         &sysfs_block_size,
 687         &sysfs_tree_depth,
 688         &sysfs_root_usage_percent,
 689         &sysfs_btree_cache_size,
 690         &sysfs_cache_available_percent,
 691
 692         &sysfs_average_key_size,
 693
 694         &sysfs_errors,
 695         &sysfs_io_error_limit,
 696         &sysfs_io_error_halflife,
 697         &sysfs_congested,
 698         &sysfs_congested_read_threshold_us,
 699         &sysfs_congested_write_threshold_us,
 700         &sysfs_clear_stats,
 701         NULL
 702 };
 703 KTYPE(bch_cache_set);
 704
 705 static struct attribute *bch_cache_set_internal_files[] = {
 706         &sysfs_active_journal_entries,
 707
 708         sysfs_time_stats_attribute_list(btree_gc, sec, ms)
 709         sysfs_time_stats_attribute_list(btree_split, sec, us)
 710         sysfs_time_stats_attribute_list(btree_sort, ms, us)
 711         sysfs_time_stats_attribute_list(btree_read, ms, us)
 712         sysfs_time_stats_attribute_list(try_harder, ms, us)
 713
 714         &sysfs_btree_nodes,
 715         &sysfs_btree_used_percent,
 716         &sysfs_btree_cache_max_chain,
 717
 718         &sysfs_bset_tree_stats,
 719         &sysfs_cache_read_races,
 720         &sysfs_writeback_keys_done,
 721         &sysfs_writeback_keys_failed,
 722
 723         &sysfs_trigger_gc,
 724         &sysfs_prune_cache,
 725 #ifdef CONFIG_BCACHE_DEBUG
 726         &sysfs_verify,
 727         &sysfs_key_merging_disabled,
 728         &sysfs_expensive_debug_checks,
 729 #endif
 730         &sysfs_gc_always_rewrite,
 731         &sysfs_btree_shrinker_disabled,
 732         &sysfs_copy_gc_enabled,
 733         NULL
 734 };
 735 KTYPE(bch_cache_set_internal);
 736
 737 SHOW(__bch_cache)
 738 {
 739         struct cache *ca = container_of(kobj, struct cache, kobj);
 740
 741         sysfs_hprint(bucket_size,       bucket_bytes(ca));
 742         sysfs_hprint(block_size,        block_bytes(ca));
 743         sysfs_print(nbuckets,           ca->sb.nbuckets);
 744         sysfs_print(discard,            ca->discard);
 745         sysfs_hprint(written, atomic_long_read(&ca->sectors_written) << 9);
 746         sysfs_hprint(btree_written,
 747                      atomic_long_read(&ca->btree_sectors_written) << 9);
 748         sysfs_hprint(metadata_written,
 749                      (atomic_long_read(&ca->meta_sectors_written) +
 750                       atomic_long_read(&ca->btree_sectors_written)) << 9);
 751
 752         sysfs_print(io_errors,
 753                     atomic_read(&ca->io_errors) >> IO_ERROR_SHIFT);
 754
 755         if (attr == &sysfs_cache_replacement_policy)
 756                 return bch_snprint_string_list(buf, PAGE_SIZE,
 757                                                cache_replacement_policies,
 758                                                CACHE_REPLACEMENT(&ca->sb));
 759
 760         if (attr == &sysfs_priority_stats) {
 761                 int cmp(const void *l, const void *r)
 762                 {       return *((uint16_t *) r) - *((uint16_t *) l); }
 763
 764                 size_t n = ca->sb.nbuckets, i, unused, btree;
 765                 uint64_t sum = 0;
 766                 /* Compute 31 quantiles */
 767                 uint16_t q[31], *p, *cached;
 768                 ssize_t ret;
 769
 770                 cached = p = vmalloc(ca->sb.nbuckets * sizeof(uint16_t));
 771                 if (!p)
 772                         return -ENOMEM;
 773
 774                 mutex_lock(&ca->set->bucket_lock);
 775                 for (i = ca->sb.first_bucket; i < n; i++)
 776                         p[i] = ca->buckets[i].prio;
 777                 mutex_unlock(&ca->set->bucket_lock);
 778
 779                 sort(p, n, sizeof(uint16_t), cmp, NULL);
 780
 781                 while (n &&
 782                        !cached[n - 1])
 783                         --n;
 784
 785                 unused = ca->sb.nbuckets - n;
 786
 787                 while (cached < p + n &&
 788                        *cached == BTREE_PRIO)
 789                         cached++;
 790
 791                 btree = cached - p;
 792                 n -= btree;
 793
 794                 for (i = 0; i < n; i++)
 795                         sum += INITIAL_PRIO - cached[i];
 796
 797                 if (n)
 798                         do_div(sum, n);
 799
 800                 for (i = 0; i < ARRAY_SIZE(q); i++)
 801                         q[i] = INITIAL_PRIO - cached[n * (i + 1) /
 802                                 (ARRAY_SIZE(q) + 1)];
 803
 804                 vfree(p);
 805
 806                 ret = scnprintf(buf, PAGE_SIZE,
 807                                 "Unused:                %zu%%\n"
 808                                 "Metadata:      %zu%%\n"
 809                                 "Average:       %llu\n"
 810                                 "Sectors per Q: %zu\n"
 811                                 "Quantiles:     [",
 812                                 unused * 100 / (size_t) ca->sb.nbuckets,
 813                                 btree * 100 / (size_t) ca->sb.nbuckets, sum,
 814                                 n * ca->sb.bucket_size / (ARRAY_SIZE(q) + 1));
 815
 816                 for (i = 0; i < ARRAY_SIZE(q); i++)
 817                         ret += scnprintf(buf + ret, PAGE_SIZE - ret,
 818                                          "%u ", q[i]);
 819                 ret--;
 820
 821                 ret += scnprintf(buf + ret, PAGE_SIZE - ret, "]\n");
 822
 823                 return ret;
 824         }
 825
 826         return 0;
 827 }
 828 SHOW_LOCKED(bch_cache)
 829
 830 STORE(__bch_cache)
 831 {
 832         struct cache *ca = container_of(kobj, struct cache, kobj);
 833
 834         if (attr == &sysfs_discard) {
 835                 bool v = strtoul_or_return(buf);
 836
 837                 if (blk_queue_discard(bdev_get_queue(ca->bdev)))
 838                         ca->discard = v;
 839
 840                 if (v != CACHE_DISCARD(&ca->sb)) {
 841                         SET_CACHE_DISCARD(&ca->sb, v);
 842                         bcache_write_super(ca->set);
 843                 }
 844         }
 845
 846         if (attr == &sysfs_cache_replacement_policy) {
 847                 ssize_t v = bch_read_string_list(buf, cache_replacement_policies);
 848
 849                 if (v < 0)
 850                         return v;
 851
 852                 if ((unsigned) v != CACHE_REPLACEMENT(&ca->sb)) {
 853                         mutex_lock(&ca->set->bucket_lock);
 854                         SET_CACHE_REPLACEMENT(&ca->sb, v);
 855                         mutex_unlock(&ca->set->bucket_lock);
 856
 857                         bcache_write_super(ca->set);
 858                 }
 859         }
 860
 861         if (attr == &sysfs_clear_stats) {
 862                 atomic_long_set(&ca->sectors_written, 0);
 863                 atomic_long_set(&ca->btree_sectors_written, 0);
 864                 atomic_long_set(&ca->meta_sectors_written, 0);
 865                 atomic_set(&ca->io_count, 0);
 866                 atomic_set(&ca->io_errors, 0);
 867         }
 868
 869         return size;
 870 }
 871 STORE_LOCKED(bch_cache)
 872
 873 static struct attribute *bch_cache_files[] = {
 874         &sysfs_bucket_size,
 875         &sysfs_block_size,
 876         &sysfs_nbuckets,
 877         &sysfs_priority_stats,
 878         &sysfs_discard,
 879         &sysfs_written,
 880         &sysfs_btree_written,
 881         &sysfs_metadata_written,
 882         &sysfs_io_errors,
 883         &sysfs_clear_stats,
 884         &sysfs_cache_replacement_policy,
 885         NULL
 886 };
 887 KTYPE(bch_cache);