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