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(freelist_percent);
 106 rw_attribute(cache_replacement_policy);
 107 rw_attribute(btree_shrinker_disabled);
 108 rw_attribute(copy_gc_enabled);
 109 rw_attribute(size);
 110
 111 SHOW(__bch_cached_dev)
 112 {
 113         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 114                                              disk.kobj);
 115         const char *states[] = { "no cache", "clean", "dirty", "inconsistent" };
 116
 117 #define var(stat)               (dc->stat)
 118
 119         if (attr == &sysfs_cache_mode)
 120                 return bch_snprint_string_list(buf, PAGE_SIZE,
 121                                                bch_cache_modes + 1,
 122                                                BDEV_CACHE_MODE(&dc->sb));
 123
 124         sysfs_printf(data_csum,         "%i", dc->disk.data_csum);
 125         var_printf(verify,              "%i");
 126         var_printf(bypass_torture_test, "%i");
 127         var_printf(writeback_metadata,  "%i");
 128         var_printf(writeback_running,   "%i");
 129         var_print(writeback_delay);
 130         var_print(writeback_percent);
 131         sysfs_hprint(writeback_rate,    dc->writeback_rate.rate << 9);
 132
 133         var_print(writeback_rate_update_seconds);
 134         var_print(writeback_rate_d_term);
 135         var_print(writeback_rate_p_term_inverse);
 136
 137         if (attr == &sysfs_writeback_rate_debug) {
 138                 char rate[20];
 139                 char dirty[20];
 140                 char target[20];
 141                 char proportional[20];
 142                 char derivative[20];
 143                 char change[20];
 144                 s64 next_io;
 145
 146                 bch_hprint(rate,        dc->writeback_rate.rate << 9);
 147                 bch_hprint(dirty,       bcache_dev_sectors_dirty(&dc->disk) << 9);
 148                 bch_hprint(target,      dc->writeback_rate_target << 9);
 149                 bch_hprint(proportional,dc->writeback_rate_proportional << 9);
 150                 bch_hprint(derivative,  dc->writeback_rate_derivative << 9);
 151                 bch_hprint(change,      dc->writeback_rate_change << 9);
 152
 153                 next_io = div64_s64(dc->writeback_rate.next - local_clock(),
 154                                     NSEC_PER_MSEC);
 155
 156                 return sprintf(buf,
 157                                "rate:\t\t%s/sec\n"
 158                                "dirty:\t\t%s\n"
 159                                "target:\t\t%s\n"
 160                                "proportional:\t%s\n"
 161                                "derivative:\t%s\n"
 162                                "change:\t\t%s/sec\n"
 163                                "next io:\t%llims\n",
 164                                rate, dirty, target, proportional,
 165                                derivative, change, next_io);
 166         }
 167
 168         sysfs_hprint(dirty_data,
 169                      bcache_dev_sectors_dirty(&dc->disk) << 9);
 170
 171         sysfs_hprint(stripe_size,       dc->disk.stripe_size << 9);
 172         var_printf(partial_stripes_expensive,   "%u");
 173
 174         var_hprint(sequential_cutoff);
 175         var_hprint(readahead);
 176
 177         sysfs_print(running,            atomic_read(&dc->running));
 178         sysfs_print(state,              states[BDEV_STATE(&dc->sb)]);
 179
 180         if (attr == &sysfs_label) {
 181                 memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
 182                 buf[SB_LABEL_SIZE + 1] = '\0';
 183                 strcat(buf, "\n");
 184                 return strlen(buf);
 185         }
 186
 187 #undef var
 188         return 0;
 189 }
 190 SHOW_LOCKED(bch_cached_dev)
 191
 192 STORE(__cached_dev)
 193 {
 194         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 195                                              disk.kobj);
 196         unsigned v = size;
 197         struct cache_set *c;
 198         struct kobj_uevent_env *env;
 199
 200 #define d_strtoul(var)          sysfs_strtoul(var, dc->var)
 201 #define d_strtoul_nonzero(var)  sysfs_strtoul_clamp(var, dc->var, 1, INT_MAX)
 202 #define d_strtoi_h(var)         sysfs_hatoi(var, dc->var)
 203
 204         sysfs_strtoul(data_csum,        dc->disk.data_csum);
 205         d_strtoul(verify);
 206         d_strtoul(bypass_torture_test);
 207         d_strtoul(writeback_metadata);
 208         d_strtoul(writeback_running);
 209         d_strtoul(writeback_delay);
 210
 211         sysfs_strtoul_clamp(writeback_percent, dc->writeback_percent, 0, 40);
 212
 213         sysfs_strtoul_clamp(writeback_rate,
 214                             dc->writeback_rate.rate, 1, INT_MAX);
 215
 216         d_strtoul_nonzero(writeback_rate_update_seconds);
 217         d_strtoul(writeback_rate_d_term);
 218         d_strtoul_nonzero(writeback_rate_p_term_inverse);
 219
 220         d_strtoi_h(sequential_cutoff);
 221         d_strtoi_h(readahead);
 222
 223         if (attr == &sysfs_clear_stats)
 224                 bch_cache_accounting_clear(&dc->accounting);
 225
 226         if (attr == &sysfs_running &&
 227             strtoul_or_return(buf))
 228                 bch_cached_dev_run(dc);
 229
 230         if (attr == &sysfs_cache_mode) {
 231                 ssize_t v = bch_read_string_list(buf, bch_cache_modes + 1);
 232
 233                 if (v < 0)
 234                         return v;
 235
 236                 if ((unsigned) v != BDEV_CACHE_MODE(&dc->sb)) {
 237                         SET_BDEV_CACHE_MODE(&dc->sb, v);
 238                         bch_write_bdev_super(dc, NULL);
 239                 }
 240         }
 241
 242         if (attr == &sysfs_label) {
 243                 if (size > SB_LABEL_SIZE)
 244                         return -EINVAL;
 245                 memcpy(dc->sb.label, buf, size);
 246                 if (size < SB_LABEL_SIZE)
 247                         dc->sb.label[size] = '\0';
 248                 if (size && dc->sb.label[size - 1] == '\n')
 249                         dc->sb.label[size - 1] = '\0';
 250                 bch_write_bdev_super(dc, NULL);
 251                 if (dc->disk.c) {
 252                         memcpy(dc->disk.c->uuids[dc->disk.id].label,
 253                                buf, SB_LABEL_SIZE);
 254                         bch_uuid_write(dc->disk.c);
 255                 }
 256                 env = kzalloc(sizeof(struct kobj_uevent_env), GFP_KERNEL);
 257                 if (!env)
 258                         return -ENOMEM;
 259                 add_uevent_var(env, "DRIVER=bcache");
 260                 add_uevent_var(env, "CACHED_UUID=%pU", dc->sb.uuid),
 261                 add_uevent_var(env, "CACHED_LABEL=%s", buf);
 262                 kobject_uevent_env(
 263                         &disk_to_dev(dc->disk.disk)->kobj, KOBJ_CHANGE, env->envp);
 264                 kfree(env);
 265         }
 266
 267         if (attr == &sysfs_attach) {
 268                 if (bch_parse_uuid(buf, dc->sb.set_uuid) < 16)
 269                         return -EINVAL;
 270
 271                 list_for_each_entry(c, &bch_cache_sets, list) {
 272                         v = bch_cached_dev_attach(dc, c);
 273                         if (!v)
 274                                 return size;
 275                 }
 276
 277                 pr_err("Can't attach %s: cache set not found", buf);
 278                 size = v;
 279         }
 280
 281         if (attr == &sysfs_detach && dc->disk.c)
 282                 bch_cached_dev_detach(dc);
 283
 284         if (attr == &sysfs_stop)
 285                 bcache_device_stop(&dc->disk);
 286
 287         return size;
 288 }
 289
 290 STORE(bch_cached_dev)
 291 {
 292         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 293                                              disk.kobj);
 294
 295         mutex_lock(&bch_register_lock);
 296         size = __cached_dev_store(kobj, attr, buf, size);
 297
 298         if (attr == &sysfs_writeback_running)
 299                 bch_writeback_queue(dc);
 300
 301         if (attr == &sysfs_writeback_percent)
 302                 schedule_delayed_work(&dc->writeback_rate_update,
 303                                       dc->writeback_rate_update_seconds * HZ);
 304
 305         mutex_unlock(&bch_register_lock);
 306         return size;
 307 }
 308
 309 static struct attribute *bch_cached_dev_files[] = {
 310         &sysfs_attach,
 311         &sysfs_detach,
 312         &sysfs_stop,
 313 #if 0
 314         &sysfs_data_csum,
 315 #endif
 316         &sysfs_cache_mode,
 317         &sysfs_writeback_metadata,
 318         &sysfs_writeback_running,
 319         &sysfs_writeback_delay,
 320         &sysfs_writeback_percent,
 321         &sysfs_writeback_rate,
 322         &sysfs_writeback_rate_update_seconds,
 323         &sysfs_writeback_rate_d_term,
 324         &sysfs_writeback_rate_p_term_inverse,
 325         &sysfs_writeback_rate_debug,
 326         &sysfs_dirty_data,
 327         &sysfs_stripe_size,
 328         &sysfs_partial_stripes_expensive,
 329         &sysfs_sequential_cutoff,
 330         &sysfs_clear_stats,
 331         &sysfs_running,
 332         &sysfs_state,
 333         &sysfs_label,
 334         &sysfs_readahead,
 335 #ifdef CONFIG_BCACHE_DEBUG
 336         &sysfs_verify,
 337         &sysfs_bypass_torture_test,
 338 #endif
 339         NULL
 340 };
 341 KTYPE(bch_cached_dev);
 342
 343 SHOW(bch_flash_dev)
 344 {
 345         struct bcache_device *d = container_of(kobj, struct bcache_device,
 346                                                kobj);
 347         struct uuid_entry *u = &d->c->uuids[d->id];
 348
 349         sysfs_printf(data_csum, "%i", d->data_csum);
 350         sysfs_hprint(size,      u->sectors << 9);
 351
 352         if (attr == &sysfs_label) {
 353                 memcpy(buf, u->label, SB_LABEL_SIZE);
 354                 buf[SB_LABEL_SIZE + 1] = '\0';
 355                 strcat(buf, "\n");
 356                 return strlen(buf);
 357         }
 358
 359         return 0;
 360 }
 361
 362 STORE(__bch_flash_dev)
 363 {
 364         struct bcache_device *d = container_of(kobj, struct bcache_device,
 365                                                kobj);
 366         struct uuid_entry *u = &d->c->uuids[d->id];
 367
 368         sysfs_strtoul(data_csum,        d->data_csum);
 369
 370         if (attr == &sysfs_size) {
 371                 uint64_t v;
 372                 strtoi_h_or_return(buf, v);
 373
 374                 u->sectors = v >> 9;
 375                 bch_uuid_write(d->c);
 376                 set_capacity(d->disk, u->sectors);
 377         }
 378
 379         if (attr == &sysfs_label) {
 380                 memcpy(u->label, buf, SB_LABEL_SIZE);
 381                 bch_uuid_write(d->c);
 382         }
 383
 384         if (attr == &sysfs_unregister) {
 385                 set_bit(BCACHE_DEV_DETACHING, &d->flags);
 386                 bcache_device_stop(d);
 387         }
 388
 389         return size;
 390 }
 391 STORE_LOCKED(bch_flash_dev)
 392
 393 static struct attribute *bch_flash_dev_files[] = {
 394         &sysfs_unregister,
 395 #if 0
 396         &sysfs_data_csum,
 397 #endif
 398         &sysfs_label,
 399         &sysfs_size,
 400         NULL
 401 };
 402 KTYPE(bch_flash_dev);
 403
 404 SHOW(__bch_cache_set)
 405 {
 406         unsigned root_usage(struct cache_set *c)
 407         {
 408                 unsigned bytes = 0;
 409                 struct bkey *k;
 410                 struct btree *b;
 411                 struct btree_iter iter;
 412
 413                 goto lock_root;
 414
 415                 do {
 416                         rw_unlock(false, b);
 417 lock_root:
 418                         b = c->root;
 419                         rw_lock(false, b, b->level);
 420                 } while (b != c->root);
 421
 422                 for_each_key_filter(b, k, &iter, bch_ptr_bad)
 423                         bytes += bkey_bytes(k);
 424
 425                 rw_unlock(false, b);
 426
 427                 return (bytes * 100) / btree_bytes(c);
 428         }
 429
 430         size_t cache_size(struct cache_set *c)
 431         {
 432                 size_t ret = 0;
 433                 struct btree *b;
 434
 435                 mutex_lock(&c->bucket_lock);
 436                 list_for_each_entry(b, &c->btree_cache, list)
 437                         ret += 1 << (b->page_order + PAGE_SHIFT);
 438
 439                 mutex_unlock(&c->bucket_lock);
 440                 return ret;
 441         }
 442
 443         unsigned cache_max_chain(struct cache_set *c)
 444         {
 445                 unsigned ret = 0;
 446                 struct hlist_head *h;
 447
 448                 mutex_lock(&c->bucket_lock);
 449
 450                 for (h = c->bucket_hash;
 451                      h < c->bucket_hash + (1 << BUCKET_HASH_BITS);
 452                      h++) {
 453                         unsigned i = 0;
 454                         struct hlist_node *p;
 455
 456                         hlist_for_each(p, h)
 457                                 i++;
 458
 459                         ret = max(ret, i);
 460                 }
 461
 462                 mutex_unlock(&c->bucket_lock);
 463                 return ret;
 464         }
 465
 466         unsigned btree_used(struct cache_set *c)
 467         {
 468                 return div64_u64(c->gc_stats.key_bytes * 100,
 469                                  (c->gc_stats.nodes ?: 1) * btree_bytes(c));
 470         }
 471
 472         unsigned average_key_size(struct cache_set *c)
 473         {
 474                 return c->gc_stats.nkeys
 475                         ? div64_u64(c->gc_stats.data, c->gc_stats.nkeys)
 476                         : 0;
 477         }
 478
 479         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 480
 481         sysfs_print(synchronous,                CACHE_SYNC(&c->sb));
 482         sysfs_print(journal_delay_ms,           c->journal_delay_ms);
 483         sysfs_hprint(bucket_size,               bucket_bytes(c));
 484         sysfs_hprint(block_size,                block_bytes(c));
 485         sysfs_print(tree_depth,                 c->root->level);
 486         sysfs_print(root_usage_percent,         root_usage(c));
 487
 488         sysfs_hprint(btree_cache_size,          cache_size(c));
 489         sysfs_print(btree_cache_max_chain,      cache_max_chain(c));
 490         sysfs_print(cache_available_percent,    100 - c->gc_stats.in_use);
 491
 492         sysfs_print_time_stats(&c->btree_gc_time,       btree_gc, sec, ms);
 493         sysfs_print_time_stats(&c->btree_split_time,    btree_split, sec, us);
 494         sysfs_print_time_stats(&c->sort_time,           btree_sort, ms, us);
 495         sysfs_print_time_stats(&c->btree_read_time,     btree_read, ms, us);
 496         sysfs_print_time_stats(&c->try_harder_time,     try_harder, ms, us);
 497
 498         sysfs_print(btree_used_percent, btree_used(c));
 499         sysfs_print(btree_nodes,        c->gc_stats.nodes);
 500         sysfs_hprint(average_key_size,  average_key_size(c));
 501
 502         sysfs_print(cache_read_races,
 503                     atomic_long_read(&c->cache_read_races));
 504
 505         sysfs_print(writeback_keys_done,
 506                     atomic_long_read(&c->writeback_keys_done));
 507         sysfs_print(writeback_keys_failed,
 508                     atomic_long_read(&c->writeback_keys_failed));
 509
 510         if (attr == &sysfs_errors)
 511                 return bch_snprint_string_list(buf, PAGE_SIZE, error_actions,
 512                                                c->on_error);
 513
 514         /* See count_io_errors for why 88 */
 515         sysfs_print(io_error_halflife,  c->error_decay * 88);
 516         sysfs_print(io_error_limit,     c->error_limit >> IO_ERROR_SHIFT);
 517
 518         sysfs_hprint(congested,
 519                      ((uint64_t) bch_get_congested(c)) << 9);
 520         sysfs_print(congested_read_threshold_us,
 521                     c->congested_read_threshold_us);
 522         sysfs_print(congested_write_threshold_us,
 523                     c->congested_write_threshold_us);
 524
 525         sysfs_print(active_journal_entries,     fifo_used(&c->journal.pin));
 526         sysfs_printf(verify,                    "%i", c->verify);
 527         sysfs_printf(key_merging_disabled,      "%i", c->key_merging_disabled);
 528         sysfs_printf(expensive_debug_checks,
 529                      "%i", c->expensive_debug_checks);
 530         sysfs_printf(gc_always_rewrite,         "%i", c->gc_always_rewrite);
 531         sysfs_printf(btree_shrinker_disabled,   "%i", c->shrinker_disabled);
 532         sysfs_printf(copy_gc_enabled,           "%i", c->copy_gc_enabled);
 533
 534         if (attr == &sysfs_bset_tree_stats)
 535                 return bch_bset_print_stats(c, buf);
 536
 537         return 0;
 538 }
 539 SHOW_LOCKED(bch_cache_set)
 540
 541 STORE(__bch_cache_set)
 542 {
 543         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 544
 545         if (attr == &sysfs_unregister)
 546                 bch_cache_set_unregister(c);
 547
 548         if (attr == &sysfs_stop)
 549                 bch_cache_set_stop(c);
 550
 551         if (attr == &sysfs_synchronous) {
 552                 bool sync = strtoul_or_return(buf);
 553
 554                 if (sync != CACHE_SYNC(&c->sb)) {
 555                         SET_CACHE_SYNC(&c->sb, sync);
 556                         bcache_write_super(c);
 557                 }
 558         }
 559
 560         if (attr == &sysfs_flash_vol_create) {
 561                 int r;
 562                 uint64_t v;
 563                 strtoi_h_or_return(buf, v);
 564
 565                 r = bch_flash_dev_create(c, v);
 566                 if (r)
 567                         return r;
 568         }
 569
 570         if (attr == &sysfs_clear_stats) {
 571                 atomic_long_set(&c->writeback_keys_done,        0);
 572                 atomic_long_set(&c->writeback_keys_failed,      0);
 573
 574                 memset(&c->gc_stats, 0, sizeof(struct gc_stat));
 575                 bch_cache_accounting_clear(&c->accounting);
 576         }
 577
 578         if (attr == &sysfs_trigger_gc)
 579                 wake_up_gc(c);
 580
 581         if (attr == &sysfs_prune_cache) {
 582                 struct shrink_control sc;
 583                 sc.gfp_mask = GFP_KERNEL;
 584                 sc.nr_to_scan = strtoul_or_return(buf);
 585                 c->shrink.scan_objects(&c->shrink, &sc);
 586         }
 587
 588         sysfs_strtoul(congested_read_threshold_us,
 589                       c->congested_read_threshold_us);
 590         sysfs_strtoul(congested_write_threshold_us,
 591                       c->congested_write_threshold_us);
 592
 593         if (attr == &sysfs_errors) {
 594                 ssize_t v = bch_read_string_list(buf, error_actions);
 595
 596                 if (v < 0)
 597                         return v;
 598
 599                 c->on_error = v;
 600         }
 601
 602         if (attr == &sysfs_io_error_limit)
 603                 c->error_limit = strtoul_or_return(buf) << IO_ERROR_SHIFT;
 604
 605         /* See count_io_errors() for why 88 */
 606         if (attr == &sysfs_io_error_halflife)
 607                 c->error_decay = strtoul_or_return(buf) / 88;
 608
 609         sysfs_strtoul(journal_delay_ms,         c->journal_delay_ms);
 610         sysfs_strtoul(verify,                   c->verify);
 611         sysfs_strtoul(key_merging_disabled,     c->key_merging_disabled);
 612         sysfs_strtoul(expensive_debug_checks,   c->expensive_debug_checks);
 613         sysfs_strtoul(gc_always_rewrite,        c->gc_always_rewrite);
 614         sysfs_strtoul(btree_shrinker_disabled,  c->shrinker_disabled);
 615         sysfs_strtoul(copy_gc_enabled,          c->copy_gc_enabled);
 616
 617         return size;
 618 }
 619 STORE_LOCKED(bch_cache_set)
 620
 621 SHOW(bch_cache_set_internal)
 622 {
 623         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 624         return bch_cache_set_show(&c->kobj, attr, buf);
 625 }
 626
 627 STORE(bch_cache_set_internal)
 628 {
 629         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 630         return bch_cache_set_store(&c->kobj, attr, buf, size);
 631 }
 632
 633 static void bch_cache_set_internal_release(struct kobject *k)
 634 {
 635 }
 636
 637 static struct attribute *bch_cache_set_files[] = {
 638         &sysfs_unregister,
 639         &sysfs_stop,
 640         &sysfs_synchronous,
 641         &sysfs_journal_delay_ms,
 642         &sysfs_flash_vol_create,
 643
 644         &sysfs_bucket_size,
 645         &sysfs_block_size,
 646         &sysfs_tree_depth,
 647         &sysfs_root_usage_percent,
 648         &sysfs_btree_cache_size,
 649         &sysfs_cache_available_percent,
 650
 651         &sysfs_average_key_size,
 652
 653         &sysfs_errors,
 654         &sysfs_io_error_limit,
 655         &sysfs_io_error_halflife,
 656         &sysfs_congested,
 657         &sysfs_congested_read_threshold_us,
 658         &sysfs_congested_write_threshold_us,
 659         &sysfs_clear_stats,
 660         NULL
 661 };
 662 KTYPE(bch_cache_set);
 663
 664 static struct attribute *bch_cache_set_internal_files[] = {
 665         &sysfs_active_journal_entries,
 666
 667         sysfs_time_stats_attribute_list(btree_gc, sec, ms)
 668         sysfs_time_stats_attribute_list(btree_split, sec, us)
 669         sysfs_time_stats_attribute_list(btree_sort, ms, us)
 670         sysfs_time_stats_attribute_list(btree_read, ms, us)
 671         sysfs_time_stats_attribute_list(try_harder, ms, us)
 672
 673         &sysfs_btree_nodes,
 674         &sysfs_btree_used_percent,
 675         &sysfs_btree_cache_max_chain,
 676
 677         &sysfs_bset_tree_stats,
 678         &sysfs_cache_read_races,
 679         &sysfs_writeback_keys_done,
 680         &sysfs_writeback_keys_failed,
 681
 682         &sysfs_trigger_gc,
 683         &sysfs_prune_cache,
 684 #ifdef CONFIG_BCACHE_DEBUG
 685         &sysfs_verify,
 686         &sysfs_key_merging_disabled,
 687         &sysfs_expensive_debug_checks,
 688 #endif
 689         &sysfs_gc_always_rewrite,
 690         &sysfs_btree_shrinker_disabled,
 691         &sysfs_copy_gc_enabled,
 692         NULL
 693 };
 694 KTYPE(bch_cache_set_internal);
 695
 696 SHOW(__bch_cache)
 697 {
 698         struct cache *ca = container_of(kobj, struct cache, kobj);
 699
 700         sysfs_hprint(bucket_size,       bucket_bytes(ca));
 701         sysfs_hprint(block_size,        block_bytes(ca));
 702         sysfs_print(nbuckets,           ca->sb.nbuckets);
 703         sysfs_print(discard,            ca->discard);
 704         sysfs_hprint(written, atomic_long_read(&ca->sectors_written) << 9);
 705         sysfs_hprint(btree_written,
 706                      atomic_long_read(&ca->btree_sectors_written) << 9);
 707         sysfs_hprint(metadata_written,
 708                      (atomic_long_read(&ca->meta_sectors_written) +
 709                       atomic_long_read(&ca->btree_sectors_written)) << 9);
 710
 711         sysfs_print(io_errors,
 712                     atomic_read(&ca->io_errors) >> IO_ERROR_SHIFT);
 713
 714         sysfs_print(freelist_percent, ca->free.size * 100 /
 715                     ((size_t) ca->sb.nbuckets));
 716
 717         if (attr == &sysfs_cache_replacement_policy)
 718                 return bch_snprint_string_list(buf, PAGE_SIZE,
 719                                                cache_replacement_policies,
 720                                                CACHE_REPLACEMENT(&ca->sb));
 721
 722         if (attr == &sysfs_priority_stats) {
 723                 int cmp(const void *l, const void *r)
 724                 {       return *((uint16_t *) r) - *((uint16_t *) l); }
 725
 726                 size_t n = ca->sb.nbuckets, i, unused, btree;
 727                 uint64_t sum = 0;
 728                 /* Compute 31 quantiles */
 729                 uint16_t q[31], *p, *cached;
 730                 ssize_t ret;
 731
 732                 cached = p = vmalloc(ca->sb.nbuckets * sizeof(uint16_t));
 733                 if (!p)
 734                         return -ENOMEM;
 735
 736                 mutex_lock(&ca->set->bucket_lock);
 737                 for (i = ca->sb.first_bucket; i < n; i++)
 738                         p[i] = ca->buckets[i].prio;
 739                 mutex_unlock(&ca->set->bucket_lock);
 740
 741                 sort(p, n, sizeof(uint16_t), cmp, NULL);
 742
 743                 while (n &&
 744                        !cached[n - 1])
 745                         --n;
 746
 747                 unused = ca->sb.nbuckets - n;
 748
 749                 while (cached < p + n &&
 750                        *cached == BTREE_PRIO)
 751                         cached++;
 752
 753                 btree = cached - p;
 754                 n -= btree;
 755
 756                 for (i = 0; i < n; i++)
 757                         sum += INITIAL_PRIO - cached[i];
 758
 759                 if (n)
 760                         do_div(sum, n);
 761
 762                 for (i = 0; i < ARRAY_SIZE(q); i++)
 763                         q[i] = INITIAL_PRIO - cached[n * (i + 1) /
 764                                 (ARRAY_SIZE(q) + 1)];
 765
 766                 vfree(p);
 767
 768                 ret = scnprintf(buf, PAGE_SIZE,
 769                                 "Unused:                %zu%%\n"
 770                                 "Metadata:      %zu%%\n"
 771                                 "Average:       %llu\n"
 772                                 "Sectors per Q: %zu\n"
 773                                 "Quantiles:     [",
 774                                 unused * 100 / (size_t) ca->sb.nbuckets,
 775                                 btree * 100 / (size_t) ca->sb.nbuckets, sum,
 776                                 n * ca->sb.bucket_size / (ARRAY_SIZE(q) + 1));
 777
 778                 for (i = 0; i < ARRAY_SIZE(q); i++)
 779                         ret += scnprintf(buf + ret, PAGE_SIZE - ret,
 780                                          "%u ", q[i]);
 781                 ret--;
 782
 783                 ret += scnprintf(buf + ret, PAGE_SIZE - ret, "]\n");
 784
 785                 return ret;
 786         }
 787
 788         return 0;
 789 }
 790 SHOW_LOCKED(bch_cache)
 791
 792 STORE(__bch_cache)
 793 {
 794         struct cache *ca = container_of(kobj, struct cache, kobj);
 795
 796         if (attr == &sysfs_discard) {
 797                 bool v = strtoul_or_return(buf);
 798
 799                 if (blk_queue_discard(bdev_get_queue(ca->bdev)))
 800                         ca->discard = v;
 801
 802                 if (v != CACHE_DISCARD(&ca->sb)) {
 803                         SET_CACHE_DISCARD(&ca->sb, v);
 804                         bcache_write_super(ca->set);
 805                 }
 806         }
 807
 808         if (attr == &sysfs_cache_replacement_policy) {
 809                 ssize_t v = bch_read_string_list(buf, cache_replacement_policies);
 810
 811                 if (v < 0)
 812                         return v;
 813
 814                 if ((unsigned) v != CACHE_REPLACEMENT(&ca->sb)) {
 815                         mutex_lock(&ca->set->bucket_lock);
 816                         SET_CACHE_REPLACEMENT(&ca->sb, v);
 817                         mutex_unlock(&ca->set->bucket_lock);
 818
 819                         bcache_write_super(ca->set);
 820                 }
 821         }
 822
 823         if (attr == &sysfs_freelist_percent) {
 824                 DECLARE_FIFO(long, free);
 825                 long i;
 826                 size_t p = strtoul_or_return(buf);
 827
 828                 p = clamp_t(size_t,
 829                             ((size_t) ca->sb.nbuckets * p) / 100,
 830                             roundup_pow_of_two(ca->sb.nbuckets) >> 9,
 831                             ca->sb.nbuckets / 2);
 832
 833                 if (!init_fifo_exact(&free, p, GFP_KERNEL))
 834                         return -ENOMEM;
 835
 836                 mutex_lock(&ca->set->bucket_lock);
 837
 838                 fifo_move(&free, &ca->free);
 839                 fifo_swap(&free, &ca->free);
 840
 841                 mutex_unlock(&ca->set->bucket_lock);
 842
 843                 while (fifo_pop(&free, i))
 844                         atomic_dec(&ca->buckets[i].pin);
 845
 846                 free_fifo(&free);
 847         }
 848
 849         if (attr == &sysfs_clear_stats) {
 850                 atomic_long_set(&ca->sectors_written, 0);
 851                 atomic_long_set(&ca->btree_sectors_written, 0);
 852                 atomic_long_set(&ca->meta_sectors_written, 0);
 853                 atomic_set(&ca->io_count, 0);
 854                 atomic_set(&ca->io_errors, 0);
 855         }
 856
 857         return size;
 858 }
 859 STORE_LOCKED(bch_cache)
 860
 861 static struct attribute *bch_cache_files[] = {
 862         &sysfs_bucket_size,
 863         &sysfs_block_size,
 864         &sysfs_nbuckets,
 865         &sysfs_priority_stats,
 866         &sysfs_discard,
 867         &sysfs_written,
 868         &sysfs_btree_written,
 869         &sysfs_metadata_written,
 870         &sysfs_io_errors,
 871         &sysfs_clear_stats,
 872         &sysfs_freelist_percent,
 873         &sysfs_cache_replacement_policy,
 874         NULL
 875 };
 876 KTYPE(bch_cache);