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 #include <linux/sched/clock.h>
  17
  18 static const char * const cache_replacement_policies[] = {
  19         "lru",
  20         "fifo",
  21         "random",
  22         NULL
  23 };
  24
  25 static const char * const error_actions[] = {
  26         "unregister",
  27         "panic",
  28         NULL
  29 };
  30
  31 write_attribute(attach);
  32 write_attribute(detach);
  33 write_attribute(unregister);
  34 write_attribute(stop);
  35 write_attribute(clear_stats);
  36 write_attribute(trigger_gc);
  37 write_attribute(prune_cache);
  38 write_attribute(flash_vol_create);
  39
  40 read_attribute(bucket_size);
  41 read_attribute(block_size);
  42 read_attribute(nbuckets);
  43 read_attribute(tree_depth);
  44 read_attribute(root_usage_percent);
  45 read_attribute(priority_stats);
  46 read_attribute(btree_cache_size);
  47 read_attribute(btree_cache_max_chain);
  48 read_attribute(cache_available_percent);
  49 read_attribute(written);
  50 read_attribute(btree_written);
  51 read_attribute(metadata_written);
  52 read_attribute(active_journal_entries);
  53
  54 sysfs_time_stats_attribute(btree_gc,    sec, ms);
  55 sysfs_time_stats_attribute(btree_split, sec, us);
  56 sysfs_time_stats_attribute(btree_sort,  ms,  us);
  57 sysfs_time_stats_attribute(btree_read,  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         ssize_t 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                 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 bch_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, bch_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 static unsigned bch_root_usage(struct cache_set *c)
 446 {
 447         unsigned bytes = 0;
 448         struct bkey *k;
 449         struct btree *b;
 450         struct btree_iter iter;
 451
 452         goto lock_root;
 453
 454         do {
 455                 rw_unlock(false, b);
 456 lock_root:
 457                 b = c->root;
 458                 rw_lock(false, b, b->level);
 459         } while (b != c->root);
 460
 461         for_each_key_filter(&b->keys, k, &iter, bch_ptr_bad)
 462                 bytes += bkey_bytes(k);
 463
 464         rw_unlock(false, b);
 465
 466         return (bytes * 100) / btree_bytes(c);
 467 }
 468
 469 static size_t bch_cache_size(struct cache_set *c)
 470 {
 471         size_t ret = 0;
 472         struct btree *b;
 473
 474         mutex_lock(&c->bucket_lock);
 475         list_for_each_entry(b, &c->btree_cache, list)
 476                 ret += 1 << (b->keys.page_order + PAGE_SHIFT);
 477
 478         mutex_unlock(&c->bucket_lock);
 479         return ret;
 480 }
 481
 482 static unsigned bch_cache_max_chain(struct cache_set *c)
 483 {
 484         unsigned ret = 0;
 485         struct hlist_head *h;
 486
 487         mutex_lock(&c->bucket_lock);
 488
 489         for (h = c->bucket_hash;
 490              h < c->bucket_hash + (1 << BUCKET_HASH_BITS);
 491              h++) {
 492                 unsigned i = 0;
 493                 struct hlist_node *p;
 494
 495                 hlist_for_each(p, h)
 496                         i++;
 497
 498                 ret = max(ret, i);
 499         }
 500
 501         mutex_unlock(&c->bucket_lock);
 502         return ret;
 503 }
 504
 505 static unsigned bch_btree_used(struct cache_set *c)
 506 {
 507         return div64_u64(c->gc_stats.key_bytes * 100,
 508                          (c->gc_stats.nodes ?: 1) * btree_bytes(c));
 509 }
 510
 511 static unsigned bch_average_key_size(struct cache_set *c)
 512 {
 513         return c->gc_stats.nkeys
 514                 ? div64_u64(c->gc_stats.data, c->gc_stats.nkeys)
 515                 : 0;
 516 }
 517
 518 SHOW(__bch_cache_set)
 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,         bch_root_usage(c));
 528
 529         sysfs_hprint(btree_cache_size,          bch_cache_size(c));
 530         sysfs_print(btree_cache_max_chain,      bch_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
 538         sysfs_print(btree_used_percent, bch_btree_used(c));
 539         sysfs_print(btree_nodes,        c->gc_stats.nodes);
 540         sysfs_hprint(average_key_size,  bch_average_key_size(c));
 541
 542         sysfs_print(cache_read_races,
 543                     atomic_long_read(&c->cache_read_races));
 544
 545         sysfs_print(writeback_keys_done,
 546                     atomic_long_read(&c->writeback_keys_done));
 547         sysfs_print(writeback_keys_failed,
 548                     atomic_long_read(&c->writeback_keys_failed));
 549
 550         if (attr == &sysfs_errors)
 551                 return bch_snprint_string_list(buf, PAGE_SIZE, error_actions,
 552                                                c->on_error);
 553
 554         /* See count_io_errors for why 88 */
 555         sysfs_print(io_error_halflife,  c->error_decay * 88);
 556         sysfs_print(io_error_limit,     c->error_limit >> IO_ERROR_SHIFT);
 557
 558         sysfs_hprint(congested,
 559                      ((uint64_t) bch_get_congested(c)) << 9);
 560         sysfs_print(congested_read_threshold_us,
 561                     c->congested_read_threshold_us);
 562         sysfs_print(congested_write_threshold_us,
 563                     c->congested_write_threshold_us);
 564
 565         sysfs_print(active_journal_entries,     fifo_used(&c->journal.pin));
 566         sysfs_printf(verify,                    "%i", c->verify);
 567         sysfs_printf(key_merging_disabled,      "%i", c->key_merging_disabled);
 568         sysfs_printf(expensive_debug_checks,
 569                      "%i", c->expensive_debug_checks);
 570         sysfs_printf(gc_always_rewrite,         "%i", c->gc_always_rewrite);
 571         sysfs_printf(btree_shrinker_disabled,   "%i", c->shrinker_disabled);
 572         sysfs_printf(copy_gc_enabled,           "%i", c->copy_gc_enabled);
 573
 574         if (attr == &sysfs_bset_tree_stats)
 575                 return bch_bset_print_stats(c, buf);
 576
 577         return 0;
 578 }
 579 SHOW_LOCKED(bch_cache_set)
 580
 581 STORE(__bch_cache_set)
 582 {
 583         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 584
 585         if (attr == &sysfs_unregister)
 586                 bch_cache_set_unregister(c);
 587
 588         if (attr == &sysfs_stop)
 589                 bch_cache_set_stop(c);
 590
 591         if (attr == &sysfs_synchronous) {
 592                 bool sync = strtoul_or_return(buf);
 593
 594                 if (sync != CACHE_SYNC(&c->sb)) {
 595                         SET_CACHE_SYNC(&c->sb, sync);
 596                         bcache_write_super(c);
 597                 }
 598         }
 599
 600         if (attr == &sysfs_flash_vol_create) {
 601                 int r;
 602                 uint64_t v;
 603                 strtoi_h_or_return(buf, v);
 604
 605                 r = bch_flash_dev_create(c, v);
 606                 if (r)
 607                         return r;
 608         }
 609
 610         if (attr == &sysfs_clear_stats) {
 611                 atomic_long_set(&c->writeback_keys_done,        0);
 612                 atomic_long_set(&c->writeback_keys_failed,      0);
 613
 614                 memset(&c->gc_stats, 0, sizeof(struct gc_stat));
 615                 bch_cache_accounting_clear(&c->accounting);
 616         }
 617
 618         if (attr == &sysfs_trigger_gc) {
 619                 /*
 620                  * Garbage collection thread only works when sectors_to_gc < 0,
 621                  * when users write to sysfs entry trigger_gc, most of time
 622                  * they want to forcibly triger gargage collection. Here -1 is
 623                  * set to c->sectors_to_gc, to make gc_should_run() give a
 624                  * chance to permit gc thread to run. "give a chance" means
 625                  * before going into gc_should_run(), there is still chance
 626                  * that c->sectors_to_gc being set to other positive value. So
 627                  * writing sysfs entry trigger_gc won't always make sure gc
 628                  * thread takes effect.
 629                  */
 630                 atomic_set(&c->sectors_to_gc, -1);
 631                 wake_up_gc(c);
 632         }
 633
 634         if (attr == &sysfs_prune_cache) {
 635                 struct shrink_control sc;
 636                 sc.gfp_mask = GFP_KERNEL;
 637                 sc.nr_to_scan = strtoul_or_return(buf);
 638                 c->shrink.scan_objects(&c->shrink, &sc);
 639         }
 640
 641         sysfs_strtoul(congested_read_threshold_us,
 642                       c->congested_read_threshold_us);
 643         sysfs_strtoul(congested_write_threshold_us,
 644                       c->congested_write_threshold_us);
 645
 646         if (attr == &sysfs_errors) {
 647                 ssize_t v = bch_read_string_list(buf, error_actions);
 648
 649                 if (v < 0)
 650                         return v;
 651
 652                 c->on_error = v;
 653         }
 654
 655         if (attr == &sysfs_io_error_limit)
 656                 c->error_limit = strtoul_or_return(buf) << IO_ERROR_SHIFT;
 657
 658         /* See count_io_errors() for why 88 */
 659         if (attr == &sysfs_io_error_halflife)
 660                 c->error_decay = strtoul_or_return(buf) / 88;
 661
 662         sysfs_strtoul(journal_delay_ms,         c->journal_delay_ms);
 663         sysfs_strtoul(verify,                   c->verify);
 664         sysfs_strtoul(key_merging_disabled,     c->key_merging_disabled);
 665         sysfs_strtoul(expensive_debug_checks,   c->expensive_debug_checks);
 666         sysfs_strtoul(gc_always_rewrite,        c->gc_always_rewrite);
 667         sysfs_strtoul(btree_shrinker_disabled,  c->shrinker_disabled);
 668         sysfs_strtoul(copy_gc_enabled,          c->copy_gc_enabled);
 669
 670         return size;
 671 }
 672 STORE_LOCKED(bch_cache_set)
 673
 674 SHOW(bch_cache_set_internal)
 675 {
 676         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 677         return bch_cache_set_show(&c->kobj, attr, buf);
 678 }
 679
 680 STORE(bch_cache_set_internal)
 681 {
 682         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 683         return bch_cache_set_store(&c->kobj, attr, buf, size);
 684 }
 685
 686 static void bch_cache_set_internal_release(struct kobject *k)
 687 {
 688 }
 689
 690 static struct attribute *bch_cache_set_files[] = {
 691         &sysfs_unregister,
 692         &sysfs_stop,
 693         &sysfs_synchronous,
 694         &sysfs_journal_delay_ms,
 695         &sysfs_flash_vol_create,
 696
 697         &sysfs_bucket_size,
 698         &sysfs_block_size,
 699         &sysfs_tree_depth,
 700         &sysfs_root_usage_percent,
 701         &sysfs_btree_cache_size,
 702         &sysfs_cache_available_percent,
 703
 704         &sysfs_average_key_size,
 705
 706         &sysfs_errors,
 707         &sysfs_io_error_limit,
 708         &sysfs_io_error_halflife,
 709         &sysfs_congested,
 710         &sysfs_congested_read_threshold_us,
 711         &sysfs_congested_write_threshold_us,
 712         &sysfs_clear_stats,
 713         NULL
 714 };
 715 KTYPE(bch_cache_set);
 716
 717 static struct attribute *bch_cache_set_internal_files[] = {
 718         &sysfs_active_journal_entries,
 719
 720         sysfs_time_stats_attribute_list(btree_gc, sec, ms)
 721         sysfs_time_stats_attribute_list(btree_split, sec, us)
 722         sysfs_time_stats_attribute_list(btree_sort, ms, us)
 723         sysfs_time_stats_attribute_list(btree_read, ms, us)
 724
 725         &sysfs_btree_nodes,
 726         &sysfs_btree_used_percent,
 727         &sysfs_btree_cache_max_chain,
 728
 729         &sysfs_bset_tree_stats,
 730         &sysfs_cache_read_races,
 731         &sysfs_writeback_keys_done,
 732         &sysfs_writeback_keys_failed,
 733
 734         &sysfs_trigger_gc,
 735         &sysfs_prune_cache,
 736 #ifdef CONFIG_BCACHE_DEBUG
 737         &sysfs_verify,
 738         &sysfs_key_merging_disabled,
 739         &sysfs_expensive_debug_checks,
 740 #endif
 741         &sysfs_gc_always_rewrite,
 742         &sysfs_btree_shrinker_disabled,
 743         &sysfs_copy_gc_enabled,
 744         NULL
 745 };
 746 KTYPE(bch_cache_set_internal);
 747
 748 SHOW(__bch_cache)
 749 {
 750         struct cache *ca = container_of(kobj, struct cache, kobj);
 751
 752         sysfs_hprint(bucket_size,       bucket_bytes(ca));
 753         sysfs_hprint(block_size,        block_bytes(ca));
 754         sysfs_print(nbuckets,           ca->sb.nbuckets);
 755         sysfs_print(discard,            ca->discard);
 756         sysfs_hprint(written, atomic_long_read(&ca->sectors_written) << 9);
 757         sysfs_hprint(btree_written,
 758                      atomic_long_read(&ca->btree_sectors_written) << 9);
 759         sysfs_hprint(metadata_written,
 760                      (atomic_long_read(&ca->meta_sectors_written) +
 761                       atomic_long_read(&ca->btree_sectors_written)) << 9);
 762
 763         sysfs_print(io_errors,
 764                     atomic_read(&ca->io_errors) >> IO_ERROR_SHIFT);
 765
 766         if (attr == &sysfs_cache_replacement_policy)
 767                 return bch_snprint_string_list(buf, PAGE_SIZE,
 768                                                cache_replacement_policies,
 769                                                CACHE_REPLACEMENT(&ca->sb));
 770
 771         if (attr == &sysfs_priority_stats) {
 772                 int cmp(const void *l, const void *r)
 773                 {       return *((uint16_t *) r) - *((uint16_t *) l); }
 774
 775                 struct bucket *b;
 776                 size_t n = ca->sb.nbuckets, i;
 777                 size_t unused = 0, available = 0, dirty = 0, meta = 0;
 778                 uint64_t sum = 0;
 779                 /* Compute 31 quantiles */
 780                 uint16_t q[31], *p, *cached;
 781                 ssize_t ret;
 782
 783                 cached = p = vmalloc(ca->sb.nbuckets * sizeof(uint16_t));
 784                 if (!p)
 785                         return -ENOMEM;
 786
 787                 mutex_lock(&ca->set->bucket_lock);
 788                 for_each_bucket(b, ca) {
 789                         if (!GC_SECTORS_USED(b))
 790                                 unused++;
 791                         if (GC_MARK(b) == GC_MARK_RECLAIMABLE)
 792                                 available++;
 793                         if (GC_MARK(b) == GC_MARK_DIRTY)
 794                                 dirty++;
 795                         if (GC_MARK(b) == GC_MARK_METADATA)
 796                                 meta++;
 797                 }
 798
 799                 for (i = ca->sb.first_bucket; i < n; i++)
 800                         p[i] = ca->buckets[i].prio;
 801                 mutex_unlock(&ca->set->bucket_lock);
 802
 803                 sort(p, n, sizeof(uint16_t), cmp, NULL);
 804
 805                 while (n &&
 806                        !cached[n - 1])
 807                         --n;
 808
 809                 unused = ca->sb.nbuckets - n;
 810
 811                 while (cached < p + n &&
 812                        *cached == BTREE_PRIO)
 813                         cached++, n--;
 814
 815                 for (i = 0; i < n; i++)
 816                         sum += INITIAL_PRIO - cached[i];
 817
 818                 if (n)
 819                         do_div(sum, n);
 820
 821                 for (i = 0; i < ARRAY_SIZE(q); i++)
 822                         q[i] = INITIAL_PRIO - cached[n * (i + 1) /
 823                                 (ARRAY_SIZE(q) + 1)];
 824
 825                 vfree(p);
 826
 827                 ret = scnprintf(buf, PAGE_SIZE,
 828                                 "Unused:                %zu%%\n"
 829                                 "Clean:         %zu%%\n"
 830                                 "Dirty:         %zu%%\n"
 831                                 "Metadata:      %zu%%\n"
 832                                 "Average:       %llu\n"
 833                                 "Sectors per Q: %zu\n"
 834                                 "Quantiles:     [",
 835                                 unused * 100 / (size_t) ca->sb.nbuckets,
 836                                 available * 100 / (size_t) ca->sb.nbuckets,
 837                                 dirty * 100 / (size_t) ca->sb.nbuckets,
 838                                 meta * 100 / (size_t) ca->sb.nbuckets, sum,
 839                                 n * ca->sb.bucket_size / (ARRAY_SIZE(q) + 1));
 840
 841                 for (i = 0; i < ARRAY_SIZE(q); i++)
 842                         ret += scnprintf(buf + ret, PAGE_SIZE - ret,
 843                                          "%u ", q[i]);
 844                 ret--;
 845
 846                 ret += scnprintf(buf + ret, PAGE_SIZE - ret, "]\n");
 847
 848                 return ret;
 849         }
 850
 851         return 0;
 852 }
 853 SHOW_LOCKED(bch_cache)
 854
 855 STORE(__bch_cache)
 856 {
 857         struct cache *ca = container_of(kobj, struct cache, kobj);
 858
 859         if (attr == &sysfs_discard) {
 860                 bool v = strtoul_or_return(buf);
 861
 862                 if (blk_queue_discard(bdev_get_queue(ca->bdev)))
 863                         ca->discard = v;
 864
 865                 if (v != CACHE_DISCARD(&ca->sb)) {
 866                         SET_CACHE_DISCARD(&ca->sb, v);
 867                         bcache_write_super(ca->set);
 868                 }
 869         }
 870
 871         if (attr == &sysfs_cache_replacement_policy) {
 872                 ssize_t v = bch_read_string_list(buf, cache_replacement_policies);
 873
 874                 if (v < 0)
 875                         return v;
 876
 877                 if ((unsigned) v != CACHE_REPLACEMENT(&ca->sb)) {
 878                         mutex_lock(&ca->set->bucket_lock);
 879                         SET_CACHE_REPLACEMENT(&ca->sb, v);
 880                         mutex_unlock(&ca->set->bucket_lock);
 881
 882                         bcache_write_super(ca->set);
 883                 }
 884         }
 885
 886         if (attr == &sysfs_clear_stats) {
 887                 atomic_long_set(&ca->sectors_written, 0);
 888                 atomic_long_set(&ca->btree_sectors_written, 0);
 889                 atomic_long_set(&ca->meta_sectors_written, 0);
 890                 atomic_set(&ca->io_count, 0);
 891                 atomic_set(&ca->io_errors, 0);
 892         }
 893
 894         return size;
 895 }
 896 STORE_LOCKED(bch_cache)
 897
 898 static struct attribute *bch_cache_files[] = {
 899         &sysfs_bucket_size,
 900         &sysfs_block_size,
 901         &sysfs_nbuckets,
 902         &sysfs_priority_stats,
 903         &sysfs_discard,
 904         &sysfs_written,
 905         &sysfs_btree_written,
 906         &sysfs_metadata_written,
 907         &sysfs_io_errors,
 908         &sysfs_clear_stats,
 909         &sysfs_cache_replacement_policy,
 910         NULL
 911 };
 912 KTYPE(bch_cache);