drivers/md/bcache/sysfs.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * bcache sysfs interfaces
   4  *
   5  * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
   6  * Copyright 2012 Google, Inc.
   7  */
   8
   9 #include "bcache.h"
  10 #include "sysfs.h"
  11 #include "btree.h"
  12 #include "request.h"
  13 #include "writeback.h"
  14
  15 #include <linux/blkdev.h>
  16 #include <linux/sort.h>
  17 #include <linux/sched/clock.h>
  18
  19 static const char * const cache_replacement_policies[] = {
  20         "lru",
  21         "fifo",
  22         "random",
  23         NULL
  24 };
  25
  26 static const char * const error_actions[] = {
  27         "unregister",
  28         "panic",
  29         NULL
  30 };
  31
  32 write_attribute(attach);
  33 write_attribute(detach);
  34 write_attribute(unregister);
  35 write_attribute(stop);
  36 write_attribute(clear_stats);
  37 write_attribute(trigger_gc);
  38 write_attribute(prune_cache);
  39 write_attribute(flash_vol_create);
  40
  41 read_attribute(bucket_size);
  42 read_attribute(block_size);
  43 read_attribute(nbuckets);
  44 read_attribute(tree_depth);
  45 read_attribute(root_usage_percent);
  46 read_attribute(priority_stats);
  47 read_attribute(btree_cache_size);
  48 read_attribute(btree_cache_max_chain);
  49 read_attribute(cache_available_percent);
  50 read_attribute(written);
  51 read_attribute(btree_written);
  52 read_attribute(metadata_written);
  53 read_attribute(active_journal_entries);
  54
  55 sysfs_time_stats_attribute(btree_gc,    sec, ms);
  56 sysfs_time_stats_attribute(btree_split, sec, us);
  57 sysfs_time_stats_attribute(btree_sort,  ms,  us);
  58 sysfs_time_stats_attribute(btree_read,  ms,  us);
  59
  60 read_attribute(btree_nodes);
  61 read_attribute(btree_used_percent);
  62 read_attribute(average_key_size);
  63 read_attribute(dirty_data);
  64 read_attribute(bset_tree_stats);
  65
  66 read_attribute(state);
  67 read_attribute(cache_read_races);
  68 read_attribute(reclaim);
  69 read_attribute(flush_write);
  70 read_attribute(retry_flush_write);
  71 read_attribute(writeback_keys_done);
  72 read_attribute(writeback_keys_failed);
  73 read_attribute(io_errors);
  74 read_attribute(congested);
  75 rw_attribute(congested_read_threshold_us);
  76 rw_attribute(congested_write_threshold_us);
  77
  78 rw_attribute(sequential_cutoff);
  79 rw_attribute(data_csum);
  80 rw_attribute(cache_mode);
  81 rw_attribute(writeback_metadata);
  82 rw_attribute(writeback_running);
  83 rw_attribute(writeback_percent);
  84 rw_attribute(writeback_delay);
  85 rw_attribute(writeback_rate);
  86
  87 rw_attribute(writeback_rate_update_seconds);
  88 rw_attribute(writeback_rate_i_term_inverse);
  89 rw_attribute(writeback_rate_p_term_inverse);
  90 rw_attribute(writeback_rate_minimum);
  91 read_attribute(writeback_rate_debug);
  92
  93 read_attribute(stripe_size);
  94 read_attribute(partial_stripes_expensive);
  95
  96 rw_attribute(synchronous);
  97 rw_attribute(journal_delay_ms);
  98 rw_attribute(discard);
  99 rw_attribute(running);
 100 rw_attribute(label);
 101 rw_attribute(readahead);
 102 rw_attribute(errors);
 103 rw_attribute(io_error_limit);
 104 rw_attribute(io_error_halflife);
 105 rw_attribute(verify);
 106 rw_attribute(bypass_torture_test);
 107 rw_attribute(key_merging_disabled);
 108 rw_attribute(gc_always_rewrite);
 109 rw_attribute(expensive_debug_checks);
 110 rw_attribute(cache_replacement_policy);
 111 rw_attribute(btree_shrinker_disabled);
 112 rw_attribute(copy_gc_enabled);
 113 rw_attribute(size);
 114
 115 SHOW(__bch_cached_dev)
 116 {
 117         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 118                                              disk.kobj);
 119         const char *states[] = { "no cache", "clean", "dirty", "inconsistent" };
 120
 121 #define var(stat)               (dc->stat)
 122
 123         if (attr == &sysfs_cache_mode)
 124                 return bch_snprint_string_list(buf, PAGE_SIZE,
 125                                                bch_cache_modes + 1,
 126                                                BDEV_CACHE_MODE(&dc->sb));
 127
 128         sysfs_printf(data_csum,         "%i", dc->disk.data_csum);
 129         var_printf(verify,              "%i");
 130         var_printf(bypass_torture_test, "%i");
 131         var_printf(writeback_metadata,  "%i");
 132         var_printf(writeback_running,   "%i");
 133         var_print(writeback_delay);
 134         var_print(writeback_percent);
 135         sysfs_hprint(writeback_rate,    dc->writeback_rate.rate << 9);
 136
 137         var_print(writeback_rate_update_seconds);
 138         var_print(writeback_rate_i_term_inverse);
 139         var_print(writeback_rate_p_term_inverse);
 140         var_print(writeback_rate_minimum);
 141
 142         if (attr == &sysfs_writeback_rate_debug) {
 143                 char rate[20];
 144                 char dirty[20];
 145                 char target[20];
 146                 char proportional[20];
 147                 char integral[20];
 148                 char change[20];
 149                 s64 next_io;
 150
 151                 bch_hprint(rate,        dc->writeback_rate.rate << 9);
 152                 bch_hprint(dirty,       bcache_dev_sectors_dirty(&dc->disk) << 9);
 153                 bch_hprint(target,      dc->writeback_rate_target << 9);
 154                 bch_hprint(proportional,dc->writeback_rate_proportional << 9);
 155                 bch_hprint(integral,    dc->writeback_rate_integral_scaled << 9);
 156                 bch_hprint(change,      dc->writeback_rate_change << 9);
 157
 158                 next_io = div64_s64(dc->writeback_rate.next - local_clock(),
 159                                     NSEC_PER_MSEC);
 160
 161                 return sprintf(buf,
 162                                "rate:\t\t%s/sec\n"
 163                                "dirty:\t\t%s\n"
 164                                "target:\t\t%s\n"
 165                                "proportional:\t%s\n"
 166                                "integral:\t%s\n"
 167                                "change:\t\t%s/sec\n"
 168                                "next io:\t%llims\n",
 169                                rate, dirty, target, proportional,
 170                                integral, change, next_io);
 171         }
 172
 173         sysfs_hprint(dirty_data,
 174                      bcache_dev_sectors_dirty(&dc->disk) << 9);
 175
 176         sysfs_hprint(stripe_size,       dc->disk.stripe_size << 9);
 177         var_printf(partial_stripes_expensive,   "%u");
 178
 179         var_hprint(sequential_cutoff);
 180         var_hprint(readahead);
 181
 182         sysfs_print(running,            atomic_read(&dc->running));
 183         sysfs_print(state,              states[BDEV_STATE(&dc->sb)]);
 184
 185         if (attr == &sysfs_label) {
 186                 memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
 187                 buf[SB_LABEL_SIZE + 1] = '\0';
 188                 strcat(buf, "\n");
 189                 return strlen(buf);
 190         }
 191
 192 #undef var
 193         return 0;
 194 }
 195 SHOW_LOCKED(bch_cached_dev)
 196
 197 STORE(__cached_dev)
 198 {
 199         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 200                                              disk.kobj);
 201         ssize_t v = size;
 202         struct cache_set *c;
 203         struct kobj_uevent_env *env;
 204
 205 #define d_strtoul(var)          sysfs_strtoul(var, dc->var)
 206 #define d_strtoul_nonzero(var)  sysfs_strtoul_clamp(var, dc->var, 1, INT_MAX)
 207 #define d_strtoi_h(var)         sysfs_hatoi(var, dc->var)
 208
 209         sysfs_strtoul(data_csum,        dc->disk.data_csum);
 210         d_strtoul(verify);
 211         d_strtoul(bypass_torture_test);
 212         d_strtoul(writeback_metadata);
 213         d_strtoul(writeback_running);
 214         d_strtoul(writeback_delay);
 215
 216         sysfs_strtoul_clamp(writeback_percent, dc->writeback_percent, 0, 40);
 217
 218         sysfs_strtoul_clamp(writeback_rate,
 219                             dc->writeback_rate.rate, 1, INT_MAX);
 220
 221         sysfs_strtoul_clamp(writeback_rate_update_seconds,
 222                             dc->writeback_rate_update_seconds,
 223                             1, WRITEBACK_RATE_UPDATE_SECS_MAX);
 224         d_strtoul(writeback_rate_i_term_inverse);
 225         d_strtoul_nonzero(writeback_rate_p_term_inverse);
 226
 227         d_strtoi_h(sequential_cutoff);
 228         d_strtoi_h(readahead);
 229
 230         if (attr == &sysfs_clear_stats)
 231                 bch_cache_accounting_clear(&dc->accounting);
 232
 233         if (attr == &sysfs_running &&
 234             strtoul_or_return(buf))
 235                 bch_cached_dev_run(dc);
 236
 237         if (attr == &sysfs_cache_mode) {
 238                 v = bch_read_string_list(buf, bch_cache_modes + 1);
 239
 240                 if (v < 0)
 241                         return v;
 242
 243                 if ((unsigned) v != BDEV_CACHE_MODE(&dc->sb)) {
 244                         SET_BDEV_CACHE_MODE(&dc->sb, v);
 245                         bch_write_bdev_super(dc, NULL);
 246                 }
 247         }
 248
 249         if (attr == &sysfs_label) {
 250                 if (size > SB_LABEL_SIZE)
 251                         return -EINVAL;
 252                 memcpy(dc->sb.label, buf, size);
 253                 if (size < SB_LABEL_SIZE)
 254                         dc->sb.label[size] = '\0';
 255                 if (size && dc->sb.label[size - 1] == '\n')
 256                         dc->sb.label[size - 1] = '\0';
 257                 bch_write_bdev_super(dc, NULL);
 258                 if (dc->disk.c) {
 259                         memcpy(dc->disk.c->uuids[dc->disk.id].label,
 260                                buf, SB_LABEL_SIZE);
 261                         bch_uuid_write(dc->disk.c);
 262                 }
 263                 env = kzalloc(sizeof(struct kobj_uevent_env), GFP_KERNEL);
 264                 if (!env)
 265                         return -ENOMEM;
 266                 add_uevent_var(env, "DRIVER=bcache");
 267                 add_uevent_var(env, "CACHED_UUID=%pU", dc->sb.uuid),
 268                 add_uevent_var(env, "CACHED_LABEL=%s", buf);
 269                 kobject_uevent_env(
 270                         &disk_to_dev(dc->disk.disk)->kobj, KOBJ_CHANGE, env->envp);
 271                 kfree(env);
 272         }
 273
 274         if (attr == &sysfs_attach) {
 275                 if (bch_parse_uuid(buf, dc->sb.set_uuid) < 16)
 276                         return -EINVAL;
 277
 278                 list_for_each_entry(c, &bch_cache_sets, list) {
 279                         v = bch_cached_dev_attach(dc, c);
 280                         if (!v)
 281                                 return size;
 282                 }
 283
 284                 pr_err("Can't attach %s: cache set not found", buf);
 285                 size = v;
 286         }
 287
 288         if (attr == &sysfs_detach && dc->disk.c)
 289                 bch_cached_dev_detach(dc);
 290
 291         if (attr == &sysfs_stop)
 292                 bcache_device_stop(&dc->disk);
 293
 294         return size;
 295 }
 296
 297 STORE(bch_cached_dev)
 298 {
 299         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 300                                              disk.kobj);
 301
 302         mutex_lock(&bch_register_lock);
 303         size = __cached_dev_store(kobj, attr, buf, size);
 304
 305         if (attr == &sysfs_writeback_running)
 306                 bch_writeback_queue(dc);
 307
 308         if (attr == &sysfs_writeback_percent)
 309                 schedule_delayed_work(&dc->writeback_rate_update,
 310                                       dc->writeback_rate_update_seconds * HZ);
 311
 312         mutex_unlock(&bch_register_lock);
 313         return size;
 314 }
 315
 316 static struct attribute *bch_cached_dev_files[] = {
 317         &sysfs_attach,
 318         &sysfs_detach,
 319         &sysfs_stop,
 320 #if 0
 321         &sysfs_data_csum,
 322 #endif
 323         &sysfs_cache_mode,
 324         &sysfs_writeback_metadata,
 325         &sysfs_writeback_running,
 326         &sysfs_writeback_delay,
 327         &sysfs_writeback_percent,
 328         &sysfs_writeback_rate,
 329         &sysfs_writeback_rate_update_seconds,
 330         &sysfs_writeback_rate_i_term_inverse,
 331         &sysfs_writeback_rate_p_term_inverse,
 332         &sysfs_writeback_rate_debug,
 333         &sysfs_dirty_data,
 334         &sysfs_stripe_size,
 335         &sysfs_partial_stripes_expensive,
 336         &sysfs_sequential_cutoff,
 337         &sysfs_clear_stats,
 338         &sysfs_running,
 339         &sysfs_state,
 340         &sysfs_label,
 341         &sysfs_readahead,
 342 #ifdef CONFIG_BCACHE_DEBUG
 343         &sysfs_verify,
 344         &sysfs_bypass_torture_test,
 345 #endif
 346         NULL
 347 };
 348 KTYPE(bch_cached_dev);
 349
 350 SHOW(bch_flash_dev)
 351 {
 352         struct bcache_device *d = container_of(kobj, struct bcache_device,
 353                                                kobj);
 354         struct uuid_entry *u = &d->c->uuids[d->id];
 355
 356         sysfs_printf(data_csum, "%i", d->data_csum);
 357         sysfs_hprint(size,      u->sectors << 9);
 358
 359         if (attr == &sysfs_label) {
 360                 memcpy(buf, u->label, SB_LABEL_SIZE);
 361                 buf[SB_LABEL_SIZE + 1] = '\0';
 362                 strcat(buf, "\n");
 363                 return strlen(buf);
 364         }
 365
 366         return 0;
 367 }
 368
 369 STORE(__bch_flash_dev)
 370 {
 371         struct bcache_device *d = container_of(kobj, struct bcache_device,
 372                                                kobj);
 373         struct uuid_entry *u = &d->c->uuids[d->id];
 374
 375         sysfs_strtoul(data_csum,        d->data_csum);
 376
 377         if (attr == &sysfs_size) {
 378                 uint64_t v;
 379                 strtoi_h_or_return(buf, v);
 380
 381                 u->sectors = v >> 9;
 382                 bch_uuid_write(d->c);
 383                 set_capacity(d->disk, u->sectors);
 384         }
 385
 386         if (attr == &sysfs_label) {
 387                 memcpy(u->label, buf, SB_LABEL_SIZE);
 388                 bch_uuid_write(d->c);
 389         }
 390
 391         if (attr == &sysfs_unregister) {
 392                 set_bit(BCACHE_DEV_DETACHING, &d->flags);
 393                 bcache_device_stop(d);
 394         }
 395
 396         return size;
 397 }
 398 STORE_LOCKED(bch_flash_dev)
 399
 400 static struct attribute *bch_flash_dev_files[] = {
 401         &sysfs_unregister,
 402 #if 0
 403         &sysfs_data_csum,
 404 #endif
 405         &sysfs_label,
 406         &sysfs_size,
 407         NULL
 408 };
 409 KTYPE(bch_flash_dev);
 410
 411 struct bset_stats_op {
 412         struct btree_op op;
 413         size_t nodes;
 414         struct bset_stats stats;
 415 };
 416
 417 static int bch_btree_bset_stats(struct btree_op *b_op, struct btree *b)
 418 {
 419         struct bset_stats_op *op = container_of(b_op, struct bset_stats_op, op);
 420
 421         op->nodes++;
 422         bch_btree_keys_stats(&b->keys, &op->stats);
 423
 424         return MAP_CONTINUE;
 425 }
 426
 427 static int bch_bset_print_stats(struct cache_set *c, char *buf)
 428 {
 429         struct bset_stats_op op;
 430         int ret;
 431
 432         memset(&op, 0, sizeof(op));
 433         bch_btree_op_init(&op.op, -1);
 434
 435         ret = bch_btree_map_nodes(&op.op, c, &ZERO_KEY, bch_btree_bset_stats);
 436         if (ret < 0)
 437                 return ret;
 438
 439         return snprintf(buf, PAGE_SIZE,
 440                         "btree nodes:           %zu\n"
 441                         "written sets:          %zu\n"
 442                         "unwritten sets:                %zu\n"
 443                         "written key bytes:     %zu\n"
 444                         "unwritten key bytes:   %zu\n"
 445                         "floats:                        %zu\n"
 446                         "failed:                        %zu\n",
 447                         op.nodes,
 448                         op.stats.sets_written, op.stats.sets_unwritten,
 449                         op.stats.bytes_written, op.stats.bytes_unwritten,
 450                         op.stats.floats, op.stats.failed);
 451 }
 452
 453 static unsigned bch_root_usage(struct cache_set *c)
 454 {
 455         unsigned bytes = 0;
 456         struct bkey *k;
 457         struct btree *b;
 458         struct btree_iter iter;
 459
 460         goto lock_root;
 461
 462         do {
 463                 rw_unlock(false, b);
 464 lock_root:
 465                 b = c->root;
 466                 rw_lock(false, b, b->level);
 467         } while (b != c->root);
 468
 469         for_each_key_filter(&b->keys, k, &iter, bch_ptr_bad)
 470                 bytes += bkey_bytes(k);
 471
 472         rw_unlock(false, b);
 473
 474         return (bytes * 100) / btree_bytes(c);
 475 }
 476
 477 static size_t bch_cache_size(struct cache_set *c)
 478 {
 479         size_t ret = 0;
 480         struct btree *b;
 481
 482         mutex_lock(&c->bucket_lock);
 483         list_for_each_entry(b, &c->btree_cache, list)
 484                 ret += 1 << (b->keys.page_order + PAGE_SHIFT);
 485
 486         mutex_unlock(&c->bucket_lock);
 487         return ret;
 488 }
 489
 490 static unsigned bch_cache_max_chain(struct cache_set *c)
 491 {
 492         unsigned ret = 0;
 493         struct hlist_head *h;
 494
 495         mutex_lock(&c->bucket_lock);
 496
 497         for (h = c->bucket_hash;
 498              h < c->bucket_hash + (1 << BUCKET_HASH_BITS);
 499              h++) {
 500                 unsigned i = 0;
 501                 struct hlist_node *p;
 502
 503                 hlist_for_each(p, h)
 504                         i++;
 505
 506                 ret = max(ret, i);
 507         }
 508
 509         mutex_unlock(&c->bucket_lock);
 510         return ret;
 511 }
 512
 513 static unsigned bch_btree_used(struct cache_set *c)
 514 {
 515         return div64_u64(c->gc_stats.key_bytes * 100,
 516                          (c->gc_stats.nodes ?: 1) * btree_bytes(c));
 517 }
 518
 519 static unsigned bch_average_key_size(struct cache_set *c)
 520 {
 521         return c->gc_stats.nkeys
 522                 ? div64_u64(c->gc_stats.data, c->gc_stats.nkeys)
 523                 : 0;
 524 }
 525
 526 SHOW(__bch_cache_set)
 527 {
 528         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 529
 530         sysfs_print(synchronous,                CACHE_SYNC(&c->sb));
 531         sysfs_print(journal_delay_ms,           c->journal_delay_ms);
 532         sysfs_hprint(bucket_size,               bucket_bytes(c));
 533         sysfs_hprint(block_size,                block_bytes(c));
 534         sysfs_print(tree_depth,                 c->root->level);
 535         sysfs_print(root_usage_percent,         bch_root_usage(c));
 536
 537         sysfs_hprint(btree_cache_size,          bch_cache_size(c));
 538         sysfs_print(btree_cache_max_chain,      bch_cache_max_chain(c));
 539         sysfs_print(cache_available_percent,    100 - c->gc_stats.in_use);
 540
 541         sysfs_print_time_stats(&c->btree_gc_time,       btree_gc, sec, ms);
 542         sysfs_print_time_stats(&c->btree_split_time,    btree_split, sec, us);
 543         sysfs_print_time_stats(&c->sort.time,           btree_sort, ms, us);
 544         sysfs_print_time_stats(&c->btree_read_time,     btree_read, ms, us);
 545
 546         sysfs_print(btree_used_percent, bch_btree_used(c));
 547         sysfs_print(btree_nodes,        c->gc_stats.nodes);
 548         sysfs_hprint(average_key_size,  bch_average_key_size(c));
 549
 550         sysfs_print(cache_read_races,
 551                     atomic_long_read(&c->cache_read_races));
 552
 553         sysfs_print(reclaim,
 554                     atomic_long_read(&c->reclaim));
 555
 556         sysfs_print(flush_write,
 557                     atomic_long_read(&c->flush_write));
 558
 559         sysfs_print(retry_flush_write,
 560                     atomic_long_read(&c->retry_flush_write));
 561
 562         sysfs_print(writeback_keys_done,
 563                     atomic_long_read(&c->writeback_keys_done));
 564         sysfs_print(writeback_keys_failed,
 565                     atomic_long_read(&c->writeback_keys_failed));
 566
 567         if (attr == &sysfs_errors)
 568                 return bch_snprint_string_list(buf, PAGE_SIZE, error_actions,
 569                                                c->on_error);
 570
 571         /* See count_io_errors for why 88 */
 572         sysfs_print(io_error_halflife,  c->error_decay * 88);
 573         sysfs_print(io_error_limit,     c->error_limit);
 574
 575         sysfs_hprint(congested,
 576                      ((uint64_t) bch_get_congested(c)) << 9);
 577         sysfs_print(congested_read_threshold_us,
 578                     c->congested_read_threshold_us);
 579         sysfs_print(congested_write_threshold_us,
 580                     c->congested_write_threshold_us);
 581
 582         sysfs_print(active_journal_entries,     fifo_used(&c->journal.pin));
 583         sysfs_printf(verify,                    "%i", c->verify);
 584         sysfs_printf(key_merging_disabled,      "%i", c->key_merging_disabled);
 585         sysfs_printf(expensive_debug_checks,
 586                      "%i", c->expensive_debug_checks);
 587         sysfs_printf(gc_always_rewrite,         "%i", c->gc_always_rewrite);
 588         sysfs_printf(btree_shrinker_disabled,   "%i", c->shrinker_disabled);
 589         sysfs_printf(copy_gc_enabled,           "%i", c->copy_gc_enabled);
 590
 591         if (attr == &sysfs_bset_tree_stats)
 592                 return bch_bset_print_stats(c, buf);
 593
 594         return 0;
 595 }
 596 SHOW_LOCKED(bch_cache_set)
 597
 598 STORE(__bch_cache_set)
 599 {
 600         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 601
 602         if (attr == &sysfs_unregister)
 603                 bch_cache_set_unregister(c);
 604
 605         if (attr == &sysfs_stop)
 606                 bch_cache_set_stop(c);
 607
 608         if (attr == &sysfs_synchronous) {
 609                 bool sync = strtoul_or_return(buf);
 610
 611                 if (sync != CACHE_SYNC(&c->sb)) {
 612                         SET_CACHE_SYNC(&c->sb, sync);
 613                         bcache_write_super(c);
 614                 }
 615         }
 616
 617         if (attr == &sysfs_flash_vol_create) {
 618                 int r;
 619                 uint64_t v;
 620                 strtoi_h_or_return(buf, v);
 621
 622                 r = bch_flash_dev_create(c, v);
 623                 if (r)
 624                         return r;
 625         }
 626
 627         if (attr == &sysfs_clear_stats) {
 628                 atomic_long_set(&c->writeback_keys_done,        0);
 629                 atomic_long_set(&c->writeback_keys_failed,      0);
 630
 631                 memset(&c->gc_stats, 0, sizeof(struct gc_stat));
 632                 bch_cache_accounting_clear(&c->accounting);
 633         }
 634
 635         if (attr == &sysfs_trigger_gc) {
 636                 /*
 637                  * Garbage collection thread only works when sectors_to_gc < 0,
 638                  * when users write to sysfs entry trigger_gc, most of time
 639                  * they want to forcibly triger gargage collection. Here -1 is
 640                  * set to c->sectors_to_gc, to make gc_should_run() give a
 641                  * chance to permit gc thread to run. "give a chance" means
 642                  * before going into gc_should_run(), there is still chance
 643                  * that c->sectors_to_gc being set to other positive value. So
 644                  * writing sysfs entry trigger_gc won't always make sure gc
 645                  * thread takes effect.
 646                  */
 647                 atomic_set(&c->sectors_to_gc, -1);
 648                 wake_up_gc(c);
 649         }
 650
 651         if (attr == &sysfs_prune_cache) {
 652                 struct shrink_control sc;
 653                 sc.gfp_mask = GFP_KERNEL;
 654                 sc.nr_to_scan = strtoul_or_return(buf);
 655                 c->shrink.scan_objects(&c->shrink, &sc);
 656         }
 657
 658         sysfs_strtoul(congested_read_threshold_us,
 659                       c->congested_read_threshold_us);
 660         sysfs_strtoul(congested_write_threshold_us,
 661                       c->congested_write_threshold_us);
 662
 663         if (attr == &sysfs_errors) {
 664                 ssize_t v = bch_read_string_list(buf, error_actions);
 665
 666                 if (v < 0)
 667                         return v;
 668
 669                 c->on_error = v;
 670         }
 671
 672         if (attr == &sysfs_io_error_limit)
 673                 c->error_limit = strtoul_or_return(buf);
 674
 675         /* See count_io_errors() for why 88 */
 676         if (attr == &sysfs_io_error_halflife)
 677                 c->error_decay = strtoul_or_return(buf) / 88;
 678
 679         sysfs_strtoul(journal_delay_ms,         c->journal_delay_ms);
 680         sysfs_strtoul(verify,                   c->verify);
 681         sysfs_strtoul(key_merging_disabled,     c->key_merging_disabled);
 682         sysfs_strtoul(expensive_debug_checks,   c->expensive_debug_checks);
 683         sysfs_strtoul(gc_always_rewrite,        c->gc_always_rewrite);
 684         sysfs_strtoul(btree_shrinker_disabled,  c->shrinker_disabled);
 685         sysfs_strtoul(copy_gc_enabled,          c->copy_gc_enabled);
 686
 687         return size;
 688 }
 689 STORE_LOCKED(bch_cache_set)
 690
 691 SHOW(bch_cache_set_internal)
 692 {
 693         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 694         return bch_cache_set_show(&c->kobj, attr, buf);
 695 }
 696
 697 STORE(bch_cache_set_internal)
 698 {
 699         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 700         return bch_cache_set_store(&c->kobj, attr, buf, size);
 701 }
 702
 703 static void bch_cache_set_internal_release(struct kobject *k)
 704 {
 705 }
 706
 707 static struct attribute *bch_cache_set_files[] = {
 708         &sysfs_unregister,
 709         &sysfs_stop,
 710         &sysfs_synchronous,
 711         &sysfs_journal_delay_ms,
 712         &sysfs_flash_vol_create,
 713
 714         &sysfs_bucket_size,
 715         &sysfs_block_size,
 716         &sysfs_tree_depth,
 717         &sysfs_root_usage_percent,
 718         &sysfs_btree_cache_size,
 719         &sysfs_cache_available_percent,
 720
 721         &sysfs_average_key_size,
 722
 723         &sysfs_errors,
 724         &sysfs_io_error_limit,
 725         &sysfs_io_error_halflife,
 726         &sysfs_congested,
 727         &sysfs_congested_read_threshold_us,
 728         &sysfs_congested_write_threshold_us,
 729         &sysfs_clear_stats,
 730         NULL
 731 };
 732 KTYPE(bch_cache_set);
 733
 734 static struct attribute *bch_cache_set_internal_files[] = {
 735         &sysfs_active_journal_entries,
 736
 737         sysfs_time_stats_attribute_list(btree_gc, sec, ms)
 738         sysfs_time_stats_attribute_list(btree_split, sec, us)
 739         sysfs_time_stats_attribute_list(btree_sort, ms, us)
 740         sysfs_time_stats_attribute_list(btree_read, ms, us)
 741
 742         &sysfs_btree_nodes,
 743         &sysfs_btree_used_percent,
 744         &sysfs_btree_cache_max_chain,
 745
 746         &sysfs_bset_tree_stats,
 747         &sysfs_cache_read_races,
 748         &sysfs_reclaim,
 749         &sysfs_flush_write,
 750         &sysfs_retry_flush_write,
 751         &sysfs_writeback_keys_done,
 752         &sysfs_writeback_keys_failed,
 753
 754         &sysfs_trigger_gc,
 755         &sysfs_prune_cache,
 756 #ifdef CONFIG_BCACHE_DEBUG
 757         &sysfs_verify,
 758         &sysfs_key_merging_disabled,
 759         &sysfs_expensive_debug_checks,
 760 #endif
 761         &sysfs_gc_always_rewrite,
 762         &sysfs_btree_shrinker_disabled,
 763         &sysfs_copy_gc_enabled,
 764         NULL
 765 };
 766 KTYPE(bch_cache_set_internal);
 767
 768 static int __bch_cache_cmp(const void *l, const void *r)
 769 {
 770         return *((uint16_t *)r) - *((uint16_t *)l);
 771 }
 772
 773 SHOW(__bch_cache)
 774 {
 775         struct cache *ca = container_of(kobj, struct cache, kobj);
 776
 777         sysfs_hprint(bucket_size,       bucket_bytes(ca));
 778         sysfs_hprint(block_size,        block_bytes(ca));
 779         sysfs_print(nbuckets,           ca->sb.nbuckets);
 780         sysfs_print(discard,            ca->discard);
 781         sysfs_hprint(written, atomic_long_read(&ca->sectors_written) << 9);
 782         sysfs_hprint(btree_written,
 783                      atomic_long_read(&ca->btree_sectors_written) << 9);
 784         sysfs_hprint(metadata_written,
 785                      (atomic_long_read(&ca->meta_sectors_written) +
 786                       atomic_long_read(&ca->btree_sectors_written)) << 9);
 787
 788         sysfs_print(io_errors,
 789                     atomic_read(&ca->io_errors) >> IO_ERROR_SHIFT);
 790
 791         if (attr == &sysfs_cache_replacement_policy)
 792                 return bch_snprint_string_list(buf, PAGE_SIZE,
 793                                                cache_replacement_policies,
 794                                                CACHE_REPLACEMENT(&ca->sb));
 795
 796         if (attr == &sysfs_priority_stats) {
 797                 struct bucket *b;
 798                 size_t n = ca->sb.nbuckets, i;
 799                 size_t unused = 0, available = 0, dirty = 0, meta = 0;
 800                 uint64_t sum = 0;
 801                 /* Compute 31 quantiles */
 802                 uint16_t q[31], *p, *cached;
 803                 ssize_t ret;
 804
 805                 cached = p = vmalloc(ca->sb.nbuckets * sizeof(uint16_t));
 806                 if (!p)
 807                         return -ENOMEM;
 808
 809                 mutex_lock(&ca->set->bucket_lock);
 810                 for_each_bucket(b, ca) {
 811                         if (!GC_SECTORS_USED(b))
 812                                 unused++;
 813                         if (GC_MARK(b) == GC_MARK_RECLAIMABLE)
 814                                 available++;
 815                         if (GC_MARK(b) == GC_MARK_DIRTY)
 816                                 dirty++;
 817                         if (GC_MARK(b) == GC_MARK_METADATA)
 818                                 meta++;
 819                 }
 820
 821                 for (i = ca->sb.first_bucket; i < n; i++)
 822                         p[i] = ca->buckets[i].prio;
 823                 mutex_unlock(&ca->set->bucket_lock);
 824
 825                 sort(p, n, sizeof(uint16_t), __bch_cache_cmp, NULL);
 826
 827                 while (n &&
 828                        !cached[n - 1])
 829                         --n;
 830
 831                 unused = ca->sb.nbuckets - n;
 832
 833                 while (cached < p + n &&
 834                        *cached == BTREE_PRIO)
 835                         cached++, n--;
 836
 837                 for (i = 0; i < n; i++)
 838                         sum += INITIAL_PRIO - cached[i];
 839
 840                 if (n)
 841                         do_div(sum, n);
 842
 843                 for (i = 0; i < ARRAY_SIZE(q); i++)
 844                         q[i] = INITIAL_PRIO - cached[n * (i + 1) /
 845                                 (ARRAY_SIZE(q) + 1)];
 846
 847                 vfree(p);
 848
 849                 ret = scnprintf(buf, PAGE_SIZE,
 850                                 "Unused:                %zu%%\n"
 851                                 "Clean:         %zu%%\n"
 852                                 "Dirty:         %zu%%\n"
 853                                 "Metadata:      %zu%%\n"
 854                                 "Average:       %llu\n"
 855                                 "Sectors per Q: %zu\n"
 856                                 "Quantiles:     [",
 857                                 unused * 100 / (size_t) ca->sb.nbuckets,
 858                                 available * 100 / (size_t) ca->sb.nbuckets,
 859                                 dirty * 100 / (size_t) ca->sb.nbuckets,
 860                                 meta * 100 / (size_t) ca->sb.nbuckets, sum,
 861                                 n * ca->sb.bucket_size / (ARRAY_SIZE(q) + 1));
 862
 863                 for (i = 0; i < ARRAY_SIZE(q); i++)
 864                         ret += scnprintf(buf + ret, PAGE_SIZE - ret,
 865                                          "%u ", q[i]);
 866                 ret--;
 867
 868                 ret += scnprintf(buf + ret, PAGE_SIZE - ret, "]\n");
 869
 870                 return ret;
 871         }
 872
 873         return 0;
 874 }
 875 SHOW_LOCKED(bch_cache)
 876
 877 STORE(__bch_cache)
 878 {
 879         struct cache *ca = container_of(kobj, struct cache, kobj);
 880
 881         if (attr == &sysfs_discard) {
 882                 bool v = strtoul_or_return(buf);
 883
 884                 if (blk_queue_discard(bdev_get_queue(ca->bdev)))
 885                         ca->discard = v;
 886
 887                 if (v != CACHE_DISCARD(&ca->sb)) {
 888                         SET_CACHE_DISCARD(&ca->sb, v);
 889                         bcache_write_super(ca->set);
 890                 }
 891         }
 892
 893         if (attr == &sysfs_cache_replacement_policy) {
 894                 ssize_t v = bch_read_string_list(buf, cache_replacement_policies);
 895
 896                 if (v < 0)
 897                         return v;
 898
 899                 if ((unsigned) v != CACHE_REPLACEMENT(&ca->sb)) {
 900                         mutex_lock(&ca->set->bucket_lock);
 901                         SET_CACHE_REPLACEMENT(&ca->sb, v);
 902                         mutex_unlock(&ca->set->bucket_lock);
 903
 904                         bcache_write_super(ca->set);
 905                 }
 906         }
 907
 908         if (attr == &sysfs_clear_stats) {
 909                 atomic_long_set(&ca->sectors_written, 0);
 910                 atomic_long_set(&ca->btree_sectors_written, 0);
 911                 atomic_long_set(&ca->meta_sectors_written, 0);
 912                 atomic_set(&ca->io_count, 0);
 913                 atomic_set(&ca->io_errors, 0);
 914         }
 915
 916         return size;
 917 }
 918 STORE_LOCKED(bch_cache)
 919
 920 static struct attribute *bch_cache_files[] = {
 921         &sysfs_bucket_size,
 922         &sysfs_block_size,
 923         &sysfs_nbuckets,
 924         &sysfs_priority_stats,
 925         &sysfs_discard,
 926         &sysfs_written,
 927         &sysfs_btree_written,
 928         &sysfs_metadata_written,
 929         &sysfs_io_errors,
 930         &sysfs_clear_stats,
 931         &sysfs_cache_replacement_policy,
 932         NULL
 933 };
 934 KTYPE(bch_cache);