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;
 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                 uint8_t         set_uuid[16];
 276
 277                 if (bch_parse_uuid(buf, set_uuid) < 16)
 278                         return -EINVAL;
 279
 280                 v = -ENOENT;
 281                 list_for_each_entry(c, &bch_cache_sets, list) {
 282                         v = bch_cached_dev_attach(dc, c, set_uuid);
 283                         if (!v)
 284                                 return size;
 285                 }
 286
 287                 pr_err("Can't attach %s: cache set not found", buf);
 288                 return v;
 289         }
 290
 291         if (attr == &sysfs_detach && dc->disk.c)
 292                 bch_cached_dev_detach(dc);
 293
 294         if (attr == &sysfs_stop)
 295                 bcache_device_stop(&dc->disk);
 296
 297         return size;
 298 }
 299
 300 STORE(bch_cached_dev)
 301 {
 302         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 303                                              disk.kobj);
 304
 305         mutex_lock(&bch_register_lock);
 306         size = __cached_dev_store(kobj, attr, buf, size);
 307
 308         if (attr == &sysfs_writeback_running)
 309                 bch_writeback_queue(dc);
 310
 311         if (attr == &sysfs_writeback_percent)
 312                 schedule_delayed_work(&dc->writeback_rate_update,
 313                                       dc->writeback_rate_update_seconds * HZ);
 314
 315         mutex_unlock(&bch_register_lock);
 316         return size;
 317 }
 318
 319 static struct attribute *bch_cached_dev_files[] = {
 320         &sysfs_attach,
 321         &sysfs_detach,
 322         &sysfs_stop,
 323 #if 0
 324         &sysfs_data_csum,
 325 #endif
 326         &sysfs_cache_mode,
 327         &sysfs_writeback_metadata,
 328         &sysfs_writeback_running,
 329         &sysfs_writeback_delay,
 330         &sysfs_writeback_percent,
 331         &sysfs_writeback_rate,
 332         &sysfs_writeback_rate_update_seconds,
 333         &sysfs_writeback_rate_i_term_inverse,
 334         &sysfs_writeback_rate_p_term_inverse,
 335         &sysfs_writeback_rate_debug,
 336         &sysfs_dirty_data,
 337         &sysfs_stripe_size,
 338         &sysfs_partial_stripes_expensive,
 339         &sysfs_sequential_cutoff,
 340         &sysfs_clear_stats,
 341         &sysfs_running,
 342         &sysfs_state,
 343         &sysfs_label,
 344         &sysfs_readahead,
 345 #ifdef CONFIG_BCACHE_DEBUG
 346         &sysfs_verify,
 347         &sysfs_bypass_torture_test,
 348 #endif
 349         NULL
 350 };
 351 KTYPE(bch_cached_dev);
 352
 353 SHOW(bch_flash_dev)
 354 {
 355         struct bcache_device *d = container_of(kobj, struct bcache_device,
 356                                                kobj);
 357         struct uuid_entry *u = &d->c->uuids[d->id];
 358
 359         sysfs_printf(data_csum, "%i", d->data_csum);
 360         sysfs_hprint(size,      u->sectors << 9);
 361
 362         if (attr == &sysfs_label) {
 363                 memcpy(buf, u->label, SB_LABEL_SIZE);
 364                 buf[SB_LABEL_SIZE + 1] = '\0';
 365                 strcat(buf, "\n");
 366                 return strlen(buf);
 367         }
 368
 369         return 0;
 370 }
 371
 372 STORE(__bch_flash_dev)
 373 {
 374         struct bcache_device *d = container_of(kobj, struct bcache_device,
 375                                                kobj);
 376         struct uuid_entry *u = &d->c->uuids[d->id];
 377
 378         sysfs_strtoul(data_csum,        d->data_csum);
 379
 380         if (attr == &sysfs_size) {
 381                 uint64_t v;
 382                 strtoi_h_or_return(buf, v);
 383
 384                 u->sectors = v >> 9;
 385                 bch_uuid_write(d->c);
 386                 set_capacity(d->disk, u->sectors);
 387         }
 388
 389         if (attr == &sysfs_label) {
 390                 memcpy(u->label, buf, SB_LABEL_SIZE);
 391                 bch_uuid_write(d->c);
 392         }
 393
 394         if (attr == &sysfs_unregister) {
 395                 set_bit(BCACHE_DEV_DETACHING, &d->flags);
 396                 bcache_device_stop(d);
 397         }
 398
 399         return size;
 400 }
 401 STORE_LOCKED(bch_flash_dev)
 402
 403 static struct attribute *bch_flash_dev_files[] = {
 404         &sysfs_unregister,
 405 #if 0
 406         &sysfs_data_csum,
 407 #endif
 408         &sysfs_label,
 409         &sysfs_size,
 410         NULL
 411 };
 412 KTYPE(bch_flash_dev);
 413
 414 struct bset_stats_op {
 415         struct btree_op op;
 416         size_t nodes;
 417         struct bset_stats stats;
 418 };
 419
 420 static int bch_btree_bset_stats(struct btree_op *b_op, struct btree *b)
 421 {
 422         struct bset_stats_op *op = container_of(b_op, struct bset_stats_op, op);
 423
 424         op->nodes++;
 425         bch_btree_keys_stats(&b->keys, &op->stats);
 426
 427         return MAP_CONTINUE;
 428 }
 429
 430 static int bch_bset_print_stats(struct cache_set *c, char *buf)
 431 {
 432         struct bset_stats_op op;
 433         int ret;
 434
 435         memset(&op, 0, sizeof(op));
 436         bch_btree_op_init(&op.op, -1);
 437
 438         ret = bch_btree_map_nodes(&op.op, c, &ZERO_KEY, bch_btree_bset_stats);
 439         if (ret < 0)
 440                 return ret;
 441
 442         return snprintf(buf, PAGE_SIZE,
 443                         "btree nodes:           %zu\n"
 444                         "written sets:          %zu\n"
 445                         "unwritten sets:                %zu\n"
 446                         "written key bytes:     %zu\n"
 447                         "unwritten key bytes:   %zu\n"
 448                         "floats:                        %zu\n"
 449                         "failed:                        %zu\n",
 450                         op.nodes,
 451                         op.stats.sets_written, op.stats.sets_unwritten,
 452                         op.stats.bytes_written, op.stats.bytes_unwritten,
 453                         op.stats.floats, op.stats.failed);
 454 }
 455
 456 static unsigned bch_root_usage(struct cache_set *c)
 457 {
 458         unsigned bytes = 0;
 459         struct bkey *k;
 460         struct btree *b;
 461         struct btree_iter iter;
 462
 463         goto lock_root;
 464
 465         do {
 466                 rw_unlock(false, b);
 467 lock_root:
 468                 b = c->root;
 469                 rw_lock(false, b, b->level);
 470         } while (b != c->root);
 471
 472         for_each_key_filter(&b->keys, k, &iter, bch_ptr_bad)
 473                 bytes += bkey_bytes(k);
 474
 475         rw_unlock(false, b);
 476
 477         return (bytes * 100) / btree_bytes(c);
 478 }
 479
 480 static size_t bch_cache_size(struct cache_set *c)
 481 {
 482         size_t ret = 0;
 483         struct btree *b;
 484
 485         mutex_lock(&c->bucket_lock);
 486         list_for_each_entry(b, &c->btree_cache, list)
 487                 ret += 1 << (b->keys.page_order + PAGE_SHIFT);
 488
 489         mutex_unlock(&c->bucket_lock);
 490         return ret;
 491 }
 492
 493 static unsigned bch_cache_max_chain(struct cache_set *c)
 494 {
 495         unsigned ret = 0;
 496         struct hlist_head *h;
 497
 498         mutex_lock(&c->bucket_lock);
 499
 500         for (h = c->bucket_hash;
 501              h < c->bucket_hash + (1 << BUCKET_HASH_BITS);
 502              h++) {
 503                 unsigned i = 0;
 504                 struct hlist_node *p;
 505
 506                 hlist_for_each(p, h)
 507                         i++;
 508
 509                 ret = max(ret, i);
 510         }
 511
 512         mutex_unlock(&c->bucket_lock);
 513         return ret;
 514 }
 515
 516 static unsigned bch_btree_used(struct cache_set *c)
 517 {
 518         return div64_u64(c->gc_stats.key_bytes * 100,
 519                          (c->gc_stats.nodes ?: 1) * btree_bytes(c));
 520 }
 521
 522 static unsigned bch_average_key_size(struct cache_set *c)
 523 {
 524         return c->gc_stats.nkeys
 525                 ? div64_u64(c->gc_stats.data, c->gc_stats.nkeys)
 526                 : 0;
 527 }
 528
 529 SHOW(__bch_cache_set)
 530 {
 531         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 532
 533         sysfs_print(synchronous,                CACHE_SYNC(&c->sb));
 534         sysfs_print(journal_delay_ms,           c->journal_delay_ms);
 535         sysfs_hprint(bucket_size,               bucket_bytes(c));
 536         sysfs_hprint(block_size,                block_bytes(c));
 537         sysfs_print(tree_depth,                 c->root->level);
 538         sysfs_print(root_usage_percent,         bch_root_usage(c));
 539
 540         sysfs_hprint(btree_cache_size,          bch_cache_size(c));
 541         sysfs_print(btree_cache_max_chain,      bch_cache_max_chain(c));
 542         sysfs_print(cache_available_percent,    100 - c->gc_stats.in_use);
 543
 544         sysfs_print_time_stats(&c->btree_gc_time,       btree_gc, sec, ms);
 545         sysfs_print_time_stats(&c->btree_split_time,    btree_split, sec, us);
 546         sysfs_print_time_stats(&c->sort.time,           btree_sort, ms, us);
 547         sysfs_print_time_stats(&c->btree_read_time,     btree_read, ms, us);
 548
 549         sysfs_print(btree_used_percent, bch_btree_used(c));
 550         sysfs_print(btree_nodes,        c->gc_stats.nodes);
 551         sysfs_hprint(average_key_size,  bch_average_key_size(c));
 552
 553         sysfs_print(cache_read_races,
 554                     atomic_long_read(&c->cache_read_races));
 555
 556         sysfs_print(reclaim,
 557                     atomic_long_read(&c->reclaim));
 558
 559         sysfs_print(flush_write,
 560                     atomic_long_read(&c->flush_write));
 561
 562         sysfs_print(retry_flush_write,
 563                     atomic_long_read(&c->retry_flush_write));
 564
 565         sysfs_print(writeback_keys_done,
 566                     atomic_long_read(&c->writeback_keys_done));
 567         sysfs_print(writeback_keys_failed,
 568                     atomic_long_read(&c->writeback_keys_failed));
 569
 570         if (attr == &sysfs_errors)
 571                 return bch_snprint_string_list(buf, PAGE_SIZE, error_actions,
 572                                                c->on_error);
 573
 574         /* See count_io_errors for why 88 */
 575         sysfs_print(io_error_halflife,  c->error_decay * 88);
 576         sysfs_print(io_error_limit,     c->error_limit);
 577
 578         sysfs_hprint(congested,
 579                      ((uint64_t) bch_get_congested(c)) << 9);
 580         sysfs_print(congested_read_threshold_us,
 581                     c->congested_read_threshold_us);
 582         sysfs_print(congested_write_threshold_us,
 583                     c->congested_write_threshold_us);
 584
 585         sysfs_print(active_journal_entries,     fifo_used(&c->journal.pin));
 586         sysfs_printf(verify,                    "%i", c->verify);
 587         sysfs_printf(key_merging_disabled,      "%i", c->key_merging_disabled);
 588         sysfs_printf(expensive_debug_checks,
 589                      "%i", c->expensive_debug_checks);
 590         sysfs_printf(gc_always_rewrite,         "%i", c->gc_always_rewrite);
 591         sysfs_printf(btree_shrinker_disabled,   "%i", c->shrinker_disabled);
 592         sysfs_printf(copy_gc_enabled,           "%i", c->copy_gc_enabled);
 593
 594         if (attr == &sysfs_bset_tree_stats)
 595                 return bch_bset_print_stats(c, buf);
 596
 597         return 0;
 598 }
 599 SHOW_LOCKED(bch_cache_set)
 600
 601 STORE(__bch_cache_set)
 602 {
 603         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 604
 605         if (attr == &sysfs_unregister)
 606                 bch_cache_set_unregister(c);
 607
 608         if (attr == &sysfs_stop)
 609                 bch_cache_set_stop(c);
 610
 611         if (attr == &sysfs_synchronous) {
 612                 bool sync = strtoul_or_return(buf);
 613
 614                 if (sync != CACHE_SYNC(&c->sb)) {
 615                         SET_CACHE_SYNC(&c->sb, sync);
 616                         bcache_write_super(c);
 617                 }
 618         }
 619
 620         if (attr == &sysfs_flash_vol_create) {
 621                 int r;
 622                 uint64_t v;
 623                 strtoi_h_or_return(buf, v);
 624
 625                 r = bch_flash_dev_create(c, v);
 626                 if (r)
 627                         return r;
 628         }
 629
 630         if (attr == &sysfs_clear_stats) {
 631                 atomic_long_set(&c->writeback_keys_done,        0);
 632                 atomic_long_set(&c->writeback_keys_failed,      0);
 633
 634                 memset(&c->gc_stats, 0, sizeof(struct gc_stat));
 635                 bch_cache_accounting_clear(&c->accounting);
 636         }
 637
 638         if (attr == &sysfs_trigger_gc) {
 639                 /*
 640                  * Garbage collection thread only works when sectors_to_gc < 0,
 641                  * when users write to sysfs entry trigger_gc, most of time
 642                  * they want to forcibly triger gargage collection. Here -1 is
 643                  * set to c->sectors_to_gc, to make gc_should_run() give a
 644                  * chance to permit gc thread to run. "give a chance" means
 645                  * before going into gc_should_run(), there is still chance
 646                  * that c->sectors_to_gc being set to other positive value. So
 647                  * writing sysfs entry trigger_gc won't always make sure gc
 648                  * thread takes effect.
 649                  */
 650                 atomic_set(&c->sectors_to_gc, -1);
 651                 wake_up_gc(c);
 652         }
 653
 654         if (attr == &sysfs_prune_cache) {
 655                 struct shrink_control sc;
 656                 sc.gfp_mask = GFP_KERNEL;
 657                 sc.nr_to_scan = strtoul_or_return(buf);
 658                 c->shrink.scan_objects(&c->shrink, &sc);
 659         }
 660
 661         sysfs_strtoul(congested_read_threshold_us,
 662                       c->congested_read_threshold_us);
 663         sysfs_strtoul(congested_write_threshold_us,
 664                       c->congested_write_threshold_us);
 665
 666         if (attr == &sysfs_errors) {
 667                 ssize_t v = bch_read_string_list(buf, error_actions);
 668
 669                 if (v < 0)
 670                         return v;
 671
 672                 c->on_error = v;
 673         }
 674
 675         if (attr == &sysfs_io_error_limit)
 676                 c->error_limit = strtoul_or_return(buf);
 677
 678         /* See count_io_errors() for why 88 */
 679         if (attr == &sysfs_io_error_halflife)
 680                 c->error_decay = strtoul_or_return(buf) / 88;
 681
 682         sysfs_strtoul(journal_delay_ms,         c->journal_delay_ms);
 683         sysfs_strtoul(verify,                   c->verify);
 684         sysfs_strtoul(key_merging_disabled,     c->key_merging_disabled);
 685         sysfs_strtoul(expensive_debug_checks,   c->expensive_debug_checks);
 686         sysfs_strtoul(gc_always_rewrite,        c->gc_always_rewrite);
 687         sysfs_strtoul(btree_shrinker_disabled,  c->shrinker_disabled);
 688         sysfs_strtoul(copy_gc_enabled,          c->copy_gc_enabled);
 689
 690         return size;
 691 }
 692 STORE_LOCKED(bch_cache_set)
 693
 694 SHOW(bch_cache_set_internal)
 695 {
 696         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 697         return bch_cache_set_show(&c->kobj, attr, buf);
 698 }
 699
 700 STORE(bch_cache_set_internal)
 701 {
 702         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 703         return bch_cache_set_store(&c->kobj, attr, buf, size);
 704 }
 705
 706 static void bch_cache_set_internal_release(struct kobject *k)
 707 {
 708 }
 709
 710 static struct attribute *bch_cache_set_files[] = {
 711         &sysfs_unregister,
 712         &sysfs_stop,
 713         &sysfs_synchronous,
 714         &sysfs_journal_delay_ms,
 715         &sysfs_flash_vol_create,
 716
 717         &sysfs_bucket_size,
 718         &sysfs_block_size,
 719         &sysfs_tree_depth,
 720         &sysfs_root_usage_percent,
 721         &sysfs_btree_cache_size,
 722         &sysfs_cache_available_percent,
 723
 724         &sysfs_average_key_size,
 725
 726         &sysfs_errors,
 727         &sysfs_io_error_limit,
 728         &sysfs_io_error_halflife,
 729         &sysfs_congested,
 730         &sysfs_congested_read_threshold_us,
 731         &sysfs_congested_write_threshold_us,
 732         &sysfs_clear_stats,
 733         NULL
 734 };
 735 KTYPE(bch_cache_set);
 736
 737 static struct attribute *bch_cache_set_internal_files[] = {
 738         &sysfs_active_journal_entries,
 739
 740         sysfs_time_stats_attribute_list(btree_gc, sec, ms)
 741         sysfs_time_stats_attribute_list(btree_split, sec, us)
 742         sysfs_time_stats_attribute_list(btree_sort, ms, us)
 743         sysfs_time_stats_attribute_list(btree_read, ms, us)
 744
 745         &sysfs_btree_nodes,
 746         &sysfs_btree_used_percent,
 747         &sysfs_btree_cache_max_chain,
 748
 749         &sysfs_bset_tree_stats,
 750         &sysfs_cache_read_races,
 751         &sysfs_reclaim,
 752         &sysfs_flush_write,
 753         &sysfs_retry_flush_write,
 754         &sysfs_writeback_keys_done,
 755         &sysfs_writeback_keys_failed,
 756
 757         &sysfs_trigger_gc,
 758         &sysfs_prune_cache,
 759 #ifdef CONFIG_BCACHE_DEBUG
 760         &sysfs_verify,
 761         &sysfs_key_merging_disabled,
 762         &sysfs_expensive_debug_checks,
 763 #endif
 764         &sysfs_gc_always_rewrite,
 765         &sysfs_btree_shrinker_disabled,
 766         &sysfs_copy_gc_enabled,
 767         NULL
 768 };
 769 KTYPE(bch_cache_set_internal);
 770
 771 static int __bch_cache_cmp(const void *l, const void *r)
 772 {
 773         return *((uint16_t *)r) - *((uint16_t *)l);
 774 }
 775
 776 SHOW(__bch_cache)
 777 {
 778         struct cache *ca = container_of(kobj, struct cache, kobj);
 779
 780         sysfs_hprint(bucket_size,       bucket_bytes(ca));
 781         sysfs_hprint(block_size,        block_bytes(ca));
 782         sysfs_print(nbuckets,           ca->sb.nbuckets);
 783         sysfs_print(discard,            ca->discard);
 784         sysfs_hprint(written, atomic_long_read(&ca->sectors_written) << 9);
 785         sysfs_hprint(btree_written,
 786                      atomic_long_read(&ca->btree_sectors_written) << 9);
 787         sysfs_hprint(metadata_written,
 788                      (atomic_long_read(&ca->meta_sectors_written) +
 789                       atomic_long_read(&ca->btree_sectors_written)) << 9);
 790
 791         sysfs_print(io_errors,
 792                     atomic_read(&ca->io_errors) >> IO_ERROR_SHIFT);
 793
 794         if (attr == &sysfs_cache_replacement_policy)
 795                 return bch_snprint_string_list(buf, PAGE_SIZE,
 796                                                cache_replacement_policies,
 797                                                CACHE_REPLACEMENT(&ca->sb));
 798
 799         if (attr == &sysfs_priority_stats) {
 800                 struct bucket *b;
 801                 size_t n = ca->sb.nbuckets, i;
 802                 size_t unused = 0, available = 0, dirty = 0, meta = 0;
 803                 uint64_t sum = 0;
 804                 /* Compute 31 quantiles */
 805                 uint16_t q[31], *p, *cached;
 806                 ssize_t ret;
 807
 808                 cached = p = vmalloc(ca->sb.nbuckets * sizeof(uint16_t));
 809                 if (!p)
 810                         return -ENOMEM;
 811
 812                 mutex_lock(&ca->set->bucket_lock);
 813                 for_each_bucket(b, ca) {
 814                         if (!GC_SECTORS_USED(b))
 815                                 unused++;
 816                         if (GC_MARK(b) == GC_MARK_RECLAIMABLE)
 817                                 available++;
 818                         if (GC_MARK(b) == GC_MARK_DIRTY)
 819                                 dirty++;
 820                         if (GC_MARK(b) == GC_MARK_METADATA)
 821                                 meta++;
 822                 }
 823
 824                 for (i = ca->sb.first_bucket; i < n; i++)
 825                         p[i] = ca->buckets[i].prio;
 826                 mutex_unlock(&ca->set->bucket_lock);
 827
 828                 sort(p, n, sizeof(uint16_t), __bch_cache_cmp, NULL);
 829
 830                 while (n &&
 831                        !cached[n - 1])
 832                         --n;
 833
 834                 unused = ca->sb.nbuckets - n;
 835
 836                 while (cached < p + n &&
 837                        *cached == BTREE_PRIO)
 838                         cached++, n--;
 839
 840                 for (i = 0; i < n; i++)
 841                         sum += INITIAL_PRIO - cached[i];
 842
 843                 if (n)
 844                         do_div(sum, n);
 845
 846                 for (i = 0; i < ARRAY_SIZE(q); i++)
 847                         q[i] = INITIAL_PRIO - cached[n * (i + 1) /
 848                                 (ARRAY_SIZE(q) + 1)];
 849
 850                 vfree(p);
 851
 852                 ret = scnprintf(buf, PAGE_SIZE,
 853                                 "Unused:                %zu%%\n"
 854                                 "Clean:         %zu%%\n"
 855                                 "Dirty:         %zu%%\n"
 856                                 "Metadata:      %zu%%\n"
 857                                 "Average:       %llu\n"
 858                                 "Sectors per Q: %zu\n"
 859                                 "Quantiles:     [",
 860                                 unused * 100 / (size_t) ca->sb.nbuckets,
 861                                 available * 100 / (size_t) ca->sb.nbuckets,
 862                                 dirty * 100 / (size_t) ca->sb.nbuckets,
 863                                 meta * 100 / (size_t) ca->sb.nbuckets, sum,
 864                                 n * ca->sb.bucket_size / (ARRAY_SIZE(q) + 1));
 865
 866                 for (i = 0; i < ARRAY_SIZE(q); i++)
 867                         ret += scnprintf(buf + ret, PAGE_SIZE - ret,
 868                                          "%u ", q[i]);
 869                 ret--;
 870
 871                 ret += scnprintf(buf + ret, PAGE_SIZE - ret, "]\n");
 872
 873                 return ret;
 874         }
 875
 876         return 0;
 877 }
 878 SHOW_LOCKED(bch_cache)
 879
 880 STORE(__bch_cache)
 881 {
 882         struct cache *ca = container_of(kobj, struct cache, kobj);
 883
 884         if (attr == &sysfs_discard) {
 885                 bool v = strtoul_or_return(buf);
 886
 887                 if (blk_queue_discard(bdev_get_queue(ca->bdev)))
 888                         ca->discard = v;
 889
 890                 if (v != CACHE_DISCARD(&ca->sb)) {
 891                         SET_CACHE_DISCARD(&ca->sb, v);
 892                         bcache_write_super(ca->set);
 893                 }
 894         }
 895
 896         if (attr == &sysfs_cache_replacement_policy) {
 897                 ssize_t v = bch_read_string_list(buf, cache_replacement_policies);
 898
 899                 if (v < 0)
 900                         return v;
 901
 902                 if ((unsigned) v != CACHE_REPLACEMENT(&ca->sb)) {
 903                         mutex_lock(&ca->set->bucket_lock);
 904                         SET_CACHE_REPLACEMENT(&ca->sb, v);
 905                         mutex_unlock(&ca->set->bucket_lock);
 906
 907                         bcache_write_super(ca->set);
 908                 }
 909         }
 910
 911         if (attr == &sysfs_clear_stats) {
 912                 atomic_long_set(&ca->sectors_written, 0);
 913                 atomic_long_set(&ca->btree_sectors_written, 0);
 914                 atomic_long_set(&ca->meta_sectors_written, 0);
 915                 atomic_set(&ca->io_count, 0);
 916                 atomic_set(&ca->io_errors, 0);
 917         }
 918
 919         return size;
 920 }
 921 STORE_LOCKED(bch_cache)
 922
 923 static struct attribute *bch_cache_files[] = {
 924         &sysfs_bucket_size,
 925         &sysfs_block_size,
 926         &sysfs_nbuckets,
 927         &sysfs_priority_stats,
 928         &sysfs_discard,
 929         &sysfs_written,
 930         &sysfs_btree_written,
 931         &sysfs_metadata_written,
 932         &sysfs_io_errors,
 933         &sysfs_clear_stats,
 934         &sysfs_cache_replacement_policy,
 935         NULL
 936 };
 937 KTYPE(bch_cache);