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 /* Default is -1; we skip past it for struct cached_dev's cache mode */
  20 static const char * const bch_cache_modes[] = {
  21         "writethrough",
  22         "writeback",
  23         "writearound",
  24         "none",
  25         NULL
  26 };
  27
  28 /* Default is -1; we skip past it for stop_when_cache_set_failed */
  29 static const char * const bch_stop_on_failure_modes[] = {
  30         "auto",
  31         "always",
  32         NULL
  33 };
  34
  35 static const char * const cache_replacement_policies[] = {
  36         "lru",
  37         "fifo",
  38         "random",
  39         NULL
  40 };
  41
  42 static const char * const error_actions[] = {
  43         "unregister",
  44         "panic",
  45         NULL
  46 };
  47
  48 write_attribute(attach);
  49 write_attribute(detach);
  50 write_attribute(unregister);
  51 write_attribute(stop);
  52 write_attribute(clear_stats);
  53 write_attribute(trigger_gc);
  54 write_attribute(prune_cache);
  55 write_attribute(flash_vol_create);
  56
  57 read_attribute(bucket_size);
  58 read_attribute(block_size);
  59 read_attribute(nbuckets);
  60 read_attribute(tree_depth);
  61 read_attribute(root_usage_percent);
  62 read_attribute(priority_stats);
  63 read_attribute(btree_cache_size);
  64 read_attribute(btree_cache_max_chain);
  65 read_attribute(cache_available_percent);
  66 read_attribute(written);
  67 read_attribute(btree_written);
  68 read_attribute(metadata_written);
  69 read_attribute(active_journal_entries);
  70
  71 sysfs_time_stats_attribute(btree_gc,    sec, ms);
  72 sysfs_time_stats_attribute(btree_split, sec, us);
  73 sysfs_time_stats_attribute(btree_sort,  ms,  us);
  74 sysfs_time_stats_attribute(btree_read,  ms,  us);
  75
  76 read_attribute(btree_nodes);
  77 read_attribute(btree_used_percent);
  78 read_attribute(average_key_size);
  79 read_attribute(dirty_data);
  80 read_attribute(bset_tree_stats);
  81
  82 read_attribute(state);
  83 read_attribute(cache_read_races);
  84 read_attribute(reclaim);
  85 read_attribute(flush_write);
  86 read_attribute(retry_flush_write);
  87 read_attribute(writeback_keys_done);
  88 read_attribute(writeback_keys_failed);
  89 read_attribute(io_errors);
  90 read_attribute(congested);
  91 rw_attribute(congested_read_threshold_us);
  92 rw_attribute(congested_write_threshold_us);
  93
  94 rw_attribute(sequential_cutoff);
  95 rw_attribute(data_csum);
  96 rw_attribute(cache_mode);
  97 rw_attribute(stop_when_cache_set_failed);
  98 rw_attribute(writeback_metadata);
  99 rw_attribute(writeback_running);
 100 rw_attribute(writeback_percent);
 101 rw_attribute(writeback_delay);
 102 rw_attribute(writeback_rate);
 103
 104 rw_attribute(writeback_rate_update_seconds);
 105 rw_attribute(writeback_rate_i_term_inverse);
 106 rw_attribute(writeback_rate_p_term_inverse);
 107 rw_attribute(writeback_rate_minimum);
 108 read_attribute(writeback_rate_debug);
 109
 110 read_attribute(stripe_size);
 111 read_attribute(partial_stripes_expensive);
 112
 113 rw_attribute(synchronous);
 114 rw_attribute(journal_delay_ms);
 115 rw_attribute(io_disable);
 116 rw_attribute(discard);
 117 rw_attribute(running);
 118 rw_attribute(label);
 119 rw_attribute(readahead);
 120 rw_attribute(errors);
 121 rw_attribute(io_error_limit);
 122 rw_attribute(io_error_halflife);
 123 rw_attribute(verify);
 124 rw_attribute(bypass_torture_test);
 125 rw_attribute(key_merging_disabled);
 126 rw_attribute(gc_always_rewrite);
 127 rw_attribute(expensive_debug_checks);
 128 rw_attribute(cache_replacement_policy);
 129 rw_attribute(btree_shrinker_disabled);
 130 rw_attribute(copy_gc_enabled);
 131 rw_attribute(size);
 132
 133 static ssize_t bch_snprint_string_list(char *buf, size_t size, const char * const list[],
 134                             size_t selected)
 135 {
 136         char *out = buf;
 137         size_t i;
 138
 139         for (i = 0; list[i]; i++)
 140                 out += snprintf(out, buf + size - out,
 141                                 i == selected ? "[%s] " : "%s ", list[i]);
 142
 143         out[-1] = '\n';
 144         return out - buf;
 145 }
 146
 147 SHOW(__bch_cached_dev)
 148 {
 149         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 150                                              disk.kobj);
 151         const char *states[] = { "no cache", "clean", "dirty", "inconsistent" };
 152         int wb = dc->writeback_running;
 153
 154 #define var(stat)               (dc->stat)
 155
 156         if (attr == &sysfs_cache_mode)
 157                 return bch_snprint_string_list(buf, PAGE_SIZE,
 158                                                bch_cache_modes,
 159                                                BDEV_CACHE_MODE(&dc->sb));
 160
 161         if (attr == &sysfs_stop_when_cache_set_failed)
 162                 return bch_snprint_string_list(buf, PAGE_SIZE,
 163                                                bch_stop_on_failure_modes,
 164                                                dc->stop_when_cache_set_failed);
 165
 166
 167         sysfs_printf(data_csum,         "%i", dc->disk.data_csum);
 168         var_printf(verify,              "%i");
 169         var_printf(bypass_torture_test, "%i");
 170         var_printf(writeback_metadata,  "%i");
 171         var_printf(writeback_running,   "%i");
 172         var_print(writeback_delay);
 173         var_print(writeback_percent);
 174         sysfs_hprint(writeback_rate,
 175                      wb ? atomic_long_read(&dc->writeback_rate.rate) << 9 : 0);
 176         sysfs_hprint(io_errors,         atomic_read(&dc->io_errors));
 177         sysfs_printf(io_error_limit,    "%i", dc->error_limit);
 178         sysfs_printf(io_disable,        "%i", dc->io_disable);
 179         var_print(writeback_rate_update_seconds);
 180         var_print(writeback_rate_i_term_inverse);
 181         var_print(writeback_rate_p_term_inverse);
 182         var_print(writeback_rate_minimum);
 183
 184         if (attr == &sysfs_writeback_rate_debug) {
 185                 char rate[20];
 186                 char dirty[20];
 187                 char target[20];
 188                 char proportional[20];
 189                 char integral[20];
 190                 char change[20];
 191                 s64 next_io;
 192
 193                 /*
 194                  * Except for dirty and target, other values should
 195                  * be 0 if writeback is not running.
 196                  */
 197                 bch_hprint(rate,
 198                            wb ? atomic_long_read(&dc->writeback_rate.rate) << 9
 199                               : 0);
 200                 bch_hprint(dirty, bcache_dev_sectors_dirty(&dc->disk) << 9);
 201                 bch_hprint(target, dc->writeback_rate_target << 9);
 202                 bch_hprint(proportional,
 203                            wb ? dc->writeback_rate_proportional << 9 : 0);
 204                 bch_hprint(integral,
 205                            wb ? dc->writeback_rate_integral_scaled << 9 : 0);
 206                 bch_hprint(change, wb ? dc->writeback_rate_change << 9 : 0);
 207                 next_io = wb ? div64_s64(dc->writeback_rate.next-local_clock(),
 208                                          NSEC_PER_MSEC) : 0;
 209
 210                 return sprintf(buf,
 211                                "rate:\t\t%s/sec\n"
 212                                "dirty:\t\t%s\n"
 213                                "target:\t\t%s\n"
 214                                "proportional:\t%s\n"
 215                                "integral:\t%s\n"
 216                                "change:\t\t%s/sec\n"
 217                                "next io:\t%llims\n",
 218                                rate, dirty, target, proportional,
 219                                integral, change, next_io);
 220         }
 221
 222         sysfs_hprint(dirty_data,
 223                      bcache_dev_sectors_dirty(&dc->disk) << 9);
 224
 225         sysfs_hprint(stripe_size,        ((uint64_t)dc->disk.stripe_size) << 9);
 226         var_printf(partial_stripes_expensive,   "%u");
 227
 228         var_hprint(sequential_cutoff);
 229         var_hprint(readahead);
 230
 231         sysfs_print(running,            atomic_read(&dc->running));
 232         sysfs_print(state,              states[BDEV_STATE(&dc->sb)]);
 233
 234         if (attr == &sysfs_label) {
 235                 memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
 236                 buf[SB_LABEL_SIZE + 1] = '\0';
 237                 strcat(buf, "\n");
 238                 return strlen(buf);
 239         }
 240
 241 #undef var
 242         return 0;
 243 }
 244 SHOW_LOCKED(bch_cached_dev)
 245
 246 STORE(__cached_dev)
 247 {
 248         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 249                                              disk.kobj);
 250         ssize_t v;
 251         struct cache_set *c;
 252         struct kobj_uevent_env *env;
 253
 254 #define d_strtoul(var)          sysfs_strtoul(var, dc->var)
 255 #define d_strtoul_nonzero(var)  sysfs_strtoul_clamp(var, dc->var, 1, INT_MAX)
 256 #define d_strtoi_h(var)         sysfs_hatoi(var, dc->var)
 257
 258         sysfs_strtoul(data_csum,        dc->disk.data_csum);
 259         d_strtoul(verify);
 260         d_strtoul(bypass_torture_test);
 261         d_strtoul(writeback_metadata);
 262         d_strtoul(writeback_running);
 263         d_strtoul(writeback_delay);
 264
 265         sysfs_strtoul_clamp(writeback_percent, dc->writeback_percent, 0, 40);
 266
 267         if (attr == &sysfs_writeback_rate) {
 268                 ssize_t ret;
 269                 long int v = atomic_long_read(&dc->writeback_rate.rate);
 270
 271                 ret = strtoul_safe_clamp(buf, v, 1, INT_MAX);
 272
 273                 if (!ret) {
 274                         atomic_long_set(&dc->writeback_rate.rate, v);
 275                         ret = size;
 276                 }
 277
 278                 return ret;
 279         }
 280
 281         sysfs_strtoul_clamp(writeback_rate_update_seconds,
 282                             dc->writeback_rate_update_seconds,
 283                             1, WRITEBACK_RATE_UPDATE_SECS_MAX);
 284         d_strtoul(writeback_rate_i_term_inverse);
 285         d_strtoul_nonzero(writeback_rate_p_term_inverse);
 286
 287         sysfs_strtoul_clamp(io_error_limit, dc->error_limit, 0, INT_MAX);
 288
 289         if (attr == &sysfs_io_disable) {
 290                 int v = strtoul_or_return(buf);
 291
 292                 dc->io_disable = v ? 1 : 0;
 293         }
 294
 295         d_strtoi_h(sequential_cutoff);
 296         d_strtoi_h(readahead);
 297
 298         if (attr == &sysfs_clear_stats)
 299                 bch_cache_accounting_clear(&dc->accounting);
 300
 301         if (attr == &sysfs_running &&
 302             strtoul_or_return(buf))
 303                 bch_cached_dev_run(dc);
 304
 305         if (attr == &sysfs_cache_mode) {
 306                 v = __sysfs_match_string(bch_cache_modes, -1, buf);
 307                 if (v < 0)
 308                         return v;
 309
 310                 if ((unsigned) v != BDEV_CACHE_MODE(&dc->sb)) {
 311                         SET_BDEV_CACHE_MODE(&dc->sb, v);
 312                         bch_write_bdev_super(dc, NULL);
 313                 }
 314         }
 315
 316         if (attr == &sysfs_stop_when_cache_set_failed) {
 317                 v = __sysfs_match_string(bch_stop_on_failure_modes, -1, buf);
 318                 if (v < 0)
 319                         return v;
 320
 321                 dc->stop_when_cache_set_failed = v;
 322         }
 323
 324         if (attr == &sysfs_label) {
 325                 if (size > SB_LABEL_SIZE)
 326                         return -EINVAL;
 327                 memcpy(dc->sb.label, buf, size);
 328                 if (size < SB_LABEL_SIZE)
 329                         dc->sb.label[size] = '\0';
 330                 if (size && dc->sb.label[size - 1] == '\n')
 331                         dc->sb.label[size - 1] = '\0';
 332                 bch_write_bdev_super(dc, NULL);
 333                 if (dc->disk.c) {
 334                         memcpy(dc->disk.c->uuids[dc->disk.id].label,
 335                                buf, SB_LABEL_SIZE);
 336                         bch_uuid_write(dc->disk.c);
 337                 }
 338                 env = kzalloc(sizeof(struct kobj_uevent_env), GFP_KERNEL);
 339                 if (!env)
 340                         return -ENOMEM;
 341                 add_uevent_var(env, "DRIVER=bcache");
 342                 add_uevent_var(env, "CACHED_UUID=%pU", dc->sb.uuid),
 343                 add_uevent_var(env, "CACHED_LABEL=%s", buf);
 344                 kobject_uevent_env(
 345                         &disk_to_dev(dc->disk.disk)->kobj, KOBJ_CHANGE, env->envp);
 346                 kfree(env);
 347         }
 348
 349         if (attr == &sysfs_attach) {
 350                 uint8_t         set_uuid[16];
 351
 352                 if (bch_parse_uuid(buf, set_uuid) < 16)
 353                         return -EINVAL;
 354
 355                 v = -ENOENT;
 356                 list_for_each_entry(c, &bch_cache_sets, list) {
 357                         v = bch_cached_dev_attach(dc, c, set_uuid);
 358                         if (!v)
 359                                 return size;
 360                 }
 361                 if (v == -ENOENT)
 362                         pr_err("Can't attach %s: cache set not found", buf);
 363                 return v;
 364         }
 365
 366         if (attr == &sysfs_detach && dc->disk.c)
 367                 bch_cached_dev_detach(dc);
 368
 369         if (attr == &sysfs_stop)
 370                 bcache_device_stop(&dc->disk);
 371
 372         return size;
 373 }
 374
 375 STORE(bch_cached_dev)
 376 {
 377         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 378                                              disk.kobj);
 379
 380         mutex_lock(&bch_register_lock);
 381         size = __cached_dev_store(kobj, attr, buf, size);
 382
 383         if (attr == &sysfs_writeback_running)
 384                 bch_writeback_queue(dc);
 385
 386         if (attr == &sysfs_writeback_percent)
 387                 if (!test_and_set_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags))
 388                         schedule_delayed_work(&dc->writeback_rate_update,
 389                                       dc->writeback_rate_update_seconds * HZ);
 390
 391         mutex_unlock(&bch_register_lock);
 392         return size;
 393 }
 394
 395 static struct attribute *bch_cached_dev_files[] = {
 396         &sysfs_attach,
 397         &sysfs_detach,
 398         &sysfs_stop,
 399 #if 0
 400         &sysfs_data_csum,
 401 #endif
 402         &sysfs_cache_mode,
 403         &sysfs_stop_when_cache_set_failed,
 404         &sysfs_writeback_metadata,
 405         &sysfs_writeback_running,
 406         &sysfs_writeback_delay,
 407         &sysfs_writeback_percent,
 408         &sysfs_writeback_rate,
 409         &sysfs_writeback_rate_update_seconds,
 410         &sysfs_writeback_rate_i_term_inverse,
 411         &sysfs_writeback_rate_p_term_inverse,
 412         &sysfs_writeback_rate_debug,
 413         &sysfs_errors,
 414         &sysfs_io_error_limit,
 415         &sysfs_io_disable,
 416         &sysfs_dirty_data,
 417         &sysfs_stripe_size,
 418         &sysfs_partial_stripes_expensive,
 419         &sysfs_sequential_cutoff,
 420         &sysfs_clear_stats,
 421         &sysfs_running,
 422         &sysfs_state,
 423         &sysfs_label,
 424         &sysfs_readahead,
 425 #ifdef CONFIG_BCACHE_DEBUG
 426         &sysfs_verify,
 427         &sysfs_bypass_torture_test,
 428 #endif
 429         NULL
 430 };
 431 KTYPE(bch_cached_dev);
 432
 433 SHOW(bch_flash_dev)
 434 {
 435         struct bcache_device *d = container_of(kobj, struct bcache_device,
 436                                                kobj);
 437         struct uuid_entry *u = &d->c->uuids[d->id];
 438
 439         sysfs_printf(data_csum, "%i", d->data_csum);
 440         sysfs_hprint(size,      u->sectors << 9);
 441
 442         if (attr == &sysfs_label) {
 443                 memcpy(buf, u->label, SB_LABEL_SIZE);
 444                 buf[SB_LABEL_SIZE + 1] = '\0';
 445                 strcat(buf, "\n");
 446                 return strlen(buf);
 447         }
 448
 449         return 0;
 450 }
 451
 452 STORE(__bch_flash_dev)
 453 {
 454         struct bcache_device *d = container_of(kobj, struct bcache_device,
 455                                                kobj);
 456         struct uuid_entry *u = &d->c->uuids[d->id];
 457
 458         sysfs_strtoul(data_csum,        d->data_csum);
 459
 460         if (attr == &sysfs_size) {
 461                 uint64_t v;
 462                 strtoi_h_or_return(buf, v);
 463
 464                 u->sectors = v >> 9;
 465                 bch_uuid_write(d->c);
 466                 set_capacity(d->disk, u->sectors);
 467         }
 468
 469         if (attr == &sysfs_label) {
 470                 memcpy(u->label, buf, SB_LABEL_SIZE);
 471                 bch_uuid_write(d->c);
 472         }
 473
 474         if (attr == &sysfs_unregister) {
 475                 set_bit(BCACHE_DEV_DETACHING, &d->flags);
 476                 bcache_device_stop(d);
 477         }
 478
 479         return size;
 480 }
 481 STORE_LOCKED(bch_flash_dev)
 482
 483 static struct attribute *bch_flash_dev_files[] = {
 484         &sysfs_unregister,
 485 #if 0
 486         &sysfs_data_csum,
 487 #endif
 488         &sysfs_label,
 489         &sysfs_size,
 490         NULL
 491 };
 492 KTYPE(bch_flash_dev);
 493
 494 struct bset_stats_op {
 495         struct btree_op op;
 496         size_t nodes;
 497         struct bset_stats stats;
 498 };
 499
 500 static int bch_btree_bset_stats(struct btree_op *b_op, struct btree *b)
 501 {
 502         struct bset_stats_op *op = container_of(b_op, struct bset_stats_op, op);
 503
 504         op->nodes++;
 505         bch_btree_keys_stats(&b->keys, &op->stats);
 506
 507         return MAP_CONTINUE;
 508 }
 509
 510 static int bch_bset_print_stats(struct cache_set *c, char *buf)
 511 {
 512         struct bset_stats_op op;
 513         int ret;
 514
 515         memset(&op, 0, sizeof(op));
 516         bch_btree_op_init(&op.op, -1);
 517
 518         ret = bch_btree_map_nodes(&op.op, c, &ZERO_KEY, bch_btree_bset_stats);
 519         if (ret < 0)
 520                 return ret;
 521
 522         return snprintf(buf, PAGE_SIZE,
 523                         "btree nodes:           %zu\n"
 524                         "written sets:          %zu\n"
 525                         "unwritten sets:                %zu\n"
 526                         "written key bytes:     %zu\n"
 527                         "unwritten key bytes:   %zu\n"
 528                         "floats:                        %zu\n"
 529                         "failed:                        %zu\n",
 530                         op.nodes,
 531                         op.stats.sets_written, op.stats.sets_unwritten,
 532                         op.stats.bytes_written, op.stats.bytes_unwritten,
 533                         op.stats.floats, op.stats.failed);
 534 }
 535
 536 static unsigned bch_root_usage(struct cache_set *c)
 537 {
 538         unsigned bytes = 0;
 539         struct bkey *k;
 540         struct btree *b;
 541         struct btree_iter iter;
 542
 543         goto lock_root;
 544
 545         do {
 546                 rw_unlock(false, b);
 547 lock_root:
 548                 b = c->root;
 549                 rw_lock(false, b, b->level);
 550         } while (b != c->root);
 551
 552         for_each_key_filter(&b->keys, k, &iter, bch_ptr_bad)
 553                 bytes += bkey_bytes(k);
 554
 555         rw_unlock(false, b);
 556
 557         return (bytes * 100) / btree_bytes(c);
 558 }
 559
 560 static size_t bch_cache_size(struct cache_set *c)
 561 {
 562         size_t ret = 0;
 563         struct btree *b;
 564
 565         mutex_lock(&c->bucket_lock);
 566         list_for_each_entry(b, &c->btree_cache, list)
 567                 ret += 1 << (b->keys.page_order + PAGE_SHIFT);
 568
 569         mutex_unlock(&c->bucket_lock);
 570         return ret;
 571 }
 572
 573 static unsigned bch_cache_max_chain(struct cache_set *c)
 574 {
 575         unsigned ret = 0;
 576         struct hlist_head *h;
 577
 578         mutex_lock(&c->bucket_lock);
 579
 580         for (h = c->bucket_hash;
 581              h < c->bucket_hash + (1 << BUCKET_HASH_BITS);
 582              h++) {
 583                 unsigned i = 0;
 584                 struct hlist_node *p;
 585
 586                 hlist_for_each(p, h)
 587                         i++;
 588
 589                 ret = max(ret, i);
 590         }
 591
 592         mutex_unlock(&c->bucket_lock);
 593         return ret;
 594 }
 595
 596 static unsigned bch_btree_used(struct cache_set *c)
 597 {
 598         return div64_u64(c->gc_stats.key_bytes * 100,
 599                          (c->gc_stats.nodes ?: 1) * btree_bytes(c));
 600 }
 601
 602 static unsigned bch_average_key_size(struct cache_set *c)
 603 {
 604         return c->gc_stats.nkeys
 605                 ? div64_u64(c->gc_stats.data, c->gc_stats.nkeys)
 606                 : 0;
 607 }
 608
 609 SHOW(__bch_cache_set)
 610 {
 611         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 612
 613         sysfs_print(synchronous,                CACHE_SYNC(&c->sb));
 614         sysfs_print(journal_delay_ms,           c->journal_delay_ms);
 615         sysfs_hprint(bucket_size,               bucket_bytes(c));
 616         sysfs_hprint(block_size,                block_bytes(c));
 617         sysfs_print(tree_depth,                 c->root->level);
 618         sysfs_print(root_usage_percent,         bch_root_usage(c));
 619
 620         sysfs_hprint(btree_cache_size,          bch_cache_size(c));
 621         sysfs_print(btree_cache_max_chain,      bch_cache_max_chain(c));
 622         sysfs_print(cache_available_percent,    100 - c->gc_stats.in_use);
 623
 624         sysfs_print_time_stats(&c->btree_gc_time,       btree_gc, sec, ms);
 625         sysfs_print_time_stats(&c->btree_split_time,    btree_split, sec, us);
 626         sysfs_print_time_stats(&c->sort.time,           btree_sort, ms, us);
 627         sysfs_print_time_stats(&c->btree_read_time,     btree_read, ms, us);
 628
 629         sysfs_print(btree_used_percent, bch_btree_used(c));
 630         sysfs_print(btree_nodes,        c->gc_stats.nodes);
 631         sysfs_hprint(average_key_size,  bch_average_key_size(c));
 632
 633         sysfs_print(cache_read_races,
 634                     atomic_long_read(&c->cache_read_races));
 635
 636         sysfs_print(reclaim,
 637                     atomic_long_read(&c->reclaim));
 638
 639         sysfs_print(flush_write,
 640                     atomic_long_read(&c->flush_write));
 641
 642         sysfs_print(retry_flush_write,
 643                     atomic_long_read(&c->retry_flush_write));
 644
 645         sysfs_print(writeback_keys_done,
 646                     atomic_long_read(&c->writeback_keys_done));
 647         sysfs_print(writeback_keys_failed,
 648                     atomic_long_read(&c->writeback_keys_failed));
 649
 650         if (attr == &sysfs_errors)
 651                 return bch_snprint_string_list(buf, PAGE_SIZE, error_actions,
 652                                                c->on_error);
 653
 654         /* See count_io_errors for why 88 */
 655         sysfs_print(io_error_halflife,  c->error_decay * 88);
 656         sysfs_print(io_error_limit,     c->error_limit);
 657
 658         sysfs_hprint(congested,
 659                      ((uint64_t) bch_get_congested(c)) << 9);
 660         sysfs_print(congested_read_threshold_us,
 661                     c->congested_read_threshold_us);
 662         sysfs_print(congested_write_threshold_us,
 663                     c->congested_write_threshold_us);
 664
 665         sysfs_print(active_journal_entries,     fifo_used(&c->journal.pin));
 666         sysfs_printf(verify,                    "%i", c->verify);
 667         sysfs_printf(key_merging_disabled,      "%i", c->key_merging_disabled);
 668         sysfs_printf(expensive_debug_checks,
 669                      "%i", c->expensive_debug_checks);
 670         sysfs_printf(gc_always_rewrite,         "%i", c->gc_always_rewrite);
 671         sysfs_printf(btree_shrinker_disabled,   "%i", c->shrinker_disabled);
 672         sysfs_printf(copy_gc_enabled,           "%i", c->copy_gc_enabled);
 673         sysfs_printf(io_disable,                "%i",
 674                      test_bit(CACHE_SET_IO_DISABLE, &c->flags));
 675
 676         if (attr == &sysfs_bset_tree_stats)
 677                 return bch_bset_print_stats(c, buf);
 678
 679         return 0;
 680 }
 681 SHOW_LOCKED(bch_cache_set)
 682
 683 STORE(__bch_cache_set)
 684 {
 685         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 686         ssize_t v;
 687
 688         if (attr == &sysfs_unregister)
 689                 bch_cache_set_unregister(c);
 690
 691         if (attr == &sysfs_stop)
 692                 bch_cache_set_stop(c);
 693
 694         if (attr == &sysfs_synchronous) {
 695                 bool sync = strtoul_or_return(buf);
 696
 697                 if (sync != CACHE_SYNC(&c->sb)) {
 698                         SET_CACHE_SYNC(&c->sb, sync);
 699                         bcache_write_super(c);
 700                 }
 701         }
 702
 703         if (attr == &sysfs_flash_vol_create) {
 704                 int r;
 705                 uint64_t v;
 706                 strtoi_h_or_return(buf, v);
 707
 708                 r = bch_flash_dev_create(c, v);
 709                 if (r)
 710                         return r;
 711         }
 712
 713         if (attr == &sysfs_clear_stats) {
 714                 atomic_long_set(&c->writeback_keys_done,        0);
 715                 atomic_long_set(&c->writeback_keys_failed,      0);
 716
 717                 memset(&c->gc_stats, 0, sizeof(struct gc_stat));
 718                 bch_cache_accounting_clear(&c->accounting);
 719         }
 720
 721         if (attr == &sysfs_trigger_gc) {
 722                 /*
 723                  * Garbage collection thread only works when sectors_to_gc < 0,
 724                  * when users write to sysfs entry trigger_gc, most of time
 725                  * they want to forcibly triger gargage collection. Here -1 is
 726                  * set to c->sectors_to_gc, to make gc_should_run() give a
 727                  * chance to permit gc thread to run. "give a chance" means
 728                  * before going into gc_should_run(), there is still chance
 729                  * that c->sectors_to_gc being set to other positive value. So
 730                  * writing sysfs entry trigger_gc won't always make sure gc
 731                  * thread takes effect.
 732                  */
 733                 atomic_set(&c->sectors_to_gc, -1);
 734                 wake_up_gc(c);
 735         }
 736
 737         if (attr == &sysfs_prune_cache) {
 738                 struct shrink_control sc;
 739                 sc.gfp_mask = GFP_KERNEL;
 740                 sc.nr_to_scan = strtoul_or_return(buf);
 741                 c->shrink.scan_objects(&c->shrink, &sc);
 742         }
 743
 744         sysfs_strtoul(congested_read_threshold_us,
 745                       c->congested_read_threshold_us);
 746         sysfs_strtoul(congested_write_threshold_us,
 747                       c->congested_write_threshold_us);
 748
 749         if (attr == &sysfs_errors) {
 750                 v = __sysfs_match_string(error_actions, -1, buf);
 751                 if (v < 0)
 752                         return v;
 753
 754                 c->on_error = v;
 755         }
 756
 757         if (attr == &sysfs_io_error_limit)
 758                 c->error_limit = strtoul_or_return(buf);
 759
 760         /* See count_io_errors() for why 88 */
 761         if (attr == &sysfs_io_error_halflife)
 762                 c->error_decay = strtoul_or_return(buf) / 88;
 763
 764         if (attr == &sysfs_io_disable) {
 765                 v = strtoul_or_return(buf);
 766                 if (v) {
 767                         if (test_and_set_bit(CACHE_SET_IO_DISABLE,
 768                                              &c->flags))
 769                                 pr_warn("CACHE_SET_IO_DISABLE already set");
 770                 } else {
 771                         if (!test_and_clear_bit(CACHE_SET_IO_DISABLE,
 772                                                 &c->flags))
 773                                 pr_warn("CACHE_SET_IO_DISABLE already cleared");
 774                 }
 775         }
 776
 777         sysfs_strtoul(journal_delay_ms,         c->journal_delay_ms);
 778         sysfs_strtoul(verify,                   c->verify);
 779         sysfs_strtoul(key_merging_disabled,     c->key_merging_disabled);
 780         sysfs_strtoul(expensive_debug_checks,   c->expensive_debug_checks);
 781         sysfs_strtoul(gc_always_rewrite,        c->gc_always_rewrite);
 782         sysfs_strtoul(btree_shrinker_disabled,  c->shrinker_disabled);
 783         sysfs_strtoul(copy_gc_enabled,          c->copy_gc_enabled);
 784
 785         return size;
 786 }
 787 STORE_LOCKED(bch_cache_set)
 788
 789 SHOW(bch_cache_set_internal)
 790 {
 791         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 792         return bch_cache_set_show(&c->kobj, attr, buf);
 793 }
 794
 795 STORE(bch_cache_set_internal)
 796 {
 797         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 798         return bch_cache_set_store(&c->kobj, attr, buf, size);
 799 }
 800
 801 static void bch_cache_set_internal_release(struct kobject *k)
 802 {
 803 }
 804
 805 static struct attribute *bch_cache_set_files[] = {
 806         &sysfs_unregister,
 807         &sysfs_stop,
 808         &sysfs_synchronous,
 809         &sysfs_journal_delay_ms,
 810         &sysfs_flash_vol_create,
 811
 812         &sysfs_bucket_size,
 813         &sysfs_block_size,
 814         &sysfs_tree_depth,
 815         &sysfs_root_usage_percent,
 816         &sysfs_btree_cache_size,
 817         &sysfs_cache_available_percent,
 818
 819         &sysfs_average_key_size,
 820
 821         &sysfs_errors,
 822         &sysfs_io_error_limit,
 823         &sysfs_io_error_halflife,
 824         &sysfs_congested,
 825         &sysfs_congested_read_threshold_us,
 826         &sysfs_congested_write_threshold_us,
 827         &sysfs_clear_stats,
 828         NULL
 829 };
 830 KTYPE(bch_cache_set);
 831
 832 static struct attribute *bch_cache_set_internal_files[] = {
 833         &sysfs_active_journal_entries,
 834
 835         sysfs_time_stats_attribute_list(btree_gc, sec, ms)
 836         sysfs_time_stats_attribute_list(btree_split, sec, us)
 837         sysfs_time_stats_attribute_list(btree_sort, ms, us)
 838         sysfs_time_stats_attribute_list(btree_read, ms, us)
 839
 840         &sysfs_btree_nodes,
 841         &sysfs_btree_used_percent,
 842         &sysfs_btree_cache_max_chain,
 843
 844         &sysfs_bset_tree_stats,
 845         &sysfs_cache_read_races,
 846         &sysfs_reclaim,
 847         &sysfs_flush_write,
 848         &sysfs_retry_flush_write,
 849         &sysfs_writeback_keys_done,
 850         &sysfs_writeback_keys_failed,
 851
 852         &sysfs_trigger_gc,
 853         &sysfs_prune_cache,
 854 #ifdef CONFIG_BCACHE_DEBUG
 855         &sysfs_verify,
 856         &sysfs_key_merging_disabled,
 857         &sysfs_expensive_debug_checks,
 858 #endif
 859         &sysfs_gc_always_rewrite,
 860         &sysfs_btree_shrinker_disabled,
 861         &sysfs_copy_gc_enabled,
 862         &sysfs_io_disable,
 863         NULL
 864 };
 865 KTYPE(bch_cache_set_internal);
 866
 867 static int __bch_cache_cmp(const void *l, const void *r)
 868 {
 869         return *((uint16_t *)r) - *((uint16_t *)l);
 870 }
 871
 872 SHOW(__bch_cache)
 873 {
 874         struct cache *ca = container_of(kobj, struct cache, kobj);
 875
 876         sysfs_hprint(bucket_size,       bucket_bytes(ca));
 877         sysfs_hprint(block_size,        block_bytes(ca));
 878         sysfs_print(nbuckets,           ca->sb.nbuckets);
 879         sysfs_print(discard,            ca->discard);
 880         sysfs_hprint(written, atomic_long_read(&ca->sectors_written) << 9);
 881         sysfs_hprint(btree_written,
 882                      atomic_long_read(&ca->btree_sectors_written) << 9);
 883         sysfs_hprint(metadata_written,
 884                      (atomic_long_read(&ca->meta_sectors_written) +
 885                       atomic_long_read(&ca->btree_sectors_written)) << 9);
 886
 887         sysfs_print(io_errors,
 888                     atomic_read(&ca->io_errors) >> IO_ERROR_SHIFT);
 889
 890         if (attr == &sysfs_cache_replacement_policy)
 891                 return bch_snprint_string_list(buf, PAGE_SIZE,
 892                                                cache_replacement_policies,
 893                                                CACHE_REPLACEMENT(&ca->sb));
 894
 895         if (attr == &sysfs_priority_stats) {
 896                 struct bucket *b;
 897                 size_t n = ca->sb.nbuckets, i;
 898                 size_t unused = 0, available = 0, dirty = 0, meta = 0;
 899                 uint64_t sum = 0;
 900                 /* Compute 31 quantiles */
 901                 uint16_t q[31], *p, *cached;
 902                 ssize_t ret;
 903
 904                 cached = p = vmalloc(array_size(sizeof(uint16_t),
 905                                                 ca->sb.nbuckets));
 906                 if (!p)
 907                         return -ENOMEM;
 908
 909                 mutex_lock(&ca->set->bucket_lock);
 910                 for_each_bucket(b, ca) {
 911                         if (!GC_SECTORS_USED(b))
 912                                 unused++;
 913                         if (GC_MARK(b) == GC_MARK_RECLAIMABLE)
 914                                 available++;
 915                         if (GC_MARK(b) == GC_MARK_DIRTY)
 916                                 dirty++;
 917                         if (GC_MARK(b) == GC_MARK_METADATA)
 918                                 meta++;
 919                 }
 920
 921                 for (i = ca->sb.first_bucket; i < n; i++)
 922                         p[i] = ca->buckets[i].prio;
 923                 mutex_unlock(&ca->set->bucket_lock);
 924
 925                 sort(p, n, sizeof(uint16_t), __bch_cache_cmp, NULL);
 926
 927                 while (n &&
 928                        !cached[n - 1])
 929                         --n;
 930
 931                 unused = ca->sb.nbuckets - n;
 932
 933                 while (cached < p + n &&
 934                        *cached == BTREE_PRIO)
 935                         cached++, n--;
 936
 937                 for (i = 0; i < n; i++)
 938                         sum += INITIAL_PRIO - cached[i];
 939
 940                 if (n)
 941                         do_div(sum, n);
 942
 943                 for (i = 0; i < ARRAY_SIZE(q); i++)
 944                         q[i] = INITIAL_PRIO - cached[n * (i + 1) /
 945                                 (ARRAY_SIZE(q) + 1)];
 946
 947                 vfree(p);
 948
 949                 ret = scnprintf(buf, PAGE_SIZE,
 950                                 "Unused:                %zu%%\n"
 951                                 "Clean:         %zu%%\n"
 952                                 "Dirty:         %zu%%\n"
 953                                 "Metadata:      %zu%%\n"
 954                                 "Average:       %llu\n"
 955                                 "Sectors per Q: %zu\n"
 956                                 "Quantiles:     [",
 957                                 unused * 100 / (size_t) ca->sb.nbuckets,
 958                                 available * 100 / (size_t) ca->sb.nbuckets,
 959                                 dirty * 100 / (size_t) ca->sb.nbuckets,
 960                                 meta * 100 / (size_t) ca->sb.nbuckets, sum,
 961                                 n * ca->sb.bucket_size / (ARRAY_SIZE(q) + 1));
 962
 963                 for (i = 0; i < ARRAY_SIZE(q); i++)
 964                         ret += scnprintf(buf + ret, PAGE_SIZE - ret,
 965                                          "%u ", q[i]);
 966                 ret--;
 967
 968                 ret += scnprintf(buf + ret, PAGE_SIZE - ret, "]\n");
 969
 970                 return ret;
 971         }
 972
 973         return 0;
 974 }
 975 SHOW_LOCKED(bch_cache)
 976
 977 STORE(__bch_cache)
 978 {
 979         struct cache *ca = container_of(kobj, struct cache, kobj);
 980         ssize_t v;
 981
 982         if (attr == &sysfs_discard) {
 983                 bool v = strtoul_or_return(buf);
 984
 985                 if (blk_queue_discard(bdev_get_queue(ca->bdev)))
 986                         ca->discard = v;
 987
 988                 if (v != CACHE_DISCARD(&ca->sb)) {
 989                         SET_CACHE_DISCARD(&ca->sb, v);
 990                         bcache_write_super(ca->set);
 991                 }
 992         }
 993
 994         if (attr == &sysfs_cache_replacement_policy) {
 995                 v = __sysfs_match_string(cache_replacement_policies, -1, buf);
 996                 if (v < 0)
 997                         return v;
 998
 999                 if ((unsigned) v != CACHE_REPLACEMENT(&ca->sb)) {
1000                         mutex_lock(&ca->set->bucket_lock);
1001                         SET_CACHE_REPLACEMENT(&ca->sb, v);
1002                         mutex_unlock(&ca->set->bucket_lock);
1003
1004                         bcache_write_super(ca->set);
1005                 }
1006         }
1007
1008         if (attr == &sysfs_clear_stats) {
1009                 atomic_long_set(&ca->sectors_written, 0);
1010                 atomic_long_set(&ca->btree_sectors_written, 0);
1011                 atomic_long_set(&ca->meta_sectors_written, 0);
1012                 atomic_set(&ca->io_count, 0);
1013                 atomic_set(&ca->io_errors, 0);
1014         }
1015
1016         return size;
1017 }
1018 STORE_LOCKED(bch_cache)
1019
1020 static struct attribute *bch_cache_files[] = {
1021         &sysfs_bucket_size,
1022         &sysfs_block_size,
1023         &sysfs_nbuckets,
1024         &sysfs_priority_stats,
1025         &sysfs_discard,
1026         &sysfs_written,
1027         &sysfs_btree_written,
1028         &sysfs_metadata_written,
1029         &sysfs_io_errors,
1030         &sysfs_clear_stats,
1031         &sysfs_cache_replacement_policy,
1032         NULL
1033 };
1034 KTYPE(bch_cache);