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 static const char * const bch_reada_cache_policies[] = {
  29         "all",
  30         "meta-only",
  31         NULL
  32 };
  33
  34 /* Default is -1; we skip past it for stop_when_cache_set_failed */
  35 static const char * const bch_stop_on_failure_modes[] = {
  36         "auto",
  37         "always",
  38         NULL
  39 };
  40
  41 static const char * const cache_replacement_policies[] = {
  42         "lru",
  43         "fifo",
  44         "random",
  45         NULL
  46 };
  47
  48 static const char * const error_actions[] = {
  49         "unregister",
  50         "panic",
  51         NULL
  52 };
  53
  54 write_attribute(attach);
  55 write_attribute(detach);
  56 write_attribute(unregister);
  57 write_attribute(stop);
  58 write_attribute(clear_stats);
  59 write_attribute(trigger_gc);
  60 write_attribute(prune_cache);
  61 write_attribute(flash_vol_create);
  62
  63 read_attribute(bucket_size);
  64 read_attribute(block_size);
  65 read_attribute(nbuckets);
  66 read_attribute(tree_depth);
  67 read_attribute(root_usage_percent);
  68 read_attribute(priority_stats);
  69 read_attribute(btree_cache_size);
  70 read_attribute(btree_cache_max_chain);
  71 read_attribute(cache_available_percent);
  72 read_attribute(written);
  73 read_attribute(btree_written);
  74 read_attribute(metadata_written);
  75 read_attribute(active_journal_entries);
  76
  77 sysfs_time_stats_attribute(btree_gc,    sec, ms);
  78 sysfs_time_stats_attribute(btree_split, sec, us);
  79 sysfs_time_stats_attribute(btree_sort,  ms,  us);
  80 sysfs_time_stats_attribute(btree_read,  ms,  us);
  81
  82 read_attribute(btree_nodes);
  83 read_attribute(btree_used_percent);
  84 read_attribute(average_key_size);
  85 read_attribute(dirty_data);
  86 read_attribute(bset_tree_stats);
  87
  88 read_attribute(state);
  89 read_attribute(cache_read_races);
  90 read_attribute(reclaim);
  91 read_attribute(flush_write);
  92 read_attribute(retry_flush_write);
  93 read_attribute(writeback_keys_done);
  94 read_attribute(writeback_keys_failed);
  95 read_attribute(io_errors);
  96 read_attribute(congested);
  97 rw_attribute(congested_read_threshold_us);
  98 rw_attribute(congested_write_threshold_us);
  99
 100 rw_attribute(sequential_cutoff);
 101 rw_attribute(data_csum);
 102 rw_attribute(cache_mode);
 103 rw_attribute(readahead_cache_policy);
 104 rw_attribute(stop_when_cache_set_failed);
 105 rw_attribute(writeback_metadata);
 106 rw_attribute(writeback_running);
 107 rw_attribute(writeback_percent);
 108 rw_attribute(writeback_delay);
 109 rw_attribute(writeback_rate);
 110
 111 rw_attribute(writeback_rate_update_seconds);
 112 rw_attribute(writeback_rate_i_term_inverse);
 113 rw_attribute(writeback_rate_p_term_inverse);
 114 rw_attribute(writeback_rate_minimum);
 115 read_attribute(writeback_rate_debug);
 116
 117 read_attribute(stripe_size);
 118 read_attribute(partial_stripes_expensive);
 119
 120 rw_attribute(synchronous);
 121 rw_attribute(journal_delay_ms);
 122 rw_attribute(io_disable);
 123 rw_attribute(discard);
 124 rw_attribute(running);
 125 rw_attribute(label);
 126 rw_attribute(readahead);
 127 rw_attribute(errors);
 128 rw_attribute(io_error_limit);
 129 rw_attribute(io_error_halflife);
 130 rw_attribute(verify);
 131 rw_attribute(bypass_torture_test);
 132 rw_attribute(key_merging_disabled);
 133 rw_attribute(gc_always_rewrite);
 134 rw_attribute(expensive_debug_checks);
 135 rw_attribute(cache_replacement_policy);
 136 rw_attribute(btree_shrinker_disabled);
 137 rw_attribute(copy_gc_enabled);
 138 rw_attribute(size);
 139
 140 static ssize_t bch_snprint_string_list(char *buf,
 141                                        size_t size,
 142                                        const char * const list[],
 143                                        size_t selected)
 144 {
 145         char *out = buf;
 146         size_t i;
 147
 148         for (i = 0; list[i]; i++)
 149                 out += snprintf(out, buf + size - out,
 150                                 i == selected ? "[%s] " : "%s ", list[i]);
 151
 152         out[-1] = '\n';
 153         return out - buf;
 154 }
 155
 156 SHOW(__bch_cached_dev)
 157 {
 158         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 159                                              disk.kobj);
 160         char const *states[] = { "no cache", "clean", "dirty", "inconsistent" };
 161         int wb = dc->writeback_running;
 162
 163 #define var(stat)               (dc->stat)
 164
 165         if (attr == &sysfs_cache_mode)
 166                 return bch_snprint_string_list(buf, PAGE_SIZE,
 167                                                bch_cache_modes,
 168                                                BDEV_CACHE_MODE(&dc->sb));
 169
 170         if (attr == &sysfs_readahead_cache_policy)
 171                 return bch_snprint_string_list(buf, PAGE_SIZE,
 172                                               bch_reada_cache_policies,
 173                                               dc->cache_readahead_policy);
 174
 175         if (attr == &sysfs_stop_when_cache_set_failed)
 176                 return bch_snprint_string_list(buf, PAGE_SIZE,
 177                                                bch_stop_on_failure_modes,
 178                                                dc->stop_when_cache_set_failed);
 179
 180
 181         sysfs_printf(data_csum,         "%i", dc->disk.data_csum);
 182         var_printf(verify,              "%i");
 183         var_printf(bypass_torture_test, "%i");
 184         var_printf(writeback_metadata,  "%i");
 185         var_printf(writeback_running,   "%i");
 186         var_print(writeback_delay);
 187         var_print(writeback_percent);
 188         sysfs_hprint(writeback_rate,
 189                      wb ? atomic_long_read(&dc->writeback_rate.rate) << 9 : 0);
 190         sysfs_printf(io_errors,         "%i", atomic_read(&dc->io_errors));
 191         sysfs_printf(io_error_limit,    "%i", dc->error_limit);
 192         sysfs_printf(io_disable,        "%i", dc->io_disable);
 193         var_print(writeback_rate_update_seconds);
 194         var_print(writeback_rate_i_term_inverse);
 195         var_print(writeback_rate_p_term_inverse);
 196         var_print(writeback_rate_minimum);
 197
 198         if (attr == &sysfs_writeback_rate_debug) {
 199                 char rate[20];
 200                 char dirty[20];
 201                 char target[20];
 202                 char proportional[20];
 203                 char integral[20];
 204                 char change[20];
 205                 s64 next_io;
 206
 207                 /*
 208                  * Except for dirty and target, other values should
 209                  * be 0 if writeback is not running.
 210                  */
 211                 bch_hprint(rate,
 212                            wb ? atomic_long_read(&dc->writeback_rate.rate) << 9
 213                               : 0);
 214                 bch_hprint(dirty, bcache_dev_sectors_dirty(&dc->disk) << 9);
 215                 bch_hprint(target, dc->writeback_rate_target << 9);
 216                 bch_hprint(proportional,
 217                            wb ? dc->writeback_rate_proportional << 9 : 0);
 218                 bch_hprint(integral,
 219                            wb ? dc->writeback_rate_integral_scaled << 9 : 0);
 220                 bch_hprint(change, wb ? dc->writeback_rate_change << 9 : 0);
 221                 next_io = wb ? div64_s64(dc->writeback_rate.next-local_clock(),
 222                                          NSEC_PER_MSEC) : 0;
 223
 224                 return sprintf(buf,
 225                                "rate:\t\t%s/sec\n"
 226                                "dirty:\t\t%s\n"
 227                                "target:\t\t%s\n"
 228                                "proportional:\t%s\n"
 229                                "integral:\t%s\n"
 230                                "change:\t\t%s/sec\n"
 231                                "next io:\t%llims\n",
 232                                rate, dirty, target, proportional,
 233                                integral, change, next_io);
 234         }
 235
 236         sysfs_hprint(dirty_data,
 237                      bcache_dev_sectors_dirty(&dc->disk) << 9);
 238
 239         sysfs_hprint(stripe_size,        ((uint64_t)dc->disk.stripe_size) << 9);
 240         var_printf(partial_stripes_expensive,   "%u");
 241
 242         var_hprint(sequential_cutoff);
 243         var_hprint(readahead);
 244
 245         sysfs_print(running,            atomic_read(&dc->running));
 246         sysfs_print(state,              states[BDEV_STATE(&dc->sb)]);
 247
 248         if (attr == &sysfs_label) {
 249                 memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
 250                 buf[SB_LABEL_SIZE + 1] = '\0';
 251                 strcat(buf, "\n");
 252                 return strlen(buf);
 253         }
 254
 255 #undef var
 256         return 0;
 257 }
 258 SHOW_LOCKED(bch_cached_dev)
 259
 260 STORE(__cached_dev)
 261 {
 262         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 263                                              disk.kobj);
 264         ssize_t v;
 265         struct cache_set *c;
 266         struct kobj_uevent_env *env;
 267
 268 #define d_strtoul(var)          sysfs_strtoul(var, dc->var)
 269 #define d_strtoul_nonzero(var)  sysfs_strtoul_clamp(var, dc->var, 1, INT_MAX)
 270 #define d_strtoi_h(var)         sysfs_hatoi(var, dc->var)
 271
 272         sysfs_strtoul(data_csum,        dc->disk.data_csum);
 273         d_strtoul(verify);
 274         d_strtoul(bypass_torture_test);
 275         d_strtoul(writeback_metadata);
 276         d_strtoul(writeback_running);
 277         d_strtoul(writeback_delay);
 278
 279         sysfs_strtoul_clamp(writeback_percent, dc->writeback_percent, 0, 40);
 280
 281         if (attr == &sysfs_writeback_rate) {
 282                 ssize_t ret;
 283                 long int v = atomic_long_read(&dc->writeback_rate.rate);
 284
 285                 ret = strtoul_safe_clamp(buf, v, 1, INT_MAX);
 286
 287                 if (!ret) {
 288                         atomic_long_set(&dc->writeback_rate.rate, v);
 289                         ret = size;
 290                 }
 291
 292                 return ret;
 293         }
 294
 295         sysfs_strtoul_clamp(writeback_rate_update_seconds,
 296                             dc->writeback_rate_update_seconds,
 297                             1, WRITEBACK_RATE_UPDATE_SECS_MAX);
 298         sysfs_strtoul_clamp(writeback_rate_i_term_inverse,
 299                             dc->writeback_rate_i_term_inverse,
 300                             1, UINT_MAX);
 301         sysfs_strtoul_clamp(writeback_rate_p_term_inverse,
 302                             dc->writeback_rate_p_term_inverse,
 303                             1, UINT_MAX);
 304         sysfs_strtoul_clamp(writeback_rate_minimum,
 305                             dc->writeback_rate_minimum,
 306                             1, UINT_MAX);
 307
 308         sysfs_strtoul_clamp(io_error_limit, dc->error_limit, 0, INT_MAX);
 309
 310         if (attr == &sysfs_io_disable) {
 311                 int v = strtoul_or_return(buf);
 312
 313                 dc->io_disable = v ? 1 : 0;
 314         }
 315
 316         sysfs_strtoul_clamp(sequential_cutoff,
 317                             dc->sequential_cutoff,
 318                             0, UINT_MAX);
 319         d_strtoi_h(readahead);
 320
 321         if (attr == &sysfs_clear_stats)
 322                 bch_cache_accounting_clear(&dc->accounting);
 323
 324         if (attr == &sysfs_running &&
 325             strtoul_or_return(buf))
 326                 bch_cached_dev_run(dc);
 327
 328         if (attr == &sysfs_cache_mode) {
 329                 v = __sysfs_match_string(bch_cache_modes, -1, buf);
 330                 if (v < 0)
 331                         return v;
 332
 333                 if ((unsigned int) v != BDEV_CACHE_MODE(&dc->sb)) {
 334                         SET_BDEV_CACHE_MODE(&dc->sb, v);
 335                         bch_write_bdev_super(dc, NULL);
 336                 }
 337         }
 338
 339         if (attr == &sysfs_readahead_cache_policy) {
 340                 v = __sysfs_match_string(bch_reada_cache_policies, -1, buf);
 341                 if (v < 0)
 342                         return v;
 343
 344                 if ((unsigned int) v != dc->cache_readahead_policy)
 345                         dc->cache_readahead_policy = v;
 346         }
 347
 348         if (attr == &sysfs_stop_when_cache_set_failed) {
 349                 v = __sysfs_match_string(bch_stop_on_failure_modes, -1, buf);
 350                 if (v < 0)
 351                         return v;
 352
 353                 dc->stop_when_cache_set_failed = v;
 354         }
 355
 356         if (attr == &sysfs_label) {
 357                 if (size > SB_LABEL_SIZE)
 358                         return -EINVAL;
 359                 memcpy(dc->sb.label, buf, size);
 360                 if (size < SB_LABEL_SIZE)
 361                         dc->sb.label[size] = '\0';
 362                 if (size && dc->sb.label[size - 1] == '\n')
 363                         dc->sb.label[size - 1] = '\0';
 364                 bch_write_bdev_super(dc, NULL);
 365                 if (dc->disk.c) {
 366                         memcpy(dc->disk.c->uuids[dc->disk.id].label,
 367                                buf, SB_LABEL_SIZE);
 368                         bch_uuid_write(dc->disk.c);
 369                 }
 370                 env = kzalloc(sizeof(struct kobj_uevent_env), GFP_KERNEL);
 371                 if (!env)
 372                         return -ENOMEM;
 373                 add_uevent_var(env, "DRIVER=bcache");
 374                 add_uevent_var(env, "CACHED_UUID=%pU", dc->sb.uuid),
 375                 add_uevent_var(env, "CACHED_LABEL=%s", buf);
 376                 kobject_uevent_env(&disk_to_dev(dc->disk.disk)->kobj,
 377                                    KOBJ_CHANGE,
 378                                    env->envp);
 379                 kfree(env);
 380         }
 381
 382         if (attr == &sysfs_attach) {
 383                 uint8_t         set_uuid[16];
 384
 385                 if (bch_parse_uuid(buf, set_uuid) < 16)
 386                         return -EINVAL;
 387
 388                 v = -ENOENT;
 389                 list_for_each_entry(c, &bch_cache_sets, list) {
 390                         v = bch_cached_dev_attach(dc, c, set_uuid);
 391                         if (!v)
 392                                 return size;
 393                 }
 394                 if (v == -ENOENT)
 395                         pr_err("Can't attach %s: cache set not found", buf);
 396                 return v;
 397         }
 398
 399         if (attr == &sysfs_detach && dc->disk.c)
 400                 bch_cached_dev_detach(dc);
 401
 402         if (attr == &sysfs_stop)
 403                 bcache_device_stop(&dc->disk);
 404
 405         return size;
 406 }
 407
 408 STORE(bch_cached_dev)
 409 {
 410         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 411                                              disk.kobj);
 412
 413         mutex_lock(&bch_register_lock);
 414         size = __cached_dev_store(kobj, attr, buf, size);
 415
 416         if (attr == &sysfs_writeback_running)
 417                 bch_writeback_queue(dc);
 418
 419         /*
 420          * Only set BCACHE_DEV_WB_RUNNING when cached device attached to
 421          * a cache set, otherwise it doesn't make sense.
 422          */
 423         if (attr == &sysfs_writeback_percent)
 424                 if ((dc->disk.c != NULL) &&
 425                     (!test_and_set_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags)))
 426                         schedule_delayed_work(&dc->writeback_rate_update,
 427                                       dc->writeback_rate_update_seconds * HZ);
 428
 429         mutex_unlock(&bch_register_lock);
 430         return size;
 431 }
 432
 433 static struct attribute *bch_cached_dev_files[] = {
 434         &sysfs_attach,
 435         &sysfs_detach,
 436         &sysfs_stop,
 437 #if 0
 438         &sysfs_data_csum,
 439 #endif
 440         &sysfs_cache_mode,
 441         &sysfs_readahead_cache_policy,
 442         &sysfs_stop_when_cache_set_failed,
 443         &sysfs_writeback_metadata,
 444         &sysfs_writeback_running,
 445         &sysfs_writeback_delay,
 446         &sysfs_writeback_percent,
 447         &sysfs_writeback_rate,
 448         &sysfs_writeback_rate_update_seconds,
 449         &sysfs_writeback_rate_i_term_inverse,
 450         &sysfs_writeback_rate_p_term_inverse,
 451         &sysfs_writeback_rate_minimum,
 452         &sysfs_writeback_rate_debug,
 453         &sysfs_io_errors,
 454         &sysfs_io_error_limit,
 455         &sysfs_io_disable,
 456         &sysfs_dirty_data,
 457         &sysfs_stripe_size,
 458         &sysfs_partial_stripes_expensive,
 459         &sysfs_sequential_cutoff,
 460         &sysfs_clear_stats,
 461         &sysfs_running,
 462         &sysfs_state,
 463         &sysfs_label,
 464         &sysfs_readahead,
 465 #ifdef CONFIG_BCACHE_DEBUG
 466         &sysfs_verify,
 467         &sysfs_bypass_torture_test,
 468 #endif
 469         NULL
 470 };
 471 KTYPE(bch_cached_dev);
 472
 473 SHOW(bch_flash_dev)
 474 {
 475         struct bcache_device *d = container_of(kobj, struct bcache_device,
 476                                                kobj);
 477         struct uuid_entry *u = &d->c->uuids[d->id];
 478
 479         sysfs_printf(data_csum, "%i", d->data_csum);
 480         sysfs_hprint(size,      u->sectors << 9);
 481
 482         if (attr == &sysfs_label) {
 483                 memcpy(buf, u->label, SB_LABEL_SIZE);
 484                 buf[SB_LABEL_SIZE + 1] = '\0';
 485                 strcat(buf, "\n");
 486                 return strlen(buf);
 487         }
 488
 489         return 0;
 490 }
 491
 492 STORE(__bch_flash_dev)
 493 {
 494         struct bcache_device *d = container_of(kobj, struct bcache_device,
 495                                                kobj);
 496         struct uuid_entry *u = &d->c->uuids[d->id];
 497
 498         sysfs_strtoul(data_csum,        d->data_csum);
 499
 500         if (attr == &sysfs_size) {
 501                 uint64_t v;
 502
 503                 strtoi_h_or_return(buf, v);
 504
 505                 u->sectors = v >> 9;
 506                 bch_uuid_write(d->c);
 507                 set_capacity(d->disk, u->sectors);
 508         }
 509
 510         if (attr == &sysfs_label) {
 511                 memcpy(u->label, buf, SB_LABEL_SIZE);
 512                 bch_uuid_write(d->c);
 513         }
 514
 515         if (attr == &sysfs_unregister) {
 516                 set_bit(BCACHE_DEV_DETACHING, &d->flags);
 517                 bcache_device_stop(d);
 518         }
 519
 520         return size;
 521 }
 522 STORE_LOCKED(bch_flash_dev)
 523
 524 static struct attribute *bch_flash_dev_files[] = {
 525         &sysfs_unregister,
 526 #if 0
 527         &sysfs_data_csum,
 528 #endif
 529         &sysfs_label,
 530         &sysfs_size,
 531         NULL
 532 };
 533 KTYPE(bch_flash_dev);
 534
 535 struct bset_stats_op {
 536         struct btree_op op;
 537         size_t nodes;
 538         struct bset_stats stats;
 539 };
 540
 541 static int bch_btree_bset_stats(struct btree_op *b_op, struct btree *b)
 542 {
 543         struct bset_stats_op *op = container_of(b_op, struct bset_stats_op, op);
 544
 545         op->nodes++;
 546         bch_btree_keys_stats(&b->keys, &op->stats);
 547
 548         return MAP_CONTINUE;
 549 }
 550
 551 static int bch_bset_print_stats(struct cache_set *c, char *buf)
 552 {
 553         struct bset_stats_op op;
 554         int ret;
 555
 556         memset(&op, 0, sizeof(op));
 557         bch_btree_op_init(&op.op, -1);
 558
 559         ret = bch_btree_map_nodes(&op.op, c, &ZERO_KEY, bch_btree_bset_stats);
 560         if (ret < 0)
 561                 return ret;
 562
 563         return snprintf(buf, PAGE_SIZE,
 564                         "btree nodes:           %zu\n"
 565                         "written sets:          %zu\n"
 566                         "unwritten sets:                %zu\n"
 567                         "written key bytes:     %zu\n"
 568                         "unwritten key bytes:   %zu\n"
 569                         "floats:                        %zu\n"
 570                         "failed:                        %zu\n",
 571                         op.nodes,
 572                         op.stats.sets_written, op.stats.sets_unwritten,
 573                         op.stats.bytes_written, op.stats.bytes_unwritten,
 574                         op.stats.floats, op.stats.failed);
 575 }
 576
 577 static unsigned int bch_root_usage(struct cache_set *c)
 578 {
 579         unsigned int bytes = 0;
 580         struct bkey *k;
 581         struct btree *b;
 582         struct btree_iter iter;
 583
 584         goto lock_root;
 585
 586         do {
 587                 rw_unlock(false, b);
 588 lock_root:
 589                 b = c->root;
 590                 rw_lock(false, b, b->level);
 591         } while (b != c->root);
 592
 593         for_each_key_filter(&b->keys, k, &iter, bch_ptr_bad)
 594                 bytes += bkey_bytes(k);
 595
 596         rw_unlock(false, b);
 597
 598         return (bytes * 100) / btree_bytes(c);
 599 }
 600
 601 static size_t bch_cache_size(struct cache_set *c)
 602 {
 603         size_t ret = 0;
 604         struct btree *b;
 605
 606         mutex_lock(&c->bucket_lock);
 607         list_for_each_entry(b, &c->btree_cache, list)
 608                 ret += 1 << (b->keys.page_order + PAGE_SHIFT);
 609
 610         mutex_unlock(&c->bucket_lock);
 611         return ret;
 612 }
 613
 614 static unsigned int bch_cache_max_chain(struct cache_set *c)
 615 {
 616         unsigned int ret = 0;
 617         struct hlist_head *h;
 618
 619         mutex_lock(&c->bucket_lock);
 620
 621         for (h = c->bucket_hash;
 622              h < c->bucket_hash + (1 << BUCKET_HASH_BITS);
 623              h++) {
 624                 unsigned int i = 0;
 625                 struct hlist_node *p;
 626
 627                 hlist_for_each(p, h)
 628                         i++;
 629
 630                 ret = max(ret, i);
 631         }
 632
 633         mutex_unlock(&c->bucket_lock);
 634         return ret;
 635 }
 636
 637 static unsigned int bch_btree_used(struct cache_set *c)
 638 {
 639         return div64_u64(c->gc_stats.key_bytes * 100,
 640                          (c->gc_stats.nodes ?: 1) * btree_bytes(c));
 641 }
 642
 643 static unsigned int bch_average_key_size(struct cache_set *c)
 644 {
 645         return c->gc_stats.nkeys
 646                 ? div64_u64(c->gc_stats.data, c->gc_stats.nkeys)
 647                 : 0;
 648 }
 649
 650 SHOW(__bch_cache_set)
 651 {
 652         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 653
 654         sysfs_print(synchronous,                CACHE_SYNC(&c->sb));
 655         sysfs_print(journal_delay_ms,           c->journal_delay_ms);
 656         sysfs_hprint(bucket_size,               bucket_bytes(c));
 657         sysfs_hprint(block_size,                block_bytes(c));
 658         sysfs_print(tree_depth,                 c->root->level);
 659         sysfs_print(root_usage_percent,         bch_root_usage(c));
 660
 661         sysfs_hprint(btree_cache_size,          bch_cache_size(c));
 662         sysfs_print(btree_cache_max_chain,      bch_cache_max_chain(c));
 663         sysfs_print(cache_available_percent,    100 - c->gc_stats.in_use);
 664
 665         sysfs_print_time_stats(&c->btree_gc_time,       btree_gc, sec, ms);
 666         sysfs_print_time_stats(&c->btree_split_time,    btree_split, sec, us);
 667         sysfs_print_time_stats(&c->sort.time,           btree_sort, ms, us);
 668         sysfs_print_time_stats(&c->btree_read_time,     btree_read, ms, us);
 669
 670         sysfs_print(btree_used_percent, bch_btree_used(c));
 671         sysfs_print(btree_nodes,        c->gc_stats.nodes);
 672         sysfs_hprint(average_key_size,  bch_average_key_size(c));
 673
 674         sysfs_print(cache_read_races,
 675                     atomic_long_read(&c->cache_read_races));
 676
 677         sysfs_print(reclaim,
 678                     atomic_long_read(&c->reclaim));
 679
 680         sysfs_print(flush_write,
 681                     atomic_long_read(&c->flush_write));
 682
 683         sysfs_print(retry_flush_write,
 684                     atomic_long_read(&c->retry_flush_write));
 685
 686         sysfs_print(writeback_keys_done,
 687                     atomic_long_read(&c->writeback_keys_done));
 688         sysfs_print(writeback_keys_failed,
 689                     atomic_long_read(&c->writeback_keys_failed));
 690
 691         if (attr == &sysfs_errors)
 692                 return bch_snprint_string_list(buf, PAGE_SIZE, error_actions,
 693                                                c->on_error);
 694
 695         /* See count_io_errors for why 88 */
 696         sysfs_print(io_error_halflife,  c->error_decay * 88);
 697         sysfs_print(io_error_limit,     c->error_limit);
 698
 699         sysfs_hprint(congested,
 700                      ((uint64_t) bch_get_congested(c)) << 9);
 701         sysfs_print(congested_read_threshold_us,
 702                     c->congested_read_threshold_us);
 703         sysfs_print(congested_write_threshold_us,
 704                     c->congested_write_threshold_us);
 705
 706         sysfs_print(active_journal_entries,     fifo_used(&c->journal.pin));
 707         sysfs_printf(verify,                    "%i", c->verify);
 708         sysfs_printf(key_merging_disabled,      "%i", c->key_merging_disabled);
 709         sysfs_printf(expensive_debug_checks,
 710                      "%i", c->expensive_debug_checks);
 711         sysfs_printf(gc_always_rewrite,         "%i", c->gc_always_rewrite);
 712         sysfs_printf(btree_shrinker_disabled,   "%i", c->shrinker_disabled);
 713         sysfs_printf(copy_gc_enabled,           "%i", c->copy_gc_enabled);
 714         sysfs_printf(io_disable,                "%i",
 715                      test_bit(CACHE_SET_IO_DISABLE, &c->flags));
 716
 717         if (attr == &sysfs_bset_tree_stats)
 718                 return bch_bset_print_stats(c, buf);
 719
 720         return 0;
 721 }
 722 SHOW_LOCKED(bch_cache_set)
 723
 724 STORE(__bch_cache_set)
 725 {
 726         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 727         ssize_t v;
 728
 729         if (attr == &sysfs_unregister)
 730                 bch_cache_set_unregister(c);
 731
 732         if (attr == &sysfs_stop)
 733                 bch_cache_set_stop(c);
 734
 735         if (attr == &sysfs_synchronous) {
 736                 bool sync = strtoul_or_return(buf);
 737
 738                 if (sync != CACHE_SYNC(&c->sb)) {
 739                         SET_CACHE_SYNC(&c->sb, sync);
 740                         bcache_write_super(c);
 741                 }
 742         }
 743
 744         if (attr == &sysfs_flash_vol_create) {
 745                 int r;
 746                 uint64_t v;
 747
 748                 strtoi_h_or_return(buf, v);
 749
 750                 r = bch_flash_dev_create(c, v);
 751                 if (r)
 752                         return r;
 753         }
 754
 755         if (attr == &sysfs_clear_stats) {
 756                 atomic_long_set(&c->writeback_keys_done,        0);
 757                 atomic_long_set(&c->writeback_keys_failed,      0);
 758
 759                 memset(&c->gc_stats, 0, sizeof(struct gc_stat));
 760                 bch_cache_accounting_clear(&c->accounting);
 761         }
 762
 763         if (attr == &sysfs_trigger_gc) {
 764                 /*
 765                  * Garbage collection thread only works when sectors_to_gc < 0,
 766                  * when users write to sysfs entry trigger_gc, most of time
 767                  * they want to forcibly triger gargage collection. Here -1 is
 768                  * set to c->sectors_to_gc, to make gc_should_run() give a
 769                  * chance to permit gc thread to run. "give a chance" means
 770                  * before going into gc_should_run(), there is still chance
 771                  * that c->sectors_to_gc being set to other positive value. So
 772                  * writing sysfs entry trigger_gc won't always make sure gc
 773                  * thread takes effect.
 774                  */
 775                 atomic_set(&c->sectors_to_gc, -1);
 776                 wake_up_gc(c);
 777         }
 778
 779         if (attr == &sysfs_prune_cache) {
 780                 struct shrink_control sc;
 781
 782                 sc.gfp_mask = GFP_KERNEL;
 783                 sc.nr_to_scan = strtoul_or_return(buf);
 784                 c->shrink.scan_objects(&c->shrink, &sc);
 785         }
 786
 787         sysfs_strtoul(congested_read_threshold_us,
 788                       c->congested_read_threshold_us);
 789         sysfs_strtoul(congested_write_threshold_us,
 790                       c->congested_write_threshold_us);
 791
 792         if (attr == &sysfs_errors) {
 793                 v = __sysfs_match_string(error_actions, -1, buf);
 794                 if (v < 0)
 795                         return v;
 796
 797                 c->on_error = v;
 798         }
 799
 800         if (attr == &sysfs_io_error_limit)
 801                 c->error_limit = strtoul_or_return(buf);
 802
 803         /* See count_io_errors() for why 88 */
 804         if (attr == &sysfs_io_error_halflife) {
 805                 unsigned long v = 0;
 806                 ssize_t ret;
 807
 808                 ret = strtoul_safe_clamp(buf, v, 0, UINT_MAX);
 809                 if (!ret) {
 810                         c->error_decay = v / 88;
 811                         return size;
 812                 }
 813                 return ret;
 814         }
 815
 816         if (attr == &sysfs_io_disable) {
 817                 v = strtoul_or_return(buf);
 818                 if (v) {
 819                         if (test_and_set_bit(CACHE_SET_IO_DISABLE,
 820                                              &c->flags))
 821                                 pr_warn("CACHE_SET_IO_DISABLE already set");
 822                 } else {
 823                         if (!test_and_clear_bit(CACHE_SET_IO_DISABLE,
 824                                                 &c->flags))
 825                                 pr_warn("CACHE_SET_IO_DISABLE already cleared");
 826                 }
 827         }
 828
 829         sysfs_strtoul(journal_delay_ms,         c->journal_delay_ms);
 830         sysfs_strtoul(verify,                   c->verify);
 831         sysfs_strtoul(key_merging_disabled,     c->key_merging_disabled);
 832         sysfs_strtoul(expensive_debug_checks,   c->expensive_debug_checks);
 833         sysfs_strtoul(gc_always_rewrite,        c->gc_always_rewrite);
 834         sysfs_strtoul(btree_shrinker_disabled,  c->shrinker_disabled);
 835         sysfs_strtoul(copy_gc_enabled,          c->copy_gc_enabled);
 836
 837         return size;
 838 }
 839 STORE_LOCKED(bch_cache_set)
 840
 841 SHOW(bch_cache_set_internal)
 842 {
 843         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 844
 845         return bch_cache_set_show(&c->kobj, attr, buf);
 846 }
 847
 848 STORE(bch_cache_set_internal)
 849 {
 850         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 851
 852         return bch_cache_set_store(&c->kobj, attr, buf, size);
 853 }
 854
 855 static void bch_cache_set_internal_release(struct kobject *k)
 856 {
 857 }
 858
 859 static struct attribute *bch_cache_set_files[] = {
 860         &sysfs_unregister,
 861         &sysfs_stop,
 862         &sysfs_synchronous,
 863         &sysfs_journal_delay_ms,
 864         &sysfs_flash_vol_create,
 865
 866         &sysfs_bucket_size,
 867         &sysfs_block_size,
 868         &sysfs_tree_depth,
 869         &sysfs_root_usage_percent,
 870         &sysfs_btree_cache_size,
 871         &sysfs_cache_available_percent,
 872
 873         &sysfs_average_key_size,
 874
 875         &sysfs_errors,
 876         &sysfs_io_error_limit,
 877         &sysfs_io_error_halflife,
 878         &sysfs_congested,
 879         &sysfs_congested_read_threshold_us,
 880         &sysfs_congested_write_threshold_us,
 881         &sysfs_clear_stats,
 882         NULL
 883 };
 884 KTYPE(bch_cache_set);
 885
 886 static struct attribute *bch_cache_set_internal_files[] = {
 887         &sysfs_active_journal_entries,
 888
 889         sysfs_time_stats_attribute_list(btree_gc, sec, ms)
 890         sysfs_time_stats_attribute_list(btree_split, sec, us)
 891         sysfs_time_stats_attribute_list(btree_sort, ms, us)
 892         sysfs_time_stats_attribute_list(btree_read, ms, us)
 893
 894         &sysfs_btree_nodes,
 895         &sysfs_btree_used_percent,
 896         &sysfs_btree_cache_max_chain,
 897
 898         &sysfs_bset_tree_stats,
 899         &sysfs_cache_read_races,
 900         &sysfs_reclaim,
 901         &sysfs_flush_write,
 902         &sysfs_retry_flush_write,
 903         &sysfs_writeback_keys_done,
 904         &sysfs_writeback_keys_failed,
 905
 906         &sysfs_trigger_gc,
 907         &sysfs_prune_cache,
 908 #ifdef CONFIG_BCACHE_DEBUG
 909         &sysfs_verify,
 910         &sysfs_key_merging_disabled,
 911         &sysfs_expensive_debug_checks,
 912 #endif
 913         &sysfs_gc_always_rewrite,
 914         &sysfs_btree_shrinker_disabled,
 915         &sysfs_copy_gc_enabled,
 916         &sysfs_io_disable,
 917         NULL
 918 };
 919 KTYPE(bch_cache_set_internal);
 920
 921 static int __bch_cache_cmp(const void *l, const void *r)
 922 {
 923         return *((uint16_t *)r) - *((uint16_t *)l);
 924 }
 925
 926 SHOW(__bch_cache)
 927 {
 928         struct cache *ca = container_of(kobj, struct cache, kobj);
 929
 930         sysfs_hprint(bucket_size,       bucket_bytes(ca));
 931         sysfs_hprint(block_size,        block_bytes(ca));
 932         sysfs_print(nbuckets,           ca->sb.nbuckets);
 933         sysfs_print(discard,            ca->discard);
 934         sysfs_hprint(written, atomic_long_read(&ca->sectors_written) << 9);
 935         sysfs_hprint(btree_written,
 936                      atomic_long_read(&ca->btree_sectors_written) << 9);
 937         sysfs_hprint(metadata_written,
 938                      (atomic_long_read(&ca->meta_sectors_written) +
 939                       atomic_long_read(&ca->btree_sectors_written)) << 9);
 940
 941         sysfs_print(io_errors,
 942                     atomic_read(&ca->io_errors) >> IO_ERROR_SHIFT);
 943
 944         if (attr == &sysfs_cache_replacement_policy)
 945                 return bch_snprint_string_list(buf, PAGE_SIZE,
 946                                                cache_replacement_policies,
 947                                                CACHE_REPLACEMENT(&ca->sb));
 948
 949         if (attr == &sysfs_priority_stats) {
 950                 struct bucket *b;
 951                 size_t n = ca->sb.nbuckets, i;
 952                 size_t unused = 0, available = 0, dirty = 0, meta = 0;
 953                 uint64_t sum = 0;
 954                 /* Compute 31 quantiles */
 955                 uint16_t q[31], *p, *cached;
 956                 ssize_t ret;
 957
 958                 cached = p = vmalloc(array_size(sizeof(uint16_t),
 959                                                 ca->sb.nbuckets));
 960                 if (!p)
 961                         return -ENOMEM;
 962
 963                 mutex_lock(&ca->set->bucket_lock);
 964                 for_each_bucket(b, ca) {
 965                         if (!GC_SECTORS_USED(b))
 966                                 unused++;
 967                         if (GC_MARK(b) == GC_MARK_RECLAIMABLE)
 968                                 available++;
 969                         if (GC_MARK(b) == GC_MARK_DIRTY)
 970                                 dirty++;
 971                         if (GC_MARK(b) == GC_MARK_METADATA)
 972                                 meta++;
 973                 }
 974
 975                 for (i = ca->sb.first_bucket; i < n; i++)
 976                         p[i] = ca->buckets[i].prio;
 977                 mutex_unlock(&ca->set->bucket_lock);
 978
 979                 sort(p, n, sizeof(uint16_t), __bch_cache_cmp, NULL);
 980
 981                 while (n &&
 982                        !cached[n - 1])
 983                         --n;
 984
 985                 unused = ca->sb.nbuckets - n;
 986
 987                 while (cached < p + n &&
 988                        *cached == BTREE_PRIO)
 989                         cached++, n--;
 990
 991                 for (i = 0; i < n; i++)
 992                         sum += INITIAL_PRIO - cached[i];
 993
 994                 if (n)
 995                         sum = div64_u64(sum, n);
 996
 997                 for (i = 0; i < ARRAY_SIZE(q); i++)
 998                         q[i] = INITIAL_PRIO - cached[n * (i + 1) /
 999                                 (ARRAY_SIZE(q) + 1)];
1000
1001                 vfree(p);
1002
1003                 ret = scnprintf(buf, PAGE_SIZE,
1004                                 "Unused:                %zu%%\n"
1005                                 "Clean:         %zu%%\n"
1006                                 "Dirty:         %zu%%\n"
1007                                 "Metadata:      %zu%%\n"
1008                                 "Average:       %llu\n"
1009                                 "Sectors per Q: %zu\n"
1010                                 "Quantiles:     [",
1011                                 unused * 100 / (size_t) ca->sb.nbuckets,
1012                                 available * 100 / (size_t) ca->sb.nbuckets,
1013                                 dirty * 100 / (size_t) ca->sb.nbuckets,
1014                                 meta * 100 / (size_t) ca->sb.nbuckets, sum,
1015                                 n * ca->sb.bucket_size / (ARRAY_SIZE(q) + 1));
1016
1017                 for (i = 0; i < ARRAY_SIZE(q); i++)
1018                         ret += scnprintf(buf + ret, PAGE_SIZE - ret,
1019                                          "%u ", q[i]);
1020                 ret--;
1021
1022                 ret += scnprintf(buf + ret, PAGE_SIZE - ret, "]\n");
1023
1024                 return ret;
1025         }
1026
1027         return 0;
1028 }
1029 SHOW_LOCKED(bch_cache)
1030
1031 STORE(__bch_cache)
1032 {
1033         struct cache *ca = container_of(kobj, struct cache, kobj);
1034         ssize_t v;
1035
1036         if (attr == &sysfs_discard) {
1037                 bool v = strtoul_or_return(buf);
1038
1039                 if (blk_queue_discard(bdev_get_queue(ca->bdev)))
1040                         ca->discard = v;
1041
1042                 if (v != CACHE_DISCARD(&ca->sb)) {
1043                         SET_CACHE_DISCARD(&ca->sb, v);
1044                         bcache_write_super(ca->set);
1045                 }
1046         }
1047
1048         if (attr == &sysfs_cache_replacement_policy) {
1049                 v = __sysfs_match_string(cache_replacement_policies, -1, buf);
1050                 if (v < 0)
1051                         return v;
1052
1053                 if ((unsigned int) v != CACHE_REPLACEMENT(&ca->sb)) {
1054                         mutex_lock(&ca->set->bucket_lock);
1055                         SET_CACHE_REPLACEMENT(&ca->sb, v);
1056                         mutex_unlock(&ca->set->bucket_lock);
1057
1058                         bcache_write_super(ca->set);
1059                 }
1060         }
1061
1062         if (attr == &sysfs_clear_stats) {
1063                 atomic_long_set(&ca->sectors_written, 0);
1064                 atomic_long_set(&ca->btree_sectors_written, 0);
1065                 atomic_long_set(&ca->meta_sectors_written, 0);
1066                 atomic_set(&ca->io_count, 0);
1067                 atomic_set(&ca->io_errors, 0);
1068         }
1069
1070         return size;
1071 }
1072 STORE_LOCKED(bch_cache)
1073
1074 static struct attribute *bch_cache_files[] = {
1075         &sysfs_bucket_size,
1076         &sysfs_block_size,
1077         &sysfs_nbuckets,
1078         &sysfs_priority_stats,
1079         &sysfs_discard,
1080         &sysfs_written,
1081         &sysfs_btree_written,
1082         &sysfs_metadata_written,
1083         &sysfs_io_errors,
1084         &sysfs_clear_stats,
1085         &sysfs_cache_replacement_policy,
1086         NULL
1087 };
1088 KTYPE(bch_cache);