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