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