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 0 ("writethrough") */
  20 static const char * const bch_cache_modes[] = {
  21         "writethrough",
  22         "writeback",
  23         "writearound",
  24         "none",
  25         NULL
  26 };
  27
  28 /* Default is 0 ("auto") */
  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 read_attribute(backing_dev_name);
  71 read_attribute(backing_dev_uuid);
  72
  73 sysfs_time_stats_attribute(btree_gc,    sec, ms);
  74 sysfs_time_stats_attribute(btree_split, sec, us);
  75 sysfs_time_stats_attribute(btree_sort,  ms,  us);
  76 sysfs_time_stats_attribute(btree_read,  ms,  us);
  77
  78 read_attribute(btree_nodes);
  79 read_attribute(btree_used_percent);
  80 read_attribute(average_key_size);
  81 read_attribute(dirty_data);
  82 read_attribute(bset_tree_stats);
  83
  84 read_attribute(state);
  85 read_attribute(cache_read_races);
  86 read_attribute(reclaim);
  87 read_attribute(flush_write);
  88 read_attribute(retry_flush_write);
  89 read_attribute(writeback_keys_done);
  90 read_attribute(writeback_keys_failed);
  91 read_attribute(io_errors);
  92 read_attribute(congested);
  93 read_attribute(cutoff_writeback);
  94 read_attribute(cutoff_writeback_sync);
  95 rw_attribute(congested_read_threshold_us);
  96 rw_attribute(congested_write_threshold_us);
  97
  98 rw_attribute(sequential_cutoff);
  99 rw_attribute(data_csum);
 100 rw_attribute(cache_mode);
 101 rw_attribute(stop_when_cache_set_failed);
 102 rw_attribute(writeback_metadata);
 103 rw_attribute(writeback_running);
 104 rw_attribute(writeback_percent);
 105 rw_attribute(writeback_delay);
 106 rw_attribute(writeback_rate);
 107
 108 rw_attribute(writeback_rate_update_seconds);
 109 rw_attribute(writeback_rate_i_term_inverse);
 110 rw_attribute(writeback_rate_p_term_inverse);
 111 rw_attribute(writeback_rate_minimum);
 112 read_attribute(writeback_rate_debug);
 113
 114 read_attribute(stripe_size);
 115 read_attribute(partial_stripes_expensive);
 116
 117 rw_attribute(synchronous);
 118 rw_attribute(journal_delay_ms);
 119 rw_attribute(io_disable);
 120 rw_attribute(discard);
 121 rw_attribute(running);
 122 rw_attribute(label);
 123 rw_attribute(readahead);
 124 rw_attribute(errors);
 125 rw_attribute(io_error_limit);
 126 rw_attribute(io_error_halflife);
 127 rw_attribute(verify);
 128 rw_attribute(bypass_torture_test);
 129 rw_attribute(key_merging_disabled);
 130 rw_attribute(gc_always_rewrite);
 131 rw_attribute(expensive_debug_checks);
 132 rw_attribute(cache_replacement_policy);
 133 rw_attribute(btree_shrinker_disabled);
 134 rw_attribute(copy_gc_enabled);
 135 rw_attribute(gc_after_writeback);
 136 rw_attribute(size);
 137
 138 static ssize_t bch_snprint_string_list(char *buf,
 139                                        size_t size,
 140                                        const char * const list[],
 141                                        size_t selected)
 142 {
 143         char *out = buf;
 144         size_t i;
 145
 146         for (i = 0; list[i]; i++)
 147                 out += snprintf(out, buf + size - out,
 148                                 i == selected ? "[%s] " : "%s ", list[i]);
 149
 150         out[-1] = '\n';
 151         return out - buf;
 152 }
 153
 154 SHOW(__bch_cached_dev)
 155 {
 156         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 157                                              disk.kobj);
 158         char const *states[] = { "no cache", "clean", "dirty", "inconsistent" };
 159         int wb = dc->writeback_running;
 160
 161 #define var(stat)               (dc->stat)
 162
 163         if (attr == &sysfs_cache_mode)
 164                 return bch_snprint_string_list(buf, PAGE_SIZE,
 165                                                bch_cache_modes,
 166                                                BDEV_CACHE_MODE(&dc->sb));
 167
 168         if (attr == &sysfs_stop_when_cache_set_failed)
 169                 return bch_snprint_string_list(buf, PAGE_SIZE,
 170                                                bch_stop_on_failure_modes,
 171                                                dc->stop_when_cache_set_failed);
 172
 173
 174         sysfs_printf(data_csum,         "%i", dc->disk.data_csum);
 175         var_printf(verify,              "%i");
 176         var_printf(bypass_torture_test, "%i");
 177         var_printf(writeback_metadata,  "%i");
 178         var_printf(writeback_running,   "%i");
 179         var_print(writeback_delay);
 180         var_print(writeback_percent);
 181         sysfs_hprint(writeback_rate,
 182                      wb ? atomic_long_read(&dc->writeback_rate.rate) << 9 : 0);
 183         sysfs_hprint(io_errors,         atomic_read(&dc->io_errors));
 184         sysfs_printf(io_error_limit,    "%i", dc->error_limit);
 185         sysfs_printf(io_disable,        "%i", dc->io_disable);
 186         var_print(writeback_rate_update_seconds);
 187         var_print(writeback_rate_i_term_inverse);
 188         var_print(writeback_rate_p_term_inverse);
 189         var_print(writeback_rate_minimum);
 190
 191         if (attr == &sysfs_writeback_rate_debug) {
 192                 char rate[20];
 193                 char dirty[20];
 194                 char target[20];
 195                 char proportional[20];
 196                 char integral[20];
 197                 char change[20];
 198                 s64 next_io;
 199
 200                 /*
 201                  * Except for dirty and target, other values should
 202                  * be 0 if writeback is not running.
 203                  */
 204                 bch_hprint(rate,
 205                            wb ? atomic_long_read(&dc->writeback_rate.rate) << 9
 206                               : 0);
 207                 bch_hprint(dirty, bcache_dev_sectors_dirty(&dc->disk) << 9);
 208                 bch_hprint(target, dc->writeback_rate_target << 9);
 209                 bch_hprint(proportional,
 210                            wb ? dc->writeback_rate_proportional << 9 : 0);
 211                 bch_hprint(integral,
 212                            wb ? dc->writeback_rate_integral_scaled << 9 : 0);
 213                 bch_hprint(change, wb ? dc->writeback_rate_change << 9 : 0);
 214                 next_io = wb ? div64_s64(dc->writeback_rate.next-local_clock(),
 215                                          NSEC_PER_MSEC) : 0;
 216
 217                 return sprintf(buf,
 218                                "rate:\t\t%s/sec\n"
 219                                "dirty:\t\t%s\n"
 220                                "target:\t\t%s\n"
 221                                "proportional:\t%s\n"
 222                                "integral:\t%s\n"
 223                                "change:\t\t%s/sec\n"
 224                                "next io:\t%llims\n",
 225                                rate, dirty, target, proportional,
 226                                integral, change, next_io);
 227         }
 228
 229         sysfs_hprint(dirty_data,
 230                      bcache_dev_sectors_dirty(&dc->disk) << 9);
 231
 232         sysfs_hprint(stripe_size,        ((uint64_t)dc->disk.stripe_size) << 9);
 233         var_printf(partial_stripes_expensive,   "%u");
 234
 235         var_hprint(sequential_cutoff);
 236         var_hprint(readahead);
 237
 238         sysfs_print(running,            atomic_read(&dc->running));
 239         sysfs_print(state,              states[BDEV_STATE(&dc->sb)]);
 240
 241         if (attr == &sysfs_label) {
 242                 memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
 243                 buf[SB_LABEL_SIZE + 1] = '\0';
 244                 strcat(buf, "\n");
 245                 return strlen(buf);
 246         }
 247
 248         if (attr == &sysfs_backing_dev_name) {
 249                 snprintf(buf, BDEVNAME_SIZE + 1, "%s", dc->backing_dev_name);
 250                 strcat(buf, "\n");
 251                 return strlen(buf);
 252         }
 253
 254         if (attr == &sysfs_backing_dev_uuid) {
 255                 /* convert binary uuid into 36-byte string plus '\0' */
 256                 snprintf(buf, 36+1, "%pU", dc->sb.uuid);
 257                 strcat(buf, "\n");
 258                 return strlen(buf);
 259         }
 260
 261 #undef var
 262         return 0;
 263 }
 264 SHOW_LOCKED(bch_cached_dev)
 265
 266 STORE(__cached_dev)
 267 {
 268         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 269                                              disk.kobj);
 270         ssize_t v;
 271         struct cache_set *c;
 272         struct kobj_uevent_env *env;
 273
 274 #define d_strtoul(var)          sysfs_strtoul(var, dc->var)
 275 #define d_strtoul_nonzero(var)  sysfs_strtoul_clamp(var, dc->var, 1, INT_MAX)
 276 #define d_strtoi_h(var)         sysfs_hatoi(var, dc->var)
 277
 278         sysfs_strtoul(data_csum,        dc->disk.data_csum);
 279         d_strtoul(verify);
 280         sysfs_strtoul_bool(bypass_torture_test, dc->bypass_torture_test);
 281         sysfs_strtoul_bool(writeback_metadata, dc->writeback_metadata);
 282         sysfs_strtoul_bool(writeback_running, dc->writeback_running);
 283         sysfs_strtoul_clamp(writeback_delay, dc->writeback_delay, 0, UINT_MAX);
 284
 285         sysfs_strtoul_clamp(writeback_percent, dc->writeback_percent,
 286                             0, bch_cutoff_writeback);
 287
 288         if (attr == &sysfs_writeback_rate) {
 289                 ssize_t ret;
 290                 long int v = atomic_long_read(&dc->writeback_rate.rate);
 291
 292                 ret = strtoul_safe_clamp(buf, v, 1, INT_MAX);
 293
 294                 if (!ret) {
 295                         atomic_long_set(&dc->writeback_rate.rate, v);
 296                         ret = size;
 297                 }
 298
 299                 return ret;
 300         }
 301
 302         sysfs_strtoul_clamp(writeback_rate_update_seconds,
 303                             dc->writeback_rate_update_seconds,
 304                             1, WRITEBACK_RATE_UPDATE_SECS_MAX);
 305         sysfs_strtoul_clamp(writeback_rate_i_term_inverse,
 306                             dc->writeback_rate_i_term_inverse,
 307                             1, UINT_MAX);
 308         sysfs_strtoul_clamp(writeback_rate_p_term_inverse,
 309                             dc->writeback_rate_p_term_inverse,
 310                             1, UINT_MAX);
 311         sysfs_strtoul_clamp(writeback_rate_minimum,
 312                             dc->writeback_rate_minimum,
 313                             1, UINT_MAX);
 314
 315         sysfs_strtoul_clamp(io_error_limit, dc->error_limit, 0, INT_MAX);
 316
 317         if (attr == &sysfs_io_disable) {
 318                 int v = strtoul_or_return(buf);
 319
 320                 dc->io_disable = v ? 1 : 0;
 321         }
 322
 323         sysfs_strtoul_clamp(sequential_cutoff,
 324                             dc->sequential_cutoff,
 325                             0, UINT_MAX);
 326         d_strtoi_h(readahead);
 327
 328         if (attr == &sysfs_clear_stats)
 329                 bch_cache_accounting_clear(&dc->accounting);
 330
 331         if (attr == &sysfs_running &&
 332             strtoul_or_return(buf))
 333                 bch_cached_dev_run(dc);
 334
 335         if (attr == &sysfs_cache_mode) {
 336                 v = __sysfs_match_string(bch_cache_modes, -1, buf);
 337                 if (v < 0)
 338                         return v;
 339
 340                 if ((unsigned int) v != BDEV_CACHE_MODE(&dc->sb)) {
 341                         SET_BDEV_CACHE_MODE(&dc->sb, v);
 342                         bch_write_bdev_super(dc, NULL);
 343                 }
 344         }
 345
 346         if (attr == &sysfs_stop_when_cache_set_failed) {
 347                 v = __sysfs_match_string(bch_stop_on_failure_modes, -1, buf);
 348                 if (v < 0)
 349                         return v;
 350
 351                 dc->stop_when_cache_set_failed = v;
 352         }
 353
 354         if (attr == &sysfs_label) {
 355                 if (size > SB_LABEL_SIZE)
 356                         return -EINVAL;
 357                 memcpy(dc->sb.label, buf, size);
 358                 if (size < SB_LABEL_SIZE)
 359                         dc->sb.label[size] = '\0';
 360                 if (size && dc->sb.label[size - 1] == '\n')
 361                         dc->sb.label[size - 1] = '\0';
 362                 bch_write_bdev_super(dc, NULL);
 363                 if (dc->disk.c) {
 364                         memcpy(dc->disk.c->uuids[dc->disk.id].label,
 365                                buf, SB_LABEL_SIZE);
 366                         bch_uuid_write(dc->disk.c);
 367                 }
 368                 env = kzalloc(sizeof(struct kobj_uevent_env), GFP_KERNEL);
 369                 if (!env)
 370                         return -ENOMEM;
 371                 add_uevent_var(env, "DRIVER=bcache");
 372                 add_uevent_var(env, "CACHED_UUID=%pU", dc->sb.uuid),
 373                 add_uevent_var(env, "CACHED_LABEL=%s", buf);
 374                 kobject_uevent_env(&disk_to_dev(dc->disk.disk)->kobj,
 375                                    KOBJ_CHANGE,
 376                                    env->envp);
 377                 kfree(env);
 378         }
 379
 380         if (attr == &sysfs_attach) {
 381                 uint8_t         set_uuid[16];
 382
 383                 if (bch_parse_uuid(buf, set_uuid) < 16)
 384                         return -EINVAL;
 385
 386                 v = -ENOENT;
 387                 list_for_each_entry(c, &bch_cache_sets, list) {
 388                         v = bch_cached_dev_attach(dc, c, set_uuid);
 389                         if (!v)
 390                                 return size;
 391                 }
 392                 if (v == -ENOENT)
 393                         pr_err("Can't attach %s: cache set not found", buf);
 394                 return v;
 395         }
 396
 397         if (attr == &sysfs_detach && dc->disk.c)
 398                 bch_cached_dev_detach(dc);
 399
 400         if (attr == &sysfs_stop)
 401                 bcache_device_stop(&dc->disk);
 402
 403         return size;
 404 }
 405
 406 STORE(bch_cached_dev)
 407 {
 408         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 409                                              disk.kobj);
 410
 411         mutex_lock(&bch_register_lock);
 412         size = __cached_dev_store(kobj, attr, buf, size);
 413
 414         if (attr == &sysfs_writeback_running) {
 415                 /* dc->writeback_running changed in __cached_dev_store() */
 416                 if (IS_ERR_OR_NULL(dc->writeback_thread)) {
 417                         /*
 418                          * reject setting it to 1 via sysfs if writeback
 419                          * kthread is not created yet.
 420                          */
 421                         if (dc->writeback_running) {
 422                                 dc->writeback_running = false;
 423                                 pr_err("%s: failed to run non-existent writeback thread",
 424                                                 dc->disk.disk->disk_name);
 425                         }
 426                 } else
 427                         /*
 428                          * writeback kthread will check if dc->writeback_running
 429                          * is true or false.
 430                          */
 431                         bch_writeback_queue(dc);
 432         }
 433
 434         if (attr == &sysfs_writeback_percent)
 435                 if (!test_and_set_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags))
 436                         schedule_delayed_work(&dc->writeback_rate_update,
 437                                       dc->writeback_rate_update_seconds * HZ);
 438
 439         mutex_unlock(&bch_register_lock);
 440         return size;
 441 }
 442
 443 static struct attribute *bch_cached_dev_files[] = {
 444         &sysfs_attach,
 445         &sysfs_detach,
 446         &sysfs_stop,
 447 #if 0
 448         &sysfs_data_csum,
 449 #endif
 450         &sysfs_cache_mode,
 451         &sysfs_stop_when_cache_set_failed,
 452         &sysfs_writeback_metadata,
 453         &sysfs_writeback_running,
 454         &sysfs_writeback_delay,
 455         &sysfs_writeback_percent,
 456         &sysfs_writeback_rate,
 457         &sysfs_writeback_rate_update_seconds,
 458         &sysfs_writeback_rate_i_term_inverse,
 459         &sysfs_writeback_rate_p_term_inverse,
 460         &sysfs_writeback_rate_minimum,
 461         &sysfs_writeback_rate_debug,
 462         &sysfs_errors,
 463         &sysfs_io_error_limit,
 464         &sysfs_io_disable,
 465         &sysfs_dirty_data,
 466         &sysfs_stripe_size,
 467         &sysfs_partial_stripes_expensive,
 468         &sysfs_sequential_cutoff,
 469         &sysfs_clear_stats,
 470         &sysfs_running,
 471         &sysfs_state,
 472         &sysfs_label,
 473         &sysfs_readahead,
 474 #ifdef CONFIG_BCACHE_DEBUG
 475         &sysfs_verify,
 476         &sysfs_bypass_torture_test,
 477 #endif
 478         &sysfs_backing_dev_name,
 479         &sysfs_backing_dev_uuid,
 480         NULL
 481 };
 482 KTYPE(bch_cached_dev);
 483
 484 SHOW(bch_flash_dev)
 485 {
 486         struct bcache_device *d = container_of(kobj, struct bcache_device,
 487                                                kobj);
 488         struct uuid_entry *u = &d->c->uuids[d->id];
 489
 490         sysfs_printf(data_csum, "%i", d->data_csum);
 491         sysfs_hprint(size,      u->sectors << 9);
 492
 493         if (attr == &sysfs_label) {
 494                 memcpy(buf, u->label, SB_LABEL_SIZE);
 495                 buf[SB_LABEL_SIZE + 1] = '\0';
 496                 strcat(buf, "\n");
 497                 return strlen(buf);
 498         }
 499
 500         return 0;
 501 }
 502
 503 STORE(__bch_flash_dev)
 504 {
 505         struct bcache_device *d = container_of(kobj, struct bcache_device,
 506                                                kobj);
 507         struct uuid_entry *u = &d->c->uuids[d->id];
 508
 509         sysfs_strtoul(data_csum,        d->data_csum);
 510
 511         if (attr == &sysfs_size) {
 512                 uint64_t v;
 513
 514                 strtoi_h_or_return(buf, v);
 515
 516                 u->sectors = v >> 9;
 517                 bch_uuid_write(d->c);
 518                 set_capacity(d->disk, u->sectors);
 519         }
 520
 521         if (attr == &sysfs_label) {
 522                 memcpy(u->label, buf, SB_LABEL_SIZE);
 523                 bch_uuid_write(d->c);
 524         }
 525
 526         if (attr == &sysfs_unregister) {
 527                 set_bit(BCACHE_DEV_DETACHING, &d->flags);
 528                 bcache_device_stop(d);
 529         }
 530
 531         return size;
 532 }
 533 STORE_LOCKED(bch_flash_dev)
 534
 535 static struct attribute *bch_flash_dev_files[] = {
 536         &sysfs_unregister,
 537 #if 0
 538         &sysfs_data_csum,
 539 #endif
 540         &sysfs_label,
 541         &sysfs_size,
 542         NULL
 543 };
 544 KTYPE(bch_flash_dev);
 545
 546 struct bset_stats_op {
 547         struct btree_op op;
 548         size_t nodes;
 549         struct bset_stats stats;
 550 };
 551
 552 static int bch_btree_bset_stats(struct btree_op *b_op, struct btree *b)
 553 {
 554         struct bset_stats_op *op = container_of(b_op, struct bset_stats_op, op);
 555
 556         op->nodes++;
 557         bch_btree_keys_stats(&b->keys, &op->stats);
 558
 559         return MAP_CONTINUE;
 560 }
 561
 562 static int bch_bset_print_stats(struct cache_set *c, char *buf)
 563 {
 564         struct bset_stats_op op;
 565         int ret;
 566
 567         memset(&op, 0, sizeof(op));
 568         bch_btree_op_init(&op.op, -1);
 569
 570         ret = bch_btree_map_nodes(&op.op, c, &ZERO_KEY, bch_btree_bset_stats);
 571         if (ret < 0)
 572                 return ret;
 573
 574         return snprintf(buf, PAGE_SIZE,
 575                         "btree nodes:           %zu\n"
 576                         "written sets:          %zu\n"
 577                         "unwritten sets:                %zu\n"
 578                         "written key bytes:     %zu\n"
 579                         "unwritten key bytes:   %zu\n"
 580                         "floats:                        %zu\n"
 581                         "failed:                        %zu\n",
 582                         op.nodes,
 583                         op.stats.sets_written, op.stats.sets_unwritten,
 584                         op.stats.bytes_written, op.stats.bytes_unwritten,
 585                         op.stats.floats, op.stats.failed);
 586 }
 587
 588 static unsigned int bch_root_usage(struct cache_set *c)
 589 {
 590         unsigned int bytes = 0;
 591         struct bkey *k;
 592         struct btree *b;
 593         struct btree_iter iter;
 594
 595         goto lock_root;
 596
 597         do {
 598                 rw_unlock(false, b);
 599 lock_root:
 600                 b = c->root;
 601                 rw_lock(false, b, b->level);
 602         } while (b != c->root);
 603
 604         for_each_key_filter(&b->keys, k, &iter, bch_ptr_bad)
 605                 bytes += bkey_bytes(k);
 606
 607         rw_unlock(false, b);
 608
 609         return (bytes * 100) / btree_bytes(c);
 610 }
 611
 612 static size_t bch_cache_size(struct cache_set *c)
 613 {
 614         size_t ret = 0;
 615         struct btree *b;
 616
 617         mutex_lock(&c->bucket_lock);
 618         list_for_each_entry(b, &c->btree_cache, list)
 619                 ret += 1 << (b->keys.page_order + PAGE_SHIFT);
 620
 621         mutex_unlock(&c->bucket_lock);
 622         return ret;
 623 }
 624
 625 static unsigned int bch_cache_max_chain(struct cache_set *c)
 626 {
 627         unsigned int ret = 0;
 628         struct hlist_head *h;
 629
 630         mutex_lock(&c->bucket_lock);
 631
 632         for (h = c->bucket_hash;
 633              h < c->bucket_hash + (1 << BUCKET_HASH_BITS);
 634              h++) {
 635                 unsigned int i = 0;
 636                 struct hlist_node *p;
 637
 638                 hlist_for_each(p, h)
 639                         i++;
 640
 641                 ret = max(ret, i);
 642         }
 643
 644         mutex_unlock(&c->bucket_lock);
 645         return ret;
 646 }
 647
 648 static unsigned int bch_btree_used(struct cache_set *c)
 649 {
 650         return div64_u64(c->gc_stats.key_bytes * 100,
 651                          (c->gc_stats.nodes ?: 1) * btree_bytes(c));
 652 }
 653
 654 static unsigned int bch_average_key_size(struct cache_set *c)
 655 {
 656         return c->gc_stats.nkeys
 657                 ? div64_u64(c->gc_stats.data, c->gc_stats.nkeys)
 658                 : 0;
 659 }
 660
 661 SHOW(__bch_cache_set)
 662 {
 663         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 664
 665         sysfs_print(synchronous,                CACHE_SYNC(&c->sb));
 666         sysfs_print(journal_delay_ms,           c->journal_delay_ms);
 667         sysfs_hprint(bucket_size,               bucket_bytes(c));
 668         sysfs_hprint(block_size,                block_bytes(c));
 669         sysfs_print(tree_depth,                 c->root->level);
 670         sysfs_print(root_usage_percent,         bch_root_usage(c));
 671
 672         sysfs_hprint(btree_cache_size,          bch_cache_size(c));
 673         sysfs_print(btree_cache_max_chain,      bch_cache_max_chain(c));
 674         sysfs_print(cache_available_percent,    100 - c->gc_stats.in_use);
 675
 676         sysfs_print_time_stats(&c->btree_gc_time,       btree_gc, sec, ms);
 677         sysfs_print_time_stats(&c->btree_split_time,    btree_split, sec, us);
 678         sysfs_print_time_stats(&c->sort.time,           btree_sort, ms, us);
 679         sysfs_print_time_stats(&c->btree_read_time,     btree_read, ms, us);
 680
 681         sysfs_print(btree_used_percent, bch_btree_used(c));
 682         sysfs_print(btree_nodes,        c->gc_stats.nodes);
 683         sysfs_hprint(average_key_size,  bch_average_key_size(c));
 684
 685         sysfs_print(cache_read_races,
 686                     atomic_long_read(&c->cache_read_races));
 687
 688         sysfs_print(reclaim,
 689                     atomic_long_read(&c->reclaim));
 690
 691         sysfs_print(flush_write,
 692                     atomic_long_read(&c->flush_write));
 693
 694         sysfs_print(retry_flush_write,
 695                     atomic_long_read(&c->retry_flush_write));
 696
 697         sysfs_print(writeback_keys_done,
 698                     atomic_long_read(&c->writeback_keys_done));
 699         sysfs_print(writeback_keys_failed,
 700                     atomic_long_read(&c->writeback_keys_failed));
 701
 702         if (attr == &sysfs_errors)
 703                 return bch_snprint_string_list(buf, PAGE_SIZE, error_actions,
 704                                                c->on_error);
 705
 706         /* See count_io_errors for why 88 */
 707         sysfs_print(io_error_halflife,  c->error_decay * 88);
 708         sysfs_print(io_error_limit,     c->error_limit);
 709
 710         sysfs_hprint(congested,
 711                      ((uint64_t) bch_get_congested(c)) << 9);
 712         sysfs_print(congested_read_threshold_us,
 713                     c->congested_read_threshold_us);
 714         sysfs_print(congested_write_threshold_us,
 715                     c->congested_write_threshold_us);
 716
 717         sysfs_print(cutoff_writeback, bch_cutoff_writeback);
 718         sysfs_print(cutoff_writeback_sync, bch_cutoff_writeback_sync);
 719
 720         sysfs_print(active_journal_entries,     fifo_used(&c->journal.pin));
 721         sysfs_printf(verify,                    "%i", c->verify);
 722         sysfs_printf(key_merging_disabled,      "%i", c->key_merging_disabled);
 723         sysfs_printf(expensive_debug_checks,
 724                      "%i", c->expensive_debug_checks);
 725         sysfs_printf(gc_always_rewrite,         "%i", c->gc_always_rewrite);
 726         sysfs_printf(btree_shrinker_disabled,   "%i", c->shrinker_disabled);
 727         sysfs_printf(copy_gc_enabled,           "%i", c->copy_gc_enabled);
 728         sysfs_printf(gc_after_writeback,        "%i", c->gc_after_writeback);
 729         sysfs_printf(io_disable,                "%i",
 730                      test_bit(CACHE_SET_IO_DISABLE, &c->flags));
 731
 732         if (attr == &sysfs_bset_tree_stats)
 733                 return bch_bset_print_stats(c, buf);
 734
 735         return 0;
 736 }
 737 SHOW_LOCKED(bch_cache_set)
 738
 739 STORE(__bch_cache_set)
 740 {
 741         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 742         ssize_t v;
 743
 744         if (attr == &sysfs_unregister)
 745                 bch_cache_set_unregister(c);
 746
 747         if (attr == &sysfs_stop)
 748                 bch_cache_set_stop(c);
 749
 750         if (attr == &sysfs_synchronous) {
 751                 bool sync = strtoul_or_return(buf);
 752
 753                 if (sync != CACHE_SYNC(&c->sb)) {
 754                         SET_CACHE_SYNC(&c->sb, sync);
 755                         bcache_write_super(c);
 756                 }
 757         }
 758
 759         if (attr == &sysfs_flash_vol_create) {
 760                 int r;
 761                 uint64_t v;
 762
 763                 strtoi_h_or_return(buf, v);
 764
 765                 r = bch_flash_dev_create(c, v);
 766                 if (r)
 767                         return r;
 768         }
 769
 770         if (attr == &sysfs_clear_stats) {
 771                 atomic_long_set(&c->writeback_keys_done,        0);
 772                 atomic_long_set(&c->writeback_keys_failed,      0);
 773
 774                 memset(&c->gc_stats, 0, sizeof(struct gc_stat));
 775                 bch_cache_accounting_clear(&c->accounting);
 776         }
 777
 778         if (attr == &sysfs_trigger_gc)
 779                 force_wake_up_gc(c);
 780
 781         if (attr == &sysfs_prune_cache) {
 782                 struct shrink_control sc;
 783
 784                 sc.gfp_mask = GFP_KERNEL;
 785                 sc.nr_to_scan = strtoul_or_return(buf);
 786                 c->shrink.scan_objects(&c->shrink, &sc);
 787         }
 788
 789         sysfs_strtoul_clamp(congested_read_threshold_us,
 790                             c->congested_read_threshold_us,
 791                             0, UINT_MAX);
 792         sysfs_strtoul_clamp(congested_write_threshold_us,
 793                             c->congested_write_threshold_us,
 794                             0, UINT_MAX);
 795
 796         if (attr == &sysfs_errors) {
 797                 v = __sysfs_match_string(error_actions, -1, buf);
 798                 if (v < 0)
 799                         return v;
 800
 801                 c->on_error = v;
 802         }
 803
 804         sysfs_strtoul_clamp(io_error_limit, c->error_limit, 0, UINT_MAX);
 805
 806         /* See count_io_errors() for why 88 */
 807         if (attr == &sysfs_io_error_halflife) {
 808                 unsigned long v = 0;
 809                 ssize_t ret;
 810
 811                 ret = strtoul_safe_clamp(buf, v, 0, UINT_MAX);
 812                 if (!ret) {
 813                         c->error_decay = v / 88;
 814                         return size;
 815                 }
 816                 return ret;
 817         }
 818
 819         if (attr == &sysfs_io_disable) {
 820                 v = strtoul_or_return(buf);
 821                 if (v) {
 822                         if (test_and_set_bit(CACHE_SET_IO_DISABLE,
 823                                              &c->flags))
 824                                 pr_warn("CACHE_SET_IO_DISABLE already set");
 825                 } else {
 826                         if (!test_and_clear_bit(CACHE_SET_IO_DISABLE,
 827                                                 &c->flags))
 828                                 pr_warn("CACHE_SET_IO_DISABLE already cleared");
 829                 }
 830         }
 831
 832         sysfs_strtoul_clamp(journal_delay_ms,
 833                             c->journal_delay_ms,
 834                             0, USHRT_MAX);
 835         sysfs_strtoul_bool(verify,              c->verify);
 836         sysfs_strtoul_bool(key_merging_disabled, c->key_merging_disabled);
 837         sysfs_strtoul(expensive_debug_checks,   c->expensive_debug_checks);
 838         sysfs_strtoul_bool(gc_always_rewrite,   c->gc_always_rewrite);
 839         sysfs_strtoul_bool(btree_shrinker_disabled, c->shrinker_disabled);
 840         sysfs_strtoul_bool(copy_gc_enabled,     c->copy_gc_enabled);
 841         /*
 842          * write gc_after_writeback here may overwrite an already set
 843          * BCH_DO_AUTO_GC, it doesn't matter because this flag will be
 844          * set in next chance.
 845          */
 846         sysfs_strtoul_clamp(gc_after_writeback, c->gc_after_writeback, 0, 1);
 847
 848         return size;
 849 }
 850 STORE_LOCKED(bch_cache_set)
 851
 852 SHOW(bch_cache_set_internal)
 853 {
 854         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 855
 856         return bch_cache_set_show(&c->kobj, attr, buf);
 857 }
 858
 859 STORE(bch_cache_set_internal)
 860 {
 861         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 862
 863         return bch_cache_set_store(&c->kobj, attr, buf, size);
 864 }
 865
 866 static void bch_cache_set_internal_release(struct kobject *k)
 867 {
 868 }
 869
 870 static struct attribute *bch_cache_set_files[] = {
 871         &sysfs_unregister,
 872         &sysfs_stop,
 873         &sysfs_synchronous,
 874         &sysfs_journal_delay_ms,
 875         &sysfs_flash_vol_create,
 876
 877         &sysfs_bucket_size,
 878         &sysfs_block_size,
 879         &sysfs_tree_depth,
 880         &sysfs_root_usage_percent,
 881         &sysfs_btree_cache_size,
 882         &sysfs_cache_available_percent,
 883
 884         &sysfs_average_key_size,
 885
 886         &sysfs_errors,
 887         &sysfs_io_error_limit,
 888         &sysfs_io_error_halflife,
 889         &sysfs_congested,
 890         &sysfs_congested_read_threshold_us,
 891         &sysfs_congested_write_threshold_us,
 892         &sysfs_clear_stats,
 893         NULL
 894 };
 895 KTYPE(bch_cache_set);
 896
 897 static struct attribute *bch_cache_set_internal_files[] = {
 898         &sysfs_active_journal_entries,
 899
 900         sysfs_time_stats_attribute_list(btree_gc, sec, ms)
 901         sysfs_time_stats_attribute_list(btree_split, sec, us)
 902         sysfs_time_stats_attribute_list(btree_sort, ms, us)
 903         sysfs_time_stats_attribute_list(btree_read, ms, us)
 904
 905         &sysfs_btree_nodes,
 906         &sysfs_btree_used_percent,
 907         &sysfs_btree_cache_max_chain,
 908
 909         &sysfs_bset_tree_stats,
 910         &sysfs_cache_read_races,
 911         &sysfs_reclaim,
 912         &sysfs_flush_write,
 913         &sysfs_retry_flush_write,
 914         &sysfs_writeback_keys_done,
 915         &sysfs_writeback_keys_failed,
 916
 917         &sysfs_trigger_gc,
 918         &sysfs_prune_cache,
 919 #ifdef CONFIG_BCACHE_DEBUG
 920         &sysfs_verify,
 921         &sysfs_key_merging_disabled,
 922         &sysfs_expensive_debug_checks,
 923 #endif
 924         &sysfs_gc_always_rewrite,
 925         &sysfs_btree_shrinker_disabled,
 926         &sysfs_copy_gc_enabled,
 927         &sysfs_gc_after_writeback,
 928         &sysfs_io_disable,
 929         &sysfs_cutoff_writeback,
 930         &sysfs_cutoff_writeback_sync,
 931         NULL
 932 };
 933 KTYPE(bch_cache_set_internal);
 934
 935 static int __bch_cache_cmp(const void *l, const void *r)
 936 {
 937         return *((uint16_t *)r) - *((uint16_t *)l);
 938 }
 939
 940 SHOW(__bch_cache)
 941 {
 942         struct cache *ca = container_of(kobj, struct cache, kobj);
 943
 944         sysfs_hprint(bucket_size,       bucket_bytes(ca));
 945         sysfs_hprint(block_size,        block_bytes(ca));
 946         sysfs_print(nbuckets,           ca->sb.nbuckets);
 947         sysfs_print(discard,            ca->discard);
 948         sysfs_hprint(written, atomic_long_read(&ca->sectors_written) << 9);
 949         sysfs_hprint(btree_written,
 950                      atomic_long_read(&ca->btree_sectors_written) << 9);
 951         sysfs_hprint(metadata_written,
 952                      (atomic_long_read(&ca->meta_sectors_written) +
 953                       atomic_long_read(&ca->btree_sectors_written)) << 9);
 954
 955         sysfs_print(io_errors,
 956                     atomic_read(&ca->io_errors) >> IO_ERROR_SHIFT);
 957
 958         if (attr == &sysfs_cache_replacement_policy)
 959                 return bch_snprint_string_list(buf, PAGE_SIZE,
 960                                                cache_replacement_policies,
 961                                                CACHE_REPLACEMENT(&ca->sb));
 962
 963         if (attr == &sysfs_priority_stats) {
 964                 struct bucket *b;
 965                 size_t n = ca->sb.nbuckets, i;
 966                 size_t unused = 0, available = 0, dirty = 0, meta = 0;
 967                 uint64_t sum = 0;
 968                 /* Compute 31 quantiles */
 969                 uint16_t q[31], *p, *cached;
 970                 ssize_t ret;
 971
 972                 cached = p = vmalloc(array_size(sizeof(uint16_t),
 973                                                 ca->sb.nbuckets));
 974                 if (!p)
 975                         return -ENOMEM;
 976
 977                 mutex_lock(&ca->set->bucket_lock);
 978                 for_each_bucket(b, ca) {
 979                         if (!GC_SECTORS_USED(b))
 980                                 unused++;
 981                         if (GC_MARK(b) == GC_MARK_RECLAIMABLE)
 982                                 available++;
 983                         if (GC_MARK(b) == GC_MARK_DIRTY)
 984                                 dirty++;
 985                         if (GC_MARK(b) == GC_MARK_METADATA)
 986                                 meta++;
 987                 }
 988
 989                 for (i = ca->sb.first_bucket; i < n; i++)
 990                         p[i] = ca->buckets[i].prio;
 991                 mutex_unlock(&ca->set->bucket_lock);
 992
 993                 sort(p, n, sizeof(uint16_t), __bch_cache_cmp, NULL);
 994
 995                 while (n &&
 996                        !cached[n - 1])
 997                         --n;
 998
 999                 while (cached < p + n &&
1000                        *cached == BTREE_PRIO)
1001                         cached++, n--;
1002
1003                 for (i = 0; i < n; i++)
1004                         sum += INITIAL_PRIO - cached[i];
1005
1006                 if (n)
1007                         do_div(sum, n);
1008
1009                 for (i = 0; i < ARRAY_SIZE(q); i++)
1010                         q[i] = INITIAL_PRIO - cached[n * (i + 1) /
1011                                 (ARRAY_SIZE(q) + 1)];
1012
1013                 vfree(p);
1014
1015                 ret = scnprintf(buf, PAGE_SIZE,
1016                                 "Unused:                %zu%%\n"
1017                                 "Clean:         %zu%%\n"
1018                                 "Dirty:         %zu%%\n"
1019                                 "Metadata:      %zu%%\n"
1020                                 "Average:       %llu\n"
1021                                 "Sectors per Q: %zu\n"
1022                                 "Quantiles:     [",
1023                                 unused * 100 / (size_t) ca->sb.nbuckets,
1024                                 available * 100 / (size_t) ca->sb.nbuckets,
1025                                 dirty * 100 / (size_t) ca->sb.nbuckets,
1026                                 meta * 100 / (size_t) ca->sb.nbuckets, sum,
1027                                 n * ca->sb.bucket_size / (ARRAY_SIZE(q) + 1));
1028
1029                 for (i = 0; i < ARRAY_SIZE(q); i++)
1030                         ret += scnprintf(buf + ret, PAGE_SIZE - ret,
1031                                          "%u ", q[i]);
1032                 ret--;
1033
1034                 ret += scnprintf(buf + ret, PAGE_SIZE - ret, "]\n");
1035
1036                 return ret;
1037         }
1038
1039         return 0;
1040 }
1041 SHOW_LOCKED(bch_cache)
1042
1043 STORE(__bch_cache)
1044 {
1045         struct cache *ca = container_of(kobj, struct cache, kobj);
1046         ssize_t v;
1047
1048         if (attr == &sysfs_discard) {
1049                 bool v = strtoul_or_return(buf);
1050
1051                 if (blk_queue_discard(bdev_get_queue(ca->bdev)))
1052                         ca->discard = v;
1053
1054                 if (v != CACHE_DISCARD(&ca->sb)) {
1055                         SET_CACHE_DISCARD(&ca->sb, v);
1056                         bcache_write_super(ca->set);
1057                 }
1058         }
1059
1060         if (attr == &sysfs_cache_replacement_policy) {
1061                 v = __sysfs_match_string(cache_replacement_policies, -1, buf);
1062                 if (v < 0)
1063                         return v;
1064
1065                 if ((unsigned int) v != CACHE_REPLACEMENT(&ca->sb)) {
1066                         mutex_lock(&ca->set->bucket_lock);
1067                         SET_CACHE_REPLACEMENT(&ca->sb, v);
1068                         mutex_unlock(&ca->set->bucket_lock);
1069
1070                         bcache_write_super(ca->set);
1071                 }
1072         }
1073
1074         if (attr == &sysfs_clear_stats) {
1075                 atomic_long_set(&ca->sectors_written, 0);
1076                 atomic_long_set(&ca->btree_sectors_written, 0);
1077                 atomic_long_set(&ca->meta_sectors_written, 0);
1078                 atomic_set(&ca->io_count, 0);
1079                 atomic_set(&ca->io_errors, 0);
1080         }
1081
1082         return size;
1083 }
1084 STORE_LOCKED(bch_cache)
1085
1086 static struct attribute *bch_cache_files[] = {
1087         &sysfs_bucket_size,
1088         &sysfs_block_size,
1089         &sysfs_nbuckets,
1090         &sysfs_priority_stats,
1091         &sysfs_discard,
1092         &sysfs_written,
1093         &sysfs_btree_written,
1094         &sysfs_metadata_written,
1095         &sysfs_io_errors,
1096         &sysfs_clear_stats,
1097         &sysfs_cache_replacement_policy,
1098         NULL
1099 };
1100 KTYPE(bch_cache);