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         d_strtoul_nonzero(writeback_rate_minimum);
 312
 313         sysfs_strtoul_clamp(io_error_limit, dc->error_limit, 0, INT_MAX);
 314
 315         if (attr == &sysfs_io_disable) {
 316                 int v = strtoul_or_return(buf);
 317
 318                 dc->io_disable = v ? 1 : 0;
 319         }
 320
 321         sysfs_strtoul_clamp(sequential_cutoff,
 322                             dc->sequential_cutoff,
 323                             0, UINT_MAX);
 324         d_strtoi_h(readahead);
 325
 326         if (attr == &sysfs_clear_stats)
 327                 bch_cache_accounting_clear(&dc->accounting);
 328
 329         if (attr == &sysfs_running &&
 330             strtoul_or_return(buf))
 331                 bch_cached_dev_run(dc);
 332
 333         if (attr == &sysfs_cache_mode) {
 334                 v = __sysfs_match_string(bch_cache_modes, -1, buf);
 335                 if (v < 0)
 336                         return v;
 337
 338                 if ((unsigned int) v != BDEV_CACHE_MODE(&dc->sb)) {
 339                         SET_BDEV_CACHE_MODE(&dc->sb, v);
 340                         bch_write_bdev_super(dc, NULL);
 341                 }
 342         }
 343
 344         if (attr == &sysfs_stop_when_cache_set_failed) {
 345                 v = __sysfs_match_string(bch_stop_on_failure_modes, -1, buf);
 346                 if (v < 0)
 347                         return v;
 348
 349                 dc->stop_when_cache_set_failed = v;
 350         }
 351
 352         if (attr == &sysfs_label) {
 353                 if (size > SB_LABEL_SIZE)
 354                         return -EINVAL;
 355                 memcpy(dc->sb.label, buf, size);
 356                 if (size < SB_LABEL_SIZE)
 357                         dc->sb.label[size] = '\0';
 358                 if (size && dc->sb.label[size - 1] == '\n')
 359                         dc->sb.label[size - 1] = '\0';
 360                 bch_write_bdev_super(dc, NULL);
 361                 if (dc->disk.c) {
 362                         memcpy(dc->disk.c->uuids[dc->disk.id].label,
 363                                buf, SB_LABEL_SIZE);
 364                         bch_uuid_write(dc->disk.c);
 365                 }
 366                 env = kzalloc(sizeof(struct kobj_uevent_env), GFP_KERNEL);
 367                 if (!env)
 368                         return -ENOMEM;
 369                 add_uevent_var(env, "DRIVER=bcache");
 370                 add_uevent_var(env, "CACHED_UUID=%pU", dc->sb.uuid),
 371                 add_uevent_var(env, "CACHED_LABEL=%s", buf);
 372                 kobject_uevent_env(&disk_to_dev(dc->disk.disk)->kobj,
 373                                    KOBJ_CHANGE,
 374                                    env->envp);
 375                 kfree(env);
 376         }
 377
 378         if (attr == &sysfs_attach) {
 379                 uint8_t         set_uuid[16];
 380
 381                 if (bch_parse_uuid(buf, set_uuid) < 16)
 382                         return -EINVAL;
 383
 384                 v = -ENOENT;
 385                 list_for_each_entry(c, &bch_cache_sets, list) {
 386                         v = bch_cached_dev_attach(dc, c, set_uuid);
 387                         if (!v)
 388                                 return size;
 389                 }
 390                 if (v == -ENOENT)
 391                         pr_err("Can't attach %s: cache set not found", buf);
 392                 return v;
 393         }
 394
 395         if (attr == &sysfs_detach && dc->disk.c)
 396                 bch_cached_dev_detach(dc);
 397
 398         if (attr == &sysfs_stop)
 399                 bcache_device_stop(&dc->disk);
 400
 401         return size;
 402 }
 403
 404 STORE(bch_cached_dev)
 405 {
 406         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 407                                              disk.kobj);
 408
 409         mutex_lock(&bch_register_lock);
 410         size = __cached_dev_store(kobj, attr, buf, size);
 411
 412         if (attr == &sysfs_writeback_running) {
 413                 /* dc->writeback_running changed in __cached_dev_store() */
 414                 if (IS_ERR_OR_NULL(dc->writeback_thread)) {
 415                         /*
 416                          * reject setting it to 1 via sysfs if writeback
 417                          * kthread is not created yet.
 418                          */
 419                         if (dc->writeback_running) {
 420                                 dc->writeback_running = false;
 421                                 pr_err("%s: failed to run non-existent writeback thread",
 422                                                 dc->disk.disk->disk_name);
 423                         }
 424                 } else
 425                         /*
 426                          * writeback kthread will check if dc->writeback_running
 427                          * is true or false.
 428                          */
 429                         bch_writeback_queue(dc);
 430         }
 431
 432         if (attr == &sysfs_writeback_percent)
 433                 if (!test_and_set_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags))
 434                         schedule_delayed_work(&dc->writeback_rate_update,
 435                                       dc->writeback_rate_update_seconds * HZ);
 436
 437         mutex_unlock(&bch_register_lock);
 438         return size;
 439 }
 440
 441 static struct attribute *bch_cached_dev_files[] = {
 442         &sysfs_attach,
 443         &sysfs_detach,
 444         &sysfs_stop,
 445 #if 0
 446         &sysfs_data_csum,
 447 #endif
 448         &sysfs_cache_mode,
 449         &sysfs_stop_when_cache_set_failed,
 450         &sysfs_writeback_metadata,
 451         &sysfs_writeback_running,
 452         &sysfs_writeback_delay,
 453         &sysfs_writeback_percent,
 454         &sysfs_writeback_rate,
 455         &sysfs_writeback_rate_update_seconds,
 456         &sysfs_writeback_rate_i_term_inverse,
 457         &sysfs_writeback_rate_p_term_inverse,
 458         &sysfs_writeback_rate_minimum,
 459         &sysfs_writeback_rate_debug,
 460         &sysfs_errors,
 461         &sysfs_io_error_limit,
 462         &sysfs_io_disable,
 463         &sysfs_dirty_data,
 464         &sysfs_stripe_size,
 465         &sysfs_partial_stripes_expensive,
 466         &sysfs_sequential_cutoff,
 467         &sysfs_clear_stats,
 468         &sysfs_running,
 469         &sysfs_state,
 470         &sysfs_label,
 471         &sysfs_readahead,
 472 #ifdef CONFIG_BCACHE_DEBUG
 473         &sysfs_verify,
 474         &sysfs_bypass_torture_test,
 475 #endif
 476         &sysfs_backing_dev_name,
 477         &sysfs_backing_dev_uuid,
 478         NULL
 479 };
 480 KTYPE(bch_cached_dev);
 481
 482 SHOW(bch_flash_dev)
 483 {
 484         struct bcache_device *d = container_of(kobj, struct bcache_device,
 485                                                kobj);
 486         struct uuid_entry *u = &d->c->uuids[d->id];
 487
 488         sysfs_printf(data_csum, "%i", d->data_csum);
 489         sysfs_hprint(size,      u->sectors << 9);
 490
 491         if (attr == &sysfs_label) {
 492                 memcpy(buf, u->label, SB_LABEL_SIZE);
 493                 buf[SB_LABEL_SIZE + 1] = '\0';
 494                 strcat(buf, "\n");
 495                 return strlen(buf);
 496         }
 497
 498         return 0;
 499 }
 500
 501 STORE(__bch_flash_dev)
 502 {
 503         struct bcache_device *d = container_of(kobj, struct bcache_device,
 504                                                kobj);
 505         struct uuid_entry *u = &d->c->uuids[d->id];
 506
 507         sysfs_strtoul(data_csum,        d->data_csum);
 508
 509         if (attr == &sysfs_size) {
 510                 uint64_t v;
 511
 512                 strtoi_h_or_return(buf, v);
 513
 514                 u->sectors = v >> 9;
 515                 bch_uuid_write(d->c);
 516                 set_capacity(d->disk, u->sectors);
 517         }
 518
 519         if (attr == &sysfs_label) {
 520                 memcpy(u->label, buf, SB_LABEL_SIZE);
 521                 bch_uuid_write(d->c);
 522         }
 523
 524         if (attr == &sysfs_unregister) {
 525                 set_bit(BCACHE_DEV_DETACHING, &d->flags);
 526                 bcache_device_stop(d);
 527         }
 528
 529         return size;
 530 }
 531 STORE_LOCKED(bch_flash_dev)
 532
 533 static struct attribute *bch_flash_dev_files[] = {
 534         &sysfs_unregister,
 535 #if 0
 536         &sysfs_data_csum,
 537 #endif
 538         &sysfs_label,
 539         &sysfs_size,
 540         NULL
 541 };
 542 KTYPE(bch_flash_dev);
 543
 544 struct bset_stats_op {
 545         struct btree_op op;
 546         size_t nodes;
 547         struct bset_stats stats;
 548 };
 549
 550 static int bch_btree_bset_stats(struct btree_op *b_op, struct btree *b)
 551 {
 552         struct bset_stats_op *op = container_of(b_op, struct bset_stats_op, op);
 553
 554         op->nodes++;
 555         bch_btree_keys_stats(&b->keys, &op->stats);
 556
 557         return MAP_CONTINUE;
 558 }
 559
 560 static int bch_bset_print_stats(struct cache_set *c, char *buf)
 561 {
 562         struct bset_stats_op op;
 563         int ret;
 564
 565         memset(&op, 0, sizeof(op));
 566         bch_btree_op_init(&op.op, -1);
 567
 568         ret = bch_btree_map_nodes(&op.op, c, &ZERO_KEY, bch_btree_bset_stats);
 569         if (ret < 0)
 570                 return ret;
 571
 572         return snprintf(buf, PAGE_SIZE,
 573                         "btree nodes:           %zu\n"
 574                         "written sets:          %zu\n"
 575                         "unwritten sets:                %zu\n"
 576                         "written key bytes:     %zu\n"
 577                         "unwritten key bytes:   %zu\n"
 578                         "floats:                        %zu\n"
 579                         "failed:                        %zu\n",
 580                         op.nodes,
 581                         op.stats.sets_written, op.stats.sets_unwritten,
 582                         op.stats.bytes_written, op.stats.bytes_unwritten,
 583                         op.stats.floats, op.stats.failed);
 584 }
 585
 586 static unsigned int bch_root_usage(struct cache_set *c)
 587 {
 588         unsigned int bytes = 0;
 589         struct bkey *k;
 590         struct btree *b;
 591         struct btree_iter iter;
 592
 593         goto lock_root;
 594
 595         do {
 596                 rw_unlock(false, b);
 597 lock_root:
 598                 b = c->root;
 599                 rw_lock(false, b, b->level);
 600         } while (b != c->root);
 601
 602         for_each_key_filter(&b->keys, k, &iter, bch_ptr_bad)
 603                 bytes += bkey_bytes(k);
 604
 605         rw_unlock(false, b);
 606
 607         return (bytes * 100) / btree_bytes(c);
 608 }
 609
 610 static size_t bch_cache_size(struct cache_set *c)
 611 {
 612         size_t ret = 0;
 613         struct btree *b;
 614
 615         mutex_lock(&c->bucket_lock);
 616         list_for_each_entry(b, &c->btree_cache, list)
 617                 ret += 1 << (b->keys.page_order + PAGE_SHIFT);
 618
 619         mutex_unlock(&c->bucket_lock);
 620         return ret;
 621 }
 622
 623 static unsigned int bch_cache_max_chain(struct cache_set *c)
 624 {
 625         unsigned int ret = 0;
 626         struct hlist_head *h;
 627
 628         mutex_lock(&c->bucket_lock);
 629
 630         for (h = c->bucket_hash;
 631              h < c->bucket_hash + (1 << BUCKET_HASH_BITS);
 632              h++) {
 633                 unsigned int i = 0;
 634                 struct hlist_node *p;
 635
 636                 hlist_for_each(p, h)
 637                         i++;
 638
 639                 ret = max(ret, i);
 640         }
 641
 642         mutex_unlock(&c->bucket_lock);
 643         return ret;
 644 }
 645
 646 static unsigned int bch_btree_used(struct cache_set *c)
 647 {
 648         return div64_u64(c->gc_stats.key_bytes * 100,
 649                          (c->gc_stats.nodes ?: 1) * btree_bytes(c));
 650 }
 651
 652 static unsigned int bch_average_key_size(struct cache_set *c)
 653 {
 654         return c->gc_stats.nkeys
 655                 ? div64_u64(c->gc_stats.data, c->gc_stats.nkeys)
 656                 : 0;
 657 }
 658
 659 SHOW(__bch_cache_set)
 660 {
 661         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 662
 663         sysfs_print(synchronous,                CACHE_SYNC(&c->sb));
 664         sysfs_print(journal_delay_ms,           c->journal_delay_ms);
 665         sysfs_hprint(bucket_size,               bucket_bytes(c));
 666         sysfs_hprint(block_size,                block_bytes(c));
 667         sysfs_print(tree_depth,                 c->root->level);
 668         sysfs_print(root_usage_percent,         bch_root_usage(c));
 669
 670         sysfs_hprint(btree_cache_size,          bch_cache_size(c));
 671         sysfs_print(btree_cache_max_chain,      bch_cache_max_chain(c));
 672         sysfs_print(cache_available_percent,    100 - c->gc_stats.in_use);
 673
 674         sysfs_print_time_stats(&c->btree_gc_time,       btree_gc, sec, ms);
 675         sysfs_print_time_stats(&c->btree_split_time,    btree_split, sec, us);
 676         sysfs_print_time_stats(&c->sort.time,           btree_sort, ms, us);
 677         sysfs_print_time_stats(&c->btree_read_time,     btree_read, ms, us);
 678
 679         sysfs_print(btree_used_percent, bch_btree_used(c));
 680         sysfs_print(btree_nodes,        c->gc_stats.nodes);
 681         sysfs_hprint(average_key_size,  bch_average_key_size(c));
 682
 683         sysfs_print(cache_read_races,
 684                     atomic_long_read(&c->cache_read_races));
 685
 686         sysfs_print(reclaim,
 687                     atomic_long_read(&c->reclaim));
 688
 689         sysfs_print(flush_write,
 690                     atomic_long_read(&c->flush_write));
 691
 692         sysfs_print(retry_flush_write,
 693                     atomic_long_read(&c->retry_flush_write));
 694
 695         sysfs_print(writeback_keys_done,
 696                     atomic_long_read(&c->writeback_keys_done));
 697         sysfs_print(writeback_keys_failed,
 698                     atomic_long_read(&c->writeback_keys_failed));
 699
 700         if (attr == &sysfs_errors)
 701                 return bch_snprint_string_list(buf, PAGE_SIZE, error_actions,
 702                                                c->on_error);
 703
 704         /* See count_io_errors for why 88 */
 705         sysfs_print(io_error_halflife,  c->error_decay * 88);
 706         sysfs_print(io_error_limit,     c->error_limit);
 707
 708         sysfs_hprint(congested,
 709                      ((uint64_t) bch_get_congested(c)) << 9);
 710         sysfs_print(congested_read_threshold_us,
 711                     c->congested_read_threshold_us);
 712         sysfs_print(congested_write_threshold_us,
 713                     c->congested_write_threshold_us);
 714
 715         sysfs_print(cutoff_writeback, bch_cutoff_writeback);
 716         sysfs_print(cutoff_writeback_sync, bch_cutoff_writeback_sync);
 717
 718         sysfs_print(active_journal_entries,     fifo_used(&c->journal.pin));
 719         sysfs_printf(verify,                    "%i", c->verify);
 720         sysfs_printf(key_merging_disabled,      "%i", c->key_merging_disabled);
 721         sysfs_printf(expensive_debug_checks,
 722                      "%i", c->expensive_debug_checks);
 723         sysfs_printf(gc_always_rewrite,         "%i", c->gc_always_rewrite);
 724         sysfs_printf(btree_shrinker_disabled,   "%i", c->shrinker_disabled);
 725         sysfs_printf(copy_gc_enabled,           "%i", c->copy_gc_enabled);
 726         sysfs_printf(gc_after_writeback,        "%i", c->gc_after_writeback);
 727         sysfs_printf(io_disable,                "%i",
 728                      test_bit(CACHE_SET_IO_DISABLE, &c->flags));
 729
 730         if (attr == &sysfs_bset_tree_stats)
 731                 return bch_bset_print_stats(c, buf);
 732
 733         return 0;
 734 }
 735 SHOW_LOCKED(bch_cache_set)
 736
 737 STORE(__bch_cache_set)
 738 {
 739         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 740         ssize_t v;
 741
 742         if (attr == &sysfs_unregister)
 743                 bch_cache_set_unregister(c);
 744
 745         if (attr == &sysfs_stop)
 746                 bch_cache_set_stop(c);
 747
 748         if (attr == &sysfs_synchronous) {
 749                 bool sync = strtoul_or_return(buf);
 750
 751                 if (sync != CACHE_SYNC(&c->sb)) {
 752                         SET_CACHE_SYNC(&c->sb, sync);
 753                         bcache_write_super(c);
 754                 }
 755         }
 756
 757         if (attr == &sysfs_flash_vol_create) {
 758                 int r;
 759                 uint64_t v;
 760
 761                 strtoi_h_or_return(buf, v);
 762
 763                 r = bch_flash_dev_create(c, v);
 764                 if (r)
 765                         return r;
 766         }
 767
 768         if (attr == &sysfs_clear_stats) {
 769                 atomic_long_set(&c->writeback_keys_done,        0);
 770                 atomic_long_set(&c->writeback_keys_failed,      0);
 771
 772                 memset(&c->gc_stats, 0, sizeof(struct gc_stat));
 773                 bch_cache_accounting_clear(&c->accounting);
 774         }
 775
 776         if (attr == &sysfs_trigger_gc)
 777                 force_wake_up_gc(c);
 778
 779         if (attr == &sysfs_prune_cache) {
 780                 struct shrink_control sc;
 781
 782                 sc.gfp_mask = GFP_KERNEL;
 783                 sc.nr_to_scan = strtoul_or_return(buf);
 784                 c->shrink.scan_objects(&c->shrink, &sc);
 785         }
 786
 787         sysfs_strtoul_clamp(congested_read_threshold_us,
 788                             c->congested_read_threshold_us,
 789                             0, UINT_MAX);
 790         sysfs_strtoul_clamp(congested_write_threshold_us,
 791                             c->congested_write_threshold_us,
 792                             0, UINT_MAX);
 793
 794         if (attr == &sysfs_errors) {
 795                 v = __sysfs_match_string(error_actions, -1, buf);
 796                 if (v < 0)
 797                         return v;
 798
 799                 c->on_error = v;
 800         }
 801
 802         if (attr == &sysfs_io_error_limit)
 803                 c->error_limit = strtoul_or_return(buf);
 804
 805         /* See count_io_errors() for why 88 */
 806         if (attr == &sysfs_io_error_halflife)
 807                 c->error_decay = strtoul_or_return(buf) / 88;
 808
 809         if (attr == &sysfs_io_disable) {
 810                 v = strtoul_or_return(buf);
 811                 if (v) {
 812                         if (test_and_set_bit(CACHE_SET_IO_DISABLE,
 813                                              &c->flags))
 814                                 pr_warn("CACHE_SET_IO_DISABLE already set");
 815                 } else {
 816                         if (!test_and_clear_bit(CACHE_SET_IO_DISABLE,
 817                                                 &c->flags))
 818                                 pr_warn("CACHE_SET_IO_DISABLE already cleared");
 819                 }
 820         }
 821
 822         sysfs_strtoul(journal_delay_ms,         c->journal_delay_ms);
 823         sysfs_strtoul_bool(verify,              c->verify);
 824         sysfs_strtoul_bool(key_merging_disabled, c->key_merging_disabled);
 825         sysfs_strtoul(expensive_debug_checks,   c->expensive_debug_checks);
 826         sysfs_strtoul_bool(gc_always_rewrite,   c->gc_always_rewrite);
 827         sysfs_strtoul_bool(btree_shrinker_disabled, c->shrinker_disabled);
 828         sysfs_strtoul_bool(copy_gc_enabled,     c->copy_gc_enabled);
 829         /*
 830          * write gc_after_writeback here may overwrite an already set
 831          * BCH_DO_AUTO_GC, it doesn't matter because this flag will be
 832          * set in next chance.
 833          */
 834         sysfs_strtoul_clamp(gc_after_writeback, c->gc_after_writeback, 0, 1);
 835
 836         return size;
 837 }
 838 STORE_LOCKED(bch_cache_set)
 839
 840 SHOW(bch_cache_set_internal)
 841 {
 842         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 843
 844         return bch_cache_set_show(&c->kobj, attr, buf);
 845 }
 846
 847 STORE(bch_cache_set_internal)
 848 {
 849         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 850
 851         return bch_cache_set_store(&c->kobj, attr, buf, size);
 852 }
 853
 854 static void bch_cache_set_internal_release(struct kobject *k)
 855 {
 856 }
 857
 858 static struct attribute *bch_cache_set_files[] = {
 859         &sysfs_unregister,
 860         &sysfs_stop,
 861         &sysfs_synchronous,
 862         &sysfs_journal_delay_ms,
 863         &sysfs_flash_vol_create,
 864
 865         &sysfs_bucket_size,
 866         &sysfs_block_size,
 867         &sysfs_tree_depth,
 868         &sysfs_root_usage_percent,
 869         &sysfs_btree_cache_size,
 870         &sysfs_cache_available_percent,
 871
 872         &sysfs_average_key_size,
 873
 874         &sysfs_errors,
 875         &sysfs_io_error_limit,
 876         &sysfs_io_error_halflife,
 877         &sysfs_congested,
 878         &sysfs_congested_read_threshold_us,
 879         &sysfs_congested_write_threshold_us,
 880         &sysfs_clear_stats,
 881         NULL
 882 };
 883 KTYPE(bch_cache_set);
 884
 885 static struct attribute *bch_cache_set_internal_files[] = {
 886         &sysfs_active_journal_entries,
 887
 888         sysfs_time_stats_attribute_list(btree_gc, sec, ms)
 889         sysfs_time_stats_attribute_list(btree_split, sec, us)
 890         sysfs_time_stats_attribute_list(btree_sort, ms, us)
 891         sysfs_time_stats_attribute_list(btree_read, ms, us)
 892
 893         &sysfs_btree_nodes,
 894         &sysfs_btree_used_percent,
 895         &sysfs_btree_cache_max_chain,
 896
 897         &sysfs_bset_tree_stats,
 898         &sysfs_cache_read_races,
 899         &sysfs_reclaim,
 900         &sysfs_flush_write,
 901         &sysfs_retry_flush_write,
 902         &sysfs_writeback_keys_done,
 903         &sysfs_writeback_keys_failed,
 904
 905         &sysfs_trigger_gc,
 906         &sysfs_prune_cache,
 907 #ifdef CONFIG_BCACHE_DEBUG
 908         &sysfs_verify,
 909         &sysfs_key_merging_disabled,
 910         &sysfs_expensive_debug_checks,
 911 #endif
 912         &sysfs_gc_always_rewrite,
 913         &sysfs_btree_shrinker_disabled,
 914         &sysfs_copy_gc_enabled,
 915         &sysfs_gc_after_writeback,
 916         &sysfs_io_disable,
 917         &sysfs_cutoff_writeback,
 918         &sysfs_cutoff_writeback_sync,
 919         NULL
 920 };
 921 KTYPE(bch_cache_set_internal);
 922
 923 static int __bch_cache_cmp(const void *l, const void *r)
 924 {
 925         return *((uint16_t *)r) - *((uint16_t *)l);
 926 }
 927
 928 SHOW(__bch_cache)
 929 {
 930         struct cache *ca = container_of(kobj, struct cache, kobj);
 931
 932         sysfs_hprint(bucket_size,       bucket_bytes(ca));
 933         sysfs_hprint(block_size,        block_bytes(ca));
 934         sysfs_print(nbuckets,           ca->sb.nbuckets);
 935         sysfs_print(discard,            ca->discard);
 936         sysfs_hprint(written, atomic_long_read(&ca->sectors_written) << 9);
 937         sysfs_hprint(btree_written,
 938                      atomic_long_read(&ca->btree_sectors_written) << 9);
 939         sysfs_hprint(metadata_written,
 940                      (atomic_long_read(&ca->meta_sectors_written) +
 941                       atomic_long_read(&ca->btree_sectors_written)) << 9);
 942
 943         sysfs_print(io_errors,
 944                     atomic_read(&ca->io_errors) >> IO_ERROR_SHIFT);
 945
 946         if (attr == &sysfs_cache_replacement_policy)
 947                 return bch_snprint_string_list(buf, PAGE_SIZE,
 948                                                cache_replacement_policies,
 949                                                CACHE_REPLACEMENT(&ca->sb));
 950
 951         if (attr == &sysfs_priority_stats) {
 952                 struct bucket *b;
 953                 size_t n = ca->sb.nbuckets, i;
 954                 size_t unused = 0, available = 0, dirty = 0, meta = 0;
 955                 uint64_t sum = 0;
 956                 /* Compute 31 quantiles */
 957                 uint16_t q[31], *p, *cached;
 958                 ssize_t ret;
 959
 960                 cached = p = vmalloc(array_size(sizeof(uint16_t),
 961                                                 ca->sb.nbuckets));
 962                 if (!p)
 963                         return -ENOMEM;
 964
 965                 mutex_lock(&ca->set->bucket_lock);
 966                 for_each_bucket(b, ca) {
 967                         if (!GC_SECTORS_USED(b))
 968                                 unused++;
 969                         if (GC_MARK(b) == GC_MARK_RECLAIMABLE)
 970                                 available++;
 971                         if (GC_MARK(b) == GC_MARK_DIRTY)
 972                                 dirty++;
 973                         if (GC_MARK(b) == GC_MARK_METADATA)
 974                                 meta++;
 975                 }
 976
 977                 for (i = ca->sb.first_bucket; i < n; i++)
 978                         p[i] = ca->buckets[i].prio;
 979                 mutex_unlock(&ca->set->bucket_lock);
 980
 981                 sort(p, n, sizeof(uint16_t), __bch_cache_cmp, NULL);
 982
 983                 while (n &&
 984                        !cached[n - 1])
 985                         --n;
 986
 987                 unused = ca->sb.nbuckets - n;
 988
 989                 while (cached < p + n &&
 990                        *cached == BTREE_PRIO)
 991                         cached++, n--;
 992
 993                 for (i = 0; i < n; i++)
 994                         sum += INITIAL_PRIO - cached[i];
 995
 996                 if (n)
 997                         do_div(sum, n);
 998
 999                 for (i = 0; i < ARRAY_SIZE(q); i++)
1000                         q[i] = INITIAL_PRIO - cached[n * (i + 1) /
1001                                 (ARRAY_SIZE(q) + 1)];
1002
1003                 vfree(p);
1004
1005                 ret = scnprintf(buf, PAGE_SIZE,
1006                                 "Unused:                %zu%%\n"
1007                                 "Clean:         %zu%%\n"
1008                                 "Dirty:         %zu%%\n"
1009                                 "Metadata:      %zu%%\n"
1010                                 "Average:       %llu\n"
1011                                 "Sectors per Q: %zu\n"
1012                                 "Quantiles:     [",
1013                                 unused * 100 / (size_t) ca->sb.nbuckets,
1014                                 available * 100 / (size_t) ca->sb.nbuckets,
1015                                 dirty * 100 / (size_t) ca->sb.nbuckets,
1016                                 meta * 100 / (size_t) ca->sb.nbuckets, sum,
1017                                 n * ca->sb.bucket_size / (ARRAY_SIZE(q) + 1));
1018
1019                 for (i = 0; i < ARRAY_SIZE(q); i++)
1020                         ret += scnprintf(buf + ret, PAGE_SIZE - ret,
1021                                          "%u ", q[i]);
1022                 ret--;
1023
1024                 ret += scnprintf(buf + ret, PAGE_SIZE - ret, "]\n");
1025
1026                 return ret;
1027         }
1028
1029         return 0;
1030 }
1031 SHOW_LOCKED(bch_cache)
1032
1033 STORE(__bch_cache)
1034 {
1035         struct cache *ca = container_of(kobj, struct cache, kobj);
1036         ssize_t v;
1037
1038         if (attr == &sysfs_discard) {
1039                 bool v = strtoul_or_return(buf);
1040
1041                 if (blk_queue_discard(bdev_get_queue(ca->bdev)))
1042                         ca->discard = v;
1043
1044                 if (v != CACHE_DISCARD(&ca->sb)) {
1045                         SET_CACHE_DISCARD(&ca->sb, v);
1046                         bcache_write_super(ca->set);
1047                 }
1048         }
1049
1050         if (attr == &sysfs_cache_replacement_policy) {
1051                 v = __sysfs_match_string(cache_replacement_policies, -1, buf);
1052                 if (v < 0)
1053                         return v;
1054
1055                 if ((unsigned int) v != CACHE_REPLACEMENT(&ca->sb)) {
1056                         mutex_lock(&ca->set->bucket_lock);
1057                         SET_CACHE_REPLACEMENT(&ca->sb, v);
1058                         mutex_unlock(&ca->set->bucket_lock);
1059
1060                         bcache_write_super(ca->set);
1061                 }
1062         }
1063
1064         if (attr == &sysfs_clear_stats) {
1065                 atomic_long_set(&ca->sectors_written, 0);
1066                 atomic_long_set(&ca->btree_sectors_written, 0);
1067                 atomic_long_set(&ca->meta_sectors_written, 0);
1068                 atomic_set(&ca->io_count, 0);
1069                 atomic_set(&ca->io_errors, 0);
1070         }
1071
1072         return size;
1073 }
1074 STORE_LOCKED(bch_cache)
1075
1076 static struct attribute *bch_cache_files[] = {
1077         &sysfs_bucket_size,
1078         &sysfs_block_size,
1079         &sysfs_nbuckets,
1080         &sysfs_priority_stats,
1081         &sysfs_discard,
1082         &sysfs_written,
1083         &sysfs_btree_written,
1084         &sysfs_metadata_written,
1085         &sysfs_io_errors,
1086         &sysfs_clear_stats,
1087         &sysfs_cache_replacement_policy,
1088         NULL
1089 };
1090 KTYPE(bch_cache);