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 };
  26
  27 /* Default is 0 ("auto") */
  28 static const char * const bch_stop_on_failure_modes[] = {
  29         "auto",
  30         "always"
  31 };
  32
  33 static const char * const cache_replacement_policies[] = {
  34         "lru",
  35         "fifo",
  36         "random"
  37 };
  38
  39 static const char * const error_actions[] = {
  40         "unregister",
  41         "panic"
  42 };
  43
  44 write_attribute(attach);
  45 write_attribute(detach);
  46 write_attribute(unregister);
  47 write_attribute(stop);
  48 write_attribute(clear_stats);
  49 write_attribute(trigger_gc);
  50 write_attribute(prune_cache);
  51 write_attribute(flash_vol_create);
  52
  53 read_attribute(bucket_size);
  54 read_attribute(block_size);
  55 read_attribute(nbuckets);
  56 read_attribute(tree_depth);
  57 read_attribute(root_usage_percent);
  58 read_attribute(priority_stats);
  59 read_attribute(btree_cache_size);
  60 read_attribute(btree_cache_max_chain);
  61 read_attribute(cache_available_percent);
  62 read_attribute(written);
  63 read_attribute(btree_written);
  64 read_attribute(metadata_written);
  65 read_attribute(active_journal_entries);
  66 read_attribute(backing_dev_name);
  67 read_attribute(backing_dev_uuid);
  68
  69 sysfs_time_stats_attribute(btree_gc,    sec, ms);
  70 sysfs_time_stats_attribute(btree_split, sec, us);
  71 sysfs_time_stats_attribute(btree_sort,  ms,  us);
  72 sysfs_time_stats_attribute(btree_read,  ms,  us);
  73
  74 read_attribute(btree_nodes);
  75 read_attribute(btree_used_percent);
  76 read_attribute(average_key_size);
  77 read_attribute(dirty_data);
  78 read_attribute(bset_tree_stats);
  79
  80 read_attribute(state);
  81 read_attribute(cache_read_races);
  82 read_attribute(reclaim);
  83 read_attribute(flush_write);
  84 read_attribute(retry_flush_write);
  85 read_attribute(writeback_keys_done);
  86 read_attribute(writeback_keys_failed);
  87 read_attribute(io_errors);
  88 read_attribute(congested);
  89 read_attribute(cutoff_writeback);
  90 read_attribute(cutoff_writeback_sync);
  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(gc_after_writeback);
 132 rw_attribute(size);
 133
 134 static ssize_t bch_snprint_string_list(char *buf,
 135                                        size_t size,
 136                                        const char * const list[],
 137                                        size_t selected)
 138 {
 139         char *out = buf;
 140         size_t i;
 141
 142         for (i = 0; list[i]; i++)
 143                 out += snprintf(out, buf + size - out,
 144                                 i == selected ? "[%s] " : "%s ", list[i]);
 145
 146         out[-1] = '\n';
 147         return out - buf;
 148 }
 149
 150 SHOW(__bch_cached_dev)
 151 {
 152         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 153                                              disk.kobj);
 154         char const *states[] = { "no cache", "clean", "dirty", "inconsistent" };
 155         int wb = dc->writeback_running;
 156
 157 #define var(stat)               (dc->stat)
 158
 159         if (attr == &sysfs_cache_mode)
 160                 return bch_snprint_string_list(buf, PAGE_SIZE,
 161                                                bch_cache_modes,
 162                                                BDEV_CACHE_MODE(&dc->sb));
 163
 164         if (attr == &sysfs_stop_when_cache_set_failed)
 165                 return bch_snprint_string_list(buf, PAGE_SIZE,
 166                                                bch_stop_on_failure_modes,
 167                                                dc->stop_when_cache_set_failed);
 168
 169
 170         sysfs_printf(data_csum,         "%i", dc->disk.data_csum);
 171         var_printf(verify,              "%i");
 172         var_printf(bypass_torture_test, "%i");
 173         var_printf(writeback_metadata,  "%i");
 174         var_printf(writeback_running,   "%i");
 175         var_print(writeback_delay);
 176         var_print(writeback_percent);
 177         sysfs_hprint(writeback_rate,
 178                      wb ? atomic_long_read(&dc->writeback_rate.rate) << 9 : 0);
 179         sysfs_hprint(io_errors,         atomic_read(&dc->io_errors));
 180         sysfs_printf(io_error_limit,    "%i", dc->error_limit);
 181         sysfs_printf(io_disable,        "%i", dc->io_disable);
 182         var_print(writeback_rate_update_seconds);
 183         var_print(writeback_rate_i_term_inverse);
 184         var_print(writeback_rate_p_term_inverse);
 185         var_print(writeback_rate_minimum);
 186
 187         if (attr == &sysfs_writeback_rate_debug) {
 188                 char rate[20];
 189                 char dirty[20];
 190                 char target[20];
 191                 char proportional[20];
 192                 char integral[20];
 193                 char change[20];
 194                 s64 next_io;
 195
 196                 /*
 197                  * Except for dirty and target, other values should
 198                  * be 0 if writeback is not running.
 199                  */
 200                 bch_hprint(rate,
 201                            wb ? atomic_long_read(&dc->writeback_rate.rate) << 9
 202                               : 0);
 203                 bch_hprint(dirty, bcache_dev_sectors_dirty(&dc->disk) << 9);
 204                 bch_hprint(target, dc->writeback_rate_target << 9);
 205                 bch_hprint(proportional,
 206                            wb ? dc->writeback_rate_proportional << 9 : 0);
 207                 bch_hprint(integral,
 208                            wb ? dc->writeback_rate_integral_scaled << 9 : 0);
 209                 bch_hprint(change, wb ? dc->writeback_rate_change << 9 : 0);
 210                 next_io = wb ? div64_s64(dc->writeback_rate.next-local_clock(),
 211                                          NSEC_PER_MSEC) : 0;
 212
 213                 return sprintf(buf,
 214                                "rate:\t\t%s/sec\n"
 215                                "dirty:\t\t%s\n"
 216                                "target:\t\t%s\n"
 217                                "proportional:\t%s\n"
 218                                "integral:\t%s\n"
 219                                "change:\t\t%s/sec\n"
 220                                "next io:\t%llims\n",
 221                                rate, dirty, target, proportional,
 222                                integral, change, next_io);
 223         }
 224
 225         sysfs_hprint(dirty_data,
 226                      bcache_dev_sectors_dirty(&dc->disk) << 9);
 227
 228         sysfs_hprint(stripe_size,        ((uint64_t)dc->disk.stripe_size) << 9);
 229         var_printf(partial_stripes_expensive,   "%u");
 230
 231         var_hprint(sequential_cutoff);
 232         var_hprint(readahead);
 233
 234         sysfs_print(running,            atomic_read(&dc->running));
 235         sysfs_print(state,              states[BDEV_STATE(&dc->sb)]);
 236
 237         if (attr == &sysfs_label) {
 238                 memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
 239                 buf[SB_LABEL_SIZE + 1] = '\0';
 240                 strcat(buf, "\n");
 241                 return strlen(buf);
 242         }
 243
 244         if (attr == &sysfs_backing_dev_name) {
 245                 snprintf(buf, BDEVNAME_SIZE + 1, "%s", dc->backing_dev_name);
 246                 strcat(buf, "\n");
 247                 return strlen(buf);
 248         }
 249
 250         if (attr == &sysfs_backing_dev_uuid) {
 251                 /* convert binary uuid into 36-byte string plus '\0' */
 252                 snprintf(buf, 36+1, "%pU", dc->sb.uuid);
 253                 strcat(buf, "\n");
 254                 return strlen(buf);
 255         }
 256
 257 #undef var
 258         return 0;
 259 }
 260 SHOW_LOCKED(bch_cached_dev)
 261
 262 STORE(__cached_dev)
 263 {
 264         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 265                                              disk.kobj);
 266         ssize_t v;
 267         struct cache_set *c;
 268         struct kobj_uevent_env *env;
 269
 270 #define d_strtoul(var)          sysfs_strtoul(var, dc->var)
 271 #define d_strtoul_nonzero(var)  sysfs_strtoul_clamp(var, dc->var, 1, INT_MAX)
 272 #define d_strtoi_h(var)         sysfs_hatoi(var, dc->var)
 273
 274         sysfs_strtoul(data_csum,        dc->disk.data_csum);
 275         d_strtoul(verify);
 276         sysfs_strtoul_bool(bypass_torture_test, dc->bypass_torture_test);
 277         sysfs_strtoul_bool(writeback_metadata, dc->writeback_metadata);
 278         sysfs_strtoul_bool(writeback_running, dc->writeback_running);
 279         sysfs_strtoul_clamp(writeback_delay, dc->writeback_delay, 0, UINT_MAX);
 280
 281         sysfs_strtoul_clamp(writeback_percent, dc->writeback_percent,
 282                             0, bch_cutoff_writeback);
 283
 284         if (attr == &sysfs_writeback_rate) {
 285                 ssize_t ret;
 286                 long int v = atomic_long_read(&dc->writeback_rate.rate);
 287
 288                 ret = strtoul_safe_clamp(buf, v, 1, INT_MAX);
 289
 290                 if (!ret) {
 291                         atomic_long_set(&dc->writeback_rate.rate, v);
 292                         ret = size;
 293                 }
 294
 295                 return ret;
 296         }
 297
 298         sysfs_strtoul_clamp(writeback_rate_update_seconds,
 299                             dc->writeback_rate_update_seconds,
 300                             1, WRITEBACK_RATE_UPDATE_SECS_MAX);
 301         sysfs_strtoul_clamp(writeback_rate_i_term_inverse,
 302                             dc->writeback_rate_i_term_inverse,
 303                             1, UINT_MAX);
 304         sysfs_strtoul_clamp(writeback_rate_p_term_inverse,
 305                             dc->writeback_rate_p_term_inverse,
 306                             1, UINT_MAX);
 307         sysfs_strtoul_clamp(writeback_rate_minimum,
 308                             dc->writeback_rate_minimum,
 309                             1, UINT_MAX);
 310
 311         sysfs_strtoul_clamp(io_error_limit, dc->error_limit, 0, INT_MAX);
 312
 313         if (attr == &sysfs_io_disable) {
 314                 int v = strtoul_or_return(buf);
 315
 316                 dc->io_disable = v ? 1 : 0;
 317         }
 318
 319         sysfs_strtoul_clamp(sequential_cutoff,
 320                             dc->sequential_cutoff,
 321                             0, UINT_MAX);
 322         d_strtoi_h(readahead);
 323
 324         if (attr == &sysfs_clear_stats)
 325                 bch_cache_accounting_clear(&dc->accounting);
 326
 327         if (attr == &sysfs_running &&
 328             strtoul_or_return(buf)) {
 329                 v = bch_cached_dev_run(dc);
 330                 if (v)
 331                         return v;
 332         }
 333
 334         if (attr == &sysfs_cache_mode) {
 335                 v = sysfs_match_string(bch_cache_modes, buf);
 336                 if (v < 0)
 337                         return v;
 338
 339                 if ((unsigned int) v != BDEV_CACHE_MODE(&dc->sb)) {
 340                         SET_BDEV_CACHE_MODE(&dc->sb, v);
 341                         bch_write_bdev_super(dc, NULL);
 342                 }
 343         }
 344
 345         if (attr == &sysfs_stop_when_cache_set_failed) {
 346                 v = sysfs_match_string(bch_stop_on_failure_modes, buf);
 347                 if (v < 0)
 348                         return v;
 349
 350                 dc->stop_when_cache_set_failed = v;
 351         }
 352
 353         if (attr == &sysfs_label) {
 354                 if (size > SB_LABEL_SIZE)
 355                         return -EINVAL;
 356                 memcpy(dc->sb.label, buf, size);
 357                 if (size < SB_LABEL_SIZE)
 358                         dc->sb.label[size] = '\0';
 359                 if (size && dc->sb.label[size - 1] == '\n')
 360                         dc->sb.label[size - 1] = '\0';
 361                 bch_write_bdev_super(dc, NULL);
 362                 if (dc->disk.c) {
 363                         memcpy(dc->disk.c->uuids[dc->disk.id].label,
 364                                buf, SB_LABEL_SIZE);
 365                         bch_uuid_write(dc->disk.c);
 366                 }
 367                 env = kzalloc(sizeof(struct kobj_uevent_env), GFP_KERNEL);
 368                 if (!env)
 369                         return -ENOMEM;
 370                 add_uevent_var(env, "DRIVER=bcache");
 371                 add_uevent_var(env, "CACHED_UUID=%pU", dc->sb.uuid),
 372                 add_uevent_var(env, "CACHED_LABEL=%s", buf);
 373                 kobject_uevent_env(&disk_to_dev(dc->disk.disk)->kobj,
 374                                    KOBJ_CHANGE,
 375                                    env->envp);
 376                 kfree(env);
 377         }
 378
 379         if (attr == &sysfs_attach) {
 380                 uint8_t         set_uuid[16];
 381
 382                 if (bch_parse_uuid(buf, set_uuid) < 16)
 383                         return -EINVAL;
 384
 385                 v = -ENOENT;
 386                 list_for_each_entry(c, &bch_cache_sets, list) {
 387                         v = bch_cached_dev_attach(dc, c, set_uuid);
 388                         if (!v)
 389                                 return size;
 390                 }
 391                 if (v == -ENOENT)
 392                         pr_err("Can't attach %s: cache set not found", buf);
 393                 return v;
 394         }
 395
 396         if (attr == &sysfs_detach && dc->disk.c)
 397                 bch_cached_dev_detach(dc);
 398
 399         if (attr == &sysfs_stop)
 400                 bcache_device_stop(&dc->disk);
 401
 402         return size;
 403 }
 404
 405 STORE(bch_cached_dev)
 406 {
 407         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 408                                              disk.kobj);
 409
 410         mutex_lock(&bch_register_lock);
 411         size = __cached_dev_store(kobj, attr, buf, size);
 412
 413         if (attr == &sysfs_writeback_running) {
 414                 /* dc->writeback_running changed in __cached_dev_store() */
 415                 if (IS_ERR_OR_NULL(dc->writeback_thread)) {
 416                         /*
 417                          * reject setting it to 1 via sysfs if writeback
 418                          * kthread is not created yet.
 419                          */
 420                         if (dc->writeback_running) {
 421                                 dc->writeback_running = false;
 422                                 pr_err("%s: failed to run non-existent writeback thread",
 423                                                 dc->disk.disk->disk_name);
 424                         }
 425                 } else
 426                         /*
 427                          * writeback kthread will check if dc->writeback_running
 428                          * is true or false.
 429                          */
 430                         bch_writeback_queue(dc);
 431         }
 432
 433         /*
 434          * Only set BCACHE_DEV_WB_RUNNING when cached device attached to
 435          * a cache set, otherwise it doesn't make sense.
 436          */
 437         if (attr == &sysfs_writeback_percent)
 438                 if ((dc->disk.c != NULL) &&
 439                     (!test_and_set_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags)))
 440                         schedule_delayed_work(&dc->writeback_rate_update,
 441                                       dc->writeback_rate_update_seconds * HZ);
 442
 443         mutex_unlock(&bch_register_lock);
 444         return size;
 445 }
 446
 447 static struct attribute *bch_cached_dev_files[] = {
 448         &sysfs_attach,
 449         &sysfs_detach,
 450         &sysfs_stop,
 451 #if 0
 452         &sysfs_data_csum,
 453 #endif
 454         &sysfs_cache_mode,
 455         &sysfs_stop_when_cache_set_failed,
 456         &sysfs_writeback_metadata,
 457         &sysfs_writeback_running,
 458         &sysfs_writeback_delay,
 459         &sysfs_writeback_percent,
 460         &sysfs_writeback_rate,
 461         &sysfs_writeback_rate_update_seconds,
 462         &sysfs_writeback_rate_i_term_inverse,
 463         &sysfs_writeback_rate_p_term_inverse,
 464         &sysfs_writeback_rate_minimum,
 465         &sysfs_writeback_rate_debug,
 466         &sysfs_errors,
 467         &sysfs_io_error_limit,
 468         &sysfs_io_disable,
 469         &sysfs_dirty_data,
 470         &sysfs_stripe_size,
 471         &sysfs_partial_stripes_expensive,
 472         &sysfs_sequential_cutoff,
 473         &sysfs_clear_stats,
 474         &sysfs_running,
 475         &sysfs_state,
 476         &sysfs_label,
 477         &sysfs_readahead,
 478 #ifdef CONFIG_BCACHE_DEBUG
 479         &sysfs_verify,
 480         &sysfs_bypass_torture_test,
 481 #endif
 482         &sysfs_backing_dev_name,
 483         &sysfs_backing_dev_uuid,
 484         NULL
 485 };
 486 KTYPE(bch_cached_dev);
 487
 488 SHOW(bch_flash_dev)
 489 {
 490         struct bcache_device *d = container_of(kobj, struct bcache_device,
 491                                                kobj);
 492         struct uuid_entry *u = &d->c->uuids[d->id];
 493
 494         sysfs_printf(data_csum, "%i", d->data_csum);
 495         sysfs_hprint(size,      u->sectors << 9);
 496
 497         if (attr == &sysfs_label) {
 498                 memcpy(buf, u->label, SB_LABEL_SIZE);
 499                 buf[SB_LABEL_SIZE + 1] = '\0';
 500                 strcat(buf, "\n");
 501                 return strlen(buf);
 502         }
 503
 504         return 0;
 505 }
 506
 507 STORE(__bch_flash_dev)
 508 {
 509         struct bcache_device *d = container_of(kobj, struct bcache_device,
 510                                                kobj);
 511         struct uuid_entry *u = &d->c->uuids[d->id];
 512
 513         sysfs_strtoul(data_csum,        d->data_csum);
 514
 515         if (attr == &sysfs_size) {
 516                 uint64_t v;
 517
 518                 strtoi_h_or_return(buf, v);
 519
 520                 u->sectors = v >> 9;
 521                 bch_uuid_write(d->c);
 522                 set_capacity(d->disk, u->sectors);
 523         }
 524
 525         if (attr == &sysfs_label) {
 526                 memcpy(u->label, buf, SB_LABEL_SIZE);
 527                 bch_uuid_write(d->c);
 528         }
 529
 530         if (attr == &sysfs_unregister) {
 531                 set_bit(BCACHE_DEV_DETACHING, &d->flags);
 532                 bcache_device_stop(d);
 533         }
 534
 535         return size;
 536 }
 537 STORE_LOCKED(bch_flash_dev)
 538
 539 static struct attribute *bch_flash_dev_files[] = {
 540         &sysfs_unregister,
 541 #if 0
 542         &sysfs_data_csum,
 543 #endif
 544         &sysfs_label,
 545         &sysfs_size,
 546         NULL
 547 };
 548 KTYPE(bch_flash_dev);
 549
 550 struct bset_stats_op {
 551         struct btree_op op;
 552         size_t nodes;
 553         struct bset_stats stats;
 554 };
 555
 556 static int bch_btree_bset_stats(struct btree_op *b_op, struct btree *b)
 557 {
 558         struct bset_stats_op *op = container_of(b_op, struct bset_stats_op, op);
 559
 560         op->nodes++;
 561         bch_btree_keys_stats(&b->keys, &op->stats);
 562
 563         return MAP_CONTINUE;
 564 }
 565
 566 static int bch_bset_print_stats(struct cache_set *c, char *buf)
 567 {
 568         struct bset_stats_op op;
 569         int ret;
 570
 571         memset(&op, 0, sizeof(op));
 572         bch_btree_op_init(&op.op, -1);
 573
 574         ret = bch_btree_map_nodes(&op.op, c, &ZERO_KEY, bch_btree_bset_stats);
 575         if (ret < 0)
 576                 return ret;
 577
 578         return snprintf(buf, PAGE_SIZE,
 579                         "btree nodes:           %zu\n"
 580                         "written sets:          %zu\n"
 581                         "unwritten sets:                %zu\n"
 582                         "written key bytes:     %zu\n"
 583                         "unwritten key bytes:   %zu\n"
 584                         "floats:                        %zu\n"
 585                         "failed:                        %zu\n",
 586                         op.nodes,
 587                         op.stats.sets_written, op.stats.sets_unwritten,
 588                         op.stats.bytes_written, op.stats.bytes_unwritten,
 589                         op.stats.floats, op.stats.failed);
 590 }
 591
 592 static unsigned int bch_root_usage(struct cache_set *c)
 593 {
 594         unsigned int bytes = 0;
 595         struct bkey *k;
 596         struct btree *b;
 597         struct btree_iter iter;
 598
 599         goto lock_root;
 600
 601         do {
 602                 rw_unlock(false, b);
 603 lock_root:
 604                 b = c->root;
 605                 rw_lock(false, b, b->level);
 606         } while (b != c->root);
 607
 608         for_each_key_filter(&b->keys, k, &iter, bch_ptr_bad)
 609                 bytes += bkey_bytes(k);
 610
 611         rw_unlock(false, b);
 612
 613         return (bytes * 100) / btree_bytes(c);
 614 }
 615
 616 static size_t bch_cache_size(struct cache_set *c)
 617 {
 618         size_t ret = 0;
 619         struct btree *b;
 620
 621         mutex_lock(&c->bucket_lock);
 622         list_for_each_entry(b, &c->btree_cache, list)
 623                 ret += 1 << (b->keys.page_order + PAGE_SHIFT);
 624
 625         mutex_unlock(&c->bucket_lock);
 626         return ret;
 627 }
 628
 629 static unsigned int bch_cache_max_chain(struct cache_set *c)
 630 {
 631         unsigned int ret = 0;
 632         struct hlist_head *h;
 633
 634         mutex_lock(&c->bucket_lock);
 635
 636         for (h = c->bucket_hash;
 637              h < c->bucket_hash + (1 << BUCKET_HASH_BITS);
 638              h++) {
 639                 unsigned int i = 0;
 640                 struct hlist_node *p;
 641
 642                 hlist_for_each(p, h)
 643                         i++;
 644
 645                 ret = max(ret, i);
 646         }
 647
 648         mutex_unlock(&c->bucket_lock);
 649         return ret;
 650 }
 651
 652 static unsigned int bch_btree_used(struct cache_set *c)
 653 {
 654         return div64_u64(c->gc_stats.key_bytes * 100,
 655                          (c->gc_stats.nodes ?: 1) * btree_bytes(c));
 656 }
 657
 658 static unsigned int bch_average_key_size(struct cache_set *c)
 659 {
 660         return c->gc_stats.nkeys
 661                 ? div64_u64(c->gc_stats.data, c->gc_stats.nkeys)
 662                 : 0;
 663 }
 664
 665 SHOW(__bch_cache_set)
 666 {
 667         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 668
 669         sysfs_print(synchronous,                CACHE_SYNC(&c->sb));
 670         sysfs_print(journal_delay_ms,           c->journal_delay_ms);
 671         sysfs_hprint(bucket_size,               bucket_bytes(c));
 672         sysfs_hprint(block_size,                block_bytes(c));
 673         sysfs_print(tree_depth,                 c->root->level);
 674         sysfs_print(root_usage_percent,         bch_root_usage(c));
 675
 676         sysfs_hprint(btree_cache_size,          bch_cache_size(c));
 677         sysfs_print(btree_cache_max_chain,      bch_cache_max_chain(c));
 678         sysfs_print(cache_available_percent,    100 - c->gc_stats.in_use);
 679
 680         sysfs_print_time_stats(&c->btree_gc_time,       btree_gc, sec, ms);
 681         sysfs_print_time_stats(&c->btree_split_time,    btree_split, sec, us);
 682         sysfs_print_time_stats(&c->sort.time,           btree_sort, ms, us);
 683         sysfs_print_time_stats(&c->btree_read_time,     btree_read, ms, us);
 684
 685         sysfs_print(btree_used_percent, bch_btree_used(c));
 686         sysfs_print(btree_nodes,        c->gc_stats.nodes);
 687         sysfs_hprint(average_key_size,  bch_average_key_size(c));
 688
 689         sysfs_print(cache_read_races,
 690                     atomic_long_read(&c->cache_read_races));
 691
 692         sysfs_print(reclaim,
 693                     atomic_long_read(&c->reclaim));
 694
 695         sysfs_print(flush_write,
 696                     atomic_long_read(&c->flush_write));
 697
 698         sysfs_print(retry_flush_write,
 699                     atomic_long_read(&c->retry_flush_write));
 700
 701         sysfs_print(writeback_keys_done,
 702                     atomic_long_read(&c->writeback_keys_done));
 703         sysfs_print(writeback_keys_failed,
 704                     atomic_long_read(&c->writeback_keys_failed));
 705
 706         if (attr == &sysfs_errors)
 707                 return bch_snprint_string_list(buf, PAGE_SIZE, error_actions,
 708                                                c->on_error);
 709
 710         /* See count_io_errors for why 88 */
 711         sysfs_print(io_error_halflife,  c->error_decay * 88);
 712         sysfs_print(io_error_limit,     c->error_limit);
 713
 714         sysfs_hprint(congested,
 715                      ((uint64_t) bch_get_congested(c)) << 9);
 716         sysfs_print(congested_read_threshold_us,
 717                     c->congested_read_threshold_us);
 718         sysfs_print(congested_write_threshold_us,
 719                     c->congested_write_threshold_us);
 720
 721         sysfs_print(cutoff_writeback, bch_cutoff_writeback);
 722         sysfs_print(cutoff_writeback_sync, bch_cutoff_writeback_sync);
 723
 724         sysfs_print(active_journal_entries,     fifo_used(&c->journal.pin));
 725         sysfs_printf(verify,                    "%i", c->verify);
 726         sysfs_printf(key_merging_disabled,      "%i", c->key_merging_disabled);
 727         sysfs_printf(expensive_debug_checks,
 728                      "%i", c->expensive_debug_checks);
 729         sysfs_printf(gc_always_rewrite,         "%i", c->gc_always_rewrite);
 730         sysfs_printf(btree_shrinker_disabled,   "%i", c->shrinker_disabled);
 731         sysfs_printf(copy_gc_enabled,           "%i", c->copy_gc_enabled);
 732         sysfs_printf(gc_after_writeback,        "%i", c->gc_after_writeback);
 733         sysfs_printf(io_disable,                "%i",
 734                      test_bit(CACHE_SET_IO_DISABLE, &c->flags));
 735
 736         if (attr == &sysfs_bset_tree_stats)
 737                 return bch_bset_print_stats(c, buf);
 738
 739         return 0;
 740 }
 741 SHOW_LOCKED(bch_cache_set)
 742
 743 STORE(__bch_cache_set)
 744 {
 745         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 746         ssize_t v;
 747
 748         if (attr == &sysfs_unregister)
 749                 bch_cache_set_unregister(c);
 750
 751         if (attr == &sysfs_stop)
 752                 bch_cache_set_stop(c);
 753
 754         if (attr == &sysfs_synchronous) {
 755                 bool sync = strtoul_or_return(buf);
 756
 757                 if (sync != CACHE_SYNC(&c->sb)) {
 758                         SET_CACHE_SYNC(&c->sb, sync);
 759                         bcache_write_super(c);
 760                 }
 761         }
 762
 763         if (attr == &sysfs_flash_vol_create) {
 764                 int r;
 765                 uint64_t v;
 766
 767                 strtoi_h_or_return(buf, v);
 768
 769                 r = bch_flash_dev_create(c, v);
 770                 if (r)
 771                         return r;
 772         }
 773
 774         if (attr == &sysfs_clear_stats) {
 775                 atomic_long_set(&c->writeback_keys_done,        0);
 776                 atomic_long_set(&c->writeback_keys_failed,      0);
 777
 778                 memset(&c->gc_stats, 0, sizeof(struct gc_stat));
 779                 bch_cache_accounting_clear(&c->accounting);
 780         }
 781
 782         if (attr == &sysfs_trigger_gc)
 783                 force_wake_up_gc(c);
 784
 785         if (attr == &sysfs_prune_cache) {
 786                 struct shrink_control sc;
 787
 788                 sc.gfp_mask = GFP_KERNEL;
 789                 sc.nr_to_scan = strtoul_or_return(buf);
 790                 c->shrink.scan_objects(&c->shrink, &sc);
 791         }
 792
 793         sysfs_strtoul_clamp(congested_read_threshold_us,
 794                             c->congested_read_threshold_us,
 795                             0, UINT_MAX);
 796         sysfs_strtoul_clamp(congested_write_threshold_us,
 797                             c->congested_write_threshold_us,
 798                             0, UINT_MAX);
 799
 800         if (attr == &sysfs_errors) {
 801                 v = sysfs_match_string(error_actions, buf);
 802                 if (v < 0)
 803                         return v;
 804
 805                 c->on_error = v;
 806         }
 807
 808         sysfs_strtoul_clamp(io_error_limit, c->error_limit, 0, UINT_MAX);
 809
 810         /* See count_io_errors() for why 88 */
 811         if (attr == &sysfs_io_error_halflife) {
 812                 unsigned long v = 0;
 813                 ssize_t ret;
 814
 815                 ret = strtoul_safe_clamp(buf, v, 0, UINT_MAX);
 816                 if (!ret) {
 817                         c->error_decay = v / 88;
 818                         return size;
 819                 }
 820                 return ret;
 821         }
 822
 823         if (attr == &sysfs_io_disable) {
 824                 v = strtoul_or_return(buf);
 825                 if (v) {
 826                         if (test_and_set_bit(CACHE_SET_IO_DISABLE,
 827                                              &c->flags))
 828                                 pr_warn("CACHE_SET_IO_DISABLE already set");
 829                 } else {
 830                         if (!test_and_clear_bit(CACHE_SET_IO_DISABLE,
 831                                                 &c->flags))
 832                                 pr_warn("CACHE_SET_IO_DISABLE already cleared");
 833                 }
 834         }
 835
 836         sysfs_strtoul_clamp(journal_delay_ms,
 837                             c->journal_delay_ms,
 838                             0, USHRT_MAX);
 839         sysfs_strtoul_bool(verify,              c->verify);
 840         sysfs_strtoul_bool(key_merging_disabled, c->key_merging_disabled);
 841         sysfs_strtoul(expensive_debug_checks,   c->expensive_debug_checks);
 842         sysfs_strtoul_bool(gc_always_rewrite,   c->gc_always_rewrite);
 843         sysfs_strtoul_bool(btree_shrinker_disabled, c->shrinker_disabled);
 844         sysfs_strtoul_bool(copy_gc_enabled,     c->copy_gc_enabled);
 845         /*
 846          * write gc_after_writeback here may overwrite an already set
 847          * BCH_DO_AUTO_GC, it doesn't matter because this flag will be
 848          * set in next chance.
 849          */
 850         sysfs_strtoul_clamp(gc_after_writeback, c->gc_after_writeback, 0, 1);
 851
 852         return size;
 853 }
 854 STORE_LOCKED(bch_cache_set)
 855
 856 SHOW(bch_cache_set_internal)
 857 {
 858         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 859
 860         return bch_cache_set_show(&c->kobj, attr, buf);
 861 }
 862
 863 STORE(bch_cache_set_internal)
 864 {
 865         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 866
 867         return bch_cache_set_store(&c->kobj, attr, buf, size);
 868 }
 869
 870 static void bch_cache_set_internal_release(struct kobject *k)
 871 {
 872 }
 873
 874 static struct attribute *bch_cache_set_files[] = {
 875         &sysfs_unregister,
 876         &sysfs_stop,
 877         &sysfs_synchronous,
 878         &sysfs_journal_delay_ms,
 879         &sysfs_flash_vol_create,
 880
 881         &sysfs_bucket_size,
 882         &sysfs_block_size,
 883         &sysfs_tree_depth,
 884         &sysfs_root_usage_percent,
 885         &sysfs_btree_cache_size,
 886         &sysfs_cache_available_percent,
 887
 888         &sysfs_average_key_size,
 889
 890         &sysfs_errors,
 891         &sysfs_io_error_limit,
 892         &sysfs_io_error_halflife,
 893         &sysfs_congested,
 894         &sysfs_congested_read_threshold_us,
 895         &sysfs_congested_write_threshold_us,
 896         &sysfs_clear_stats,
 897         NULL
 898 };
 899 KTYPE(bch_cache_set);
 900
 901 static struct attribute *bch_cache_set_internal_files[] = {
 902         &sysfs_active_journal_entries,
 903
 904         sysfs_time_stats_attribute_list(btree_gc, sec, ms)
 905         sysfs_time_stats_attribute_list(btree_split, sec, us)
 906         sysfs_time_stats_attribute_list(btree_sort, ms, us)
 907         sysfs_time_stats_attribute_list(btree_read, ms, us)
 908
 909         &sysfs_btree_nodes,
 910         &sysfs_btree_used_percent,
 911         &sysfs_btree_cache_max_chain,
 912
 913         &sysfs_bset_tree_stats,
 914         &sysfs_cache_read_races,
 915         &sysfs_reclaim,
 916         &sysfs_flush_write,
 917         &sysfs_retry_flush_write,
 918         &sysfs_writeback_keys_done,
 919         &sysfs_writeback_keys_failed,
 920
 921         &sysfs_trigger_gc,
 922         &sysfs_prune_cache,
 923 #ifdef CONFIG_BCACHE_DEBUG
 924         &sysfs_verify,
 925         &sysfs_key_merging_disabled,
 926         &sysfs_expensive_debug_checks,
 927 #endif
 928         &sysfs_gc_always_rewrite,
 929         &sysfs_btree_shrinker_disabled,
 930         &sysfs_copy_gc_enabled,
 931         &sysfs_gc_after_writeback,
 932         &sysfs_io_disable,
 933         &sysfs_cutoff_writeback,
 934         &sysfs_cutoff_writeback_sync,
 935         NULL
 936 };
 937 KTYPE(bch_cache_set_internal);
 938
 939 static int __bch_cache_cmp(const void *l, const void *r)
 940 {
 941         return *((uint16_t *)r) - *((uint16_t *)l);
 942 }
 943
 944 SHOW(__bch_cache)
 945 {
 946         struct cache *ca = container_of(kobj, struct cache, kobj);
 947
 948         sysfs_hprint(bucket_size,       bucket_bytes(ca));
 949         sysfs_hprint(block_size,        block_bytes(ca));
 950         sysfs_print(nbuckets,           ca->sb.nbuckets);
 951         sysfs_print(discard,            ca->discard);
 952         sysfs_hprint(written, atomic_long_read(&ca->sectors_written) << 9);
 953         sysfs_hprint(btree_written,
 954                      atomic_long_read(&ca->btree_sectors_written) << 9);
 955         sysfs_hprint(metadata_written,
 956                      (atomic_long_read(&ca->meta_sectors_written) +
 957                       atomic_long_read(&ca->btree_sectors_written)) << 9);
 958
 959         sysfs_print(io_errors,
 960                     atomic_read(&ca->io_errors) >> IO_ERROR_SHIFT);
 961
 962         if (attr == &sysfs_cache_replacement_policy)
 963                 return bch_snprint_string_list(buf, PAGE_SIZE,
 964                                                cache_replacement_policies,
 965                                                CACHE_REPLACEMENT(&ca->sb));
 966
 967         if (attr == &sysfs_priority_stats) {
 968                 struct bucket *b;
 969                 size_t n = ca->sb.nbuckets, i;
 970                 size_t unused = 0, available = 0, dirty = 0, meta = 0;
 971                 uint64_t sum = 0;
 972                 /* Compute 31 quantiles */
 973                 uint16_t q[31], *p, *cached;
 974                 ssize_t ret;
 975
 976                 cached = p = vmalloc(array_size(sizeof(uint16_t),
 977                                                 ca->sb.nbuckets));
 978                 if (!p)
 979                         return -ENOMEM;
 980
 981                 mutex_lock(&ca->set->bucket_lock);
 982                 for_each_bucket(b, ca) {
 983                         if (!GC_SECTORS_USED(b))
 984                                 unused++;
 985                         if (GC_MARK(b) == GC_MARK_RECLAIMABLE)
 986                                 available++;
 987                         if (GC_MARK(b) == GC_MARK_DIRTY)
 988                                 dirty++;
 989                         if (GC_MARK(b) == GC_MARK_METADATA)
 990                                 meta++;
 991                 }
 992
 993                 for (i = ca->sb.first_bucket; i < n; i++)
 994                         p[i] = ca->buckets[i].prio;
 995                 mutex_unlock(&ca->set->bucket_lock);
 996
 997                 sort(p, n, sizeof(uint16_t), __bch_cache_cmp, NULL);
 998
 999                 while (n &&
1000                        !cached[n - 1])
1001                         --n;
1002
1003                 while (cached < p + n &&
1004                        *cached == BTREE_PRIO)
1005                         cached++, n--;
1006
1007                 for (i = 0; i < n; i++)
1008                         sum += INITIAL_PRIO - cached[i];
1009
1010                 if (n)
1011                         do_div(sum, n);
1012
1013                 for (i = 0; i < ARRAY_SIZE(q); i++)
1014                         q[i] = INITIAL_PRIO - cached[n * (i + 1) /
1015                                 (ARRAY_SIZE(q) + 1)];
1016
1017                 vfree(p);
1018
1019                 ret = scnprintf(buf, PAGE_SIZE,
1020                                 "Unused:                %zu%%\n"
1021                                 "Clean:         %zu%%\n"
1022                                 "Dirty:         %zu%%\n"
1023                                 "Metadata:      %zu%%\n"
1024                                 "Average:       %llu\n"
1025                                 "Sectors per Q: %zu\n"
1026                                 "Quantiles:     [",
1027                                 unused * 100 / (size_t) ca->sb.nbuckets,
1028                                 available * 100 / (size_t) ca->sb.nbuckets,
1029                                 dirty * 100 / (size_t) ca->sb.nbuckets,
1030                                 meta * 100 / (size_t) ca->sb.nbuckets, sum,
1031                                 n * ca->sb.bucket_size / (ARRAY_SIZE(q) + 1));
1032
1033                 for (i = 0; i < ARRAY_SIZE(q); i++)
1034                         ret += scnprintf(buf + ret, PAGE_SIZE - ret,
1035                                          "%u ", q[i]);
1036                 ret--;
1037
1038                 ret += scnprintf(buf + ret, PAGE_SIZE - ret, "]\n");
1039
1040                 return ret;
1041         }
1042
1043         return 0;
1044 }
1045 SHOW_LOCKED(bch_cache)
1046
1047 STORE(__bch_cache)
1048 {
1049         struct cache *ca = container_of(kobj, struct cache, kobj);
1050         ssize_t v;
1051
1052         if (attr == &sysfs_discard) {
1053                 bool v = strtoul_or_return(buf);
1054
1055                 if (blk_queue_discard(bdev_get_queue(ca->bdev)))
1056                         ca->discard = v;
1057
1058                 if (v != CACHE_DISCARD(&ca->sb)) {
1059                         SET_CACHE_DISCARD(&ca->sb, v);
1060                         bcache_write_super(ca->set);
1061                 }
1062         }
1063
1064         if (attr == &sysfs_cache_replacement_policy) {
1065                 v = sysfs_match_string(cache_replacement_policies, buf);
1066                 if (v < 0)
1067                         return v;
1068
1069                 if ((unsigned int) v != CACHE_REPLACEMENT(&ca->sb)) {
1070                         mutex_lock(&ca->set->bucket_lock);
1071                         SET_CACHE_REPLACEMENT(&ca->sb, v);
1072                         mutex_unlock(&ca->set->bucket_lock);
1073
1074                         bcache_write_super(ca->set);
1075                 }
1076         }
1077
1078         if (attr == &sysfs_clear_stats) {
1079                 atomic_long_set(&ca->sectors_written, 0);
1080                 atomic_long_set(&ca->btree_sectors_written, 0);
1081                 atomic_long_set(&ca->meta_sectors_written, 0);
1082                 atomic_set(&ca->io_count, 0);
1083                 atomic_set(&ca->io_errors, 0);
1084         }
1085
1086         return size;
1087 }
1088 STORE_LOCKED(bch_cache)
1089
1090 static struct attribute *bch_cache_files[] = {
1091         &sysfs_bucket_size,
1092         &sysfs_block_size,
1093         &sysfs_nbuckets,
1094         &sysfs_priority_stats,
1095         &sysfs_discard,
1096         &sysfs_written,
1097         &sysfs_btree_written,
1098         &sysfs_metadata_written,
1099         &sysfs_io_errors,
1100         &sysfs_clear_stats,
1101         &sysfs_cache_replacement_policy,
1102         NULL
1103 };
1104 KTYPE(bch_cache);