linux/drivers/md/bcache/sysfs.c
<<
>>
Prefs
   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
  19static const char * const cache_replacement_policies[] = {
  20        "lru",
  21        "fifo",
  22        "random",
  23        NULL
  24};
  25
  26static const char * const error_actions[] = {
  27        "unregister",
  28        "panic",
  29        NULL
  30};
  31
  32write_attribute(attach);
  33write_attribute(detach);
  34write_attribute(unregister);
  35write_attribute(stop);
  36write_attribute(clear_stats);
  37write_attribute(trigger_gc);
  38write_attribute(prune_cache);
  39write_attribute(flash_vol_create);
  40
  41read_attribute(bucket_size);
  42read_attribute(block_size);
  43read_attribute(nbuckets);
  44read_attribute(tree_depth);
  45read_attribute(root_usage_percent);
  46read_attribute(priority_stats);
  47read_attribute(btree_cache_size);
  48read_attribute(btree_cache_max_chain);
  49read_attribute(cache_available_percent);
  50read_attribute(written);
  51read_attribute(btree_written);
  52read_attribute(metadata_written);
  53read_attribute(active_journal_entries);
  54
  55sysfs_time_stats_attribute(btree_gc,    sec, ms);
  56sysfs_time_stats_attribute(btree_split, sec, us);
  57sysfs_time_stats_attribute(btree_sort,  ms,  us);
  58sysfs_time_stats_attribute(btree_read,  ms,  us);
  59
  60read_attribute(btree_nodes);
  61read_attribute(btree_used_percent);
  62read_attribute(average_key_size);
  63read_attribute(dirty_data);
  64read_attribute(bset_tree_stats);
  65
  66read_attribute(state);
  67read_attribute(cache_read_races);
  68read_attribute(writeback_keys_done);
  69read_attribute(writeback_keys_failed);
  70read_attribute(io_errors);
  71read_attribute(congested);
  72rw_attribute(congested_read_threshold_us);
  73rw_attribute(congested_write_threshold_us);
  74
  75rw_attribute(sequential_cutoff);
  76rw_attribute(data_csum);
  77rw_attribute(cache_mode);
  78rw_attribute(writeback_metadata);
  79rw_attribute(writeback_running);
  80rw_attribute(writeback_percent);
  81rw_attribute(writeback_delay);
  82rw_attribute(writeback_rate);
  83
  84rw_attribute(writeback_rate_update_seconds);
  85rw_attribute(writeback_rate_d_term);
  86rw_attribute(writeback_rate_p_term_inverse);
  87read_attribute(writeback_rate_debug);
  88
  89read_attribute(stripe_size);
  90read_attribute(partial_stripes_expensive);
  91
  92rw_attribute(synchronous);
  93rw_attribute(journal_delay_ms);
  94rw_attribute(discard);
  95rw_attribute(running);
  96rw_attribute(label);
  97rw_attribute(readahead);
  98rw_attribute(errors);
  99rw_attribute(io_error_limit);
 100rw_attribute(io_error_halflife);
 101rw_attribute(verify);
 102rw_attribute(bypass_torture_test);
 103rw_attribute(key_merging_disabled);
 104rw_attribute(gc_always_rewrite);
 105rw_attribute(expensive_debug_checks);
 106rw_attribute(cache_replacement_policy);
 107rw_attribute(btree_shrinker_disabled);
 108rw_attribute(copy_gc_enabled);
 109rw_attribute(size);
 110
 111SHOW(__bch_cached_dev)
 112{
 113        struct cached_dev *dc = container_of(kobj, struct cached_dev,
 114                                             disk.kobj);
 115        const char *states[] = { "no cache", "clean", "dirty", "inconsistent" };
 116
 117#define var(stat)               (dc->stat)
 118
 119        if (attr == &sysfs_cache_mode)
 120                return bch_snprint_string_list(buf, PAGE_SIZE,
 121                                               bch_cache_modes + 1,
 122                                               BDEV_CACHE_MODE(&dc->sb));
 123
 124        sysfs_printf(data_csum,         "%i", dc->disk.data_csum);
 125        var_printf(verify,              "%i");
 126        var_printf(bypass_torture_test, "%i");
 127        var_printf(writeback_metadata,  "%i");
 128        var_printf(writeback_running,   "%i");
 129        var_print(writeback_delay);
 130        var_print(writeback_percent);
 131        sysfs_hprint(writeback_rate,    dc->writeback_rate.rate << 9);
 132
 133        var_print(writeback_rate_update_seconds);
 134        var_print(writeback_rate_d_term);
 135        var_print(writeback_rate_p_term_inverse);
 136
 137        if (attr == &sysfs_writeback_rate_debug) {
 138                char rate[20];
 139                char dirty[20];
 140                char target[20];
 141                char proportional[20];
 142                char derivative[20];
 143                char change[20];
 144                s64 next_io;
 145
 146                bch_hprint(rate,        dc->writeback_rate.rate << 9);
 147                bch_hprint(dirty,       bcache_dev_sectors_dirty(&dc->disk) << 9);
 148                bch_hprint(target,      dc->writeback_rate_target << 9);
 149                bch_hprint(proportional,dc->writeback_rate_proportional << 9);
 150                bch_hprint(derivative,  dc->writeback_rate_derivative << 9);
 151                bch_hprint(change,      dc->writeback_rate_change << 9);
 152
 153                next_io = div64_s64(dc->writeback_rate.next - local_clock(),
 154                                    NSEC_PER_MSEC);
 155
 156                return sprintf(buf,
 157                               "rate:\t\t%s/sec\n"
 158                               "dirty:\t\t%s\n"
 159                               "target:\t\t%s\n"
 160                               "proportional:\t%s\n"
 161                               "derivative:\t%s\n"
 162                               "change:\t\t%s/sec\n"
 163                               "next io:\t%llims\n",
 164                               rate, dirty, target, proportional,
 165                               derivative, change, next_io);
 166        }
 167
 168        sysfs_hprint(dirty_data,
 169                     bcache_dev_sectors_dirty(&dc->disk) << 9);
 170
 171        sysfs_hprint(stripe_size,       dc->disk.stripe_size << 9);
 172        var_printf(partial_stripes_expensive,   "%u");
 173
 174        var_hprint(sequential_cutoff);
 175        var_hprint(readahead);
 176
 177        sysfs_print(running,            atomic_read(&dc->running));
 178        sysfs_print(state,              states[BDEV_STATE(&dc->sb)]);
 179
 180        if (attr == &sysfs_label) {
 181                memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
 182                buf[SB_LABEL_SIZE + 1] = '\0';
 183                strcat(buf, "\n");
 184                return strlen(buf);
 185        }
 186
 187#undef var
 188        return 0;
 189}
 190SHOW_LOCKED(bch_cached_dev)
 191
 192STORE(__cached_dev)
 193{
 194        struct cached_dev *dc = container_of(kobj, struct cached_dev,
 195                                             disk.kobj);
 196        ssize_t v = size;
 197        struct cache_set *c;
 198        struct kobj_uevent_env *env;
 199
 200#define d_strtoul(var)          sysfs_strtoul(var, dc->var)
 201#define d_strtoul_nonzero(var)  sysfs_strtoul_clamp(var, dc->var, 1, INT_MAX)
 202#define d_strtoi_h(var)         sysfs_hatoi(var, dc->var)
 203
 204        sysfs_strtoul(data_csum,        dc->disk.data_csum);
 205        d_strtoul(verify);
 206        d_strtoul(bypass_torture_test);
 207        d_strtoul(writeback_metadata);
 208        d_strtoul(writeback_running);
 209        d_strtoul(writeback_delay);
 210
 211        sysfs_strtoul_clamp(writeback_percent, dc->writeback_percent, 0, 40);
 212
 213        sysfs_strtoul_clamp(writeback_rate,
 214                            dc->writeback_rate.rate, 1, INT_MAX);
 215
 216        d_strtoul_nonzero(writeback_rate_update_seconds);
 217        d_strtoul(writeback_rate_d_term);
 218        d_strtoul_nonzero(writeback_rate_p_term_inverse);
 219
 220        d_strtoi_h(sequential_cutoff);
 221        d_strtoi_h(readahead);
 222
 223        if (attr == &sysfs_clear_stats)
 224                bch_cache_accounting_clear(&dc->accounting);
 225
 226        if (attr == &sysfs_running &&
 227            strtoul_or_return(buf))
 228                bch_cached_dev_run(dc);
 229
 230        if (attr == &sysfs_cache_mode) {
 231                v = bch_read_string_list(buf, bch_cache_modes + 1);
 232
 233                if (v < 0)
 234                        return v;
 235
 236                if ((unsigned) v != BDEV_CACHE_MODE(&dc->sb)) {
 237                        SET_BDEV_CACHE_MODE(&dc->sb, v);
 238                        bch_write_bdev_super(dc, NULL);
 239                }
 240        }
 241
 242        if (attr == &sysfs_label) {
 243                if (size > SB_LABEL_SIZE)
 244                        return -EINVAL;
 245                memcpy(dc->sb.label, buf, size);
 246                if (size < SB_LABEL_SIZE)
 247                        dc->sb.label[size] = '\0';
 248                if (size && dc->sb.label[size - 1] == '\n')
 249                        dc->sb.label[size - 1] = '\0';
 250                bch_write_bdev_super(dc, NULL);
 251                if (dc->disk.c) {
 252                        memcpy(dc->disk.c->uuids[dc->disk.id].label,
 253                               buf, SB_LABEL_SIZE);
 254                        bch_uuid_write(dc->disk.c);
 255                }
 256                env = kzalloc(sizeof(struct kobj_uevent_env), GFP_KERNEL);
 257                if (!env)
 258                        return -ENOMEM;
 259                add_uevent_var(env, "DRIVER=bcache");
 260                add_uevent_var(env, "CACHED_UUID=%pU", dc->sb.uuid),
 261                add_uevent_var(env, "CACHED_LABEL=%s", buf);
 262                kobject_uevent_env(
 263                        &disk_to_dev(dc->disk.disk)->kobj, KOBJ_CHANGE, env->envp);
 264                kfree(env);
 265        }
 266
 267        if (attr == &sysfs_attach) {
 268                if (bch_parse_uuid(buf, dc->sb.set_uuid) < 16)
 269                        return -EINVAL;
 270
 271                list_for_each_entry(c, &bch_cache_sets, list) {
 272                        v = bch_cached_dev_attach(dc, c);
 273                        if (!v)
 274                                return size;
 275                }
 276
 277                pr_err("Can't attach %s: cache set not found", buf);
 278                size = v;
 279        }
 280
 281        if (attr == &sysfs_detach && dc->disk.c)
 282                bch_cached_dev_detach(dc);
 283
 284        if (attr == &sysfs_stop)
 285                bcache_device_stop(&dc->disk);
 286
 287        return size;
 288}
 289
 290STORE(bch_cached_dev)
 291{
 292        struct cached_dev *dc = container_of(kobj, struct cached_dev,
 293                                             disk.kobj);
 294
 295        mutex_lock(&bch_register_lock);
 296        size = __cached_dev_store(kobj, attr, buf, size);
 297
 298        if (attr == &sysfs_writeback_running)
 299                bch_writeback_queue(dc);
 300
 301        if (attr == &sysfs_writeback_percent)
 302                schedule_delayed_work(&dc->writeback_rate_update,
 303                                      dc->writeback_rate_update_seconds * HZ);
 304
 305        mutex_unlock(&bch_register_lock);
 306        return size;
 307}
 308
 309static struct attribute *bch_cached_dev_files[] = {
 310        &sysfs_attach,
 311        &sysfs_detach,
 312        &sysfs_stop,
 313#if 0
 314        &sysfs_data_csum,
 315#endif
 316        &sysfs_cache_mode,
 317        &sysfs_writeback_metadata,
 318        &sysfs_writeback_running,
 319        &sysfs_writeback_delay,
 320        &sysfs_writeback_percent,
 321        &sysfs_writeback_rate,
 322        &sysfs_writeback_rate_update_seconds,
 323        &sysfs_writeback_rate_d_term,
 324        &sysfs_writeback_rate_p_term_inverse,
 325        &sysfs_writeback_rate_debug,
 326        &sysfs_dirty_data,
 327        &sysfs_stripe_size,
 328        &sysfs_partial_stripes_expensive,
 329        &sysfs_sequential_cutoff,
 330        &sysfs_clear_stats,
 331        &sysfs_running,
 332        &sysfs_state,
 333        &sysfs_label,
 334        &sysfs_readahead,
 335#ifdef CONFIG_BCACHE_DEBUG
 336        &sysfs_verify,
 337        &sysfs_bypass_torture_test,
 338#endif
 339        NULL
 340};
 341KTYPE(bch_cached_dev);
 342
 343SHOW(bch_flash_dev)
 344{
 345        struct bcache_device *d = container_of(kobj, struct bcache_device,
 346                                               kobj);
 347        struct uuid_entry *u = &d->c->uuids[d->id];
 348
 349        sysfs_printf(data_csum, "%i", d->data_csum);
 350        sysfs_hprint(size,      u->sectors << 9);
 351
 352        if (attr == &sysfs_label) {
 353                memcpy(buf, u->label, SB_LABEL_SIZE);
 354                buf[SB_LABEL_SIZE + 1] = '\0';
 355                strcat(buf, "\n");
 356                return strlen(buf);
 357        }
 358
 359        return 0;
 360}
 361
 362STORE(__bch_flash_dev)
 363{
 364        struct bcache_device *d = container_of(kobj, struct bcache_device,
 365                                               kobj);
 366        struct uuid_entry *u = &d->c->uuids[d->id];
 367
 368        sysfs_strtoul(data_csum,        d->data_csum);
 369
 370        if (attr == &sysfs_size) {
 371                uint64_t v;
 372                strtoi_h_or_return(buf, v);
 373
 374                u->sectors = v >> 9;
 375                bch_uuid_write(d->c);
 376                set_capacity(d->disk, u->sectors);
 377        }
 378
 379        if (attr == &sysfs_label) {
 380                memcpy(u->label, buf, SB_LABEL_SIZE);
 381                bch_uuid_write(d->c);
 382        }
 383
 384        if (attr == &sysfs_unregister) {
 385                set_bit(BCACHE_DEV_DETACHING, &d->flags);
 386                bcache_device_stop(d);
 387        }
 388
 389        return size;
 390}
 391STORE_LOCKED(bch_flash_dev)
 392
 393static struct attribute *bch_flash_dev_files[] = {
 394        &sysfs_unregister,
 395#if 0
 396        &sysfs_data_csum,
 397#endif
 398        &sysfs_label,
 399        &sysfs_size,
 400        NULL
 401};
 402KTYPE(bch_flash_dev);
 403
 404struct bset_stats_op {
 405        struct btree_op op;
 406        size_t nodes;
 407        struct bset_stats stats;
 408};
 409
 410static int bch_btree_bset_stats(struct btree_op *b_op, struct btree *b)
 411{
 412        struct bset_stats_op *op = container_of(b_op, struct bset_stats_op, op);
 413
 414        op->nodes++;
 415        bch_btree_keys_stats(&b->keys, &op->stats);
 416
 417        return MAP_CONTINUE;
 418}
 419
 420static int bch_bset_print_stats(struct cache_set *c, char *buf)
 421{
 422        struct bset_stats_op op;
 423        int ret;
 424
 425        memset(&op, 0, sizeof(op));
 426        bch_btree_op_init(&op.op, -1);
 427
 428        ret = bch_btree_map_nodes(&op.op, c, &ZERO_KEY, bch_btree_bset_stats);
 429        if (ret < 0)
 430                return ret;
 431
 432        return snprintf(buf, PAGE_SIZE,
 433                        "btree nodes:           %zu\n"
 434                        "written sets:          %zu\n"
 435                        "unwritten sets:                %zu\n"
 436                        "written key bytes:     %zu\n"
 437                        "unwritten key bytes:   %zu\n"
 438                        "floats:                        %zu\n"
 439                        "failed:                        %zu\n",
 440                        op.nodes,
 441                        op.stats.sets_written, op.stats.sets_unwritten,
 442                        op.stats.bytes_written, op.stats.bytes_unwritten,
 443                        op.stats.floats, op.stats.failed);
 444}
 445
 446static unsigned bch_root_usage(struct cache_set *c)
 447{
 448        unsigned bytes = 0;
 449        struct bkey *k;
 450        struct btree *b;
 451        struct btree_iter iter;
 452
 453        goto lock_root;
 454
 455        do {
 456                rw_unlock(false, b);
 457lock_root:
 458                b = c->root;
 459                rw_lock(false, b, b->level);
 460        } while (b != c->root);
 461
 462        for_each_key_filter(&b->keys, k, &iter, bch_ptr_bad)
 463                bytes += bkey_bytes(k);
 464
 465        rw_unlock(false, b);
 466
 467        return (bytes * 100) / btree_bytes(c);
 468}
 469
 470static size_t bch_cache_size(struct cache_set *c)
 471{
 472        size_t ret = 0;
 473        struct btree *b;
 474
 475        mutex_lock(&c->bucket_lock);
 476        list_for_each_entry(b, &c->btree_cache, list)
 477                ret += 1 << (b->keys.page_order + PAGE_SHIFT);
 478
 479        mutex_unlock(&c->bucket_lock);
 480        return ret;
 481}
 482
 483static unsigned bch_cache_max_chain(struct cache_set *c)
 484{
 485        unsigned ret = 0;
 486        struct hlist_head *h;
 487
 488        mutex_lock(&c->bucket_lock);
 489
 490        for (h = c->bucket_hash;
 491             h < c->bucket_hash + (1 << BUCKET_HASH_BITS);
 492             h++) {
 493                unsigned i = 0;
 494                struct hlist_node *p;
 495
 496                hlist_for_each(p, h)
 497                        i++;
 498
 499                ret = max(ret, i);
 500        }
 501
 502        mutex_unlock(&c->bucket_lock);
 503        return ret;
 504}
 505
 506static unsigned bch_btree_used(struct cache_set *c)
 507{
 508        return div64_u64(c->gc_stats.key_bytes * 100,
 509                         (c->gc_stats.nodes ?: 1) * btree_bytes(c));
 510}
 511
 512static unsigned bch_average_key_size(struct cache_set *c)
 513{
 514        return c->gc_stats.nkeys
 515                ? div64_u64(c->gc_stats.data, c->gc_stats.nkeys)
 516                : 0;
 517}
 518
 519SHOW(__bch_cache_set)
 520{
 521        struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 522
 523        sysfs_print(synchronous,                CACHE_SYNC(&c->sb));
 524        sysfs_print(journal_delay_ms,           c->journal_delay_ms);
 525        sysfs_hprint(bucket_size,               bucket_bytes(c));
 526        sysfs_hprint(block_size,                block_bytes(c));
 527        sysfs_print(tree_depth,                 c->root->level);
 528        sysfs_print(root_usage_percent,         bch_root_usage(c));
 529
 530        sysfs_hprint(btree_cache_size,          bch_cache_size(c));
 531        sysfs_print(btree_cache_max_chain,      bch_cache_max_chain(c));
 532        sysfs_print(cache_available_percent,    100 - c->gc_stats.in_use);
 533
 534        sysfs_print_time_stats(&c->btree_gc_time,       btree_gc, sec, ms);
 535        sysfs_print_time_stats(&c->btree_split_time,    btree_split, sec, us);
 536        sysfs_print_time_stats(&c->sort.time,           btree_sort, ms, us);
 537        sysfs_print_time_stats(&c->btree_read_time,     btree_read, ms, us);
 538
 539        sysfs_print(btree_used_percent, bch_btree_used(c));
 540        sysfs_print(btree_nodes,        c->gc_stats.nodes);
 541        sysfs_hprint(average_key_size,  bch_average_key_size(c));
 542
 543        sysfs_print(cache_read_races,
 544                    atomic_long_read(&c->cache_read_races));
 545
 546        sysfs_print(writeback_keys_done,
 547                    atomic_long_read(&c->writeback_keys_done));
 548        sysfs_print(writeback_keys_failed,
 549                    atomic_long_read(&c->writeback_keys_failed));
 550
 551        if (attr == &sysfs_errors)
 552                return bch_snprint_string_list(buf, PAGE_SIZE, error_actions,
 553                                               c->on_error);
 554
 555        /* See count_io_errors for why 88 */
 556        sysfs_print(io_error_halflife,  c->error_decay * 88);
 557        sysfs_print(io_error_limit,     c->error_limit >> IO_ERROR_SHIFT);
 558
 559        sysfs_hprint(congested,
 560                     ((uint64_t) bch_get_congested(c)) << 9);
 561        sysfs_print(congested_read_threshold_us,
 562                    c->congested_read_threshold_us);
 563        sysfs_print(congested_write_threshold_us,
 564                    c->congested_write_threshold_us);
 565
 566        sysfs_print(active_journal_entries,     fifo_used(&c->journal.pin));
 567        sysfs_printf(verify,                    "%i", c->verify);
 568        sysfs_printf(key_merging_disabled,      "%i", c->key_merging_disabled);
 569        sysfs_printf(expensive_debug_checks,
 570                     "%i", c->expensive_debug_checks);
 571        sysfs_printf(gc_always_rewrite,         "%i", c->gc_always_rewrite);
 572        sysfs_printf(btree_shrinker_disabled,   "%i", c->shrinker_disabled);
 573        sysfs_printf(copy_gc_enabled,           "%i", c->copy_gc_enabled);
 574
 575        if (attr == &sysfs_bset_tree_stats)
 576                return bch_bset_print_stats(c, buf);
 577
 578        return 0;
 579}
 580SHOW_LOCKED(bch_cache_set)
 581
 582STORE(__bch_cache_set)
 583{
 584        struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 585
 586        if (attr == &sysfs_unregister)
 587                bch_cache_set_unregister(c);
 588
 589        if (attr == &sysfs_stop)
 590                bch_cache_set_stop(c);
 591
 592        if (attr == &sysfs_synchronous) {
 593                bool sync = strtoul_or_return(buf);
 594
 595                if (sync != CACHE_SYNC(&c->sb)) {
 596                        SET_CACHE_SYNC(&c->sb, sync);
 597                        bcache_write_super(c);
 598                }
 599        }
 600
 601        if (attr == &sysfs_flash_vol_create) {
 602                int r;
 603                uint64_t v;
 604                strtoi_h_or_return(buf, v);
 605
 606                r = bch_flash_dev_create(c, v);
 607                if (r)
 608                        return r;
 609        }
 610
 611        if (attr == &sysfs_clear_stats) {
 612                atomic_long_set(&c->writeback_keys_done,        0);
 613                atomic_long_set(&c->writeback_keys_failed,      0);
 614
 615                memset(&c->gc_stats, 0, sizeof(struct gc_stat));
 616                bch_cache_accounting_clear(&c->accounting);
 617        }
 618
 619        if (attr == &sysfs_trigger_gc) {
 620                /*
 621                 * Garbage collection thread only works when sectors_to_gc < 0,
 622                 * when users write to sysfs entry trigger_gc, most of time
 623                 * they want to forcibly triger gargage collection. Here -1 is
 624                 * set to c->sectors_to_gc, to make gc_should_run() give a
 625                 * chance to permit gc thread to run. "give a chance" means
 626                 * before going into gc_should_run(), there is still chance
 627                 * that c->sectors_to_gc being set to other positive value. So
 628                 * writing sysfs entry trigger_gc won't always make sure gc
 629                 * thread takes effect.
 630                 */
 631                atomic_set(&c->sectors_to_gc, -1);
 632                wake_up_gc(c);
 633        }
 634
 635        if (attr == &sysfs_prune_cache) {
 636                struct shrink_control sc;
 637                sc.gfp_mask = GFP_KERNEL;
 638                sc.nr_to_scan = strtoul_or_return(buf);
 639                c->shrink.scan_objects(&c->shrink, &sc);
 640        }
 641
 642        sysfs_strtoul(congested_read_threshold_us,
 643                      c->congested_read_threshold_us);
 644        sysfs_strtoul(congested_write_threshold_us,
 645                      c->congested_write_threshold_us);
 646
 647        if (attr == &sysfs_errors) {
 648                ssize_t v = bch_read_string_list(buf, error_actions);
 649
 650                if (v < 0)
 651                        return v;
 652
 653                c->on_error = v;
 654        }
 655
 656        if (attr == &sysfs_io_error_limit)
 657                c->error_limit = strtoul_or_return(buf) << IO_ERROR_SHIFT;
 658
 659        /* See count_io_errors() for why 88 */
 660        if (attr == &sysfs_io_error_halflife)
 661                c->error_decay = strtoul_or_return(buf) / 88;
 662
 663        sysfs_strtoul(journal_delay_ms,         c->journal_delay_ms);
 664        sysfs_strtoul(verify,                   c->verify);
 665        sysfs_strtoul(key_merging_disabled,     c->key_merging_disabled);
 666        sysfs_strtoul(expensive_debug_checks,   c->expensive_debug_checks);
 667        sysfs_strtoul(gc_always_rewrite,        c->gc_always_rewrite);
 668        sysfs_strtoul(btree_shrinker_disabled,  c->shrinker_disabled);
 669        sysfs_strtoul(copy_gc_enabled,          c->copy_gc_enabled);
 670
 671        return size;
 672}
 673STORE_LOCKED(bch_cache_set)
 674
 675SHOW(bch_cache_set_internal)
 676{
 677        struct cache_set *c = container_of(kobj, struct cache_set, internal);
 678        return bch_cache_set_show(&c->kobj, attr, buf);
 679}
 680
 681STORE(bch_cache_set_internal)
 682{
 683        struct cache_set *c = container_of(kobj, struct cache_set, internal);
 684        return bch_cache_set_store(&c->kobj, attr, buf, size);
 685}
 686
 687static void bch_cache_set_internal_release(struct kobject *k)
 688{
 689}
 690
 691static struct attribute *bch_cache_set_files[] = {
 692        &sysfs_unregister,
 693        &sysfs_stop,
 694        &sysfs_synchronous,
 695        &sysfs_journal_delay_ms,
 696        &sysfs_flash_vol_create,
 697
 698        &sysfs_bucket_size,
 699        &sysfs_block_size,
 700        &sysfs_tree_depth,
 701        &sysfs_root_usage_percent,
 702        &sysfs_btree_cache_size,
 703        &sysfs_cache_available_percent,
 704
 705        &sysfs_average_key_size,
 706
 707        &sysfs_errors,
 708        &sysfs_io_error_limit,
 709        &sysfs_io_error_halflife,
 710        &sysfs_congested,
 711        &sysfs_congested_read_threshold_us,
 712        &sysfs_congested_write_threshold_us,
 713        &sysfs_clear_stats,
 714        NULL
 715};
 716KTYPE(bch_cache_set);
 717
 718static struct attribute *bch_cache_set_internal_files[] = {
 719        &sysfs_active_journal_entries,
 720
 721        sysfs_time_stats_attribute_list(btree_gc, sec, ms)
 722        sysfs_time_stats_attribute_list(btree_split, sec, us)
 723        sysfs_time_stats_attribute_list(btree_sort, ms, us)
 724        sysfs_time_stats_attribute_list(btree_read, ms, us)
 725
 726        &sysfs_btree_nodes,
 727        &sysfs_btree_used_percent,
 728        &sysfs_btree_cache_max_chain,
 729
 730        &sysfs_bset_tree_stats,
 731        &sysfs_cache_read_races,
 732        &sysfs_writeback_keys_done,
 733        &sysfs_writeback_keys_failed,
 734
 735        &sysfs_trigger_gc,
 736        &sysfs_prune_cache,
 737#ifdef CONFIG_BCACHE_DEBUG
 738        &sysfs_verify,
 739        &sysfs_key_merging_disabled,
 740        &sysfs_expensive_debug_checks,
 741#endif
 742        &sysfs_gc_always_rewrite,
 743        &sysfs_btree_shrinker_disabled,
 744        &sysfs_copy_gc_enabled,
 745        NULL
 746};
 747KTYPE(bch_cache_set_internal);
 748
 749SHOW(__bch_cache)
 750{
 751        struct cache *ca = container_of(kobj, struct cache, kobj);
 752
 753        sysfs_hprint(bucket_size,       bucket_bytes(ca));
 754        sysfs_hprint(block_size,        block_bytes(ca));
 755        sysfs_print(nbuckets,           ca->sb.nbuckets);
 756        sysfs_print(discard,            ca->discard);
 757        sysfs_hprint(written, atomic_long_read(&ca->sectors_written) << 9);
 758        sysfs_hprint(btree_written,
 759                     atomic_long_read(&ca->btree_sectors_written) << 9);
 760        sysfs_hprint(metadata_written,
 761                     (atomic_long_read(&ca->meta_sectors_written) +
 762                      atomic_long_read(&ca->btree_sectors_written)) << 9);
 763
 764        sysfs_print(io_errors,
 765                    atomic_read(&ca->io_errors) >> IO_ERROR_SHIFT);
 766
 767        if (attr == &sysfs_cache_replacement_policy)
 768                return bch_snprint_string_list(buf, PAGE_SIZE,
 769                                               cache_replacement_policies,
 770                                               CACHE_REPLACEMENT(&ca->sb));
 771
 772        if (attr == &sysfs_priority_stats) {
 773                int cmp(const void *l, const void *r)
 774                {       return *((uint16_t *) r) - *((uint16_t *) l); }
 775
 776                struct bucket *b;
 777                size_t n = ca->sb.nbuckets, i;
 778                size_t unused = 0, available = 0, dirty = 0, meta = 0;
 779                uint64_t sum = 0;
 780                /* Compute 31 quantiles */
 781                uint16_t q[31], *p, *cached;
 782                ssize_t ret;
 783
 784                cached = p = vmalloc(ca->sb.nbuckets * sizeof(uint16_t));
 785                if (!p)
 786                        return -ENOMEM;
 787
 788                mutex_lock(&ca->set->bucket_lock);
 789                for_each_bucket(b, ca) {
 790                        if (!GC_SECTORS_USED(b))
 791                                unused++;
 792                        if (GC_MARK(b) == GC_MARK_RECLAIMABLE)
 793                                available++;
 794                        if (GC_MARK(b) == GC_MARK_DIRTY)
 795                                dirty++;
 796                        if (GC_MARK(b) == GC_MARK_METADATA)
 797                                meta++;
 798                }
 799
 800                for (i = ca->sb.first_bucket; i < n; i++)
 801                        p[i] = ca->buckets[i].prio;
 802                mutex_unlock(&ca->set->bucket_lock);
 803
 804                sort(p, n, sizeof(uint16_t), cmp, NULL);
 805
 806                while (n &&
 807                       !cached[n - 1])
 808                        --n;
 809
 810                unused = ca->sb.nbuckets - n;
 811
 812                while (cached < p + n &&
 813                       *cached == BTREE_PRIO)
 814                        cached++, n--;
 815
 816                for (i = 0; i < n; i++)
 817                        sum += INITIAL_PRIO - cached[i];
 818
 819                if (n)
 820                        do_div(sum, n);
 821
 822                for (i = 0; i < ARRAY_SIZE(q); i++)
 823                        q[i] = INITIAL_PRIO - cached[n * (i + 1) /
 824                                (ARRAY_SIZE(q) + 1)];
 825
 826                vfree(p);
 827
 828                ret = scnprintf(buf, PAGE_SIZE,
 829                                "Unused:                %zu%%\n"
 830                                "Clean:         %zu%%\n"
 831                                "Dirty:         %zu%%\n"
 832                                "Metadata:      %zu%%\n"
 833                                "Average:       %llu\n"
 834                                "Sectors per Q: %zu\n"
 835                                "Quantiles:     [",
 836                                unused * 100 / (size_t) ca->sb.nbuckets,
 837                                available * 100 / (size_t) ca->sb.nbuckets,
 838                                dirty * 100 / (size_t) ca->sb.nbuckets,
 839                                meta * 100 / (size_t) ca->sb.nbuckets, sum,
 840                                n * ca->sb.bucket_size / (ARRAY_SIZE(q) + 1));
 841
 842                for (i = 0; i < ARRAY_SIZE(q); i++)
 843                        ret += scnprintf(buf + ret, PAGE_SIZE - ret,
 844                                         "%u ", q[i]);
 845                ret--;
 846
 847                ret += scnprintf(buf + ret, PAGE_SIZE - ret, "]\n");
 848
 849                return ret;
 850        }
 851
 852        return 0;
 853}
 854SHOW_LOCKED(bch_cache)
 855
 856STORE(__bch_cache)
 857{
 858        struct cache *ca = container_of(kobj, struct cache, kobj);
 859
 860        if (attr == &sysfs_discard) {
 861                bool v = strtoul_or_return(buf);
 862
 863                if (blk_queue_discard(bdev_get_queue(ca->bdev)))
 864                        ca->discard = v;
 865
 866                if (v != CACHE_DISCARD(&ca->sb)) {
 867                        SET_CACHE_DISCARD(&ca->sb, v);
 868                        bcache_write_super(ca->set);
 869                }
 870        }
 871
 872        if (attr == &sysfs_cache_replacement_policy) {
 873                ssize_t v = bch_read_string_list(buf, cache_replacement_policies);
 874
 875                if (v < 0)
 876                        return v;
 877
 878                if ((unsigned) v != CACHE_REPLACEMENT(&ca->sb)) {
 879                        mutex_lock(&ca->set->bucket_lock);
 880                        SET_CACHE_REPLACEMENT(&ca->sb, v);
 881                        mutex_unlock(&ca->set->bucket_lock);
 882
 883                        bcache_write_super(ca->set);
 884                }
 885        }
 886
 887        if (attr == &sysfs_clear_stats) {
 888                atomic_long_set(&ca->sectors_written, 0);
 889                atomic_long_set(&ca->btree_sectors_written, 0);
 890                atomic_long_set(&ca->meta_sectors_written, 0);
 891                atomic_set(&ca->io_count, 0);
 892                atomic_set(&ca->io_errors, 0);
 893        }
 894
 895        return size;
 896}
 897STORE_LOCKED(bch_cache)
 898
 899static struct attribute *bch_cache_files[] = {
 900        &sysfs_bucket_size,
 901        &sysfs_block_size,
 902        &sysfs_nbuckets,
 903        &sysfs_priority_stats,
 904        &sysfs_discard,
 905        &sysfs_written,
 906        &sysfs_btree_written,
 907        &sysfs_metadata_written,
 908        &sysfs_io_errors,
 909        &sysfs_clear_stats,
 910        &sysfs_cache_replacement_policy,
 911        NULL
 912};
 913KTYPE(bch_cache);
 914