drivers/md/bcache/sysfs.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * bcache sysfs interfaces
   4  *
   5  * Copyright 2010, 2011 Kent Overstreet <[email protected]>
   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 #include "features.h"
  15
  16 #include <linux/blkdev.h>
  17 #include <linux/sort.h>
  18 #include <linux/sched/clock.h>
  19
  20 extern bool bcache_is_reboot;
  21
  22 /* Default is 0 ("writethrough") */
  23 static const char * const bch_cache_modes[] = {
  24         "writethrough",
  25         "writeback",
  26         "writearound",
  27         "none",
  28         NULL
  29 };
  30
  31 static const char * const bch_reada_cache_policies[] = {
  32         "all",
  33         "meta-only",
  34         NULL
  35 };
  36
  37 /* Default is 0 ("auto") */
  38 static const char * const bch_stop_on_failure_modes[] = {
  39         "auto",
  40         "always",
  41         NULL
  42 };
  43
  44 static const char * const cache_replacement_policies[] = {
  45         "lru",
  46         "fifo",
  47         "random",
  48         NULL
  49 };
  50
  51 static const char * const error_actions[] = {
  52         "unregister",
  53         "panic",
  54         NULL
  55 };
  56
  57 write_attribute(attach);
  58 write_attribute(detach);
  59 write_attribute(unregister);
  60 write_attribute(stop);
  61 write_attribute(clear_stats);
  62 write_attribute(trigger_gc);
  63 write_attribute(prune_cache);
  64 write_attribute(flash_vol_create);
  65
  66 read_attribute(bucket_size);
  67 read_attribute(block_size);
  68 read_attribute(nbuckets);
  69 read_attribute(tree_depth);
  70 read_attribute(root_usage_percent);
  71 read_attribute(priority_stats);
  72 read_attribute(btree_cache_size);
  73 read_attribute(btree_cache_max_chain);
  74 read_attribute(cache_available_percent);
  75 read_attribute(written);
  76 read_attribute(btree_written);
  77 read_attribute(metadata_written);
  78 read_attribute(active_journal_entries);
  79 read_attribute(backing_dev_name);
  80 read_attribute(backing_dev_uuid);
  81
  82 sysfs_time_stats_attribute(btree_gc,    sec, ms);
  83 sysfs_time_stats_attribute(btree_split, sec, us);
  84 sysfs_time_stats_attribute(btree_sort,  ms,  us);
  85 sysfs_time_stats_attribute(btree_read,  ms,  us);
  86
  87 read_attribute(btree_nodes);
  88 read_attribute(btree_used_percent);
  89 read_attribute(average_key_size);
  90 read_attribute(dirty_data);
  91 read_attribute(bset_tree_stats);
  92 read_attribute(feature_compat);
  93 read_attribute(feature_ro_compat);
  94 read_attribute(feature_incompat);
  95
  96 read_attribute(state);
  97 read_attribute(cache_read_races);
  98 read_attribute(reclaim);
  99 read_attribute(reclaimed_journal_buckets);
 100 read_attribute(flush_write);
 101 read_attribute(writeback_keys_done);
 102 read_attribute(writeback_keys_failed);
 103 read_attribute(io_errors);
 104 read_attribute(congested);
 105 read_attribute(cutoff_writeback);
 106 read_attribute(cutoff_writeback_sync);
 107 rw_attribute(congested_read_threshold_us);
 108 rw_attribute(congested_write_threshold_us);
 109
 110 rw_attribute(sequential_cutoff);
 111 rw_attribute(data_csum);
 112 rw_attribute(cache_mode);
 113 rw_attribute(readahead_cache_policy);
 114 rw_attribute(stop_when_cache_set_failed);
 115 rw_attribute(writeback_metadata);
 116 rw_attribute(writeback_running);
 117 rw_attribute(writeback_percent);
 118 rw_attribute(writeback_delay);
 119 rw_attribute(writeback_rate);
 120 rw_attribute(writeback_consider_fragment);
 121
 122 rw_attribute(writeback_rate_update_seconds);
 123 rw_attribute(writeback_rate_i_term_inverse);
 124 rw_attribute(writeback_rate_p_term_inverse);
 125 rw_attribute(writeback_rate_fp_term_low);
 126 rw_attribute(writeback_rate_fp_term_mid);
 127 rw_attribute(writeback_rate_fp_term_high);
 128 rw_attribute(writeback_rate_minimum);
 129 read_attribute(writeback_rate_debug);
 130
 131 read_attribute(stripe_size);
 132 read_attribute(partial_stripes_expensive);
 133
 134 rw_attribute(synchronous);
 135 rw_attribute(journal_delay_ms);
 136 rw_attribute(io_disable);
 137 rw_attribute(discard);
 138 rw_attribute(running);
 139 rw_attribute(label);
 140 rw_attribute(errors);
 141 rw_attribute(io_error_limit);
 142 rw_attribute(io_error_halflife);
 143 rw_attribute(verify);
 144 rw_attribute(bypass_torture_test);
 145 rw_attribute(key_merging_disabled);
 146 rw_attribute(gc_always_rewrite);
 147 rw_attribute(expensive_debug_checks);
 148 rw_attribute(cache_replacement_policy);
 149 rw_attribute(btree_shrinker_disabled);
 150 rw_attribute(copy_gc_enabled);
 151 rw_attribute(idle_max_writeback_rate);
 152 rw_attribute(gc_after_writeback);
 153 rw_attribute(size);
 154
 155 static ssize_t bch_snprint_string_list(char *buf,
 156                                        size_t size,
 157                                        const char * const list[],
 158                                        size_t selected)
 159 {
 160         char *out = buf;
 161         size_t i;
 162
 163         for (i = 0; list[i]; i++)
 164                 out += scnprintf(out, buf + size - out,
 165                                 i == selected ? "[%s] " : "%s ", list[i]);
 166
 167         out[-1] = '\n';
 168         return out - buf;
 169 }
 170
 171 SHOW(__bch_cached_dev)
 172 {
 173         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 174                                              disk.kobj);
 175         char const *states[] = { "no cache", "clean", "dirty", "inconsistent" };
 176         int wb = dc->writeback_running;
 177
 178 #define var(stat)               (dc->stat)
 179
 180         if (attr == &sysfs_cache_mode)
 181                 return bch_snprint_string_list(buf, PAGE_SIZE,
 182                                                bch_cache_modes,
 183                                                BDEV_CACHE_MODE(&dc->sb));
 184
 185         if (attr == &sysfs_readahead_cache_policy)
 186                 return bch_snprint_string_list(buf, PAGE_SIZE,
 187                                               bch_reada_cache_policies,
 188                                               dc->cache_readahead_policy);
 189
 190         if (attr == &sysfs_stop_when_cache_set_failed)
 191                 return bch_snprint_string_list(buf, PAGE_SIZE,
 192                                                bch_stop_on_failure_modes,
 193                                                dc->stop_when_cache_set_failed);
 194
 195
 196         sysfs_printf(data_csum,         "%i", dc->disk.data_csum);
 197         var_printf(verify,              "%i");
 198         var_printf(bypass_torture_test, "%i");
 199         var_printf(writeback_metadata,  "%i");
 200         var_printf(writeback_running,   "%i");
 201         var_printf(writeback_consider_fragment, "%i");
 202         var_print(writeback_delay);
 203         var_print(writeback_percent);
 204         sysfs_hprint(writeback_rate,
 205                      wb ? atomic_long_read(&dc->writeback_rate.rate) << 9 : 0);
 206         sysfs_printf(io_errors,         "%i", atomic_read(&dc->io_errors));
 207         sysfs_printf(io_error_limit,    "%i", dc->error_limit);
 208         sysfs_printf(io_disable,        "%i", dc->io_disable);
 209         var_print(writeback_rate_update_seconds);
 210         var_print(writeback_rate_i_term_inverse);
 211         var_print(writeback_rate_p_term_inverse);
 212         var_print(writeback_rate_fp_term_low);
 213         var_print(writeback_rate_fp_term_mid);
 214         var_print(writeback_rate_fp_term_high);
 215         var_print(writeback_rate_minimum);
 216
 217         if (attr == &sysfs_writeback_rate_debug) {
 218                 char rate[20];
 219                 char dirty[20];
 220                 char target[20];
 221                 char proportional[20];
 222                 char integral[20];
 223                 char change[20];
 224                 s64 next_io;
 225
 226                 /*
 227                  * Except for dirty and target, other values should
 228                  * be 0 if writeback is not running.
 229                  */
 230                 bch_hprint(rate,
 231                            wb ? atomic_long_read(&dc->writeback_rate.rate) << 9
 232                               : 0);
 233                 bch_hprint(dirty, bcache_dev_sectors_dirty(&dc->disk) << 9);
 234                 bch_hprint(target, dc->writeback_rate_target << 9);
 235                 bch_hprint(proportional,
 236                            wb ? dc->writeback_rate_proportional << 9 : 0);
 237                 bch_hprint(integral,
 238                            wb ? dc->writeback_rate_integral_scaled << 9 : 0);
 239                 bch_hprint(change, wb ? dc->writeback_rate_change << 9 : 0);
 240                 next_io = wb ? div64_s64(dc->writeback_rate.next-local_clock(),
 241                                          NSEC_PER_MSEC) : 0;
 242
 243                 return sprintf(buf,
 244                                "rate:\t\t%s/sec\n"
 245                                "dirty:\t\t%s\n"
 246                                "target:\t\t%s\n"
 247                                "proportional:\t%s\n"
 248                                "integral:\t%s\n"
 249                                "change:\t\t%s/sec\n"
 250                                "next io:\t%llims\n",
 251                                rate, dirty, target, proportional,
 252                                integral, change, next_io);
 253         }
 254
 255         sysfs_hprint(dirty_data,
 256                      bcache_dev_sectors_dirty(&dc->disk) << 9);
 257
 258         sysfs_hprint(stripe_size,        ((uint64_t)dc->disk.stripe_size) << 9);
 259         var_printf(partial_stripes_expensive,   "%u");
 260
 261         var_hprint(sequential_cutoff);
 262
 263         sysfs_print(running,            atomic_read(&dc->running));
 264         sysfs_print(state,              states[BDEV_STATE(&dc->sb)]);
 265
 266         if (attr == &sysfs_label) {
 267                 memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
 268                 buf[SB_LABEL_SIZE + 1] = '\0';
 269                 strcat(buf, "\n");
 270                 return strlen(buf);
 271         }
 272
 273         if (attr == &sysfs_backing_dev_name) {
 274                 snprintf(buf, BDEVNAME_SIZE + 1, "%pg", dc->bdev);
 275                 strcat(buf, "\n");
 276                 return strlen(buf);
 277         }
 278
 279         if (attr == &sysfs_backing_dev_uuid) {
 280                 /* convert binary uuid into 36-byte string plus '\0' */
 281                 snprintf(buf, 36+1, "%pU", dc->sb.uuid);
 282                 strcat(buf, "\n");
 283                 return strlen(buf);
 284         }
 285
 286 #undef var
 287         return 0;
 288 }
 289 SHOW_LOCKED(bch_cached_dev)
 290
 291 STORE(__cached_dev)
 292 {
 293         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 294                                              disk.kobj);
 295         ssize_t v;
 296         struct cache_set *c;
 297         struct kobj_uevent_env *env;
 298
 299         /* no user space access if system is rebooting */
 300         if (bcache_is_reboot)
 301                 return -EBUSY;
 302
 303 #define d_strtoul(var)          sysfs_strtoul(var, dc->var)
 304 #define d_strtoul_nonzero(var)  sysfs_strtoul_clamp(var, dc->var, 1, INT_MAX)
 305 #define d_strtoi_h(var)         sysfs_hatoi(var, dc->var)
 306
 307         sysfs_strtoul(data_csum,        dc->disk.data_csum);
 308         d_strtoul(verify);
 309         sysfs_strtoul_bool(bypass_torture_test, dc->bypass_torture_test);
 310         sysfs_strtoul_bool(writeback_metadata, dc->writeback_metadata);
 311         sysfs_strtoul_bool(writeback_running, dc->writeback_running);
 312         sysfs_strtoul_bool(writeback_consider_fragment, dc->writeback_consider_fragment);
 313         sysfs_strtoul_clamp(writeback_delay, dc->writeback_delay, 0, UINT_MAX);
 314
 315         sysfs_strtoul_clamp(writeback_percent, dc->writeback_percent,
 316                             0, bch_cutoff_writeback);
 317
 318         if (attr == &sysfs_writeback_rate) {
 319                 ssize_t ret;
 320                 long int v = atomic_long_read(&dc->writeback_rate.rate);
 321
 322                 ret = strtoul_safe_clamp(buf, v, 1, INT_MAX);
 323
 324                 if (!ret) {
 325                         atomic_long_set(&dc->writeback_rate.rate, v);
 326                         ret = size;
 327                 }
 328
 329                 return ret;
 330         }
 331
 332         sysfs_strtoul_clamp(writeback_rate_update_seconds,
 333                             dc->writeback_rate_update_seconds,
 334                             1, WRITEBACK_RATE_UPDATE_SECS_MAX);
 335         sysfs_strtoul_clamp(writeback_rate_i_term_inverse,
 336                             dc->writeback_rate_i_term_inverse,
 337                             1, UINT_MAX);
 338         sysfs_strtoul_clamp(writeback_rate_p_term_inverse,
 339                             dc->writeback_rate_p_term_inverse,
 340                             1, UINT_MAX);
 341         sysfs_strtoul_clamp(writeback_rate_fp_term_low,
 342                             dc->writeback_rate_fp_term_low,
 343                             1, dc->writeback_rate_fp_term_mid - 1);
 344         sysfs_strtoul_clamp(writeback_rate_fp_term_mid,
 345                             dc->writeback_rate_fp_term_mid,
 346                             dc->writeback_rate_fp_term_low + 1,
 347                             dc->writeback_rate_fp_term_high - 1);
 348         sysfs_strtoul_clamp(writeback_rate_fp_term_high,
 349                             dc->writeback_rate_fp_term_high,
 350                             dc->writeback_rate_fp_term_mid + 1, UINT_MAX);
 351         sysfs_strtoul_clamp(writeback_rate_minimum,
 352                             dc->writeback_rate_minimum,
 353                             1, UINT_MAX);
 354
 355         sysfs_strtoul_clamp(io_error_limit, dc->error_limit, 0, INT_MAX);
 356
 357         if (attr == &sysfs_io_disable) {
 358                 int v = strtoul_or_return(buf);
 359
 360                 dc->io_disable = v ? 1 : 0;
 361         }
 362
 363         sysfs_strtoul_clamp(sequential_cutoff,
 364                             dc->sequential_cutoff,
 365                             0, UINT_MAX);
 366
 367         if (attr == &sysfs_clear_stats)
 368                 bch_cache_accounting_clear(&dc->accounting);
 369
 370         if (attr == &sysfs_running &&
 371             strtoul_or_return(buf)) {
 372                 v = bch_cached_dev_run(dc);
 373                 if (v)
 374                         return v;
 375         }
 376
 377         if (attr == &sysfs_cache_mode) {
 378                 v = __sysfs_match_string(bch_cache_modes, -1, buf);
 379                 if (v < 0)
 380                         return v;
 381
 382                 if ((unsigned int) v != BDEV_CACHE_MODE(&dc->sb)) {
 383                         SET_BDEV_CACHE_MODE(&dc->sb, v);
 384                         bch_write_bdev_super(dc, NULL);
 385                 }
 386         }
 387
 388         if (attr == &sysfs_readahead_cache_policy) {
 389                 v = __sysfs_match_string(bch_reada_cache_policies, -1, buf);
 390                 if (v < 0)
 391                         return v;
 392
 393                 if ((unsigned int) v != dc->cache_readahead_policy)
 394                         dc->cache_readahead_policy = v;
 395         }
 396
 397         if (attr == &sysfs_stop_when_cache_set_failed) {
 398                 v = __sysfs_match_string(bch_stop_on_failure_modes, -1, buf);
 399                 if (v < 0)
 400                         return v;
 401
 402                 dc->stop_when_cache_set_failed = v;
 403         }
 404
 405         if (attr == &sysfs_label) {
 406                 if (size > SB_LABEL_SIZE)
 407                         return -EINVAL;
 408                 memcpy(dc->sb.label, buf, size);
 409                 if (size < SB_LABEL_SIZE)
 410                         dc->sb.label[size] = '\0';
 411                 if (size && dc->sb.label[size - 1] == '\n')
 412                         dc->sb.label[size - 1] = '\0';
 413                 bch_write_bdev_super(dc, NULL);
 414                 if (dc->disk.c) {
 415                         memcpy(dc->disk.c->uuids[dc->disk.id].label,
 416                                buf, SB_LABEL_SIZE);
 417                         bch_uuid_write(dc->disk.c);
 418                 }
 419                 env = kzalloc(sizeof(struct kobj_uevent_env), GFP_KERNEL);
 420                 if (!env)
 421                         return -ENOMEM;
 422                 add_uevent_var(env, "DRIVER=bcache");
 423                 add_uevent_var(env, "CACHED_UUID=%pU", dc->sb.uuid);
 424                 add_uevent_var(env, "CACHED_LABEL=%s", buf);
 425                 kobject_uevent_env(&disk_to_dev(dc->disk.disk)->kobj,
 426                                    KOBJ_CHANGE,
 427                                    env->envp);
 428                 kfree(env);
 429         }
 430
 431         if (attr == &sysfs_attach) {
 432                 uint8_t         set_uuid[16];
 433
 434                 if (bch_parse_uuid(buf, set_uuid) < 16)
 435                         return -EINVAL;
 436
 437                 v = -ENOENT;
 438                 list_for_each_entry(c, &bch_cache_sets, list) {
 439                         v = bch_cached_dev_attach(dc, c, set_uuid);
 440                         if (!v)
 441                                 return size;
 442                 }
 443                 if (v == -ENOENT)
 444                         pr_err("Can't attach %s: cache set not found\n", buf);
 445                 return v;
 446         }
 447
 448         if (attr == &sysfs_detach && dc->disk.c)
 449                 bch_cached_dev_detach(dc);
 450
 451         if (attr == &sysfs_stop)
 452                 bcache_device_stop(&dc->disk);
 453
 454         return size;
 455 }
 456
 457 STORE(bch_cached_dev)
 458 {
 459         struct cached_dev *dc = container_of(kobj, struct cached_dev,
 460                                              disk.kobj);
 461
 462         /* no user space access if system is rebooting */
 463         if (bcache_is_reboot)
 464                 return -EBUSY;
 465
 466         mutex_lock(&bch_register_lock);
 467         size = __cached_dev_store(kobj, attr, buf, size);
 468
 469         if (attr == &sysfs_writeback_running) {
 470                 /* dc->writeback_running changed in __cached_dev_store() */
 471                 if (IS_ERR_OR_NULL(dc->writeback_thread)) {
 472                         /*
 473                          * reject setting it to 1 via sysfs if writeback
 474                          * kthread is not created yet.
 475                          */
 476                         if (dc->writeback_running) {
 477                                 dc->writeback_running = false;
 478                                 pr_err("%s: failed to run non-existent writeback thread\n",
 479                                                 dc->disk.disk->disk_name);
 480                         }
 481                 } else
 482                         /*
 483                          * writeback kthread will check if dc->writeback_running
 484                          * is true or false.
 485                          */
 486                         bch_writeback_queue(dc);
 487         }
 488
 489         /*
 490          * Only set BCACHE_DEV_WB_RUNNING when cached device attached to
 491          * a cache set, otherwise it doesn't make sense.
 492          */
 493         if (attr == &sysfs_writeback_percent)
 494                 if ((dc->disk.c != NULL) &&
 495                     (!test_and_set_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags)))
 496                         schedule_delayed_work(&dc->writeback_rate_update,
 497                                       dc->writeback_rate_update_seconds * HZ);
 498
 499         mutex_unlock(&bch_register_lock);
 500         return size;
 501 }
 502
 503 static struct attribute *bch_cached_dev_attrs[] = {
 504         &sysfs_attach,
 505         &sysfs_detach,
 506         &sysfs_stop,
 507 #if 0
 508         &sysfs_data_csum,
 509 #endif
 510         &sysfs_cache_mode,
 511         &sysfs_readahead_cache_policy,
 512         &sysfs_stop_when_cache_set_failed,
 513         &sysfs_writeback_metadata,
 514         &sysfs_writeback_running,
 515         &sysfs_writeback_delay,
 516         &sysfs_writeback_percent,
 517         &sysfs_writeback_rate,
 518         &sysfs_writeback_consider_fragment,
 519         &sysfs_writeback_rate_update_seconds,
 520         &sysfs_writeback_rate_i_term_inverse,
 521         &sysfs_writeback_rate_p_term_inverse,
 522         &sysfs_writeback_rate_fp_term_low,
 523         &sysfs_writeback_rate_fp_term_mid,
 524         &sysfs_writeback_rate_fp_term_high,
 525         &sysfs_writeback_rate_minimum,
 526         &sysfs_writeback_rate_debug,
 527         &sysfs_io_errors,
 528         &sysfs_io_error_limit,
 529         &sysfs_io_disable,
 530         &sysfs_dirty_data,
 531         &sysfs_stripe_size,
 532         &sysfs_partial_stripes_expensive,
 533         &sysfs_sequential_cutoff,
 534         &sysfs_clear_stats,
 535         &sysfs_running,
 536         &sysfs_state,
 537         &sysfs_label,
 538 #ifdef CONFIG_BCACHE_DEBUG
 539         &sysfs_verify,
 540         &sysfs_bypass_torture_test,
 541 #endif
 542         &sysfs_backing_dev_name,
 543         &sysfs_backing_dev_uuid,
 544         NULL
 545 };
 546 ATTRIBUTE_GROUPS(bch_cached_dev);
 547 KTYPE(bch_cached_dev);
 548
 549 SHOW(bch_flash_dev)
 550 {
 551         struct bcache_device *d = container_of(kobj, struct bcache_device,
 552                                                kobj);
 553         struct uuid_entry *u = &d->c->uuids[d->id];
 554
 555         sysfs_printf(data_csum, "%i", d->data_csum);
 556         sysfs_hprint(size,      u->sectors << 9);
 557
 558         if (attr == &sysfs_label) {
 559                 memcpy(buf, u->label, SB_LABEL_SIZE);
 560                 buf[SB_LABEL_SIZE + 1] = '\0';
 561                 strcat(buf, "\n");
 562                 return strlen(buf);
 563         }
 564
 565         return 0;
 566 }
 567
 568 STORE(__bch_flash_dev)
 569 {
 570         struct bcache_device *d = container_of(kobj, struct bcache_device,
 571                                                kobj);
 572         struct uuid_entry *u = &d->c->uuids[d->id];
 573
 574         /* no user space access if system is rebooting */
 575         if (bcache_is_reboot)
 576                 return -EBUSY;
 577
 578         sysfs_strtoul(data_csum,        d->data_csum);
 579
 580         if (attr == &sysfs_size) {
 581                 uint64_t v;
 582
 583                 strtoi_h_or_return(buf, v);
 584
 585                 u->sectors = v >> 9;
 586                 bch_uuid_write(d->c);
 587                 set_capacity(d->disk, u->sectors);
 588         }
 589
 590         if (attr == &sysfs_label) {
 591                 memcpy(u->label, buf, SB_LABEL_SIZE);
 592                 bch_uuid_write(d->c);
 593         }
 594
 595         if (attr == &sysfs_unregister) {
 596                 set_bit(BCACHE_DEV_DETACHING, &d->flags);
 597                 bcache_device_stop(d);
 598         }
 599
 600         return size;
 601 }
 602 STORE_LOCKED(bch_flash_dev)
 603
 604 static struct attribute *bch_flash_dev_attrs[] = {
 605         &sysfs_unregister,
 606 #if 0
 607         &sysfs_data_csum,
 608 #endif
 609         &sysfs_label,
 610         &sysfs_size,
 611         NULL
 612 };
 613 ATTRIBUTE_GROUPS(bch_flash_dev);
 614 KTYPE(bch_flash_dev);
 615
 616 struct bset_stats_op {
 617         struct btree_op op;
 618         size_t nodes;
 619         struct bset_stats stats;
 620 };
 621
 622 static int bch_btree_bset_stats(struct btree_op *b_op, struct btree *b)
 623 {
 624         struct bset_stats_op *op = container_of(b_op, struct bset_stats_op, op);
 625
 626         op->nodes++;
 627         bch_btree_keys_stats(&b->keys, &op->stats);
 628
 629         return MAP_CONTINUE;
 630 }
 631
 632 static int bch_bset_print_stats(struct cache_set *c, char *buf)
 633 {
 634         struct bset_stats_op op;
 635         int ret;
 636
 637         memset(&op, 0, sizeof(op));
 638         bch_btree_op_init(&op.op, -1);
 639
 640         ret = bch_btree_map_nodes(&op.op, c, &ZERO_KEY, bch_btree_bset_stats);
 641         if (ret < 0)
 642                 return ret;
 643
 644         return snprintf(buf, PAGE_SIZE,
 645                         "btree nodes:           %zu\n"
 646                         "written sets:          %zu\n"
 647                         "unwritten sets:                %zu\n"
 648                         "written key bytes:     %zu\n"
 649                         "unwritten key bytes:   %zu\n"
 650                         "floats:                        %zu\n"
 651                         "failed:                        %zu\n",
 652                         op.nodes,
 653                         op.stats.sets_written, op.stats.sets_unwritten,
 654                         op.stats.bytes_written, op.stats.bytes_unwritten,
 655                         op.stats.floats, op.stats.failed);
 656 }
 657
 658 static unsigned int bch_root_usage(struct cache_set *c)
 659 {
 660         unsigned int bytes = 0;
 661         struct bkey *k;
 662         struct btree *b;
 663         struct btree_iter iter;
 664
 665         min_heap_init(&iter.heap, NULL, MAX_BSETS);
 666
 667         goto lock_root;
 668
 669         do {
 670                 rw_unlock(false, b);
 671 lock_root:
 672                 b = c->root;
 673                 rw_lock(false, b, b->level);
 674         } while (b != c->root);
 675
 676         for_each_key_filter(&b->keys, k, &iter, bch_ptr_bad)
 677                 bytes += bkey_bytes(k);
 678
 679         rw_unlock(false, b);
 680
 681         return (bytes * 100) / btree_bytes(c);
 682 }
 683
 684 static size_t bch_cache_size(struct cache_set *c)
 685 {
 686         size_t ret = 0;
 687         struct btree *b;
 688
 689         mutex_lock(&c->bucket_lock);
 690         list_for_each_entry(b, &c->btree_cache, list)
 691                 ret += 1 << (b->keys.page_order + PAGE_SHIFT);
 692
 693         mutex_unlock(&c->bucket_lock);
 694         return ret;
 695 }
 696
 697 static unsigned int bch_cache_max_chain(struct cache_set *c)
 698 {
 699         unsigned int ret = 0;
 700         struct hlist_head *h;
 701
 702         mutex_lock(&c->bucket_lock);
 703
 704         for (h = c->bucket_hash;
 705              h < c->bucket_hash + (1 << BUCKET_HASH_BITS);
 706              h++) {
 707                 ret = max(ret, hlist_count_nodes(h));
 708         }
 709
 710         mutex_unlock(&c->bucket_lock);
 711         return ret;
 712 }
 713
 714 static unsigned int bch_btree_used(struct cache_set *c)
 715 {
 716         return div64_u64(c->gc_stats.key_bytes * 100,
 717                          (c->gc_stats.nodes ?: 1) * btree_bytes(c));
 718 }
 719
 720 static unsigned int bch_average_key_size(struct cache_set *c)
 721 {
 722         return c->gc_stats.nkeys
 723                 ? div64_u64(c->gc_stats.data, c->gc_stats.nkeys)
 724                 : 0;
 725 }
 726
 727 SHOW(__bch_cache_set)
 728 {
 729         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 730
 731         sysfs_print(synchronous,                CACHE_SYNC(&c->cache->sb));
 732         sysfs_print(journal_delay_ms,           c->journal_delay_ms);
 733         sysfs_hprint(bucket_size,               bucket_bytes(c->cache));
 734         sysfs_hprint(block_size,                block_bytes(c->cache));
 735         sysfs_print(tree_depth,                 c->root->level);
 736         sysfs_print(root_usage_percent,         bch_root_usage(c));
 737
 738         sysfs_hprint(btree_cache_size,          bch_cache_size(c));
 739         sysfs_print(btree_cache_max_chain,      bch_cache_max_chain(c));
 740         sysfs_print(cache_available_percent,    100 - c->gc_stats.in_use);
 741
 742         sysfs_print_time_stats(&c->btree_gc_time,       btree_gc, sec, ms);
 743         sysfs_print_time_stats(&c->btree_split_time,    btree_split, sec, us);
 744         sysfs_print_time_stats(&c->sort.time,           btree_sort, ms, us);
 745         sysfs_print_time_stats(&c->btree_read_time,     btree_read, ms, us);
 746
 747         sysfs_print(btree_used_percent, bch_btree_used(c));
 748         sysfs_print(btree_nodes,        c->gc_stats.nodes);
 749         sysfs_hprint(average_key_size,  bch_average_key_size(c));
 750
 751         sysfs_print(cache_read_races,
 752                     atomic_long_read(&c->cache_read_races));
 753
 754         sysfs_print(reclaim,
 755                     atomic_long_read(&c->reclaim));
 756
 757         sysfs_print(reclaimed_journal_buckets,
 758                     atomic_long_read(&c->reclaimed_journal_buckets));
 759
 760         sysfs_print(flush_write,
 761                     atomic_long_read(&c->flush_write));
 762
 763         sysfs_print(writeback_keys_done,
 764                     atomic_long_read(&c->writeback_keys_done));
 765         sysfs_print(writeback_keys_failed,
 766                     atomic_long_read(&c->writeback_keys_failed));
 767
 768         if (attr == &sysfs_errors)
 769                 return bch_snprint_string_list(buf, PAGE_SIZE, error_actions,
 770                                                c->on_error);
 771
 772         /* See count_io_errors for why 88 */
 773         sysfs_print(io_error_halflife,  c->error_decay * 88);
 774         sysfs_print(io_error_limit,     c->error_limit);
 775
 776         sysfs_hprint(congested,
 777                      ((uint64_t) bch_get_congested(c)) << 9);
 778         sysfs_print(congested_read_threshold_us,
 779                     c->congested_read_threshold_us);
 780         sysfs_print(congested_write_threshold_us,
 781                     c->congested_write_threshold_us);
 782
 783         sysfs_print(cutoff_writeback, bch_cutoff_writeback);
 784         sysfs_print(cutoff_writeback_sync, bch_cutoff_writeback_sync);
 785
 786         sysfs_print(active_journal_entries,     fifo_used(&c->journal.pin));
 787         sysfs_printf(verify,                    "%i", c->verify);
 788         sysfs_printf(key_merging_disabled,      "%i", c->key_merging_disabled);
 789         sysfs_printf(expensive_debug_checks,
 790                      "%i", c->expensive_debug_checks);
 791         sysfs_printf(gc_always_rewrite,         "%i", c->gc_always_rewrite);
 792         sysfs_printf(btree_shrinker_disabled,   "%i", c->shrinker_disabled);
 793         sysfs_printf(copy_gc_enabled,           "%i", c->copy_gc_enabled);
 794         sysfs_printf(idle_max_writeback_rate,   "%i",
 795                      c->idle_max_writeback_rate_enabled);
 796         sysfs_printf(gc_after_writeback,        "%i", c->gc_after_writeback);
 797         sysfs_printf(io_disable,                "%i",
 798                      test_bit(CACHE_SET_IO_DISABLE, &c->flags));
 799
 800         if (attr == &sysfs_bset_tree_stats)
 801                 return bch_bset_print_stats(c, buf);
 802
 803         if (attr == &sysfs_feature_compat)
 804                 return bch_print_cache_set_feature_compat(c, buf, PAGE_SIZE);
 805         if (attr == &sysfs_feature_ro_compat)
 806                 return bch_print_cache_set_feature_ro_compat(c, buf, PAGE_SIZE);
 807         if (attr == &sysfs_feature_incompat)
 808                 return bch_print_cache_set_feature_incompat(c, buf, PAGE_SIZE);
 809
 810         return 0;
 811 }
 812 SHOW_LOCKED(bch_cache_set)
 813
 814 STORE(__bch_cache_set)
 815 {
 816         struct cache_set *c = container_of(kobj, struct cache_set, kobj);
 817         ssize_t v;
 818
 819         /* no user space access if system is rebooting */
 820         if (bcache_is_reboot)
 821                 return -EBUSY;
 822
 823         if (attr == &sysfs_unregister)
 824                 bch_cache_set_unregister(c);
 825
 826         if (attr == &sysfs_stop)
 827                 bch_cache_set_stop(c);
 828
 829         if (attr == &sysfs_synchronous) {
 830                 bool sync = strtoul_or_return(buf);
 831
 832                 if (sync != CACHE_SYNC(&c->cache->sb)) {
 833                         SET_CACHE_SYNC(&c->cache->sb, sync);
 834                         bcache_write_super(c);
 835                 }
 836         }
 837
 838         if (attr == &sysfs_flash_vol_create) {
 839                 int r;
 840                 uint64_t v;
 841
 842                 strtoi_h_or_return(buf, v);
 843
 844                 r = bch_flash_dev_create(c, v);
 845                 if (r)
 846                         return r;
 847         }
 848
 849         if (attr == &sysfs_clear_stats) {
 850                 atomic_long_set(&c->writeback_keys_done,        0);
 851                 atomic_long_set(&c->writeback_keys_failed,      0);
 852
 853                 memset(&c->gc_stats, 0, sizeof(struct gc_stat));
 854                 bch_cache_accounting_clear(&c->accounting);
 855         }
 856
 857         if (attr == &sysfs_trigger_gc)
 858                 force_wake_up_gc(c);
 859
 860         if (attr == &sysfs_prune_cache) {
 861                 struct shrink_control sc;
 862
 863                 sc.gfp_mask = GFP_KERNEL;
 864                 sc.nr_to_scan = strtoul_or_return(buf);
 865                 if (c->shrink)
 866                         c->shrink->scan_objects(c->shrink, &sc);
 867         }
 868
 869         sysfs_strtoul_clamp(congested_read_threshold_us,
 870                             c->congested_read_threshold_us,
 871                             0, UINT_MAX);
 872         sysfs_strtoul_clamp(congested_write_threshold_us,
 873                             c->congested_write_threshold_us,
 874                             0, UINT_MAX);
 875
 876         if (attr == &sysfs_errors) {
 877                 v = __sysfs_match_string(error_actions, -1, buf);
 878                 if (v < 0)
 879                         return v;
 880
 881                 c->on_error = v;
 882         }
 883
 884         sysfs_strtoul_clamp(io_error_limit, c->error_limit, 0, UINT_MAX);
 885
 886         /* See count_io_errors() for why 88 */
 887         if (attr == &sysfs_io_error_halflife) {
 888                 unsigned long v = 0;
 889                 ssize_t ret;
 890
 891                 ret = strtoul_safe_clamp(buf, v, 0, UINT_MAX);
 892                 if (!ret) {
 893                         c->error_decay = v / 88;
 894                         return size;
 895                 }
 896                 return ret;
 897         }
 898
 899         if (attr == &sysfs_io_disable) {
 900                 v = strtoul_or_return(buf);
 901                 if (v) {
 902                         if (test_and_set_bit(CACHE_SET_IO_DISABLE,
 903                                              &c->flags))
 904                                 pr_warn("CACHE_SET_IO_DISABLE already set\n");
 905                 } else {
 906                         if (!test_and_clear_bit(CACHE_SET_IO_DISABLE,
 907                                                 &c->flags))
 908                                 pr_warn("CACHE_SET_IO_DISABLE already cleared\n");
 909                 }
 910         }
 911
 912         sysfs_strtoul_clamp(journal_delay_ms,
 913                             c->journal_delay_ms,
 914                             0, USHRT_MAX);
 915         sysfs_strtoul_bool(verify,              c->verify);
 916         sysfs_strtoul_bool(key_merging_disabled, c->key_merging_disabled);
 917         sysfs_strtoul(expensive_debug_checks,   c->expensive_debug_checks);
 918         sysfs_strtoul_bool(gc_always_rewrite,   c->gc_always_rewrite);
 919         sysfs_strtoul_bool(btree_shrinker_disabled, c->shrinker_disabled);
 920         sysfs_strtoul_bool(copy_gc_enabled,     c->copy_gc_enabled);
 921         sysfs_strtoul_bool(idle_max_writeback_rate,
 922                            c->idle_max_writeback_rate_enabled);
 923
 924         /*
 925          * write gc_after_writeback here may overwrite an already set
 926          * BCH_DO_AUTO_GC, it doesn't matter because this flag will be
 927          * set in next chance.
 928          */
 929         sysfs_strtoul_clamp(gc_after_writeback, c->gc_after_writeback, 0, 1);
 930
 931         return size;
 932 }
 933 STORE_LOCKED(bch_cache_set)
 934
 935 SHOW(bch_cache_set_internal)
 936 {
 937         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 938
 939         return bch_cache_set_show(&c->kobj, attr, buf);
 940 }
 941
 942 STORE(bch_cache_set_internal)
 943 {
 944         struct cache_set *c = container_of(kobj, struct cache_set, internal);
 945
 946         /* no user space access if system is rebooting */
 947         if (bcache_is_reboot)
 948                 return -EBUSY;
 949
 950         return bch_cache_set_store(&c->kobj, attr, buf, size);
 951 }
 952
 953 static void bch_cache_set_internal_release(struct kobject *k)
 954 {
 955 }
 956
 957 static struct attribute *bch_cache_set_attrs[] = {
 958         &sysfs_unregister,
 959         &sysfs_stop,
 960         &sysfs_synchronous,
 961         &sysfs_journal_delay_ms,
 962         &sysfs_flash_vol_create,
 963
 964         &sysfs_bucket_size,
 965         &sysfs_block_size,
 966         &sysfs_tree_depth,
 967         &sysfs_root_usage_percent,
 968         &sysfs_btree_cache_size,
 969         &sysfs_cache_available_percent,
 970
 971         &sysfs_average_key_size,
 972
 973         &sysfs_errors,
 974         &sysfs_io_error_limit,
 975         &sysfs_io_error_halflife,
 976         &sysfs_congested,
 977         &sysfs_congested_read_threshold_us,
 978         &sysfs_congested_write_threshold_us,
 979         &sysfs_clear_stats,
 980         NULL
 981 };
 982 ATTRIBUTE_GROUPS(bch_cache_set);
 983 KTYPE(bch_cache_set);
 984
 985 static struct attribute *bch_cache_set_internal_attrs[] = {
 986         &sysfs_active_journal_entries,
 987
 988         sysfs_time_stats_attribute_list(btree_gc, sec, ms)
 989         sysfs_time_stats_attribute_list(btree_split, sec, us)
 990         sysfs_time_stats_attribute_list(btree_sort, ms, us)
 991         sysfs_time_stats_attribute_list(btree_read, ms, us)
 992
 993         &sysfs_btree_nodes,
 994         &sysfs_btree_used_percent,
 995         &sysfs_btree_cache_max_chain,
 996
 997         &sysfs_bset_tree_stats,
 998         &sysfs_cache_read_races,
 999         &sysfs_reclaim,
1000         &sysfs_reclaimed_journal_buckets,
1001         &sysfs_flush_write,
1002         &sysfs_writeback_keys_done,
1003         &sysfs_writeback_keys_failed,
1004
1005         &sysfs_trigger_gc,
1006         &sysfs_prune_cache,
1007 #ifdef CONFIG_BCACHE_DEBUG
1008         &sysfs_verify,
1009         &sysfs_key_merging_disabled,
1010         &sysfs_expensive_debug_checks,
1011 #endif
1012         &sysfs_gc_always_rewrite,
1013         &sysfs_btree_shrinker_disabled,
1014         &sysfs_copy_gc_enabled,
1015         &sysfs_idle_max_writeback_rate,
1016         &sysfs_gc_after_writeback,
1017         &sysfs_io_disable,
1018         &sysfs_cutoff_writeback,
1019         &sysfs_cutoff_writeback_sync,
1020         &sysfs_feature_compat,
1021         &sysfs_feature_ro_compat,
1022         &sysfs_feature_incompat,
1023         NULL
1024 };
1025 ATTRIBUTE_GROUPS(bch_cache_set_internal);
1026 KTYPE(bch_cache_set_internal);
1027
1028 static int __bch_cache_cmp(const void *l, const void *r)
1029 {
1030         cond_resched();
1031         return *((uint16_t *)r) - *((uint16_t *)l);
1032 }
1033
1034 SHOW(__bch_cache)
1035 {
1036         struct cache *ca = container_of(kobj, struct cache, kobj);
1037
1038         sysfs_hprint(bucket_size,       bucket_bytes(ca));
1039         sysfs_hprint(block_size,        block_bytes(ca));
1040         sysfs_print(nbuckets,           ca->sb.nbuckets);
1041         sysfs_print(discard,            ca->discard);
1042         sysfs_hprint(written, atomic_long_read(&ca->sectors_written) << 9);
1043         sysfs_hprint(btree_written,
1044                      atomic_long_read(&ca->btree_sectors_written) << 9);
1045         sysfs_hprint(metadata_written,
1046                      (atomic_long_read(&ca->meta_sectors_written) +
1047                       atomic_long_read(&ca->btree_sectors_written)) << 9);
1048
1049         sysfs_print(io_errors,
1050                     atomic_read(&ca->io_errors) >> IO_ERROR_SHIFT);
1051
1052         if (attr == &sysfs_cache_replacement_policy)
1053                 return bch_snprint_string_list(buf, PAGE_SIZE,
1054                                                cache_replacement_policies,
1055                                                CACHE_REPLACEMENT(&ca->sb));
1056
1057         if (attr == &sysfs_priority_stats) {
1058                 struct bucket *b;
1059                 size_t n = ca->sb.nbuckets, i;
1060                 size_t unused = 0, available = 0, dirty = 0, meta = 0;
1061                 uint64_t sum = 0;
1062                 /* Compute 31 quantiles */
1063                 uint16_t q[31], *p, *cached;
1064                 ssize_t ret;
1065
1066                 cached = p = vmalloc(array_size(sizeof(uint16_t),
1067                                                 ca->sb.nbuckets));
1068                 if (!p)
1069                         return -ENOMEM;
1070
1071                 mutex_lock(&ca->set->bucket_lock);
1072                 for_each_bucket(b, ca) {
1073                         if (!GC_SECTORS_USED(b))
1074                                 unused++;
1075                         if (GC_MARK(b) == GC_MARK_RECLAIMABLE)
1076                                 available++;
1077                         if (GC_MARK(b) == GC_MARK_DIRTY)
1078                                 dirty++;
1079                         if (GC_MARK(b) == GC_MARK_METADATA)
1080                                 meta++;
1081                 }
1082
1083                 for (i = ca->sb.first_bucket; i < n; i++)
1084                         p[i] = ca->buckets[i].prio;
1085                 mutex_unlock(&ca->set->bucket_lock);
1086
1087                 sort(p, n, sizeof(uint16_t), __bch_cache_cmp, NULL);
1088
1089                 while (n &&
1090                        !cached[n - 1])
1091                         --n;
1092
1093                 while (cached < p + n &&
1094                        *cached == BTREE_PRIO) {
1095                         cached++;
1096                         n--;
1097                 }
1098
1099                 for (i = 0; i < n; i++)
1100                         sum += INITIAL_PRIO - cached[i];
1101
1102                 if (n)
1103                         sum = div64_u64(sum, n);
1104
1105                 for (i = 0; i < ARRAY_SIZE(q); i++)
1106                         q[i] = INITIAL_PRIO - cached[n * (i + 1) /
1107                                 (ARRAY_SIZE(q) + 1)];
1108
1109                 vfree(p);
1110
1111                 ret = sysfs_emit(buf,
1112                                  "Unused:               %zu%%\n"
1113                                  "Clean:                %zu%%\n"
1114                                  "Dirty:                %zu%%\n"
1115                                  "Metadata:     %zu%%\n"
1116                                  "Average:      %llu\n"
1117                                  "Sectors per Q:        %zu\n"
1118                                  "Quantiles:    [",
1119                                  unused * 100 / (size_t) ca->sb.nbuckets,
1120                                  available * 100 / (size_t) ca->sb.nbuckets,
1121                                  dirty * 100 / (size_t) ca->sb.nbuckets,
1122                                  meta * 100 / (size_t) ca->sb.nbuckets, sum,
1123                                  n * ca->sb.bucket_size / (ARRAY_SIZE(q) + 1));
1124
1125                 for (i = 0; i < ARRAY_SIZE(q); i++)
1126                         ret += sysfs_emit_at(buf, ret, "%u ", q[i]);
1127                 ret--;
1128
1129                 ret += sysfs_emit_at(buf, ret, "]\n");
1130
1131                 return ret;
1132         }
1133
1134         return 0;
1135 }
1136 SHOW_LOCKED(bch_cache)
1137
1138 STORE(__bch_cache)
1139 {
1140         struct cache *ca = container_of(kobj, struct cache, kobj);
1141         ssize_t v;
1142
1143         /* no user space access if system is rebooting */
1144         if (bcache_is_reboot)
1145                 return -EBUSY;
1146
1147         if (attr == &sysfs_discard) {
1148                 bool v = strtoul_or_return(buf);
1149
1150                 if (bdev_max_discard_sectors(ca->bdev))
1151                         ca->discard = v;
1152
1153                 if (v != CACHE_DISCARD(&ca->sb)) {
1154                         SET_CACHE_DISCARD(&ca->sb, v);
1155                         bcache_write_super(ca->set);
1156                 }
1157         }
1158
1159         if (attr == &sysfs_cache_replacement_policy) {
1160                 v = __sysfs_match_string(cache_replacement_policies, -1, buf);
1161                 if (v < 0)
1162                         return v;
1163
1164                 if ((unsigned int) v != CACHE_REPLACEMENT(&ca->sb)) {
1165                         mutex_lock(&ca->set->bucket_lock);
1166                         SET_CACHE_REPLACEMENT(&ca->sb, v);
1167                         mutex_unlock(&ca->set->bucket_lock);
1168
1169                         bcache_write_super(ca->set);
1170                 }
1171         }
1172
1173         if (attr == &sysfs_clear_stats) {
1174                 atomic_long_set(&ca->sectors_written, 0);
1175                 atomic_long_set(&ca->btree_sectors_written, 0);
1176                 atomic_long_set(&ca->meta_sectors_written, 0);
1177                 atomic_set(&ca->io_count, 0);
1178                 atomic_set(&ca->io_errors, 0);
1179         }
1180
1181         return size;
1182 }
1183 STORE_LOCKED(bch_cache)
1184
1185 static struct attribute *bch_cache_attrs[] = {
1186         &sysfs_bucket_size,
1187         &sysfs_block_size,
1188         &sysfs_nbuckets,
1189         &sysfs_priority_stats,
1190         &sysfs_discard,
1191         &sysfs_written,
1192         &sysfs_btree_written,
1193         &sysfs_metadata_written,
1194         &sysfs_io_errors,
1195         &sysfs_clear_stats,
1196         &sysfs_cache_replacement_policy,
1197         NULL
1198 };
1199 ATTRIBUTE_GROUPS(bch_cache);
1200 KTYPE(bch_cache);