git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/sysfs.c

   1 /*
   2  * bcache sysfs interfaces
   3  *
   4  * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
   5  * Copyright 2012 Google, Inc.
   6  */
   7
   8 #include "bcachefs.h"
   9 #include "alloc.h"
  10 #include "compress.h"
  11 #include "sysfs.h"
  12 #include "btree_cache.h"
  13 #include "btree_iter.h"
  14 #include "btree_update.h"
  15 #include "btree_gc.h"
  16 #include "buckets.h"
  17 #include "inode.h"
  18 #include "journal.h"
  19 #include "keylist.h"
  20 #include "move.h"
  21 #include "opts.h"
  22 #include "super-io.h"
  23 #include "tier.h"
  24
  25 #include <linux/blkdev.h>
  26 #include <linux/sort.h>
  27 #include <linux/sched/clock.h>
  28
  29 #include "util.h"
  30
  31 #define SYSFS_OPS(type)                                                 \
  32 struct sysfs_ops type ## _sysfs_ops = {                                 \
  33         .show   = type ## _show,                                        \
  34         .store  = type ## _store                                        \
  35 }
  36
  37 #define SHOW(fn)                                                        \
  38 static ssize_t fn ## _show(struct kobject *kobj, struct attribute *attr,\
  39                            char *buf)                                   \
  40
  41 #define STORE(fn)                                                       \
  42 static ssize_t fn ## _store(struct kobject *kobj, struct attribute *attr,\
  43                             const char *buf, size_t size)               \
  44
  45 #define __sysfs_attribute(_name, _mode)                                 \
  46         static struct attribute sysfs_##_name =                         \
  47                 { .name = #_name, .mode = _mode }
  48
  49 #define write_attribute(n)      __sysfs_attribute(n, S_IWUSR)
  50 #define read_attribute(n)       __sysfs_attribute(n, S_IRUGO)
  51 #define rw_attribute(n)         __sysfs_attribute(n, S_IRUGO|S_IWUSR)
  52
  53 #define sysfs_printf(file, fmt, ...)                                    \
  54 do {                                                                    \
  55         if (attr == &sysfs_ ## file)                                    \
  56                 return snprintf(buf, PAGE_SIZE, fmt "\n", __VA_ARGS__); \
  57 } while (0)
  58
  59 #define sysfs_print(file, var)                                          \
  60 do {                                                                    \
  61         if (attr == &sysfs_ ## file)                                    \
  62                 return snprint(buf, PAGE_SIZE, var);                    \
  63 } while (0)
  64
  65 #define sysfs_hprint(file, val)                                         \
  66 do {                                                                    \
  67         if (attr == &sysfs_ ## file) {                                  \
  68                 ssize_t ret = bch2_hprint(buf, val);                    \
  69                 strcat(buf, "\n");                                      \
  70                 return ret + 1;                                         \
  71         }                                                               \
  72 } while (0)
  73
  74 #define var_printf(_var, fmt)   sysfs_printf(_var, fmt, var(_var))
  75 #define var_print(_var)         sysfs_print(_var, var(_var))
  76 #define var_hprint(_var)        sysfs_hprint(_var, var(_var))
  77
  78 #define sysfs_strtoul(file, var)                                        \
  79 do {                                                                    \
  80         if (attr == &sysfs_ ## file)                                    \
  81                 return strtoul_safe(buf, var) ?: (ssize_t) size;        \
  82 } while (0)
  83
  84 #define sysfs_strtoul_clamp(file, var, min, max)                        \
  85 do {                                                                    \
  86         if (attr == &sysfs_ ## file)                                    \
  87                 return strtoul_safe_clamp(buf, var, min, max)           \
  88                         ?: (ssize_t) size;                              \
  89 } while (0)
  90
  91 #define strtoul_or_return(cp)                                           \
  92 ({                                                                      \
  93         unsigned long _v;                                               \
  94         int _r = kstrtoul(cp, 10, &_v);                                 \
  95         if (_r)                                                         \
  96                 return _r;                                              \
  97         _v;                                                             \
  98 })
  99
 100 #define strtoul_restrict_or_return(cp, min, max)                        \
 101 ({                                                                      \
 102         unsigned long __v = 0;                                          \
 103         int _r = strtoul_safe_restrict(cp, __v, min, max);              \
 104         if (_r)                                                         \
 105                 return _r;                                              \
 106         __v;                                                            \
 107 })
 108
 109 #define strtoi_h_or_return(cp)                                          \
 110 ({                                                                      \
 111         u64 _v;                                                         \
 112         int _r = strtoi_h(cp, &_v);                                     \
 113         if (_r)                                                         \
 114                 return _r;                                              \
 115         _v;                                                             \
 116 })
 117
 118 #define sysfs_hatoi(file, var)                                          \
 119 do {                                                                    \
 120         if (attr == &sysfs_ ## file)                                    \
 121                 return strtoi_h(buf, &var) ?: (ssize_t) size;           \
 122 } while (0)
 123
 124 write_attribute(trigger_journal_flush);
 125 write_attribute(trigger_btree_coalesce);
 126 write_attribute(trigger_gc);
 127 write_attribute(prune_cache);
 128 rw_attribute(btree_gc_periodic);
 129
 130 read_attribute(uuid);
 131 read_attribute(minor);
 132 read_attribute(bucket_size);
 133 read_attribute(block_size);
 134 read_attribute(btree_node_size);
 135 read_attribute(first_bucket);
 136 read_attribute(nbuckets);
 137 read_attribute(read_priority_stats);
 138 read_attribute(write_priority_stats);
 139 read_attribute(fragmentation_stats);
 140 read_attribute(oldest_gen_stats);
 141 read_attribute(reserve_stats);
 142 read_attribute(btree_cache_size);
 143 read_attribute(compression_stats);
 144 read_attribute(written);
 145 read_attribute(btree_written);
 146 read_attribute(metadata_written);
 147 read_attribute(journal_debug);
 148 read_attribute(journal_pins);
 149
 150 read_attribute(internal_uuid);
 151
 152 read_attribute(available_buckets);
 153 read_attribute(free_buckets);
 154 read_attribute(dirty_data);
 155 read_attribute(dirty_bytes);
 156 read_attribute(dirty_buckets);
 157 read_attribute(cached_data);
 158 read_attribute(cached_bytes);
 159 read_attribute(cached_buckets);
 160 read_attribute(meta_buckets);
 161 read_attribute(alloc_buckets);
 162 read_attribute(has_data);
 163 read_attribute(has_metadata);
 164 read_attribute(alloc_debug);
 165
 166 read_attribute(read_realloc_races);
 167
 168 rw_attribute(journal_write_delay_ms);
 169 rw_attribute(journal_reclaim_delay_ms);
 170
 171 rw_attribute(discard);
 172 rw_attribute(cache_replacement_policy);
 173
 174 rw_attribute(foreground_write_ratelimit_enabled);
 175 rw_attribute(copy_gc_enabled);
 176 sysfs_pd_controller_attribute(copy_gc);
 177
 178 rw_attribute(tier);
 179 rw_attribute(tiering_enabled);
 180 rw_attribute(tiering_percent);
 181 sysfs_pd_controller_attribute(tiering);
 182
 183 sysfs_pd_controller_attribute(foreground_write);
 184
 185 rw_attribute(pd_controllers_update_seconds);
 186
 187 rw_attribute(foreground_target_percent);
 188
 189 read_attribute(meta_replicas_have);
 190 read_attribute(data_replicas_have);
 191
 192 #define BCH_DEBUG_PARAM(name, description)                              \
 193         rw_attribute(name);
 194
 195         BCH_DEBUG_PARAMS()
 196 #undef BCH_DEBUG_PARAM
 197
 198 #define BCH_OPT(_name, _mode, ...)                                      \
 199         static struct attribute sysfs_opt_##_name = {                   \
 200                 .name = #_name, .mode = _mode,                          \
 201         };
 202
 203         BCH_VISIBLE_OPTS()
 204 #undef BCH_OPT
 205
 206 #define BCH_TIME_STAT(name, frequency_units, duration_units)            \
 207         sysfs_time_stats_attribute(name, frequency_units, duration_units);
 208         BCH_TIME_STATS()
 209 #undef BCH_TIME_STAT
 210
 211 static struct attribute sysfs_state_rw = {
 212         .name = "state",
 213         .mode = S_IRUGO
 214 };
 215
 216 static size_t bch2_btree_cache_size(struct bch_fs *c)
 217 {
 218         size_t ret = 0;
 219         struct btree *b;
 220
 221         mutex_lock(&c->btree_cache_lock);
 222         list_for_each_entry(b, &c->btree_cache, list)
 223                 ret += btree_bytes(c);
 224
 225         mutex_unlock(&c->btree_cache_lock);
 226         return ret;
 227 }
 228
 229 static ssize_t show_fs_alloc_debug(struct bch_fs *c, char *buf)
 230 {
 231         struct bch_fs_usage stats = bch2_fs_usage_read(c);
 232
 233         return scnprintf(buf, PAGE_SIZE,
 234                          "capacity:\t\t%llu\n"
 235                          "compressed:\n"
 236                          "\tmeta:\t\t%llu\n"
 237                          "\tdirty:\t\t%llu\n"
 238                          "\tcached:\t\t%llu\n"
 239                          "uncompressed:\n"
 240                          "\tmeta:\t\t%llu\n"
 241                          "\tdirty:\t\t%llu\n"
 242                          "\tcached:\t\t%llu\n"
 243                          "persistent reserved sectors:\t%llu\n"
 244                          "online reserved sectors:\t%llu\n",
 245                          c->capacity,
 246                          stats.s[S_COMPRESSED][S_META],
 247                          stats.s[S_COMPRESSED][S_DIRTY],
 248                          stats.s[S_COMPRESSED][S_CACHED],
 249                          stats.s[S_UNCOMPRESSED][S_META],
 250                          stats.s[S_UNCOMPRESSED][S_DIRTY],
 251                          stats.s[S_UNCOMPRESSED][S_CACHED],
 252                          stats.persistent_reserved,
 253                          stats.online_reserved);
 254 }
 255
 256 static ssize_t bch2_compression_stats(struct bch_fs *c, char *buf)
 257 {
 258         struct btree_iter iter;
 259         struct bkey_s_c k;
 260         u64 nr_uncompressed_extents = 0, uncompressed_sectors = 0,
 261             nr_compressed_extents = 0,
 262             compressed_sectors_compressed = 0,
 263             compressed_sectors_uncompressed = 0;
 264
 265         if (!bch2_fs_running(c))
 266                 return -EPERM;
 267
 268         for_each_btree_key(&iter, c, BTREE_ID_EXTENTS, POS_MIN, 0, k)
 269                 if (k.k->type == BCH_EXTENT) {
 270                         struct bkey_s_c_extent e = bkey_s_c_to_extent(k);
 271                         const struct bch_extent_ptr *ptr;
 272                         const union bch_extent_crc *crc;
 273
 274                         extent_for_each_ptr_crc(e, ptr, crc) {
 275                                 if (crc_compression_type(crc) == BCH_COMPRESSION_NONE) {
 276                                         nr_uncompressed_extents++;
 277                                         uncompressed_sectors += e.k->size;
 278                                 } else {
 279                                         nr_compressed_extents++;
 280                                         compressed_sectors_compressed +=
 281                                                 crc_compressed_size(e.k, crc);
 282                                         compressed_sectors_uncompressed +=
 283                                                 crc_uncompressed_size(e.k, crc);
 284                                 }
 285
 286                                 /* only looking at the first ptr */
 287                                 break;
 288                         }
 289                 }
 290         bch2_btree_iter_unlock(&iter);
 291
 292         return snprintf(buf, PAGE_SIZE,
 293                         "uncompressed data:\n"
 294                         "       nr extents:                     %llu\n"
 295                         "       size (bytes):                   %llu\n"
 296                         "compressed data:\n"
 297                         "       nr extents:                     %llu\n"
 298                         "       compressed size (bytes):        %llu\n"
 299                         "       uncompressed size (bytes):      %llu\n",
 300                         nr_uncompressed_extents,
 301                         uncompressed_sectors << 9,
 302                         nr_compressed_extents,
 303                         compressed_sectors_compressed << 9,
 304                         compressed_sectors_uncompressed << 9);
 305 }
 306
 307 SHOW(bch2_fs)
 308 {
 309         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 310
 311         sysfs_print(minor,                      c->minor);
 312         sysfs_printf(internal_uuid, "%pU",      c->sb.uuid.b);
 313
 314         sysfs_print(journal_write_delay_ms,     c->journal.write_delay_ms);
 315         sysfs_print(journal_reclaim_delay_ms,   c->journal.reclaim_delay_ms);
 316
 317         sysfs_print(block_size,                 block_bytes(c));
 318         sysfs_print(btree_node_size,            btree_bytes(c));
 319         sysfs_hprint(btree_cache_size,          bch2_btree_cache_size(c));
 320
 321         sysfs_print(read_realloc_races,
 322                     atomic_long_read(&c->read_realloc_races));
 323
 324         sysfs_printf(btree_gc_periodic, "%u",   (int) c->btree_gc_periodic);
 325
 326         sysfs_printf(foreground_write_ratelimit_enabled, "%i",
 327                      c->foreground_write_ratelimit_enabled);
 328         sysfs_printf(copy_gc_enabled, "%i", c->copy_gc_enabled);
 329         sysfs_pd_controller_show(foreground_write, &c->foreground_write_pd);
 330
 331         sysfs_print(pd_controllers_update_seconds,
 332                     c->pd_controllers_update_seconds);
 333         sysfs_print(foreground_target_percent, c->foreground_target_percent);
 334
 335         sysfs_printf(tiering_enabled,           "%i", c->tiering_enabled);
 336         sysfs_print(tiering_percent,            c->tiering_percent);
 337
 338         sysfs_pd_controller_show(tiering,       &c->tiers[1].pd); /* XXX */
 339
 340         sysfs_printf(meta_replicas_have, "%u",  c->sb.meta_replicas_have);
 341         sysfs_printf(data_replicas_have, "%u",  c->sb.data_replicas_have);
 342
 343         /* Debugging: */
 344
 345         if (attr == &sysfs_alloc_debug)
 346                 return show_fs_alloc_debug(c, buf);
 347
 348         if (attr == &sysfs_journal_debug)
 349                 return bch2_journal_print_debug(&c->journal, buf);
 350
 351         if (attr == &sysfs_journal_pins)
 352                 return bch2_journal_print_pins(&c->journal, buf);
 353
 354         if (attr == &sysfs_compression_stats)
 355                 return bch2_compression_stats(c, buf);
 356
 357 #define BCH_DEBUG_PARAM(name, description) sysfs_print(name, c->name);
 358         BCH_DEBUG_PARAMS()
 359 #undef BCH_DEBUG_PARAM
 360
 361         return 0;
 362 }
 363
 364 STORE(__bch2_fs)
 365 {
 366         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 367
 368         sysfs_strtoul(journal_write_delay_ms, c->journal.write_delay_ms);
 369         sysfs_strtoul(journal_reclaim_delay_ms, c->journal.reclaim_delay_ms);
 370
 371         sysfs_strtoul(foreground_write_ratelimit_enabled,
 372                       c->foreground_write_ratelimit_enabled);
 373
 374         if (attr == &sysfs_btree_gc_periodic) {
 375                 ssize_t ret = strtoul_safe(buf, c->btree_gc_periodic)
 376                         ?: (ssize_t) size;
 377
 378                 wake_up_process(c->gc_thread);
 379                 return ret;
 380         }
 381
 382         if (attr == &sysfs_copy_gc_enabled) {
 383                 struct bch_dev *ca;
 384                 unsigned i;
 385                 ssize_t ret = strtoul_safe(buf, c->copy_gc_enabled)
 386                         ?: (ssize_t) size;
 387
 388                 for_each_member_device(ca, c, i)
 389                         if (ca->moving_gc_read)
 390                                 wake_up_process(ca->moving_gc_read);
 391                 return ret;
 392         }
 393
 394         if (attr == &sysfs_tiering_enabled) {
 395                 ssize_t ret = strtoul_safe(buf, c->tiering_enabled)
 396                         ?: (ssize_t) size;
 397
 398                 bch2_tiering_start(c); /* issue wakeups */
 399                 return ret;
 400         }
 401
 402         sysfs_pd_controller_store(foreground_write, &c->foreground_write_pd);
 403
 404         sysfs_strtoul(pd_controllers_update_seconds,
 405                       c->pd_controllers_update_seconds);
 406         sysfs_strtoul(foreground_target_percent, c->foreground_target_percent);
 407
 408         sysfs_strtoul(tiering_percent,          c->tiering_percent);
 409         sysfs_pd_controller_store(tiering,      &c->tiers[1].pd); /* XXX */
 410
 411         /* Debugging: */
 412
 413 #define BCH_DEBUG_PARAM(name, description) sysfs_strtoul(name, c->name);
 414         BCH_DEBUG_PARAMS()
 415 #undef BCH_DEBUG_PARAM
 416
 417         if (!bch2_fs_running(c))
 418                 return -EPERM;
 419
 420         /* Debugging: */
 421
 422         if (attr == &sysfs_trigger_journal_flush)
 423                 bch2_journal_meta_async(&c->journal, NULL);
 424
 425         if (attr == &sysfs_trigger_btree_coalesce)
 426                 bch2_coalesce(c);
 427
 428         if (attr == &sysfs_trigger_gc)
 429                 bch2_gc(c);
 430
 431         if (attr == &sysfs_prune_cache) {
 432                 struct shrink_control sc;
 433
 434                 sc.gfp_mask = GFP_KERNEL;
 435                 sc.nr_to_scan = strtoul_or_return(buf);
 436                 c->btree_cache_shrink.scan_objects(&c->btree_cache_shrink, &sc);
 437         }
 438
 439         return size;
 440 }
 441
 442 STORE(bch2_fs)
 443 {
 444         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 445
 446         mutex_lock(&c->state_lock);
 447         size = __bch2_fs_store(kobj, attr, buf, size);
 448         mutex_unlock(&c->state_lock);
 449
 450         return size;
 451 }
 452 SYSFS_OPS(bch2_fs);
 453
 454 struct attribute *bch2_fs_files[] = {
 455         &sysfs_minor,
 456         &sysfs_block_size,
 457         &sysfs_btree_node_size,
 458         &sysfs_btree_cache_size,
 459
 460         &sysfs_meta_replicas_have,
 461         &sysfs_data_replicas_have,
 462
 463         &sysfs_journal_write_delay_ms,
 464         &sysfs_journal_reclaim_delay_ms,
 465
 466         &sysfs_foreground_target_percent,
 467         &sysfs_tiering_percent,
 468
 469         &sysfs_compression_stats,
 470         NULL
 471 };
 472
 473 /* internal dir - just a wrapper */
 474
 475 SHOW(bch2_fs_internal)
 476 {
 477         struct bch_fs *c = container_of(kobj, struct bch_fs, internal);
 478         return bch2_fs_show(&c->kobj, attr, buf);
 479 }
 480
 481 STORE(bch2_fs_internal)
 482 {
 483         struct bch_fs *c = container_of(kobj, struct bch_fs, internal);
 484         return bch2_fs_store(&c->kobj, attr, buf, size);
 485 }
 486 SYSFS_OPS(bch2_fs_internal);
 487
 488 struct attribute *bch2_fs_internal_files[] = {
 489         &sysfs_alloc_debug,
 490         &sysfs_journal_debug,
 491         &sysfs_journal_pins,
 492
 493         &sysfs_read_realloc_races,
 494
 495         &sysfs_trigger_journal_flush,
 496         &sysfs_trigger_btree_coalesce,
 497         &sysfs_trigger_gc,
 498         &sysfs_prune_cache,
 499
 500         &sysfs_foreground_write_ratelimit_enabled,
 501         &sysfs_copy_gc_enabled,
 502         &sysfs_tiering_enabled,
 503         sysfs_pd_controller_files(tiering),
 504         sysfs_pd_controller_files(foreground_write),
 505         &sysfs_internal_uuid,
 506
 507 #define BCH_DEBUG_PARAM(name, description) &sysfs_##name,
 508         BCH_DEBUG_PARAMS()
 509 #undef BCH_DEBUG_PARAM
 510
 511         NULL
 512 };
 513
 514 /* options */
 515
 516 SHOW(bch2_fs_opts_dir)
 517 {
 518         struct bch_fs *c = container_of(kobj, struct bch_fs, opts_dir);
 519
 520         return bch2_opt_show(&c->opts, attr->name, buf, PAGE_SIZE);
 521 }
 522
 523 STORE(bch2_fs_opts_dir)
 524 {
 525         struct bch_fs *c = container_of(kobj, struct bch_fs, opts_dir);
 526         const struct bch_option *opt;
 527         int id;
 528         u64 v;
 529
 530         id = bch2_parse_sysfs_opt(attr->name, buf, &v);
 531         if (id < 0)
 532                 return id;
 533
 534         opt = &bch2_opt_table[id];
 535
 536         mutex_lock(&c->sb_lock);
 537
 538         if (id == Opt_compression) {
 539                 int ret = bch2_check_set_has_compressed_data(c, v);
 540                 if (ret) {
 541                         mutex_unlock(&c->sb_lock);
 542                         return ret;
 543                 }
 544         }
 545
 546         if (opt->set_sb != SET_NO_SB_OPT) {
 547                 opt->set_sb(c->disk_sb, v);
 548                 bch2_write_super(c);
 549         }
 550
 551         bch2_opt_set(&c->opts, id, v);
 552
 553         mutex_unlock(&c->sb_lock);
 554
 555         return size;
 556 }
 557 SYSFS_OPS(bch2_fs_opts_dir);
 558
 559 struct attribute *bch2_fs_opts_dir_files[] = {
 560 #define BCH_OPT(_name, ...)                                             \
 561         &sysfs_opt_##_name,
 562
 563         BCH_VISIBLE_OPTS()
 564 #undef BCH_OPT
 565
 566         NULL
 567 };
 568
 569 /* time stats */
 570
 571 SHOW(bch2_fs_time_stats)
 572 {
 573         struct bch_fs *c = container_of(kobj, struct bch_fs, time_stats);
 574
 575 #define BCH_TIME_STAT(name, frequency_units, duration_units)            \
 576         sysfs_print_time_stats(&c->name##_time, name,                   \
 577                                frequency_units, duration_units);
 578         BCH_TIME_STATS()
 579 #undef BCH_TIME_STAT
 580
 581         return 0;
 582 }
 583
 584 STORE(bch2_fs_time_stats)
 585 {
 586         struct bch_fs *c = container_of(kobj, struct bch_fs, time_stats);
 587
 588 #define BCH_TIME_STAT(name, frequency_units, duration_units)            \
 589         sysfs_clear_time_stats(&c->name##_time, name);
 590         BCH_TIME_STATS()
 591 #undef BCH_TIME_STAT
 592
 593         return size;
 594 }
 595 SYSFS_OPS(bch2_fs_time_stats);
 596
 597 struct attribute *bch2_fs_time_stats_files[] = {
 598 #define BCH_TIME_STAT(name, frequency_units, duration_units)            \
 599         sysfs_time_stats_attribute_list(name, frequency_units, duration_units)
 600         BCH_TIME_STATS()
 601 #undef BCH_TIME_STAT
 602
 603         NULL
 604 };
 605
 606 typedef unsigned (bucket_map_fn)(struct bch_dev *, struct bucket *, void *);
 607
 608 static unsigned bucket_priority_fn(struct bch_dev *ca, struct bucket *g,
 609                                    void *private)
 610 {
 611         int rw = (private ? 1 : 0);
 612
 613         return ca->fs->prio_clock[rw].hand - g->prio[rw];
 614 }
 615
 616 static unsigned bucket_sectors_used_fn(struct bch_dev *ca, struct bucket *g,
 617                                        void *private)
 618 {
 619         return bucket_sectors_used(g->mark);
 620 }
 621
 622 static unsigned bucket_oldest_gen_fn(struct bch_dev *ca, struct bucket *g,
 623                                      void *private)
 624 {
 625         return bucket_gc_gen(ca, g);
 626 }
 627
 628 static ssize_t show_quantiles(struct bch_dev *ca, char *buf,
 629                               bucket_map_fn *fn, void *private)
 630 {
 631         int cmp(const void *l, const void *r)
 632         {       return *((unsigned *) r) - *((unsigned *) l); }
 633
 634         size_t n = ca->mi.nbuckets, i;
 635         /* Compute 31 quantiles */
 636         unsigned q[31], *p;
 637         ssize_t ret = 0;
 638
 639         p = vzalloc(ca->mi.nbuckets * sizeof(unsigned));
 640         if (!p)
 641                 return -ENOMEM;
 642
 643         for (i = ca->mi.first_bucket; i < n; i++)
 644                 p[i] = fn(ca, &ca->buckets[i], private);
 645
 646         sort(p, n, sizeof(unsigned), cmp, NULL);
 647
 648         while (n &&
 649                !p[n - 1])
 650                 --n;
 651
 652         for (i = 0; i < ARRAY_SIZE(q); i++)
 653                 q[i] = p[n * (i + 1) / (ARRAY_SIZE(q) + 1)];
 654
 655         vfree(p);
 656
 657         for (i = 0; i < ARRAY_SIZE(q); i++)
 658                 ret += scnprintf(buf + ret, PAGE_SIZE - ret,
 659                                  "%u ", q[i]);
 660         buf[ret - 1] = '\n';
 661
 662         return ret;
 663
 664 }
 665
 666 static ssize_t show_reserve_stats(struct bch_dev *ca, char *buf)
 667 {
 668         enum alloc_reserve i;
 669         ssize_t ret;
 670
 671         spin_lock(&ca->freelist_lock);
 672
 673         ret = scnprintf(buf, PAGE_SIZE,
 674                         "free_inc:\t%zu\t%zu\n",
 675                         fifo_used(&ca->free_inc),
 676                         ca->free_inc.size);
 677
 678         for (i = 0; i < RESERVE_NR; i++)
 679                 ret += scnprintf(buf + ret, PAGE_SIZE - ret,
 680                                  "free[%u]:\t%zu\t%zu\n", i,
 681                                  fifo_used(&ca->free[i]),
 682                                  ca->free[i].size);
 683
 684         spin_unlock(&ca->freelist_lock);
 685
 686         return ret;
 687 }
 688
 689 static ssize_t show_dev_alloc_debug(struct bch_dev *ca, char *buf)
 690 {
 691         struct bch_fs *c = ca->fs;
 692         struct bch_dev_usage stats = bch2_dev_usage_read(ca);
 693
 694         return scnprintf(buf, PAGE_SIZE,
 695                 "free_inc:               %zu/%zu\n"
 696                 "free[RESERVE_PRIO]:     %zu/%zu\n"
 697                 "free[RESERVE_BTREE]:    %zu/%zu\n"
 698                 "free[RESERVE_MOVINGGC]: %zu/%zu\n"
 699                 "free[RESERVE_NONE]:     %zu/%zu\n"
 700                 "alloc:                  %llu/%llu\n"
 701                 "meta:                   %llu/%llu\n"
 702                 "dirty:                  %llu/%llu\n"
 703                 "available:              %llu/%llu\n"
 704                 "freelist_wait:          %s\n"
 705                 "open buckets:           %u/%u (reserved %u)\n"
 706                 "open_buckets_wait:      %s\n",
 707                 fifo_used(&ca->free_inc),               ca->free_inc.size,
 708                 fifo_used(&ca->free[RESERVE_PRIO]),     ca->free[RESERVE_PRIO].size,
 709                 fifo_used(&ca->free[RESERVE_BTREE]),    ca->free[RESERVE_BTREE].size,
 710                 fifo_used(&ca->free[RESERVE_MOVINGGC]), ca->free[RESERVE_MOVINGGC].size,
 711                 fifo_used(&ca->free[RESERVE_NONE]),     ca->free[RESERVE_NONE].size,
 712                 stats.buckets_alloc,                    ca->mi.nbuckets - ca->mi.first_bucket,
 713                 stats.buckets_meta,                     ca->mi.nbuckets - ca->mi.first_bucket,
 714                 stats.buckets_dirty,                    ca->mi.nbuckets - ca->mi.first_bucket,
 715                 __dev_buckets_available(ca, stats),     ca->mi.nbuckets - ca->mi.first_bucket,
 716                 c->freelist_wait.list.first             ? "waiting" : "empty",
 717                 c->open_buckets_nr_free, OPEN_BUCKETS_COUNT, BTREE_NODE_RESERVE,
 718                 c->open_buckets_wait.list.first         ? "waiting" : "empty");
 719 }
 720
 721 static u64 sectors_written(struct bch_dev *ca)
 722 {
 723         u64 ret = 0;
 724         int cpu;
 725
 726         for_each_possible_cpu(cpu)
 727                 ret += *per_cpu_ptr(ca->sectors_written, cpu);
 728
 729         return ret;
 730 }
 731
 732 SHOW(bch2_dev)
 733 {
 734         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
 735         struct bch_fs *c = ca->fs;
 736         struct bch_dev_usage stats = bch2_dev_usage_read(ca);
 737
 738         sysfs_printf(uuid,              "%pU\n", ca->uuid.b);
 739
 740         sysfs_print(bucket_size,        bucket_bytes(ca));
 741         sysfs_print(block_size,         block_bytes(c));
 742         sysfs_print(first_bucket,       ca->mi.first_bucket);
 743         sysfs_print(nbuckets,           ca->mi.nbuckets);
 744         sysfs_print(discard,            ca->mi.discard);
 745         sysfs_hprint(written, sectors_written(ca) << 9);
 746         sysfs_hprint(btree_written,
 747                      atomic64_read(&ca->btree_sectors_written) << 9);
 748         sysfs_hprint(metadata_written,
 749                      (atomic64_read(&ca->meta_sectors_written) +
 750                       atomic64_read(&ca->btree_sectors_written)) << 9);
 751
 752         sysfs_hprint(dirty_data,        stats.sectors[S_DIRTY] << 9);
 753         sysfs_print(dirty_bytes,        stats.sectors[S_DIRTY] << 9);
 754         sysfs_print(dirty_buckets,      stats.buckets_dirty);
 755         sysfs_hprint(cached_data,       stats.sectors[S_CACHED] << 9);
 756         sysfs_print(cached_bytes,       stats.sectors[S_CACHED] << 9);
 757         sysfs_print(cached_buckets,     stats.buckets_cached);
 758         sysfs_print(meta_buckets,       stats.buckets_meta);
 759         sysfs_print(alloc_buckets,      stats.buckets_alloc);
 760         sysfs_print(available_buckets,  dev_buckets_available(ca));
 761         sysfs_print(free_buckets,       dev_buckets_free(ca));
 762         sysfs_print(has_data,           ca->mi.has_data);
 763         sysfs_print(has_metadata,       ca->mi.has_metadata);
 764
 765         sysfs_pd_controller_show(copy_gc, &ca->moving_gc_pd);
 766
 767         if (attr == &sysfs_cache_replacement_policy)
 768                 return bch2_snprint_string_list(buf, PAGE_SIZE,
 769                                                 bch2_cache_replacement_policies,
 770                                                 ca->mi.replacement);
 771
 772         sysfs_print(tier,               ca->mi.tier);
 773
 774         if (attr == &sysfs_state_rw)
 775                 return bch2_snprint_string_list(buf, PAGE_SIZE,
 776                                                 bch2_dev_state,
 777                                                 ca->mi.state);
 778
 779         if (attr == &sysfs_read_priority_stats)
 780                 return show_quantiles(ca, buf, bucket_priority_fn, (void *) 0);
 781         if (attr == &sysfs_write_priority_stats)
 782                 return show_quantiles(ca, buf, bucket_priority_fn, (void *) 1);
 783         if (attr == &sysfs_fragmentation_stats)
 784                 return show_quantiles(ca, buf, bucket_sectors_used_fn, NULL);
 785         if (attr == &sysfs_oldest_gen_stats)
 786                 return show_quantiles(ca, buf, bucket_oldest_gen_fn, NULL);
 787         if (attr == &sysfs_reserve_stats)
 788                 return show_reserve_stats(ca, buf);
 789         if (attr == &sysfs_alloc_debug)
 790                 return show_dev_alloc_debug(ca, buf);
 791
 792         return 0;
 793 }
 794
 795 STORE(bch2_dev)
 796 {
 797         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
 798         struct bch_fs *c = ca->fs;
 799         struct bch_member *mi;
 800
 801         sysfs_pd_controller_store(copy_gc, &ca->moving_gc_pd);
 802
 803         if (attr == &sysfs_discard) {
 804                 bool v = strtoul_or_return(buf);
 805
 806                 mutex_lock(&c->sb_lock);
 807                 mi = &bch2_sb_get_members(c->disk_sb)->members[ca->dev_idx];
 808
 809                 if (v != BCH_MEMBER_DISCARD(mi)) {
 810                         SET_BCH_MEMBER_DISCARD(mi, v);
 811                         bch2_write_super(c);
 812                 }
 813                 mutex_unlock(&c->sb_lock);
 814         }
 815
 816         if (attr == &sysfs_cache_replacement_policy) {
 817                 ssize_t v = bch2_read_string_list(buf, bch2_cache_replacement_policies);
 818
 819                 if (v < 0)
 820                         return v;
 821
 822                 mutex_lock(&c->sb_lock);
 823                 mi = &bch2_sb_get_members(c->disk_sb)->members[ca->dev_idx];
 824
 825                 if ((unsigned) v != BCH_MEMBER_REPLACEMENT(mi)) {
 826                         SET_BCH_MEMBER_REPLACEMENT(mi, v);
 827                         bch2_write_super(c);
 828                 }
 829                 mutex_unlock(&c->sb_lock);
 830         }
 831
 832         if (attr == &sysfs_tier) {
 833                 unsigned prev_tier;
 834                 unsigned v = strtoul_restrict_or_return(buf,
 835                                         0, BCH_TIER_MAX - 1);
 836
 837                 mutex_lock(&c->sb_lock);
 838                 prev_tier = ca->mi.tier;
 839
 840                 if (v == ca->mi.tier) {
 841                         mutex_unlock(&c->sb_lock);
 842                         return size;
 843                 }
 844
 845                 mi = &bch2_sb_get_members(c->disk_sb)->members[ca->dev_idx];
 846                 SET_BCH_MEMBER_TIER(mi, v);
 847                 bch2_write_super(c);
 848
 849                 bch2_dev_group_remove(&c->tiers[prev_tier].devs, ca);
 850                 bch2_dev_group_add(&c->tiers[ca->mi.tier].devs, ca);
 851                 mutex_unlock(&c->sb_lock);
 852
 853                 bch2_recalc_capacity(c);
 854                 bch2_tiering_start(c);
 855         }
 856
 857         return size;
 858 }
 859 SYSFS_OPS(bch2_dev);
 860
 861 struct attribute *bch2_dev_files[] = {
 862         &sysfs_uuid,
 863         &sysfs_bucket_size,
 864         &sysfs_block_size,
 865         &sysfs_first_bucket,
 866         &sysfs_nbuckets,
 867
 868         /* settings: */
 869         &sysfs_discard,
 870         &sysfs_cache_replacement_policy,
 871         &sysfs_tier,
 872         &sysfs_state_rw,
 873
 874         &sysfs_has_data,
 875         &sysfs_has_metadata,
 876
 877         /* io stats: */
 878         &sysfs_written,
 879         &sysfs_btree_written,
 880         &sysfs_metadata_written,
 881
 882         /* alloc info - data: */
 883         &sysfs_dirty_data,
 884         &sysfs_dirty_bytes,
 885         &sysfs_cached_data,
 886         &sysfs_cached_bytes,
 887
 888         /* alloc info - buckets: */
 889         &sysfs_available_buckets,
 890         &sysfs_free_buckets,
 891         &sysfs_dirty_buckets,
 892         &sysfs_cached_buckets,
 893         &sysfs_meta_buckets,
 894         &sysfs_alloc_buckets,
 895
 896         /* alloc info - other stats: */
 897         &sysfs_read_priority_stats,
 898         &sysfs_write_priority_stats,
 899         &sysfs_fragmentation_stats,
 900         &sysfs_oldest_gen_stats,
 901         &sysfs_reserve_stats,
 902
 903         /* debug: */
 904         &sysfs_alloc_debug,
 905
 906         sysfs_pd_controller_files(copy_gc),
 907         NULL
 908 };