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 #ifndef NO_BCACHEFS_SYSFS
   9
  10 #include "bcachefs.h"
  11 #include "alloc.h"
  12 #include "compress.h"
  13 #include "sysfs.h"
  14 #include "btree_cache.h"
  15 #include "btree_iter.h"
  16 #include "btree_update.h"
  17 #include "btree_gc.h"
  18 #include "buckets.h"
  19 #include "inode.h"
  20 #include "journal.h"
  21 #include "keylist.h"
  22 #include "move.h"
  23 #include "opts.h"
  24 #include "super-io.h"
  25 #include "tier.h"
  26
  27 #include <linux/blkdev.h>
  28 #include <linux/sort.h>
  29 #include <linux/sched/clock.h>
  30
  31 #include "util.h"
  32
  33 #define SYSFS_OPS(type)                                                 \
  34 struct sysfs_ops type ## _sysfs_ops = {                                 \
  35         .show   = type ## _show,                                        \
  36         .store  = type ## _store                                        \
  37 }
  38
  39 #define SHOW(fn)                                                        \
  40 static ssize_t fn ## _show(struct kobject *kobj, struct attribute *attr,\
  41                            char *buf)                                   \
  42
  43 #define STORE(fn)                                                       \
  44 static ssize_t fn ## _store(struct kobject *kobj, struct attribute *attr,\
  45                             const char *buf, size_t size)               \
  46
  47 #define __sysfs_attribute(_name, _mode)                                 \
  48         static struct attribute sysfs_##_name =                         \
  49                 { .name = #_name, .mode = _mode }
  50
  51 #define write_attribute(n)      __sysfs_attribute(n, S_IWUSR)
  52 #define read_attribute(n)       __sysfs_attribute(n, S_IRUGO)
  53 #define rw_attribute(n)         __sysfs_attribute(n, S_IRUGO|S_IWUSR)
  54
  55 #define sysfs_printf(file, fmt, ...)                                    \
  56 do {                                                                    \
  57         if (attr == &sysfs_ ## file)                                    \
  58                 return scnprintf(buf, PAGE_SIZE, fmt "\n", __VA_ARGS__);\
  59 } while (0)
  60
  61 #define sysfs_print(file, var)                                          \
  62 do {                                                                    \
  63         if (attr == &sysfs_ ## file)                                    \
  64                 return snprint(buf, PAGE_SIZE, var);                    \
  65 } while (0)
  66
  67 #define sysfs_hprint(file, val)                                         \
  68 do {                                                                    \
  69         if (attr == &sysfs_ ## file) {                                  \
  70                 ssize_t ret = bch2_hprint(buf, val);                    \
  71                 strcat(buf, "\n");                                      \
  72                 return ret + 1;                                         \
  73         }                                                               \
  74 } while (0)
  75
  76 #define var_printf(_var, fmt)   sysfs_printf(_var, fmt, var(_var))
  77 #define var_print(_var)         sysfs_print(_var, var(_var))
  78 #define var_hprint(_var)        sysfs_hprint(_var, var(_var))
  79
  80 #define sysfs_strtoul(file, var)                                        \
  81 do {                                                                    \
  82         if (attr == &sysfs_ ## file)                                    \
  83                 return strtoul_safe(buf, var) ?: (ssize_t) size;        \
  84 } while (0)
  85
  86 #define sysfs_strtoul_clamp(file, var, min, max)                        \
  87 do {                                                                    \
  88         if (attr == &sysfs_ ## file)                                    \
  89                 return strtoul_safe_clamp(buf, var, min, max)           \
  90                         ?: (ssize_t) size;                              \
  91 } while (0)
  92
  93 #define strtoul_or_return(cp)                                           \
  94 ({                                                                      \
  95         unsigned long _v;                                               \
  96         int _r = kstrtoul(cp, 10, &_v);                                 \
  97         if (_r)                                                         \
  98                 return _r;                                              \
  99         _v;                                                             \
 100 })
 101
 102 #define strtoul_restrict_or_return(cp, min, max)                        \
 103 ({                                                                      \
 104         unsigned long __v = 0;                                          \
 105         int _r = strtoul_safe_restrict(cp, __v, min, max);              \
 106         if (_r)                                                         \
 107                 return _r;                                              \
 108         __v;                                                            \
 109 })
 110
 111 #define strtoi_h_or_return(cp)                                          \
 112 ({                                                                      \
 113         u64 _v;                                                         \
 114         int _r = strtoi_h(cp, &_v);                                     \
 115         if (_r)                                                         \
 116                 return _r;                                              \
 117         _v;                                                             \
 118 })
 119
 120 #define sysfs_hatoi(file, var)                                          \
 121 do {                                                                    \
 122         if (attr == &sysfs_ ## file)                                    \
 123                 return strtoi_h(buf, &var) ?: (ssize_t) size;           \
 124 } while (0)
 125
 126 write_attribute(trigger_journal_flush);
 127 write_attribute(trigger_btree_coalesce);
 128 write_attribute(trigger_gc);
 129 write_attribute(prune_cache);
 130 rw_attribute(btree_gc_periodic);
 131
 132 read_attribute(uuid);
 133 read_attribute(minor);
 134 read_attribute(bucket_size);
 135 read_attribute(block_size);
 136 read_attribute(btree_node_size);
 137 read_attribute(first_bucket);
 138 read_attribute(nbuckets);
 139 read_attribute(iostats);
 140 read_attribute(read_priority_stats);
 141 read_attribute(write_priority_stats);
 142 read_attribute(fragmentation_stats);
 143 read_attribute(oldest_gen_stats);
 144 read_attribute(reserve_stats);
 145 read_attribute(btree_cache_size);
 146 read_attribute(compression_stats);
 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(alloc_debug);
 164
 165 read_attribute(read_realloc_races);
 166
 167 rw_attribute(journal_write_delay_ms);
 168 rw_attribute(journal_reclaim_delay_ms);
 169
 170 rw_attribute(discard);
 171 rw_attribute(cache_replacement_policy);
 172
 173 rw_attribute(foreground_write_ratelimit_enabled);
 174 rw_attribute(copy_gc_enabled);
 175 sysfs_pd_controller_attribute(copy_gc);
 176
 177 rw_attribute(tier);
 178 rw_attribute(tiering_enabled);
 179 rw_attribute(tiering_percent);
 180 sysfs_pd_controller_attribute(tiering);
 181
 182 sysfs_pd_controller_attribute(foreground_write);
 183
 184 rw_attribute(pd_controllers_update_seconds);
 185
 186 rw_attribute(foreground_target_percent);
 187
 188 read_attribute(meta_replicas_have);
 189 read_attribute(data_replicas_have);
 190
 191 #define BCH_DEBUG_PARAM(name, description)                              \
 192         rw_attribute(name);
 193
 194         BCH_DEBUG_PARAMS()
 195 #undef BCH_DEBUG_PARAM
 196
 197 #define BCH_OPT(_name, _mode, ...)                                      \
 198         static struct attribute sysfs_opt_##_name = {                   \
 199                 .name = #_name, .mode = _mode,                          \
 200         };
 201
 202         BCH_VISIBLE_OPTS()
 203 #undef BCH_OPT
 204
 205 #define BCH_TIME_STAT(name, frequency_units, duration_units)            \
 206         sysfs_time_stats_attribute(name, frequency_units, duration_units);
 207         BCH_TIME_STATS()
 208 #undef BCH_TIME_STAT
 209
 210 static struct attribute sysfs_state_rw = {
 211         .name = "state",
 212         .mode = S_IRUGO
 213 };
 214
 215 static size_t bch2_btree_cache_size(struct bch_fs *c)
 216 {
 217         size_t ret = 0;
 218         struct btree *b;
 219
 220         mutex_lock(&c->btree_cache_lock);
 221         list_for_each_entry(b, &c->btree_cache, list)
 222                 ret += btree_bytes(c);
 223
 224         mutex_unlock(&c->btree_cache_lock);
 225         return ret;
 226 }
 227
 228 static ssize_t show_fs_alloc_debug(struct bch_fs *c, char *buf)
 229 {
 230         struct bch_fs_usage stats = bch2_fs_usage_read(c);
 231
 232         return scnprintf(buf, PAGE_SIZE,
 233                          "capacity:\t\t%llu\n"
 234                          "1 replicas:\n"
 235                          "\tmeta:\t\t%llu\n"
 236                          "\tdirty:\t\t%llu\n"
 237                          "\treserved:\t%llu\n"
 238                          "2 replicas:\n"
 239                          "\tmeta:\t\t%llu\n"
 240                          "\tdirty:\t\t%llu\n"
 241                          "\treserved:\t%llu\n"
 242                          "3 replicas:\n"
 243                          "\tmeta:\t\t%llu\n"
 244                          "\tdirty:\t\t%llu\n"
 245                          "\treserved:\t%llu\n"
 246                          "4 replicas:\n"
 247                          "\tmeta:\t\t%llu\n"
 248                          "\tdirty:\t\t%llu\n"
 249                          "\treserved:\t%llu\n"
 250                          "online reserved:\t%llu\n",
 251                          c->capacity,
 252                          stats.s[0].data[S_META],
 253                          stats.s[0].data[S_DIRTY],
 254                          stats.s[0].persistent_reserved,
 255                          stats.s[1].data[S_META],
 256                          stats.s[1].data[S_DIRTY],
 257                          stats.s[1].persistent_reserved,
 258                          stats.s[2].data[S_META],
 259                          stats.s[2].data[S_DIRTY],
 260                          stats.s[2].persistent_reserved,
 261                          stats.s[3].data[S_META],
 262                          stats.s[3].data[S_DIRTY],
 263                          stats.s[3].persistent_reserved,
 264                          stats.online_reserved);
 265 }
 266
 267 static ssize_t bch2_compression_stats(struct bch_fs *c, char *buf)
 268 {
 269         struct btree_iter iter;
 270         struct bkey_s_c k;
 271         u64 nr_uncompressed_extents = 0, uncompressed_sectors = 0,
 272             nr_compressed_extents = 0,
 273             compressed_sectors_compressed = 0,
 274             compressed_sectors_uncompressed = 0;
 275
 276         if (!bch2_fs_running(c))
 277                 return -EPERM;
 278
 279         for_each_btree_key(&iter, c, BTREE_ID_EXTENTS, POS_MIN, 0, k)
 280                 if (k.k->type == BCH_EXTENT) {
 281                         struct bkey_s_c_extent e = bkey_s_c_to_extent(k);
 282                         const struct bch_extent_ptr *ptr;
 283                         const union bch_extent_crc *crc;
 284
 285                         extent_for_each_ptr_crc(e, ptr, crc) {
 286                                 if (crc_compression_type(crc) == BCH_COMPRESSION_NONE) {
 287                                         nr_uncompressed_extents++;
 288                                         uncompressed_sectors += e.k->size;
 289                                 } else {
 290                                         nr_compressed_extents++;
 291                                         compressed_sectors_compressed +=
 292                                                 crc_compressed_size(e.k, crc);
 293                                         compressed_sectors_uncompressed +=
 294                                                 crc_uncompressed_size(e.k, crc);
 295                                 }
 296
 297                                 /* only looking at the first ptr */
 298                                 break;
 299                         }
 300                 }
 301         bch2_btree_iter_unlock(&iter);
 302
 303         return scnprintf(buf, PAGE_SIZE,
 304                         "uncompressed data:\n"
 305                         "       nr extents:                     %llu\n"
 306                         "       size (bytes):                   %llu\n"
 307                         "compressed data:\n"
 308                         "       nr extents:                     %llu\n"
 309                         "       compressed size (bytes):        %llu\n"
 310                         "       uncompressed size (bytes):      %llu\n",
 311                         nr_uncompressed_extents,
 312                         uncompressed_sectors << 9,
 313                         nr_compressed_extents,
 314                         compressed_sectors_compressed << 9,
 315                         compressed_sectors_uncompressed << 9);
 316 }
 317
 318 SHOW(bch2_fs)
 319 {
 320         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 321
 322         sysfs_print(minor,                      c->minor);
 323         sysfs_printf(internal_uuid, "%pU",      c->sb.uuid.b);
 324
 325         sysfs_print(journal_write_delay_ms,     c->journal.write_delay_ms);
 326         sysfs_print(journal_reclaim_delay_ms,   c->journal.reclaim_delay_ms);
 327
 328         sysfs_print(block_size,                 block_bytes(c));
 329         sysfs_print(btree_node_size,            btree_bytes(c));
 330         sysfs_hprint(btree_cache_size,          bch2_btree_cache_size(c));
 331
 332         sysfs_print(read_realloc_races,
 333                     atomic_long_read(&c->read_realloc_races));
 334
 335         sysfs_printf(btree_gc_periodic, "%u",   (int) c->btree_gc_periodic);
 336
 337         sysfs_printf(foreground_write_ratelimit_enabled, "%i",
 338                      c->foreground_write_ratelimit_enabled);
 339         sysfs_printf(copy_gc_enabled, "%i", c->copy_gc_enabled);
 340         sysfs_pd_controller_show(foreground_write, &c->foreground_write_pd);
 341
 342         sysfs_print(pd_controllers_update_seconds,
 343                     c->pd_controllers_update_seconds);
 344         sysfs_print(foreground_target_percent, c->foreground_target_percent);
 345
 346         sysfs_printf(tiering_enabled,           "%i", c->tiering_enabled);
 347         sysfs_print(tiering_percent,            c->tiering_percent);
 348
 349         sysfs_pd_controller_show(tiering,       &c->tiers[1].pd); /* XXX */
 350
 351         sysfs_printf(meta_replicas_have, "%u",  bch2_replicas_online(c, true));
 352         sysfs_printf(data_replicas_have, "%u",  bch2_replicas_online(c, false));
 353
 354         /* Debugging: */
 355
 356         if (attr == &sysfs_alloc_debug)
 357                 return show_fs_alloc_debug(c, buf);
 358
 359         if (attr == &sysfs_journal_debug)
 360                 return bch2_journal_print_debug(&c->journal, buf);
 361
 362         if (attr == &sysfs_journal_pins)
 363                 return bch2_journal_print_pins(&c->journal, buf);
 364
 365         if (attr == &sysfs_compression_stats)
 366                 return bch2_compression_stats(c, buf);
 367
 368 #define BCH_DEBUG_PARAM(name, description) sysfs_print(name, c->name);
 369         BCH_DEBUG_PARAMS()
 370 #undef BCH_DEBUG_PARAM
 371
 372         return 0;
 373 }
 374
 375 STORE(__bch2_fs)
 376 {
 377         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 378
 379         sysfs_strtoul(journal_write_delay_ms, c->journal.write_delay_ms);
 380         sysfs_strtoul(journal_reclaim_delay_ms, c->journal.reclaim_delay_ms);
 381
 382         sysfs_strtoul(foreground_write_ratelimit_enabled,
 383                       c->foreground_write_ratelimit_enabled);
 384
 385         if (attr == &sysfs_btree_gc_periodic) {
 386                 ssize_t ret = strtoul_safe(buf, c->btree_gc_periodic)
 387                         ?: (ssize_t) size;
 388
 389                 wake_up_process(c->gc_thread);
 390                 return ret;
 391         }
 392
 393         if (attr == &sysfs_copy_gc_enabled) {
 394                 struct bch_dev *ca;
 395                 unsigned i;
 396                 ssize_t ret = strtoul_safe(buf, c->copy_gc_enabled)
 397                         ?: (ssize_t) size;
 398
 399                 for_each_member_device(ca, c, i)
 400                         if (ca->moving_gc_read)
 401                                 wake_up_process(ca->moving_gc_read);
 402                 return ret;
 403         }
 404
 405         if (attr == &sysfs_tiering_enabled) {
 406                 ssize_t ret = strtoul_safe(buf, c->tiering_enabled)
 407                         ?: (ssize_t) size;
 408
 409                 bch2_tiering_start(c); /* issue wakeups */
 410                 return ret;
 411         }
 412
 413         sysfs_pd_controller_store(foreground_write, &c->foreground_write_pd);
 414
 415         sysfs_strtoul(pd_controllers_update_seconds,
 416                       c->pd_controllers_update_seconds);
 417         sysfs_strtoul(foreground_target_percent, c->foreground_target_percent);
 418
 419         sysfs_strtoul(tiering_percent,          c->tiering_percent);
 420         sysfs_pd_controller_store(tiering,      &c->tiers[1].pd); /* XXX */
 421
 422         /* Debugging: */
 423
 424 #define BCH_DEBUG_PARAM(name, description) sysfs_strtoul(name, c->name);
 425         BCH_DEBUG_PARAMS()
 426 #undef BCH_DEBUG_PARAM
 427
 428         if (!bch2_fs_running(c))
 429                 return -EPERM;
 430
 431         /* Debugging: */
 432
 433         if (attr == &sysfs_trigger_journal_flush)
 434                 bch2_journal_meta_async(&c->journal, NULL);
 435
 436         if (attr == &sysfs_trigger_btree_coalesce)
 437                 bch2_coalesce(c);
 438
 439         if (attr == &sysfs_trigger_gc)
 440                 bch2_gc(c);
 441
 442         if (attr == &sysfs_prune_cache) {
 443                 struct shrink_control sc;
 444
 445                 sc.gfp_mask = GFP_KERNEL;
 446                 sc.nr_to_scan = strtoul_or_return(buf);
 447                 c->btree_cache_shrink.scan_objects(&c->btree_cache_shrink, &sc);
 448         }
 449
 450         return size;
 451 }
 452
 453 STORE(bch2_fs)
 454 {
 455         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 456
 457         mutex_lock(&c->state_lock);
 458         size = __bch2_fs_store(kobj, attr, buf, size);
 459         mutex_unlock(&c->state_lock);
 460
 461         return size;
 462 }
 463 SYSFS_OPS(bch2_fs);
 464
 465 struct attribute *bch2_fs_files[] = {
 466         &sysfs_minor,
 467         &sysfs_block_size,
 468         &sysfs_btree_node_size,
 469         &sysfs_btree_cache_size,
 470
 471         &sysfs_meta_replicas_have,
 472         &sysfs_data_replicas_have,
 473
 474         &sysfs_journal_write_delay_ms,
 475         &sysfs_journal_reclaim_delay_ms,
 476
 477         &sysfs_foreground_target_percent,
 478         &sysfs_tiering_percent,
 479
 480         &sysfs_compression_stats,
 481         NULL
 482 };
 483
 484 /* internal dir - just a wrapper */
 485
 486 SHOW(bch2_fs_internal)
 487 {
 488         struct bch_fs *c = container_of(kobj, struct bch_fs, internal);
 489         return bch2_fs_show(&c->kobj, attr, buf);
 490 }
 491
 492 STORE(bch2_fs_internal)
 493 {
 494         struct bch_fs *c = container_of(kobj, struct bch_fs, internal);
 495         return bch2_fs_store(&c->kobj, attr, buf, size);
 496 }
 497 SYSFS_OPS(bch2_fs_internal);
 498
 499 struct attribute *bch2_fs_internal_files[] = {
 500         &sysfs_alloc_debug,
 501         &sysfs_journal_debug,
 502         &sysfs_journal_pins,
 503
 504         &sysfs_read_realloc_races,
 505
 506         &sysfs_trigger_journal_flush,
 507         &sysfs_trigger_btree_coalesce,
 508         &sysfs_trigger_gc,
 509         &sysfs_prune_cache,
 510
 511         &sysfs_foreground_write_ratelimit_enabled,
 512         &sysfs_copy_gc_enabled,
 513         &sysfs_tiering_enabled,
 514         sysfs_pd_controller_files(tiering),
 515         sysfs_pd_controller_files(foreground_write),
 516         &sysfs_internal_uuid,
 517
 518 #define BCH_DEBUG_PARAM(name, description) &sysfs_##name,
 519         BCH_DEBUG_PARAMS()
 520 #undef BCH_DEBUG_PARAM
 521
 522         NULL
 523 };
 524
 525 /* options */
 526
 527 SHOW(bch2_fs_opts_dir)
 528 {
 529         char *out = buf, *end = buf + PAGE_SIZE;
 530         struct bch_fs *c = container_of(kobj, struct bch_fs, opts_dir);
 531
 532         out += bch2_opt_show(&c->opts, attr->name, out, end - out);
 533         out += scnprintf(out, end - out, "\n");
 534
 535         return out - buf;
 536 }
 537
 538 STORE(bch2_fs_opts_dir)
 539 {
 540         struct bch_fs *c = container_of(kobj, struct bch_fs, opts_dir);
 541         const struct bch_option *opt;
 542         int id;
 543         u64 v;
 544
 545         id = bch2_parse_sysfs_opt(attr->name, buf, &v);
 546         if (id < 0)
 547                 return id;
 548
 549         opt = &bch2_opt_table[id];
 550
 551         mutex_lock(&c->sb_lock);
 552
 553         if (id == Opt_compression) {
 554                 int ret = bch2_check_set_has_compressed_data(c, v);
 555                 if (ret) {
 556                         mutex_unlock(&c->sb_lock);
 557                         return ret;
 558                 }
 559         }
 560
 561         if (opt->set_sb != SET_NO_SB_OPT) {
 562                 opt->set_sb(c->disk_sb, v);
 563                 bch2_write_super(c);
 564         }
 565
 566         bch2_opt_set(&c->opts, id, v);
 567
 568         mutex_unlock(&c->sb_lock);
 569
 570         return size;
 571 }
 572 SYSFS_OPS(bch2_fs_opts_dir);
 573
 574 struct attribute *bch2_fs_opts_dir_files[] = {
 575 #define BCH_OPT(_name, ...)                                             \
 576         &sysfs_opt_##_name,
 577
 578         BCH_VISIBLE_OPTS()
 579 #undef BCH_OPT
 580
 581         NULL
 582 };
 583
 584 /* time stats */
 585
 586 SHOW(bch2_fs_time_stats)
 587 {
 588         struct bch_fs *c = container_of(kobj, struct bch_fs, time_stats);
 589
 590 #define BCH_TIME_STAT(name, frequency_units, duration_units)            \
 591         sysfs_print_time_stats(&c->name##_time, name,                   \
 592                                frequency_units, duration_units);
 593         BCH_TIME_STATS()
 594 #undef BCH_TIME_STAT
 595
 596         return 0;
 597 }
 598
 599 STORE(bch2_fs_time_stats)
 600 {
 601         struct bch_fs *c = container_of(kobj, struct bch_fs, time_stats);
 602
 603 #define BCH_TIME_STAT(name, frequency_units, duration_units)            \
 604         sysfs_clear_time_stats(&c->name##_time, name);
 605         BCH_TIME_STATS()
 606 #undef BCH_TIME_STAT
 607
 608         return size;
 609 }
 610 SYSFS_OPS(bch2_fs_time_stats);
 611
 612 struct attribute *bch2_fs_time_stats_files[] = {
 613 #define BCH_TIME_STAT(name, frequency_units, duration_units)            \
 614         sysfs_time_stats_attribute_list(name, frequency_units, duration_units)
 615         BCH_TIME_STATS()
 616 #undef BCH_TIME_STAT
 617
 618         NULL
 619 };
 620
 621 typedef unsigned (bucket_map_fn)(struct bch_dev *, struct bucket *, void *);
 622
 623 static unsigned bucket_priority_fn(struct bch_dev *ca, struct bucket *g,
 624                                    void *private)
 625 {
 626         int rw = (private ? 1 : 0);
 627
 628         return ca->fs->prio_clock[rw].hand - g->prio[rw];
 629 }
 630
 631 static unsigned bucket_sectors_used_fn(struct bch_dev *ca, struct bucket *g,
 632                                        void *private)
 633 {
 634         return bucket_sectors_used(g->mark);
 635 }
 636
 637 static unsigned bucket_oldest_gen_fn(struct bch_dev *ca, struct bucket *g,
 638                                      void *private)
 639 {
 640         return bucket_gc_gen(ca, g);
 641 }
 642
 643 static ssize_t show_quantiles(struct bch_dev *ca, char *buf,
 644                               bucket_map_fn *fn, void *private)
 645 {
 646         int cmp(const void *l, const void *r)
 647         {       return *((unsigned *) r) - *((unsigned *) l); }
 648
 649         size_t n = ca->mi.nbuckets, i;
 650         /* Compute 31 quantiles */
 651         unsigned q[31], *p;
 652         ssize_t ret = 0;
 653
 654         p = vzalloc(ca->mi.nbuckets * sizeof(unsigned));
 655         if (!p)
 656                 return -ENOMEM;
 657
 658         for (i = ca->mi.first_bucket; i < n; i++)
 659                 p[i] = fn(ca, &ca->buckets[i], private);
 660
 661         sort(p, n, sizeof(unsigned), cmp, NULL);
 662
 663         while (n &&
 664                !p[n - 1])
 665                 --n;
 666
 667         for (i = 0; i < ARRAY_SIZE(q); i++)
 668                 q[i] = p[n * (i + 1) / (ARRAY_SIZE(q) + 1)];
 669
 670         vfree(p);
 671
 672         for (i = 0; i < ARRAY_SIZE(q); i++)
 673                 ret += scnprintf(buf + ret, PAGE_SIZE - ret,
 674                                  "%u ", q[i]);
 675         buf[ret - 1] = '\n';
 676
 677         return ret;
 678
 679 }
 680
 681 static ssize_t show_reserve_stats(struct bch_dev *ca, char *buf)
 682 {
 683         enum alloc_reserve i;
 684         ssize_t ret;
 685
 686         spin_lock(&ca->freelist_lock);
 687
 688         ret = scnprintf(buf, PAGE_SIZE,
 689                         "free_inc:\t%zu\t%zu\n",
 690                         fifo_used(&ca->free_inc),
 691                         ca->free_inc.size);
 692
 693         for (i = 0; i < RESERVE_NR; i++)
 694                 ret += scnprintf(buf + ret, PAGE_SIZE - ret,
 695                                  "free[%u]:\t%zu\t%zu\n", i,
 696                                  fifo_used(&ca->free[i]),
 697                                  ca->free[i].size);
 698
 699         spin_unlock(&ca->freelist_lock);
 700
 701         return ret;
 702 }
 703
 704 static ssize_t show_dev_alloc_debug(struct bch_dev *ca, char *buf)
 705 {
 706         struct bch_fs *c = ca->fs;
 707         struct bch_dev_usage stats = bch2_dev_usage_read(ca);
 708
 709         return scnprintf(buf, PAGE_SIZE,
 710                 "free_inc:               %zu/%zu\n"
 711                 "free[RESERVE_BTREE]:    %zu/%zu\n"
 712                 "free[RESERVE_MOVINGGC]: %zu/%zu\n"
 713                 "free[RESERVE_NONE]:     %zu/%zu\n"
 714                 "alloc:                  %llu/%llu\n"
 715                 "meta:                   %llu/%llu\n"
 716                 "dirty:                  %llu/%llu\n"
 717                 "available:              %llu/%llu\n"
 718                 "freelist_wait:          %s\n"
 719                 "open buckets:           %u/%u (reserved %u)\n"
 720                 "open_buckets_wait:      %s\n",
 721                 fifo_used(&ca->free_inc),               ca->free_inc.size,
 722                 fifo_used(&ca->free[RESERVE_BTREE]),    ca->free[RESERVE_BTREE].size,
 723                 fifo_used(&ca->free[RESERVE_MOVINGGC]), ca->free[RESERVE_MOVINGGC].size,
 724                 fifo_used(&ca->free[RESERVE_NONE]),     ca->free[RESERVE_NONE].size,
 725                 stats.buckets_alloc,                    ca->mi.nbuckets - ca->mi.first_bucket,
 726                 stats.buckets[S_META],                  ca->mi.nbuckets - ca->mi.first_bucket,
 727                 stats.buckets[S_DIRTY],                 ca->mi.nbuckets - ca->mi.first_bucket,
 728                 __dev_buckets_available(ca, stats),     ca->mi.nbuckets - ca->mi.first_bucket,
 729                 c->freelist_wait.list.first             ? "waiting" : "empty",
 730                 c->open_buckets_nr_free, OPEN_BUCKETS_COUNT, BTREE_NODE_RESERVE,
 731                 c->open_buckets_wait.list.first         ? "waiting" : "empty");
 732 }
 733
 734 const char * const bch2_rw[] = {
 735         "read",
 736         "write",
 737         NULL
 738 };
 739
 740 static ssize_t show_dev_iostats(struct bch_dev *ca, char *buf)
 741 {
 742         char *out = buf, *end = buf + PAGE_SIZE;
 743         int rw, i, cpu;
 744
 745         for (rw = 0; rw < 2; rw++) {
 746                 out += scnprintf(out, end - out, "%s:\n", bch2_rw[rw]);
 747
 748                 for (i = 1; i < BCH_DATA_NR; i++) {
 749                         u64 n = 0;
 750
 751                         for_each_possible_cpu(cpu)
 752                                 n += per_cpu_ptr(ca->io_done, cpu)->sectors[rw][i];
 753
 754                         out += scnprintf(out, end - out, "%-12s:%12llu\n",
 755                                          bch2_data_types[i], n << 9);
 756                 }
 757         }
 758
 759         return out - buf;
 760 }
 761
 762 SHOW(bch2_dev)
 763 {
 764         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
 765         struct bch_fs *c = ca->fs;
 766         struct bch_dev_usage stats = bch2_dev_usage_read(ca);
 767         char *out = buf, *end = buf + PAGE_SIZE;
 768
 769         sysfs_printf(uuid,              "%pU\n", ca->uuid.b);
 770
 771         sysfs_print(bucket_size,        bucket_bytes(ca));
 772         sysfs_print(block_size,         block_bytes(c));
 773         sysfs_print(first_bucket,       ca->mi.first_bucket);
 774         sysfs_print(nbuckets,           ca->mi.nbuckets);
 775         sysfs_print(discard,            ca->mi.discard);
 776
 777         sysfs_hprint(dirty_data,        stats.sectors[S_DIRTY] << 9);
 778         sysfs_print(dirty_bytes,        stats.sectors[S_DIRTY] << 9);
 779         sysfs_print(dirty_buckets,      stats.buckets[S_DIRTY]);
 780         sysfs_hprint(cached_data,       stats.sectors_cached << 9);
 781         sysfs_print(cached_bytes,       stats.sectors_cached << 9);
 782         sysfs_print(cached_buckets,     stats.buckets_cached);
 783         sysfs_print(meta_buckets,       stats.buckets[S_META]);
 784         sysfs_print(alloc_buckets,      stats.buckets_alloc);
 785         sysfs_print(available_buckets,  dev_buckets_available(ca));
 786         sysfs_print(free_buckets,       dev_buckets_free(ca));
 787
 788         if (attr == &sysfs_has_data) {
 789                 out += bch2_scnprint_flag_list(out, end - out,
 790                                                bch2_data_types,
 791                                                bch2_dev_has_data(c, ca));
 792                 out += scnprintf(out, end - out, "\n");
 793                 return out - buf;
 794         }
 795
 796         sysfs_pd_controller_show(copy_gc, &ca->moving_gc_pd);
 797
 798         if (attr == &sysfs_cache_replacement_policy) {
 799                 out += bch2_scnprint_string_list(out, end - out,
 800                                                  bch2_cache_replacement_policies,
 801                                                  ca->mi.replacement);
 802                 out += scnprintf(out, end - out, "\n");
 803                 return out - buf;
 804         }
 805
 806         sysfs_print(tier,               ca->mi.tier);
 807
 808         if (attr == &sysfs_state_rw) {
 809                 out += bch2_scnprint_string_list(out, end - out,
 810                                                  bch2_dev_state,
 811                                                  ca->mi.state);
 812                 out += scnprintf(out, end - out, "\n");
 813                 return out - buf;
 814         }
 815
 816         if (attr == &sysfs_iostats)
 817                 return show_dev_iostats(ca, buf);
 818         if (attr == &sysfs_read_priority_stats)
 819                 return show_quantiles(ca, buf, bucket_priority_fn, (void *) 0);
 820         if (attr == &sysfs_write_priority_stats)
 821                 return show_quantiles(ca, buf, bucket_priority_fn, (void *) 1);
 822         if (attr == &sysfs_fragmentation_stats)
 823                 return show_quantiles(ca, buf, bucket_sectors_used_fn, NULL);
 824         if (attr == &sysfs_oldest_gen_stats)
 825                 return show_quantiles(ca, buf, bucket_oldest_gen_fn, NULL);
 826         if (attr == &sysfs_reserve_stats)
 827                 return show_reserve_stats(ca, buf);
 828         if (attr == &sysfs_alloc_debug)
 829                 return show_dev_alloc_debug(ca, buf);
 830
 831         return 0;
 832 }
 833
 834 STORE(bch2_dev)
 835 {
 836         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
 837         struct bch_fs *c = ca->fs;
 838         struct bch_member *mi;
 839
 840         sysfs_pd_controller_store(copy_gc, &ca->moving_gc_pd);
 841
 842         if (attr == &sysfs_discard) {
 843                 bool v = strtoul_or_return(buf);
 844
 845                 mutex_lock(&c->sb_lock);
 846                 mi = &bch2_sb_get_members(c->disk_sb)->members[ca->dev_idx];
 847
 848                 if (v != BCH_MEMBER_DISCARD(mi)) {
 849                         SET_BCH_MEMBER_DISCARD(mi, v);
 850                         bch2_write_super(c);
 851                 }
 852                 mutex_unlock(&c->sb_lock);
 853         }
 854
 855         if (attr == &sysfs_cache_replacement_policy) {
 856                 ssize_t v = bch2_read_string_list(buf, bch2_cache_replacement_policies);
 857
 858                 if (v < 0)
 859                         return v;
 860
 861                 mutex_lock(&c->sb_lock);
 862                 mi = &bch2_sb_get_members(c->disk_sb)->members[ca->dev_idx];
 863
 864                 if ((unsigned) v != BCH_MEMBER_REPLACEMENT(mi)) {
 865                         SET_BCH_MEMBER_REPLACEMENT(mi, v);
 866                         bch2_write_super(c);
 867                 }
 868                 mutex_unlock(&c->sb_lock);
 869         }
 870
 871         if (attr == &sysfs_tier) {
 872                 unsigned prev_tier;
 873                 unsigned v = strtoul_restrict_or_return(buf,
 874                                         0, BCH_TIER_MAX - 1);
 875
 876                 mutex_lock(&c->sb_lock);
 877                 prev_tier = ca->mi.tier;
 878
 879                 if (v == ca->mi.tier) {
 880                         mutex_unlock(&c->sb_lock);
 881                         return size;
 882                 }
 883
 884                 mi = &bch2_sb_get_members(c->disk_sb)->members[ca->dev_idx];
 885                 SET_BCH_MEMBER_TIER(mi, v);
 886                 bch2_write_super(c);
 887
 888                 clear_bit(ca->dev_idx, c->tiers[prev_tier].devs.d);
 889                 set_bit(ca->dev_idx, c->tiers[ca->mi.tier].devs.d);
 890                 mutex_unlock(&c->sb_lock);
 891
 892                 bch2_recalc_capacity(c);
 893                 bch2_tiering_start(c);
 894         }
 895
 896         return size;
 897 }
 898 SYSFS_OPS(bch2_dev);
 899
 900 struct attribute *bch2_dev_files[] = {
 901         &sysfs_uuid,
 902         &sysfs_bucket_size,
 903         &sysfs_block_size,
 904         &sysfs_first_bucket,
 905         &sysfs_nbuckets,
 906
 907         /* settings: */
 908         &sysfs_discard,
 909         &sysfs_cache_replacement_policy,
 910         &sysfs_tier,
 911         &sysfs_state_rw,
 912
 913         &sysfs_has_data,
 914         &sysfs_iostats,
 915
 916         /* alloc info - data: */
 917         &sysfs_dirty_data,
 918         &sysfs_dirty_bytes,
 919         &sysfs_cached_data,
 920         &sysfs_cached_bytes,
 921
 922         /* alloc info - buckets: */
 923         &sysfs_available_buckets,
 924         &sysfs_free_buckets,
 925         &sysfs_dirty_buckets,
 926         &sysfs_cached_buckets,
 927         &sysfs_meta_buckets,
 928         &sysfs_alloc_buckets,
 929
 930         /* alloc info - other stats: */
 931         &sysfs_read_priority_stats,
 932         &sysfs_write_priority_stats,
 933         &sysfs_fragmentation_stats,
 934         &sysfs_oldest_gen_stats,
 935         &sysfs_reserve_stats,
 936
 937         /* debug: */
 938         &sysfs_alloc_debug,
 939
 940         sysfs_pd_controller_files(copy_gc),
 941         NULL
 942 };
 943
 944 #endif  /* _BCACHEFS_SYSFS_H_ */