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

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * bcache sysfs interfaces
   4  *
   5  * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
   6  * Copyright 2012 Google, Inc.
   7  */
   8
   9 #ifndef NO_BCACHEFS_SYSFS
  10
  11 #include "bcachefs.h"
  12 #include "alloc_background.h"
  13 #include "sysfs.h"
  14 #include "btree_cache.h"
  15 #include "btree_io.h"
  16 #include "btree_iter.h"
  17 #include "btree_key_cache.h"
  18 #include "btree_update.h"
  19 #include "btree_update_interior.h"
  20 #include "btree_gc.h"
  21 #include "buckets.h"
  22 #include "clock.h"
  23 #include "disk_groups.h"
  24 #include "ec.h"
  25 #include "inode.h"
  26 #include "journal.h"
  27 #include "keylist.h"
  28 #include "move.h"
  29 #include "opts.h"
  30 #include "rebalance.h"
  31 #include "replicas.h"
  32 #include "super-io.h"
  33 #include "tests.h"
  34
  35 #include <linux/blkdev.h>
  36 #include <linux/sort.h>
  37 #include <linux/sched/clock.h>
  38
  39 #include "util.h"
  40
  41 #define SYSFS_OPS(type)                                                 \
  42 struct sysfs_ops type ## _sysfs_ops = {                                 \
  43         .show   = type ## _show,                                        \
  44         .store  = type ## _store                                        \
  45 }
  46
  47 #define SHOW(fn)                                                        \
  48 static ssize_t fn ## _show(struct kobject *kobj, struct attribute *attr,\
  49                            char *buf)                                   \
  50
  51 #define STORE(fn)                                                       \
  52 static ssize_t fn ## _store(struct kobject *kobj, struct attribute *attr,\
  53                             const char *buf, size_t size)               \
  54
  55 #define __sysfs_attribute(_name, _mode)                                 \
  56         static struct attribute sysfs_##_name =                         \
  57                 { .name = #_name, .mode = _mode }
  58
  59 #define write_attribute(n)      __sysfs_attribute(n, S_IWUSR)
  60 #define read_attribute(n)       __sysfs_attribute(n, S_IRUGO)
  61 #define rw_attribute(n)         __sysfs_attribute(n, S_IRUGO|S_IWUSR)
  62
  63 #define sysfs_printf(file, fmt, ...)                                    \
  64 do {                                                                    \
  65         if (attr == &sysfs_ ## file)                                    \
  66                 return scnprintf(buf, PAGE_SIZE, fmt "\n", __VA_ARGS__);\
  67 } while (0)
  68
  69 #define sysfs_print(file, var)                                          \
  70 do {                                                                    \
  71         if (attr == &sysfs_ ## file)                                    \
  72                 return snprint(buf, PAGE_SIZE, var);                    \
  73 } while (0)
  74
  75 #define sysfs_hprint(file, val)                                         \
  76 do {                                                                    \
  77         if (attr == &sysfs_ ## file) {                                  \
  78                 bch2_hprint(&out, val);                                 \
  79                 pr_buf(&out, "\n");                                     \
  80                 return out.pos - buf;                                   \
  81         }                                                               \
  82 } while (0)
  83
  84 #define var_printf(_var, fmt)   sysfs_printf(_var, fmt, var(_var))
  85 #define var_print(_var)         sysfs_print(_var, var(_var))
  86 #define var_hprint(_var)        sysfs_hprint(_var, var(_var))
  87
  88 #define sysfs_strtoul(file, var)                                        \
  89 do {                                                                    \
  90         if (attr == &sysfs_ ## file)                                    \
  91                 return strtoul_safe(buf, var) ?: (ssize_t) size;        \
  92 } while (0)
  93
  94 #define sysfs_strtoul_clamp(file, var, min, max)                        \
  95 do {                                                                    \
  96         if (attr == &sysfs_ ## file)                                    \
  97                 return strtoul_safe_clamp(buf, var, min, max)           \
  98                         ?: (ssize_t) size;                              \
  99 } while (0)
 100
 101 #define strtoul_or_return(cp)                                           \
 102 ({                                                                      \
 103         unsigned long _v;                                               \
 104         int _r = kstrtoul(cp, 10, &_v);                                 \
 105         if (_r)                                                         \
 106                 return _r;                                              \
 107         _v;                                                             \
 108 })
 109
 110 #define strtoul_restrict_or_return(cp, min, max)                        \
 111 ({                                                                      \
 112         unsigned long __v = 0;                                          \
 113         int _r = strtoul_safe_restrict(cp, __v, min, max);              \
 114         if (_r)                                                         \
 115                 return _r;                                              \
 116         __v;                                                            \
 117 })
 118
 119 #define strtoi_h_or_return(cp)                                          \
 120 ({                                                                      \
 121         u64 _v;                                                         \
 122         int _r = strtoi_h(cp, &_v);                                     \
 123         if (_r)                                                         \
 124                 return _r;                                              \
 125         _v;                                                             \
 126 })
 127
 128 #define sysfs_hatoi(file, var)                                          \
 129 do {                                                                    \
 130         if (attr == &sysfs_ ## file)                                    \
 131                 return strtoi_h(buf, &var) ?: (ssize_t) size;           \
 132 } while (0)
 133
 134 write_attribute(trigger_journal_flush);
 135 write_attribute(trigger_btree_coalesce);
 136 write_attribute(trigger_gc);
 137 write_attribute(prune_cache);
 138 rw_attribute(btree_gc_periodic);
 139
 140 read_attribute(uuid);
 141 read_attribute(minor);
 142 read_attribute(bucket_size);
 143 read_attribute(block_size);
 144 read_attribute(btree_node_size);
 145 read_attribute(first_bucket);
 146 read_attribute(nbuckets);
 147 read_attribute(durability);
 148 read_attribute(iodone);
 149
 150 read_attribute(io_latency_read);
 151 read_attribute(io_latency_write);
 152 read_attribute(io_latency_stats_read);
 153 read_attribute(io_latency_stats_write);
 154 read_attribute(congested);
 155
 156 read_attribute(bucket_quantiles_last_read);
 157 read_attribute(bucket_quantiles_last_write);
 158 read_attribute(bucket_quantiles_fragmentation);
 159 read_attribute(bucket_quantiles_oldest_gen);
 160
 161 read_attribute(reserve_stats);
 162 read_attribute(btree_cache_size);
 163 read_attribute(compression_stats);
 164 read_attribute(journal_debug);
 165 read_attribute(journal_pins);
 166 read_attribute(btree_updates);
 167 read_attribute(dirty_btree_nodes);
 168 read_attribute(btree_cache);
 169 read_attribute(btree_key_cache);
 170 read_attribute(btree_transactions);
 171 read_attribute(stripes_heap);
 172
 173 read_attribute(internal_uuid);
 174
 175 read_attribute(has_data);
 176 read_attribute(alloc_debug);
 177 write_attribute(wake_allocator);
 178
 179 read_attribute(read_realloc_races);
 180 read_attribute(extent_migrate_done);
 181 read_attribute(extent_migrate_raced);
 182
 183 rw_attribute(journal_write_delay_ms);
 184 rw_attribute(journal_reclaim_delay_ms);
 185
 186 rw_attribute(discard);
 187 rw_attribute(cache_replacement_policy);
 188 rw_attribute(label);
 189
 190 rw_attribute(copy_gc_enabled);
 191 sysfs_pd_controller_attribute(copy_gc);
 192
 193 rw_attribute(rebalance_enabled);
 194 sysfs_pd_controller_attribute(rebalance);
 195 read_attribute(rebalance_work);
 196 rw_attribute(promote_whole_extents);
 197
 198 read_attribute(new_stripes);
 199
 200 rw_attribute(pd_controllers_update_seconds);
 201
 202 read_attribute(meta_replicas_have);
 203 read_attribute(data_replicas_have);
 204
 205 read_attribute(io_timers_read);
 206 read_attribute(io_timers_write);
 207
 208 #ifdef CONFIG_BCACHEFS_TESTS
 209 write_attribute(perf_test);
 210 #endif /* CONFIG_BCACHEFS_TESTS */
 211
 212 #define x(_name)                                                \
 213         static struct attribute sysfs_time_stat_##_name =               \
 214                 { .name = #_name, .mode = S_IRUGO };
 215         BCH_TIME_STATS()
 216 #undef x
 217
 218 static struct attribute sysfs_state_rw = {
 219         .name = "state",
 220         .mode = S_IRUGO
 221 };
 222
 223 static size_t bch2_btree_cache_size(struct bch_fs *c)
 224 {
 225         size_t ret = 0;
 226         struct btree *b;
 227
 228         mutex_lock(&c->btree_cache.lock);
 229         list_for_each_entry(b, &c->btree_cache.live, list)
 230                 ret += btree_bytes(c);
 231
 232         mutex_unlock(&c->btree_cache.lock);
 233         return ret;
 234 }
 235
 236 static int fs_alloc_debug_to_text(struct printbuf *out, struct bch_fs *c)
 237 {
 238         struct bch_fs_usage *fs_usage = bch2_fs_usage_read(c);
 239
 240         if (!fs_usage)
 241                 return -ENOMEM;
 242
 243         bch2_fs_usage_to_text(out, c, fs_usage);
 244
 245         percpu_up_read(&c->mark_lock);
 246
 247         kfree(fs_usage);
 248         return 0;
 249 }
 250
 251 static int bch2_compression_stats_to_text(struct printbuf *out, struct bch_fs *c)
 252 {
 253         struct btree_trans trans;
 254         struct btree_iter *iter;
 255         struct bkey_s_c k;
 256         u64 nr_uncompressed_extents = 0, uncompressed_sectors = 0,
 257             nr_compressed_extents = 0,
 258             compressed_sectors_compressed = 0,
 259             compressed_sectors_uncompressed = 0;
 260         int ret;
 261
 262         if (!test_bit(BCH_FS_STARTED, &c->flags))
 263                 return -EPERM;
 264
 265         bch2_trans_init(&trans, c, 0, 0);
 266
 267         for_each_btree_key(&trans, iter, BTREE_ID_EXTENTS, POS_MIN, 0, k, ret)
 268                 if (k.k->type == KEY_TYPE_extent) {
 269                         struct bkey_s_c_extent e = bkey_s_c_to_extent(k);
 270                         const union bch_extent_entry *entry;
 271                         struct extent_ptr_decoded p;
 272
 273                         extent_for_each_ptr_decode(e, p, entry) {
 274                                 if (!crc_is_compressed(p.crc)) {
 275                                         nr_uncompressed_extents++;
 276                                         uncompressed_sectors += e.k->size;
 277                                 } else {
 278                                         nr_compressed_extents++;
 279                                         compressed_sectors_compressed +=
 280                                                 p.crc.compressed_size;
 281                                         compressed_sectors_uncompressed +=
 282                                                 p.crc.uncompressed_size;
 283                                 }
 284
 285                                 /* only looking at the first ptr */
 286                                 break;
 287                         }
 288                 }
 289
 290         ret = bch2_trans_exit(&trans) ?: ret;
 291         if (ret)
 292                 return ret;
 293
 294         pr_buf(out,
 295                "uncompressed data:\n"
 296                "        nr extents:                     %llu\n"
 297                "        size (bytes):                   %llu\n"
 298                "compressed data:\n"
 299                "        nr extents:                     %llu\n"
 300                "        compressed size (bytes):        %llu\n"
 301                "        uncompressed size (bytes):      %llu\n",
 302                nr_uncompressed_extents,
 303                uncompressed_sectors << 9,
 304                nr_compressed_extents,
 305                compressed_sectors_compressed << 9,
 306                compressed_sectors_uncompressed << 9);
 307         return 0;
 308 }
 309
 310 SHOW(bch2_fs)
 311 {
 312         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 313         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 314
 315         sysfs_print(minor,                      c->minor);
 316         sysfs_printf(internal_uuid, "%pU",      c->sb.uuid.b);
 317
 318         sysfs_print(journal_write_delay_ms,     c->journal.write_delay_ms);
 319         sysfs_print(journal_reclaim_delay_ms,   c->journal.reclaim_delay_ms);
 320
 321         sysfs_print(block_size,                 block_bytes(c));
 322         sysfs_print(btree_node_size,            btree_bytes(c));
 323         sysfs_hprint(btree_cache_size,          bch2_btree_cache_size(c));
 324
 325         sysfs_print(read_realloc_races,
 326                     atomic_long_read(&c->read_realloc_races));
 327         sysfs_print(extent_migrate_done,
 328                     atomic_long_read(&c->extent_migrate_done));
 329         sysfs_print(extent_migrate_raced,
 330                     atomic_long_read(&c->extent_migrate_raced));
 331
 332         sysfs_printf(btree_gc_periodic, "%u",   (int) c->btree_gc_periodic);
 333
 334         sysfs_printf(copy_gc_enabled, "%i", c->copy_gc_enabled);
 335
 336         sysfs_print(pd_controllers_update_seconds,
 337                     c->pd_controllers_update_seconds);
 338
 339         sysfs_printf(rebalance_enabled,         "%i", c->rebalance.enabled);
 340         sysfs_pd_controller_show(rebalance,     &c->rebalance.pd); /* XXX */
 341         sysfs_pd_controller_show(copy_gc,       &c->copygc_pd);
 342
 343         if (attr == &sysfs_rebalance_work) {
 344                 bch2_rebalance_work_to_text(&out, c);
 345                 return out.pos - buf;
 346         }
 347
 348         sysfs_print(promote_whole_extents,      c->promote_whole_extents);
 349
 350         sysfs_printf(meta_replicas_have, "%i",  bch2_replicas_online(c, true));
 351         sysfs_printf(data_replicas_have, "%i",  bch2_replicas_online(c, false));
 352
 353         /* Debugging: */
 354
 355         if (attr == &sysfs_alloc_debug)
 356                 return fs_alloc_debug_to_text(&out, c) ?: out.pos - buf;
 357
 358         if (attr == &sysfs_journal_debug) {
 359                 bch2_journal_debug_to_text(&out, &c->journal);
 360                 return out.pos - buf;
 361         }
 362
 363         if (attr == &sysfs_journal_pins) {
 364                 bch2_journal_pins_to_text(&out, &c->journal);
 365                 return out.pos - buf;
 366         }
 367
 368         if (attr == &sysfs_btree_updates) {
 369                 bch2_btree_updates_to_text(&out, c);
 370                 return out.pos - buf;
 371         }
 372
 373         if (attr == &sysfs_dirty_btree_nodes) {
 374                 bch2_dirty_btree_nodes_to_text(&out, c);
 375                 return out.pos - buf;
 376         }
 377
 378         if (attr == &sysfs_btree_cache) {
 379                 bch2_btree_cache_to_text(&out, c);
 380                 return out.pos - buf;
 381         }
 382
 383         if (attr == &sysfs_btree_key_cache) {
 384                 bch2_btree_key_cache_to_text(&out, &c->btree_key_cache);
 385                 return out.pos - buf;
 386         }
 387
 388         if (attr == &sysfs_btree_transactions) {
 389                 bch2_btree_trans_to_text(&out, c);
 390                 return out.pos - buf;
 391         }
 392
 393         if (attr == &sysfs_stripes_heap) {
 394                 bch2_stripes_heap_to_text(&out, c);
 395                 return out.pos - buf;
 396         }
 397
 398         if (attr == &sysfs_compression_stats) {
 399                 bch2_compression_stats_to_text(&out, c);
 400                 return out.pos - buf;
 401         }
 402
 403         if (attr == &sysfs_new_stripes) {
 404                 bch2_new_stripes_to_text(&out, c);
 405                 return out.pos - buf;
 406         }
 407
 408         if (attr == &sysfs_io_timers_read) {
 409                 bch2_io_timers_to_text(&out, &c->io_clock[READ]);
 410                 return out.pos - buf;
 411         }
 412         if (attr == &sysfs_io_timers_write) {
 413                 bch2_io_timers_to_text(&out, &c->io_clock[WRITE]);
 414                 return out.pos - buf;
 415         }
 416
 417         return 0;
 418 }
 419
 420 STORE(bch2_fs)
 421 {
 422         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 423
 424         sysfs_strtoul(journal_write_delay_ms, c->journal.write_delay_ms);
 425         sysfs_strtoul(journal_reclaim_delay_ms, c->journal.reclaim_delay_ms);
 426
 427         if (attr == &sysfs_btree_gc_periodic) {
 428                 ssize_t ret = strtoul_safe(buf, c->btree_gc_periodic)
 429                         ?: (ssize_t) size;
 430
 431                 wake_up_process(c->gc_thread);
 432                 return ret;
 433         }
 434
 435         if (attr == &sysfs_copy_gc_enabled) {
 436                 ssize_t ret = strtoul_safe(buf, c->copy_gc_enabled)
 437                         ?: (ssize_t) size;
 438
 439                 if (c->copygc_thread)
 440                         wake_up_process(c->copygc_thread);
 441                 return ret;
 442         }
 443
 444         if (attr == &sysfs_rebalance_enabled) {
 445                 ssize_t ret = strtoul_safe(buf, c->rebalance.enabled)
 446                         ?: (ssize_t) size;
 447
 448                 rebalance_wakeup(c);
 449                 return ret;
 450         }
 451
 452         sysfs_strtoul(pd_controllers_update_seconds,
 453                       c->pd_controllers_update_seconds);
 454         sysfs_pd_controller_store(rebalance,    &c->rebalance.pd);
 455         sysfs_pd_controller_store(copy_gc,      &c->copygc_pd);
 456
 457         sysfs_strtoul(promote_whole_extents,    c->promote_whole_extents);
 458
 459         /* Debugging: */
 460
 461         if (!test_bit(BCH_FS_STARTED, &c->flags))
 462                 return -EPERM;
 463
 464         /* Debugging: */
 465
 466         if (attr == &sysfs_trigger_journal_flush)
 467                 bch2_journal_meta(&c->journal);
 468
 469         if (attr == &sysfs_trigger_btree_coalesce)
 470                 bch2_coalesce(c);
 471
 472         if (attr == &sysfs_trigger_gc) {
 473                 /*
 474                  * Full gc is currently incompatible with btree key cache:
 475                  */
 476 #if 0
 477                 down_read(&c->state_lock);
 478                 bch2_gc(c, NULL, false, false);
 479                 up_read(&c->state_lock);
 480 #else
 481                 bch2_gc_gens(c);
 482 #endif
 483         }
 484
 485         if (attr == &sysfs_prune_cache) {
 486                 struct shrink_control sc;
 487
 488                 sc.gfp_mask = GFP_KERNEL;
 489                 sc.nr_to_scan = strtoul_or_return(buf);
 490                 c->btree_cache.shrink.scan_objects(&c->btree_cache.shrink, &sc);
 491         }
 492
 493 #ifdef CONFIG_BCACHEFS_TESTS
 494         if (attr == &sysfs_perf_test) {
 495                 char *tmp = kstrdup(buf, GFP_KERNEL), *p = tmp;
 496                 char *test              = strsep(&p, " \t\n");
 497                 char *nr_str            = strsep(&p, " \t\n");
 498                 char *threads_str       = strsep(&p, " \t\n");
 499                 unsigned threads;
 500                 u64 nr;
 501                 int ret = -EINVAL;
 502
 503                 if (threads_str &&
 504                     !(ret = kstrtouint(threads_str, 10, &threads)) &&
 505                     !(ret = bch2_strtoull_h(nr_str, &nr)))
 506                         bch2_btree_perf_test(c, test, nr, threads);
 507                 else
 508                         size = ret;
 509                 kfree(tmp);
 510         }
 511 #endif
 512         return size;
 513 }
 514 SYSFS_OPS(bch2_fs);
 515
 516 struct attribute *bch2_fs_files[] = {
 517         &sysfs_minor,
 518         &sysfs_block_size,
 519         &sysfs_btree_node_size,
 520         &sysfs_btree_cache_size,
 521
 522         &sysfs_meta_replicas_have,
 523         &sysfs_data_replicas_have,
 524
 525         &sysfs_journal_write_delay_ms,
 526         &sysfs_journal_reclaim_delay_ms,
 527
 528         &sysfs_promote_whole_extents,
 529
 530         &sysfs_compression_stats,
 531
 532 #ifdef CONFIG_BCACHEFS_TESTS
 533         &sysfs_perf_test,
 534 #endif
 535         NULL
 536 };
 537
 538 /* internal dir - just a wrapper */
 539
 540 SHOW(bch2_fs_internal)
 541 {
 542         struct bch_fs *c = container_of(kobj, struct bch_fs, internal);
 543         return bch2_fs_show(&c->kobj, attr, buf);
 544 }
 545
 546 STORE(bch2_fs_internal)
 547 {
 548         struct bch_fs *c = container_of(kobj, struct bch_fs, internal);
 549         return bch2_fs_store(&c->kobj, attr, buf, size);
 550 }
 551 SYSFS_OPS(bch2_fs_internal);
 552
 553 struct attribute *bch2_fs_internal_files[] = {
 554         &sysfs_alloc_debug,
 555         &sysfs_journal_debug,
 556         &sysfs_journal_pins,
 557         &sysfs_btree_updates,
 558         &sysfs_dirty_btree_nodes,
 559         &sysfs_btree_cache,
 560         &sysfs_btree_key_cache,
 561         &sysfs_btree_transactions,
 562         &sysfs_stripes_heap,
 563
 564         &sysfs_read_realloc_races,
 565         &sysfs_extent_migrate_done,
 566         &sysfs_extent_migrate_raced,
 567
 568         &sysfs_trigger_journal_flush,
 569         &sysfs_trigger_btree_coalesce,
 570         &sysfs_trigger_gc,
 571         &sysfs_prune_cache,
 572
 573         &sysfs_copy_gc_enabled,
 574
 575         &sysfs_rebalance_enabled,
 576         &sysfs_rebalance_work,
 577         sysfs_pd_controller_files(rebalance),
 578         sysfs_pd_controller_files(copy_gc),
 579
 580         &sysfs_new_stripes,
 581
 582         &sysfs_io_timers_read,
 583         &sysfs_io_timers_write,
 584
 585         &sysfs_internal_uuid,
 586         NULL
 587 };
 588
 589 /* options */
 590
 591 SHOW(bch2_fs_opts_dir)
 592 {
 593         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 594         struct bch_fs *c = container_of(kobj, struct bch_fs, opts_dir);
 595         const struct bch_option *opt = container_of(attr, struct bch_option, attr);
 596         int id = opt - bch2_opt_table;
 597         u64 v = bch2_opt_get_by_id(&c->opts, id);
 598
 599         bch2_opt_to_text(&out, c, opt, v, OPT_SHOW_FULL_LIST);
 600         pr_buf(&out, "\n");
 601
 602         return out.pos - buf;
 603 }
 604
 605 STORE(bch2_fs_opts_dir)
 606 {
 607         struct bch_fs *c = container_of(kobj, struct bch_fs, opts_dir);
 608         const struct bch_option *opt = container_of(attr, struct bch_option, attr);
 609         int ret, id = opt - bch2_opt_table;
 610         char *tmp;
 611         u64 v;
 612
 613         tmp = kstrdup(buf, GFP_KERNEL);
 614         if (!tmp)
 615                 return -ENOMEM;
 616
 617         ret = bch2_opt_parse(c, opt, strim(tmp), &v);
 618         kfree(tmp);
 619
 620         if (ret < 0)
 621                 return ret;
 622
 623         ret = bch2_opt_check_may_set(c, id, v);
 624         if (ret < 0)
 625                 return ret;
 626
 627         if (opt->set_sb != SET_NO_SB_OPT) {
 628                 mutex_lock(&c->sb_lock);
 629                 opt->set_sb(c->disk_sb.sb, v);
 630                 bch2_write_super(c);
 631                 mutex_unlock(&c->sb_lock);
 632         }
 633
 634         bch2_opt_set_by_id(&c->opts, id, v);
 635
 636         if ((id == Opt_background_target ||
 637              id == Opt_background_compression) && v) {
 638                 bch2_rebalance_add_work(c, S64_MAX);
 639                 rebalance_wakeup(c);
 640         }
 641
 642         return size;
 643 }
 644 SYSFS_OPS(bch2_fs_opts_dir);
 645
 646 struct attribute *bch2_fs_opts_dir_files[] = { NULL };
 647
 648 int bch2_opts_create_sysfs_files(struct kobject *kobj)
 649 {
 650         const struct bch_option *i;
 651         int ret;
 652
 653         for (i = bch2_opt_table;
 654              i < bch2_opt_table + bch2_opts_nr;
 655              i++) {
 656                 if (!(i->mode & (OPT_FORMAT|OPT_MOUNT|OPT_RUNTIME)))
 657                         continue;
 658
 659                 ret = sysfs_create_file(kobj, &i->attr);
 660                 if (ret)
 661                         return ret;
 662         }
 663
 664         return 0;
 665 }
 666
 667 /* time stats */
 668
 669 SHOW(bch2_fs_time_stats)
 670 {
 671         struct bch_fs *c = container_of(kobj, struct bch_fs, time_stats);
 672         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 673
 674 #define x(name)                                                         \
 675         if (attr == &sysfs_time_stat_##name) {                          \
 676                 bch2_time_stats_to_text(&out, &c->times[BCH_TIME_##name]);\
 677                 return out.pos - buf;                                   \
 678         }
 679         BCH_TIME_STATS()
 680 #undef x
 681
 682         return 0;
 683 }
 684
 685 STORE(bch2_fs_time_stats)
 686 {
 687         return size;
 688 }
 689 SYSFS_OPS(bch2_fs_time_stats);
 690
 691 struct attribute *bch2_fs_time_stats_files[] = {
 692 #define x(name)                                         \
 693         &sysfs_time_stat_##name,
 694         BCH_TIME_STATS()
 695 #undef x
 696         NULL
 697 };
 698
 699 typedef unsigned (bucket_map_fn)(struct bch_fs *, struct bch_dev *,
 700                                  size_t, void *);
 701
 702 static unsigned bucket_last_io_fn(struct bch_fs *c, struct bch_dev *ca,
 703                                   size_t b, void *private)
 704 {
 705         int rw = (private ? 1 : 0);
 706
 707         return bucket_last_io(c, bucket(ca, b), rw);
 708 }
 709
 710 static unsigned bucket_sectors_used_fn(struct bch_fs *c, struct bch_dev *ca,
 711                                        size_t b, void *private)
 712 {
 713         struct bucket *g = bucket(ca, b);
 714         return bucket_sectors_used(g->mark);
 715 }
 716
 717 static unsigned bucket_oldest_gen_fn(struct bch_fs *c, struct bch_dev *ca,
 718                                      size_t b, void *private)
 719 {
 720         return bucket_gc_gen(ca, b);
 721 }
 722
 723 static int unsigned_cmp(const void *_l, const void *_r)
 724 {
 725         const unsigned *l = _l;
 726         const unsigned *r = _r;
 727
 728         return cmp_int(*l, *r);
 729 }
 730
 731 static int quantiles_to_text(struct printbuf *out,
 732                              struct bch_fs *c, struct bch_dev *ca,
 733                              bucket_map_fn *fn, void *private)
 734 {
 735         size_t i, n;
 736         /* Compute 31 quantiles */
 737         unsigned q[31], *p;
 738
 739         down_read(&ca->bucket_lock);
 740         n = ca->mi.nbuckets;
 741
 742         p = vzalloc(n * sizeof(unsigned));
 743         if (!p) {
 744                 up_read(&ca->bucket_lock);
 745                 return -ENOMEM;
 746         }
 747
 748         for (i = ca->mi.first_bucket; i < n; i++)
 749                 p[i] = fn(c, ca, i, private);
 750
 751         sort(p, n, sizeof(unsigned), unsigned_cmp, NULL);
 752         up_read(&ca->bucket_lock);
 753
 754         while (n &&
 755                !p[n - 1])
 756                 --n;
 757
 758         for (i = 0; i < ARRAY_SIZE(q); i++)
 759                 q[i] = p[n * (i + 1) / (ARRAY_SIZE(q) + 1)];
 760
 761         vfree(p);
 762
 763         for (i = 0; i < ARRAY_SIZE(q); i++)
 764                 pr_buf(out, "%u ", q[i]);
 765         pr_buf(out, "\n");
 766         return 0;
 767 }
 768
 769 static void reserve_stats_to_text(struct printbuf *out, struct bch_dev *ca)
 770 {
 771         enum alloc_reserve i;
 772
 773         spin_lock(&ca->fs->freelist_lock);
 774
 775         pr_buf(out, "free_inc:\t%zu\t%zu\n",
 776                fifo_used(&ca->free_inc),
 777                ca->free_inc.size);
 778
 779         for (i = 0; i < RESERVE_NR; i++)
 780                 pr_buf(out, "free[%u]:\t%zu\t%zu\n", i,
 781                        fifo_used(&ca->free[i]),
 782                        ca->free[i].size);
 783
 784         spin_unlock(&ca->fs->freelist_lock);
 785 }
 786
 787 static void dev_alloc_debug_to_text(struct printbuf *out, struct bch_dev *ca)
 788 {
 789         struct bch_fs *c = ca->fs;
 790         struct bch_dev_usage stats = bch2_dev_usage_read(ca);
 791         unsigned i, nr[BCH_DATA_NR];
 792
 793         memset(nr, 0, sizeof(nr));
 794
 795         for (i = 0; i < ARRAY_SIZE(c->open_buckets); i++)
 796                 nr[c->open_buckets[i].type]++;
 797
 798         pr_buf(out,
 799                 "free_inc:               %zu/%zu\n"
 800                 "free[RESERVE_BTREE]:    %zu/%zu\n"
 801                 "free[RESERVE_MOVINGGC]: %zu/%zu\n"
 802                 "free[RESERVE_NONE]:     %zu/%zu\n"
 803                 "buckets:\n"
 804                 "    capacity:           %llu\n"
 805                 "    alloc:              %llu\n"
 806                 "    sb:                 %llu\n"
 807                 "    journal:            %llu\n"
 808                 "    meta:               %llu\n"
 809                 "    user:               %llu\n"
 810                 "    cached:             %llu\n"
 811                 "    erasure coded:      %llu\n"
 812                 "    available:          %lli\n"
 813                 "sectors:\n"
 814                 "    sb:                 %llu\n"
 815                 "    journal:            %llu\n"
 816                 "    meta:               %llu\n"
 817                 "    user:               %llu\n"
 818                 "    cached:             %llu\n"
 819                 "    erasure coded:      %llu\n"
 820                 "    fragmented:         %llu\n"
 821                 "    copygc threshold:   %llu\n"
 822                 "freelist_wait:          %s\n"
 823                 "open buckets:           %u/%u (reserved %u)\n"
 824                 "open_buckets_wait:      %s\n"
 825                 "open_buckets_btree:     %u\n"
 826                 "open_buckets_user:      %u\n"
 827                 "btree reserve cache:    %u\n",
 828                 fifo_used(&ca->free_inc),               ca->free_inc.size,
 829                 fifo_used(&ca->free[RESERVE_BTREE]),    ca->free[RESERVE_BTREE].size,
 830                 fifo_used(&ca->free[RESERVE_MOVINGGC]), ca->free[RESERVE_MOVINGGC].size,
 831                 fifo_used(&ca->free[RESERVE_NONE]),     ca->free[RESERVE_NONE].size,
 832                 ca->mi.nbuckets - ca->mi.first_bucket,
 833                 stats.buckets_alloc,
 834                 stats.buckets[BCH_DATA_sb],
 835                 stats.buckets[BCH_DATA_journal],
 836                 stats.buckets[BCH_DATA_btree],
 837                 stats.buckets[BCH_DATA_user],
 838                 stats.buckets[BCH_DATA_cached],
 839                 stats.buckets_ec,
 840                 __dev_buckets_available(ca, stats),
 841                 stats.sectors[BCH_DATA_sb],
 842                 stats.sectors[BCH_DATA_journal],
 843                 stats.sectors[BCH_DATA_btree],
 844                 stats.sectors[BCH_DATA_user],
 845                 stats.sectors[BCH_DATA_cached],
 846                 stats.sectors_ec,
 847                 stats.sectors_fragmented,
 848                 c->copygc_threshold,
 849                 c->freelist_wait.list.first             ? "waiting" : "empty",
 850                 c->open_buckets_nr_free, OPEN_BUCKETS_COUNT,
 851                 BTREE_NODE_OPEN_BUCKET_RESERVE,
 852                 c->open_buckets_wait.list.first         ? "waiting" : "empty",
 853                 nr[BCH_DATA_btree],
 854                 nr[BCH_DATA_user],
 855                 c->btree_reserve_cache_nr);
 856 }
 857
 858 static const char * const bch2_rw[] = {
 859         "read",
 860         "write",
 861         NULL
 862 };
 863
 864 static void dev_iodone_to_text(struct printbuf *out, struct bch_dev *ca)
 865 {
 866         int rw, i;
 867
 868         for (rw = 0; rw < 2; rw++) {
 869                 pr_buf(out, "%s:\n", bch2_rw[rw]);
 870
 871                 for (i = 1; i < BCH_DATA_NR; i++)
 872                         pr_buf(out, "%-12s:%12llu\n",
 873                                bch2_data_types[i],
 874                                percpu_u64_get(&ca->io_done->sectors[rw][i]) << 9);
 875         }
 876 }
 877
 878 SHOW(bch2_dev)
 879 {
 880         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
 881         struct bch_fs *c = ca->fs;
 882         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 883
 884         sysfs_printf(uuid,              "%pU\n", ca->uuid.b);
 885
 886         sysfs_print(bucket_size,        bucket_bytes(ca));
 887         sysfs_print(block_size,         block_bytes(c));
 888         sysfs_print(first_bucket,       ca->mi.first_bucket);
 889         sysfs_print(nbuckets,           ca->mi.nbuckets);
 890         sysfs_print(durability,         ca->mi.durability);
 891         sysfs_print(discard,            ca->mi.discard);
 892
 893         if (attr == &sysfs_label) {
 894                 if (ca->mi.group) {
 895                         mutex_lock(&c->sb_lock);
 896                         bch2_disk_path_to_text(&out, &c->disk_sb,
 897                                                ca->mi.group - 1);
 898                         mutex_unlock(&c->sb_lock);
 899                 }
 900
 901                 pr_buf(&out, "\n");
 902                 return out.pos - buf;
 903         }
 904
 905         if (attr == &sysfs_has_data) {
 906                 bch2_flags_to_text(&out, bch2_data_types,
 907                                    bch2_dev_has_data(c, ca));
 908                 pr_buf(&out, "\n");
 909                 return out.pos - buf;
 910         }
 911
 912         if (attr == &sysfs_cache_replacement_policy) {
 913                 bch2_string_opt_to_text(&out,
 914                                         bch2_cache_replacement_policies,
 915                                         ca->mi.replacement);
 916                 pr_buf(&out, "\n");
 917                 return out.pos - buf;
 918         }
 919
 920         if (attr == &sysfs_state_rw) {
 921                 bch2_string_opt_to_text(&out, bch2_dev_state,
 922                                         ca->mi.state);
 923                 pr_buf(&out, "\n");
 924                 return out.pos - buf;
 925         }
 926
 927         if (attr == &sysfs_iodone) {
 928                 dev_iodone_to_text(&out, ca);
 929                 return out.pos - buf;
 930         }
 931
 932         sysfs_print(io_latency_read,            atomic64_read(&ca->cur_latency[READ]));
 933         sysfs_print(io_latency_write,           atomic64_read(&ca->cur_latency[WRITE]));
 934
 935         if (attr == &sysfs_io_latency_stats_read) {
 936                 bch2_time_stats_to_text(&out, &ca->io_latency[READ]);
 937                 return out.pos - buf;
 938         }
 939         if (attr == &sysfs_io_latency_stats_write) {
 940                 bch2_time_stats_to_text(&out, &ca->io_latency[WRITE]);
 941                 return out.pos - buf;
 942         }
 943
 944         sysfs_printf(congested,                 "%u%%",
 945                      clamp(atomic_read(&ca->congested), 0, CONGESTED_MAX)
 946                      * 100 / CONGESTED_MAX);
 947
 948         if (attr == &sysfs_bucket_quantiles_last_read)
 949                 return quantiles_to_text(&out, c, ca, bucket_last_io_fn, (void *) 0) ?: out.pos - buf;
 950         if (attr == &sysfs_bucket_quantiles_last_write)
 951                 return quantiles_to_text(&out, c, ca, bucket_last_io_fn, (void *) 1) ?: out.pos - buf;
 952         if (attr == &sysfs_bucket_quantiles_fragmentation)
 953                 return quantiles_to_text(&out, c, ca, bucket_sectors_used_fn, NULL)  ?: out.pos - buf;
 954         if (attr == &sysfs_bucket_quantiles_oldest_gen)
 955                 return quantiles_to_text(&out, c, ca, bucket_oldest_gen_fn, NULL)    ?: out.pos - buf;
 956
 957         if (attr == &sysfs_reserve_stats) {
 958                 reserve_stats_to_text(&out, ca);
 959                 return out.pos - buf;
 960         }
 961         if (attr == &sysfs_alloc_debug) {
 962                 dev_alloc_debug_to_text(&out, ca);
 963                 return out.pos - buf;
 964         }
 965
 966         return 0;
 967 }
 968
 969 STORE(bch2_dev)
 970 {
 971         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
 972         struct bch_fs *c = ca->fs;
 973         struct bch_member *mi;
 974
 975         if (attr == &sysfs_discard) {
 976                 bool v = strtoul_or_return(buf);
 977
 978                 mutex_lock(&c->sb_lock);
 979                 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
 980
 981                 if (v != BCH_MEMBER_DISCARD(mi)) {
 982                         SET_BCH_MEMBER_DISCARD(mi, v);
 983                         bch2_write_super(c);
 984                 }
 985                 mutex_unlock(&c->sb_lock);
 986         }
 987
 988         if (attr == &sysfs_cache_replacement_policy) {
 989                 ssize_t v = __sysfs_match_string(bch2_cache_replacement_policies, -1, buf);
 990
 991                 if (v < 0)
 992                         return v;
 993
 994                 mutex_lock(&c->sb_lock);
 995                 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
 996
 997                 if ((unsigned) v != BCH_MEMBER_REPLACEMENT(mi)) {
 998                         SET_BCH_MEMBER_REPLACEMENT(mi, v);
 999                         bch2_write_super(c);
1000                 }
1001                 mutex_unlock(&c->sb_lock);
1002         }
1003
1004         if (attr == &sysfs_label) {
1005                 char *tmp;
1006                 int ret;
1007
1008                 tmp = kstrdup(buf, GFP_KERNEL);
1009                 if (!tmp)
1010                         return -ENOMEM;
1011
1012                 ret = bch2_dev_group_set(c, ca, strim(tmp));
1013                 kfree(tmp);
1014                 if (ret)
1015                         return ret;
1016         }
1017
1018         if (attr == &sysfs_wake_allocator)
1019                 bch2_wake_allocator(ca);
1020
1021         return size;
1022 }
1023 SYSFS_OPS(bch2_dev);
1024
1025 struct attribute *bch2_dev_files[] = {
1026         &sysfs_uuid,
1027         &sysfs_bucket_size,
1028         &sysfs_block_size,
1029         &sysfs_first_bucket,
1030         &sysfs_nbuckets,
1031         &sysfs_durability,
1032
1033         /* settings: */
1034         &sysfs_discard,
1035         &sysfs_cache_replacement_policy,
1036         &sysfs_state_rw,
1037         &sysfs_label,
1038
1039         &sysfs_has_data,
1040         &sysfs_iodone,
1041
1042         &sysfs_io_latency_read,
1043         &sysfs_io_latency_write,
1044         &sysfs_io_latency_stats_read,
1045         &sysfs_io_latency_stats_write,
1046         &sysfs_congested,
1047
1048         /* alloc info - other stats: */
1049         &sysfs_bucket_quantiles_last_read,
1050         &sysfs_bucket_quantiles_last_write,
1051         &sysfs_bucket_quantiles_fragmentation,
1052         &sysfs_bucket_quantiles_oldest_gen,
1053
1054         &sysfs_reserve_stats,
1055
1056         /* debug: */
1057         &sysfs_alloc_debug,
1058         &sysfs_wake_allocator,
1059         NULL
1060 };
1061
1062 #endif  /* _BCACHEFS_SYSFS_H_ */