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_gc);
 136 write_attribute(prune_cache);
 137 rw_attribute(btree_gc_periodic);
 138 rw_attribute(gc_gens_pos);
 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(btree_avg_write_size);
 157
 158 read_attribute(bucket_quantiles_last_read);
 159 read_attribute(bucket_quantiles_last_write);
 160 read_attribute(bucket_quantiles_fragmentation);
 161 read_attribute(bucket_quantiles_oldest_gen);
 162
 163 read_attribute(reserve_stats);
 164 read_attribute(btree_cache_size);
 165 read_attribute(compression_stats);
 166 read_attribute(journal_debug);
 167 read_attribute(journal_pins);
 168 read_attribute(btree_updates);
 169 read_attribute(dirty_btree_nodes);
 170 read_attribute(btree_cache);
 171 read_attribute(btree_key_cache);
 172 read_attribute(btree_transactions);
 173 read_attribute(stripes_heap);
 174 read_attribute(open_buckets);
 175
 176 read_attribute(internal_uuid);
 177
 178 read_attribute(has_data);
 179 read_attribute(alloc_debug);
 180 write_attribute(wake_allocator);
 181
 182 read_attribute(read_realloc_races);
 183 read_attribute(extent_migrate_done);
 184 read_attribute(extent_migrate_raced);
 185
 186 rw_attribute(journal_write_delay_ms);
 187 rw_attribute(journal_reclaim_delay_ms);
 188
 189 rw_attribute(discard);
 190 rw_attribute(cache_replacement_policy);
 191 rw_attribute(label);
 192
 193 rw_attribute(copy_gc_enabled);
 194 read_attribute(copy_gc_wait);
 195
 196 rw_attribute(rebalance_enabled);
 197 sysfs_pd_controller_attribute(rebalance);
 198 read_attribute(rebalance_work);
 199 rw_attribute(promote_whole_extents);
 200
 201 read_attribute(new_stripes);
 202
 203 read_attribute(io_timers_read);
 204 read_attribute(io_timers_write);
 205
 206 #ifdef CONFIG_BCACHEFS_TESTS
 207 write_attribute(perf_test);
 208 #endif /* CONFIG_BCACHEFS_TESTS */
 209
 210 #define x(_name)                                                \
 211         static struct attribute sysfs_time_stat_##_name =               \
 212                 { .name = #_name, .mode = S_IRUGO };
 213         BCH_TIME_STATS()
 214 #undef x
 215
 216 static struct attribute sysfs_state_rw = {
 217         .name = "state",
 218         .mode = S_IRUGO
 219 };
 220
 221 static size_t bch2_btree_cache_size(struct bch_fs *c)
 222 {
 223         size_t ret = 0;
 224         struct btree *b;
 225
 226         mutex_lock(&c->btree_cache.lock);
 227         list_for_each_entry(b, &c->btree_cache.live, list)
 228                 ret += btree_bytes(c);
 229
 230         mutex_unlock(&c->btree_cache.lock);
 231         return ret;
 232 }
 233
 234 static size_t bch2_btree_avg_write_size(struct bch_fs *c)
 235 {
 236         u64 nr = atomic64_read(&c->btree_writes_nr);
 237         u64 sectors = atomic64_read(&c->btree_writes_sectors);
 238
 239         return nr ? div64_u64(sectors, nr) : 0;
 240 }
 241
 242 static int fs_alloc_debug_to_text(struct printbuf *out, struct bch_fs *c)
 243 {
 244         struct bch_fs_usage_online *fs_usage = bch2_fs_usage_read(c);
 245
 246         if (!fs_usage)
 247                 return -ENOMEM;
 248
 249         bch2_fs_usage_to_text(out, c, fs_usage);
 250
 251         percpu_up_read(&c->mark_lock);
 252
 253         kfree(fs_usage);
 254         return 0;
 255 }
 256
 257 static int bch2_compression_stats_to_text(struct printbuf *out, struct bch_fs *c)
 258 {
 259         struct btree_trans trans;
 260         struct btree_iter *iter;
 261         struct bkey_s_c k;
 262         u64 nr_uncompressed_extents = 0, uncompressed_sectors = 0,
 263             nr_compressed_extents = 0,
 264             compressed_sectors_compressed = 0,
 265             compressed_sectors_uncompressed = 0;
 266         int ret;
 267
 268         if (!test_bit(BCH_FS_STARTED, &c->flags))
 269                 return -EPERM;
 270
 271         bch2_trans_init(&trans, c, 0, 0);
 272
 273         for_each_btree_key(&trans, iter, BTREE_ID_extents, POS_MIN, 0, k, ret)
 274                 if (k.k->type == KEY_TYPE_extent) {
 275                         struct bkey_s_c_extent e = bkey_s_c_to_extent(k);
 276                         const union bch_extent_entry *entry;
 277                         struct extent_ptr_decoded p;
 278
 279                         extent_for_each_ptr_decode(e, p, entry) {
 280                                 if (!crc_is_compressed(p.crc)) {
 281                                         nr_uncompressed_extents++;
 282                                         uncompressed_sectors += e.k->size;
 283                                 } else {
 284                                         nr_compressed_extents++;
 285                                         compressed_sectors_compressed +=
 286                                                 p.crc.compressed_size;
 287                                         compressed_sectors_uncompressed +=
 288                                                 p.crc.uncompressed_size;
 289                                 }
 290
 291                                 /* only looking at the first ptr */
 292                                 break;
 293                         }
 294                 }
 295
 296         ret = bch2_trans_exit(&trans) ?: ret;
 297         if (ret)
 298                 return ret;
 299
 300         pr_buf(out,
 301                "uncompressed data:\n"
 302                "        nr extents:                     %llu\n"
 303                "        size (bytes):                   %llu\n"
 304                "compressed data:\n"
 305                "        nr extents:                     %llu\n"
 306                "        compressed size (bytes):        %llu\n"
 307                "        uncompressed size (bytes):      %llu\n",
 308                nr_uncompressed_extents,
 309                uncompressed_sectors << 9,
 310                nr_compressed_extents,
 311                compressed_sectors_compressed << 9,
 312                compressed_sectors_uncompressed << 9);
 313         return 0;
 314 }
 315
 316 static void bch2_gc_gens_pos_to_text(struct printbuf *out, struct bch_fs *c)
 317 {
 318         pr_buf(out, "%s: ", bch2_btree_ids[c->gc_gens_btree]);
 319         bch2_bpos_to_text(out, c->gc_gens_pos);
 320         pr_buf(out, "\n");
 321 }
 322
 323 SHOW(bch2_fs)
 324 {
 325         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 326         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 327
 328         sysfs_print(minor,                      c->minor);
 329         sysfs_printf(internal_uuid, "%pU",      c->sb.uuid.b);
 330
 331         sysfs_print(journal_write_delay_ms,     c->journal.write_delay_ms);
 332         sysfs_print(journal_reclaim_delay_ms,   c->journal.reclaim_delay_ms);
 333
 334         sysfs_print(block_size,                 block_bytes(c));
 335         sysfs_print(btree_node_size,            btree_bytes(c));
 336         sysfs_hprint(btree_cache_size,          bch2_btree_cache_size(c));
 337         sysfs_hprint(btree_avg_write_size,      bch2_btree_avg_write_size(c));
 338
 339         sysfs_print(read_realloc_races,
 340                     atomic_long_read(&c->read_realloc_races));
 341         sysfs_print(extent_migrate_done,
 342                     atomic_long_read(&c->extent_migrate_done));
 343         sysfs_print(extent_migrate_raced,
 344                     atomic_long_read(&c->extent_migrate_raced));
 345
 346         sysfs_printf(btree_gc_periodic, "%u",   (int) c->btree_gc_periodic);
 347
 348         if (attr == &sysfs_gc_gens_pos) {
 349                 bch2_gc_gens_pos_to_text(&out, c);
 350                 return out.pos - buf;
 351         }
 352
 353         sysfs_printf(copy_gc_enabled, "%i", c->copy_gc_enabled);
 354
 355         sysfs_printf(rebalance_enabled,         "%i", c->rebalance.enabled);
 356         sysfs_pd_controller_show(rebalance,     &c->rebalance.pd); /* XXX */
 357         sysfs_hprint(copy_gc_wait,
 358                      max(0LL, c->copygc_wait -
 359                          atomic64_read(&c->io_clock[WRITE].now)) << 9);
 360
 361         if (attr == &sysfs_rebalance_work) {
 362                 bch2_rebalance_work_to_text(&out, c);
 363                 return out.pos - buf;
 364         }
 365
 366         sysfs_print(promote_whole_extents,      c->promote_whole_extents);
 367
 368         /* Debugging: */
 369
 370         if (attr == &sysfs_alloc_debug)
 371                 return fs_alloc_debug_to_text(&out, c) ?: out.pos - buf;
 372
 373         if (attr == &sysfs_journal_debug) {
 374                 bch2_journal_debug_to_text(&out, &c->journal);
 375                 return out.pos - buf;
 376         }
 377
 378         if (attr == &sysfs_journal_pins) {
 379                 bch2_journal_pins_to_text(&out, &c->journal);
 380                 return out.pos - buf;
 381         }
 382
 383         if (attr == &sysfs_btree_updates) {
 384                 bch2_btree_updates_to_text(&out, c);
 385                 return out.pos - buf;
 386         }
 387
 388         if (attr == &sysfs_dirty_btree_nodes) {
 389                 bch2_dirty_btree_nodes_to_text(&out, c);
 390                 return out.pos - buf;
 391         }
 392
 393         if (attr == &sysfs_btree_cache) {
 394                 bch2_btree_cache_to_text(&out, c);
 395                 return out.pos - buf;
 396         }
 397
 398         if (attr == &sysfs_btree_key_cache) {
 399                 bch2_btree_key_cache_to_text(&out, &c->btree_key_cache);
 400                 return out.pos - buf;
 401         }
 402
 403         if (attr == &sysfs_btree_transactions) {
 404                 bch2_btree_trans_to_text(&out, c);
 405                 return out.pos - buf;
 406         }
 407
 408         if (attr == &sysfs_stripes_heap) {
 409                 bch2_stripes_heap_to_text(&out, c);
 410                 return out.pos - buf;
 411         }
 412
 413         if (attr == &sysfs_open_buckets) {
 414                 bch2_open_buckets_to_text(&out, c);
 415                 return out.pos - buf;
 416         }
 417
 418         if (attr == &sysfs_compression_stats) {
 419                 bch2_compression_stats_to_text(&out, c);
 420                 return out.pos - buf;
 421         }
 422
 423         if (attr == &sysfs_new_stripes) {
 424                 bch2_new_stripes_to_text(&out, c);
 425                 return out.pos - buf;
 426         }
 427
 428         if (attr == &sysfs_io_timers_read) {
 429                 bch2_io_timers_to_text(&out, &c->io_clock[READ]);
 430                 return out.pos - buf;
 431         }
 432         if (attr == &sysfs_io_timers_write) {
 433                 bch2_io_timers_to_text(&out, &c->io_clock[WRITE]);
 434                 return out.pos - buf;
 435         }
 436
 437         return 0;
 438 }
 439
 440 STORE(bch2_fs)
 441 {
 442         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 443
 444         sysfs_strtoul(journal_write_delay_ms, c->journal.write_delay_ms);
 445         sysfs_strtoul(journal_reclaim_delay_ms, c->journal.reclaim_delay_ms);
 446
 447         if (attr == &sysfs_btree_gc_periodic) {
 448                 ssize_t ret = strtoul_safe(buf, c->btree_gc_periodic)
 449                         ?: (ssize_t) size;
 450
 451                 wake_up_process(c->gc_thread);
 452                 return ret;
 453         }
 454
 455         if (attr == &sysfs_copy_gc_enabled) {
 456                 ssize_t ret = strtoul_safe(buf, c->copy_gc_enabled)
 457                         ?: (ssize_t) size;
 458
 459                 if (c->copygc_thread)
 460                         wake_up_process(c->copygc_thread);
 461                 return ret;
 462         }
 463
 464         if (attr == &sysfs_rebalance_enabled) {
 465                 ssize_t ret = strtoul_safe(buf, c->rebalance.enabled)
 466                         ?: (ssize_t) size;
 467
 468                 rebalance_wakeup(c);
 469                 return ret;
 470         }
 471
 472         sysfs_pd_controller_store(rebalance,    &c->rebalance.pd);
 473
 474         sysfs_strtoul(promote_whole_extents,    c->promote_whole_extents);
 475
 476         /* Debugging: */
 477
 478         if (!test_bit(BCH_FS_STARTED, &c->flags))
 479                 return -EPERM;
 480
 481         /* Debugging: */
 482
 483         if (attr == &sysfs_trigger_journal_flush)
 484                 bch2_journal_meta(&c->journal);
 485
 486         if (attr == &sysfs_trigger_gc) {
 487                 /*
 488                  * Full gc is currently incompatible with btree key cache:
 489                  */
 490 #if 0
 491                 down_read(&c->state_lock);
 492                 bch2_gc(c, false, false);
 493                 up_read(&c->state_lock);
 494 #else
 495                 bch2_gc_gens(c);
 496 #endif
 497         }
 498
 499         if (attr == &sysfs_prune_cache) {
 500                 struct shrink_control sc;
 501
 502                 sc.gfp_mask = GFP_KERNEL;
 503                 sc.nr_to_scan = strtoul_or_return(buf);
 504                 c->btree_cache.shrink.scan_objects(&c->btree_cache.shrink, &sc);
 505         }
 506
 507 #ifdef CONFIG_BCACHEFS_TESTS
 508         if (attr == &sysfs_perf_test) {
 509                 char *tmp = kstrdup(buf, GFP_KERNEL), *p = tmp;
 510                 char *test              = strsep(&p, " \t\n");
 511                 char *nr_str            = strsep(&p, " \t\n");
 512                 char *threads_str       = strsep(&p, " \t\n");
 513                 unsigned threads;
 514                 u64 nr;
 515                 int ret = -EINVAL;
 516
 517                 if (threads_str &&
 518                     !(ret = kstrtouint(threads_str, 10, &threads)) &&
 519                     !(ret = bch2_strtoull_h(nr_str, &nr)))
 520                         ret = bch2_btree_perf_test(c, test, nr, threads);
 521                 kfree(tmp);
 522
 523                 if (ret)
 524                         size = ret;
 525         }
 526 #endif
 527         return size;
 528 }
 529 SYSFS_OPS(bch2_fs);
 530
 531 struct attribute *bch2_fs_files[] = {
 532         &sysfs_minor,
 533         &sysfs_block_size,
 534         &sysfs_btree_node_size,
 535         &sysfs_btree_cache_size,
 536         &sysfs_btree_avg_write_size,
 537
 538         &sysfs_journal_write_delay_ms,
 539         &sysfs_journal_reclaim_delay_ms,
 540
 541         &sysfs_promote_whole_extents,
 542
 543         &sysfs_compression_stats,
 544
 545 #ifdef CONFIG_BCACHEFS_TESTS
 546         &sysfs_perf_test,
 547 #endif
 548         NULL
 549 };
 550
 551 /* internal dir - just a wrapper */
 552
 553 SHOW(bch2_fs_internal)
 554 {
 555         struct bch_fs *c = container_of(kobj, struct bch_fs, internal);
 556         return bch2_fs_show(&c->kobj, attr, buf);
 557 }
 558
 559 STORE(bch2_fs_internal)
 560 {
 561         struct bch_fs *c = container_of(kobj, struct bch_fs, internal);
 562         return bch2_fs_store(&c->kobj, attr, buf, size);
 563 }
 564 SYSFS_OPS(bch2_fs_internal);
 565
 566 struct attribute *bch2_fs_internal_files[] = {
 567         &sysfs_alloc_debug,
 568         &sysfs_journal_debug,
 569         &sysfs_journal_pins,
 570         &sysfs_btree_updates,
 571         &sysfs_dirty_btree_nodes,
 572         &sysfs_btree_cache,
 573         &sysfs_btree_key_cache,
 574         &sysfs_btree_transactions,
 575         &sysfs_stripes_heap,
 576         &sysfs_open_buckets,
 577
 578         &sysfs_read_realloc_races,
 579         &sysfs_extent_migrate_done,
 580         &sysfs_extent_migrate_raced,
 581
 582         &sysfs_trigger_journal_flush,
 583         &sysfs_trigger_gc,
 584         &sysfs_gc_gens_pos,
 585         &sysfs_prune_cache,
 586
 587         &sysfs_copy_gc_enabled,
 588         &sysfs_copy_gc_wait,
 589
 590         &sysfs_rebalance_enabled,
 591         &sysfs_rebalance_work,
 592         sysfs_pd_controller_files(rebalance),
 593
 594         &sysfs_new_stripes,
 595
 596         &sysfs_io_timers_read,
 597         &sysfs_io_timers_write,
 598
 599         &sysfs_internal_uuid,
 600         NULL
 601 };
 602
 603 /* options */
 604
 605 SHOW(bch2_fs_opts_dir)
 606 {
 607         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 608         struct bch_fs *c = container_of(kobj, struct bch_fs, opts_dir);
 609         const struct bch_option *opt = container_of(attr, struct bch_option, attr);
 610         int id = opt - bch2_opt_table;
 611         u64 v = bch2_opt_get_by_id(&c->opts, id);
 612
 613         bch2_opt_to_text(&out, c, opt, v, OPT_SHOW_FULL_LIST);
 614         pr_buf(&out, "\n");
 615
 616         return out.pos - buf;
 617 }
 618
 619 STORE(bch2_fs_opts_dir)
 620 {
 621         struct bch_fs *c = container_of(kobj, struct bch_fs, opts_dir);
 622         const struct bch_option *opt = container_of(attr, struct bch_option, attr);
 623         int ret, id = opt - bch2_opt_table;
 624         char *tmp;
 625         u64 v;
 626
 627         tmp = kstrdup(buf, GFP_KERNEL);
 628         if (!tmp)
 629                 return -ENOMEM;
 630
 631         ret = bch2_opt_parse(c, opt, strim(tmp), &v);
 632         kfree(tmp);
 633
 634         if (ret < 0)
 635                 return ret;
 636
 637         ret = bch2_opt_check_may_set(c, id, v);
 638         if (ret < 0)
 639                 return ret;
 640
 641         if (opt->set_sb != SET_NO_SB_OPT) {
 642                 mutex_lock(&c->sb_lock);
 643                 opt->set_sb(c->disk_sb.sb, v);
 644                 bch2_write_super(c);
 645                 mutex_unlock(&c->sb_lock);
 646         }
 647
 648         bch2_opt_set_by_id(&c->opts, id, v);
 649
 650         if ((id == Opt_background_target ||
 651              id == Opt_background_compression) && v) {
 652                 bch2_rebalance_add_work(c, S64_MAX);
 653                 rebalance_wakeup(c);
 654         }
 655
 656         return size;
 657 }
 658 SYSFS_OPS(bch2_fs_opts_dir);
 659
 660 struct attribute *bch2_fs_opts_dir_files[] = { NULL };
 661
 662 int bch2_opts_create_sysfs_files(struct kobject *kobj)
 663 {
 664         const struct bch_option *i;
 665         int ret;
 666
 667         for (i = bch2_opt_table;
 668              i < bch2_opt_table + bch2_opts_nr;
 669              i++) {
 670                 if (!(i->mode & (OPT_FORMAT|OPT_MOUNT|OPT_RUNTIME)))
 671                         continue;
 672
 673                 ret = sysfs_create_file(kobj, &i->attr);
 674                 if (ret)
 675                         return ret;
 676         }
 677
 678         return 0;
 679 }
 680
 681 /* time stats */
 682
 683 SHOW(bch2_fs_time_stats)
 684 {
 685         struct bch_fs *c = container_of(kobj, struct bch_fs, time_stats);
 686         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 687
 688 #define x(name)                                                         \
 689         if (attr == &sysfs_time_stat_##name) {                          \
 690                 bch2_time_stats_to_text(&out, &c->times[BCH_TIME_##name]);\
 691                 return out.pos - buf;                                   \
 692         }
 693         BCH_TIME_STATS()
 694 #undef x
 695
 696         return 0;
 697 }
 698
 699 STORE(bch2_fs_time_stats)
 700 {
 701         return size;
 702 }
 703 SYSFS_OPS(bch2_fs_time_stats);
 704
 705 struct attribute *bch2_fs_time_stats_files[] = {
 706 #define x(name)                                         \
 707         &sysfs_time_stat_##name,
 708         BCH_TIME_STATS()
 709 #undef x
 710         NULL
 711 };
 712
 713 typedef unsigned (bucket_map_fn)(struct bch_fs *, struct bch_dev *,
 714                                  size_t, void *);
 715
 716 static unsigned bucket_last_io_fn(struct bch_fs *c, struct bch_dev *ca,
 717                                   size_t b, void *private)
 718 {
 719         int rw = (private ? 1 : 0);
 720
 721         return atomic64_read(&c->io_clock[rw].now) - bucket(ca, b)->io_time[rw];
 722 }
 723
 724 static unsigned bucket_sectors_used_fn(struct bch_fs *c, struct bch_dev *ca,
 725                                        size_t b, void *private)
 726 {
 727         struct bucket *g = bucket(ca, b);
 728         return bucket_sectors_used(g->mark);
 729 }
 730
 731 static unsigned bucket_oldest_gen_fn(struct bch_fs *c, struct bch_dev *ca,
 732                                      size_t b, void *private)
 733 {
 734         return bucket_gc_gen(bucket(ca, b));
 735 }
 736
 737 static int unsigned_cmp(const void *_l, const void *_r)
 738 {
 739         const unsigned *l = _l;
 740         const unsigned *r = _r;
 741
 742         return cmp_int(*l, *r);
 743 }
 744
 745 static int quantiles_to_text(struct printbuf *out,
 746                              struct bch_fs *c, struct bch_dev *ca,
 747                              bucket_map_fn *fn, void *private)
 748 {
 749         size_t i, n;
 750         /* Compute 31 quantiles */
 751         unsigned q[31], *p;
 752
 753         down_read(&ca->bucket_lock);
 754         n = ca->mi.nbuckets;
 755
 756         p = vzalloc(n * sizeof(unsigned));
 757         if (!p) {
 758                 up_read(&ca->bucket_lock);
 759                 return -ENOMEM;
 760         }
 761
 762         for (i = ca->mi.first_bucket; i < n; i++)
 763                 p[i] = fn(c, ca, i, private);
 764
 765         sort(p, n, sizeof(unsigned), unsigned_cmp, NULL);
 766         up_read(&ca->bucket_lock);
 767
 768         while (n &&
 769                !p[n - 1])
 770                 --n;
 771
 772         for (i = 0; i < ARRAY_SIZE(q); i++)
 773                 q[i] = p[n * (i + 1) / (ARRAY_SIZE(q) + 1)];
 774
 775         vfree(p);
 776
 777         for (i = 0; i < ARRAY_SIZE(q); i++)
 778                 pr_buf(out, "%u ", q[i]);
 779         pr_buf(out, "\n");
 780         return 0;
 781 }
 782
 783 static void reserve_stats_to_text(struct printbuf *out, struct bch_dev *ca)
 784 {
 785         enum alloc_reserve i;
 786
 787         spin_lock(&ca->fs->freelist_lock);
 788
 789         pr_buf(out, "free_inc:\t%zu\t%zu\n",
 790                fifo_used(&ca->free_inc),
 791                ca->free_inc.size);
 792
 793         for (i = 0; i < RESERVE_NR; i++)
 794                 pr_buf(out, "free[%u]:\t%zu\t%zu\n", i,
 795                        fifo_used(&ca->free[i]),
 796                        ca->free[i].size);
 797
 798         spin_unlock(&ca->fs->freelist_lock);
 799 }
 800
 801 static void dev_alloc_debug_to_text(struct printbuf *out, struct bch_dev *ca)
 802 {
 803         struct bch_fs *c = ca->fs;
 804         struct bch_dev_usage stats = bch2_dev_usage_read(ca);
 805         unsigned i, nr[BCH_DATA_NR];
 806
 807         memset(nr, 0, sizeof(nr));
 808
 809         for (i = 0; i < ARRAY_SIZE(c->open_buckets); i++)
 810                 nr[c->open_buckets[i].type]++;
 811
 812         pr_buf(out,
 813                "\t\t buckets\t sectors      fragmented\n"
 814                "capacity%16llu\n",
 815                ca->mi.nbuckets - ca->mi.first_bucket);
 816
 817         for (i = 1; i < BCH_DATA_NR; i++)
 818                 pr_buf(out, "%-8s%16llu%16llu%16llu\n",
 819                        bch2_data_types[i], stats.d[i].buckets,
 820                        stats.d[i].sectors, stats.d[i].fragmented);
 821
 822         pr_buf(out,
 823                "ec\t%16llu\n"
 824                "available%15llu\n"
 825                "\n"
 826                "free_inc\t\t%zu/%zu\n"
 827                "free[RESERVE_MOVINGGC]\t%zu/%zu\n"
 828                "free[RESERVE_NONE]\t%zu/%zu\n"
 829                "freelist_wait\t\t%s\n"
 830                "open buckets allocated\t%u\n"
 831                "open buckets this dev\t%u\n"
 832                "open buckets total\t%u\n"
 833                "open_buckets_wait\t%s\n"
 834                "open_buckets_btree\t%u\n"
 835                "open_buckets_user\t%u\n"
 836                "btree reserve cache\t%u\n"
 837                "thread state:\t\t%s\n",
 838                stats.buckets_ec,
 839                __dev_buckets_available(ca, stats),
 840                fifo_used(&ca->free_inc),                ca->free_inc.size,
 841                fifo_used(&ca->free[RESERVE_MOVINGGC]),  ca->free[RESERVE_MOVINGGC].size,
 842                fifo_used(&ca->free[RESERVE_NONE]),      ca->free[RESERVE_NONE].size,
 843                c->freelist_wait.list.first              ? "waiting" : "empty",
 844                OPEN_BUCKETS_COUNT - c->open_buckets_nr_free,
 845                ca->nr_open_buckets,
 846                OPEN_BUCKETS_COUNT,
 847                c->open_buckets_wait.list.first          ? "waiting" : "empty",
 848                nr[BCH_DATA_btree],
 849                nr[BCH_DATA_user],
 850                c->btree_reserve_cache_nr,
 851                bch2_allocator_states[ca->allocator_state]);
 852 }
 853
 854 static const char * const bch2_rw[] = {
 855         "read",
 856         "write",
 857         NULL
 858 };
 859
 860 static void dev_iodone_to_text(struct printbuf *out, struct bch_dev *ca)
 861 {
 862         int rw, i;
 863
 864         for (rw = 0; rw < 2; rw++) {
 865                 pr_buf(out, "%s:\n", bch2_rw[rw]);
 866
 867                 for (i = 1; i < BCH_DATA_NR; i++)
 868                         pr_buf(out, "%-12s:%12llu\n",
 869                                bch2_data_types[i],
 870                                percpu_u64_get(&ca->io_done->sectors[rw][i]) << 9);
 871         }
 872 }
 873
 874 SHOW(bch2_dev)
 875 {
 876         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
 877         struct bch_fs *c = ca->fs;
 878         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 879
 880         sysfs_printf(uuid,              "%pU\n", ca->uuid.b);
 881
 882         sysfs_print(bucket_size,        bucket_bytes(ca));
 883         sysfs_print(block_size,         block_bytes(c));
 884         sysfs_print(first_bucket,       ca->mi.first_bucket);
 885         sysfs_print(nbuckets,           ca->mi.nbuckets);
 886         sysfs_print(durability,         ca->mi.durability);
 887         sysfs_print(discard,            ca->mi.discard);
 888
 889         if (attr == &sysfs_label) {
 890                 if (ca->mi.group) {
 891                         mutex_lock(&c->sb_lock);
 892                         bch2_disk_path_to_text(&out, &c->disk_sb,
 893                                                ca->mi.group - 1);
 894                         mutex_unlock(&c->sb_lock);
 895                 }
 896
 897                 pr_buf(&out, "\n");
 898                 return out.pos - buf;
 899         }
 900
 901         if (attr == &sysfs_has_data) {
 902                 bch2_flags_to_text(&out, bch2_data_types,
 903                                    bch2_dev_has_data(c, ca));
 904                 pr_buf(&out, "\n");
 905                 return out.pos - buf;
 906         }
 907
 908         if (attr == &sysfs_cache_replacement_policy) {
 909                 bch2_string_opt_to_text(&out,
 910                                         bch2_cache_replacement_policies,
 911                                         ca->mi.replacement);
 912                 pr_buf(&out, "\n");
 913                 return out.pos - buf;
 914         }
 915
 916         if (attr == &sysfs_state_rw) {
 917                 bch2_string_opt_to_text(&out, bch2_member_states,
 918                                         ca->mi.state);
 919                 pr_buf(&out, "\n");
 920                 return out.pos - buf;
 921         }
 922
 923         if (attr == &sysfs_iodone) {
 924                 dev_iodone_to_text(&out, ca);
 925                 return out.pos - buf;
 926         }
 927
 928         sysfs_print(io_latency_read,            atomic64_read(&ca->cur_latency[READ]));
 929         sysfs_print(io_latency_write,           atomic64_read(&ca->cur_latency[WRITE]));
 930
 931         if (attr == &sysfs_io_latency_stats_read) {
 932                 bch2_time_stats_to_text(&out, &ca->io_latency[READ]);
 933                 return out.pos - buf;
 934         }
 935         if (attr == &sysfs_io_latency_stats_write) {
 936                 bch2_time_stats_to_text(&out, &ca->io_latency[WRITE]);
 937                 return out.pos - buf;
 938         }
 939
 940         sysfs_printf(congested,                 "%u%%",
 941                      clamp(atomic_read(&ca->congested), 0, CONGESTED_MAX)
 942                      * 100 / CONGESTED_MAX);
 943
 944         if (attr == &sysfs_bucket_quantiles_last_read)
 945                 return quantiles_to_text(&out, c, ca, bucket_last_io_fn, (void *) 0) ?: out.pos - buf;
 946         if (attr == &sysfs_bucket_quantiles_last_write)
 947                 return quantiles_to_text(&out, c, ca, bucket_last_io_fn, (void *) 1) ?: out.pos - buf;
 948         if (attr == &sysfs_bucket_quantiles_fragmentation)
 949                 return quantiles_to_text(&out, c, ca, bucket_sectors_used_fn, NULL)  ?: out.pos - buf;
 950         if (attr == &sysfs_bucket_quantiles_oldest_gen)
 951                 return quantiles_to_text(&out, c, ca, bucket_oldest_gen_fn, NULL)    ?: out.pos - buf;
 952
 953         if (attr == &sysfs_reserve_stats) {
 954                 reserve_stats_to_text(&out, ca);
 955                 return out.pos - buf;
 956         }
 957         if (attr == &sysfs_alloc_debug) {
 958                 dev_alloc_debug_to_text(&out, ca);
 959                 return out.pos - buf;
 960         }
 961
 962         return 0;
 963 }
 964
 965 STORE(bch2_dev)
 966 {
 967         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
 968         struct bch_fs *c = ca->fs;
 969         struct bch_member *mi;
 970
 971         if (attr == &sysfs_discard) {
 972                 bool v = strtoul_or_return(buf);
 973
 974                 mutex_lock(&c->sb_lock);
 975                 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
 976
 977                 if (v != BCH_MEMBER_DISCARD(mi)) {
 978                         SET_BCH_MEMBER_DISCARD(mi, v);
 979                         bch2_write_super(c);
 980                 }
 981                 mutex_unlock(&c->sb_lock);
 982         }
 983
 984         if (attr == &sysfs_cache_replacement_policy) {
 985                 ssize_t v = __sysfs_match_string(bch2_cache_replacement_policies, -1, buf);
 986
 987                 if (v < 0)
 988                         return v;
 989
 990                 mutex_lock(&c->sb_lock);
 991                 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
 992
 993                 if ((unsigned) v != BCH_MEMBER_REPLACEMENT(mi)) {
 994                         SET_BCH_MEMBER_REPLACEMENT(mi, v);
 995                         bch2_write_super(c);
 996                 }
 997                 mutex_unlock(&c->sb_lock);
 998         }
 999
1000         if (attr == &sysfs_label) {
1001                 char *tmp;
1002                 int ret;
1003
1004                 tmp = kstrdup(buf, GFP_KERNEL);
1005                 if (!tmp)
1006                         return -ENOMEM;
1007
1008                 ret = bch2_dev_group_set(c, ca, strim(tmp));
1009                 kfree(tmp);
1010                 if (ret)
1011                         return ret;
1012         }
1013
1014         if (attr == &sysfs_wake_allocator)
1015                 bch2_wake_allocator(ca);
1016
1017         return size;
1018 }
1019 SYSFS_OPS(bch2_dev);
1020
1021 struct attribute *bch2_dev_files[] = {
1022         &sysfs_uuid,
1023         &sysfs_bucket_size,
1024         &sysfs_block_size,
1025         &sysfs_first_bucket,
1026         &sysfs_nbuckets,
1027         &sysfs_durability,
1028
1029         /* settings: */
1030         &sysfs_discard,
1031         &sysfs_cache_replacement_policy,
1032         &sysfs_state_rw,
1033         &sysfs_label,
1034
1035         &sysfs_has_data,
1036         &sysfs_iodone,
1037
1038         &sysfs_io_latency_read,
1039         &sysfs_io_latency_write,
1040         &sysfs_io_latency_stats_read,
1041         &sysfs_io_latency_stats_write,
1042         &sysfs_congested,
1043
1044         /* alloc info - other stats: */
1045         &sysfs_bucket_quantiles_last_read,
1046         &sysfs_bucket_quantiles_last_write,
1047         &sysfs_bucket_quantiles_fragmentation,
1048         &sysfs_bucket_quantiles_oldest_gen,
1049
1050         &sysfs_reserve_stats,
1051
1052         /* debug: */
1053         &sysfs_alloc_debug,
1054         &sysfs_wake_allocator,
1055         NULL
1056 };
1057
1058 #endif  /* _BCACHEFS_SYSFS_H_ */