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(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
 175 read_attribute(internal_uuid);
 176
 177 read_attribute(has_data);
 178 read_attribute(alloc_debug);
 179 write_attribute(wake_allocator);
 180
 181 read_attribute(read_realloc_races);
 182 read_attribute(extent_migrate_done);
 183 read_attribute(extent_migrate_raced);
 184
 185 rw_attribute(journal_write_delay_ms);
 186 rw_attribute(journal_reclaim_delay_ms);
 187
 188 rw_attribute(discard);
 189 rw_attribute(cache_replacement_policy);
 190 rw_attribute(label);
 191
 192 rw_attribute(copy_gc_enabled);
 193 sysfs_pd_controller_attribute(copy_gc);
 194
 195 rw_attribute(rebalance_enabled);
 196 sysfs_pd_controller_attribute(rebalance);
 197 read_attribute(rebalance_work);
 198 rw_attribute(promote_whole_extents);
 199
 200 read_attribute(new_stripes);
 201
 202 rw_attribute(pd_controllers_update_seconds);
 203
 204 read_attribute(io_timers_read);
 205 read_attribute(io_timers_write);
 206
 207 #ifdef CONFIG_BCACHEFS_TESTS
 208 write_attribute(perf_test);
 209 #endif /* CONFIG_BCACHEFS_TESTS */
 210
 211 #define x(_name)                                                \
 212         static struct attribute sysfs_time_stat_##_name =               \
 213                 { .name = #_name, .mode = S_IRUGO };
 214         BCH_TIME_STATS()
 215 #undef x
 216
 217 static struct attribute sysfs_state_rw = {
 218         .name = "state",
 219         .mode = S_IRUGO
 220 };
 221
 222 static size_t bch2_btree_cache_size(struct bch_fs *c)
 223 {
 224         size_t ret = 0;
 225         struct btree *b;
 226
 227         mutex_lock(&c->btree_cache.lock);
 228         list_for_each_entry(b, &c->btree_cache.live, list)
 229                 ret += btree_bytes(c);
 230
 231         mutex_unlock(&c->btree_cache.lock);
 232         return ret;
 233 }
 234
 235 static size_t bch2_btree_avg_write_size(struct bch_fs *c)
 236 {
 237         u64 nr = atomic64_read(&c->btree_writes_nr);
 238         u64 sectors = atomic64_read(&c->btree_writes_sectors);
 239
 240         return nr ? div64_u64(sectors, nr) : 0;
 241 }
 242
 243 static int fs_alloc_debug_to_text(struct printbuf *out, struct bch_fs *c)
 244 {
 245         struct bch_fs_usage_online *fs_usage = bch2_fs_usage_read(c);
 246
 247         if (!fs_usage)
 248                 return -ENOMEM;
 249
 250         bch2_fs_usage_to_text(out, c, fs_usage);
 251
 252         percpu_up_read(&c->mark_lock);
 253
 254         kfree(fs_usage);
 255         return 0;
 256 }
 257
 258 static int bch2_compression_stats_to_text(struct printbuf *out, struct bch_fs *c)
 259 {
 260         struct btree_trans trans;
 261         struct btree_iter *iter;
 262         struct bkey_s_c k;
 263         u64 nr_uncompressed_extents = 0, uncompressed_sectors = 0,
 264             nr_compressed_extents = 0,
 265             compressed_sectors_compressed = 0,
 266             compressed_sectors_uncompressed = 0;
 267         int ret;
 268
 269         if (!test_bit(BCH_FS_STARTED, &c->flags))
 270                 return -EPERM;
 271
 272         bch2_trans_init(&trans, c, 0, 0);
 273
 274         for_each_btree_key(&trans, iter, BTREE_ID_extents, POS_MIN, 0, k, ret)
 275                 if (k.k->type == KEY_TYPE_extent) {
 276                         struct bkey_s_c_extent e = bkey_s_c_to_extent(k);
 277                         const union bch_extent_entry *entry;
 278                         struct extent_ptr_decoded p;
 279
 280                         extent_for_each_ptr_decode(e, p, entry) {
 281                                 if (!crc_is_compressed(p.crc)) {
 282                                         nr_uncompressed_extents++;
 283                                         uncompressed_sectors += e.k->size;
 284                                 } else {
 285                                         nr_compressed_extents++;
 286                                         compressed_sectors_compressed +=
 287                                                 p.crc.compressed_size;
 288                                         compressed_sectors_uncompressed +=
 289                                                 p.crc.uncompressed_size;
 290                                 }
 291
 292                                 /* only looking at the first ptr */
 293                                 break;
 294                         }
 295                 }
 296
 297         ret = bch2_trans_exit(&trans) ?: ret;
 298         if (ret)
 299                 return ret;
 300
 301         pr_buf(out,
 302                "uncompressed data:\n"
 303                "        nr extents:                     %llu\n"
 304                "        size (bytes):                   %llu\n"
 305                "compressed data:\n"
 306                "        nr extents:                     %llu\n"
 307                "        compressed size (bytes):        %llu\n"
 308                "        uncompressed size (bytes):      %llu\n",
 309                nr_uncompressed_extents,
 310                uncompressed_sectors << 9,
 311                nr_compressed_extents,
 312                compressed_sectors_compressed << 9,
 313                compressed_sectors_uncompressed << 9);
 314         return 0;
 315 }
 316
 317 SHOW(bch2_fs)
 318 {
 319         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 320         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 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         sysfs_hprint(btree_avg_write_size,      bch2_btree_avg_write_size(c));
 332
 333         sysfs_print(read_realloc_races,
 334                     atomic_long_read(&c->read_realloc_races));
 335         sysfs_print(extent_migrate_done,
 336                     atomic_long_read(&c->extent_migrate_done));
 337         sysfs_print(extent_migrate_raced,
 338                     atomic_long_read(&c->extent_migrate_raced));
 339
 340         sysfs_printf(btree_gc_periodic, "%u",   (int) c->btree_gc_periodic);
 341
 342         sysfs_printf(copy_gc_enabled, "%i", c->copy_gc_enabled);
 343
 344         sysfs_print(pd_controllers_update_seconds,
 345                     c->pd_controllers_update_seconds);
 346
 347         sysfs_printf(rebalance_enabled,         "%i", c->rebalance.enabled);
 348         sysfs_pd_controller_show(rebalance,     &c->rebalance.pd); /* XXX */
 349         sysfs_pd_controller_show(copy_gc,       &c->copygc_pd);
 350
 351         if (attr == &sysfs_rebalance_work) {
 352                 bch2_rebalance_work_to_text(&out, c);
 353                 return out.pos - buf;
 354         }
 355
 356         sysfs_print(promote_whole_extents,      c->promote_whole_extents);
 357
 358         /* Debugging: */
 359
 360         if (attr == &sysfs_alloc_debug)
 361                 return fs_alloc_debug_to_text(&out, c) ?: out.pos - buf;
 362
 363         if (attr == &sysfs_journal_debug) {
 364                 bch2_journal_debug_to_text(&out, &c->journal);
 365                 return out.pos - buf;
 366         }
 367
 368         if (attr == &sysfs_journal_pins) {
 369                 bch2_journal_pins_to_text(&out, &c->journal);
 370                 return out.pos - buf;
 371         }
 372
 373         if (attr == &sysfs_btree_updates) {
 374                 bch2_btree_updates_to_text(&out, c);
 375                 return out.pos - buf;
 376         }
 377
 378         if (attr == &sysfs_dirty_btree_nodes) {
 379                 bch2_dirty_btree_nodes_to_text(&out, c);
 380                 return out.pos - buf;
 381         }
 382
 383         if (attr == &sysfs_btree_cache) {
 384                 bch2_btree_cache_to_text(&out, c);
 385                 return out.pos - buf;
 386         }
 387
 388         if (attr == &sysfs_btree_key_cache) {
 389                 bch2_btree_key_cache_to_text(&out, &c->btree_key_cache);
 390                 return out.pos - buf;
 391         }
 392
 393         if (attr == &sysfs_btree_transactions) {
 394                 bch2_btree_trans_to_text(&out, c);
 395                 return out.pos - buf;
 396         }
 397
 398         if (attr == &sysfs_stripes_heap) {
 399                 bch2_stripes_heap_to_text(&out, c);
 400                 return out.pos - buf;
 401         }
 402
 403         if (attr == &sysfs_compression_stats) {
 404                 bch2_compression_stats_to_text(&out, c);
 405                 return out.pos - buf;
 406         }
 407
 408         if (attr == &sysfs_new_stripes) {
 409                 bch2_new_stripes_to_text(&out, c);
 410                 return out.pos - buf;
 411         }
 412
 413         if (attr == &sysfs_io_timers_read) {
 414                 bch2_io_timers_to_text(&out, &c->io_clock[READ]);
 415                 return out.pos - buf;
 416         }
 417         if (attr == &sysfs_io_timers_write) {
 418                 bch2_io_timers_to_text(&out, &c->io_clock[WRITE]);
 419                 return out.pos - buf;
 420         }
 421
 422         return 0;
 423 }
 424
 425 STORE(bch2_fs)
 426 {
 427         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 428
 429         sysfs_strtoul(journal_write_delay_ms, c->journal.write_delay_ms);
 430         sysfs_strtoul(journal_reclaim_delay_ms, c->journal.reclaim_delay_ms);
 431
 432         if (attr == &sysfs_btree_gc_periodic) {
 433                 ssize_t ret = strtoul_safe(buf, c->btree_gc_periodic)
 434                         ?: (ssize_t) size;
 435
 436                 wake_up_process(c->gc_thread);
 437                 return ret;
 438         }
 439
 440         if (attr == &sysfs_copy_gc_enabled) {
 441                 ssize_t ret = strtoul_safe(buf, c->copy_gc_enabled)
 442                         ?: (ssize_t) size;
 443
 444                 if (c->copygc_thread)
 445                         wake_up_process(c->copygc_thread);
 446                 return ret;
 447         }
 448
 449         if (attr == &sysfs_rebalance_enabled) {
 450                 ssize_t ret = strtoul_safe(buf, c->rebalance.enabled)
 451                         ?: (ssize_t) size;
 452
 453                 rebalance_wakeup(c);
 454                 return ret;
 455         }
 456
 457         sysfs_strtoul(pd_controllers_update_seconds,
 458                       c->pd_controllers_update_seconds);
 459         sysfs_pd_controller_store(rebalance,    &c->rebalance.pd);
 460         sysfs_pd_controller_store(copy_gc,      &c->copygc_pd);
 461
 462         sysfs_strtoul(promote_whole_extents,    c->promote_whole_extents);
 463
 464         /* Debugging: */
 465
 466         if (!test_bit(BCH_FS_STARTED, &c->flags))
 467                 return -EPERM;
 468
 469         /* Debugging: */
 470
 471         if (attr == &sysfs_trigger_journal_flush)
 472                 bch2_journal_meta(&c->journal);
 473
 474         if (attr == &sysfs_trigger_btree_coalesce)
 475                 bch2_coalesce(c);
 476
 477         if (attr == &sysfs_trigger_gc) {
 478                 /*
 479                  * Full gc is currently incompatible with btree key cache:
 480                  */
 481 #if 0
 482                 down_read(&c->state_lock);
 483                 bch2_gc(c, false, false);
 484                 up_read(&c->state_lock);
 485 #else
 486                 bch2_gc_gens(c);
 487 #endif
 488         }
 489
 490         if (attr == &sysfs_prune_cache) {
 491                 struct shrink_control sc;
 492
 493                 sc.gfp_mask = GFP_KERNEL;
 494                 sc.nr_to_scan = strtoul_or_return(buf);
 495                 c->btree_cache.shrink.scan_objects(&c->btree_cache.shrink, &sc);
 496         }
 497
 498 #ifdef CONFIG_BCACHEFS_TESTS
 499         if (attr == &sysfs_perf_test) {
 500                 char *tmp = kstrdup(buf, GFP_KERNEL), *p = tmp;
 501                 char *test              = strsep(&p, " \t\n");
 502                 char *nr_str            = strsep(&p, " \t\n");
 503                 char *threads_str       = strsep(&p, " \t\n");
 504                 unsigned threads;
 505                 u64 nr;
 506                 int ret = -EINVAL;
 507
 508                 if (threads_str &&
 509                     !(ret = kstrtouint(threads_str, 10, &threads)) &&
 510                     !(ret = bch2_strtoull_h(nr_str, &nr)))
 511                         ret = bch2_btree_perf_test(c, test, nr, threads);
 512                 kfree(tmp);
 513
 514                 if (ret)
 515                         size = ret;
 516         }
 517 #endif
 518         return size;
 519 }
 520 SYSFS_OPS(bch2_fs);
 521
 522 struct attribute *bch2_fs_files[] = {
 523         &sysfs_minor,
 524         &sysfs_block_size,
 525         &sysfs_btree_node_size,
 526         &sysfs_btree_cache_size,
 527         &sysfs_btree_avg_write_size,
 528
 529         &sysfs_journal_write_delay_ms,
 530         &sysfs_journal_reclaim_delay_ms,
 531
 532         &sysfs_promote_whole_extents,
 533
 534         &sysfs_compression_stats,
 535
 536 #ifdef CONFIG_BCACHEFS_TESTS
 537         &sysfs_perf_test,
 538 #endif
 539         NULL
 540 };
 541
 542 /* internal dir - just a wrapper */
 543
 544 SHOW(bch2_fs_internal)
 545 {
 546         struct bch_fs *c = container_of(kobj, struct bch_fs, internal);
 547         return bch2_fs_show(&c->kobj, attr, buf);
 548 }
 549
 550 STORE(bch2_fs_internal)
 551 {
 552         struct bch_fs *c = container_of(kobj, struct bch_fs, internal);
 553         return bch2_fs_store(&c->kobj, attr, buf, size);
 554 }
 555 SYSFS_OPS(bch2_fs_internal);
 556
 557 struct attribute *bch2_fs_internal_files[] = {
 558         &sysfs_alloc_debug,
 559         &sysfs_journal_debug,
 560         &sysfs_journal_pins,
 561         &sysfs_btree_updates,
 562         &sysfs_dirty_btree_nodes,
 563         &sysfs_btree_cache,
 564         &sysfs_btree_key_cache,
 565         &sysfs_btree_transactions,
 566         &sysfs_stripes_heap,
 567
 568         &sysfs_read_realloc_races,
 569         &sysfs_extent_migrate_done,
 570         &sysfs_extent_migrate_raced,
 571
 572         &sysfs_trigger_journal_flush,
 573         &sysfs_trigger_btree_coalesce,
 574         &sysfs_trigger_gc,
 575         &sysfs_prune_cache,
 576
 577         &sysfs_copy_gc_enabled,
 578
 579         &sysfs_rebalance_enabled,
 580         &sysfs_rebalance_work,
 581         sysfs_pd_controller_files(rebalance),
 582         sysfs_pd_controller_files(copy_gc),
 583
 584         &sysfs_new_stripes,
 585
 586         &sysfs_io_timers_read,
 587         &sysfs_io_timers_write,
 588
 589         &sysfs_internal_uuid,
 590         NULL
 591 };
 592
 593 /* options */
 594
 595 SHOW(bch2_fs_opts_dir)
 596 {
 597         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 598         struct bch_fs *c = container_of(kobj, struct bch_fs, opts_dir);
 599         const struct bch_option *opt = container_of(attr, struct bch_option, attr);
 600         int id = opt - bch2_opt_table;
 601         u64 v = bch2_opt_get_by_id(&c->opts, id);
 602
 603         bch2_opt_to_text(&out, c, opt, v, OPT_SHOW_FULL_LIST);
 604         pr_buf(&out, "\n");
 605
 606         return out.pos - buf;
 607 }
 608
 609 STORE(bch2_fs_opts_dir)
 610 {
 611         struct bch_fs *c = container_of(kobj, struct bch_fs, opts_dir);
 612         const struct bch_option *opt = container_of(attr, struct bch_option, attr);
 613         int ret, id = opt - bch2_opt_table;
 614         char *tmp;
 615         u64 v;
 616
 617         tmp = kstrdup(buf, GFP_KERNEL);
 618         if (!tmp)
 619                 return -ENOMEM;
 620
 621         ret = bch2_opt_parse(c, opt, strim(tmp), &v);
 622         kfree(tmp);
 623
 624         if (ret < 0)
 625                 return ret;
 626
 627         ret = bch2_opt_check_may_set(c, id, v);
 628         if (ret < 0)
 629                 return ret;
 630
 631         if (opt->set_sb != SET_NO_SB_OPT) {
 632                 mutex_lock(&c->sb_lock);
 633                 opt->set_sb(c->disk_sb.sb, v);
 634                 bch2_write_super(c);
 635                 mutex_unlock(&c->sb_lock);
 636         }
 637
 638         bch2_opt_set_by_id(&c->opts, id, v);
 639
 640         if ((id == Opt_background_target ||
 641              id == Opt_background_compression) && v) {
 642                 bch2_rebalance_add_work(c, S64_MAX);
 643                 rebalance_wakeup(c);
 644         }
 645
 646         return size;
 647 }
 648 SYSFS_OPS(bch2_fs_opts_dir);
 649
 650 struct attribute *bch2_fs_opts_dir_files[] = { NULL };
 651
 652 int bch2_opts_create_sysfs_files(struct kobject *kobj)
 653 {
 654         const struct bch_option *i;
 655         int ret;
 656
 657         for (i = bch2_opt_table;
 658              i < bch2_opt_table + bch2_opts_nr;
 659              i++) {
 660                 if (!(i->mode & (OPT_FORMAT|OPT_MOUNT|OPT_RUNTIME)))
 661                         continue;
 662
 663                 ret = sysfs_create_file(kobj, &i->attr);
 664                 if (ret)
 665                         return ret;
 666         }
 667
 668         return 0;
 669 }
 670
 671 /* time stats */
 672
 673 SHOW(bch2_fs_time_stats)
 674 {
 675         struct bch_fs *c = container_of(kobj, struct bch_fs, time_stats);
 676         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 677
 678 #define x(name)                                                         \
 679         if (attr == &sysfs_time_stat_##name) {                          \
 680                 bch2_time_stats_to_text(&out, &c->times[BCH_TIME_##name]);\
 681                 return out.pos - buf;                                   \
 682         }
 683         BCH_TIME_STATS()
 684 #undef x
 685
 686         return 0;
 687 }
 688
 689 STORE(bch2_fs_time_stats)
 690 {
 691         return size;
 692 }
 693 SYSFS_OPS(bch2_fs_time_stats);
 694
 695 struct attribute *bch2_fs_time_stats_files[] = {
 696 #define x(name)                                         \
 697         &sysfs_time_stat_##name,
 698         BCH_TIME_STATS()
 699 #undef x
 700         NULL
 701 };
 702
 703 typedef unsigned (bucket_map_fn)(struct bch_fs *, struct bch_dev *,
 704                                  size_t, void *);
 705
 706 static unsigned bucket_last_io_fn(struct bch_fs *c, struct bch_dev *ca,
 707                                   size_t b, void *private)
 708 {
 709         int rw = (private ? 1 : 0);
 710
 711         return atomic64_read(&c->io_clock[rw].now) - bucket(ca, b)->io_time[rw];
 712 }
 713
 714 static unsigned bucket_sectors_used_fn(struct bch_fs *c, struct bch_dev *ca,
 715                                        size_t b, void *private)
 716 {
 717         struct bucket *g = bucket(ca, b);
 718         return bucket_sectors_used(g->mark);
 719 }
 720
 721 static unsigned bucket_oldest_gen_fn(struct bch_fs *c, struct bch_dev *ca,
 722                                      size_t b, void *private)
 723 {
 724         return bucket_gc_gen(bucket(ca, b));
 725 }
 726
 727 static int unsigned_cmp(const void *_l, const void *_r)
 728 {
 729         const unsigned *l = _l;
 730         const unsigned *r = _r;
 731
 732         return cmp_int(*l, *r);
 733 }
 734
 735 static int quantiles_to_text(struct printbuf *out,
 736                              struct bch_fs *c, struct bch_dev *ca,
 737                              bucket_map_fn *fn, void *private)
 738 {
 739         size_t i, n;
 740         /* Compute 31 quantiles */
 741         unsigned q[31], *p;
 742
 743         down_read(&ca->bucket_lock);
 744         n = ca->mi.nbuckets;
 745
 746         p = vzalloc(n * sizeof(unsigned));
 747         if (!p) {
 748                 up_read(&ca->bucket_lock);
 749                 return -ENOMEM;
 750         }
 751
 752         for (i = ca->mi.first_bucket; i < n; i++)
 753                 p[i] = fn(c, ca, i, private);
 754
 755         sort(p, n, sizeof(unsigned), unsigned_cmp, NULL);
 756         up_read(&ca->bucket_lock);
 757
 758         while (n &&
 759                !p[n - 1])
 760                 --n;
 761
 762         for (i = 0; i < ARRAY_SIZE(q); i++)
 763                 q[i] = p[n * (i + 1) / (ARRAY_SIZE(q) + 1)];
 764
 765         vfree(p);
 766
 767         for (i = 0; i < ARRAY_SIZE(q); i++)
 768                 pr_buf(out, "%u ", q[i]);
 769         pr_buf(out, "\n");
 770         return 0;
 771 }
 772
 773 static void reserve_stats_to_text(struct printbuf *out, struct bch_dev *ca)
 774 {
 775         enum alloc_reserve i;
 776
 777         spin_lock(&ca->fs->freelist_lock);
 778
 779         pr_buf(out, "free_inc:\t%zu\t%zu\n",
 780                fifo_used(&ca->free_inc),
 781                ca->free_inc.size);
 782
 783         for (i = 0; i < RESERVE_NR; i++)
 784                 pr_buf(out, "free[%u]:\t%zu\t%zu\n", i,
 785                        fifo_used(&ca->free[i]),
 786                        ca->free[i].size);
 787
 788         spin_unlock(&ca->fs->freelist_lock);
 789 }
 790
 791 static void dev_alloc_debug_to_text(struct printbuf *out, struct bch_dev *ca)
 792 {
 793         struct bch_fs *c = ca->fs;
 794         struct bch_dev_usage stats = bch2_dev_usage_read(ca);
 795         unsigned i, nr[BCH_DATA_NR];
 796
 797         memset(nr, 0, sizeof(nr));
 798
 799         for (i = 0; i < ARRAY_SIZE(c->open_buckets); i++)
 800                 nr[c->open_buckets[i].type]++;
 801
 802         pr_buf(out,
 803                "\t\t buckets\t sectors      fragmented\n"
 804                "capacity%16llu\n",
 805                ca->mi.nbuckets - ca->mi.first_bucket);
 806
 807         for (i = 1; i < BCH_DATA_NR; i++)
 808                 pr_buf(out, "%-8s%16llu%16llu%16llu\n",
 809                        bch2_data_types[i], stats.d[i].buckets,
 810                        stats.d[i].sectors, stats.d[i].fragmented);
 811
 812         pr_buf(out,
 813                "ec\t%16llu\n"
 814                "available%15llu\n"
 815                "\n"
 816                "free_inc\t\t%zu/%zu\n"
 817                "free[RESERVE_MOVINGGC]\t%zu/%zu\n"
 818                "free[RESERVE_NONE]\t%zu/%zu\n"
 819                "freelist_wait\t\t%s\n"
 820                "open buckets\t\t%u/%u (reserved %u)\n"
 821                "open_buckets_wait\t%s\n"
 822                "open_buckets_btree\t%u\n"
 823                "open_buckets_user\t%u\n"
 824                "btree reserve cache\t%u\n",
 825                stats.buckets_ec,
 826                __dev_buckets_available(ca, stats),
 827                fifo_used(&ca->free_inc),                ca->free_inc.size,
 828                fifo_used(&ca->free[RESERVE_MOVINGGC]),  ca->free[RESERVE_MOVINGGC].size,
 829                fifo_used(&ca->free[RESERVE_NONE]),      ca->free[RESERVE_NONE].size,
 830                c->freelist_wait.list.first              ? "waiting" : "empty",
 831                c->open_buckets_nr_free, OPEN_BUCKETS_COUNT,
 832                BTREE_NODE_OPEN_BUCKET_RESERVE,
 833                c->open_buckets_wait.list.first          ? "waiting" : "empty",
 834                nr[BCH_DATA_btree],
 835                nr[BCH_DATA_user],
 836                c->btree_reserve_cache_nr);
 837 }
 838
 839 static const char * const bch2_rw[] = {
 840         "read",
 841         "write",
 842         NULL
 843 };
 844
 845 static void dev_iodone_to_text(struct printbuf *out, struct bch_dev *ca)
 846 {
 847         int rw, i;
 848
 849         for (rw = 0; rw < 2; rw++) {
 850                 pr_buf(out, "%s:\n", bch2_rw[rw]);
 851
 852                 for (i = 1; i < BCH_DATA_NR; i++)
 853                         pr_buf(out, "%-12s:%12llu\n",
 854                                bch2_data_types[i],
 855                                percpu_u64_get(&ca->io_done->sectors[rw][i]) << 9);
 856         }
 857 }
 858
 859 SHOW(bch2_dev)
 860 {
 861         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
 862         struct bch_fs *c = ca->fs;
 863         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 864
 865         sysfs_printf(uuid,              "%pU\n", ca->uuid.b);
 866
 867         sysfs_print(bucket_size,        bucket_bytes(ca));
 868         sysfs_print(block_size,         block_bytes(c));
 869         sysfs_print(first_bucket,       ca->mi.first_bucket);
 870         sysfs_print(nbuckets,           ca->mi.nbuckets);
 871         sysfs_print(durability,         ca->mi.durability);
 872         sysfs_print(discard,            ca->mi.discard);
 873
 874         if (attr == &sysfs_label) {
 875                 if (ca->mi.group) {
 876                         mutex_lock(&c->sb_lock);
 877                         bch2_disk_path_to_text(&out, &c->disk_sb,
 878                                                ca->mi.group - 1);
 879                         mutex_unlock(&c->sb_lock);
 880                 }
 881
 882                 pr_buf(&out, "\n");
 883                 return out.pos - buf;
 884         }
 885
 886         if (attr == &sysfs_has_data) {
 887                 bch2_flags_to_text(&out, bch2_data_types,
 888                                    bch2_dev_has_data(c, ca));
 889                 pr_buf(&out, "\n");
 890                 return out.pos - buf;
 891         }
 892
 893         if (attr == &sysfs_cache_replacement_policy) {
 894                 bch2_string_opt_to_text(&out,
 895                                         bch2_cache_replacement_policies,
 896                                         ca->mi.replacement);
 897                 pr_buf(&out, "\n");
 898                 return out.pos - buf;
 899         }
 900
 901         if (attr == &sysfs_state_rw) {
 902                 bch2_string_opt_to_text(&out, bch2_member_states,
 903                                         ca->mi.state);
 904                 pr_buf(&out, "\n");
 905                 return out.pos - buf;
 906         }
 907
 908         if (attr == &sysfs_iodone) {
 909                 dev_iodone_to_text(&out, ca);
 910                 return out.pos - buf;
 911         }
 912
 913         sysfs_print(io_latency_read,            atomic64_read(&ca->cur_latency[READ]));
 914         sysfs_print(io_latency_write,           atomic64_read(&ca->cur_latency[WRITE]));
 915
 916         if (attr == &sysfs_io_latency_stats_read) {
 917                 bch2_time_stats_to_text(&out, &ca->io_latency[READ]);
 918                 return out.pos - buf;
 919         }
 920         if (attr == &sysfs_io_latency_stats_write) {
 921                 bch2_time_stats_to_text(&out, &ca->io_latency[WRITE]);
 922                 return out.pos - buf;
 923         }
 924
 925         sysfs_printf(congested,                 "%u%%",
 926                      clamp(atomic_read(&ca->congested), 0, CONGESTED_MAX)
 927                      * 100 / CONGESTED_MAX);
 928
 929         if (attr == &sysfs_bucket_quantiles_last_read)
 930                 return quantiles_to_text(&out, c, ca, bucket_last_io_fn, (void *) 0) ?: out.pos - buf;
 931         if (attr == &sysfs_bucket_quantiles_last_write)
 932                 return quantiles_to_text(&out, c, ca, bucket_last_io_fn, (void *) 1) ?: out.pos - buf;
 933         if (attr == &sysfs_bucket_quantiles_fragmentation)
 934                 return quantiles_to_text(&out, c, ca, bucket_sectors_used_fn, NULL)  ?: out.pos - buf;
 935         if (attr == &sysfs_bucket_quantiles_oldest_gen)
 936                 return quantiles_to_text(&out, c, ca, bucket_oldest_gen_fn, NULL)    ?: out.pos - buf;
 937
 938         if (attr == &sysfs_reserve_stats) {
 939                 reserve_stats_to_text(&out, ca);
 940                 return out.pos - buf;
 941         }
 942         if (attr == &sysfs_alloc_debug) {
 943                 dev_alloc_debug_to_text(&out, ca);
 944                 return out.pos - buf;
 945         }
 946
 947         return 0;
 948 }
 949
 950 STORE(bch2_dev)
 951 {
 952         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
 953         struct bch_fs *c = ca->fs;
 954         struct bch_member *mi;
 955
 956         if (attr == &sysfs_discard) {
 957                 bool v = strtoul_or_return(buf);
 958
 959                 mutex_lock(&c->sb_lock);
 960                 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
 961
 962                 if (v != BCH_MEMBER_DISCARD(mi)) {
 963                         SET_BCH_MEMBER_DISCARD(mi, v);
 964                         bch2_write_super(c);
 965                 }
 966                 mutex_unlock(&c->sb_lock);
 967         }
 968
 969         if (attr == &sysfs_cache_replacement_policy) {
 970                 ssize_t v = __sysfs_match_string(bch2_cache_replacement_policies, -1, buf);
 971
 972                 if (v < 0)
 973                         return v;
 974
 975                 mutex_lock(&c->sb_lock);
 976                 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
 977
 978                 if ((unsigned) v != BCH_MEMBER_REPLACEMENT(mi)) {
 979                         SET_BCH_MEMBER_REPLACEMENT(mi, v);
 980                         bch2_write_super(c);
 981                 }
 982                 mutex_unlock(&c->sb_lock);
 983         }
 984
 985         if (attr == &sysfs_label) {
 986                 char *tmp;
 987                 int ret;
 988
 989                 tmp = kstrdup(buf, GFP_KERNEL);
 990                 if (!tmp)
 991                         return -ENOMEM;
 992
 993                 ret = bch2_dev_group_set(c, ca, strim(tmp));
 994                 kfree(tmp);
 995                 if (ret)
 996                         return ret;
 997         }
 998
 999         if (attr == &sysfs_wake_allocator)
1000                 bch2_wake_allocator(ca);
1001
1002         return size;
1003 }
1004 SYSFS_OPS(bch2_dev);
1005
1006 struct attribute *bch2_dev_files[] = {
1007         &sysfs_uuid,
1008         &sysfs_bucket_size,
1009         &sysfs_block_size,
1010         &sysfs_first_bucket,
1011         &sysfs_nbuckets,
1012         &sysfs_durability,
1013
1014         /* settings: */
1015         &sysfs_discard,
1016         &sysfs_cache_replacement_policy,
1017         &sysfs_state_rw,
1018         &sysfs_label,
1019
1020         &sysfs_has_data,
1021         &sysfs_iodone,
1022
1023         &sysfs_io_latency_read,
1024         &sysfs_io_latency_write,
1025         &sysfs_io_latency_stats_read,
1026         &sysfs_io_latency_stats_write,
1027         &sysfs_congested,
1028
1029         /* alloc info - other stats: */
1030         &sysfs_bucket_quantiles_last_read,
1031         &sysfs_bucket_quantiles_last_write,
1032         &sysfs_bucket_quantiles_fragmentation,
1033         &sysfs_bucket_quantiles_oldest_gen,
1034
1035         &sysfs_reserve_stats,
1036
1037         /* debug: */
1038         &sysfs_alloc_debug,
1039         &sysfs_wake_allocator,
1040         NULL
1041 };
1042
1043 #endif  /* _BCACHEFS_SYSFS_H_ */