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_update.h"
  18 #include "btree_update_interior.h"
  19 #include "btree_gc.h"
  20 #include "buckets.h"
  21 #include "clock.h"
  22 #include "disk_groups.h"
  23 #include "ec.h"
  24 #include "inode.h"
  25 #include "journal.h"
  26 #include "keylist.h"
  27 #include "move.h"
  28 #include "opts.h"
  29 #include "rebalance.h"
  30 #include "replicas.h"
  31 #include "super-io.h"
  32 #include "tests.h"
  33
  34 #include <linux/blkdev.h>
  35 #include <linux/sort.h>
  36 #include <linux/sched/clock.h>
  37
  38 #include "util.h"
  39
  40 #define SYSFS_OPS(type)                                                 \
  41 struct sysfs_ops type ## _sysfs_ops = {                                 \
  42         .show   = type ## _show,                                        \
  43         .store  = type ## _store                                        \
  44 }
  45
  46 #define SHOW(fn)                                                        \
  47 static ssize_t fn ## _show(struct kobject *kobj, struct attribute *attr,\
  48                            char *buf)                                   \
  49
  50 #define STORE(fn)                                                       \
  51 static ssize_t fn ## _store(struct kobject *kobj, struct attribute *attr,\
  52                             const char *buf, size_t size)               \
  53
  54 #define __sysfs_attribute(_name, _mode)                                 \
  55         static struct attribute sysfs_##_name =                         \
  56                 { .name = #_name, .mode = _mode }
  57
  58 #define write_attribute(n)      __sysfs_attribute(n, S_IWUSR)
  59 #define read_attribute(n)       __sysfs_attribute(n, S_IRUGO)
  60 #define rw_attribute(n)         __sysfs_attribute(n, S_IRUGO|S_IWUSR)
  61
  62 #define sysfs_printf(file, fmt, ...)                                    \
  63 do {                                                                    \
  64         if (attr == &sysfs_ ## file)                                    \
  65                 return scnprintf(buf, PAGE_SIZE, fmt "\n", __VA_ARGS__);\
  66 } while (0)
  67
  68 #define sysfs_print(file, var)                                          \
  69 do {                                                                    \
  70         if (attr == &sysfs_ ## file)                                    \
  71                 return snprint(buf, PAGE_SIZE, var);                    \
  72 } while (0)
  73
  74 #define sysfs_hprint(file, val)                                         \
  75 do {                                                                    \
  76         if (attr == &sysfs_ ## file) {                                  \
  77                 struct printbuf out = _PBUF(buf, PAGE_SIZE);            \
  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(trigger_alloc_write);
 138 write_attribute(prune_cache);
 139 rw_attribute(btree_gc_periodic);
 140
 141 read_attribute(uuid);
 142 read_attribute(minor);
 143 read_attribute(bucket_size);
 144 read_attribute(block_size);
 145 read_attribute(btree_node_size);
 146 read_attribute(first_bucket);
 147 read_attribute(nbuckets);
 148 read_attribute(durability);
 149 read_attribute(iodone);
 150
 151 read_attribute(io_latency_read);
 152 read_attribute(io_latency_write);
 153 read_attribute(io_latency_stats_read);
 154 read_attribute(io_latency_stats_write);
 155 read_attribute(congested);
 156
 157 read_attribute(bucket_quantiles_last_read);
 158 read_attribute(bucket_quantiles_last_write);
 159 read_attribute(bucket_quantiles_fragmentation);
 160 read_attribute(bucket_quantiles_oldest_gen);
 161
 162 read_attribute(reserve_stats);
 163 read_attribute(btree_cache_size);
 164 read_attribute(compression_stats);
 165 read_attribute(journal_debug);
 166 read_attribute(journal_pins);
 167 read_attribute(btree_updates);
 168 read_attribute(dirty_btree_nodes);
 169
 170 read_attribute(internal_uuid);
 171
 172 read_attribute(has_data);
 173 read_attribute(alloc_debug);
 174 write_attribute(wake_allocator);
 175
 176 read_attribute(read_realloc_races);
 177 read_attribute(extent_migrate_done);
 178 read_attribute(extent_migrate_raced);
 179
 180 rw_attribute(journal_write_delay_ms);
 181 rw_attribute(journal_reclaim_delay_ms);
 182
 183 rw_attribute(discard);
 184 rw_attribute(cache_replacement_policy);
 185 rw_attribute(label);
 186
 187 rw_attribute(copy_gc_enabled);
 188 sysfs_pd_controller_attribute(copy_gc);
 189
 190 rw_attribute(rebalance_enabled);
 191 sysfs_pd_controller_attribute(rebalance);
 192 read_attribute(rebalance_work);
 193 rw_attribute(promote_whole_extents);
 194
 195 read_attribute(new_stripes);
 196
 197 rw_attribute(pd_controllers_update_seconds);
 198
 199 read_attribute(meta_replicas_have);
 200 read_attribute(data_replicas_have);
 201
 202 read_attribute(io_timers_read);
 203 read_attribute(io_timers_write);
 204
 205 #ifdef CONFIG_BCACHEFS_TESTS
 206 write_attribute(perf_test);
 207 #endif /* CONFIG_BCACHEFS_TESTS */
 208
 209 #define BCH_DEBUG_PARAM(name, description)                              \
 210         rw_attribute(name);
 211
 212         BCH_DEBUG_PARAMS()
 213 #undef BCH_DEBUG_PARAM
 214
 215 #define x(_name)                                                \
 216         static struct attribute sysfs_time_stat_##_name =               \
 217                 { .name = #_name, .mode = S_IRUGO };
 218         BCH_TIME_STATS()
 219 #undef x
 220
 221 static struct attribute sysfs_state_rw = {
 222         .name = "state",
 223         .mode = S_IRUGO
 224 };
 225
 226 static size_t bch2_btree_cache_size(struct bch_fs *c)
 227 {
 228         size_t ret = 0;
 229         struct btree *b;
 230
 231         mutex_lock(&c->btree_cache.lock);
 232         list_for_each_entry(b, &c->btree_cache.live, list)
 233                 ret += btree_bytes(c);
 234
 235         mutex_unlock(&c->btree_cache.lock);
 236         return ret;
 237 }
 238
 239 static ssize_t show_fs_alloc_debug(struct bch_fs *c, char *buf)
 240 {
 241         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 242         struct bch_fs_usage *fs_usage = bch2_fs_usage_read(c);
 243
 244         if (!fs_usage)
 245                 return -ENOMEM;
 246
 247         bch2_fs_usage_to_text(&out, c, fs_usage);
 248
 249         percpu_up_read(&c->mark_lock);
 250
 251         kfree(fs_usage);
 252
 253         return out.pos - buf;
 254 }
 255
 256 static ssize_t bch2_compression_stats(struct bch_fs *c, char *buf)
 257 {
 258         struct btree_trans trans;
 259         struct btree_iter *iter;
 260         struct bkey_s_c k;
 261         u64 nr_uncompressed_extents = 0, uncompressed_sectors = 0,
 262             nr_compressed_extents = 0,
 263             compressed_sectors_compressed = 0,
 264             compressed_sectors_uncompressed = 0;
 265         int ret;
 266
 267         if (!test_bit(BCH_FS_STARTED, &c->flags))
 268                 return -EPERM;
 269
 270         bch2_trans_init(&trans, c, 0, 0);
 271
 272         for_each_btree_key(&trans, iter, BTREE_ID_EXTENTS, POS_MIN, 0, k, ret)
 273                 if (k.k->type == KEY_TYPE_extent) {
 274                         struct bkey_s_c_extent e = bkey_s_c_to_extent(k);
 275                         const union bch_extent_entry *entry;
 276                         struct extent_ptr_decoded p;
 277
 278                         extent_for_each_ptr_decode(e, p, entry) {
 279                                 if (!crc_is_compressed(p.crc)) {
 280                                         nr_uncompressed_extents++;
 281                                         uncompressed_sectors += e.k->size;
 282                                 } else {
 283                                         nr_compressed_extents++;
 284                                         compressed_sectors_compressed +=
 285                                                 p.crc.compressed_size;
 286                                         compressed_sectors_uncompressed +=
 287                                                 p.crc.uncompressed_size;
 288                                 }
 289
 290                                 /* only looking at the first ptr */
 291                                 break;
 292                         }
 293                 }
 294
 295         ret = bch2_trans_exit(&trans) ?: ret;
 296         if (ret)
 297                 return ret;
 298
 299         return scnprintf(buf, PAGE_SIZE,
 300                         "uncompressed data:\n"
 301                         "       nr extents:                     %llu\n"
 302                         "       size (bytes):                   %llu\n"
 303                         "compressed data:\n"
 304                         "       nr extents:                     %llu\n"
 305                         "       compressed size (bytes):        %llu\n"
 306                         "       uncompressed size (bytes):      %llu\n",
 307                         nr_uncompressed_extents,
 308                         uncompressed_sectors << 9,
 309                         nr_compressed_extents,
 310                         compressed_sectors_compressed << 9,
 311                         compressed_sectors_uncompressed << 9);
 312 }
 313
 314 static ssize_t bch2_new_stripes(struct bch_fs *c, char *buf)
 315 {
 316         char *out = buf, *end = buf + PAGE_SIZE;
 317         struct ec_stripe_head *h;
 318         struct ec_stripe_new *s;
 319
 320         mutex_lock(&c->ec_new_stripe_lock);
 321         list_for_each_entry(h, &c->ec_new_stripe_list, list) {
 322                 out += scnprintf(out, end - out,
 323                                  "target %u algo %u redundancy %u:\n",
 324                                  h->target, h->algo, h->redundancy);
 325
 326                 if (h->s)
 327                         out += scnprintf(out, end - out,
 328                                          "\tpending: blocks %u allocated %u\n",
 329                                          h->s->blocks.nr,
 330                                          bitmap_weight(h->s->blocks_allocated,
 331                                                        h->s->blocks.nr));
 332
 333                 mutex_lock(&h->lock);
 334                 list_for_each_entry(s, &h->stripes, list)
 335                         out += scnprintf(out, end - out,
 336                                          "\tin flight: blocks %u allocated %u pin %u\n",
 337                                          s->blocks.nr,
 338                                          bitmap_weight(s->blocks_allocated,
 339                                                        s->blocks.nr),
 340                                          atomic_read(&s->pin));
 341                 mutex_unlock(&h->lock);
 342
 343         }
 344         mutex_unlock(&c->ec_new_stripe_lock);
 345
 346         return out - buf;
 347 }
 348
 349 SHOW(bch2_fs)
 350 {
 351         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 352
 353         sysfs_print(minor,                      c->minor);
 354         sysfs_printf(internal_uuid, "%pU",      c->sb.uuid.b);
 355
 356         sysfs_print(journal_write_delay_ms,     c->journal.write_delay_ms);
 357         sysfs_print(journal_reclaim_delay_ms,   c->journal.reclaim_delay_ms);
 358
 359         sysfs_print(block_size,                 block_bytes(c));
 360         sysfs_print(btree_node_size,            btree_bytes(c));
 361         sysfs_hprint(btree_cache_size,          bch2_btree_cache_size(c));
 362
 363         sysfs_print(read_realloc_races,
 364                     atomic_long_read(&c->read_realloc_races));
 365         sysfs_print(extent_migrate_done,
 366                     atomic_long_read(&c->extent_migrate_done));
 367         sysfs_print(extent_migrate_raced,
 368                     atomic_long_read(&c->extent_migrate_raced));
 369
 370         sysfs_printf(btree_gc_periodic, "%u",   (int) c->btree_gc_periodic);
 371
 372         sysfs_printf(copy_gc_enabled, "%i", c->copy_gc_enabled);
 373
 374         sysfs_print(pd_controllers_update_seconds,
 375                     c->pd_controllers_update_seconds);
 376
 377         sysfs_printf(rebalance_enabled,         "%i", c->rebalance.enabled);
 378         sysfs_pd_controller_show(rebalance,     &c->rebalance.pd); /* XXX */
 379
 380         if (attr == &sysfs_rebalance_work)
 381                 return bch2_rebalance_work_show(c, buf);
 382
 383         sysfs_print(promote_whole_extents,      c->promote_whole_extents);
 384
 385         sysfs_printf(meta_replicas_have, "%i",  bch2_replicas_online(c, true));
 386         sysfs_printf(data_replicas_have, "%i",  bch2_replicas_online(c, false));
 387
 388         /* Debugging: */
 389
 390         if (attr == &sysfs_alloc_debug)
 391                 return show_fs_alloc_debug(c, buf);
 392
 393         if (attr == &sysfs_journal_debug)
 394                 return bch2_journal_print_debug(&c->journal, buf);
 395
 396         if (attr == &sysfs_journal_pins)
 397                 return bch2_journal_print_pins(&c->journal, buf);
 398
 399         if (attr == &sysfs_btree_updates)
 400                 return bch2_btree_updates_print(c, buf);
 401
 402         if (attr == &sysfs_dirty_btree_nodes)
 403                 return bch2_dirty_btree_nodes_print(c, buf);
 404
 405         if (attr == &sysfs_compression_stats)
 406                 return bch2_compression_stats(c, buf);
 407
 408         if (attr == &sysfs_new_stripes)
 409                 return bch2_new_stripes(c, buf);
 410
 411         if (attr == &sysfs_io_timers_read)
 412                 return bch2_io_timers_show(&c->io_clock[READ], buf);
 413         if (attr == &sysfs_io_timers_write)
 414                 return bch2_io_timers_show(&c->io_clock[WRITE], buf);
 415
 416 #define BCH_DEBUG_PARAM(name, description) sysfs_print(name, c->name);
 417         BCH_DEBUG_PARAMS()
 418 #undef BCH_DEBUG_PARAM
 419
 420         return 0;
 421 }
 422
 423 STORE(__bch2_fs)
 424 {
 425         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 426
 427         sysfs_strtoul(journal_write_delay_ms, c->journal.write_delay_ms);
 428         sysfs_strtoul(journal_reclaim_delay_ms, c->journal.reclaim_delay_ms);
 429
 430         if (attr == &sysfs_btree_gc_periodic) {
 431                 ssize_t ret = strtoul_safe(buf, c->btree_gc_periodic)
 432                         ?: (ssize_t) size;
 433
 434                 wake_up_process(c->gc_thread);
 435                 return ret;
 436         }
 437
 438         if (attr == &sysfs_copy_gc_enabled) {
 439                 struct bch_dev *ca;
 440                 unsigned i;
 441                 ssize_t ret = strtoul_safe(buf, c->copy_gc_enabled)
 442                         ?: (ssize_t) size;
 443
 444                 for_each_member_device(ca, c, i)
 445                         if (ca->copygc_thread)
 446                                 wake_up_process(ca->copygc_thread);
 447                 return ret;
 448         }
 449
 450         if (attr == &sysfs_rebalance_enabled) {
 451                 ssize_t ret = strtoul_safe(buf, c->rebalance.enabled)
 452                         ?: (ssize_t) size;
 453
 454                 rebalance_wakeup(c);
 455                 return ret;
 456         }
 457
 458         sysfs_strtoul(pd_controllers_update_seconds,
 459                       c->pd_controllers_update_seconds);
 460         sysfs_pd_controller_store(rebalance,    &c->rebalance.pd);
 461
 462         sysfs_strtoul(promote_whole_extents,    c->promote_whole_extents);
 463
 464         /* Debugging: */
 465
 466 #define BCH_DEBUG_PARAM(name, description) sysfs_strtoul(name, c->name);
 467         BCH_DEBUG_PARAMS()
 468 #undef BCH_DEBUG_PARAM
 469
 470         if (!test_bit(BCH_FS_STARTED, &c->flags))
 471                 return -EPERM;
 472
 473         /* Debugging: */
 474
 475         if (attr == &sysfs_trigger_journal_flush)
 476                 bch2_journal_meta_async(&c->journal, NULL);
 477
 478         if (attr == &sysfs_trigger_btree_coalesce)
 479                 bch2_coalesce(c);
 480
 481         if (attr == &sysfs_trigger_gc)
 482                 bch2_gc(c, NULL, false, false);
 483
 484         if (attr == &sysfs_trigger_alloc_write) {
 485                 bool wrote;
 486
 487                 bch2_alloc_write(c, 0, &wrote);
 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 #ifdef CONFIG_BCACHEFS_TESTS
 498         if (attr == &sysfs_perf_test) {
 499                 char *tmp = kstrdup(buf, GFP_KERNEL), *p = tmp;
 500                 char *test              = strsep(&p, " \t\n");
 501                 char *nr_str            = strsep(&p, " \t\n");
 502                 char *threads_str       = strsep(&p, " \t\n");
 503                 unsigned threads;
 504                 u64 nr;
 505                 int ret = -EINVAL;
 506
 507                 if (threads_str &&
 508                     !(ret = kstrtouint(threads_str, 10, &threads)) &&
 509                     !(ret = bch2_strtoull_h(nr_str, &nr)))
 510                         bch2_btree_perf_test(c, test, nr, threads);
 511                 else
 512                         size = ret;
 513                 kfree(tmp);
 514         }
 515 #endif
 516         return size;
 517 }
 518
 519 STORE(bch2_fs)
 520 {
 521         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 522
 523         mutex_lock(&c->state_lock);
 524         size = __bch2_fs_store(kobj, attr, buf, size);
 525         mutex_unlock(&c->state_lock);
 526
 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
 537         &sysfs_meta_replicas_have,
 538         &sysfs_data_replicas_have,
 539
 540         &sysfs_journal_write_delay_ms,
 541         &sysfs_journal_reclaim_delay_ms,
 542
 543         &sysfs_promote_whole_extents,
 544
 545         &sysfs_compression_stats,
 546
 547 #ifdef CONFIG_BCACHEFS_TESTS
 548         &sysfs_perf_test,
 549 #endif
 550         NULL
 551 };
 552
 553 /* internal dir - just a wrapper */
 554
 555 SHOW(bch2_fs_internal)
 556 {
 557         struct bch_fs *c = container_of(kobj, struct bch_fs, internal);
 558         return bch2_fs_show(&c->kobj, attr, buf);
 559 }
 560
 561 STORE(bch2_fs_internal)
 562 {
 563         struct bch_fs *c = container_of(kobj, struct bch_fs, internal);
 564         return bch2_fs_store(&c->kobj, attr, buf, size);
 565 }
 566 SYSFS_OPS(bch2_fs_internal);
 567
 568 struct attribute *bch2_fs_internal_files[] = {
 569         &sysfs_alloc_debug,
 570         &sysfs_journal_debug,
 571         &sysfs_journal_pins,
 572         &sysfs_btree_updates,
 573         &sysfs_dirty_btree_nodes,
 574
 575         &sysfs_read_realloc_races,
 576         &sysfs_extent_migrate_done,
 577         &sysfs_extent_migrate_raced,
 578
 579         &sysfs_trigger_journal_flush,
 580         &sysfs_trigger_btree_coalesce,
 581         &sysfs_trigger_gc,
 582         &sysfs_trigger_alloc_write,
 583         &sysfs_prune_cache,
 584
 585         &sysfs_copy_gc_enabled,
 586
 587         &sysfs_rebalance_enabled,
 588         &sysfs_rebalance_work,
 589         sysfs_pd_controller_files(rebalance),
 590
 591         &sysfs_new_stripes,
 592
 593         &sysfs_io_timers_read,
 594         &sysfs_io_timers_write,
 595
 596         &sysfs_internal_uuid,
 597
 598 #define BCH_DEBUG_PARAM(name, description) &sysfs_##name,
 599         BCH_DEBUG_PARAMS()
 600 #undef BCH_DEBUG_PARAM
 601
 602         NULL
 603 };
 604
 605 /* options */
 606
 607 SHOW(bch2_fs_opts_dir)
 608 {
 609         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 610         struct bch_fs *c = container_of(kobj, struct bch_fs, opts_dir);
 611         const struct bch_option *opt = container_of(attr, struct bch_option, attr);
 612         int id = opt - bch2_opt_table;
 613         u64 v = bch2_opt_get_by_id(&c->opts, id);
 614
 615         bch2_opt_to_text(&out, c, opt, v, OPT_SHOW_FULL_LIST);
 616         pr_buf(&out, "\n");
 617
 618         return out.pos - buf;
 619 }
 620
 621 STORE(bch2_fs_opts_dir)
 622 {
 623         struct bch_fs *c = container_of(kobj, struct bch_fs, opts_dir);
 624         const struct bch_option *opt = container_of(attr, struct bch_option, attr);
 625         int ret, id = opt - bch2_opt_table;
 626         char *tmp;
 627         u64 v;
 628
 629         tmp = kstrdup(buf, GFP_KERNEL);
 630         if (!tmp)
 631                 return -ENOMEM;
 632
 633         ret = bch2_opt_parse(c, opt, strim(tmp), &v);
 634         kfree(tmp);
 635
 636         if (ret < 0)
 637                 return ret;
 638
 639         ret = bch2_opt_check_may_set(c, id, v);
 640         if (ret < 0)
 641                 return ret;
 642
 643         if (opt->set_sb != SET_NO_SB_OPT) {
 644                 mutex_lock(&c->sb_lock);
 645                 opt->set_sb(c->disk_sb.sb, v);
 646                 bch2_write_super(c);
 647                 mutex_unlock(&c->sb_lock);
 648         }
 649
 650         bch2_opt_set_by_id(&c->opts, id, v);
 651
 652         if ((id == Opt_background_target ||
 653              id == Opt_background_compression) && v) {
 654                 bch2_rebalance_add_work(c, S64_MAX);
 655                 rebalance_wakeup(c);
 656         }
 657
 658         return size;
 659 }
 660 SYSFS_OPS(bch2_fs_opts_dir);
 661
 662 struct attribute *bch2_fs_opts_dir_files[] = { NULL };
 663
 664 int bch2_opts_create_sysfs_files(struct kobject *kobj)
 665 {
 666         const struct bch_option *i;
 667         int ret;
 668
 669         for (i = bch2_opt_table;
 670              i < bch2_opt_table + bch2_opts_nr;
 671              i++) {
 672                 if (!(i->mode & (OPT_FORMAT|OPT_MOUNT|OPT_RUNTIME)))
 673                         continue;
 674
 675                 ret = sysfs_create_file(kobj, &i->attr);
 676                 if (ret)
 677                         return ret;
 678         }
 679
 680         return 0;
 681 }
 682
 683 /* time stats */
 684
 685 SHOW(bch2_fs_time_stats)
 686 {
 687         struct bch_fs *c = container_of(kobj, struct bch_fs, time_stats);
 688
 689 #define x(name)                                         \
 690         if (attr == &sysfs_time_stat_##name)                            \
 691                 return bch2_time_stats_print(&c->times[BCH_TIME_##name],\
 692                                              buf, PAGE_SIZE);
 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 bucket_last_io(c, bucket(ca, b), 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(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 ssize_t show_quantiles(struct bch_fs *c, struct bch_dev *ca,
 746                               char *buf, bucket_map_fn *fn, void *private)
 747 {
 748         size_t i, n;
 749         /* Compute 31 quantiles */
 750         unsigned q[31], *p;
 751         ssize_t ret = 0;
 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                 ret += scnprintf(buf + ret, PAGE_SIZE - ret,
 779                                  "%u ", q[i]);
 780         buf[ret - 1] = '\n';
 781
 782         return ret;
 783 }
 784
 785 static ssize_t show_reserve_stats(struct bch_dev *ca, char *buf)
 786 {
 787         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 788         enum alloc_reserve i;
 789
 790         spin_lock(&ca->fs->freelist_lock);
 791
 792         pr_buf(&out, "free_inc:\t%zu\t%zu\n",
 793                fifo_used(&ca->free_inc),
 794                ca->free_inc.size);
 795
 796         for (i = 0; i < RESERVE_NR; i++)
 797                 pr_buf(&out, "free[%u]:\t%zu\t%zu\n", i,
 798                        fifo_used(&ca->free[i]),
 799                        ca->free[i].size);
 800
 801         spin_unlock(&ca->fs->freelist_lock);
 802
 803         return out.pos - buf;
 804 }
 805
 806 static ssize_t show_dev_alloc_debug(struct bch_dev *ca, char *buf)
 807 {
 808         struct bch_fs *c = ca->fs;
 809         struct bch_dev_usage stats = bch2_dev_usage_read(c, ca);
 810         unsigned i, nr[BCH_DATA_NR];
 811
 812         memset(nr, 0, sizeof(nr));
 813
 814         for (i = 0; i < ARRAY_SIZE(c->open_buckets); i++)
 815                 nr[c->open_buckets[i].type]++;
 816
 817         return scnprintf(buf, PAGE_SIZE,
 818                 "free_inc:               %zu/%zu\n"
 819                 "free[RESERVE_BTREE]:    %zu/%zu\n"
 820                 "free[RESERVE_MOVINGGC]: %zu/%zu\n"
 821                 "free[RESERVE_NONE]:     %zu/%zu\n"
 822                 "buckets:\n"
 823                 "    capacity:           %llu\n"
 824                 "    alloc:              %llu\n"
 825                 "    sb:                 %llu\n"
 826                 "    journal:            %llu\n"
 827                 "    meta:               %llu\n"
 828                 "    user:               %llu\n"
 829                 "    cached:             %llu\n"
 830                 "    erasure coded:      %llu\n"
 831                 "    available:          %lli\n"
 832                 "sectors:\n"
 833                 "    sb:                 %llu\n"
 834                 "    journal:            %llu\n"
 835                 "    meta:               %llu\n"
 836                 "    user:               %llu\n"
 837                 "    cached:             %llu\n"
 838                 "    fragmented:         %llu\n"
 839                 "    copygc threshold:   %llu\n"
 840                 "freelist_wait:          %s\n"
 841                 "open buckets:           %u/%u (reserved %u)\n"
 842                 "open_buckets_wait:      %s\n"
 843                 "open_buckets_btree:     %u\n"
 844                 "open_buckets_user:      %u\n"
 845                 "btree reserve cache:    %u\n",
 846                 fifo_used(&ca->free_inc),               ca->free_inc.size,
 847                 fifo_used(&ca->free[RESERVE_BTREE]),    ca->free[RESERVE_BTREE].size,
 848                 fifo_used(&ca->free[RESERVE_MOVINGGC]), ca->free[RESERVE_MOVINGGC].size,
 849                 fifo_used(&ca->free[RESERVE_NONE]),     ca->free[RESERVE_NONE].size,
 850                 ca->mi.nbuckets - ca->mi.first_bucket,
 851                 stats.buckets_alloc,
 852                 stats.buckets[BCH_DATA_SB],
 853                 stats.buckets[BCH_DATA_JOURNAL],
 854                 stats.buckets[BCH_DATA_BTREE],
 855                 stats.buckets[BCH_DATA_USER],
 856                 stats.buckets[BCH_DATA_CACHED],
 857                 stats.buckets_ec,
 858                 ca->mi.nbuckets - ca->mi.first_bucket - stats.buckets_unavailable,
 859                 stats.sectors[BCH_DATA_SB],
 860                 stats.sectors[BCH_DATA_JOURNAL],
 861                 stats.sectors[BCH_DATA_BTREE],
 862                 stats.sectors[BCH_DATA_USER],
 863                 stats.sectors[BCH_DATA_CACHED],
 864                 stats.sectors_fragmented,
 865                 ca->copygc_threshold,
 866                 c->freelist_wait.list.first             ? "waiting" : "empty",
 867                 c->open_buckets_nr_free, OPEN_BUCKETS_COUNT,
 868                 BTREE_NODE_OPEN_BUCKET_RESERVE,
 869                 c->open_buckets_wait.list.first         ? "waiting" : "empty",
 870                 nr[BCH_DATA_BTREE],
 871                 nr[BCH_DATA_USER],
 872                 c->btree_reserve_cache_nr);
 873 }
 874
 875 static const char * const bch2_rw[] = {
 876         "read",
 877         "write",
 878         NULL
 879 };
 880
 881 static ssize_t show_dev_iodone(struct bch_dev *ca, char *buf)
 882 {
 883         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 884         int rw, i;
 885
 886         for (rw = 0; rw < 2; rw++) {
 887                 pr_buf(&out, "%s:\n", bch2_rw[rw]);
 888
 889                 for (i = 1; i < BCH_DATA_NR; i++)
 890                         pr_buf(&out, "%-12s:%12llu\n",
 891                                bch2_data_types[i],
 892                                percpu_u64_get(&ca->io_done->sectors[rw][i]) << 9);
 893         }
 894
 895         return out.pos - buf;
 896 }
 897
 898 SHOW(bch2_dev)
 899 {
 900         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
 901         struct bch_fs *c = ca->fs;
 902         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 903
 904         sysfs_printf(uuid,              "%pU\n", ca->uuid.b);
 905
 906         sysfs_print(bucket_size,        bucket_bytes(ca));
 907         sysfs_print(block_size,         block_bytes(c));
 908         sysfs_print(first_bucket,       ca->mi.first_bucket);
 909         sysfs_print(nbuckets,           ca->mi.nbuckets);
 910         sysfs_print(durability,         ca->mi.durability);
 911         sysfs_print(discard,            ca->mi.discard);
 912
 913         if (attr == &sysfs_label) {
 914                 if (ca->mi.group) {
 915                         mutex_lock(&c->sb_lock);
 916                         bch2_disk_path_to_text(&out, &c->disk_sb,
 917                                                ca->mi.group - 1);
 918                         mutex_unlock(&c->sb_lock);
 919                 }
 920
 921                 pr_buf(&out, "\n");
 922                 return out.pos - buf;
 923         }
 924
 925         if (attr == &sysfs_has_data) {
 926                 bch2_flags_to_text(&out, bch2_data_types,
 927                                    bch2_dev_has_data(c, ca));
 928                 pr_buf(&out, "\n");
 929                 return out.pos - buf;
 930         }
 931
 932         sysfs_pd_controller_show(copy_gc, &ca->copygc_pd);
 933
 934         if (attr == &sysfs_cache_replacement_policy) {
 935                 bch2_string_opt_to_text(&out,
 936                                         bch2_cache_replacement_policies,
 937                                         ca->mi.replacement);
 938                 pr_buf(&out, "\n");
 939                 return out.pos - buf;
 940         }
 941
 942         if (attr == &sysfs_state_rw) {
 943                 bch2_string_opt_to_text(&out, bch2_dev_state,
 944                                         ca->mi.state);
 945                 pr_buf(&out, "\n");
 946                 return out.pos - buf;
 947         }
 948
 949         if (attr == &sysfs_iodone)
 950                 return show_dev_iodone(ca, buf);
 951
 952         sysfs_print(io_latency_read,            atomic64_read(&ca->cur_latency[READ]));
 953         sysfs_print(io_latency_write,           atomic64_read(&ca->cur_latency[WRITE]));
 954
 955         if (attr == &sysfs_io_latency_stats_read)
 956                 return bch2_time_stats_print(&ca->io_latency[READ], buf, PAGE_SIZE);
 957         if (attr == &sysfs_io_latency_stats_write)
 958                 return bch2_time_stats_print(&ca->io_latency[WRITE], buf, PAGE_SIZE);
 959
 960         sysfs_printf(congested,                 "%u%%",
 961                      clamp(atomic_read(&ca->congested), 0, CONGESTED_MAX)
 962                      * 100 / CONGESTED_MAX);
 963
 964         if (attr == &sysfs_bucket_quantiles_last_read)
 965                 return show_quantiles(c, ca, buf, bucket_last_io_fn, (void *) 0);
 966         if (attr == &sysfs_bucket_quantiles_last_write)
 967                 return show_quantiles(c, ca, buf, bucket_last_io_fn, (void *) 1);
 968         if (attr == &sysfs_bucket_quantiles_fragmentation)
 969                 return show_quantiles(c, ca, buf, bucket_sectors_used_fn, NULL);
 970         if (attr == &sysfs_bucket_quantiles_oldest_gen)
 971                 return show_quantiles(c, ca, buf, bucket_oldest_gen_fn, NULL);
 972
 973         if (attr == &sysfs_reserve_stats)
 974                 return show_reserve_stats(ca, buf);
 975         if (attr == &sysfs_alloc_debug)
 976                 return show_dev_alloc_debug(ca, buf);
 977
 978         return 0;
 979 }
 980
 981 STORE(bch2_dev)
 982 {
 983         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
 984         struct bch_fs *c = ca->fs;
 985         struct bch_member *mi;
 986
 987         sysfs_pd_controller_store(copy_gc, &ca->copygc_pd);
 988
 989         if (attr == &sysfs_discard) {
 990                 bool v = strtoul_or_return(buf);
 991
 992                 mutex_lock(&c->sb_lock);
 993                 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
 994
 995                 if (v != BCH_MEMBER_DISCARD(mi)) {
 996                         SET_BCH_MEMBER_DISCARD(mi, v);
 997                         bch2_write_super(c);
 998                 }
 999                 mutex_unlock(&c->sb_lock);
1000         }
1001
1002         if (attr == &sysfs_cache_replacement_policy) {
1003                 ssize_t v = __sysfs_match_string(bch2_cache_replacement_policies, -1, buf);
1004
1005                 if (v < 0)
1006                         return v;
1007
1008                 mutex_lock(&c->sb_lock);
1009                 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1010
1011                 if ((unsigned) v != BCH_MEMBER_REPLACEMENT(mi)) {
1012                         SET_BCH_MEMBER_REPLACEMENT(mi, v);
1013                         bch2_write_super(c);
1014                 }
1015                 mutex_unlock(&c->sb_lock);
1016         }
1017
1018         if (attr == &sysfs_label) {
1019                 char *tmp;
1020                 int ret;
1021
1022                 tmp = kstrdup(buf, GFP_KERNEL);
1023                 if (!tmp)
1024                         return -ENOMEM;
1025
1026                 ret = bch2_dev_group_set(c, ca, strim(tmp));
1027                 kfree(tmp);
1028                 if (ret)
1029                         return ret;
1030         }
1031
1032         if (attr == &sysfs_wake_allocator)
1033                 bch2_wake_allocator(ca);
1034
1035         return size;
1036 }
1037 SYSFS_OPS(bch2_dev);
1038
1039 struct attribute *bch2_dev_files[] = {
1040         &sysfs_uuid,
1041         &sysfs_bucket_size,
1042         &sysfs_block_size,
1043         &sysfs_first_bucket,
1044         &sysfs_nbuckets,
1045         &sysfs_durability,
1046
1047         /* settings: */
1048         &sysfs_discard,
1049         &sysfs_cache_replacement_policy,
1050         &sysfs_state_rw,
1051         &sysfs_label,
1052
1053         &sysfs_has_data,
1054         &sysfs_iodone,
1055
1056         &sysfs_io_latency_read,
1057         &sysfs_io_latency_write,
1058         &sysfs_io_latency_stats_read,
1059         &sysfs_io_latency_stats_write,
1060         &sysfs_congested,
1061
1062         /* alloc info - other stats: */
1063         &sysfs_bucket_quantiles_last_read,
1064         &sysfs_bucket_quantiles_last_write,
1065         &sysfs_bucket_quantiles_fragmentation,
1066         &sysfs_bucket_quantiles_oldest_gen,
1067
1068         &sysfs_reserve_stats,
1069
1070         /* debug: */
1071         &sysfs_alloc_debug,
1072         &sysfs_wake_allocator,
1073
1074         sysfs_pd_controller_files(copy_gc),
1075         NULL
1076 };
1077
1078 #endif  /* _BCACHEFS_SYSFS_H_ */