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 read_attribute(btree_transactions);
 170
 171 read_attribute(internal_uuid);
 172
 173 read_attribute(has_data);
 174 read_attribute(alloc_debug);
 175 write_attribute(wake_allocator);
 176
 177 read_attribute(read_realloc_races);
 178 read_attribute(extent_migrate_done);
 179 read_attribute(extent_migrate_raced);
 180
 181 rw_attribute(journal_write_delay_ms);
 182 rw_attribute(journal_reclaim_delay_ms);
 183
 184 rw_attribute(discard);
 185 rw_attribute(cache_replacement_policy);
 186 rw_attribute(label);
 187
 188 rw_attribute(copy_gc_enabled);
 189 sysfs_pd_controller_attribute(copy_gc);
 190
 191 rw_attribute(rebalance_enabled);
 192 sysfs_pd_controller_attribute(rebalance);
 193 read_attribute(rebalance_work);
 194 rw_attribute(promote_whole_extents);
 195
 196 read_attribute(new_stripes);
 197
 198 rw_attribute(pd_controllers_update_seconds);
 199
 200 read_attribute(meta_replicas_have);
 201 read_attribute(data_replicas_have);
 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 BCH_DEBUG_PARAM(name, description)                              \
 211         rw_attribute(name);
 212
 213         BCH_DEBUG_PARAMS()
 214 #undef BCH_DEBUG_PARAM
 215
 216 #define x(_name)                                                \
 217         static struct attribute sysfs_time_stat_##_name =               \
 218                 { .name = #_name, .mode = S_IRUGO };
 219         BCH_TIME_STATS()
 220 #undef x
 221
 222 static struct attribute sysfs_state_rw = {
 223         .name = "state",
 224         .mode = S_IRUGO
 225 };
 226
 227 static size_t bch2_btree_cache_size(struct bch_fs *c)
 228 {
 229         size_t ret = 0;
 230         struct btree *b;
 231
 232         mutex_lock(&c->btree_cache.lock);
 233         list_for_each_entry(b, &c->btree_cache.live, list)
 234                 ret += btree_bytes(c);
 235
 236         mutex_unlock(&c->btree_cache.lock);
 237         return ret;
 238 }
 239
 240 static ssize_t show_fs_alloc_debug(struct bch_fs *c, char *buf)
 241 {
 242         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 243         struct bch_fs_usage *fs_usage = bch2_fs_usage_read(c);
 244
 245         if (!fs_usage)
 246                 return -ENOMEM;
 247
 248         bch2_fs_usage_to_text(&out, c, fs_usage);
 249
 250         percpu_up_read(&c->mark_lock);
 251
 252         kfree(fs_usage);
 253
 254         return out.pos - buf;
 255 }
 256
 257 static ssize_t bch2_compression_stats(struct bch_fs *c, char *buf)
 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         return scnprintf(buf, PAGE_SIZE,
 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 }
 314
 315 static ssize_t bch2_new_stripes(struct bch_fs *c, char *buf)
 316 {
 317         char *out = buf, *end = buf + PAGE_SIZE;
 318         struct ec_stripe_head *h;
 319         struct ec_stripe_new *s;
 320
 321         mutex_lock(&c->ec_new_stripe_lock);
 322         list_for_each_entry(h, &c->ec_new_stripe_list, list) {
 323                 out += scnprintf(out, end - out,
 324                                  "target %u algo %u redundancy %u:\n",
 325                                  h->target, h->algo, h->redundancy);
 326
 327                 if (h->s)
 328                         out += scnprintf(out, end - out,
 329                                          "\tpending: blocks %u allocated %u\n",
 330                                          h->s->blocks.nr,
 331                                          bitmap_weight(h->s->blocks_allocated,
 332                                                        h->s->blocks.nr));
 333
 334                 mutex_lock(&h->lock);
 335                 list_for_each_entry(s, &h->stripes, list)
 336                         out += scnprintf(out, end - out,
 337                                          "\tin flight: blocks %u allocated %u pin %u\n",
 338                                          s->blocks.nr,
 339                                          bitmap_weight(s->blocks_allocated,
 340                                                        s->blocks.nr),
 341                                          atomic_read(&s->pin));
 342                 mutex_unlock(&h->lock);
 343
 344         }
 345         mutex_unlock(&c->ec_new_stripe_lock);
 346
 347         return out - buf;
 348 }
 349
 350 SHOW(bch2_fs)
 351 {
 352         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 353
 354         sysfs_print(minor,                      c->minor);
 355         sysfs_printf(internal_uuid, "%pU",      c->sb.uuid.b);
 356
 357         sysfs_print(journal_write_delay_ms,     c->journal.write_delay_ms);
 358         sysfs_print(journal_reclaim_delay_ms,   c->journal.reclaim_delay_ms);
 359
 360         sysfs_print(block_size,                 block_bytes(c));
 361         sysfs_print(btree_node_size,            btree_bytes(c));
 362         sysfs_hprint(btree_cache_size,          bch2_btree_cache_size(c));
 363
 364         sysfs_print(read_realloc_races,
 365                     atomic_long_read(&c->read_realloc_races));
 366         sysfs_print(extent_migrate_done,
 367                     atomic_long_read(&c->extent_migrate_done));
 368         sysfs_print(extent_migrate_raced,
 369                     atomic_long_read(&c->extent_migrate_raced));
 370
 371         sysfs_printf(btree_gc_periodic, "%u",   (int) c->btree_gc_periodic);
 372
 373         sysfs_printf(copy_gc_enabled, "%i", c->copy_gc_enabled);
 374
 375         sysfs_print(pd_controllers_update_seconds,
 376                     c->pd_controllers_update_seconds);
 377
 378         sysfs_printf(rebalance_enabled,         "%i", c->rebalance.enabled);
 379         sysfs_pd_controller_show(rebalance,     &c->rebalance.pd); /* XXX */
 380
 381         if (attr == &sysfs_rebalance_work)
 382                 return bch2_rebalance_work_show(c, buf);
 383
 384         sysfs_print(promote_whole_extents,      c->promote_whole_extents);
 385
 386         sysfs_printf(meta_replicas_have, "%i",  bch2_replicas_online(c, true));
 387         sysfs_printf(data_replicas_have, "%i",  bch2_replicas_online(c, false));
 388
 389         /* Debugging: */
 390
 391         if (attr == &sysfs_alloc_debug)
 392                 return show_fs_alloc_debug(c, buf);
 393
 394         if (attr == &sysfs_journal_debug)
 395                 return bch2_journal_print_debug(&c->journal, buf);
 396
 397         if (attr == &sysfs_journal_pins)
 398                 return bch2_journal_print_pins(&c->journal, buf);
 399
 400         if (attr == &sysfs_btree_updates)
 401                 return bch2_btree_updates_print(c, buf);
 402
 403         if (attr == &sysfs_dirty_btree_nodes)
 404                 return bch2_dirty_btree_nodes_print(c, buf);
 405         if (attr == &sysfs_btree_transactions) {
 406                 struct printbuf out = _PBUF(buf, PAGE_SIZE);
 407
 408                 bch2_btree_trans_to_text(&out, c);
 409                 return out.pos - buf;
 410         }
 411
 412         if (attr == &sysfs_compression_stats)
 413                 return bch2_compression_stats(c, buf);
 414
 415         if (attr == &sysfs_new_stripes)
 416                 return bch2_new_stripes(c, buf);
 417
 418         if (attr == &sysfs_io_timers_read)
 419                 return bch2_io_timers_show(&c->io_clock[READ], buf);
 420         if (attr == &sysfs_io_timers_write)
 421                 return bch2_io_timers_show(&c->io_clock[WRITE], buf);
 422
 423 #define BCH_DEBUG_PARAM(name, description) sysfs_print(name, c->name);
 424         BCH_DEBUG_PARAMS()
 425 #undef BCH_DEBUG_PARAM
 426
 427         return 0;
 428 }
 429
 430 STORE(__bch2_fs)
 431 {
 432         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 433
 434         sysfs_strtoul(journal_write_delay_ms, c->journal.write_delay_ms);
 435         sysfs_strtoul(journal_reclaim_delay_ms, c->journal.reclaim_delay_ms);
 436
 437         if (attr == &sysfs_btree_gc_periodic) {
 438                 ssize_t ret = strtoul_safe(buf, c->btree_gc_periodic)
 439                         ?: (ssize_t) size;
 440
 441                 wake_up_process(c->gc_thread);
 442                 return ret;
 443         }
 444
 445         if (attr == &sysfs_copy_gc_enabled) {
 446                 struct bch_dev *ca;
 447                 unsigned i;
 448                 ssize_t ret = strtoul_safe(buf, c->copy_gc_enabled)
 449                         ?: (ssize_t) size;
 450
 451                 for_each_member_device(ca, c, i)
 452                         if (ca->copygc_thread)
 453                                 wake_up_process(ca->copygc_thread);
 454                 return ret;
 455         }
 456
 457         if (attr == &sysfs_rebalance_enabled) {
 458                 ssize_t ret = strtoul_safe(buf, c->rebalance.enabled)
 459                         ?: (ssize_t) size;
 460
 461                 rebalance_wakeup(c);
 462                 return ret;
 463         }
 464
 465         sysfs_strtoul(pd_controllers_update_seconds,
 466                       c->pd_controllers_update_seconds);
 467         sysfs_pd_controller_store(rebalance,    &c->rebalance.pd);
 468
 469         sysfs_strtoul(promote_whole_extents,    c->promote_whole_extents);
 470
 471         /* Debugging: */
 472
 473 #define BCH_DEBUG_PARAM(name, description) sysfs_strtoul(name, c->name);
 474         BCH_DEBUG_PARAMS()
 475 #undef BCH_DEBUG_PARAM
 476
 477         if (!test_bit(BCH_FS_STARTED, &c->flags))
 478                 return -EPERM;
 479
 480         /* Debugging: */
 481
 482         if (attr == &sysfs_trigger_journal_flush)
 483                 bch2_journal_meta_async(&c->journal, NULL);
 484
 485         if (attr == &sysfs_trigger_btree_coalesce)
 486                 bch2_coalesce(c);
 487
 488         if (attr == &sysfs_trigger_gc)
 489                 bch2_gc(c, NULL, false, false);
 490
 491         if (attr == &sysfs_trigger_alloc_write) {
 492                 bool wrote;
 493
 494                 bch2_alloc_write(c, 0, &wrote);
 495         }
 496
 497         if (attr == &sysfs_prune_cache) {
 498                 struct shrink_control sc;
 499
 500                 sc.gfp_mask = GFP_KERNEL;
 501                 sc.nr_to_scan = strtoul_or_return(buf);
 502                 c->btree_cache.shrink.scan_objects(&c->btree_cache.shrink, &sc);
 503         }
 504 #ifdef CONFIG_BCACHEFS_TESTS
 505         if (attr == &sysfs_perf_test) {
 506                 char *tmp = kstrdup(buf, GFP_KERNEL), *p = tmp;
 507                 char *test              = strsep(&p, " \t\n");
 508                 char *nr_str            = strsep(&p, " \t\n");
 509                 char *threads_str       = strsep(&p, " \t\n");
 510                 unsigned threads;
 511                 u64 nr;
 512                 int ret = -EINVAL;
 513
 514                 if (threads_str &&
 515                     !(ret = kstrtouint(threads_str, 10, &threads)) &&
 516                     !(ret = bch2_strtoull_h(nr_str, &nr)))
 517                         bch2_btree_perf_test(c, test, nr, threads);
 518                 else
 519                         size = ret;
 520                 kfree(tmp);
 521         }
 522 #endif
 523         return size;
 524 }
 525
 526 STORE(bch2_fs)
 527 {
 528         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 529
 530         mutex_lock(&c->state_lock);
 531         size = __bch2_fs_store(kobj, attr, buf, size);
 532         mutex_unlock(&c->state_lock);
 533
 534         return size;
 535 }
 536 SYSFS_OPS(bch2_fs);
 537
 538 struct attribute *bch2_fs_files[] = {
 539         &sysfs_minor,
 540         &sysfs_block_size,
 541         &sysfs_btree_node_size,
 542         &sysfs_btree_cache_size,
 543
 544         &sysfs_meta_replicas_have,
 545         &sysfs_data_replicas_have,
 546
 547         &sysfs_journal_write_delay_ms,
 548         &sysfs_journal_reclaim_delay_ms,
 549
 550         &sysfs_promote_whole_extents,
 551
 552         &sysfs_compression_stats,
 553
 554 #ifdef CONFIG_BCACHEFS_TESTS
 555         &sysfs_perf_test,
 556 #endif
 557         NULL
 558 };
 559
 560 /* internal dir - just a wrapper */
 561
 562 SHOW(bch2_fs_internal)
 563 {
 564         struct bch_fs *c = container_of(kobj, struct bch_fs, internal);
 565         return bch2_fs_show(&c->kobj, attr, buf);
 566 }
 567
 568 STORE(bch2_fs_internal)
 569 {
 570         struct bch_fs *c = container_of(kobj, struct bch_fs, internal);
 571         return bch2_fs_store(&c->kobj, attr, buf, size);
 572 }
 573 SYSFS_OPS(bch2_fs_internal);
 574
 575 struct attribute *bch2_fs_internal_files[] = {
 576         &sysfs_alloc_debug,
 577         &sysfs_journal_debug,
 578         &sysfs_journal_pins,
 579         &sysfs_btree_updates,
 580         &sysfs_dirty_btree_nodes,
 581         &sysfs_btree_transactions,
 582
 583         &sysfs_read_realloc_races,
 584         &sysfs_extent_migrate_done,
 585         &sysfs_extent_migrate_raced,
 586
 587         &sysfs_trigger_journal_flush,
 588         &sysfs_trigger_btree_coalesce,
 589         &sysfs_trigger_gc,
 590         &sysfs_trigger_alloc_write,
 591         &sysfs_prune_cache,
 592
 593         &sysfs_copy_gc_enabled,
 594
 595         &sysfs_rebalance_enabled,
 596         &sysfs_rebalance_work,
 597         sysfs_pd_controller_files(rebalance),
 598
 599         &sysfs_new_stripes,
 600
 601         &sysfs_io_timers_read,
 602         &sysfs_io_timers_write,
 603
 604         &sysfs_internal_uuid,
 605
 606 #define BCH_DEBUG_PARAM(name, description) &sysfs_##name,
 607         BCH_DEBUG_PARAMS()
 608 #undef BCH_DEBUG_PARAM
 609
 610         NULL
 611 };
 612
 613 /* options */
 614
 615 SHOW(bch2_fs_opts_dir)
 616 {
 617         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 618         struct bch_fs *c = container_of(kobj, struct bch_fs, opts_dir);
 619         const struct bch_option *opt = container_of(attr, struct bch_option, attr);
 620         int id = opt - bch2_opt_table;
 621         u64 v = bch2_opt_get_by_id(&c->opts, id);
 622
 623         bch2_opt_to_text(&out, c, opt, v, OPT_SHOW_FULL_LIST);
 624         pr_buf(&out, "\n");
 625
 626         return out.pos - buf;
 627 }
 628
 629 STORE(bch2_fs_opts_dir)
 630 {
 631         struct bch_fs *c = container_of(kobj, struct bch_fs, opts_dir);
 632         const struct bch_option *opt = container_of(attr, struct bch_option, attr);
 633         int ret, id = opt - bch2_opt_table;
 634         char *tmp;
 635         u64 v;
 636
 637         tmp = kstrdup(buf, GFP_KERNEL);
 638         if (!tmp)
 639                 return -ENOMEM;
 640
 641         ret = bch2_opt_parse(c, opt, strim(tmp), &v);
 642         kfree(tmp);
 643
 644         if (ret < 0)
 645                 return ret;
 646
 647         ret = bch2_opt_check_may_set(c, id, v);
 648         if (ret < 0)
 649                 return ret;
 650
 651         if (opt->set_sb != SET_NO_SB_OPT) {
 652                 mutex_lock(&c->sb_lock);
 653                 opt->set_sb(c->disk_sb.sb, v);
 654                 bch2_write_super(c);
 655                 mutex_unlock(&c->sb_lock);
 656         }
 657
 658         bch2_opt_set_by_id(&c->opts, id, v);
 659
 660         if ((id == Opt_background_target ||
 661              id == Opt_background_compression) && v) {
 662                 bch2_rebalance_add_work(c, S64_MAX);
 663                 rebalance_wakeup(c);
 664         }
 665
 666         return size;
 667 }
 668 SYSFS_OPS(bch2_fs_opts_dir);
 669
 670 struct attribute *bch2_fs_opts_dir_files[] = { NULL };
 671
 672 int bch2_opts_create_sysfs_files(struct kobject *kobj)
 673 {
 674         const struct bch_option *i;
 675         int ret;
 676
 677         for (i = bch2_opt_table;
 678              i < bch2_opt_table + bch2_opts_nr;
 679              i++) {
 680                 if (!(i->mode & (OPT_FORMAT|OPT_MOUNT|OPT_RUNTIME)))
 681                         continue;
 682
 683                 ret = sysfs_create_file(kobj, &i->attr);
 684                 if (ret)
 685                         return ret;
 686         }
 687
 688         return 0;
 689 }
 690
 691 /* time stats */
 692
 693 SHOW(bch2_fs_time_stats)
 694 {
 695         struct bch_fs *c = container_of(kobj, struct bch_fs, time_stats);
 696
 697 #define x(name)                                         \
 698         if (attr == &sysfs_time_stat_##name)                            \
 699                 return bch2_time_stats_print(&c->times[BCH_TIME_##name],\
 700                                              buf, PAGE_SIZE);
 701         BCH_TIME_STATS()
 702 #undef x
 703
 704         return 0;
 705 }
 706
 707 STORE(bch2_fs_time_stats)
 708 {
 709         return size;
 710 }
 711 SYSFS_OPS(bch2_fs_time_stats);
 712
 713 struct attribute *bch2_fs_time_stats_files[] = {
 714 #define x(name)                                         \
 715         &sysfs_time_stat_##name,
 716         BCH_TIME_STATS()
 717 #undef x
 718         NULL
 719 };
 720
 721 typedef unsigned (bucket_map_fn)(struct bch_fs *, struct bch_dev *,
 722                                  size_t, void *);
 723
 724 static unsigned bucket_last_io_fn(struct bch_fs *c, struct bch_dev *ca,
 725                                   size_t b, void *private)
 726 {
 727         int rw = (private ? 1 : 0);
 728
 729         return bucket_last_io(c, bucket(ca, b), rw);
 730 }
 731
 732 static unsigned bucket_sectors_used_fn(struct bch_fs *c, struct bch_dev *ca,
 733                                        size_t b, void *private)
 734 {
 735         struct bucket *g = bucket(ca, b);
 736         return bucket_sectors_used(g->mark);
 737 }
 738
 739 static unsigned bucket_oldest_gen_fn(struct bch_fs *c, struct bch_dev *ca,
 740                                      size_t b, void *private)
 741 {
 742         return bucket_gc_gen(ca, b);
 743 }
 744
 745 static int unsigned_cmp(const void *_l, const void *_r)
 746 {
 747         const unsigned *l = _l;
 748         const unsigned *r = _r;
 749
 750         return cmp_int(*l, *r);
 751 }
 752
 753 static ssize_t show_quantiles(struct bch_fs *c, struct bch_dev *ca,
 754                               char *buf, bucket_map_fn *fn, void *private)
 755 {
 756         size_t i, n;
 757         /* Compute 31 quantiles */
 758         unsigned q[31], *p;
 759         ssize_t ret = 0;
 760
 761         down_read(&ca->bucket_lock);
 762         n = ca->mi.nbuckets;
 763
 764         p = vzalloc(n * sizeof(unsigned));
 765         if (!p) {
 766                 up_read(&ca->bucket_lock);
 767                 return -ENOMEM;
 768         }
 769
 770         for (i = ca->mi.first_bucket; i < n; i++)
 771                 p[i] = fn(c, ca, i, private);
 772
 773         sort(p, n, sizeof(unsigned), unsigned_cmp, NULL);
 774         up_read(&ca->bucket_lock);
 775
 776         while (n &&
 777                !p[n - 1])
 778                 --n;
 779
 780         for (i = 0; i < ARRAY_SIZE(q); i++)
 781                 q[i] = p[n * (i + 1) / (ARRAY_SIZE(q) + 1)];
 782
 783         vfree(p);
 784
 785         for (i = 0; i < ARRAY_SIZE(q); i++)
 786                 ret += scnprintf(buf + ret, PAGE_SIZE - ret,
 787                                  "%u ", q[i]);
 788         buf[ret - 1] = '\n';
 789
 790         return ret;
 791 }
 792
 793 static ssize_t show_reserve_stats(struct bch_dev *ca, char *buf)
 794 {
 795         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 796         enum alloc_reserve i;
 797
 798         spin_lock(&ca->fs->freelist_lock);
 799
 800         pr_buf(&out, "free_inc:\t%zu\t%zu\n",
 801                fifo_used(&ca->free_inc),
 802                ca->free_inc.size);
 803
 804         for (i = 0; i < RESERVE_NR; i++)
 805                 pr_buf(&out, "free[%u]:\t%zu\t%zu\n", i,
 806                        fifo_used(&ca->free[i]),
 807                        ca->free[i].size);
 808
 809         spin_unlock(&ca->fs->freelist_lock);
 810
 811         return out.pos - buf;
 812 }
 813
 814 static ssize_t show_dev_alloc_debug(struct bch_dev *ca, char *buf)
 815 {
 816         struct bch_fs *c = ca->fs;
 817         struct bch_dev_usage stats = bch2_dev_usage_read(c, ca);
 818         unsigned i, nr[BCH_DATA_NR];
 819
 820         memset(nr, 0, sizeof(nr));
 821
 822         for (i = 0; i < ARRAY_SIZE(c->open_buckets); i++)
 823                 nr[c->open_buckets[i].type]++;
 824
 825         return scnprintf(buf, PAGE_SIZE,
 826                 "free_inc:               %zu/%zu\n"
 827                 "free[RESERVE_BTREE]:    %zu/%zu\n"
 828                 "free[RESERVE_MOVINGGC]: %zu/%zu\n"
 829                 "free[RESERVE_NONE]:     %zu/%zu\n"
 830                 "buckets:\n"
 831                 "    capacity:           %llu\n"
 832                 "    alloc:              %llu\n"
 833                 "    sb:                 %llu\n"
 834                 "    journal:            %llu\n"
 835                 "    meta:               %llu\n"
 836                 "    user:               %llu\n"
 837                 "    cached:             %llu\n"
 838                 "    erasure coded:      %llu\n"
 839                 "    available:          %lli\n"
 840                 "sectors:\n"
 841                 "    sb:                 %llu\n"
 842                 "    journal:            %llu\n"
 843                 "    meta:               %llu\n"
 844                 "    user:               %llu\n"
 845                 "    cached:             %llu\n"
 846                 "    fragmented:         %llu\n"
 847                 "    copygc threshold:   %llu\n"
 848                 "freelist_wait:          %s\n"
 849                 "open buckets:           %u/%u (reserved %u)\n"
 850                 "open_buckets_wait:      %s\n"
 851                 "open_buckets_btree:     %u\n"
 852                 "open_buckets_user:      %u\n"
 853                 "btree reserve cache:    %u\n",
 854                 fifo_used(&ca->free_inc),               ca->free_inc.size,
 855                 fifo_used(&ca->free[RESERVE_BTREE]),    ca->free[RESERVE_BTREE].size,
 856                 fifo_used(&ca->free[RESERVE_MOVINGGC]), ca->free[RESERVE_MOVINGGC].size,
 857                 fifo_used(&ca->free[RESERVE_NONE]),     ca->free[RESERVE_NONE].size,
 858                 ca->mi.nbuckets - ca->mi.first_bucket,
 859                 stats.buckets_alloc,
 860                 stats.buckets[BCH_DATA_SB],
 861                 stats.buckets[BCH_DATA_JOURNAL],
 862                 stats.buckets[BCH_DATA_BTREE],
 863                 stats.buckets[BCH_DATA_USER],
 864                 stats.buckets[BCH_DATA_CACHED],
 865                 stats.buckets_ec,
 866                 ca->mi.nbuckets - ca->mi.first_bucket - stats.buckets_unavailable,
 867                 stats.sectors[BCH_DATA_SB],
 868                 stats.sectors[BCH_DATA_JOURNAL],
 869                 stats.sectors[BCH_DATA_BTREE],
 870                 stats.sectors[BCH_DATA_USER],
 871                 stats.sectors[BCH_DATA_CACHED],
 872                 stats.sectors_fragmented,
 873                 ca->copygc_threshold,
 874                 c->freelist_wait.list.first             ? "waiting" : "empty",
 875                 c->open_buckets_nr_free, OPEN_BUCKETS_COUNT,
 876                 BTREE_NODE_OPEN_BUCKET_RESERVE,
 877                 c->open_buckets_wait.list.first         ? "waiting" : "empty",
 878                 nr[BCH_DATA_BTREE],
 879                 nr[BCH_DATA_USER],
 880                 c->btree_reserve_cache_nr);
 881 }
 882
 883 static const char * const bch2_rw[] = {
 884         "read",
 885         "write",
 886         NULL
 887 };
 888
 889 static ssize_t show_dev_iodone(struct bch_dev *ca, char *buf)
 890 {
 891         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 892         int rw, i;
 893
 894         for (rw = 0; rw < 2; rw++) {
 895                 pr_buf(&out, "%s:\n", bch2_rw[rw]);
 896
 897                 for (i = 1; i < BCH_DATA_NR; i++)
 898                         pr_buf(&out, "%-12s:%12llu\n",
 899                                bch2_data_types[i],
 900                                percpu_u64_get(&ca->io_done->sectors[rw][i]) << 9);
 901         }
 902
 903         return out.pos - buf;
 904 }
 905
 906 SHOW(bch2_dev)
 907 {
 908         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
 909         struct bch_fs *c = ca->fs;
 910         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 911
 912         sysfs_printf(uuid,              "%pU\n", ca->uuid.b);
 913
 914         sysfs_print(bucket_size,        bucket_bytes(ca));
 915         sysfs_print(block_size,         block_bytes(c));
 916         sysfs_print(first_bucket,       ca->mi.first_bucket);
 917         sysfs_print(nbuckets,           ca->mi.nbuckets);
 918         sysfs_print(durability,         ca->mi.durability);
 919         sysfs_print(discard,            ca->mi.discard);
 920
 921         if (attr == &sysfs_label) {
 922                 if (ca->mi.group) {
 923                         mutex_lock(&c->sb_lock);
 924                         bch2_disk_path_to_text(&out, &c->disk_sb,
 925                                                ca->mi.group - 1);
 926                         mutex_unlock(&c->sb_lock);
 927                 }
 928
 929                 pr_buf(&out, "\n");
 930                 return out.pos - buf;
 931         }
 932
 933         if (attr == &sysfs_has_data) {
 934                 bch2_flags_to_text(&out, bch2_data_types,
 935                                    bch2_dev_has_data(c, ca));
 936                 pr_buf(&out, "\n");
 937                 return out.pos - buf;
 938         }
 939
 940         sysfs_pd_controller_show(copy_gc, &ca->copygc_pd);
 941
 942         if (attr == &sysfs_cache_replacement_policy) {
 943                 bch2_string_opt_to_text(&out,
 944                                         bch2_cache_replacement_policies,
 945                                         ca->mi.replacement);
 946                 pr_buf(&out, "\n");
 947                 return out.pos - buf;
 948         }
 949
 950         if (attr == &sysfs_state_rw) {
 951                 bch2_string_opt_to_text(&out, bch2_dev_state,
 952                                         ca->mi.state);
 953                 pr_buf(&out, "\n");
 954                 return out.pos - buf;
 955         }
 956
 957         if (attr == &sysfs_iodone)
 958                 return show_dev_iodone(ca, buf);
 959
 960         sysfs_print(io_latency_read,            atomic64_read(&ca->cur_latency[READ]));
 961         sysfs_print(io_latency_write,           atomic64_read(&ca->cur_latency[WRITE]));
 962
 963         if (attr == &sysfs_io_latency_stats_read)
 964                 return bch2_time_stats_print(&ca->io_latency[READ], buf, PAGE_SIZE);
 965         if (attr == &sysfs_io_latency_stats_write)
 966                 return bch2_time_stats_print(&ca->io_latency[WRITE], buf, PAGE_SIZE);
 967
 968         sysfs_printf(congested,                 "%u%%",
 969                      clamp(atomic_read(&ca->congested), 0, CONGESTED_MAX)
 970                      * 100 / CONGESTED_MAX);
 971
 972         if (attr == &sysfs_bucket_quantiles_last_read)
 973                 return show_quantiles(c, ca, buf, bucket_last_io_fn, (void *) 0);
 974         if (attr == &sysfs_bucket_quantiles_last_write)
 975                 return show_quantiles(c, ca, buf, bucket_last_io_fn, (void *) 1);
 976         if (attr == &sysfs_bucket_quantiles_fragmentation)
 977                 return show_quantiles(c, ca, buf, bucket_sectors_used_fn, NULL);
 978         if (attr == &sysfs_bucket_quantiles_oldest_gen)
 979                 return show_quantiles(c, ca, buf, bucket_oldest_gen_fn, NULL);
 980
 981         if (attr == &sysfs_reserve_stats)
 982                 return show_reserve_stats(ca, buf);
 983         if (attr == &sysfs_alloc_debug)
 984                 return show_dev_alloc_debug(ca, buf);
 985
 986         return 0;
 987 }
 988
 989 STORE(bch2_dev)
 990 {
 991         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
 992         struct bch_fs *c = ca->fs;
 993         struct bch_member *mi;
 994
 995         sysfs_pd_controller_store(copy_gc, &ca->copygc_pd);
 996
 997         if (attr == &sysfs_discard) {
 998                 bool v = strtoul_or_return(buf);
 999
1000                 mutex_lock(&c->sb_lock);
1001                 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1002
1003                 if (v != BCH_MEMBER_DISCARD(mi)) {
1004                         SET_BCH_MEMBER_DISCARD(mi, v);
1005                         bch2_write_super(c);
1006                 }
1007                 mutex_unlock(&c->sb_lock);
1008         }
1009
1010         if (attr == &sysfs_cache_replacement_policy) {
1011                 ssize_t v = __sysfs_match_string(bch2_cache_replacement_policies, -1, buf);
1012
1013                 if (v < 0)
1014                         return v;
1015
1016                 mutex_lock(&c->sb_lock);
1017                 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1018
1019                 if ((unsigned) v != BCH_MEMBER_REPLACEMENT(mi)) {
1020                         SET_BCH_MEMBER_REPLACEMENT(mi, v);
1021                         bch2_write_super(c);
1022                 }
1023                 mutex_unlock(&c->sb_lock);
1024         }
1025
1026         if (attr == &sysfs_label) {
1027                 char *tmp;
1028                 int ret;
1029
1030                 tmp = kstrdup(buf, GFP_KERNEL);
1031                 if (!tmp)
1032                         return -ENOMEM;
1033
1034                 ret = bch2_dev_group_set(c, ca, strim(tmp));
1035                 kfree(tmp);
1036                 if (ret)
1037                         return ret;
1038         }
1039
1040         if (attr == &sysfs_wake_allocator)
1041                 bch2_wake_allocator(ca);
1042
1043         return size;
1044 }
1045 SYSFS_OPS(bch2_dev);
1046
1047 struct attribute *bch2_dev_files[] = {
1048         &sysfs_uuid,
1049         &sysfs_bucket_size,
1050         &sysfs_block_size,
1051         &sysfs_first_bucket,
1052         &sysfs_nbuckets,
1053         &sysfs_durability,
1054
1055         /* settings: */
1056         &sysfs_discard,
1057         &sysfs_cache_replacement_policy,
1058         &sysfs_state_rw,
1059         &sysfs_label,
1060
1061         &sysfs_has_data,
1062         &sysfs_iodone,
1063
1064         &sysfs_io_latency_read,
1065         &sysfs_io_latency_write,
1066         &sysfs_io_latency_stats_read,
1067         &sysfs_io_latency_stats_write,
1068         &sysfs_congested,
1069
1070         /* alloc info - other stats: */
1071         &sysfs_bucket_quantiles_last_read,
1072         &sysfs_bucket_quantiles_last_write,
1073         &sysfs_bucket_quantiles_fragmentation,
1074         &sysfs_bucket_quantiles_oldest_gen,
1075
1076         &sysfs_reserve_stats,
1077
1078         /* debug: */
1079         &sysfs_alloc_debug,
1080         &sysfs_wake_allocator,
1081
1082         sysfs_pd_controller_files(copy_gc),
1083         NULL
1084 };
1085
1086 #endif  /* _BCACHEFS_SYSFS_H_ */