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

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * bcache sysfs interfaces
   4  *
   5  * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
   6  * Copyright 2012 Google, Inc.
   7  */
   8
   9 #ifndef NO_BCACHEFS_SYSFS
  10
  11 #include "bcachefs.h"
  12 #include "alloc_background.h"
  13 #include "sysfs.h"
  14 #include "btree_cache.h"
  15 #include "btree_io.h"
  16 #include "btree_iter.h"
  17 #include "btree_key_cache.h"
  18 #include "btree_update.h"
  19 #include "btree_update_interior.h"
  20 #include "btree_gc.h"
  21 #include "buckets.h"
  22 #include "clock.h"
  23 #include "disk_groups.h"
  24 #include "ec.h"
  25 #include "inode.h"
  26 #include "journal.h"
  27 #include "keylist.h"
  28 #include "move.h"
  29 #include "opts.h"
  30 #include "rebalance.h"
  31 #include "replicas.h"
  32 #include "super-io.h"
  33 #include "tests.h"
  34
  35 #include <linux/blkdev.h>
  36 #include <linux/sort.h>
  37 #include <linux/sched/clock.h>
  38
  39 #include "util.h"
  40
  41 #define SYSFS_OPS(type)                                                 \
  42 struct sysfs_ops type ## _sysfs_ops = {                                 \
  43         .show   = type ## _show,                                        \
  44         .store  = type ## _store                                        \
  45 }
  46
  47 #define SHOW(fn)                                                        \
  48 static ssize_t fn ## _show(struct kobject *kobj, struct attribute *attr,\
  49                            char *buf)                                   \
  50
  51 #define STORE(fn)                                                       \
  52 static ssize_t fn ## _store(struct kobject *kobj, struct attribute *attr,\
  53                             const char *buf, size_t size)               \
  54
  55 #define __sysfs_attribute(_name, _mode)                                 \
  56         static struct attribute sysfs_##_name =                         \
  57                 { .name = #_name, .mode = _mode }
  58
  59 #define write_attribute(n)      __sysfs_attribute(n, S_IWUSR)
  60 #define read_attribute(n)       __sysfs_attribute(n, S_IRUGO)
  61 #define rw_attribute(n)         __sysfs_attribute(n, S_IRUGO|S_IWUSR)
  62
  63 #define sysfs_printf(file, fmt, ...)                                    \
  64 do {                                                                    \
  65         if (attr == &sysfs_ ## file)                                    \
  66                 return scnprintf(buf, PAGE_SIZE, fmt "\n", __VA_ARGS__);\
  67 } while (0)
  68
  69 #define sysfs_print(file, var)                                          \
  70 do {                                                                    \
  71         if (attr == &sysfs_ ## file)                                    \
  72                 return snprint(buf, PAGE_SIZE, var);                    \
  73 } while (0)
  74
  75 #define sysfs_hprint(file, val)                                         \
  76 do {                                                                    \
  77         if (attr == &sysfs_ ## file) {                                  \
  78                 bch2_hprint(&out, val);                                 \
  79                 pr_buf(&out, "\n");                                     \
  80                 return out.pos - buf;                                   \
  81         }                                                               \
  82 } while (0)
  83
  84 #define var_printf(_var, fmt)   sysfs_printf(_var, fmt, var(_var))
  85 #define var_print(_var)         sysfs_print(_var, var(_var))
  86 #define var_hprint(_var)        sysfs_hprint(_var, var(_var))
  87
  88 #define sysfs_strtoul(file, var)                                        \
  89 do {                                                                    \
  90         if (attr == &sysfs_ ## file)                                    \
  91                 return strtoul_safe(buf, var) ?: (ssize_t) size;        \
  92 } while (0)
  93
  94 #define sysfs_strtoul_clamp(file, var, min, max)                        \
  95 do {                                                                    \
  96         if (attr == &sysfs_ ## file)                                    \
  97                 return strtoul_safe_clamp(buf, var, min, max)           \
  98                         ?: (ssize_t) size;                              \
  99 } while (0)
 100
 101 #define strtoul_or_return(cp)                                           \
 102 ({                                                                      \
 103         unsigned long _v;                                               \
 104         int _r = kstrtoul(cp, 10, &_v);                                 \
 105         if (_r)                                                         \
 106                 return _r;                                              \
 107         _v;                                                             \
 108 })
 109
 110 #define strtoul_restrict_or_return(cp, min, max)                        \
 111 ({                                                                      \
 112         unsigned long __v = 0;                                          \
 113         int _r = strtoul_safe_restrict(cp, __v, min, max);              \
 114         if (_r)                                                         \
 115                 return _r;                                              \
 116         __v;                                                            \
 117 })
 118
 119 #define strtoi_h_or_return(cp)                                          \
 120 ({                                                                      \
 121         u64 _v;                                                         \
 122         int _r = strtoi_h(cp, &_v);                                     \
 123         if (_r)                                                         \
 124                 return _r;                                              \
 125         _v;                                                             \
 126 })
 127
 128 #define sysfs_hatoi(file, var)                                          \
 129 do {                                                                    \
 130         if (attr == &sysfs_ ## file)                                    \
 131                 return strtoi_h(buf, &var) ?: (ssize_t) size;           \
 132 } while (0)
 133
 134 write_attribute(trigger_journal_flush);
 135 write_attribute(trigger_btree_coalesce);
 136 write_attribute(trigger_gc);
 137 write_attribute(prune_cache);
 138 rw_attribute(btree_gc_periodic);
 139
 140 read_attribute(uuid);
 141 read_attribute(minor);
 142 read_attribute(bucket_size);
 143 read_attribute(block_size);
 144 read_attribute(btree_node_size);
 145 read_attribute(first_bucket);
 146 read_attribute(nbuckets);
 147 read_attribute(durability);
 148 read_attribute(iodone);
 149
 150 read_attribute(io_latency_read);
 151 read_attribute(io_latency_write);
 152 read_attribute(io_latency_stats_read);
 153 read_attribute(io_latency_stats_write);
 154 read_attribute(congested);
 155
 156 read_attribute(bucket_quantiles_last_read);
 157 read_attribute(bucket_quantiles_last_write);
 158 read_attribute(bucket_quantiles_fragmentation);
 159 read_attribute(bucket_quantiles_oldest_gen);
 160
 161 read_attribute(reserve_stats);
 162 read_attribute(btree_cache_size);
 163 read_attribute(compression_stats);
 164 read_attribute(journal_debug);
 165 read_attribute(journal_pins);
 166 read_attribute(btree_updates);
 167 read_attribute(dirty_btree_nodes);
 168 read_attribute(btree_key_cache);
 169 read_attribute(btree_transactions);
 170 read_attribute(stripes_heap);
 171
 172 read_attribute(internal_uuid);
 173
 174 read_attribute(has_data);
 175 read_attribute(alloc_debug);
 176 write_attribute(wake_allocator);
 177
 178 read_attribute(read_realloc_races);
 179 read_attribute(extent_migrate_done);
 180 read_attribute(extent_migrate_raced);
 181
 182 rw_attribute(journal_write_delay_ms);
 183 rw_attribute(journal_reclaim_delay_ms);
 184
 185 rw_attribute(discard);
 186 rw_attribute(cache_replacement_policy);
 187 rw_attribute(label);
 188
 189 rw_attribute(copy_gc_enabled);
 190 sysfs_pd_controller_attribute(copy_gc);
 191
 192 rw_attribute(rebalance_enabled);
 193 sysfs_pd_controller_attribute(rebalance);
 194 read_attribute(rebalance_work);
 195 rw_attribute(promote_whole_extents);
 196
 197 read_attribute(new_stripes);
 198
 199 rw_attribute(pd_controllers_update_seconds);
 200
 201 read_attribute(meta_replicas_have);
 202 read_attribute(data_replicas_have);
 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 BCH_DEBUG_PARAM(name, description)                              \
 212         rw_attribute(name);
 213
 214         BCH_DEBUG_PARAMS()
 215 #undef BCH_DEBUG_PARAM
 216
 217 #define x(_name)                                                \
 218         static struct attribute sysfs_time_stat_##_name =               \
 219                 { .name = #_name, .mode = S_IRUGO };
 220         BCH_TIME_STATS()
 221 #undef x
 222
 223 static struct attribute sysfs_state_rw = {
 224         .name = "state",
 225         .mode = S_IRUGO
 226 };
 227
 228 static size_t bch2_btree_cache_size(struct bch_fs *c)
 229 {
 230         size_t ret = 0;
 231         struct btree *b;
 232
 233         mutex_lock(&c->btree_cache.lock);
 234         list_for_each_entry(b, &c->btree_cache.live, list)
 235                 ret += btree_bytes(c);
 236
 237         mutex_unlock(&c->btree_cache.lock);
 238         return ret;
 239 }
 240
 241 static int fs_alloc_debug_to_text(struct printbuf *out, struct bch_fs *c)
 242 {
 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         return 0;
 254 }
 255
 256 static int bch2_compression_stats_to_text(struct printbuf *out, struct bch_fs *c)
 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         pr_buf(out,
 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         return 0;
 313 }
 314
 315 SHOW(bch2_fs)
 316 {
 317         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 318         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 319
 320         sysfs_print(minor,                      c->minor);
 321         sysfs_printf(internal_uuid, "%pU",      c->sb.uuid.b);
 322
 323         sysfs_print(journal_write_delay_ms,     c->journal.write_delay_ms);
 324         sysfs_print(journal_reclaim_delay_ms,   c->journal.reclaim_delay_ms);
 325
 326         sysfs_print(block_size,                 block_bytes(c));
 327         sysfs_print(btree_node_size,            btree_bytes(c));
 328         sysfs_hprint(btree_cache_size,          bch2_btree_cache_size(c));
 329
 330         sysfs_print(read_realloc_races,
 331                     atomic_long_read(&c->read_realloc_races));
 332         sysfs_print(extent_migrate_done,
 333                     atomic_long_read(&c->extent_migrate_done));
 334         sysfs_print(extent_migrate_raced,
 335                     atomic_long_read(&c->extent_migrate_raced));
 336
 337         sysfs_printf(btree_gc_periodic, "%u",   (int) c->btree_gc_periodic);
 338
 339         sysfs_printf(copy_gc_enabled, "%i", c->copy_gc_enabled);
 340
 341         sysfs_print(pd_controllers_update_seconds,
 342                     c->pd_controllers_update_seconds);
 343
 344         sysfs_printf(rebalance_enabled,         "%i", c->rebalance.enabled);
 345         sysfs_pd_controller_show(rebalance,     &c->rebalance.pd); /* XXX */
 346         sysfs_pd_controller_show(copy_gc,       &c->copygc_pd);
 347
 348         if (attr == &sysfs_rebalance_work) {
 349                 bch2_rebalance_work_to_text(&out, c);
 350                 return out.pos - buf;
 351         }
 352
 353         sysfs_print(promote_whole_extents,      c->promote_whole_extents);
 354
 355         sysfs_printf(meta_replicas_have, "%i",  bch2_replicas_online(c, true));
 356         sysfs_printf(data_replicas_have, "%i",  bch2_replicas_online(c, false));
 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_key_cache) {
 384                 bch2_btree_key_cache_to_text(&out, &c->btree_key_cache);
 385                 return out.pos - buf;
 386         }
 387
 388         if (attr == &sysfs_btree_transactions) {
 389                 bch2_btree_trans_to_text(&out, c);
 390                 return out.pos - buf;
 391         }
 392
 393         if (attr == &sysfs_stripes_heap) {
 394                 bch2_stripes_heap_to_text(&out, c);
 395                 return out.pos - buf;
 396         }
 397
 398         if (attr == &sysfs_compression_stats) {
 399                 bch2_compression_stats_to_text(&out, c);
 400                 return out.pos - buf;
 401         }
 402
 403         if (attr == &sysfs_new_stripes) {
 404                 bch2_new_stripes_to_text(&out, c);
 405                 return out.pos - buf;
 406         }
 407
 408         if (attr == &sysfs_io_timers_read) {
 409                 bch2_io_timers_to_text(&out, &c->io_clock[READ]);
 410                 return out.pos - buf;
 411         }
 412         if (attr == &sysfs_io_timers_write) {
 413                 bch2_io_timers_to_text(&out, &c->io_clock[WRITE]);
 414                 return out.pos - buf;
 415         }
 416
 417 #define BCH_DEBUG_PARAM(name, description) sysfs_print(name, c->name);
 418         BCH_DEBUG_PARAMS()
 419 #undef BCH_DEBUG_PARAM
 420
 421         return 0;
 422 }
 423
 424 STORE(bch2_fs)
 425 {
 426         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 427
 428         sysfs_strtoul(journal_write_delay_ms, c->journal.write_delay_ms);
 429         sysfs_strtoul(journal_reclaim_delay_ms, c->journal.reclaim_delay_ms);
 430
 431         if (attr == &sysfs_btree_gc_periodic) {
 432                 ssize_t ret = strtoul_safe(buf, c->btree_gc_periodic)
 433                         ?: (ssize_t) size;
 434
 435                 wake_up_process(c->gc_thread);
 436                 return ret;
 437         }
 438
 439         if (attr == &sysfs_copy_gc_enabled) {
 440                 ssize_t ret = strtoul_safe(buf, c->copy_gc_enabled)
 441                         ?: (ssize_t) size;
 442
 443                 if (c->copygc_thread)
 444                         wake_up_process(c->copygc_thread);
 445                 return ret;
 446         }
 447
 448         if (attr == &sysfs_rebalance_enabled) {
 449                 ssize_t ret = strtoul_safe(buf, c->rebalance.enabled)
 450                         ?: (ssize_t) size;
 451
 452                 rebalance_wakeup(c);
 453                 return ret;
 454         }
 455
 456         sysfs_strtoul(pd_controllers_update_seconds,
 457                       c->pd_controllers_update_seconds);
 458         sysfs_pd_controller_store(rebalance,    &c->rebalance.pd);
 459         sysfs_pd_controller_store(copy_gc,      &c->copygc_pd);
 460
 461         sysfs_strtoul(promote_whole_extents,    c->promote_whole_extents);
 462
 463         /* Debugging: */
 464
 465 #define BCH_DEBUG_PARAM(name, description) sysfs_strtoul(name, c->name);
 466         BCH_DEBUG_PARAMS()
 467 #undef BCH_DEBUG_PARAM
 468
 469         if (!test_bit(BCH_FS_STARTED, &c->flags))
 470                 return -EPERM;
 471
 472         /* Debugging: */
 473
 474         if (attr == &sysfs_trigger_journal_flush)
 475                 bch2_journal_meta_async(&c->journal, NULL);
 476
 477         if (attr == &sysfs_trigger_btree_coalesce)
 478                 bch2_coalesce(c);
 479
 480         if (attr == &sysfs_trigger_gc) {
 481                 /*
 482                  * Full gc is currently incompatible with btree key cache:
 483                  */
 484 #if 0
 485                 down_read(&c->state_lock);
 486                 bch2_gc(c, NULL, false, false);
 487                 up_read(&c->state_lock);
 488 #else
 489                 bch2_gc_gens(c);
 490 #endif
 491         }
 492
 493         if (attr == &sysfs_prune_cache) {
 494                 struct shrink_control sc;
 495
 496                 sc.gfp_mask = GFP_KERNEL;
 497                 sc.nr_to_scan = strtoul_or_return(buf);
 498                 c->btree_cache.shrink.scan_objects(&c->btree_cache.shrink, &sc);
 499         }
 500
 501 #ifdef CONFIG_BCACHEFS_TESTS
 502         if (attr == &sysfs_perf_test) {
 503                 char *tmp = kstrdup(buf, GFP_KERNEL), *p = tmp;
 504                 char *test              = strsep(&p, " \t\n");
 505                 char *nr_str            = strsep(&p, " \t\n");
 506                 char *threads_str       = strsep(&p, " \t\n");
 507                 unsigned threads;
 508                 u64 nr;
 509                 int ret = -EINVAL;
 510
 511                 if (threads_str &&
 512                     !(ret = kstrtouint(threads_str, 10, &threads)) &&
 513                     !(ret = bch2_strtoull_h(nr_str, &nr)))
 514                         bch2_btree_perf_test(c, test, nr, threads);
 515                 else
 516                         size = ret;
 517                 kfree(tmp);
 518         }
 519 #endif
 520         return size;
 521 }
 522 SYSFS_OPS(bch2_fs);
 523
 524 struct attribute *bch2_fs_files[] = {
 525         &sysfs_minor,
 526         &sysfs_block_size,
 527         &sysfs_btree_node_size,
 528         &sysfs_btree_cache_size,
 529
 530         &sysfs_meta_replicas_have,
 531         &sysfs_data_replicas_have,
 532
 533         &sysfs_journal_write_delay_ms,
 534         &sysfs_journal_reclaim_delay_ms,
 535
 536         &sysfs_promote_whole_extents,
 537
 538         &sysfs_compression_stats,
 539
 540 #ifdef CONFIG_BCACHEFS_TESTS
 541         &sysfs_perf_test,
 542 #endif
 543         NULL
 544 };
 545
 546 /* internal dir - just a wrapper */
 547
 548 SHOW(bch2_fs_internal)
 549 {
 550         struct bch_fs *c = container_of(kobj, struct bch_fs, internal);
 551         return bch2_fs_show(&c->kobj, attr, buf);
 552 }
 553
 554 STORE(bch2_fs_internal)
 555 {
 556         struct bch_fs *c = container_of(kobj, struct bch_fs, internal);
 557         return bch2_fs_store(&c->kobj, attr, buf, size);
 558 }
 559 SYSFS_OPS(bch2_fs_internal);
 560
 561 struct attribute *bch2_fs_internal_files[] = {
 562         &sysfs_alloc_debug,
 563         &sysfs_journal_debug,
 564         &sysfs_journal_pins,
 565         &sysfs_btree_updates,
 566         &sysfs_dirty_btree_nodes,
 567         &sysfs_btree_key_cache,
 568         &sysfs_btree_transactions,
 569         &sysfs_stripes_heap,
 570
 571         &sysfs_read_realloc_races,
 572         &sysfs_extent_migrate_done,
 573         &sysfs_extent_migrate_raced,
 574
 575         &sysfs_trigger_journal_flush,
 576         &sysfs_trigger_btree_coalesce,
 577         &sysfs_trigger_gc,
 578         &sysfs_prune_cache,
 579
 580         &sysfs_copy_gc_enabled,
 581
 582         &sysfs_rebalance_enabled,
 583         &sysfs_rebalance_work,
 584         sysfs_pd_controller_files(rebalance),
 585         sysfs_pd_controller_files(copy_gc),
 586
 587         &sysfs_new_stripes,
 588
 589         &sysfs_io_timers_read,
 590         &sysfs_io_timers_write,
 591
 592         &sysfs_internal_uuid,
 593
 594 #define BCH_DEBUG_PARAM(name, description) &sysfs_##name,
 595         BCH_DEBUG_PARAMS()
 596 #undef BCH_DEBUG_PARAM
 597
 598         NULL
 599 };
 600
 601 /* options */
 602
 603 SHOW(bch2_fs_opts_dir)
 604 {
 605         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 606         struct bch_fs *c = container_of(kobj, struct bch_fs, opts_dir);
 607         const struct bch_option *opt = container_of(attr, struct bch_option, attr);
 608         int id = opt - bch2_opt_table;
 609         u64 v = bch2_opt_get_by_id(&c->opts, id);
 610
 611         bch2_opt_to_text(&out, c, opt, v, OPT_SHOW_FULL_LIST);
 612         pr_buf(&out, "\n");
 613
 614         return out.pos - buf;
 615 }
 616
 617 STORE(bch2_fs_opts_dir)
 618 {
 619         struct bch_fs *c = container_of(kobj, struct bch_fs, opts_dir);
 620         const struct bch_option *opt = container_of(attr, struct bch_option, attr);
 621         int ret, id = opt - bch2_opt_table;
 622         char *tmp;
 623         u64 v;
 624
 625         tmp = kstrdup(buf, GFP_KERNEL);
 626         if (!tmp)
 627                 return -ENOMEM;
 628
 629         ret = bch2_opt_parse(c, opt, strim(tmp), &v);
 630         kfree(tmp);
 631
 632         if (ret < 0)
 633                 return ret;
 634
 635         ret = bch2_opt_check_may_set(c, id, v);
 636         if (ret < 0)
 637                 return ret;
 638
 639         if (opt->set_sb != SET_NO_SB_OPT) {
 640                 mutex_lock(&c->sb_lock);
 641                 opt->set_sb(c->disk_sb.sb, v);
 642                 bch2_write_super(c);
 643                 mutex_unlock(&c->sb_lock);
 644         }
 645
 646         bch2_opt_set_by_id(&c->opts, id, v);
 647
 648         if ((id == Opt_background_target ||
 649              id == Opt_background_compression) && v) {
 650                 bch2_rebalance_add_work(c, S64_MAX);
 651                 rebalance_wakeup(c);
 652         }
 653
 654         return size;
 655 }
 656 SYSFS_OPS(bch2_fs_opts_dir);
 657
 658 struct attribute *bch2_fs_opts_dir_files[] = { NULL };
 659
 660 int bch2_opts_create_sysfs_files(struct kobject *kobj)
 661 {
 662         const struct bch_option *i;
 663         int ret;
 664
 665         for (i = bch2_opt_table;
 666              i < bch2_opt_table + bch2_opts_nr;
 667              i++) {
 668                 if (!(i->mode & (OPT_FORMAT|OPT_MOUNT|OPT_RUNTIME)))
 669                         continue;
 670
 671                 ret = sysfs_create_file(kobj, &i->attr);
 672                 if (ret)
 673                         return ret;
 674         }
 675
 676         return 0;
 677 }
 678
 679 /* time stats */
 680
 681 SHOW(bch2_fs_time_stats)
 682 {
 683         struct bch_fs *c = container_of(kobj, struct bch_fs, time_stats);
 684         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 685
 686 #define x(name)                                                         \
 687         if (attr == &sysfs_time_stat_##name) {                          \
 688                 bch2_time_stats_to_text(&out, &c->times[BCH_TIME_##name]);\
 689                 return out.pos - buf;                                   \
 690         }
 691         BCH_TIME_STATS()
 692 #undef x
 693
 694         return 0;
 695 }
 696
 697 STORE(bch2_fs_time_stats)
 698 {
 699         return size;
 700 }
 701 SYSFS_OPS(bch2_fs_time_stats);
 702
 703 struct attribute *bch2_fs_time_stats_files[] = {
 704 #define x(name)                                         \
 705         &sysfs_time_stat_##name,
 706         BCH_TIME_STATS()
 707 #undef x
 708         NULL
 709 };
 710
 711 typedef unsigned (bucket_map_fn)(struct bch_fs *, struct bch_dev *,
 712                                  size_t, void *);
 713
 714 static unsigned bucket_last_io_fn(struct bch_fs *c, struct bch_dev *ca,
 715                                   size_t b, void *private)
 716 {
 717         int rw = (private ? 1 : 0);
 718
 719         return bucket_last_io(c, bucket(ca, b), rw);
 720 }
 721
 722 static unsigned bucket_sectors_used_fn(struct bch_fs *c, struct bch_dev *ca,
 723                                        size_t b, void *private)
 724 {
 725         struct bucket *g = bucket(ca, b);
 726         return bucket_sectors_used(g->mark);
 727 }
 728
 729 static unsigned bucket_oldest_gen_fn(struct bch_fs *c, struct bch_dev *ca,
 730                                      size_t b, void *private)
 731 {
 732         return bucket_gc_gen(ca, b);
 733 }
 734
 735 static int unsigned_cmp(const void *_l, const void *_r)
 736 {
 737         const unsigned *l = _l;
 738         const unsigned *r = _r;
 739
 740         return cmp_int(*l, *r);
 741 }
 742
 743 static int quantiles_to_text(struct printbuf *out,
 744                              struct bch_fs *c, struct bch_dev *ca,
 745                              bucket_map_fn *fn, void *private)
 746 {
 747         size_t i, n;
 748         /* Compute 31 quantiles */
 749         unsigned q[31], *p;
 750
 751         down_read(&ca->bucket_lock);
 752         n = ca->mi.nbuckets;
 753
 754         p = vzalloc(n * sizeof(unsigned));
 755         if (!p) {
 756                 up_read(&ca->bucket_lock);
 757                 return -ENOMEM;
 758         }
 759
 760         for (i = ca->mi.first_bucket; i < n; i++)
 761                 p[i] = fn(c, ca, i, private);
 762
 763         sort(p, n, sizeof(unsigned), unsigned_cmp, NULL);
 764         up_read(&ca->bucket_lock);
 765
 766         while (n &&
 767                !p[n - 1])
 768                 --n;
 769
 770         for (i = 0; i < ARRAY_SIZE(q); i++)
 771                 q[i] = p[n * (i + 1) / (ARRAY_SIZE(q) + 1)];
 772
 773         vfree(p);
 774
 775         for (i = 0; i < ARRAY_SIZE(q); i++)
 776                 pr_buf(out, "%u ", q[i]);
 777         pr_buf(out, "\n");
 778         return 0;
 779 }
 780
 781 static void reserve_stats_to_text(struct printbuf *out, struct bch_dev *ca)
 782 {
 783         enum alloc_reserve i;
 784
 785         spin_lock(&ca->fs->freelist_lock);
 786
 787         pr_buf(out, "free_inc:\t%zu\t%zu\n",
 788                fifo_used(&ca->free_inc),
 789                ca->free_inc.size);
 790
 791         for (i = 0; i < RESERVE_NR; i++)
 792                 pr_buf(out, "free[%u]:\t%zu\t%zu\n", i,
 793                        fifo_used(&ca->free[i]),
 794                        ca->free[i].size);
 795
 796         spin_unlock(&ca->fs->freelist_lock);
 797 }
 798
 799 static void dev_alloc_debug_to_text(struct printbuf *out, struct bch_dev *ca)
 800 {
 801         struct bch_fs *c = ca->fs;
 802         struct bch_dev_usage stats = bch2_dev_usage_read(ca);
 803         unsigned i, nr[BCH_DATA_NR];
 804
 805         memset(nr, 0, sizeof(nr));
 806
 807         for (i = 0; i < ARRAY_SIZE(c->open_buckets); i++)
 808                 nr[c->open_buckets[i].type]++;
 809
 810         pr_buf(out,
 811                 "free_inc:               %zu/%zu\n"
 812                 "free[RESERVE_BTREE]:    %zu/%zu\n"
 813                 "free[RESERVE_MOVINGGC]: %zu/%zu\n"
 814                 "free[RESERVE_NONE]:     %zu/%zu\n"
 815                 "buckets:\n"
 816                 "    capacity:           %llu\n"
 817                 "    alloc:              %llu\n"
 818                 "    sb:                 %llu\n"
 819                 "    journal:            %llu\n"
 820                 "    meta:               %llu\n"
 821                 "    user:               %llu\n"
 822                 "    cached:             %llu\n"
 823                 "    erasure coded:      %llu\n"
 824                 "    available:          %lli\n"
 825                 "sectors:\n"
 826                 "    sb:                 %llu\n"
 827                 "    journal:            %llu\n"
 828                 "    meta:               %llu\n"
 829                 "    user:               %llu\n"
 830                 "    cached:             %llu\n"
 831                 "    erasure coded:      %llu\n"
 832                 "    fragmented:         %llu\n"
 833                 "    copygc threshold:   %llu\n"
 834                 "freelist_wait:          %s\n"
 835                 "open buckets:           %u/%u (reserved %u)\n"
 836                 "open_buckets_wait:      %s\n"
 837                 "open_buckets_btree:     %u\n"
 838                 "open_buckets_user:      %u\n"
 839                 "btree reserve cache:    %u\n",
 840                 fifo_used(&ca->free_inc),               ca->free_inc.size,
 841                 fifo_used(&ca->free[RESERVE_BTREE]),    ca->free[RESERVE_BTREE].size,
 842                 fifo_used(&ca->free[RESERVE_MOVINGGC]), ca->free[RESERVE_MOVINGGC].size,
 843                 fifo_used(&ca->free[RESERVE_NONE]),     ca->free[RESERVE_NONE].size,
 844                 ca->mi.nbuckets - ca->mi.first_bucket,
 845                 stats.buckets_alloc,
 846                 stats.buckets[BCH_DATA_sb],
 847                 stats.buckets[BCH_DATA_journal],
 848                 stats.buckets[BCH_DATA_btree],
 849                 stats.buckets[BCH_DATA_user],
 850                 stats.buckets[BCH_DATA_cached],
 851                 stats.buckets_ec,
 852                 __dev_buckets_available(ca, stats),
 853                 stats.sectors[BCH_DATA_sb],
 854                 stats.sectors[BCH_DATA_journal],
 855                 stats.sectors[BCH_DATA_btree],
 856                 stats.sectors[BCH_DATA_user],
 857                 stats.sectors[BCH_DATA_cached],
 858                 stats.sectors_ec,
 859                 stats.sectors_fragmented,
 860                 c->copygc_threshold,
 861                 c->freelist_wait.list.first             ? "waiting" : "empty",
 862                 c->open_buckets_nr_free, OPEN_BUCKETS_COUNT,
 863                 BTREE_NODE_OPEN_BUCKET_RESERVE,
 864                 c->open_buckets_wait.list.first         ? "waiting" : "empty",
 865                 nr[BCH_DATA_btree],
 866                 nr[BCH_DATA_user],
 867                 c->btree_reserve_cache_nr);
 868 }
 869
 870 static const char * const bch2_rw[] = {
 871         "read",
 872         "write",
 873         NULL
 874 };
 875
 876 static void dev_iodone_to_text(struct printbuf *out, struct bch_dev *ca)
 877 {
 878         int rw, i;
 879
 880         for (rw = 0; rw < 2; rw++) {
 881                 pr_buf(out, "%s:\n", bch2_rw[rw]);
 882
 883                 for (i = 1; i < BCH_DATA_NR; i++)
 884                         pr_buf(out, "%-12s:%12llu\n",
 885                                bch2_data_types[i],
 886                                percpu_u64_get(&ca->io_done->sectors[rw][i]) << 9);
 887         }
 888 }
 889
 890 SHOW(bch2_dev)
 891 {
 892         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
 893         struct bch_fs *c = ca->fs;
 894         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 895
 896         sysfs_printf(uuid,              "%pU\n", ca->uuid.b);
 897
 898         sysfs_print(bucket_size,        bucket_bytes(ca));
 899         sysfs_print(block_size,         block_bytes(c));
 900         sysfs_print(first_bucket,       ca->mi.first_bucket);
 901         sysfs_print(nbuckets,           ca->mi.nbuckets);
 902         sysfs_print(durability,         ca->mi.durability);
 903         sysfs_print(discard,            ca->mi.discard);
 904
 905         if (attr == &sysfs_label) {
 906                 if (ca->mi.group) {
 907                         mutex_lock(&c->sb_lock);
 908                         bch2_disk_path_to_text(&out, &c->disk_sb,
 909                                                ca->mi.group - 1);
 910                         mutex_unlock(&c->sb_lock);
 911                 }
 912
 913                 pr_buf(&out, "\n");
 914                 return out.pos - buf;
 915         }
 916
 917         if (attr == &sysfs_has_data) {
 918                 bch2_flags_to_text(&out, bch2_data_types,
 919                                    bch2_dev_has_data(c, ca));
 920                 pr_buf(&out, "\n");
 921                 return out.pos - buf;
 922         }
 923
 924         if (attr == &sysfs_cache_replacement_policy) {
 925                 bch2_string_opt_to_text(&out,
 926                                         bch2_cache_replacement_policies,
 927                                         ca->mi.replacement);
 928                 pr_buf(&out, "\n");
 929                 return out.pos - buf;
 930         }
 931
 932         if (attr == &sysfs_state_rw) {
 933                 bch2_string_opt_to_text(&out, bch2_dev_state,
 934                                         ca->mi.state);
 935                 pr_buf(&out, "\n");
 936                 return out.pos - buf;
 937         }
 938
 939         if (attr == &sysfs_iodone) {
 940                 dev_iodone_to_text(&out, ca);
 941                 return out.pos - buf;
 942         }
 943
 944         sysfs_print(io_latency_read,            atomic64_read(&ca->cur_latency[READ]));
 945         sysfs_print(io_latency_write,           atomic64_read(&ca->cur_latency[WRITE]));
 946
 947         if (attr == &sysfs_io_latency_stats_read) {
 948                 bch2_time_stats_to_text(&out, &ca->io_latency[READ]);
 949                 return out.pos - buf;
 950         }
 951         if (attr == &sysfs_io_latency_stats_write) {
 952                 bch2_time_stats_to_text(&out, &ca->io_latency[WRITE]);
 953                 return out.pos - buf;
 954         }
 955
 956         sysfs_printf(congested,                 "%u%%",
 957                      clamp(atomic_read(&ca->congested), 0, CONGESTED_MAX)
 958                      * 100 / CONGESTED_MAX);
 959
 960         if (attr == &sysfs_bucket_quantiles_last_read)
 961                 return quantiles_to_text(&out, c, ca, bucket_last_io_fn, (void *) 0) ?: out.pos - buf;
 962         if (attr == &sysfs_bucket_quantiles_last_write)
 963                 return quantiles_to_text(&out, c, ca, bucket_last_io_fn, (void *) 1) ?: out.pos - buf;
 964         if (attr == &sysfs_bucket_quantiles_fragmentation)
 965                 return quantiles_to_text(&out, c, ca, bucket_sectors_used_fn, NULL)  ?: out.pos - buf;
 966         if (attr == &sysfs_bucket_quantiles_oldest_gen)
 967                 return quantiles_to_text(&out, c, ca, bucket_oldest_gen_fn, NULL)    ?: out.pos - buf;
 968
 969         if (attr == &sysfs_reserve_stats) {
 970                 reserve_stats_to_text(&out, ca);
 971                 return out.pos - buf;
 972         }
 973         if (attr == &sysfs_alloc_debug) {
 974                 dev_alloc_debug_to_text(&out, ca);
 975                 return out.pos - buf;
 976         }
 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         if (attr == &sysfs_discard) {
 988                 bool v = strtoul_or_return(buf);
 989
 990                 mutex_lock(&c->sb_lock);
 991                 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
 992
 993                 if (v != BCH_MEMBER_DISCARD(mi)) {
 994                         SET_BCH_MEMBER_DISCARD(mi, v);
 995                         bch2_write_super(c);
 996                 }
 997                 mutex_unlock(&c->sb_lock);
 998         }
 999
1000         if (attr == &sysfs_cache_replacement_policy) {
1001                 ssize_t v = __sysfs_match_string(bch2_cache_replacement_policies, -1, buf);
1002
1003                 if (v < 0)
1004                         return v;
1005
1006                 mutex_lock(&c->sb_lock);
1007                 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1008
1009                 if ((unsigned) v != BCH_MEMBER_REPLACEMENT(mi)) {
1010                         SET_BCH_MEMBER_REPLACEMENT(mi, v);
1011                         bch2_write_super(c);
1012                 }
1013                 mutex_unlock(&c->sb_lock);
1014         }
1015
1016         if (attr == &sysfs_label) {
1017                 char *tmp;
1018                 int ret;
1019
1020                 tmp = kstrdup(buf, GFP_KERNEL);
1021                 if (!tmp)
1022                         return -ENOMEM;
1023
1024                 ret = bch2_dev_group_set(c, ca, strim(tmp));
1025                 kfree(tmp);
1026                 if (ret)
1027                         return ret;
1028         }
1029
1030         if (attr == &sysfs_wake_allocator)
1031                 bch2_wake_allocator(ca);
1032
1033         return size;
1034 }
1035 SYSFS_OPS(bch2_dev);
1036
1037 struct attribute *bch2_dev_files[] = {
1038         &sysfs_uuid,
1039         &sysfs_bucket_size,
1040         &sysfs_block_size,
1041         &sysfs_first_bucket,
1042         &sysfs_nbuckets,
1043         &sysfs_durability,
1044
1045         /* settings: */
1046         &sysfs_discard,
1047         &sysfs_cache_replacement_policy,
1048         &sysfs_state_rw,
1049         &sysfs_label,
1050
1051         &sysfs_has_data,
1052         &sysfs_iodone,
1053
1054         &sysfs_io_latency_read,
1055         &sysfs_io_latency_write,
1056         &sysfs_io_latency_stats_read,
1057         &sysfs_io_latency_stats_write,
1058         &sysfs_congested,
1059
1060         /* alloc info - other stats: */
1061         &sysfs_bucket_quantiles_last_read,
1062         &sysfs_bucket_quantiles_last_write,
1063         &sysfs_bucket_quantiles_fragmentation,
1064         &sysfs_bucket_quantiles_oldest_gen,
1065
1066         &sysfs_reserve_stats,
1067
1068         /* debug: */
1069         &sysfs_alloc_debug,
1070         &sysfs_wake_allocator,
1071         NULL
1072 };
1073
1074 #endif  /* _BCACHEFS_SYSFS_H_ */