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

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * bcache sysfs interfaces
   4  *
   5  * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
   6  * Copyright 2012 Google, Inc.
   7  */
   8
   9 #ifndef NO_BCACHEFS_SYSFS
  10
  11 #include "bcachefs.h"
  12 #include "alloc_background.h"
  13 #include "sysfs.h"
  14 #include "btree_cache.h"
  15 #include "btree_io.h"
  16 #include "btree_iter.h"
  17 #include "btree_key_cache.h"
  18 #include "btree_update.h"
  19 #include "btree_update_interior.h"
  20 #include "btree_gc.h"
  21 #include "buckets.h"
  22 #include "clock.h"
  23 #include "disk_groups.h"
  24 #include "ec.h"
  25 #include "inode.h"
  26 #include "journal.h"
  27 #include "keylist.h"
  28 #include "move.h"
  29 #include "opts.h"
  30 #include "rebalance.h"
  31 #include "replicas.h"
  32 #include "super-io.h"
  33 #include "tests.h"
  34
  35 #include <linux/blkdev.h>
  36 #include <linux/sort.h>
  37 #include <linux/sched/clock.h>
  38
  39 #include "util.h"
  40
  41 #define SYSFS_OPS(type)                                                 \
  42 struct sysfs_ops type ## _sysfs_ops = {                                 \
  43         .show   = type ## _show,                                        \
  44         .store  = type ## _store                                        \
  45 }
  46
  47 #define SHOW(fn)                                                        \
  48 static ssize_t fn ## _show(struct kobject *kobj, struct attribute *attr,\
  49                            char *buf)                                   \
  50
  51 #define STORE(fn)                                                       \
  52 static ssize_t fn ## _store(struct kobject *kobj, struct attribute *attr,\
  53                             const char *buf, size_t size)               \
  54
  55 #define __sysfs_attribute(_name, _mode)                                 \
  56         static struct attribute sysfs_##_name =                         \
  57                 { .name = #_name, .mode = _mode }
  58
  59 #define write_attribute(n)      __sysfs_attribute(n, S_IWUSR)
  60 #define read_attribute(n)       __sysfs_attribute(n, S_IRUGO)
  61 #define rw_attribute(n)         __sysfs_attribute(n, S_IRUGO|S_IWUSR)
  62
  63 #define sysfs_printf(file, fmt, ...)                                    \
  64 do {                                                                    \
  65         if (attr == &sysfs_ ## file)                                    \
  66                 return scnprintf(buf, PAGE_SIZE, fmt "\n", __VA_ARGS__);\
  67 } while (0)
  68
  69 #define sysfs_print(file, var)                                          \
  70 do {                                                                    \
  71         if (attr == &sysfs_ ## file)                                    \
  72                 return snprint(buf, PAGE_SIZE, var);                    \
  73 } while (0)
  74
  75 #define sysfs_hprint(file, val)                                         \
  76 do {                                                                    \
  77         if (attr == &sysfs_ ## file) {                                  \
  78                 bch2_hprint(&out, val);                                 \
  79                 pr_buf(&out, "\n");                                     \
  80                 return out.pos - buf;                                   \
  81         }                                                               \
  82 } while (0)
  83
  84 #define var_printf(_var, fmt)   sysfs_printf(_var, fmt, var(_var))
  85 #define var_print(_var)         sysfs_print(_var, var(_var))
  86 #define var_hprint(_var)        sysfs_hprint(_var, var(_var))
  87
  88 #define sysfs_strtoul(file, var)                                        \
  89 do {                                                                    \
  90         if (attr == &sysfs_ ## file)                                    \
  91                 return strtoul_safe(buf, var) ?: (ssize_t) size;        \
  92 } while (0)
  93
  94 #define sysfs_strtoul_clamp(file, var, min, max)                        \
  95 do {                                                                    \
  96         if (attr == &sysfs_ ## file)                                    \
  97                 return strtoul_safe_clamp(buf, var, min, max)           \
  98                         ?: (ssize_t) size;                              \
  99 } while (0)
 100
 101 #define strtoul_or_return(cp)                                           \
 102 ({                                                                      \
 103         unsigned long _v;                                               \
 104         int _r = kstrtoul(cp, 10, &_v);                                 \
 105         if (_r)                                                         \
 106                 return _r;                                              \
 107         _v;                                                             \
 108 })
 109
 110 #define strtoul_restrict_or_return(cp, min, max)                        \
 111 ({                                                                      \
 112         unsigned long __v = 0;                                          \
 113         int _r = strtoul_safe_restrict(cp, __v, min, max);              \
 114         if (_r)                                                         \
 115                 return _r;                                              \
 116         __v;                                                            \
 117 })
 118
 119 #define strtoi_h_or_return(cp)                                          \
 120 ({                                                                      \
 121         u64 _v;                                                         \
 122         int _r = strtoi_h(cp, &_v);                                     \
 123         if (_r)                                                         \
 124                 return _r;                                              \
 125         _v;                                                             \
 126 })
 127
 128 #define sysfs_hatoi(file, var)                                          \
 129 do {                                                                    \
 130         if (attr == &sysfs_ ## file)                                    \
 131                 return strtoi_h(buf, &var) ?: (ssize_t) size;           \
 132 } while (0)
 133
 134 write_attribute(trigger_journal_flush);
 135 write_attribute(trigger_gc);
 136 write_attribute(prune_cache);
 137 rw_attribute(btree_gc_periodic);
 138 rw_attribute(gc_gens_pos);
 139
 140 read_attribute(uuid);
 141 read_attribute(minor);
 142 read_attribute(bucket_size);
 143 read_attribute(block_size);
 144 read_attribute(btree_node_size);
 145 read_attribute(first_bucket);
 146 read_attribute(nbuckets);
 147 read_attribute(durability);
 148 read_attribute(iodone);
 149
 150 read_attribute(io_latency_read);
 151 read_attribute(io_latency_write);
 152 read_attribute(io_latency_stats_read);
 153 read_attribute(io_latency_stats_write);
 154 read_attribute(congested);
 155
 156 read_attribute(btree_avg_write_size);
 157
 158 read_attribute(bucket_quantiles_last_read);
 159 read_attribute(bucket_quantiles_last_write);
 160 read_attribute(bucket_quantiles_fragmentation);
 161 read_attribute(bucket_quantiles_oldest_gen);
 162
 163 read_attribute(reserve_stats);
 164 read_attribute(btree_cache_size);
 165 read_attribute(compression_stats);
 166 read_attribute(journal_debug);
 167 read_attribute(journal_pins);
 168 read_attribute(btree_updates);
 169 read_attribute(dirty_btree_nodes);
 170 read_attribute(btree_cache);
 171 read_attribute(btree_key_cache);
 172 read_attribute(btree_transactions);
 173 read_attribute(stripes_heap);
 174
 175 read_attribute(internal_uuid);
 176
 177 read_attribute(has_data);
 178 read_attribute(alloc_debug);
 179 write_attribute(wake_allocator);
 180
 181 read_attribute(read_realloc_races);
 182 read_attribute(extent_migrate_done);
 183 read_attribute(extent_migrate_raced);
 184
 185 rw_attribute(journal_write_delay_ms);
 186 rw_attribute(journal_reclaim_delay_ms);
 187
 188 rw_attribute(discard);
 189 rw_attribute(cache_replacement_policy);
 190 rw_attribute(label);
 191
 192 rw_attribute(copy_gc_enabled);
 193 read_attribute(copy_gc_wait);
 194
 195 rw_attribute(rebalance_enabled);
 196 sysfs_pd_controller_attribute(rebalance);
 197 read_attribute(rebalance_work);
 198 rw_attribute(promote_whole_extents);
 199
 200 read_attribute(new_stripes);
 201
 202 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 x(_name)                                                \
 210         static struct attribute sysfs_time_stat_##_name =               \
 211                 { .name = #_name, .mode = S_IRUGO };
 212         BCH_TIME_STATS()
 213 #undef x
 214
 215 static struct attribute sysfs_state_rw = {
 216         .name = "state",
 217         .mode = S_IRUGO
 218 };
 219
 220 static size_t bch2_btree_cache_size(struct bch_fs *c)
 221 {
 222         size_t ret = 0;
 223         struct btree *b;
 224
 225         mutex_lock(&c->btree_cache.lock);
 226         list_for_each_entry(b, &c->btree_cache.live, list)
 227                 ret += btree_bytes(c);
 228
 229         mutex_unlock(&c->btree_cache.lock);
 230         return ret;
 231 }
 232
 233 static size_t bch2_btree_avg_write_size(struct bch_fs *c)
 234 {
 235         u64 nr = atomic64_read(&c->btree_writes_nr);
 236         u64 sectors = atomic64_read(&c->btree_writes_sectors);
 237
 238         return nr ? div64_u64(sectors, nr) : 0;
 239 }
 240
 241 static int fs_alloc_debug_to_text(struct printbuf *out, struct bch_fs *c)
 242 {
 243         struct bch_fs_usage_online *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 void bch2_gc_gens_pos_to_text(struct printbuf *out, struct bch_fs *c)
 316 {
 317         pr_buf(out, "%s: ", bch2_btree_ids[c->gc_gens_btree]);
 318         bch2_bpos_to_text(out, c->gc_gens_pos);
 319         pr_buf(out, "\n");
 320 }
 321
 322 SHOW(bch2_fs)
 323 {
 324         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 325         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 326
 327         sysfs_print(minor,                      c->minor);
 328         sysfs_printf(internal_uuid, "%pU",      c->sb.uuid.b);
 329
 330         sysfs_print(journal_write_delay_ms,     c->journal.write_delay_ms);
 331         sysfs_print(journal_reclaim_delay_ms,   c->journal.reclaim_delay_ms);
 332
 333         sysfs_print(block_size,                 block_bytes(c));
 334         sysfs_print(btree_node_size,            btree_bytes(c));
 335         sysfs_hprint(btree_cache_size,          bch2_btree_cache_size(c));
 336         sysfs_hprint(btree_avg_write_size,      bch2_btree_avg_write_size(c));
 337
 338         sysfs_print(read_realloc_races,
 339                     atomic_long_read(&c->read_realloc_races));
 340         sysfs_print(extent_migrate_done,
 341                     atomic_long_read(&c->extent_migrate_done));
 342         sysfs_print(extent_migrate_raced,
 343                     atomic_long_read(&c->extent_migrate_raced));
 344
 345         sysfs_printf(btree_gc_periodic, "%u",   (int) c->btree_gc_periodic);
 346
 347         if (attr == &sysfs_gc_gens_pos) {
 348                 bch2_gc_gens_pos_to_text(&out, c);
 349                 return out.pos - buf;
 350         }
 351
 352         sysfs_printf(copy_gc_enabled, "%i", c->copy_gc_enabled);
 353
 354         sysfs_printf(rebalance_enabled,         "%i", c->rebalance.enabled);
 355         sysfs_pd_controller_show(rebalance,     &c->rebalance.pd); /* XXX */
 356         sysfs_hprint(copy_gc_wait,
 357                      max(0LL, c->copygc_wait -
 358                          atomic64_read(&c->io_clock[WRITE].now)) << 9);
 359
 360         if (attr == &sysfs_rebalance_work) {
 361                 bch2_rebalance_work_to_text(&out, c);
 362                 return out.pos - buf;
 363         }
 364
 365         sysfs_print(promote_whole_extents,      c->promote_whole_extents);
 366
 367         /* Debugging: */
 368
 369         if (attr == &sysfs_alloc_debug)
 370                 return fs_alloc_debug_to_text(&out, c) ?: out.pos - buf;
 371
 372         if (attr == &sysfs_journal_debug) {
 373                 bch2_journal_debug_to_text(&out, &c->journal);
 374                 return out.pos - buf;
 375         }
 376
 377         if (attr == &sysfs_journal_pins) {
 378                 bch2_journal_pins_to_text(&out, &c->journal);
 379                 return out.pos - buf;
 380         }
 381
 382         if (attr == &sysfs_btree_updates) {
 383                 bch2_btree_updates_to_text(&out, c);
 384                 return out.pos - buf;
 385         }
 386
 387         if (attr == &sysfs_dirty_btree_nodes) {
 388                 bch2_dirty_btree_nodes_to_text(&out, c);
 389                 return out.pos - buf;
 390         }
 391
 392         if (attr == &sysfs_btree_cache) {
 393                 bch2_btree_cache_to_text(&out, c);
 394                 return out.pos - buf;
 395         }
 396
 397         if (attr == &sysfs_btree_key_cache) {
 398                 bch2_btree_key_cache_to_text(&out, &c->btree_key_cache);
 399                 return out.pos - buf;
 400         }
 401
 402         if (attr == &sysfs_btree_transactions) {
 403                 bch2_btree_trans_to_text(&out, c);
 404                 return out.pos - buf;
 405         }
 406
 407         if (attr == &sysfs_stripes_heap) {
 408                 bch2_stripes_heap_to_text(&out, c);
 409                 return out.pos - buf;
 410         }
 411
 412         if (attr == &sysfs_compression_stats) {
 413                 bch2_compression_stats_to_text(&out, c);
 414                 return out.pos - buf;
 415         }
 416
 417         if (attr == &sysfs_new_stripes) {
 418                 bch2_new_stripes_to_text(&out, c);
 419                 return out.pos - buf;
 420         }
 421
 422         if (attr == &sysfs_io_timers_read) {
 423                 bch2_io_timers_to_text(&out, &c->io_clock[READ]);
 424                 return out.pos - buf;
 425         }
 426         if (attr == &sysfs_io_timers_write) {
 427                 bch2_io_timers_to_text(&out, &c->io_clock[WRITE]);
 428                 return out.pos - buf;
 429         }
 430
 431         return 0;
 432 }
 433
 434 STORE(bch2_fs)
 435 {
 436         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 437
 438         sysfs_strtoul(journal_write_delay_ms, c->journal.write_delay_ms);
 439         sysfs_strtoul(journal_reclaim_delay_ms, c->journal.reclaim_delay_ms);
 440
 441         if (attr == &sysfs_btree_gc_periodic) {
 442                 ssize_t ret = strtoul_safe(buf, c->btree_gc_periodic)
 443                         ?: (ssize_t) size;
 444
 445                 wake_up_process(c->gc_thread);
 446                 return ret;
 447         }
 448
 449         if (attr == &sysfs_copy_gc_enabled) {
 450                 ssize_t ret = strtoul_safe(buf, c->copy_gc_enabled)
 451                         ?: (ssize_t) size;
 452
 453                 if (c->copygc_thread)
 454                         wake_up_process(c->copygc_thread);
 455                 return ret;
 456         }
 457
 458         if (attr == &sysfs_rebalance_enabled) {
 459                 ssize_t ret = strtoul_safe(buf, c->rebalance.enabled)
 460                         ?: (ssize_t) size;
 461
 462                 rebalance_wakeup(c);
 463                 return ret;
 464         }
 465
 466         sysfs_pd_controller_store(rebalance,    &c->rebalance.pd);
 467
 468         sysfs_strtoul(promote_whole_extents,    c->promote_whole_extents);
 469
 470         /* Debugging: */
 471
 472         if (!test_bit(BCH_FS_STARTED, &c->flags))
 473                 return -EPERM;
 474
 475         /* Debugging: */
 476
 477         if (attr == &sysfs_trigger_journal_flush)
 478                 bch2_journal_meta(&c->journal);
 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, 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                         ret = bch2_btree_perf_test(c, test, nr, threads);
 515                 kfree(tmp);
 516
 517                 if (ret)
 518                         size = ret;
 519         }
 520 #endif
 521         return size;
 522 }
 523 SYSFS_OPS(bch2_fs);
 524
 525 struct attribute *bch2_fs_files[] = {
 526         &sysfs_minor,
 527         &sysfs_block_size,
 528         &sysfs_btree_node_size,
 529         &sysfs_btree_cache_size,
 530         &sysfs_btree_avg_write_size,
 531
 532         &sysfs_journal_write_delay_ms,
 533         &sysfs_journal_reclaim_delay_ms,
 534
 535         &sysfs_promote_whole_extents,
 536
 537         &sysfs_compression_stats,
 538
 539 #ifdef CONFIG_BCACHEFS_TESTS
 540         &sysfs_perf_test,
 541 #endif
 542         NULL
 543 };
 544
 545 /* internal dir - just a wrapper */
 546
 547 SHOW(bch2_fs_internal)
 548 {
 549         struct bch_fs *c = container_of(kobj, struct bch_fs, internal);
 550         return bch2_fs_show(&c->kobj, attr, buf);
 551 }
 552
 553 STORE(bch2_fs_internal)
 554 {
 555         struct bch_fs *c = container_of(kobj, struct bch_fs, internal);
 556         return bch2_fs_store(&c->kobj, attr, buf, size);
 557 }
 558 SYSFS_OPS(bch2_fs_internal);
 559
 560 struct attribute *bch2_fs_internal_files[] = {
 561         &sysfs_alloc_debug,
 562         &sysfs_journal_debug,
 563         &sysfs_journal_pins,
 564         &sysfs_btree_updates,
 565         &sysfs_dirty_btree_nodes,
 566         &sysfs_btree_cache,
 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_gc,
 577         &sysfs_gc_gens_pos,
 578         &sysfs_prune_cache,
 579
 580         &sysfs_copy_gc_enabled,
 581         &sysfs_copy_gc_wait,
 582
 583         &sysfs_rebalance_enabled,
 584         &sysfs_rebalance_work,
 585         sysfs_pd_controller_files(rebalance),
 586
 587         &sysfs_new_stripes,
 588
 589         &sysfs_io_timers_read,
 590         &sysfs_io_timers_write,
 591
 592         &sysfs_internal_uuid,
 593         NULL
 594 };
 595
 596 /* options */
 597
 598 SHOW(bch2_fs_opts_dir)
 599 {
 600         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 601         struct bch_fs *c = container_of(kobj, struct bch_fs, opts_dir);
 602         const struct bch_option *opt = container_of(attr, struct bch_option, attr);
 603         int id = opt - bch2_opt_table;
 604         u64 v = bch2_opt_get_by_id(&c->opts, id);
 605
 606         bch2_opt_to_text(&out, c, opt, v, OPT_SHOW_FULL_LIST);
 607         pr_buf(&out, "\n");
 608
 609         return out.pos - buf;
 610 }
 611
 612 STORE(bch2_fs_opts_dir)
 613 {
 614         struct bch_fs *c = container_of(kobj, struct bch_fs, opts_dir);
 615         const struct bch_option *opt = container_of(attr, struct bch_option, attr);
 616         int ret, id = opt - bch2_opt_table;
 617         char *tmp;
 618         u64 v;
 619
 620         tmp = kstrdup(buf, GFP_KERNEL);
 621         if (!tmp)
 622                 return -ENOMEM;
 623
 624         ret = bch2_opt_parse(c, opt, strim(tmp), &v);
 625         kfree(tmp);
 626
 627         if (ret < 0)
 628                 return ret;
 629
 630         ret = bch2_opt_check_may_set(c, id, v);
 631         if (ret < 0)
 632                 return ret;
 633
 634         if (opt->set_sb != SET_NO_SB_OPT) {
 635                 mutex_lock(&c->sb_lock);
 636                 opt->set_sb(c->disk_sb.sb, v);
 637                 bch2_write_super(c);
 638                 mutex_unlock(&c->sb_lock);
 639         }
 640
 641         bch2_opt_set_by_id(&c->opts, id, v);
 642
 643         if ((id == Opt_background_target ||
 644              id == Opt_background_compression) && v) {
 645                 bch2_rebalance_add_work(c, S64_MAX);
 646                 rebalance_wakeup(c);
 647         }
 648
 649         return size;
 650 }
 651 SYSFS_OPS(bch2_fs_opts_dir);
 652
 653 struct attribute *bch2_fs_opts_dir_files[] = { NULL };
 654
 655 int bch2_opts_create_sysfs_files(struct kobject *kobj)
 656 {
 657         const struct bch_option *i;
 658         int ret;
 659
 660         for (i = bch2_opt_table;
 661              i < bch2_opt_table + bch2_opts_nr;
 662              i++) {
 663                 if (!(i->mode & (OPT_FORMAT|OPT_MOUNT|OPT_RUNTIME)))
 664                         continue;
 665
 666                 ret = sysfs_create_file(kobj, &i->attr);
 667                 if (ret)
 668                         return ret;
 669         }
 670
 671         return 0;
 672 }
 673
 674 /* time stats */
 675
 676 SHOW(bch2_fs_time_stats)
 677 {
 678         struct bch_fs *c = container_of(kobj, struct bch_fs, time_stats);
 679         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 680
 681 #define x(name)                                                         \
 682         if (attr == &sysfs_time_stat_##name) {                          \
 683                 bch2_time_stats_to_text(&out, &c->times[BCH_TIME_##name]);\
 684                 return out.pos - buf;                                   \
 685         }
 686         BCH_TIME_STATS()
 687 #undef x
 688
 689         return 0;
 690 }
 691
 692 STORE(bch2_fs_time_stats)
 693 {
 694         return size;
 695 }
 696 SYSFS_OPS(bch2_fs_time_stats);
 697
 698 struct attribute *bch2_fs_time_stats_files[] = {
 699 #define x(name)                                         \
 700         &sysfs_time_stat_##name,
 701         BCH_TIME_STATS()
 702 #undef x
 703         NULL
 704 };
 705
 706 typedef unsigned (bucket_map_fn)(struct bch_fs *, struct bch_dev *,
 707                                  size_t, void *);
 708
 709 static unsigned bucket_last_io_fn(struct bch_fs *c, struct bch_dev *ca,
 710                                   size_t b, void *private)
 711 {
 712         int rw = (private ? 1 : 0);
 713
 714         return atomic64_read(&c->io_clock[rw].now) - bucket(ca, b)->io_time[rw];
 715 }
 716
 717 static unsigned bucket_sectors_used_fn(struct bch_fs *c, struct bch_dev *ca,
 718                                        size_t b, void *private)
 719 {
 720         struct bucket *g = bucket(ca, b);
 721         return bucket_sectors_used(g->mark);
 722 }
 723
 724 static unsigned bucket_oldest_gen_fn(struct bch_fs *c, struct bch_dev *ca,
 725                                      size_t b, void *private)
 726 {
 727         return bucket_gc_gen(bucket(ca, b));
 728 }
 729
 730 static int unsigned_cmp(const void *_l, const void *_r)
 731 {
 732         const unsigned *l = _l;
 733         const unsigned *r = _r;
 734
 735         return cmp_int(*l, *r);
 736 }
 737
 738 static int quantiles_to_text(struct printbuf *out,
 739                              struct bch_fs *c, struct bch_dev *ca,
 740                              bucket_map_fn *fn, void *private)
 741 {
 742         size_t i, n;
 743         /* Compute 31 quantiles */
 744         unsigned q[31], *p;
 745
 746         down_read(&ca->bucket_lock);
 747         n = ca->mi.nbuckets;
 748
 749         p = vzalloc(n * sizeof(unsigned));
 750         if (!p) {
 751                 up_read(&ca->bucket_lock);
 752                 return -ENOMEM;
 753         }
 754
 755         for (i = ca->mi.first_bucket; i < n; i++)
 756                 p[i] = fn(c, ca, i, private);
 757
 758         sort(p, n, sizeof(unsigned), unsigned_cmp, NULL);
 759         up_read(&ca->bucket_lock);
 760
 761         while (n &&
 762                !p[n - 1])
 763                 --n;
 764
 765         for (i = 0; i < ARRAY_SIZE(q); i++)
 766                 q[i] = p[n * (i + 1) / (ARRAY_SIZE(q) + 1)];
 767
 768         vfree(p);
 769
 770         for (i = 0; i < ARRAY_SIZE(q); i++)
 771                 pr_buf(out, "%u ", q[i]);
 772         pr_buf(out, "\n");
 773         return 0;
 774 }
 775
 776 static void reserve_stats_to_text(struct printbuf *out, struct bch_dev *ca)
 777 {
 778         enum alloc_reserve i;
 779
 780         spin_lock(&ca->fs->freelist_lock);
 781
 782         pr_buf(out, "free_inc:\t%zu\t%zu\n",
 783                fifo_used(&ca->free_inc),
 784                ca->free_inc.size);
 785
 786         for (i = 0; i < RESERVE_NR; i++)
 787                 pr_buf(out, "free[%u]:\t%zu\t%zu\n", i,
 788                        fifo_used(&ca->free[i]),
 789                        ca->free[i].size);
 790
 791         spin_unlock(&ca->fs->freelist_lock);
 792 }
 793
 794 static void dev_alloc_debug_to_text(struct printbuf *out, struct bch_dev *ca)
 795 {
 796         struct bch_fs *c = ca->fs;
 797         struct bch_dev_usage stats = bch2_dev_usage_read(ca);
 798         unsigned i, nr[BCH_DATA_NR];
 799
 800         memset(nr, 0, sizeof(nr));
 801
 802         for (i = 0; i < ARRAY_SIZE(c->open_buckets); i++)
 803                 nr[c->open_buckets[i].type]++;
 804
 805         pr_buf(out,
 806                "\t\t buckets\t sectors      fragmented\n"
 807                "capacity%16llu\n",
 808                ca->mi.nbuckets - ca->mi.first_bucket);
 809
 810         for (i = 1; i < BCH_DATA_NR; i++)
 811                 pr_buf(out, "%-8s%16llu%16llu%16llu\n",
 812                        bch2_data_types[i], stats.d[i].buckets,
 813                        stats.d[i].sectors, stats.d[i].fragmented);
 814
 815         pr_buf(out,
 816                "ec\t%16llu\n"
 817                "available%15llu\n"
 818                "\n"
 819                "free_inc\t\t%zu/%zu\n"
 820                "free[RESERVE_MOVINGGC]\t%zu/%zu\n"
 821                "free[RESERVE_NONE]\t%zu/%zu\n"
 822                "freelist_wait\t\t%s\n"
 823                "open buckets allocated\t%u\n"
 824                "open buckets this dev\t%u\n"
 825                "open buckets total\t%u\n"
 826                "open_buckets_wait\t%s\n"
 827                "open_buckets_btree\t%u\n"
 828                "open_buckets_user\t%u\n"
 829                "btree reserve cache\t%u\n"
 830                "thread state:\t\t%s\n",
 831                stats.buckets_ec,
 832                __dev_buckets_available(ca, stats),
 833                fifo_used(&ca->free_inc),                ca->free_inc.size,
 834                fifo_used(&ca->free[RESERVE_MOVINGGC]),  ca->free[RESERVE_MOVINGGC].size,
 835                fifo_used(&ca->free[RESERVE_NONE]),      ca->free[RESERVE_NONE].size,
 836                c->freelist_wait.list.first              ? "waiting" : "empty",
 837                OPEN_BUCKETS_COUNT - c->open_buckets_nr_free,
 838                ca->nr_open_buckets,
 839                OPEN_BUCKETS_COUNT,
 840                c->open_buckets_wait.list.first          ? "waiting" : "empty",
 841                nr[BCH_DATA_btree],
 842                nr[BCH_DATA_user],
 843                c->btree_reserve_cache_nr,
 844                bch2_allocator_states[ca->allocator_state]);
 845 }
 846
 847 static const char * const bch2_rw[] = {
 848         "read",
 849         "write",
 850         NULL
 851 };
 852
 853 static void dev_iodone_to_text(struct printbuf *out, struct bch_dev *ca)
 854 {
 855         int rw, i;
 856
 857         for (rw = 0; rw < 2; rw++) {
 858                 pr_buf(out, "%s:\n", bch2_rw[rw]);
 859
 860                 for (i = 1; i < BCH_DATA_NR; i++)
 861                         pr_buf(out, "%-12s:%12llu\n",
 862                                bch2_data_types[i],
 863                                percpu_u64_get(&ca->io_done->sectors[rw][i]) << 9);
 864         }
 865 }
 866
 867 SHOW(bch2_dev)
 868 {
 869         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
 870         struct bch_fs *c = ca->fs;
 871         struct printbuf out = _PBUF(buf, PAGE_SIZE);
 872
 873         sysfs_printf(uuid,              "%pU\n", ca->uuid.b);
 874
 875         sysfs_print(bucket_size,        bucket_bytes(ca));
 876         sysfs_print(block_size,         block_bytes(c));
 877         sysfs_print(first_bucket,       ca->mi.first_bucket);
 878         sysfs_print(nbuckets,           ca->mi.nbuckets);
 879         sysfs_print(durability,         ca->mi.durability);
 880         sysfs_print(discard,            ca->mi.discard);
 881
 882         if (attr == &sysfs_label) {
 883                 if (ca->mi.group) {
 884                         mutex_lock(&c->sb_lock);
 885                         bch2_disk_path_to_text(&out, &c->disk_sb,
 886                                                ca->mi.group - 1);
 887                         mutex_unlock(&c->sb_lock);
 888                 }
 889
 890                 pr_buf(&out, "\n");
 891                 return out.pos - buf;
 892         }
 893
 894         if (attr == &sysfs_has_data) {
 895                 bch2_flags_to_text(&out, bch2_data_types,
 896                                    bch2_dev_has_data(c, ca));
 897                 pr_buf(&out, "\n");
 898                 return out.pos - buf;
 899         }
 900
 901         if (attr == &sysfs_cache_replacement_policy) {
 902                 bch2_string_opt_to_text(&out,
 903                                         bch2_cache_replacement_policies,
 904                                         ca->mi.replacement);
 905                 pr_buf(&out, "\n");
 906                 return out.pos - buf;
 907         }
 908
 909         if (attr == &sysfs_state_rw) {
 910                 bch2_string_opt_to_text(&out, bch2_member_states,
 911                                         ca->mi.state);
 912                 pr_buf(&out, "\n");
 913                 return out.pos - buf;
 914         }
 915
 916         if (attr == &sysfs_iodone) {
 917                 dev_iodone_to_text(&out, ca);
 918                 return out.pos - buf;
 919         }
 920
 921         sysfs_print(io_latency_read,            atomic64_read(&ca->cur_latency[READ]));
 922         sysfs_print(io_latency_write,           atomic64_read(&ca->cur_latency[WRITE]));
 923
 924         if (attr == &sysfs_io_latency_stats_read) {
 925                 bch2_time_stats_to_text(&out, &ca->io_latency[READ]);
 926                 return out.pos - buf;
 927         }
 928         if (attr == &sysfs_io_latency_stats_write) {
 929                 bch2_time_stats_to_text(&out, &ca->io_latency[WRITE]);
 930                 return out.pos - buf;
 931         }
 932
 933         sysfs_printf(congested,                 "%u%%",
 934                      clamp(atomic_read(&ca->congested), 0, CONGESTED_MAX)
 935                      * 100 / CONGESTED_MAX);
 936
 937         if (attr == &sysfs_bucket_quantiles_last_read)
 938                 return quantiles_to_text(&out, c, ca, bucket_last_io_fn, (void *) 0) ?: out.pos - buf;
 939         if (attr == &sysfs_bucket_quantiles_last_write)
 940                 return quantiles_to_text(&out, c, ca, bucket_last_io_fn, (void *) 1) ?: out.pos - buf;
 941         if (attr == &sysfs_bucket_quantiles_fragmentation)
 942                 return quantiles_to_text(&out, c, ca, bucket_sectors_used_fn, NULL)  ?: out.pos - buf;
 943         if (attr == &sysfs_bucket_quantiles_oldest_gen)
 944                 return quantiles_to_text(&out, c, ca, bucket_oldest_gen_fn, NULL)    ?: out.pos - buf;
 945
 946         if (attr == &sysfs_reserve_stats) {
 947                 reserve_stats_to_text(&out, ca);
 948                 return out.pos - buf;
 949         }
 950         if (attr == &sysfs_alloc_debug) {
 951                 dev_alloc_debug_to_text(&out, ca);
 952                 return out.pos - buf;
 953         }
 954
 955         return 0;
 956 }
 957
 958 STORE(bch2_dev)
 959 {
 960         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
 961         struct bch_fs *c = ca->fs;
 962         struct bch_member *mi;
 963
 964         if (attr == &sysfs_discard) {
 965                 bool v = strtoul_or_return(buf);
 966
 967                 mutex_lock(&c->sb_lock);
 968                 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
 969
 970                 if (v != BCH_MEMBER_DISCARD(mi)) {
 971                         SET_BCH_MEMBER_DISCARD(mi, v);
 972                         bch2_write_super(c);
 973                 }
 974                 mutex_unlock(&c->sb_lock);
 975         }
 976
 977         if (attr == &sysfs_cache_replacement_policy) {
 978                 ssize_t v = __sysfs_match_string(bch2_cache_replacement_policies, -1, buf);
 979
 980                 if (v < 0)
 981                         return v;
 982
 983                 mutex_lock(&c->sb_lock);
 984                 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
 985
 986                 if ((unsigned) v != BCH_MEMBER_REPLACEMENT(mi)) {
 987                         SET_BCH_MEMBER_REPLACEMENT(mi, v);
 988                         bch2_write_super(c);
 989                 }
 990                 mutex_unlock(&c->sb_lock);
 991         }
 992
 993         if (attr == &sysfs_label) {
 994                 char *tmp;
 995                 int ret;
 996
 997                 tmp = kstrdup(buf, GFP_KERNEL);
 998                 if (!tmp)
 999                         return -ENOMEM;
1000
1001                 ret = bch2_dev_group_set(c, ca, strim(tmp));
1002                 kfree(tmp);
1003                 if (ret)
1004                         return ret;
1005         }
1006
1007         if (attr == &sysfs_wake_allocator)
1008                 bch2_wake_allocator(ca);
1009
1010         return size;
1011 }
1012 SYSFS_OPS(bch2_dev);
1013
1014 struct attribute *bch2_dev_files[] = {
1015         &sysfs_uuid,
1016         &sysfs_bucket_size,
1017         &sysfs_block_size,
1018         &sysfs_first_bucket,
1019         &sysfs_nbuckets,
1020         &sysfs_durability,
1021
1022         /* settings: */
1023         &sysfs_discard,
1024         &sysfs_cache_replacement_policy,
1025         &sysfs_state_rw,
1026         &sysfs_label,
1027
1028         &sysfs_has_data,
1029         &sysfs_iodone,
1030
1031         &sysfs_io_latency_read,
1032         &sysfs_io_latency_write,
1033         &sysfs_io_latency_stats_read,
1034         &sysfs_io_latency_stats_write,
1035         &sysfs_congested,
1036
1037         /* alloc info - other stats: */
1038         &sysfs_bucket_quantiles_last_read,
1039         &sysfs_bucket_quantiles_last_write,
1040         &sysfs_bucket_quantiles_fragmentation,
1041         &sysfs_bucket_quantiles_oldest_gen,
1042
1043         &sysfs_reserve_stats,
1044
1045         /* debug: */
1046         &sysfs_alloc_debug,
1047         &sysfs_wake_allocator,
1048         NULL
1049 };
1050
1051 #endif  /* _BCACHEFS_SYSFS_H_ */