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

   1 /*
   2  * bcache sysfs interfaces
   3  *
   4  * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
   5  * Copyright 2012 Google, Inc.
   6  */
   7
   8 #ifndef NO_BCACHEFS_SYSFS
   9
  10 #include "bcachefs.h"
  11 #include "alloc_background.h"
  12 #include "compress.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 "disk_groups.h"
  22 #include "inode.h"
  23 #include "journal.h"
  24 #include "keylist.h"
  25 #include "move.h"
  26 #include "opts.h"
  27 #include "rebalance.h"
  28 #include "replicas.h"
  29 #include "super-io.h"
  30 #include "tests.h"
  31
  32 #include <linux/blkdev.h>
  33 #include <linux/sort.h>
  34 #include <linux/sched/clock.h>
  35
  36 #include "util.h"
  37
  38 #define SYSFS_OPS(type)                                                 \
  39 struct sysfs_ops type ## _sysfs_ops = {                                 \
  40         .show   = type ## _show,                                        \
  41         .store  = type ## _store                                        \
  42 }
  43
  44 #define SHOW(fn)                                                        \
  45 static ssize_t fn ## _show(struct kobject *kobj, struct attribute *attr,\
  46                            char *buf)                                   \
  47
  48 #define STORE(fn)                                                       \
  49 static ssize_t fn ## _store(struct kobject *kobj, struct attribute *attr,\
  50                             const char *buf, size_t size)               \
  51
  52 #define __sysfs_attribute(_name, _mode)                                 \
  53         static struct attribute sysfs_##_name =                         \
  54                 { .name = #_name, .mode = _mode }
  55
  56 #define write_attribute(n)      __sysfs_attribute(n, S_IWUSR)
  57 #define read_attribute(n)       __sysfs_attribute(n, S_IRUGO)
  58 #define rw_attribute(n)         __sysfs_attribute(n, S_IRUGO|S_IWUSR)
  59
  60 #define sysfs_printf(file, fmt, ...)                                    \
  61 do {                                                                    \
  62         if (attr == &sysfs_ ## file)                                    \
  63                 return scnprintf(buf, PAGE_SIZE, fmt "\n", __VA_ARGS__);\
  64 } while (0)
  65
  66 #define sysfs_print(file, var)                                          \
  67 do {                                                                    \
  68         if (attr == &sysfs_ ## file)                                    \
  69                 return snprint(buf, PAGE_SIZE, var);                    \
  70 } while (0)
  71
  72 #define sysfs_hprint(file, val)                                         \
  73 do {                                                                    \
  74         if (attr == &sysfs_ ## file) {                                  \
  75                 ssize_t ret = bch2_hprint(buf, val);                    \
  76                 strcat(buf, "\n");                                      \
  77                 return ret + 1;                                         \
  78         }                                                               \
  79 } while (0)
  80
  81 #define var_printf(_var, fmt)   sysfs_printf(_var, fmt, var(_var))
  82 #define var_print(_var)         sysfs_print(_var, var(_var))
  83 #define var_hprint(_var)        sysfs_hprint(_var, var(_var))
  84
  85 #define sysfs_strtoul(file, var)                                        \
  86 do {                                                                    \
  87         if (attr == &sysfs_ ## file)                                    \
  88                 return strtoul_safe(buf, var) ?: (ssize_t) size;        \
  89 } while (0)
  90
  91 #define sysfs_strtoul_clamp(file, var, min, max)                        \
  92 do {                                                                    \
  93         if (attr == &sysfs_ ## file)                                    \
  94                 return strtoul_safe_clamp(buf, var, min, max)           \
  95                         ?: (ssize_t) size;                              \
  96 } while (0)
  97
  98 #define strtoul_or_return(cp)                                           \
  99 ({                                                                      \
 100         unsigned long _v;                                               \
 101         int _r = kstrtoul(cp, 10, &_v);                                 \
 102         if (_r)                                                         \
 103                 return _r;                                              \
 104         _v;                                                             \
 105 })
 106
 107 #define strtoul_restrict_or_return(cp, min, max)                        \
 108 ({                                                                      \
 109         unsigned long __v = 0;                                          \
 110         int _r = strtoul_safe_restrict(cp, __v, min, max);              \
 111         if (_r)                                                         \
 112                 return _r;                                              \
 113         __v;                                                            \
 114 })
 115
 116 #define strtoi_h_or_return(cp)                                          \
 117 ({                                                                      \
 118         u64 _v;                                                         \
 119         int _r = strtoi_h(cp, &_v);                                     \
 120         if (_r)                                                         \
 121                 return _r;                                              \
 122         _v;                                                             \
 123 })
 124
 125 #define sysfs_hatoi(file, var)                                          \
 126 do {                                                                    \
 127         if (attr == &sysfs_ ## file)                                    \
 128                 return strtoi_h(buf, &var) ?: (ssize_t) size;           \
 129 } while (0)
 130
 131 write_attribute(trigger_journal_flush);
 132 write_attribute(trigger_btree_coalesce);
 133 write_attribute(trigger_gc);
 134 write_attribute(prune_cache);
 135 rw_attribute(btree_gc_periodic);
 136
 137 read_attribute(uuid);
 138 read_attribute(minor);
 139 read_attribute(bucket_size);
 140 read_attribute(block_size);
 141 read_attribute(btree_node_size);
 142 read_attribute(first_bucket);
 143 read_attribute(nbuckets);
 144 read_attribute(durability);
 145 read_attribute(iodone);
 146
 147 read_attribute(io_latency_read);
 148 read_attribute(io_latency_write);
 149 read_attribute(io_latency_stats_read);
 150 read_attribute(io_latency_stats_write);
 151 read_attribute(congested);
 152
 153 read_attribute(bucket_quantiles_last_read);
 154 read_attribute(bucket_quantiles_last_write);
 155 read_attribute(bucket_quantiles_fragmentation);
 156 read_attribute(bucket_quantiles_oldest_gen);
 157
 158 read_attribute(reserve_stats);
 159 read_attribute(btree_cache_size);
 160 read_attribute(compression_stats);
 161 read_attribute(journal_debug);
 162 read_attribute(journal_pins);
 163 read_attribute(btree_updates);
 164 read_attribute(dirty_btree_nodes);
 165
 166 read_attribute(internal_uuid);
 167
 168 read_attribute(has_data);
 169 read_attribute(alloc_debug);
 170 write_attribute(wake_allocator);
 171
 172 read_attribute(read_realloc_races);
 173 read_attribute(extent_migrate_done);
 174 read_attribute(extent_migrate_raced);
 175
 176 rw_attribute(journal_write_delay_ms);
 177 rw_attribute(journal_reclaim_delay_ms);
 178
 179 rw_attribute(discard);
 180 rw_attribute(cache_replacement_policy);
 181 rw_attribute(label);
 182
 183 rw_attribute(copy_gc_enabled);
 184 sysfs_pd_controller_attribute(copy_gc);
 185
 186 rw_attribute(rebalance_enabled);
 187 sysfs_pd_controller_attribute(rebalance);
 188 read_attribute(rebalance_work);
 189 rw_attribute(promote_whole_extents);
 190
 191 rw_attribute(pd_controllers_update_seconds);
 192
 193 read_attribute(meta_replicas_have);
 194 read_attribute(data_replicas_have);
 195
 196 #ifdef CONFIG_BCACHEFS_TESTS
 197 write_attribute(perf_test);
 198 #endif /* CONFIG_BCACHEFS_TESTS */
 199
 200 #define BCH_DEBUG_PARAM(name, description)                              \
 201         rw_attribute(name);
 202
 203         BCH_DEBUG_PARAMS()
 204 #undef BCH_DEBUG_PARAM
 205
 206 #define x(_name)                                                \
 207         static struct attribute sysfs_time_stat_##_name =               \
 208                 { .name = #_name, .mode = S_IRUGO };
 209         BCH_TIME_STATS()
 210 #undef x
 211
 212 static struct attribute sysfs_state_rw = {
 213         .name = "state",
 214         .mode = S_IRUGO
 215 };
 216
 217 static size_t bch2_btree_cache_size(struct bch_fs *c)
 218 {
 219         size_t ret = 0;
 220         struct btree *b;
 221
 222         mutex_lock(&c->btree_cache.lock);
 223         list_for_each_entry(b, &c->btree_cache.live, list)
 224                 ret += btree_bytes(c);
 225
 226         mutex_unlock(&c->btree_cache.lock);
 227         return ret;
 228 }
 229
 230 static ssize_t show_fs_alloc_debug(struct bch_fs *c, char *buf)
 231 {
 232         char *out = buf, *end = buf + PAGE_SIZE;
 233         struct bch_fs_usage stats = bch2_fs_usage_read(c);
 234         unsigned replicas, type;
 235
 236         out += scnprintf(out, end - out,
 237                          "capacity:\t\t%llu\n",
 238                          c->capacity);
 239
 240         for (replicas = 0; replicas < ARRAY_SIZE(stats.replicas); replicas++) {
 241                 out += scnprintf(out, end - out,
 242                                  "%u replicas:\n",
 243                                  replicas + 1);
 244
 245                 for (type = BCH_DATA_SB; type < BCH_DATA_NR; type++)
 246                         out += scnprintf(out, end - out,
 247                                          "\t%s:\t\t%llu\n",
 248                                          bch2_data_types[type],
 249                                          stats.replicas[replicas].data[type]);
 250                 out += scnprintf(out, end - out,
 251                                  "\treserved:\t%llu\n",
 252                                  stats.replicas[replicas].persistent_reserved);
 253         }
 254
 255         out += scnprintf(out, end - out, "bucket usage\n");
 256
 257         for (type = BCH_DATA_SB; type < BCH_DATA_NR; type++)
 258                 out += scnprintf(out, end - out,
 259                                  "\t%s:\t\t%llu\n",
 260                                  bch2_data_types[type],
 261                                  stats.buckets[type]);
 262
 263         out += scnprintf(out, end - out,
 264                          "online reserved:\t%llu\n",
 265                          stats.online_reserved);
 266
 267         return out - buf;
 268 }
 269
 270 static ssize_t bch2_compression_stats(struct bch_fs *c, char *buf)
 271 {
 272         struct btree_iter iter;
 273         struct bkey_s_c k;
 274         u64 nr_uncompressed_extents = 0, uncompressed_sectors = 0,
 275             nr_compressed_extents = 0,
 276             compressed_sectors_compressed = 0,
 277             compressed_sectors_uncompressed = 0;
 278
 279         if (!bch2_fs_running(c))
 280                 return -EPERM;
 281
 282         for_each_btree_key(&iter, c, BTREE_ID_EXTENTS, POS_MIN, 0, k)
 283                 if (k.k->type == BCH_EXTENT) {
 284                         struct bkey_s_c_extent e = bkey_s_c_to_extent(k);
 285                         const struct bch_extent_ptr *ptr;
 286                         struct bch_extent_crc_unpacked crc;
 287
 288                         extent_for_each_ptr_crc(e, ptr, crc) {
 289                                 if (crc.compression_type == BCH_COMPRESSION_NONE) {
 290                                         nr_uncompressed_extents++;
 291                                         uncompressed_sectors += e.k->size;
 292                                 } else {
 293                                         nr_compressed_extents++;
 294                                         compressed_sectors_compressed +=
 295                                                 crc.compressed_size;
 296                                         compressed_sectors_uncompressed +=
 297                                                 crc.uncompressed_size;
 298                                 }
 299
 300                                 /* only looking at the first ptr */
 301                                 break;
 302                         }
 303                 }
 304         bch2_btree_iter_unlock(&iter);
 305
 306         return scnprintf(buf, PAGE_SIZE,
 307                         "uncompressed data:\n"
 308                         "       nr extents:                     %llu\n"
 309                         "       size (bytes):                   %llu\n"
 310                         "compressed data:\n"
 311                         "       nr extents:                     %llu\n"
 312                         "       compressed size (bytes):        %llu\n"
 313                         "       uncompressed size (bytes):      %llu\n",
 314                         nr_uncompressed_extents,
 315                         uncompressed_sectors << 9,
 316                         nr_compressed_extents,
 317                         compressed_sectors_compressed << 9,
 318                         compressed_sectors_uncompressed << 9);
 319 }
 320
 321 SHOW(bch2_fs)
 322 {
 323         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 324
 325         sysfs_print(minor,                      c->minor);
 326         sysfs_printf(internal_uuid, "%pU",      c->sb.uuid.b);
 327
 328         sysfs_print(journal_write_delay_ms,     c->journal.write_delay_ms);
 329         sysfs_print(journal_reclaim_delay_ms,   c->journal.reclaim_delay_ms);
 330
 331         sysfs_print(block_size,                 block_bytes(c));
 332         sysfs_print(btree_node_size,            btree_bytes(c));
 333         sysfs_hprint(btree_cache_size,          bch2_btree_cache_size(c));
 334
 335         sysfs_print(read_realloc_races,
 336                     atomic_long_read(&c->read_realloc_races));
 337         sysfs_print(extent_migrate_done,
 338                     atomic_long_read(&c->extent_migrate_done));
 339         sysfs_print(extent_migrate_raced,
 340                     atomic_long_read(&c->extent_migrate_raced));
 341
 342         sysfs_printf(btree_gc_periodic, "%u",   (int) c->btree_gc_periodic);
 343
 344         sysfs_printf(copy_gc_enabled, "%i", c->copy_gc_enabled);
 345
 346         sysfs_print(pd_controllers_update_seconds,
 347                     c->pd_controllers_update_seconds);
 348
 349         sysfs_printf(rebalance_enabled,         "%i", c->rebalance.enabled);
 350         sysfs_pd_controller_show(rebalance,     &c->rebalance.pd); /* XXX */
 351
 352         if (attr == &sysfs_rebalance_work)
 353                 return bch2_rebalance_work_show(c, buf);
 354
 355         sysfs_print(promote_whole_extents,      c->promote_whole_extents);
 356
 357         sysfs_printf(meta_replicas_have, "%u",  bch2_replicas_online(c, true));
 358         sysfs_printf(data_replicas_have, "%u",  bch2_replicas_online(c, false));
 359
 360         /* Debugging: */
 361
 362         if (attr == &sysfs_alloc_debug)
 363                 return show_fs_alloc_debug(c, buf);
 364
 365         if (attr == &sysfs_journal_debug)
 366                 return bch2_journal_print_debug(&c->journal, buf);
 367
 368         if (attr == &sysfs_journal_pins)
 369                 return bch2_journal_print_pins(&c->journal, buf);
 370
 371         if (attr == &sysfs_btree_updates)
 372                 return bch2_btree_updates_print(c, buf);
 373
 374         if (attr == &sysfs_dirty_btree_nodes)
 375                 return bch2_dirty_btree_nodes_print(c, buf);
 376
 377         if (attr == &sysfs_compression_stats)
 378                 return bch2_compression_stats(c, buf);
 379
 380 #define BCH_DEBUG_PARAM(name, description) sysfs_print(name, c->name);
 381         BCH_DEBUG_PARAMS()
 382 #undef BCH_DEBUG_PARAM
 383
 384         return 0;
 385 }
 386
 387 STORE(__bch2_fs)
 388 {
 389         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 390
 391         sysfs_strtoul(journal_write_delay_ms, c->journal.write_delay_ms);
 392         sysfs_strtoul(journal_reclaim_delay_ms, c->journal.reclaim_delay_ms);
 393
 394         if (attr == &sysfs_btree_gc_periodic) {
 395                 ssize_t ret = strtoul_safe(buf, c->btree_gc_periodic)
 396                         ?: (ssize_t) size;
 397
 398                 wake_up_process(c->gc_thread);
 399                 return ret;
 400         }
 401
 402         if (attr == &sysfs_copy_gc_enabled) {
 403                 struct bch_dev *ca;
 404                 unsigned i;
 405                 ssize_t ret = strtoul_safe(buf, c->copy_gc_enabled)
 406                         ?: (ssize_t) size;
 407
 408                 for_each_member_device(ca, c, i)
 409                         if (ca->copygc_thread)
 410                                 wake_up_process(ca->copygc_thread);
 411                 return ret;
 412         }
 413
 414         if (attr == &sysfs_rebalance_enabled) {
 415                 ssize_t ret = strtoul_safe(buf, c->rebalance.enabled)
 416                         ?: (ssize_t) size;
 417
 418                 rebalance_wakeup(c);
 419                 return ret;
 420         }
 421
 422         sysfs_strtoul(pd_controllers_update_seconds,
 423                       c->pd_controllers_update_seconds);
 424         sysfs_pd_controller_store(rebalance,    &c->rebalance.pd);
 425
 426         sysfs_strtoul(promote_whole_extents,    c->promote_whole_extents);
 427
 428         /* Debugging: */
 429
 430 #define BCH_DEBUG_PARAM(name, description) sysfs_strtoul(name, c->name);
 431         BCH_DEBUG_PARAMS()
 432 #undef BCH_DEBUG_PARAM
 433
 434         if (!bch2_fs_running(c))
 435                 return -EPERM;
 436
 437         /* Debugging: */
 438
 439         if (attr == &sysfs_trigger_journal_flush)
 440                 bch2_journal_meta_async(&c->journal, NULL);
 441
 442         if (attr == &sysfs_trigger_btree_coalesce)
 443                 bch2_coalesce(c);
 444
 445         if (attr == &sysfs_trigger_gc)
 446                 bch2_gc(c);
 447
 448         if (attr == &sysfs_prune_cache) {
 449                 struct shrink_control sc;
 450
 451                 sc.gfp_mask = GFP_KERNEL;
 452                 sc.nr_to_scan = strtoul_or_return(buf);
 453                 c->btree_cache.shrink.scan_objects(&c->btree_cache.shrink, &sc);
 454         }
 455 #ifdef CONFIG_BCACHEFS_TESTS
 456         if (attr == &sysfs_perf_test) {
 457                 char *tmp = kstrdup(buf, GFP_KERNEL), *p = tmp;
 458                 char *test              = strsep(&p, " \t\n");
 459                 char *nr_str            = strsep(&p, " \t\n");
 460                 char *threads_str       = strsep(&p, " \t\n");
 461                 unsigned threads;
 462                 u64 nr;
 463                 int ret = -EINVAL;
 464
 465                 if (threads_str &&
 466                     !(ret = kstrtouint(threads_str, 10, &threads)) &&
 467                     !(ret = bch2_strtoull_h(nr_str, &nr)))
 468                         bch2_btree_perf_test(c, test, nr, threads);
 469                 else
 470                         size = ret;
 471                 kfree(tmp);
 472         }
 473 #endif
 474         return size;
 475 }
 476
 477 STORE(bch2_fs)
 478 {
 479         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 480
 481         mutex_lock(&c->state_lock);
 482         size = __bch2_fs_store(kobj, attr, buf, size);
 483         mutex_unlock(&c->state_lock);
 484
 485         return size;
 486 }
 487 SYSFS_OPS(bch2_fs);
 488
 489 struct attribute *bch2_fs_files[] = {
 490         &sysfs_minor,
 491         &sysfs_block_size,
 492         &sysfs_btree_node_size,
 493         &sysfs_btree_cache_size,
 494
 495         &sysfs_meta_replicas_have,
 496         &sysfs_data_replicas_have,
 497
 498         &sysfs_journal_write_delay_ms,
 499         &sysfs_journal_reclaim_delay_ms,
 500
 501         &sysfs_promote_whole_extents,
 502
 503         &sysfs_compression_stats,
 504
 505 #ifdef CONFIG_BCACHEFS_TESTS
 506         &sysfs_perf_test,
 507 #endif
 508         NULL
 509 };
 510
 511 /* internal dir - just a wrapper */
 512
 513 SHOW(bch2_fs_internal)
 514 {
 515         struct bch_fs *c = container_of(kobj, struct bch_fs, internal);
 516         return bch2_fs_show(&c->kobj, attr, buf);
 517 }
 518
 519 STORE(bch2_fs_internal)
 520 {
 521         struct bch_fs *c = container_of(kobj, struct bch_fs, internal);
 522         return bch2_fs_store(&c->kobj, attr, buf, size);
 523 }
 524 SYSFS_OPS(bch2_fs_internal);
 525
 526 struct attribute *bch2_fs_internal_files[] = {
 527         &sysfs_alloc_debug,
 528         &sysfs_journal_debug,
 529         &sysfs_journal_pins,
 530         &sysfs_btree_updates,
 531         &sysfs_dirty_btree_nodes,
 532
 533         &sysfs_read_realloc_races,
 534         &sysfs_extent_migrate_done,
 535         &sysfs_extent_migrate_raced,
 536
 537         &sysfs_trigger_journal_flush,
 538         &sysfs_trigger_btree_coalesce,
 539         &sysfs_trigger_gc,
 540         &sysfs_prune_cache,
 541
 542         &sysfs_copy_gc_enabled,
 543
 544         &sysfs_rebalance_enabled,
 545         &sysfs_rebalance_work,
 546         sysfs_pd_controller_files(rebalance),
 547
 548         &sysfs_internal_uuid,
 549
 550 #define BCH_DEBUG_PARAM(name, description) &sysfs_##name,
 551         BCH_DEBUG_PARAMS()
 552 #undef BCH_DEBUG_PARAM
 553
 554         NULL
 555 };
 556
 557 /* options */
 558
 559 SHOW(bch2_fs_opts_dir)
 560 {
 561         char *out = buf, *end = buf + PAGE_SIZE;
 562         struct bch_fs *c = container_of(kobj, struct bch_fs, opts_dir);
 563         const struct bch_option *opt = container_of(attr, struct bch_option, attr);
 564         int id = opt - bch2_opt_table;
 565         u64 v = bch2_opt_get_by_id(&c->opts, id);
 566
 567         out += bch2_opt_to_text(c, out, end - out, opt, v, OPT_SHOW_FULL_LIST);
 568         out += scnprintf(out, end - out, "\n");
 569
 570         return out - buf;
 571 }
 572
 573 STORE(bch2_fs_opts_dir)
 574 {
 575         struct bch_fs *c = container_of(kobj, struct bch_fs, opts_dir);
 576         const struct bch_option *opt = container_of(attr, struct bch_option, attr);
 577         int ret, id = opt - bch2_opt_table;
 578         char *tmp;
 579         u64 v;
 580
 581         tmp = kstrdup(buf, GFP_KERNEL);
 582         if (!tmp)
 583                 return -ENOMEM;
 584
 585         ret = bch2_opt_parse(c, opt, strim(tmp), &v);
 586         kfree(tmp);
 587
 588         if (ret < 0)
 589                 return ret;
 590
 591         if (id == Opt_compression ||
 592             id == Opt_background_compression) {
 593                 int ret = bch2_check_set_has_compressed_data(c, v);
 594                 if (ret) {
 595                         mutex_unlock(&c->sb_lock);
 596                         return ret;
 597                 }
 598         }
 599
 600         if (opt->set_sb != SET_NO_SB_OPT) {
 601                 mutex_lock(&c->sb_lock);
 602                 opt->set_sb(c->disk_sb.sb, v);
 603                 bch2_write_super(c);
 604                 mutex_unlock(&c->sb_lock);
 605         }
 606
 607         bch2_opt_set_by_id(&c->opts, id, v);
 608
 609         if ((id == Opt_background_target ||
 610              id == Opt_background_compression) && v) {
 611                 bch2_rebalance_add_work(c, S64_MAX);
 612                 rebalance_wakeup(c);
 613         }
 614
 615         return size;
 616 }
 617 SYSFS_OPS(bch2_fs_opts_dir);
 618
 619 struct attribute *bch2_fs_opts_dir_files[] = { NULL };
 620
 621 int bch2_opts_create_sysfs_files(struct kobject *kobj)
 622 {
 623         const struct bch_option *i;
 624         int ret;
 625
 626         for (i = bch2_opt_table;
 627              i < bch2_opt_table + bch2_opts_nr;
 628              i++) {
 629                 if (i->mode == OPT_INTERNAL)
 630                         continue;
 631
 632                 ret = sysfs_create_file(kobj, &i->attr);
 633                 if (ret)
 634                         return ret;
 635         }
 636
 637         return 0;
 638 }
 639
 640 /* time stats */
 641
 642 SHOW(bch2_fs_time_stats)
 643 {
 644         struct bch_fs *c = container_of(kobj, struct bch_fs, time_stats);
 645
 646 #define x(name)                                         \
 647         if (attr == &sysfs_time_stat_##name)                            \
 648                 return bch2_time_stats_print(&c->times[BCH_TIME_##name],\
 649                                              buf, PAGE_SIZE);
 650         BCH_TIME_STATS()
 651 #undef x
 652
 653         return 0;
 654 }
 655
 656 STORE(bch2_fs_time_stats)
 657 {
 658         return size;
 659 }
 660 SYSFS_OPS(bch2_fs_time_stats);
 661
 662 struct attribute *bch2_fs_time_stats_files[] = {
 663 #define x(name)                                         \
 664         &sysfs_time_stat_##name,
 665         BCH_TIME_STATS()
 666 #undef x
 667         NULL
 668 };
 669
 670 typedef unsigned (bucket_map_fn)(struct bch_fs *, struct bch_dev *,
 671                                  size_t, void *);
 672
 673 static unsigned bucket_last_io_fn(struct bch_fs *c, struct bch_dev *ca,
 674                                   size_t b, void *private)
 675 {
 676         int rw = (private ? 1 : 0);
 677
 678         return bucket_last_io(c, bucket(ca, b), rw);
 679 }
 680
 681 static unsigned bucket_sectors_used_fn(struct bch_fs *c, struct bch_dev *ca,
 682                                        size_t b, void *private)
 683 {
 684         struct bucket *g = bucket(ca, b);
 685         return bucket_sectors_used(g->mark);
 686 }
 687
 688 static unsigned bucket_oldest_gen_fn(struct bch_fs *c, struct bch_dev *ca,
 689                                      size_t b, void *private)
 690 {
 691         return bucket_gc_gen(ca, b);
 692 }
 693
 694 static int unsigned_cmp(const void *_l, const void *_r)
 695 {
 696         unsigned l = *((unsigned *) _l);
 697         unsigned r = *((unsigned *) _r);
 698
 699         return (l > r) - (l < r);
 700 }
 701
 702 static ssize_t show_quantiles(struct bch_fs *c, struct bch_dev *ca,
 703                               char *buf, bucket_map_fn *fn, void *private)
 704 {
 705         size_t i, n;
 706         /* Compute 31 quantiles */
 707         unsigned q[31], *p;
 708         ssize_t ret = 0;
 709
 710         down_read(&ca->bucket_lock);
 711         n = ca->mi.nbuckets;
 712
 713         p = vzalloc(n * sizeof(unsigned));
 714         if (!p) {
 715                 up_read(&ca->bucket_lock);
 716                 return -ENOMEM;
 717         }
 718
 719         for (i = ca->mi.first_bucket; i < n; i++)
 720                 p[i] = fn(c, ca, i, private);
 721
 722         sort(p, n, sizeof(unsigned), unsigned_cmp, NULL);
 723         up_read(&ca->bucket_lock);
 724
 725         while (n &&
 726                !p[n - 1])
 727                 --n;
 728
 729         for (i = 0; i < ARRAY_SIZE(q); i++)
 730                 q[i] = p[n * (i + 1) / (ARRAY_SIZE(q) + 1)];
 731
 732         vfree(p);
 733
 734         for (i = 0; i < ARRAY_SIZE(q); i++)
 735                 ret += scnprintf(buf + ret, PAGE_SIZE - ret,
 736                                  "%u ", q[i]);
 737         buf[ret - 1] = '\n';
 738
 739         return ret;
 740 }
 741
 742 static ssize_t show_reserve_stats(struct bch_dev *ca, char *buf)
 743 {
 744         enum alloc_reserve i;
 745         ssize_t ret;
 746
 747         spin_lock(&ca->freelist_lock);
 748
 749         ret = scnprintf(buf, PAGE_SIZE,
 750                         "free_inc:\t%zu\t%zu\n",
 751                         fifo_used(&ca->free_inc),
 752                         ca->free_inc.size);
 753
 754         for (i = 0; i < RESERVE_NR; i++)
 755                 ret += scnprintf(buf + ret, PAGE_SIZE - ret,
 756                                  "free[%u]:\t%zu\t%zu\n", i,
 757                                  fifo_used(&ca->free[i]),
 758                                  ca->free[i].size);
 759
 760         spin_unlock(&ca->freelist_lock);
 761
 762         return ret;
 763 }
 764
 765 static ssize_t show_dev_alloc_debug(struct bch_dev *ca, char *buf)
 766 {
 767         struct bch_fs *c = ca->fs;
 768         struct bch_dev_usage stats = bch2_dev_usage_read(c, ca);
 769
 770         return scnprintf(buf, PAGE_SIZE,
 771                 "free_inc:               %zu/%zu\n"
 772                 "free[RESERVE_BTREE]:    %zu/%zu\n"
 773                 "free[RESERVE_MOVINGGC]: %zu/%zu\n"
 774                 "free[RESERVE_NONE]:     %zu/%zu\n"
 775                 "buckets:\n"
 776                 "    capacity:           %llu\n"
 777                 "    alloc:              %llu\n"
 778                 "    sb:                 %llu\n"
 779                 "    journal:            %llu\n"
 780                 "    meta:               %llu\n"
 781                 "    user:               %llu\n"
 782                 "    cached:             %llu\n"
 783                 "    available:          %lli\n"
 784                 "sectors:\n"
 785                 "    sb:                 %llu\n"
 786                 "    journal:            %llu\n"
 787                 "    meta:               %llu\n"
 788                 "    user:               %llu\n"
 789                 "    cached:             %llu\n"
 790                 "    fragmented:         %llu\n"
 791                 "    copygc threshold:   %llu\n"
 792                 "freelist_wait:          %s\n"
 793                 "open buckets:           %u/%u (reserved %u)\n"
 794                 "open_buckets_wait:      %s\n",
 795                 fifo_used(&ca->free_inc),               ca->free_inc.size,
 796                 fifo_used(&ca->free[RESERVE_BTREE]),    ca->free[RESERVE_BTREE].size,
 797                 fifo_used(&ca->free[RESERVE_MOVINGGC]), ca->free[RESERVE_MOVINGGC].size,
 798                 fifo_used(&ca->free[RESERVE_NONE]),     ca->free[RESERVE_NONE].size,
 799                 ca->mi.nbuckets - ca->mi.first_bucket,
 800                 stats.buckets_alloc,
 801                 stats.buckets[BCH_DATA_SB],
 802                 stats.buckets[BCH_DATA_JOURNAL],
 803                 stats.buckets[BCH_DATA_BTREE],
 804                 stats.buckets[BCH_DATA_USER],
 805                 stats.buckets[BCH_DATA_CACHED],
 806                 ca->mi.nbuckets - ca->mi.first_bucket - stats.buckets_unavailable,
 807                 stats.sectors[BCH_DATA_SB],
 808                 stats.sectors[BCH_DATA_JOURNAL],
 809                 stats.sectors[BCH_DATA_BTREE],
 810                 stats.sectors[BCH_DATA_USER],
 811                 stats.sectors[BCH_DATA_CACHED],
 812                 stats.sectors_fragmented,
 813                 ca->copygc_threshold,
 814                 c->freelist_wait.list.first             ? "waiting" : "empty",
 815                 c->open_buckets_nr_free, OPEN_BUCKETS_COUNT, BTREE_NODE_RESERVE,
 816                 c->open_buckets_wait.list.first         ? "waiting" : "empty");
 817 }
 818
 819 static const char * const bch2_rw[] = {
 820         "read",
 821         "write",
 822         NULL
 823 };
 824
 825 static ssize_t show_dev_iodone(struct bch_dev *ca, char *buf)
 826 {
 827         char *out = buf, *end = buf + PAGE_SIZE;
 828         int rw, i, cpu;
 829
 830         for (rw = 0; rw < 2; rw++) {
 831                 out += scnprintf(out, end - out, "%s:\n", bch2_rw[rw]);
 832
 833                 for (i = 1; i < BCH_DATA_NR; i++) {
 834                         u64 n = 0;
 835
 836                         for_each_possible_cpu(cpu)
 837                                 n += per_cpu_ptr(ca->io_done, cpu)->sectors[rw][i];
 838
 839                         out += scnprintf(out, end - out, "%-12s:%12llu\n",
 840                                          bch2_data_types[i], n << 9);
 841                 }
 842         }
 843
 844         return out - buf;
 845 }
 846
 847 SHOW(bch2_dev)
 848 {
 849         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
 850         struct bch_fs *c = ca->fs;
 851         char *out = buf, *end = buf + PAGE_SIZE;
 852
 853         sysfs_printf(uuid,              "%pU\n", ca->uuid.b);
 854
 855         sysfs_print(bucket_size,        bucket_bytes(ca));
 856         sysfs_print(block_size,         block_bytes(c));
 857         sysfs_print(first_bucket,       ca->mi.first_bucket);
 858         sysfs_print(nbuckets,           ca->mi.nbuckets);
 859         sysfs_print(durability,         ca->mi.durability);
 860         sysfs_print(discard,            ca->mi.discard);
 861
 862         if (attr == &sysfs_label) {
 863                 if (ca->mi.group) {
 864                         mutex_lock(&c->sb_lock);
 865                         out += bch2_disk_path_print(&c->disk_sb, out, end - out,
 866                                                     ca->mi.group - 1);
 867                         mutex_unlock(&c->sb_lock);
 868                 } else {
 869                         out += scnprintf(out, end - out, "none");
 870                 }
 871
 872                 out += scnprintf(out, end - out, "\n");
 873                 return out - buf;
 874         }
 875
 876         if (attr == &sysfs_has_data) {
 877                 out += bch2_scnprint_flag_list(out, end - out,
 878                                                bch2_data_types,
 879                                                bch2_dev_has_data(c, ca));
 880                 out += scnprintf(out, end - out, "\n");
 881                 return out - buf;
 882         }
 883
 884         sysfs_pd_controller_show(copy_gc, &ca->copygc_pd);
 885
 886         if (attr == &sysfs_cache_replacement_policy) {
 887                 out += bch2_scnprint_string_list(out, end - out,
 888                                                  bch2_cache_replacement_policies,
 889                                                  ca->mi.replacement);
 890                 out += scnprintf(out, end - out, "\n");
 891                 return out - buf;
 892         }
 893
 894         if (attr == &sysfs_state_rw) {
 895                 out += bch2_scnprint_string_list(out, end - out,
 896                                                  bch2_dev_state,
 897                                                  ca->mi.state);
 898                 out += scnprintf(out, end - out, "\n");
 899                 return out - buf;
 900         }
 901
 902         if (attr == &sysfs_iodone)
 903                 return show_dev_iodone(ca, buf);
 904
 905         sysfs_print(io_latency_read,            atomic64_read(&ca->cur_latency[READ]));
 906         sysfs_print(io_latency_write,           atomic64_read(&ca->cur_latency[WRITE]));
 907
 908         if (attr == &sysfs_io_latency_stats_read)
 909                 return bch2_time_stats_print(&ca->io_latency[READ], buf, PAGE_SIZE);
 910         if (attr == &sysfs_io_latency_stats_write)
 911                 return bch2_time_stats_print(&ca->io_latency[WRITE], buf, PAGE_SIZE);
 912
 913         sysfs_printf(congested,                 "%u%%",
 914                      clamp(atomic_read(&ca->congested), 0, CONGESTED_MAX)
 915                      * 100 / CONGESTED_MAX);
 916
 917         if (attr == &sysfs_bucket_quantiles_last_read)
 918                 return show_quantiles(c, ca, buf, bucket_last_io_fn, (void *) 0);
 919         if (attr == &sysfs_bucket_quantiles_last_write)
 920                 return show_quantiles(c, ca, buf, bucket_last_io_fn, (void *) 1);
 921         if (attr == &sysfs_bucket_quantiles_fragmentation)
 922                 return show_quantiles(c, ca, buf, bucket_sectors_used_fn, NULL);
 923         if (attr == &sysfs_bucket_quantiles_oldest_gen)
 924                 return show_quantiles(c, ca, buf, bucket_oldest_gen_fn, NULL);
 925
 926         if (attr == &sysfs_reserve_stats)
 927                 return show_reserve_stats(ca, buf);
 928         if (attr == &sysfs_alloc_debug)
 929                 return show_dev_alloc_debug(ca, buf);
 930
 931         return 0;
 932 }
 933
 934 STORE(bch2_dev)
 935 {
 936         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
 937         struct bch_fs *c = ca->fs;
 938         struct bch_member *mi;
 939
 940         sysfs_pd_controller_store(copy_gc, &ca->copygc_pd);
 941
 942         if (attr == &sysfs_discard) {
 943                 bool v = strtoul_or_return(buf);
 944
 945                 mutex_lock(&c->sb_lock);
 946                 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
 947
 948                 if (v != BCH_MEMBER_DISCARD(mi)) {
 949                         SET_BCH_MEMBER_DISCARD(mi, v);
 950                         bch2_write_super(c);
 951                 }
 952                 mutex_unlock(&c->sb_lock);
 953         }
 954
 955         if (attr == &sysfs_cache_replacement_policy) {
 956                 ssize_t v = __sysfs_match_string(bch2_cache_replacement_policies, -1, buf);
 957
 958                 if (v < 0)
 959                         return v;
 960
 961                 mutex_lock(&c->sb_lock);
 962                 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
 963
 964                 if ((unsigned) v != BCH_MEMBER_REPLACEMENT(mi)) {
 965                         SET_BCH_MEMBER_REPLACEMENT(mi, v);
 966                         bch2_write_super(c);
 967                 }
 968                 mutex_unlock(&c->sb_lock);
 969         }
 970
 971         if (attr == &sysfs_label) {
 972                 char *tmp;
 973                 int ret;
 974
 975                 tmp = kstrdup(buf, GFP_KERNEL);
 976                 if (!tmp)
 977                         return -ENOMEM;
 978
 979                 ret = bch2_dev_group_set(c, ca, strim(tmp));
 980                 kfree(tmp);
 981                 if (ret)
 982                         return ret;
 983         }
 984
 985         if (attr == &sysfs_wake_allocator)
 986                 bch2_wake_allocator(ca);
 987
 988         return size;
 989 }
 990 SYSFS_OPS(bch2_dev);
 991
 992 struct attribute *bch2_dev_files[] = {
 993         &sysfs_uuid,
 994         &sysfs_bucket_size,
 995         &sysfs_block_size,
 996         &sysfs_first_bucket,
 997         &sysfs_nbuckets,
 998         &sysfs_durability,
 999
1000         /* settings: */
1001         &sysfs_discard,
1002         &sysfs_cache_replacement_policy,
1003         &sysfs_state_rw,
1004         &sysfs_label,
1005
1006         &sysfs_has_data,
1007         &sysfs_iodone,
1008
1009         &sysfs_io_latency_read,
1010         &sysfs_io_latency_write,
1011         &sysfs_io_latency_stats_read,
1012         &sysfs_io_latency_stats_write,
1013         &sysfs_congested,
1014
1015         /* alloc info - other stats: */
1016         &sysfs_bucket_quantiles_last_read,
1017         &sysfs_bucket_quantiles_last_write,
1018         &sysfs_bucket_quantiles_fragmentation,
1019         &sysfs_bucket_quantiles_oldest_gen,
1020
1021         &sysfs_reserve_stats,
1022
1023         /* debug: */
1024         &sysfs_alloc_debug,
1025         &sysfs_wake_allocator,
1026
1027         sysfs_pd_controller_files(copy_gc),
1028         NULL
1029 };
1030
1031 #endif  /* _BCACHEFS_SYSFS_H_ */