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

   1
   2 #include "bcachefs.h"
   3 #include "journal.h"
   4 #include "replicas.h"
   5 #include "super-io.h"
   6
   7 static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *,
   8                                             struct bch_replicas_cpu *);
   9
  10 /* Replicas tracking - in memory: */
  11
  12 static inline int u8_cmp(u8 l, u8 r)
  13 {
  14         return (l > r) - (l < r);
  15 }
  16
  17 static void verify_replicas_entry_sorted(struct bch_replicas_entry *e)
  18 {
  19 #ifdef CONFIG_BCACHES_DEBUG
  20         unsigned i;
  21
  22         for (i = 0; i + 1 < e->nr_devs; i++)
  23                 BUG_ON(e->devs[i] >= e->devs[i + 1]);
  24 #endif
  25 }
  26
  27 static void replicas_entry_sort(struct bch_replicas_entry *e)
  28 {
  29         bubble_sort(e->devs, e->nr_devs, u8_cmp);
  30 }
  31
  32 static void bch2_cpu_replicas_sort(struct bch_replicas_cpu *r)
  33 {
  34         eytzinger0_sort(r->entries, r->nr, r->entry_size, memcmp, NULL);
  35 }
  36
  37 void bch2_replicas_entry_to_text(struct printbuf *out,
  38                                  struct bch_replicas_entry *e)
  39 {
  40         unsigned i;
  41
  42         pr_buf(out, "%s: %u/%u [",
  43                bch2_data_types[e->data_type],
  44                e->nr_required,
  45                e->nr_devs);
  46
  47         for (i = 0; i < e->nr_devs; i++)
  48                 pr_buf(out, i ? " %u" : "%u", e->devs[i]);
  49         pr_buf(out, "]");
  50 }
  51
  52 void bch2_cpu_replicas_to_text(struct printbuf *out,
  53                               struct bch_replicas_cpu *r)
  54 {
  55         struct bch_replicas_entry *e;
  56         bool first = true;
  57
  58         for_each_cpu_replicas_entry(r, e) {
  59                 if (!first)
  60                         pr_buf(out, " ");
  61                 first = false;
  62
  63                 bch2_replicas_entry_to_text(out, e);
  64         }
  65 }
  66
  67 static void extent_to_replicas(struct bkey_s_c k,
  68                                struct bch_replicas_entry *r)
  69 {
  70         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
  71         const union bch_extent_entry *entry;
  72         struct extent_ptr_decoded p;
  73
  74         r->nr_required  = 1;
  75
  76         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
  77                 if (p.ptr.cached)
  78                         continue;
  79
  80                 if (p.ec_nr) {
  81                         r->nr_devs = 0;
  82                         break;
  83                 }
  84
  85                 r->devs[r->nr_devs++] = p.ptr.dev;
  86         }
  87 }
  88
  89 static void stripe_to_replicas(struct bkey_s_c k,
  90                                struct bch_replicas_entry *r)
  91 {
  92         struct bkey_s_c_stripe s = bkey_s_c_to_stripe(k);
  93         const struct bch_extent_ptr *ptr;
  94
  95         r->nr_required  = s.v->nr_blocks - s.v->nr_redundant;
  96
  97         for (ptr = s.v->ptrs;
  98              ptr < s.v->ptrs + s.v->nr_blocks;
  99              ptr++)
 100                 r->devs[r->nr_devs++] = ptr->dev;
 101 }
 102
 103 static void bkey_to_replicas(struct bch_replicas_entry *e,
 104                              struct bkey_s_c k)
 105 {
 106         e->nr_devs = 0;
 107
 108         switch (k.k->type) {
 109         case KEY_TYPE_btree_ptr:
 110                 e->data_type = BCH_DATA_BTREE;
 111                 extent_to_replicas(k, e);
 112                 break;
 113         case KEY_TYPE_extent:
 114                 e->data_type = BCH_DATA_USER;
 115                 extent_to_replicas(k, e);
 116                 break;
 117         case KEY_TYPE_stripe:
 118                 e->data_type = BCH_DATA_USER;
 119                 stripe_to_replicas(k, e);
 120                 break;
 121         }
 122
 123         replicas_entry_sort(e);
 124 }
 125
 126 void bch2_devlist_to_replicas(struct bch_replicas_entry *e,
 127                               enum bch_data_type data_type,
 128                               struct bch_devs_list devs)
 129 {
 130         unsigned i;
 131
 132         BUG_ON(!data_type ||
 133                data_type == BCH_DATA_SB ||
 134                data_type >= BCH_DATA_NR);
 135
 136         e->data_type    = data_type;
 137         e->nr_devs      = 0;
 138         e->nr_required  = 1;
 139
 140         for (i = 0; i < devs.nr; i++)
 141                 e->devs[e->nr_devs++] = devs.devs[i];
 142
 143         replicas_entry_sort(e);
 144 }
 145
 146 static struct bch_replicas_cpu
 147 cpu_replicas_add_entry(struct bch_replicas_cpu *old,
 148                        struct bch_replicas_entry *new_entry)
 149 {
 150         unsigned i;
 151         struct bch_replicas_cpu new = {
 152                 .nr             = old->nr + 1,
 153                 .entry_size     = max_t(unsigned, old->entry_size,
 154                                         replicas_entry_bytes(new_entry)),
 155         };
 156
 157         BUG_ON(!new_entry->data_type);
 158         verify_replicas_entry_sorted(new_entry);
 159
 160         new.entries = kcalloc(new.nr, new.entry_size, GFP_NOIO);
 161         if (!new.entries)
 162                 return new;
 163
 164         for (i = 0; i < old->nr; i++)
 165                 memcpy(cpu_replicas_entry(&new, i),
 166                        cpu_replicas_entry(old, i),
 167                        old->entry_size);
 168
 169         memcpy(cpu_replicas_entry(&new, old->nr),
 170                new_entry,
 171                replicas_entry_bytes(new_entry));
 172
 173         bch2_cpu_replicas_sort(&new);
 174         return new;
 175 }
 176
 177 static inline int __replicas_entry_idx(struct bch_replicas_cpu *r,
 178                                        struct bch_replicas_entry *search)
 179 {
 180         int idx, entry_size = replicas_entry_bytes(search);
 181
 182         if (unlikely(entry_size > r->entry_size))
 183                 return -1;
 184
 185         verify_replicas_entry_sorted(search);
 186
 187 #define entry_cmp(_l, _r, size) memcmp(_l, _r, entry_size)
 188         idx = eytzinger0_find(r->entries, r->nr, r->entry_size,
 189                               entry_cmp, search);
 190 #undef entry_cmp
 191
 192         return idx < r->nr ? idx : -1;
 193 }
 194
 195 int bch2_replicas_entry_idx(struct bch_fs *c,
 196                             struct bch_replicas_entry *search)
 197 {
 198         replicas_entry_sort(search);
 199
 200         return __replicas_entry_idx(&c->replicas, search);
 201 }
 202
 203 static bool __replicas_has_entry(struct bch_replicas_cpu *r,
 204                                  struct bch_replicas_entry *search)
 205 {
 206         return __replicas_entry_idx(r, search) >= 0;
 207 }
 208
 209 static bool bch2_replicas_marked_locked(struct bch_fs *c,
 210                           struct bch_replicas_entry *search,
 211                           bool check_gc_replicas)
 212 {
 213         if (!search->nr_devs)
 214                 return true;
 215
 216         verify_replicas_entry_sorted(search);
 217
 218         return __replicas_has_entry(&c->replicas, search) &&
 219                 (!check_gc_replicas ||
 220                  likely((!c->replicas_gc.entries)) ||
 221                  __replicas_has_entry(&c->replicas_gc, search));
 222 }
 223
 224 bool bch2_replicas_marked(struct bch_fs *c,
 225                           struct bch_replicas_entry *search,
 226                           bool check_gc_replicas)
 227 {
 228         bool marked;
 229
 230         percpu_down_read_preempt_disable(&c->mark_lock);
 231         marked = bch2_replicas_marked_locked(c, search, check_gc_replicas);
 232         percpu_up_read_preempt_enable(&c->mark_lock);
 233
 234         return marked;
 235 }
 236
 237 static void __replicas_table_update(struct bch_fs_usage __percpu *dst_p,
 238                                     struct bch_replicas_cpu *dst_r,
 239                                     struct bch_fs_usage __percpu *src_p,
 240                                     struct bch_replicas_cpu *src_r)
 241 {
 242         unsigned src_nr = sizeof(struct bch_fs_usage) / sizeof(u64) + src_r->nr;
 243         struct bch_fs_usage *dst, *src = (void *)
 244                 bch2_acc_percpu_u64s((void *) src_p, src_nr);
 245         int src_idx, dst_idx;
 246
 247         preempt_disable();
 248         dst = this_cpu_ptr(dst_p);
 249         preempt_enable();
 250
 251         *dst = *src;
 252
 253         for (src_idx = 0; src_idx < src_r->nr; src_idx++) {
 254                 if (!src->replicas[src_idx])
 255                         continue;
 256
 257                 dst_idx = __replicas_entry_idx(dst_r,
 258                                 cpu_replicas_entry(src_r, src_idx));
 259                 BUG_ON(dst_idx < 0);
 260
 261                 dst->replicas[dst_idx] = src->replicas[src_idx];
 262         }
 263 }
 264
 265 /*
 266  * Resize filesystem accounting:
 267  */
 268 static int replicas_table_update(struct bch_fs *c,
 269                                  struct bch_replicas_cpu *new_r)
 270 {
 271         struct bch_fs_usage __percpu *new_usage[2] = { NULL, NULL };
 272         struct bch_fs_usage *new_scratch = NULL;
 273         unsigned bytes = sizeof(struct bch_fs_usage) +
 274                 sizeof(u64) * new_r->nr;
 275         int ret = -ENOMEM;
 276
 277         if (!(new_usage[0] = __alloc_percpu_gfp(bytes, sizeof(u64),
 278                                                 GFP_NOIO)) ||
 279             (c->usage[1] &&
 280              !(new_usage[1] = __alloc_percpu_gfp(bytes, sizeof(u64),
 281                                                  GFP_NOIO))) ||
 282             !(new_scratch  = kmalloc(bytes, GFP_NOIO)))
 283                 goto err;
 284
 285         if (c->usage[0])
 286                 __replicas_table_update(new_usage[0],   new_r,
 287                                         c->usage[0],    &c->replicas);
 288         if (c->usage[1])
 289                 __replicas_table_update(new_usage[1],   new_r,
 290                                         c->usage[1],    &c->replicas);
 291
 292         swap(c->usage[0],       new_usage[0]);
 293         swap(c->usage[1],       new_usage[1]);
 294         swap(c->usage_scratch,  new_scratch);
 295         swap(c->replicas,       *new_r);
 296         ret = 0;
 297 err:
 298         kfree(new_scratch);
 299         free_percpu(new_usage[1]);
 300         free_percpu(new_usage[0]);
 301         return ret;
 302 }
 303
 304 static unsigned reserve_journal_replicas(struct bch_fs *c,
 305                                      struct bch_replicas_cpu *r)
 306 {
 307         struct bch_replicas_entry *e;
 308         unsigned journal_res_u64s = 0;
 309
 310         /* nr_inodes: */
 311         journal_res_u64s +=
 312                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 313
 314         /* key_version: */
 315         journal_res_u64s +=
 316                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 317
 318         /* persistent_reserved: */
 319         journal_res_u64s +=
 320                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64)) *
 321                 BCH_REPLICAS_MAX;
 322
 323         for_each_cpu_replicas_entry(r, e)
 324                 journal_res_u64s +=
 325                         DIV_ROUND_UP(sizeof(struct jset_entry_data_usage) +
 326                                      e->nr_devs, sizeof(u64));
 327         return journal_res_u64s;
 328 }
 329
 330 noinline
 331 static int bch2_mark_replicas_slowpath(struct bch_fs *c,
 332                                 struct bch_replicas_entry *new_entry)
 333 {
 334         struct bch_replicas_cpu new_r, new_gc;
 335         int ret = -ENOMEM;
 336
 337         memset(&new_r, 0, sizeof(new_r));
 338         memset(&new_gc, 0, sizeof(new_gc));
 339
 340         mutex_lock(&c->sb_lock);
 341
 342         if (c->replicas_gc.entries &&
 343             !__replicas_has_entry(&c->replicas_gc, new_entry)) {
 344                 new_gc = cpu_replicas_add_entry(&c->replicas_gc, new_entry);
 345                 if (!new_gc.entries)
 346                         goto err;
 347         }
 348
 349         if (!__replicas_has_entry(&c->replicas, new_entry)) {
 350                 new_r = cpu_replicas_add_entry(&c->replicas, new_entry);
 351                 if (!new_r.entries)
 352                         goto err;
 353
 354                 ret = bch2_cpu_replicas_to_sb_replicas(c, &new_r);
 355                 if (ret)
 356                         goto err;
 357
 358                 bch2_journal_entry_res_resize(&c->journal,
 359                                 &c->replicas_journal_res,
 360                                 reserve_journal_replicas(c, &new_r));
 361         }
 362
 363         if (!new_r.entries &&
 364             !new_gc.entries)
 365                 goto out;
 366
 367         /* allocations done, now commit: */
 368
 369         if (new_r.entries)
 370                 bch2_write_super(c);
 371
 372         /* don't update in memory replicas until changes are persistent */
 373         percpu_down_write(&c->mark_lock);
 374         if (new_r.entries)
 375                 ret = replicas_table_update(c, &new_r);
 376         if (new_gc.entries)
 377                 swap(new_gc, c->replicas_gc);
 378         percpu_up_write(&c->mark_lock);
 379 out:
 380         ret = 0;
 381 err:
 382         mutex_unlock(&c->sb_lock);
 383
 384         kfree(new_r.entries);
 385         kfree(new_gc.entries);
 386
 387         return ret;
 388 }
 389
 390 int bch2_mark_replicas(struct bch_fs *c,
 391                        struct bch_replicas_entry *r)
 392 {
 393         return likely(bch2_replicas_marked(c, r, true))
 394                 ? 0
 395                 : bch2_mark_replicas_slowpath(c, r);
 396 }
 397
 398 bool bch2_bkey_replicas_marked_locked(struct bch_fs *c,
 399                                       struct bkey_s_c k,
 400                                       bool check_gc_replicas)
 401 {
 402         struct bch_replicas_padded search;
 403         struct bch_devs_list cached = bch2_bkey_cached_devs(k);
 404         unsigned i;
 405
 406         for (i = 0; i < cached.nr; i++) {
 407                 bch2_replicas_entry_cached(&search.e, cached.devs[i]);
 408
 409                 if (!bch2_replicas_marked_locked(c, &search.e,
 410                                                  check_gc_replicas))
 411                         return false;
 412         }
 413
 414         bkey_to_replicas(&search.e, k);
 415
 416         return bch2_replicas_marked_locked(c, &search.e, check_gc_replicas);
 417 }
 418
 419 bool bch2_bkey_replicas_marked(struct bch_fs *c,
 420                                struct bkey_s_c k,
 421                                bool check_gc_replicas)
 422 {
 423         bool marked;
 424
 425         percpu_down_read_preempt_disable(&c->mark_lock);
 426         marked = bch2_bkey_replicas_marked_locked(c, k, check_gc_replicas);
 427         percpu_up_read_preempt_enable(&c->mark_lock);
 428
 429         return marked;
 430 }
 431
 432 int bch2_mark_bkey_replicas(struct bch_fs *c, struct bkey_s_c k)
 433 {
 434         struct bch_replicas_padded search;
 435         struct bch_devs_list cached = bch2_bkey_cached_devs(k);
 436         unsigned i;
 437         int ret;
 438
 439         for (i = 0; i < cached.nr; i++) {
 440                 bch2_replicas_entry_cached(&search.e, cached.devs[i]);
 441
 442                 ret = bch2_mark_replicas(c, &search.e);
 443                 if (ret)
 444                         return ret;
 445         }
 446
 447         bkey_to_replicas(&search.e, k);
 448
 449         return bch2_mark_replicas(c, &search.e);
 450 }
 451
 452 int bch2_replicas_gc_end(struct bch_fs *c, int ret)
 453 {
 454         unsigned i;
 455
 456         lockdep_assert_held(&c->replicas_gc_lock);
 457
 458         mutex_lock(&c->sb_lock);
 459
 460         if (ret)
 461                 goto err;
 462
 463         /*
 464          * this is kind of crappy; the replicas gc mechanism needs to be ripped
 465          * out
 466          */
 467
 468         for (i = 0; i < c->replicas.nr; i++) {
 469                 struct bch_replicas_entry *e =
 470                         cpu_replicas_entry(&c->replicas, i);
 471                 struct bch_replicas_cpu n;
 472                 u64 v;
 473
 474                 if (__replicas_has_entry(&c->replicas_gc, e))
 475                         continue;
 476
 477                 v = percpu_u64_get(&c->usage[0]->replicas[i]);
 478                 if (!v)
 479                         continue;
 480
 481                 n = cpu_replicas_add_entry(&c->replicas_gc, e);
 482                 if (!n.entries) {
 483                         ret = -ENOSPC;
 484                         goto err;
 485                 }
 486
 487                 percpu_down_write(&c->mark_lock);
 488                 swap(n, c->replicas_gc);
 489                 percpu_up_write(&c->mark_lock);
 490
 491                 kfree(n.entries);
 492         }
 493
 494         if (bch2_cpu_replicas_to_sb_replicas(c, &c->replicas_gc)) {
 495                 ret = -ENOSPC;
 496                 goto err;
 497         }
 498
 499         bch2_write_super(c);
 500
 501         /* don't update in memory replicas until changes are persistent */
 502 err:
 503         percpu_down_write(&c->mark_lock);
 504         if (!ret)
 505                 ret = replicas_table_update(c, &c->replicas_gc);
 506
 507         kfree(c->replicas_gc.entries);
 508         c->replicas_gc.entries = NULL;
 509         percpu_up_write(&c->mark_lock);
 510
 511         mutex_unlock(&c->sb_lock);
 512         return ret;
 513 }
 514
 515 int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask)
 516 {
 517         struct bch_replicas_entry *e;
 518         unsigned i = 0;
 519
 520         lockdep_assert_held(&c->replicas_gc_lock);
 521
 522         mutex_lock(&c->sb_lock);
 523         BUG_ON(c->replicas_gc.entries);
 524
 525         c->replicas_gc.nr               = 0;
 526         c->replicas_gc.entry_size       = 0;
 527
 528         for_each_cpu_replicas_entry(&c->replicas, e)
 529                 if (!((1 << e->data_type) & typemask)) {
 530                         c->replicas_gc.nr++;
 531                         c->replicas_gc.entry_size =
 532                                 max_t(unsigned, c->replicas_gc.entry_size,
 533                                       replicas_entry_bytes(e));
 534                 }
 535
 536         c->replicas_gc.entries = kcalloc(c->replicas_gc.nr,
 537                                          c->replicas_gc.entry_size,
 538                                          GFP_NOIO);
 539         if (!c->replicas_gc.entries) {
 540                 mutex_unlock(&c->sb_lock);
 541                 return -ENOMEM;
 542         }
 543
 544         for_each_cpu_replicas_entry(&c->replicas, e)
 545                 if (!((1 << e->data_type) & typemask))
 546                         memcpy(cpu_replicas_entry(&c->replicas_gc, i++),
 547                                e, c->replicas_gc.entry_size);
 548
 549         bch2_cpu_replicas_sort(&c->replicas_gc);
 550         mutex_unlock(&c->sb_lock);
 551
 552         return 0;
 553 }
 554
 555 int bch2_replicas_set_usage(struct bch_fs *c,
 556                             struct bch_replicas_entry *r,
 557                             u64 sectors)
 558 {
 559         int ret, idx = bch2_replicas_entry_idx(c, r);
 560
 561         if (idx < 0) {
 562                 struct bch_replicas_cpu n;
 563
 564                 n = cpu_replicas_add_entry(&c->replicas, r);
 565                 if (!n.entries)
 566                         return -ENOMEM;
 567
 568                 ret = replicas_table_update(c, &n);
 569                 if (ret)
 570                         return ret;
 571
 572                 kfree(n.entries);
 573
 574                 idx = bch2_replicas_entry_idx(c, r);
 575                 BUG_ON(ret < 0);
 576         }
 577
 578         percpu_u64_set(&c->usage[0]->replicas[idx], sectors);
 579
 580         return 0;
 581 }
 582
 583 /* Replicas tracking - superblock: */
 584
 585 static int
 586 __bch2_sb_replicas_to_cpu_replicas(struct bch_sb_field_replicas *sb_r,
 587                                    struct bch_replicas_cpu *cpu_r)
 588 {
 589         struct bch_replicas_entry *e, *dst;
 590         unsigned nr = 0, entry_size = 0, idx = 0;
 591
 592         for_each_replicas_entry(sb_r, e) {
 593                 entry_size = max_t(unsigned, entry_size,
 594                                    replicas_entry_bytes(e));
 595                 nr++;
 596         }
 597
 598         cpu_r->entries = kcalloc(nr, entry_size, GFP_NOIO);
 599         if (!cpu_r->entries)
 600                 return -ENOMEM;
 601
 602         cpu_r->nr               = nr;
 603         cpu_r->entry_size       = entry_size;
 604
 605         for_each_replicas_entry(sb_r, e) {
 606                 dst = cpu_replicas_entry(cpu_r, idx++);
 607                 memcpy(dst, e, replicas_entry_bytes(e));
 608                 replicas_entry_sort(dst);
 609         }
 610
 611         return 0;
 612 }
 613
 614 static int
 615 __bch2_sb_replicas_v0_to_cpu_replicas(struct bch_sb_field_replicas_v0 *sb_r,
 616                                       struct bch_replicas_cpu *cpu_r)
 617 {
 618         struct bch_replicas_entry_v0 *e;
 619         unsigned nr = 0, entry_size = 0, idx = 0;
 620
 621         for_each_replicas_entry(sb_r, e) {
 622                 entry_size = max_t(unsigned, entry_size,
 623                                    replicas_entry_bytes(e));
 624                 nr++;
 625         }
 626
 627         entry_size += sizeof(struct bch_replicas_entry) -
 628                 sizeof(struct bch_replicas_entry_v0);
 629
 630         cpu_r->entries = kcalloc(nr, entry_size, GFP_NOIO);
 631         if (!cpu_r->entries)
 632                 return -ENOMEM;
 633
 634         cpu_r->nr               = nr;
 635         cpu_r->entry_size       = entry_size;
 636
 637         for_each_replicas_entry(sb_r, e) {
 638                 struct bch_replicas_entry *dst =
 639                         cpu_replicas_entry(cpu_r, idx++);
 640
 641                 dst->data_type  = e->data_type;
 642                 dst->nr_devs    = e->nr_devs;
 643                 dst->nr_required = 1;
 644                 memcpy(dst->devs, e->devs, e->nr_devs);
 645                 replicas_entry_sort(dst);
 646         }
 647
 648         return 0;
 649 }
 650
 651 int bch2_sb_replicas_to_cpu_replicas(struct bch_fs *c)
 652 {
 653         struct bch_sb_field_replicas *sb_v1;
 654         struct bch_sb_field_replicas_v0 *sb_v0;
 655         struct bch_replicas_cpu new_r = { 0, 0, NULL };
 656         int ret = 0;
 657
 658         if ((sb_v1 = bch2_sb_get_replicas(c->disk_sb.sb)))
 659                 ret = __bch2_sb_replicas_to_cpu_replicas(sb_v1, &new_r);
 660         else if ((sb_v0 = bch2_sb_get_replicas_v0(c->disk_sb.sb)))
 661                 ret = __bch2_sb_replicas_v0_to_cpu_replicas(sb_v0, &new_r);
 662
 663         if (ret)
 664                 return -ENOMEM;
 665
 666         bch2_cpu_replicas_sort(&new_r);
 667
 668         percpu_down_write(&c->mark_lock);
 669
 670         ret = replicas_table_update(c, &new_r);
 671         percpu_up_write(&c->mark_lock);
 672
 673         kfree(new_r.entries);
 674
 675         return 0;
 676 }
 677
 678 static int bch2_cpu_replicas_to_sb_replicas_v0(struct bch_fs *c,
 679                                                struct bch_replicas_cpu *r)
 680 {
 681         struct bch_sb_field_replicas_v0 *sb_r;
 682         struct bch_replicas_entry_v0 *dst;
 683         struct bch_replicas_entry *src;
 684         size_t bytes;
 685
 686         bytes = sizeof(struct bch_sb_field_replicas);
 687
 688         for_each_cpu_replicas_entry(r, src)
 689                 bytes += replicas_entry_bytes(src) - 1;
 690
 691         sb_r = bch2_sb_resize_replicas_v0(&c->disk_sb,
 692                         DIV_ROUND_UP(bytes, sizeof(u64)));
 693         if (!sb_r)
 694                 return -ENOSPC;
 695
 696         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas);
 697         sb_r = bch2_sb_get_replicas_v0(c->disk_sb.sb);
 698
 699         memset(&sb_r->entries, 0,
 700                vstruct_end(&sb_r->field) -
 701                (void *) &sb_r->entries);
 702
 703         dst = sb_r->entries;
 704         for_each_cpu_replicas_entry(r, src) {
 705                 dst->data_type  = src->data_type;
 706                 dst->nr_devs    = src->nr_devs;
 707                 memcpy(dst->devs, src->devs, src->nr_devs);
 708
 709                 dst = replicas_entry_next(dst);
 710
 711                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 712         }
 713
 714         return 0;
 715 }
 716
 717 static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *c,
 718                                             struct bch_replicas_cpu *r)
 719 {
 720         struct bch_sb_field_replicas *sb_r;
 721         struct bch_replicas_entry *dst, *src;
 722         bool need_v1 = false;
 723         size_t bytes;
 724
 725         bytes = sizeof(struct bch_sb_field_replicas);
 726
 727         for_each_cpu_replicas_entry(r, src) {
 728                 bytes += replicas_entry_bytes(src);
 729                 if (src->nr_required != 1)
 730                         need_v1 = true;
 731         }
 732
 733         if (!need_v1)
 734                 return bch2_cpu_replicas_to_sb_replicas_v0(c, r);
 735
 736         sb_r = bch2_sb_resize_replicas(&c->disk_sb,
 737                         DIV_ROUND_UP(bytes, sizeof(u64)));
 738         if (!sb_r)
 739                 return -ENOSPC;
 740
 741         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas_v0);
 742         sb_r = bch2_sb_get_replicas(c->disk_sb.sb);
 743
 744         memset(&sb_r->entries, 0,
 745                vstruct_end(&sb_r->field) -
 746                (void *) &sb_r->entries);
 747
 748         dst = sb_r->entries;
 749         for_each_cpu_replicas_entry(r, src) {
 750                 memcpy(dst, src, replicas_entry_bytes(src));
 751
 752                 dst = replicas_entry_next(dst);
 753
 754                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 755         }
 756
 757         return 0;
 758 }
 759
 760 static const char *check_dup_replicas_entries(struct bch_replicas_cpu *cpu_r)
 761 {
 762         unsigned i;
 763
 764         sort_cmp_size(cpu_r->entries,
 765                       cpu_r->nr,
 766                       cpu_r->entry_size,
 767                       memcmp, NULL);
 768
 769         for (i = 0; i + 1 < cpu_r->nr; i++) {
 770                 struct bch_replicas_entry *l =
 771                         cpu_replicas_entry(cpu_r, i);
 772                 struct bch_replicas_entry *r =
 773                         cpu_replicas_entry(cpu_r, i + 1);
 774
 775                 BUG_ON(memcmp(l, r, cpu_r->entry_size) > 0);
 776
 777                 if (!memcmp(l, r, cpu_r->entry_size))
 778                         return "duplicate replicas entry";
 779         }
 780
 781         return NULL;
 782 }
 783
 784 static const char *bch2_sb_validate_replicas(struct bch_sb *sb, struct bch_sb_field *f)
 785 {
 786         struct bch_sb_field_replicas *sb_r = field_to_type(f, replicas);
 787         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 788         struct bch_replicas_cpu cpu_r = { .entries = NULL };
 789         struct bch_replicas_entry *e;
 790         const char *err;
 791         unsigned i;
 792
 793         for_each_replicas_entry(sb_r, e) {
 794                 err = "invalid replicas entry: invalid data type";
 795                 if (e->data_type >= BCH_DATA_NR)
 796                         goto err;
 797
 798                 err = "invalid replicas entry: no devices";
 799                 if (!e->nr_devs)
 800                         goto err;
 801
 802                 err = "invalid replicas entry: bad nr_required";
 803                 if (!e->nr_required ||
 804                     (e->nr_required > 1 &&
 805                      e->nr_required >= e->nr_devs))
 806                         goto err;
 807
 808                 err = "invalid replicas entry: invalid device";
 809                 for (i = 0; i < e->nr_devs; i++)
 810                         if (!bch2_dev_exists(sb, mi, e->devs[i]))
 811                                 goto err;
 812         }
 813
 814         err = "cannot allocate memory";
 815         if (__bch2_sb_replicas_to_cpu_replicas(sb_r, &cpu_r))
 816                 goto err;
 817
 818         err = check_dup_replicas_entries(&cpu_r);
 819 err:
 820         kfree(cpu_r.entries);
 821         return err;
 822 }
 823
 824 static void bch2_sb_replicas_to_text(struct printbuf *out,
 825                                      struct bch_sb *sb,
 826                                      struct bch_sb_field *f)
 827 {
 828         struct bch_sb_field_replicas *r = field_to_type(f, replicas);
 829         struct bch_replicas_entry *e;
 830         bool first = true;
 831
 832         for_each_replicas_entry(r, e) {
 833                 if (!first)
 834                         pr_buf(out, " ");
 835                 first = false;
 836
 837                 bch2_replicas_entry_to_text(out, e);
 838         }
 839 }
 840
 841 const struct bch_sb_field_ops bch_sb_field_ops_replicas = {
 842         .validate       = bch2_sb_validate_replicas,
 843         .to_text        = bch2_sb_replicas_to_text,
 844 };
 845
 846 static const char *bch2_sb_validate_replicas_v0(struct bch_sb *sb, struct bch_sb_field *f)
 847 {
 848         struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0);
 849         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 850         struct bch_replicas_cpu cpu_r = { .entries = NULL };
 851         struct bch_replicas_entry_v0 *e;
 852         const char *err;
 853         unsigned i;
 854
 855         for_each_replicas_entry_v0(sb_r, e) {
 856                 err = "invalid replicas entry: invalid data type";
 857                 if (e->data_type >= BCH_DATA_NR)
 858                         goto err;
 859
 860                 err = "invalid replicas entry: no devices";
 861                 if (!e->nr_devs)
 862                         goto err;
 863
 864                 err = "invalid replicas entry: invalid device";
 865                 for (i = 0; i < e->nr_devs; i++)
 866                         if (!bch2_dev_exists(sb, mi, e->devs[i]))
 867                                 goto err;
 868         }
 869
 870         err = "cannot allocate memory";
 871         if (__bch2_sb_replicas_v0_to_cpu_replicas(sb_r, &cpu_r))
 872                 goto err;
 873
 874         err = check_dup_replicas_entries(&cpu_r);
 875 err:
 876         kfree(cpu_r.entries);
 877         return err;
 878 }
 879
 880 const struct bch_sb_field_ops bch_sb_field_ops_replicas_v0 = {
 881         .validate       = bch2_sb_validate_replicas_v0,
 882 };
 883
 884 /* Query replicas: */
 885
 886 struct replicas_status __bch2_replicas_status(struct bch_fs *c,
 887                                               struct bch_devs_mask online_devs)
 888 {
 889         struct bch_sb_field_members *mi;
 890         struct bch_replicas_entry *e;
 891         unsigned i, nr_online, nr_offline;
 892         struct replicas_status ret;
 893
 894         memset(&ret, 0, sizeof(ret));
 895
 896         for (i = 0; i < ARRAY_SIZE(ret.replicas); i++)
 897                 ret.replicas[i].redundancy = INT_MAX;
 898
 899         mi = bch2_sb_get_members(c->disk_sb.sb);
 900
 901         percpu_down_read_preempt_disable(&c->mark_lock);
 902
 903         for_each_cpu_replicas_entry(&c->replicas, e) {
 904                 if (e->data_type >= ARRAY_SIZE(ret.replicas))
 905                         panic("e %p data_type %u\n", e, e->data_type);
 906
 907                 nr_online = nr_offline = 0;
 908
 909                 for (i = 0; i < e->nr_devs; i++) {
 910                         BUG_ON(!bch2_dev_exists(c->disk_sb.sb, mi,
 911                                                 e->devs[i]));
 912
 913                         if (test_bit(e->devs[i], online_devs.d))
 914                                 nr_online++;
 915                         else
 916                                 nr_offline++;
 917                 }
 918
 919                 ret.replicas[e->data_type].redundancy =
 920                         min(ret.replicas[e->data_type].redundancy,
 921                             (int) nr_online - (int) e->nr_required);
 922
 923                 ret.replicas[e->data_type].nr_offline =
 924                         max(ret.replicas[e->data_type].nr_offline,
 925                             nr_offline);
 926         }
 927
 928         percpu_up_read_preempt_enable(&c->mark_lock);
 929
 930         for (i = 0; i < ARRAY_SIZE(ret.replicas); i++)
 931                 if (ret.replicas[i].redundancy == INT_MAX)
 932                         ret.replicas[i].redundancy = 0;
 933
 934         return ret;
 935 }
 936
 937 struct replicas_status bch2_replicas_status(struct bch_fs *c)
 938 {
 939         return __bch2_replicas_status(c, bch2_online_devs(c));
 940 }
 941
 942 static bool have_enough_devs(struct replicas_status s,
 943                              enum bch_data_type type,
 944                              bool force_if_degraded,
 945                              bool force_if_lost)
 946 {
 947         return (!s.replicas[type].nr_offline || force_if_degraded) &&
 948                 (s.replicas[type].redundancy >= 0 || force_if_lost);
 949 }
 950
 951 bool bch2_have_enough_devs(struct replicas_status s, unsigned flags)
 952 {
 953         return (have_enough_devs(s, BCH_DATA_JOURNAL,
 954                                  flags & BCH_FORCE_IF_METADATA_DEGRADED,
 955                                  flags & BCH_FORCE_IF_METADATA_LOST) &&
 956                 have_enough_devs(s, BCH_DATA_BTREE,
 957                                  flags & BCH_FORCE_IF_METADATA_DEGRADED,
 958                                  flags & BCH_FORCE_IF_METADATA_LOST) &&
 959                 have_enough_devs(s, BCH_DATA_USER,
 960                                  flags & BCH_FORCE_IF_DATA_DEGRADED,
 961                                  flags & BCH_FORCE_IF_DATA_LOST));
 962 }
 963
 964 int bch2_replicas_online(struct bch_fs *c, bool meta)
 965 {
 966         struct replicas_status s = bch2_replicas_status(c);
 967
 968         return (meta
 969                 ? min(s.replicas[BCH_DATA_JOURNAL].redundancy,
 970                       s.replicas[BCH_DATA_BTREE].redundancy)
 971                 : s.replicas[BCH_DATA_USER].redundancy) + 1;
 972 }
 973
 974 unsigned bch2_dev_has_data(struct bch_fs *c, struct bch_dev *ca)
 975 {
 976         struct bch_replicas_entry *e;
 977         unsigned i, ret = 0;
 978
 979         percpu_down_read_preempt_disable(&c->mark_lock);
 980
 981         for_each_cpu_replicas_entry(&c->replicas, e)
 982                 for (i = 0; i < e->nr_devs; i++)
 983                         if (e->devs[i] == ca->dev_idx)
 984                                 ret |= 1 << e->data_type;
 985
 986         percpu_up_read_preempt_enable(&c->mark_lock);
 987
 988         return ret;
 989 }
 990
 991 int bch2_fs_replicas_init(struct bch_fs *c)
 992 {
 993         c->journal.entry_u64s_reserved +=
 994                 reserve_journal_replicas(c, &c->replicas);
 995
 996         return replicas_table_update(c, &c->replicas);
 997 }