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

   1
   2 #include "bcachefs.h"
   3 #include "replicas.h"
   4 #include "super-io.h"
   5
   6 struct bch_replicas_entry_padded {
   7         struct bch_replicas_entry       e;
   8         u8                              pad[BCH_SB_MEMBERS_MAX];
   9 };
  10
  11 static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *,
  12                                             struct bch_replicas_cpu *);
  13
  14 /* Replicas tracking - in memory: */
  15
  16 static inline int u8_cmp(u8 l, u8 r)
  17 {
  18         return (l > r) - (l < r);
  19 }
  20
  21 static void replicas_entry_sort(struct bch_replicas_entry *e)
  22 {
  23         bubble_sort(e->devs, e->nr_devs, u8_cmp);
  24 }
  25
  26 #define for_each_cpu_replicas_entry(_r, _i)                             \
  27         for (_i = (_r)->entries;                                        \
  28              (void *) (_i) < (void *) (_r)->entries + (_r)->nr * (_r)->entry_size;\
  29              _i = (void *) (_i) + (_r)->entry_size)
  30
  31 static inline struct bch_replicas_entry *
  32 cpu_replicas_entry(struct bch_replicas_cpu *r, unsigned i)
  33 {
  34         return (void *) r->entries + r->entry_size * i;
  35 }
  36
  37 static void bch2_cpu_replicas_sort(struct bch_replicas_cpu *r)
  38 {
  39         eytzinger0_sort(r->entries, r->nr, r->entry_size, memcmp, NULL);
  40 }
  41
  42 static void replicas_entry_to_text(struct printbuf *out,
  43                                   struct bch_replicas_entry *e)
  44 {
  45         unsigned i;
  46
  47         pr_buf(out, "%s: %u/%u [",
  48                bch2_data_types[e->data_type],
  49                e->nr_required,
  50                e->nr_devs);
  51
  52         for (i = 0; i < e->nr_devs; i++)
  53                 pr_buf(out, i ? " %u" : "%u", e->devs[i]);
  54         pr_buf(out, "]");
  55 }
  56
  57 void bch2_cpu_replicas_to_text(struct printbuf *out,
  58                               struct bch_replicas_cpu *r)
  59 {
  60         struct bch_replicas_entry *e;
  61         bool first = true;
  62
  63         for_each_cpu_replicas_entry(r, e) {
  64                 if (!first)
  65                         pr_buf(out, " ");
  66                 first = false;
  67
  68                 replicas_entry_to_text(out, e);
  69         }
  70 }
  71
  72 static void extent_to_replicas(struct bkey_s_c k,
  73                                struct bch_replicas_entry *r)
  74 {
  75         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
  76         const union bch_extent_entry *entry;
  77         struct extent_ptr_decoded p;
  78
  79         r->nr_required  = 1;
  80
  81         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
  82                 if (p.ptr.cached)
  83                         continue;
  84
  85                 if (p.ec_nr) {
  86                         r->nr_devs = 0;
  87                         break;
  88                 }
  89
  90                 r->devs[r->nr_devs++] = p.ptr.dev;
  91         }
  92 }
  93
  94 static void stripe_to_replicas(struct bkey_s_c k,
  95                                struct bch_replicas_entry *r)
  96 {
  97         struct bkey_s_c_stripe s = bkey_s_c_to_stripe(k);
  98         const struct bch_extent_ptr *ptr;
  99
 100         r->nr_required  = s.v->nr_blocks - s.v->nr_redundant;
 101
 102         for (ptr = s.v->ptrs;
 103              ptr < s.v->ptrs + s.v->nr_blocks;
 104              ptr++)
 105                 r->devs[r->nr_devs++] = ptr->dev;
 106 }
 107
 108 static void bkey_to_replicas(struct bkey_s_c k,
 109                              struct bch_replicas_entry *e)
 110 {
 111         e->nr_devs = 0;
 112
 113         switch (k.k->type) {
 114         case KEY_TYPE_btree_ptr:
 115                 e->data_type = BCH_DATA_BTREE;
 116                 extent_to_replicas(k, e);
 117                 break;
 118         case KEY_TYPE_extent:
 119                 e->data_type = BCH_DATA_USER;
 120                 extent_to_replicas(k, e);
 121                 break;
 122         case KEY_TYPE_stripe:
 123                 e->data_type = BCH_DATA_USER;
 124                 stripe_to_replicas(k, e);
 125                 break;
 126         }
 127
 128         replicas_entry_sort(e);
 129 }
 130
 131 static inline void devlist_to_replicas(struct bch_devs_list devs,
 132                                        enum bch_data_type data_type,
 133                                        struct bch_replicas_entry *e)
 134 {
 135         unsigned i;
 136
 137         BUG_ON(!data_type ||
 138                data_type == BCH_DATA_SB ||
 139                data_type >= BCH_DATA_NR);
 140
 141         e->data_type    = data_type;
 142         e->nr_devs      = 0;
 143         e->nr_required  = 1;
 144
 145         for (i = 0; i < devs.nr; i++)
 146                 e->devs[e->nr_devs++] = devs.devs[i];
 147
 148         replicas_entry_sort(e);
 149 }
 150
 151 static struct bch_replicas_cpu *
 152 cpu_replicas_add_entry(struct bch_replicas_cpu *old,
 153                        struct bch_replicas_entry *new_entry)
 154 {
 155         struct bch_replicas_cpu *new;
 156         unsigned i, nr, entry_size;
 157
 158         entry_size = max_t(unsigned, old->entry_size,
 159                            replicas_entry_bytes(new_entry));
 160         nr = old->nr + 1;
 161
 162         new = kzalloc(sizeof(struct bch_replicas_cpu) +
 163                       nr * entry_size, GFP_NOIO);
 164         if (!new)
 165                 return NULL;
 166
 167         new->nr         = nr;
 168         new->entry_size = entry_size;
 169
 170         for (i = 0; i < old->nr; i++)
 171                 memcpy(cpu_replicas_entry(new, i),
 172                        cpu_replicas_entry(old, i),
 173                        old->entry_size);
 174
 175         memcpy(cpu_replicas_entry(new, old->nr),
 176                new_entry,
 177                replicas_entry_bytes(new_entry));
 178
 179         bch2_cpu_replicas_sort(new);
 180         return new;
 181 }
 182
 183 static bool __replicas_has_entry(struct bch_replicas_cpu *r,
 184                                  struct bch_replicas_entry *search)
 185 {
 186         return replicas_entry_bytes(search) <= r->entry_size &&
 187                 eytzinger0_find(r->entries, r->nr,
 188                                 r->entry_size,
 189                                 memcmp, search) < r->nr;
 190 }
 191
 192 static bool replicas_has_entry(struct bch_fs *c,
 193                                struct bch_replicas_entry *search,
 194                                bool check_gc_replicas)
 195 {
 196         struct bch_replicas_cpu *r, *gc_r;
 197         bool marked;
 198
 199         rcu_read_lock();
 200         r = rcu_dereference(c->replicas);
 201         marked = __replicas_has_entry(r, search) &&
 202                 (!check_gc_replicas ||
 203                  likely(!(gc_r = rcu_dereference(c->replicas_gc))) ||
 204                  __replicas_has_entry(gc_r, search));
 205         rcu_read_unlock();
 206
 207         return marked;
 208 }
 209
 210 noinline
 211 static int bch2_mark_replicas_slowpath(struct bch_fs *c,
 212                                 struct bch_replicas_entry *new_entry)
 213 {
 214         struct bch_replicas_cpu *old_gc, *new_gc = NULL, *old_r, *new_r = NULL;
 215         int ret = -ENOMEM;
 216
 217         mutex_lock(&c->sb_lock);
 218
 219         old_gc = rcu_dereference_protected(c->replicas_gc,
 220                                            lockdep_is_held(&c->sb_lock));
 221         if (old_gc && !__replicas_has_entry(old_gc, new_entry)) {
 222                 new_gc = cpu_replicas_add_entry(old_gc, new_entry);
 223                 if (!new_gc)
 224                         goto err;
 225         }
 226
 227         old_r = rcu_dereference_protected(c->replicas,
 228                                           lockdep_is_held(&c->sb_lock));
 229         if (!__replicas_has_entry(old_r, new_entry)) {
 230                 new_r = cpu_replicas_add_entry(old_r, new_entry);
 231                 if (!new_r)
 232                         goto err;
 233
 234                 ret = bch2_cpu_replicas_to_sb_replicas(c, new_r);
 235                 if (ret)
 236                         goto err;
 237         }
 238
 239         /* allocations done, now commit: */
 240
 241         if (new_r)
 242                 bch2_write_super(c);
 243
 244         /* don't update in memory replicas until changes are persistent */
 245
 246         if (new_gc) {
 247                 rcu_assign_pointer(c->replicas_gc, new_gc);
 248                 kfree_rcu(old_gc, rcu);
 249         }
 250
 251         if (new_r) {
 252                 rcu_assign_pointer(c->replicas, new_r);
 253                 kfree_rcu(old_r, rcu);
 254         }
 255
 256         mutex_unlock(&c->sb_lock);
 257         return 0;
 258 err:
 259         mutex_unlock(&c->sb_lock);
 260         kfree(new_gc);
 261         kfree(new_r);
 262         return ret;
 263 }
 264
 265 static int __bch2_mark_replicas(struct bch_fs *c,
 266                                 struct bch_replicas_entry *devs)
 267 {
 268         return likely(replicas_has_entry(c, devs, true))
 269                 ? 0
 270                 : bch2_mark_replicas_slowpath(c, devs);
 271 }
 272
 273 int bch2_mark_replicas(struct bch_fs *c,
 274                        enum bch_data_type data_type,
 275                        struct bch_devs_list devs)
 276 {
 277         struct bch_replicas_entry_padded search;
 278
 279         if (!devs.nr)
 280                 return 0;
 281
 282         memset(&search, 0, sizeof(search));
 283
 284         BUG_ON(devs.nr >= BCH_REPLICAS_MAX);
 285
 286         devlist_to_replicas(devs, data_type, &search.e);
 287
 288         return __bch2_mark_replicas(c, &search.e);
 289 }
 290
 291 int bch2_mark_bkey_replicas(struct bch_fs *c, struct bkey_s_c k)
 292 {
 293         struct bch_replicas_entry_padded search;
 294         struct bch_devs_list cached = bch2_bkey_cached_devs(k);
 295         unsigned i;
 296         int ret;
 297
 298         memset(&search, 0, sizeof(search));
 299
 300         for (i = 0; i < cached.nr; i++)
 301                 if ((ret = bch2_mark_replicas(c, BCH_DATA_CACHED,
 302                                               bch2_dev_list_single(cached.devs[i]))))
 303                         return ret;
 304
 305         bkey_to_replicas(k, &search.e);
 306
 307         return search.e.nr_devs
 308                 ? __bch2_mark_replicas(c, &search.e)
 309                 : 0;
 310 }
 311
 312 int bch2_replicas_gc_end(struct bch_fs *c, int ret)
 313 {
 314         struct bch_replicas_cpu *new_r, *old_r;
 315
 316         lockdep_assert_held(&c->replicas_gc_lock);
 317
 318         mutex_lock(&c->sb_lock);
 319
 320         new_r = rcu_dereference_protected(c->replicas_gc,
 321                                           lockdep_is_held(&c->sb_lock));
 322         rcu_assign_pointer(c->replicas_gc, NULL);
 323
 324         if (ret)
 325                 goto err;
 326
 327         if (bch2_cpu_replicas_to_sb_replicas(c, new_r)) {
 328                 ret = -ENOSPC;
 329                 goto err;
 330         }
 331
 332         bch2_write_super(c);
 333
 334         /* don't update in memory replicas until changes are persistent */
 335
 336         old_r = rcu_dereference_protected(c->replicas,
 337                                           lockdep_is_held(&c->sb_lock));
 338
 339         rcu_assign_pointer(c->replicas, new_r);
 340         kfree_rcu(old_r, rcu);
 341 out:
 342         mutex_unlock(&c->sb_lock);
 343         return ret;
 344 err:
 345         kfree_rcu(new_r, rcu);
 346         goto out;
 347 }
 348
 349 int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask)
 350 {
 351         struct bch_replicas_cpu *dst, *src;
 352         struct bch_replicas_entry *e;
 353
 354         lockdep_assert_held(&c->replicas_gc_lock);
 355
 356         mutex_lock(&c->sb_lock);
 357         BUG_ON(c->replicas_gc);
 358
 359         src = rcu_dereference_protected(c->replicas,
 360                                         lockdep_is_held(&c->sb_lock));
 361
 362         dst = kzalloc(sizeof(struct bch_replicas_cpu) +
 363                       src->nr * src->entry_size, GFP_NOIO);
 364         if (!dst) {
 365                 mutex_unlock(&c->sb_lock);
 366                 return -ENOMEM;
 367         }
 368
 369         dst->nr         = 0;
 370         dst->entry_size = src->entry_size;
 371
 372         for_each_cpu_replicas_entry(src, e)
 373                 if (!((1 << e->data_type) & typemask))
 374                         memcpy(cpu_replicas_entry(dst, dst->nr++),
 375                                e, dst->entry_size);
 376
 377         bch2_cpu_replicas_sort(dst);
 378
 379         rcu_assign_pointer(c->replicas_gc, dst);
 380         mutex_unlock(&c->sb_lock);
 381
 382         return 0;
 383 }
 384
 385 /* Replicas tracking - superblock: */
 386
 387 static struct bch_replicas_cpu *
 388 __bch2_sb_replicas_to_cpu_replicas(struct bch_sb_field_replicas *sb_r)
 389 {
 390         struct bch_replicas_entry *e, *dst;
 391         struct bch_replicas_cpu *cpu_r;
 392         unsigned nr = 0, entry_size = 0, idx = 0;
 393
 394         for_each_replicas_entry(sb_r, e) {
 395                 entry_size = max_t(unsigned, entry_size,
 396                                    replicas_entry_bytes(e));
 397                 nr++;
 398         }
 399
 400         cpu_r = kzalloc(sizeof(struct bch_replicas_cpu) +
 401                         nr * entry_size, GFP_NOIO);
 402         if (!cpu_r)
 403                 return NULL;
 404
 405         cpu_r->nr               = nr;
 406         cpu_r->entry_size       = entry_size;
 407
 408         for_each_replicas_entry(sb_r, e) {
 409                 dst = cpu_replicas_entry(cpu_r, idx++);
 410                 memcpy(dst, e, replicas_entry_bytes(e));
 411                 replicas_entry_sort(dst);
 412         }
 413
 414         return cpu_r;
 415 }
 416
 417 static struct bch_replicas_cpu *
 418 __bch2_sb_replicas_v0_to_cpu_replicas(struct bch_sb_field_replicas_v0 *sb_r)
 419 {
 420         struct bch_replicas_entry_v0 *e;
 421         struct bch_replicas_cpu *cpu_r;
 422         unsigned nr = 0, entry_size = 0, idx = 0;
 423
 424         for_each_replicas_entry(sb_r, e) {
 425                 entry_size = max_t(unsigned, entry_size,
 426                                    replicas_entry_bytes(e));
 427                 nr++;
 428         }
 429
 430         entry_size += sizeof(struct bch_replicas_entry) -
 431                 sizeof(struct bch_replicas_entry_v0);
 432
 433         cpu_r = kzalloc(sizeof(struct bch_replicas_cpu) +
 434                         nr * entry_size, GFP_NOIO);
 435         if (!cpu_r)
 436                 return NULL;
 437
 438         cpu_r->nr               = nr;
 439         cpu_r->entry_size       = entry_size;
 440
 441         for_each_replicas_entry(sb_r, e) {
 442                 struct bch_replicas_entry *dst =
 443                         cpu_replicas_entry(cpu_r, idx++);
 444
 445                 dst->data_type  = e->data_type;
 446                 dst->nr_devs    = e->nr_devs;
 447                 dst->nr_required = 1;
 448                 memcpy(dst->devs, e->devs, e->nr_devs);
 449                 replicas_entry_sort(dst);
 450         }
 451
 452         return cpu_r;
 453 }
 454
 455 int bch2_sb_replicas_to_cpu_replicas(struct bch_fs *c)
 456 {
 457         struct bch_sb_field_replicas *sb_v1;
 458         struct bch_sb_field_replicas_v0 *sb_v0;
 459         struct bch_replicas_cpu *cpu_r, *old_r;
 460
 461         if ((sb_v1 = bch2_sb_get_replicas(c->disk_sb.sb)))
 462                 cpu_r = __bch2_sb_replicas_to_cpu_replicas(sb_v1);
 463         else if ((sb_v0 = bch2_sb_get_replicas_v0(c->disk_sb.sb)))
 464                 cpu_r = __bch2_sb_replicas_v0_to_cpu_replicas(sb_v0);
 465         else
 466                 cpu_r = kzalloc(sizeof(struct bch_replicas_cpu), GFP_NOIO);
 467
 468         if (!cpu_r)
 469                 return -ENOMEM;
 470
 471         bch2_cpu_replicas_sort(cpu_r);
 472
 473         old_r = rcu_dereference_check(c->replicas, lockdep_is_held(&c->sb_lock));
 474         rcu_assign_pointer(c->replicas, cpu_r);
 475         if (old_r)
 476                 kfree_rcu(old_r, rcu);
 477
 478         return 0;
 479 }
 480
 481 static int bch2_cpu_replicas_to_sb_replicas_v0(struct bch_fs *c,
 482                                                struct bch_replicas_cpu *r)
 483 {
 484         struct bch_sb_field_replicas_v0 *sb_r;
 485         struct bch_replicas_entry_v0 *dst;
 486         struct bch_replicas_entry *src;
 487         size_t bytes;
 488
 489         bytes = sizeof(struct bch_sb_field_replicas);
 490
 491         for_each_cpu_replicas_entry(r, src)
 492                 bytes += replicas_entry_bytes(src) - 1;
 493
 494         sb_r = bch2_sb_resize_replicas_v0(&c->disk_sb,
 495                         DIV_ROUND_UP(bytes, sizeof(u64)));
 496         if (!sb_r)
 497                 return -ENOSPC;
 498
 499         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas);
 500         sb_r = bch2_sb_get_replicas_v0(c->disk_sb.sb);
 501
 502         memset(&sb_r->entries, 0,
 503                vstruct_end(&sb_r->field) -
 504                (void *) &sb_r->entries);
 505
 506         dst = sb_r->entries;
 507         for_each_cpu_replicas_entry(r, src) {
 508                 dst->data_type  = src->data_type;
 509                 dst->nr_devs    = src->nr_devs;
 510                 memcpy(dst->devs, src->devs, src->nr_devs);
 511
 512                 dst = replicas_entry_next(dst);
 513
 514                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 515         }
 516
 517         return 0;
 518 }
 519
 520 static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *c,
 521                                             struct bch_replicas_cpu *r)
 522 {
 523         struct bch_sb_field_replicas *sb_r;
 524         struct bch_replicas_entry *dst, *src;
 525         bool need_v1 = false;
 526         size_t bytes;
 527
 528         bytes = sizeof(struct bch_sb_field_replicas);
 529
 530         for_each_cpu_replicas_entry(r, src) {
 531                 bytes += replicas_entry_bytes(src);
 532                 if (src->nr_required != 1)
 533                         need_v1 = true;
 534         }
 535
 536         if (!need_v1)
 537                 return bch2_cpu_replicas_to_sb_replicas_v0(c, r);
 538
 539         sb_r = bch2_sb_resize_replicas(&c->disk_sb,
 540                         DIV_ROUND_UP(bytes, sizeof(u64)));
 541         if (!sb_r)
 542                 return -ENOSPC;
 543
 544         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas_v0);
 545         sb_r = bch2_sb_get_replicas(c->disk_sb.sb);
 546
 547         memset(&sb_r->entries, 0,
 548                vstruct_end(&sb_r->field) -
 549                (void *) &sb_r->entries);
 550
 551         dst = sb_r->entries;
 552         for_each_cpu_replicas_entry(r, src) {
 553                 memcpy(dst, src, replicas_entry_bytes(src));
 554
 555                 dst = replicas_entry_next(dst);
 556
 557                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 558         }
 559
 560         return 0;
 561 }
 562
 563 static const char *check_dup_replicas_entries(struct bch_replicas_cpu *cpu_r)
 564 {
 565         unsigned i;
 566
 567         sort_cmp_size(cpu_r->entries,
 568                       cpu_r->nr,
 569                       cpu_r->entry_size,
 570                       memcmp, NULL);
 571
 572         for (i = 0; i + 1 < cpu_r->nr; i++) {
 573                 struct bch_replicas_entry *l =
 574                         cpu_replicas_entry(cpu_r, i);
 575                 struct bch_replicas_entry *r =
 576                         cpu_replicas_entry(cpu_r, i + 1);
 577
 578                 BUG_ON(memcmp(l, r, cpu_r->entry_size) > 0);
 579
 580                 if (!memcmp(l, r, cpu_r->entry_size))
 581                         return "duplicate replicas entry";
 582         }
 583
 584         return NULL;
 585 }
 586
 587 static const char *bch2_sb_validate_replicas(struct bch_sb *sb, struct bch_sb_field *f)
 588 {
 589         struct bch_sb_field_replicas *sb_r = field_to_type(f, replicas);
 590         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 591         struct bch_replicas_cpu *cpu_r = NULL;
 592         struct bch_replicas_entry *e;
 593         const char *err;
 594         unsigned i;
 595
 596         for_each_replicas_entry(sb_r, e) {
 597                 err = "invalid replicas entry: invalid data type";
 598                 if (e->data_type >= BCH_DATA_NR)
 599                         goto err;
 600
 601                 err = "invalid replicas entry: no devices";
 602                 if (!e->nr_devs)
 603                         goto err;
 604
 605                 err = "invalid replicas entry: bad nr_required";
 606                 if (!e->nr_required ||
 607                     (e->nr_required > 1 &&
 608                      e->nr_required >= e->nr_devs))
 609                         goto err;
 610
 611                 err = "invalid replicas entry: invalid device";
 612                 for (i = 0; i < e->nr_devs; i++)
 613                         if (!bch2_dev_exists(sb, mi, e->devs[i]))
 614                                 goto err;
 615         }
 616
 617         err = "cannot allocate memory";
 618         cpu_r = __bch2_sb_replicas_to_cpu_replicas(sb_r);
 619         if (!cpu_r)
 620                 goto err;
 621
 622         err = check_dup_replicas_entries(cpu_r);
 623 err:
 624         kfree(cpu_r);
 625         return err;
 626 }
 627
 628 static void bch2_sb_replicas_to_text(struct printbuf *out,
 629                                      struct bch_sb *sb,
 630                                      struct bch_sb_field *f)
 631 {
 632         struct bch_sb_field_replicas *r = field_to_type(f, replicas);
 633         struct bch_replicas_entry *e;
 634         bool first = true;
 635
 636         for_each_replicas_entry(r, e) {
 637                 if (!first)
 638                         pr_buf(out, " ");
 639                 first = false;
 640
 641                 replicas_entry_to_text(out, e);
 642         }
 643 }
 644
 645 const struct bch_sb_field_ops bch_sb_field_ops_replicas = {
 646         .validate       = bch2_sb_validate_replicas,
 647         .to_text        = bch2_sb_replicas_to_text,
 648 };
 649
 650 static const char *bch2_sb_validate_replicas_v0(struct bch_sb *sb, struct bch_sb_field *f)
 651 {
 652         struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0);
 653         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 654         struct bch_replicas_cpu *cpu_r = NULL;
 655         struct bch_replicas_entry_v0 *e;
 656         const char *err;
 657         unsigned i;
 658
 659         for_each_replicas_entry_v0(sb_r, e) {
 660                 err = "invalid replicas entry: invalid data type";
 661                 if (e->data_type >= BCH_DATA_NR)
 662                         goto err;
 663
 664                 err = "invalid replicas entry: no devices";
 665                 if (!e->nr_devs)
 666                         goto err;
 667
 668                 err = "invalid replicas entry: invalid device";
 669                 for (i = 0; i < e->nr_devs; i++)
 670                         if (!bch2_dev_exists(sb, mi, e->devs[i]))
 671                                 goto err;
 672         }
 673
 674         err = "cannot allocate memory";
 675         cpu_r = __bch2_sb_replicas_v0_to_cpu_replicas(sb_r);
 676         if (!cpu_r)
 677                 goto err;
 678
 679         err = check_dup_replicas_entries(cpu_r);
 680 err:
 681         kfree(cpu_r);
 682         return err;
 683 }
 684
 685 const struct bch_sb_field_ops bch_sb_field_ops_replicas_v0 = {
 686         .validate       = bch2_sb_validate_replicas_v0,
 687 };
 688
 689 /* Query replicas: */
 690
 691 bool bch2_replicas_marked(struct bch_fs *c,
 692                           enum bch_data_type data_type,
 693                           struct bch_devs_list devs,
 694                           bool check_gc_replicas)
 695 {
 696         struct bch_replicas_entry_padded search;
 697
 698         if (!devs.nr)
 699                 return true;
 700
 701         memset(&search, 0, sizeof(search));
 702
 703         devlist_to_replicas(devs, data_type, &search.e);
 704
 705         return replicas_has_entry(c, &search.e, check_gc_replicas);
 706 }
 707
 708 bool bch2_bkey_replicas_marked(struct bch_fs *c,
 709                                struct bkey_s_c k,
 710                                bool check_gc_replicas)
 711 {
 712         struct bch_replicas_entry_padded search;
 713         struct bch_devs_list cached = bch2_bkey_cached_devs(k);
 714         unsigned i;
 715
 716         memset(&search, 0, sizeof(search));
 717
 718         for (i = 0; i < cached.nr; i++)
 719                 if (!bch2_replicas_marked(c, BCH_DATA_CACHED,
 720                                           bch2_dev_list_single(cached.devs[i]),
 721                                           check_gc_replicas))
 722                         return false;
 723
 724         bkey_to_replicas(k, &search.e);
 725
 726         return search.e.nr_devs
 727                 ? replicas_has_entry(c, &search.e, check_gc_replicas)
 728                 : true;
 729 }
 730
 731 struct replicas_status __bch2_replicas_status(struct bch_fs *c,
 732                                               struct bch_devs_mask online_devs)
 733 {
 734         struct bch_sb_field_members *mi;
 735         struct bch_replicas_entry *e;
 736         struct bch_replicas_cpu *r;
 737         unsigned i, nr_online, nr_offline;
 738         struct replicas_status ret;
 739
 740         memset(&ret, 0, sizeof(ret));
 741
 742         for (i = 0; i < ARRAY_SIZE(ret.replicas); i++)
 743                 ret.replicas[i].redundancy = INT_MAX;
 744
 745         mi = bch2_sb_get_members(c->disk_sb.sb);
 746         rcu_read_lock();
 747         r = rcu_dereference(c->replicas);
 748
 749         for_each_cpu_replicas_entry(r, e) {
 750                 if (e->data_type >= ARRAY_SIZE(ret.replicas))
 751                         panic("e %p data_type %u\n", e, e->data_type);
 752
 753                 nr_online = nr_offline = 0;
 754
 755                 for (i = 0; i < e->nr_devs; i++) {
 756                         BUG_ON(!bch2_dev_exists(c->disk_sb.sb, mi,
 757                                                 e->devs[i]));
 758
 759                         if (test_bit(e->devs[i], online_devs.d))
 760                                 nr_online++;
 761                         else
 762                                 nr_offline++;
 763                 }
 764
 765                 ret.replicas[e->data_type].redundancy =
 766                         min(ret.replicas[e->data_type].redundancy,
 767                             (int) nr_online - (int) e->nr_required);
 768
 769                 ret.replicas[e->data_type].nr_offline =
 770                         max(ret.replicas[e->data_type].nr_offline,
 771                             nr_offline);
 772         }
 773
 774         rcu_read_unlock();
 775
 776         for (i = 0; i < ARRAY_SIZE(ret.replicas); i++)
 777                 if (ret.replicas[i].redundancy == INT_MAX)
 778                         ret.replicas[i].redundancy = 0;
 779
 780         return ret;
 781 }
 782
 783 struct replicas_status bch2_replicas_status(struct bch_fs *c)
 784 {
 785         return __bch2_replicas_status(c, bch2_online_devs(c));
 786 }
 787
 788 static bool have_enough_devs(struct replicas_status s,
 789                              enum bch_data_type type,
 790                              bool force_if_degraded,
 791                              bool force_if_lost)
 792 {
 793         return (!s.replicas[type].nr_offline || force_if_degraded) &&
 794                 (s.replicas[type].redundancy >= 0 || force_if_lost);
 795 }
 796
 797 bool bch2_have_enough_devs(struct replicas_status s, unsigned flags)
 798 {
 799         return (have_enough_devs(s, BCH_DATA_JOURNAL,
 800                                  flags & BCH_FORCE_IF_METADATA_DEGRADED,
 801                                  flags & BCH_FORCE_IF_METADATA_LOST) &&
 802                 have_enough_devs(s, BCH_DATA_BTREE,
 803                                  flags & BCH_FORCE_IF_METADATA_DEGRADED,
 804                                  flags & BCH_FORCE_IF_METADATA_LOST) &&
 805                 have_enough_devs(s, BCH_DATA_USER,
 806                                  flags & BCH_FORCE_IF_DATA_DEGRADED,
 807                                  flags & BCH_FORCE_IF_DATA_LOST));
 808 }
 809
 810 int bch2_replicas_online(struct bch_fs *c, bool meta)
 811 {
 812         struct replicas_status s = bch2_replicas_status(c);
 813
 814         return (meta
 815                 ? min(s.replicas[BCH_DATA_JOURNAL].redundancy,
 816                       s.replicas[BCH_DATA_BTREE].redundancy)
 817                 : s.replicas[BCH_DATA_USER].redundancy) + 1;
 818 }
 819
 820 unsigned bch2_dev_has_data(struct bch_fs *c, struct bch_dev *ca)
 821 {
 822         struct bch_replicas_entry *e;
 823         struct bch_replicas_cpu *r;
 824         unsigned i, ret = 0;
 825
 826         rcu_read_lock();
 827         r = rcu_dereference(c->replicas);
 828
 829         for_each_cpu_replicas_entry(r, e)
 830                 for (i = 0; i < e->nr_devs; i++)
 831                         if (e->devs[i] == ca->dev_idx)
 832                                 ret |= 1 << e->data_type;
 833
 834         rcu_read_unlock();
 835
 836         return ret;
 837 }