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 bool bch2_replicas_marked(struct bch_fs *c,
 210                           struct bch_replicas_entry *search,
 211                           bool check_gc_replicas)
 212 {
 213         bool marked;
 214
 215         if (!search->nr_devs)
 216                 return true;
 217
 218         verify_replicas_entry_sorted(search);
 219
 220         percpu_down_read_preempt_disable(&c->mark_lock);
 221         marked = __replicas_has_entry(&c->replicas, search) &&
 222                 (!check_gc_replicas ||
 223                  likely((!c->replicas_gc.entries)) ||
 224                  __replicas_has_entry(&c->replicas_gc, search));
 225         percpu_up_read_preempt_enable(&c->mark_lock);
 226
 227         return marked;
 228 }
 229
 230 static void __replicas_table_update(struct bch_fs_usage __percpu *dst_p,
 231                                     struct bch_replicas_cpu *dst_r,
 232                                     struct bch_fs_usage __percpu *src_p,
 233                                     struct bch_replicas_cpu *src_r)
 234 {
 235         unsigned src_nr = sizeof(struct bch_fs_usage) / sizeof(u64) + src_r->nr;
 236         struct bch_fs_usage *dst, *src = (void *)
 237                 bch2_acc_percpu_u64s((void *) src_p, src_nr);
 238         int src_idx, dst_idx;
 239
 240         preempt_disable();
 241         dst = this_cpu_ptr(dst_p);
 242         preempt_enable();
 243
 244         *dst = *src;
 245
 246         for (src_idx = 0; src_idx < src_r->nr; src_idx++) {
 247                 if (!src->data[src_idx])
 248                         continue;
 249
 250                 dst_idx = __replicas_entry_idx(dst_r,
 251                                 cpu_replicas_entry(src_r, src_idx));
 252                 BUG_ON(dst_idx < 0);
 253
 254                 dst->data[dst_idx] = src->data[src_idx];
 255         }
 256 }
 257
 258 /*
 259  * Resize filesystem accounting:
 260  */
 261 static int replicas_table_update(struct bch_fs *c,
 262                                  struct bch_replicas_cpu *new_r)
 263 {
 264         struct bch_fs_usage __percpu *new_usage[3] = { NULL, NULL, NULL };
 265         unsigned bytes = sizeof(struct bch_fs_usage) +
 266                 sizeof(u64) * new_r->nr;
 267         unsigned i;
 268         int ret = -ENOMEM;
 269
 270         for (i = 0; i < 3; i++) {
 271                 if (i < 2 && !c->usage[i])
 272                         continue;
 273
 274                 new_usage[i] = __alloc_percpu_gfp(bytes, sizeof(u64),
 275                                                   GFP_NOIO);
 276                 if (!new_usage[i])
 277                         goto err;
 278         }
 279
 280         for (i = 0; i < 2; i++) {
 281                 if (!c->usage[i])
 282                         continue;
 283
 284                 __replicas_table_update(new_usage[i],   new_r,
 285                                         c->usage[i],    &c->replicas);
 286
 287                 swap(c->usage[i], new_usage[i]);
 288         }
 289
 290         swap(c->usage_scratch, new_usage[2]);
 291
 292         swap(c->replicas, *new_r);
 293         ret = 0;
 294 err:
 295         for (i = 0; i < 3; i++)
 296                 free_percpu(new_usage[i]);
 297         return ret;
 298 }
 299
 300 static unsigned reserve_journal_replicas(struct bch_fs *c,
 301                                      struct bch_replicas_cpu *r)
 302 {
 303         struct bch_replicas_entry *e;
 304         unsigned journal_res_u64s = 0;
 305
 306         /* nr_inodes: */
 307         journal_res_u64s +=
 308                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 309
 310         /* key_version: */
 311         journal_res_u64s +=
 312                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 313
 314         /* persistent_reserved: */
 315         journal_res_u64s +=
 316                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64)) *
 317                 BCH_REPLICAS_MAX;
 318
 319         for_each_cpu_replicas_entry(r, e)
 320                 journal_res_u64s +=
 321                         DIV_ROUND_UP(sizeof(struct jset_entry_data_usage) +
 322                                      e->nr_devs, sizeof(u64));
 323         return journal_res_u64s;
 324 }
 325
 326 noinline
 327 static int bch2_mark_replicas_slowpath(struct bch_fs *c,
 328                                 struct bch_replicas_entry *new_entry)
 329 {
 330         struct bch_replicas_cpu new_r, new_gc;
 331         int ret = -ENOMEM;
 332
 333         memset(&new_r, 0, sizeof(new_r));
 334         memset(&new_gc, 0, sizeof(new_gc));
 335
 336         mutex_lock(&c->sb_lock);
 337
 338         if (c->replicas_gc.entries &&
 339             !__replicas_has_entry(&c->replicas_gc, new_entry)) {
 340                 new_gc = cpu_replicas_add_entry(&c->replicas_gc, new_entry);
 341                 if (!new_gc.entries)
 342                         goto err;
 343         }
 344
 345         if (!__replicas_has_entry(&c->replicas, new_entry)) {
 346                 new_r = cpu_replicas_add_entry(&c->replicas, new_entry);
 347                 if (!new_r.entries)
 348                         goto err;
 349
 350                 ret = bch2_cpu_replicas_to_sb_replicas(c, &new_r);
 351                 if (ret)
 352                         goto err;
 353
 354                 bch2_journal_entry_res_resize(&c->journal,
 355                                 &c->replicas_journal_res,
 356                                 reserve_journal_replicas(c, &new_r));
 357         }
 358
 359         if (!new_r.entries &&
 360             !new_gc.entries)
 361                 goto out;
 362
 363         /* allocations done, now commit: */
 364
 365         if (new_r.entries)
 366                 bch2_write_super(c);
 367
 368         /* don't update in memory replicas until changes are persistent */
 369         percpu_down_write(&c->mark_lock);
 370         if (new_r.entries)
 371                 ret = replicas_table_update(c, &new_r);
 372         if (new_gc.entries)
 373                 swap(new_gc, c->replicas_gc);
 374         percpu_up_write(&c->mark_lock);
 375 out:
 376         ret = 0;
 377 err:
 378         mutex_unlock(&c->sb_lock);
 379
 380         kfree(new_r.entries);
 381         kfree(new_gc.entries);
 382
 383         return ret;
 384 }
 385
 386 int bch2_mark_replicas(struct bch_fs *c,
 387                        struct bch_replicas_entry *r)
 388 {
 389         return likely(bch2_replicas_marked(c, r, true))
 390                 ? 0
 391                 : bch2_mark_replicas_slowpath(c, r);
 392 }
 393
 394 bool bch2_bkey_replicas_marked(struct bch_fs *c,
 395                                struct bkey_s_c k,
 396                                bool check_gc_replicas)
 397 {
 398         struct bch_replicas_padded search;
 399         struct bch_devs_list cached = bch2_bkey_cached_devs(k);
 400         unsigned i;
 401
 402         for (i = 0; i < cached.nr; i++) {
 403                 bch2_replicas_entry_cached(&search.e, cached.devs[i]);
 404
 405                 if (!bch2_replicas_marked(c, &search.e, check_gc_replicas))
 406                         return false;
 407         }
 408
 409         bkey_to_replicas(&search.e, k);
 410
 411         return bch2_replicas_marked(c, &search.e, check_gc_replicas);
 412 }
 413
 414 int bch2_mark_bkey_replicas(struct bch_fs *c, struct bkey_s_c k)
 415 {
 416         struct bch_replicas_padded search;
 417         struct bch_devs_list cached = bch2_bkey_cached_devs(k);
 418         unsigned i;
 419         int ret;
 420
 421         for (i = 0; i < cached.nr; i++) {
 422                 bch2_replicas_entry_cached(&search.e, cached.devs[i]);
 423
 424                 ret = bch2_mark_replicas(c, &search.e);
 425                 if (ret)
 426                         return ret;
 427         }
 428
 429         bkey_to_replicas(&search.e, k);
 430
 431         return bch2_mark_replicas(c, &search.e);
 432 }
 433
 434 int bch2_replicas_gc_end(struct bch_fs *c, int ret)
 435 {
 436         unsigned i;
 437
 438         lockdep_assert_held(&c->replicas_gc_lock);
 439
 440         mutex_lock(&c->sb_lock);
 441
 442         if (ret)
 443                 goto err;
 444
 445         /*
 446          * this is kind of crappy; the replicas gc mechanism needs to be ripped
 447          * out
 448          */
 449
 450         for (i = 0; i < c->replicas.nr; i++) {
 451                 struct bch_replicas_entry *e =
 452                         cpu_replicas_entry(&c->replicas, i);
 453                 struct bch_replicas_cpu n;
 454                 u64 v;
 455
 456                 if (__replicas_has_entry(&c->replicas_gc, e))
 457                         continue;
 458
 459                 v = percpu_u64_get(&c->usage[0]->data[i]);
 460                 if (!v)
 461                         continue;
 462
 463                 n = cpu_replicas_add_entry(&c->replicas_gc, e);
 464                 if (!n.entries) {
 465                         ret = -ENOSPC;
 466                         goto err;
 467                 }
 468
 469                 percpu_down_write(&c->mark_lock);
 470                 swap(n, c->replicas_gc);
 471                 percpu_up_write(&c->mark_lock);
 472
 473                 kfree(n.entries);
 474         }
 475
 476         if (bch2_cpu_replicas_to_sb_replicas(c, &c->replicas_gc)) {
 477                 ret = -ENOSPC;
 478                 goto err;
 479         }
 480
 481         bch2_write_super(c);
 482
 483         /* don't update in memory replicas until changes are persistent */
 484 err:
 485         percpu_down_write(&c->mark_lock);
 486         if (!ret)
 487                 ret = replicas_table_update(c, &c->replicas_gc);
 488
 489         kfree(c->replicas_gc.entries);
 490         c->replicas_gc.entries = NULL;
 491         percpu_up_write(&c->mark_lock);
 492
 493         mutex_unlock(&c->sb_lock);
 494         return ret;
 495 }
 496
 497 int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask)
 498 {
 499         struct bch_replicas_entry *e;
 500         unsigned i = 0;
 501
 502         lockdep_assert_held(&c->replicas_gc_lock);
 503
 504         mutex_lock(&c->sb_lock);
 505         BUG_ON(c->replicas_gc.entries);
 506
 507         c->replicas_gc.nr               = 0;
 508         c->replicas_gc.entry_size       = 0;
 509
 510         for_each_cpu_replicas_entry(&c->replicas, e)
 511                 if (!((1 << e->data_type) & typemask)) {
 512                         c->replicas_gc.nr++;
 513                         c->replicas_gc.entry_size =
 514                                 max_t(unsigned, c->replicas_gc.entry_size,
 515                                       replicas_entry_bytes(e));
 516                 }
 517
 518         c->replicas_gc.entries = kcalloc(c->replicas_gc.nr,
 519                                          c->replicas_gc.entry_size,
 520                                          GFP_NOIO);
 521         if (!c->replicas_gc.entries) {
 522                 mutex_unlock(&c->sb_lock);
 523                 return -ENOMEM;
 524         }
 525
 526         for_each_cpu_replicas_entry(&c->replicas, e)
 527                 if (!((1 << e->data_type) & typemask))
 528                         memcpy(cpu_replicas_entry(&c->replicas_gc, i++),
 529                                e, c->replicas_gc.entry_size);
 530
 531         bch2_cpu_replicas_sort(&c->replicas_gc);
 532         mutex_unlock(&c->sb_lock);
 533
 534         return 0;
 535 }
 536
 537 int bch2_replicas_set_usage(struct bch_fs *c,
 538                             struct bch_replicas_entry *r,
 539                             u64 sectors)
 540 {
 541         int ret, idx = bch2_replicas_entry_idx(c, r);
 542
 543         if (idx < 0) {
 544                 struct bch_replicas_cpu n;
 545
 546                 n = cpu_replicas_add_entry(&c->replicas, r);
 547                 if (!n.entries)
 548                         return -ENOMEM;
 549
 550                 ret = replicas_table_update(c, &n);
 551                 if (ret)
 552                         return ret;
 553
 554                 kfree(n.entries);
 555
 556                 idx = bch2_replicas_entry_idx(c, r);
 557                 BUG_ON(ret < 0);
 558         }
 559
 560         percpu_u64_set(&c->usage[0]->data[idx], sectors);
 561
 562         return 0;
 563 }
 564
 565 /* Replicas tracking - superblock: */
 566
 567 static int
 568 __bch2_sb_replicas_to_cpu_replicas(struct bch_sb_field_replicas *sb_r,
 569                                    struct bch_replicas_cpu *cpu_r)
 570 {
 571         struct bch_replicas_entry *e, *dst;
 572         unsigned nr = 0, entry_size = 0, idx = 0;
 573
 574         for_each_replicas_entry(sb_r, e) {
 575                 entry_size = max_t(unsigned, entry_size,
 576                                    replicas_entry_bytes(e));
 577                 nr++;
 578         }
 579
 580         cpu_r->entries = kcalloc(nr, entry_size, GFP_NOIO);
 581         if (!cpu_r->entries)
 582                 return -ENOMEM;
 583
 584         cpu_r->nr               = nr;
 585         cpu_r->entry_size       = entry_size;
 586
 587         for_each_replicas_entry(sb_r, e) {
 588                 dst = cpu_replicas_entry(cpu_r, idx++);
 589                 memcpy(dst, e, replicas_entry_bytes(e));
 590                 replicas_entry_sort(dst);
 591         }
 592
 593         return 0;
 594 }
 595
 596 static int
 597 __bch2_sb_replicas_v0_to_cpu_replicas(struct bch_sb_field_replicas_v0 *sb_r,
 598                                       struct bch_replicas_cpu *cpu_r)
 599 {
 600         struct bch_replicas_entry_v0 *e;
 601         unsigned nr = 0, entry_size = 0, idx = 0;
 602
 603         for_each_replicas_entry(sb_r, e) {
 604                 entry_size = max_t(unsigned, entry_size,
 605                                    replicas_entry_bytes(e));
 606                 nr++;
 607         }
 608
 609         entry_size += sizeof(struct bch_replicas_entry) -
 610                 sizeof(struct bch_replicas_entry_v0);
 611
 612         cpu_r->entries = kcalloc(nr, entry_size, GFP_NOIO);
 613         if (!cpu_r->entries)
 614                 return -ENOMEM;
 615
 616         cpu_r->nr               = nr;
 617         cpu_r->entry_size       = entry_size;
 618
 619         for_each_replicas_entry(sb_r, e) {
 620                 struct bch_replicas_entry *dst =
 621                         cpu_replicas_entry(cpu_r, idx++);
 622
 623                 dst->data_type  = e->data_type;
 624                 dst->nr_devs    = e->nr_devs;
 625                 dst->nr_required = 1;
 626                 memcpy(dst->devs, e->devs, e->nr_devs);
 627                 replicas_entry_sort(dst);
 628         }
 629
 630         return 0;
 631 }
 632
 633 int bch2_sb_replicas_to_cpu_replicas(struct bch_fs *c)
 634 {
 635         struct bch_sb_field_replicas *sb_v1;
 636         struct bch_sb_field_replicas_v0 *sb_v0;
 637         struct bch_replicas_cpu new_r = { 0, 0, NULL };
 638         int ret = 0;
 639
 640         if ((sb_v1 = bch2_sb_get_replicas(c->disk_sb.sb)))
 641                 ret = __bch2_sb_replicas_to_cpu_replicas(sb_v1, &new_r);
 642         else if ((sb_v0 = bch2_sb_get_replicas_v0(c->disk_sb.sb)))
 643                 ret = __bch2_sb_replicas_v0_to_cpu_replicas(sb_v0, &new_r);
 644
 645         if (ret)
 646                 return -ENOMEM;
 647
 648         bch2_cpu_replicas_sort(&new_r);
 649
 650         percpu_down_write(&c->mark_lock);
 651
 652         ret = replicas_table_update(c, &new_r);
 653         percpu_up_write(&c->mark_lock);
 654
 655         kfree(new_r.entries);
 656
 657         return 0;
 658 }
 659
 660 static int bch2_cpu_replicas_to_sb_replicas_v0(struct bch_fs *c,
 661                                                struct bch_replicas_cpu *r)
 662 {
 663         struct bch_sb_field_replicas_v0 *sb_r;
 664         struct bch_replicas_entry_v0 *dst;
 665         struct bch_replicas_entry *src;
 666         size_t bytes;
 667
 668         bytes = sizeof(struct bch_sb_field_replicas);
 669
 670         for_each_cpu_replicas_entry(r, src)
 671                 bytes += replicas_entry_bytes(src) - 1;
 672
 673         sb_r = bch2_sb_resize_replicas_v0(&c->disk_sb,
 674                         DIV_ROUND_UP(bytes, sizeof(u64)));
 675         if (!sb_r)
 676                 return -ENOSPC;
 677
 678         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas);
 679         sb_r = bch2_sb_get_replicas_v0(c->disk_sb.sb);
 680
 681         memset(&sb_r->entries, 0,
 682                vstruct_end(&sb_r->field) -
 683                (void *) &sb_r->entries);
 684
 685         dst = sb_r->entries;
 686         for_each_cpu_replicas_entry(r, src) {
 687                 dst->data_type  = src->data_type;
 688                 dst->nr_devs    = src->nr_devs;
 689                 memcpy(dst->devs, src->devs, src->nr_devs);
 690
 691                 dst = replicas_entry_next(dst);
 692
 693                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 694         }
 695
 696         return 0;
 697 }
 698
 699 static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *c,
 700                                             struct bch_replicas_cpu *r)
 701 {
 702         struct bch_sb_field_replicas *sb_r;
 703         struct bch_replicas_entry *dst, *src;
 704         bool need_v1 = false;
 705         size_t bytes;
 706
 707         bytes = sizeof(struct bch_sb_field_replicas);
 708
 709         for_each_cpu_replicas_entry(r, src) {
 710                 bytes += replicas_entry_bytes(src);
 711                 if (src->nr_required != 1)
 712                         need_v1 = true;
 713         }
 714
 715         if (!need_v1)
 716                 return bch2_cpu_replicas_to_sb_replicas_v0(c, r);
 717
 718         sb_r = bch2_sb_resize_replicas(&c->disk_sb,
 719                         DIV_ROUND_UP(bytes, sizeof(u64)));
 720         if (!sb_r)
 721                 return -ENOSPC;
 722
 723         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas_v0);
 724         sb_r = bch2_sb_get_replicas(c->disk_sb.sb);
 725
 726         memset(&sb_r->entries, 0,
 727                vstruct_end(&sb_r->field) -
 728                (void *) &sb_r->entries);
 729
 730         dst = sb_r->entries;
 731         for_each_cpu_replicas_entry(r, src) {
 732                 memcpy(dst, src, replicas_entry_bytes(src));
 733
 734                 dst = replicas_entry_next(dst);
 735
 736                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 737         }
 738
 739         return 0;
 740 }
 741
 742 static const char *check_dup_replicas_entries(struct bch_replicas_cpu *cpu_r)
 743 {
 744         unsigned i;
 745
 746         sort_cmp_size(cpu_r->entries,
 747                       cpu_r->nr,
 748                       cpu_r->entry_size,
 749                       memcmp, NULL);
 750
 751         for (i = 0; i + 1 < cpu_r->nr; i++) {
 752                 struct bch_replicas_entry *l =
 753                         cpu_replicas_entry(cpu_r, i);
 754                 struct bch_replicas_entry *r =
 755                         cpu_replicas_entry(cpu_r, i + 1);
 756
 757                 BUG_ON(memcmp(l, r, cpu_r->entry_size) > 0);
 758
 759                 if (!memcmp(l, r, cpu_r->entry_size))
 760                         return "duplicate replicas entry";
 761         }
 762
 763         return NULL;
 764 }
 765
 766 static const char *bch2_sb_validate_replicas(struct bch_sb *sb, struct bch_sb_field *f)
 767 {
 768         struct bch_sb_field_replicas *sb_r = field_to_type(f, replicas);
 769         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 770         struct bch_replicas_cpu cpu_r = { .entries = NULL };
 771         struct bch_replicas_entry *e;
 772         const char *err;
 773         unsigned i;
 774
 775         for_each_replicas_entry(sb_r, e) {
 776                 err = "invalid replicas entry: invalid data type";
 777                 if (e->data_type >= BCH_DATA_NR)
 778                         goto err;
 779
 780                 err = "invalid replicas entry: no devices";
 781                 if (!e->nr_devs)
 782                         goto err;
 783
 784                 err = "invalid replicas entry: bad nr_required";
 785                 if (!e->nr_required ||
 786                     (e->nr_required > 1 &&
 787                      e->nr_required >= e->nr_devs))
 788                         goto err;
 789
 790                 err = "invalid replicas entry: invalid device";
 791                 for (i = 0; i < e->nr_devs; i++)
 792                         if (!bch2_dev_exists(sb, mi, e->devs[i]))
 793                                 goto err;
 794         }
 795
 796         err = "cannot allocate memory";
 797         if (__bch2_sb_replicas_to_cpu_replicas(sb_r, &cpu_r))
 798                 goto err;
 799
 800         err = check_dup_replicas_entries(&cpu_r);
 801 err:
 802         kfree(cpu_r.entries);
 803         return err;
 804 }
 805
 806 static void bch2_sb_replicas_to_text(struct printbuf *out,
 807                                      struct bch_sb *sb,
 808                                      struct bch_sb_field *f)
 809 {
 810         struct bch_sb_field_replicas *r = field_to_type(f, replicas);
 811         struct bch_replicas_entry *e;
 812         bool first = true;
 813
 814         for_each_replicas_entry(r, e) {
 815                 if (!first)
 816                         pr_buf(out, " ");
 817                 first = false;
 818
 819                 bch2_replicas_entry_to_text(out, e);
 820         }
 821 }
 822
 823 const struct bch_sb_field_ops bch_sb_field_ops_replicas = {
 824         .validate       = bch2_sb_validate_replicas,
 825         .to_text        = bch2_sb_replicas_to_text,
 826 };
 827
 828 static const char *bch2_sb_validate_replicas_v0(struct bch_sb *sb, struct bch_sb_field *f)
 829 {
 830         struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0);
 831         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 832         struct bch_replicas_cpu cpu_r = { .entries = NULL };
 833         struct bch_replicas_entry_v0 *e;
 834         const char *err;
 835         unsigned i;
 836
 837         for_each_replicas_entry_v0(sb_r, e) {
 838                 err = "invalid replicas entry: invalid data type";
 839                 if (e->data_type >= BCH_DATA_NR)
 840                         goto err;
 841
 842                 err = "invalid replicas entry: no devices";
 843                 if (!e->nr_devs)
 844                         goto err;
 845
 846                 err = "invalid replicas entry: invalid device";
 847                 for (i = 0; i < e->nr_devs; i++)
 848                         if (!bch2_dev_exists(sb, mi, e->devs[i]))
 849                                 goto err;
 850         }
 851
 852         err = "cannot allocate memory";
 853         if (__bch2_sb_replicas_v0_to_cpu_replicas(sb_r, &cpu_r))
 854                 goto err;
 855
 856         err = check_dup_replicas_entries(&cpu_r);
 857 err:
 858         kfree(cpu_r.entries);
 859         return err;
 860 }
 861
 862 const struct bch_sb_field_ops bch_sb_field_ops_replicas_v0 = {
 863         .validate       = bch2_sb_validate_replicas_v0,
 864 };
 865
 866 /* Query replicas: */
 867
 868 struct replicas_status __bch2_replicas_status(struct bch_fs *c,
 869                                               struct bch_devs_mask online_devs)
 870 {
 871         struct bch_sb_field_members *mi;
 872         struct bch_replicas_entry *e;
 873         unsigned i, nr_online, nr_offline;
 874         struct replicas_status ret;
 875
 876         memset(&ret, 0, sizeof(ret));
 877
 878         for (i = 0; i < ARRAY_SIZE(ret.replicas); i++)
 879                 ret.replicas[i].redundancy = INT_MAX;
 880
 881         mi = bch2_sb_get_members(c->disk_sb.sb);
 882
 883         percpu_down_read_preempt_disable(&c->mark_lock);
 884
 885         for_each_cpu_replicas_entry(&c->replicas, e) {
 886                 if (e->data_type >= ARRAY_SIZE(ret.replicas))
 887                         panic("e %p data_type %u\n", e, e->data_type);
 888
 889                 nr_online = nr_offline = 0;
 890
 891                 for (i = 0; i < e->nr_devs; i++) {
 892                         BUG_ON(!bch2_dev_exists(c->disk_sb.sb, mi,
 893                                                 e->devs[i]));
 894
 895                         if (test_bit(e->devs[i], online_devs.d))
 896                                 nr_online++;
 897                         else
 898                                 nr_offline++;
 899                 }
 900
 901                 ret.replicas[e->data_type].redundancy =
 902                         min(ret.replicas[e->data_type].redundancy,
 903                             (int) nr_online - (int) e->nr_required);
 904
 905                 ret.replicas[e->data_type].nr_offline =
 906                         max(ret.replicas[e->data_type].nr_offline,
 907                             nr_offline);
 908         }
 909
 910         percpu_up_read_preempt_enable(&c->mark_lock);
 911
 912         for (i = 0; i < ARRAY_SIZE(ret.replicas); i++)
 913                 if (ret.replicas[i].redundancy == INT_MAX)
 914                         ret.replicas[i].redundancy = 0;
 915
 916         return ret;
 917 }
 918
 919 struct replicas_status bch2_replicas_status(struct bch_fs *c)
 920 {
 921         return __bch2_replicas_status(c, bch2_online_devs(c));
 922 }
 923
 924 static bool have_enough_devs(struct replicas_status s,
 925                              enum bch_data_type type,
 926                              bool force_if_degraded,
 927                              bool force_if_lost)
 928 {
 929         return (!s.replicas[type].nr_offline || force_if_degraded) &&
 930                 (s.replicas[type].redundancy >= 0 || force_if_lost);
 931 }
 932
 933 bool bch2_have_enough_devs(struct replicas_status s, unsigned flags)
 934 {
 935         return (have_enough_devs(s, BCH_DATA_JOURNAL,
 936                                  flags & BCH_FORCE_IF_METADATA_DEGRADED,
 937                                  flags & BCH_FORCE_IF_METADATA_LOST) &&
 938                 have_enough_devs(s, BCH_DATA_BTREE,
 939                                  flags & BCH_FORCE_IF_METADATA_DEGRADED,
 940                                  flags & BCH_FORCE_IF_METADATA_LOST) &&
 941                 have_enough_devs(s, BCH_DATA_USER,
 942                                  flags & BCH_FORCE_IF_DATA_DEGRADED,
 943                                  flags & BCH_FORCE_IF_DATA_LOST));
 944 }
 945
 946 int bch2_replicas_online(struct bch_fs *c, bool meta)
 947 {
 948         struct replicas_status s = bch2_replicas_status(c);
 949
 950         return (meta
 951                 ? min(s.replicas[BCH_DATA_JOURNAL].redundancy,
 952                       s.replicas[BCH_DATA_BTREE].redundancy)
 953                 : s.replicas[BCH_DATA_USER].redundancy) + 1;
 954 }
 955
 956 unsigned bch2_dev_has_data(struct bch_fs *c, struct bch_dev *ca)
 957 {
 958         struct bch_replicas_entry *e;
 959         unsigned i, ret = 0;
 960
 961         percpu_down_read_preempt_disable(&c->mark_lock);
 962
 963         for_each_cpu_replicas_entry(&c->replicas, e)
 964                 for (i = 0; i < e->nr_devs; i++)
 965                         if (e->devs[i] == ca->dev_idx)
 966                                 ret |= 1 << e->data_type;
 967
 968         percpu_up_read_preempt_enable(&c->mark_lock);
 969
 970         return ret;
 971 }
 972
 973 int bch2_fs_replicas_init(struct bch_fs *c)
 974 {
 975         c->journal.entry_u64s_reserved +=
 976                 reserve_journal_replicas(c, &c->replicas);
 977         return 0;
 978 }