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

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "buckets.h"
   5 #include "journal.h"
   6 #include "replicas.h"
   7 #include "super-io.h"
   8
   9 static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *,
  10                                             struct bch_replicas_cpu *);
  11
  12 /* Replicas tracking - in memory: */
  13
  14 static void verify_replicas_entry(struct bch_replicas_entry *e)
  15 {
  16 #ifdef CONFIG_BCACHEFS_DEBUG
  17         unsigned i;
  18
  19         BUG_ON(e->data_type >= BCH_DATA_NR);
  20         BUG_ON(!e->nr_devs);
  21         BUG_ON(e->nr_required > 1 &&
  22                e->nr_required >= e->nr_devs);
  23
  24         for (i = 0; i + 1 < e->nr_devs; i++)
  25                 BUG_ON(e->devs[i] >= e->devs[i + 1]);
  26 #endif
  27 }
  28
  29 void bch2_replicas_entry_sort(struct bch_replicas_entry *e)
  30 {
  31         bubble_sort(e->devs, e->nr_devs, u8_cmp);
  32 }
  33
  34 static void bch2_cpu_replicas_sort(struct bch_replicas_cpu *r)
  35 {
  36         eytzinger0_sort(r->entries, r->nr, r->entry_size, memcmp, NULL);
  37 }
  38
  39 void bch2_replicas_entry_v0_to_text(struct printbuf *out,
  40                                     struct bch_replicas_entry_v0 *e)
  41 {
  42         unsigned i;
  43
  44         if (e->data_type < BCH_DATA_NR)
  45                 pr_buf(out, "%s", bch2_data_types[e->data_type]);
  46         else
  47                 pr_buf(out, "(invalid data type %u)", e->data_type);
  48
  49         pr_buf(out, ": %u [", e->nr_devs);
  50         for (i = 0; i < e->nr_devs; i++)
  51                 pr_buf(out, i ? " %u" : "%u", e->devs[i]);
  52         pr_buf(out, "]");
  53 }
  54
  55 void bch2_replicas_entry_to_text(struct printbuf *out,
  56                                  struct bch_replicas_entry *e)
  57 {
  58         unsigned i;
  59
  60         if (e->data_type < BCH_DATA_NR)
  61                 pr_buf(out, "%s", bch2_data_types[e->data_type]);
  62         else
  63                 pr_buf(out, "(invalid data type %u)", e->data_type);
  64
  65         pr_buf(out, ": %u/%u [", e->nr_required, e->nr_devs);
  66         for (i = 0; i < e->nr_devs; i++)
  67                 pr_buf(out, i ? " %u" : "%u", e->devs[i]);
  68         pr_buf(out, "]");
  69 }
  70
  71 void bch2_cpu_replicas_to_text(struct printbuf *out,
  72                                struct bch_replicas_cpu *r)
  73 {
  74         struct bch_replicas_entry *e;
  75         bool first = true;
  76
  77         for_each_cpu_replicas_entry(r, e) {
  78                 if (!first)
  79                         pr_buf(out, " ");
  80                 first = false;
  81
  82                 bch2_replicas_entry_to_text(out, e);
  83         }
  84 }
  85
  86 static void extent_to_replicas(struct bkey_s_c k,
  87                                struct bch_replicas_entry *r)
  88 {
  89         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
  90         const union bch_extent_entry *entry;
  91         struct extent_ptr_decoded p;
  92
  93         r->nr_required  = 1;
  94
  95         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
  96                 if (p.ptr.cached)
  97                         continue;
  98
  99                 if (!p.has_ec)
 100                         r->devs[r->nr_devs++] = p.ptr.dev;
 101                 else
 102                         r->nr_required = 0;
 103         }
 104 }
 105
 106 static void stripe_to_replicas(struct bkey_s_c k,
 107                                struct bch_replicas_entry *r)
 108 {
 109         struct bkey_s_c_stripe s = bkey_s_c_to_stripe(k);
 110         const struct bch_extent_ptr *ptr;
 111
 112         r->nr_required  = s.v->nr_blocks - s.v->nr_redundant;
 113
 114         for (ptr = s.v->ptrs;
 115              ptr < s.v->ptrs + s.v->nr_blocks;
 116              ptr++)
 117                 r->devs[r->nr_devs++] = ptr->dev;
 118 }
 119
 120 void bch2_bkey_to_replicas(struct bch_replicas_entry *e,
 121                            struct bkey_s_c k)
 122 {
 123         e->nr_devs = 0;
 124
 125         switch (k.k->type) {
 126         case KEY_TYPE_btree_ptr:
 127         case KEY_TYPE_btree_ptr_v2:
 128                 e->data_type = BCH_DATA_btree;
 129                 extent_to_replicas(k, e);
 130                 break;
 131         case KEY_TYPE_extent:
 132         case KEY_TYPE_reflink_v:
 133                 e->data_type = BCH_DATA_user;
 134                 extent_to_replicas(k, e);
 135                 break;
 136         case KEY_TYPE_stripe:
 137                 e->data_type = BCH_DATA_parity;
 138                 stripe_to_replicas(k, e);
 139                 break;
 140         }
 141
 142         bch2_replicas_entry_sort(e);
 143 }
 144
 145 void bch2_devlist_to_replicas(struct bch_replicas_entry *e,
 146                               enum bch_data_type data_type,
 147                               struct bch_devs_list devs)
 148 {
 149         unsigned i;
 150
 151         BUG_ON(!data_type ||
 152                data_type == BCH_DATA_sb ||
 153                data_type >= BCH_DATA_NR);
 154
 155         e->data_type    = data_type;
 156         e->nr_devs      = 0;
 157         e->nr_required  = 1;
 158
 159         for (i = 0; i < devs.nr; i++)
 160                 e->devs[e->nr_devs++] = devs.devs[i];
 161
 162         bch2_replicas_entry_sort(e);
 163 }
 164
 165 static struct bch_replicas_cpu
 166 cpu_replicas_add_entry(struct bch_replicas_cpu *old,
 167                        struct bch_replicas_entry *new_entry)
 168 {
 169         unsigned i;
 170         struct bch_replicas_cpu new = {
 171                 .nr             = old->nr + 1,
 172                 .entry_size     = max_t(unsigned, old->entry_size,
 173                                         replicas_entry_bytes(new_entry)),
 174         };
 175
 176         BUG_ON(!new_entry->data_type);
 177         verify_replicas_entry(new_entry);
 178
 179         new.entries = kcalloc(new.nr, new.entry_size, GFP_KERNEL);
 180         if (!new.entries)
 181                 return new;
 182
 183         for (i = 0; i < old->nr; i++)
 184                 memcpy(cpu_replicas_entry(&new, i),
 185                        cpu_replicas_entry(old, i),
 186                        old->entry_size);
 187
 188         memcpy(cpu_replicas_entry(&new, old->nr),
 189                new_entry,
 190                replicas_entry_bytes(new_entry));
 191
 192         bch2_cpu_replicas_sort(&new);
 193         return new;
 194 }
 195
 196 static inline int __replicas_entry_idx(struct bch_replicas_cpu *r,
 197                                        struct bch_replicas_entry *search)
 198 {
 199         int idx, entry_size = replicas_entry_bytes(search);
 200
 201         if (unlikely(entry_size > r->entry_size))
 202                 return -1;
 203
 204         verify_replicas_entry(search);
 205
 206 #define entry_cmp(_l, _r, size) memcmp(_l, _r, entry_size)
 207         idx = eytzinger0_find(r->entries, r->nr, r->entry_size,
 208                               entry_cmp, search);
 209 #undef entry_cmp
 210
 211         return idx < r->nr ? idx : -1;
 212 }
 213
 214 int bch2_replicas_entry_idx(struct bch_fs *c,
 215                             struct bch_replicas_entry *search)
 216 {
 217         bch2_replicas_entry_sort(search);
 218
 219         return __replicas_entry_idx(&c->replicas, search);
 220 }
 221
 222 static bool __replicas_has_entry(struct bch_replicas_cpu *r,
 223                                  struct bch_replicas_entry *search)
 224 {
 225         return __replicas_entry_idx(r, search) >= 0;
 226 }
 227
 228 bool bch2_replicas_marked(struct bch_fs *c,
 229                           struct bch_replicas_entry *search)
 230 {
 231         bool marked;
 232
 233         if (!search->nr_devs)
 234                 return true;
 235
 236         verify_replicas_entry(search);
 237
 238         percpu_down_read(&c->mark_lock);
 239         marked = __replicas_has_entry(&c->replicas, search) &&
 240                 (likely((!c->replicas_gc.entries)) ||
 241                  __replicas_has_entry(&c->replicas_gc, search));
 242         percpu_up_read(&c->mark_lock);
 243
 244         return marked;
 245 }
 246
 247 static void __replicas_table_update(struct bch_fs_usage *dst,
 248                                     struct bch_replicas_cpu *dst_r,
 249                                     struct bch_fs_usage *src,
 250                                     struct bch_replicas_cpu *src_r)
 251 {
 252         int src_idx, dst_idx;
 253
 254         *dst = *src;
 255
 256         for (src_idx = 0; src_idx < src_r->nr; src_idx++) {
 257                 if (!src->replicas[src_idx])
 258                         continue;
 259
 260                 dst_idx = __replicas_entry_idx(dst_r,
 261                                 cpu_replicas_entry(src_r, src_idx));
 262                 BUG_ON(dst_idx < 0);
 263
 264                 dst->replicas[dst_idx] = src->replicas[src_idx];
 265         }
 266 }
 267
 268 static void __replicas_table_update_pcpu(struct bch_fs_usage __percpu *dst_p,
 269                                     struct bch_replicas_cpu *dst_r,
 270                                     struct bch_fs_usage __percpu *src_p,
 271                                     struct bch_replicas_cpu *src_r)
 272 {
 273         unsigned src_nr = sizeof(struct bch_fs_usage) / sizeof(u64) + src_r->nr;
 274         struct bch_fs_usage *dst, *src = (void *)
 275                 bch2_acc_percpu_u64s((void *) src_p, src_nr);
 276
 277         preempt_disable();
 278         dst = this_cpu_ptr(dst_p);
 279         preempt_enable();
 280
 281         __replicas_table_update(dst, dst_r, src, src_r);
 282 }
 283
 284 /*
 285  * Resize filesystem accounting:
 286  */
 287 static int replicas_table_update(struct bch_fs *c,
 288                                  struct bch_replicas_cpu *new_r)
 289 {
 290         struct bch_fs_usage __percpu *new_usage[JOURNAL_BUF_NR];
 291         struct bch_fs_usage_online *new_scratch = NULL;
 292         struct bch_fs_usage __percpu *new_gc = NULL;
 293         struct bch_fs_usage *new_base = NULL;
 294         unsigned i, bytes = sizeof(struct bch_fs_usage) +
 295                 sizeof(u64) * new_r->nr;
 296         unsigned scratch_bytes = sizeof(struct bch_fs_usage_online) +
 297                 sizeof(u64) * new_r->nr;
 298         int ret = 0;
 299
 300         memset(new_usage, 0, sizeof(new_usage));
 301
 302         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 303                 if (!(new_usage[i] = __alloc_percpu_gfp(bytes,
 304                                         sizeof(u64), GFP_KERNEL)))
 305                         goto err;
 306
 307         if (!(new_base = kzalloc(bytes, GFP_KERNEL)) ||
 308             !(new_scratch  = kmalloc(scratch_bytes, GFP_KERNEL)) ||
 309             (c->usage_gc &&
 310              !(new_gc = __alloc_percpu_gfp(bytes, sizeof(u64), GFP_KERNEL))))
 311                 goto err;
 312
 313         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 314                 if (c->usage[i])
 315                         __replicas_table_update_pcpu(new_usage[i], new_r,
 316                                                      c->usage[i], &c->replicas);
 317         if (c->usage_base)
 318                 __replicas_table_update(new_base,               new_r,
 319                                         c->usage_base,          &c->replicas);
 320         if (c->usage_gc)
 321                 __replicas_table_update_pcpu(new_gc,            new_r,
 322                                              c->usage_gc,       &c->replicas);
 323
 324         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 325                 swap(c->usage[i],       new_usage[i]);
 326         swap(c->usage_base,     new_base);
 327         swap(c->usage_scratch,  new_scratch);
 328         swap(c->usage_gc,       new_gc);
 329         swap(c->replicas,       *new_r);
 330 out:
 331         free_percpu(new_gc);
 332         kfree(new_scratch);
 333         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 334                 free_percpu(new_usage[i]);
 335         kfree(new_base);
 336         return ret;
 337 err:
 338         bch_err(c, "error updating replicas table: memory allocation failure");
 339         ret = -ENOMEM;
 340         goto out;
 341 }
 342
 343 static unsigned reserve_journal_replicas(struct bch_fs *c,
 344                                      struct bch_replicas_cpu *r)
 345 {
 346         struct bch_replicas_entry *e;
 347         unsigned journal_res_u64s = 0;
 348
 349         /* nr_inodes: */
 350         journal_res_u64s +=
 351                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 352
 353         /* key_version: */
 354         journal_res_u64s +=
 355                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 356
 357         /* persistent_reserved: */
 358         journal_res_u64s +=
 359                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64)) *
 360                 BCH_REPLICAS_MAX;
 361
 362         for_each_cpu_replicas_entry(r, e)
 363                 journal_res_u64s +=
 364                         DIV_ROUND_UP(sizeof(struct jset_entry_data_usage) +
 365                                      e->nr_devs, sizeof(u64));
 366         return journal_res_u64s;
 367 }
 368
 369 noinline
 370 static int bch2_mark_replicas_slowpath(struct bch_fs *c,
 371                                 struct bch_replicas_entry *new_entry)
 372 {
 373         struct bch_replicas_cpu new_r, new_gc;
 374         int ret = 0;
 375
 376         verify_replicas_entry(new_entry);
 377
 378         memset(&new_r, 0, sizeof(new_r));
 379         memset(&new_gc, 0, sizeof(new_gc));
 380
 381         mutex_lock(&c->sb_lock);
 382
 383         if (c->replicas_gc.entries &&
 384             !__replicas_has_entry(&c->replicas_gc, new_entry)) {
 385                 new_gc = cpu_replicas_add_entry(&c->replicas_gc, new_entry);
 386                 if (!new_gc.entries)
 387                         goto err;
 388         }
 389
 390         if (!__replicas_has_entry(&c->replicas, new_entry)) {
 391                 new_r = cpu_replicas_add_entry(&c->replicas, new_entry);
 392                 if (!new_r.entries)
 393                         goto err;
 394
 395                 ret = bch2_cpu_replicas_to_sb_replicas(c, &new_r);
 396                 if (ret)
 397                         goto err;
 398
 399                 bch2_journal_entry_res_resize(&c->journal,
 400                                 &c->replicas_journal_res,
 401                                 reserve_journal_replicas(c, &new_r));
 402         }
 403
 404         if (!new_r.entries &&
 405             !new_gc.entries)
 406                 goto out;
 407
 408         /* allocations done, now commit: */
 409
 410         if (new_r.entries)
 411                 bch2_write_super(c);
 412
 413         /* don't update in memory replicas until changes are persistent */
 414         percpu_down_write(&c->mark_lock);
 415         if (new_r.entries)
 416                 ret = replicas_table_update(c, &new_r);
 417         if (new_gc.entries)
 418                 swap(new_gc, c->replicas_gc);
 419         percpu_up_write(&c->mark_lock);
 420 out:
 421         mutex_unlock(&c->sb_lock);
 422
 423         kfree(new_r.entries);
 424         kfree(new_gc.entries);
 425
 426         return ret;
 427 err:
 428         bch_err(c, "error adding replicas entry: memory allocation failure");
 429         ret = -ENOMEM;
 430         goto out;
 431 }
 432
 433 int bch2_mark_replicas(struct bch_fs *c, struct bch_replicas_entry *r)
 434 {
 435         return likely(bch2_replicas_marked(c, r))
 436                 ? 0 : bch2_mark_replicas_slowpath(c, r);
 437 }
 438
 439 /* replicas delta list: */
 440
 441 int bch2_replicas_delta_list_mark(struct bch_fs *c,
 442                                   struct replicas_delta_list *r)
 443 {
 444         struct replicas_delta *d = r->d;
 445         struct replicas_delta *top = (void *) r->d + r->used;
 446         int ret = 0;
 447
 448         for (d = r->d; !ret && d != top; d = replicas_delta_next(d))
 449                 ret = bch2_mark_replicas(c, &d->r);
 450         return ret;
 451 }
 452
 453 /*
 454  * Old replicas_gc mechanism: only used for journal replicas entries now, should
 455  * die at some point:
 456  */
 457
 458 int bch2_replicas_gc_end(struct bch_fs *c, int ret)
 459 {
 460         unsigned i;
 461
 462         lockdep_assert_held(&c->replicas_gc_lock);
 463
 464         mutex_lock(&c->sb_lock);
 465         percpu_down_write(&c->mark_lock);
 466
 467         /*
 468          * this is kind of crappy; the replicas gc mechanism needs to be ripped
 469          * out
 470          */
 471
 472         for (i = 0; i < c->replicas.nr; i++) {
 473                 struct bch_replicas_entry *e =
 474                         cpu_replicas_entry(&c->replicas, i);
 475                 struct bch_replicas_cpu n;
 476
 477                 if (!__replicas_has_entry(&c->replicas_gc, e) &&
 478                     bch2_fs_usage_read_one(c, &c->usage_base->replicas[i])) {
 479                         n = cpu_replicas_add_entry(&c->replicas_gc, e);
 480                         if (!n.entries) {
 481                                 ret = -ENOSPC;
 482                                 goto err;
 483                         }
 484
 485                         swap(n, c->replicas_gc);
 486                         kfree(n.entries);
 487                 }
 488         }
 489
 490         if (bch2_cpu_replicas_to_sb_replicas(c, &c->replicas_gc)) {
 491                 ret = -ENOSPC;
 492                 goto err;
 493         }
 494
 495         ret = replicas_table_update(c, &c->replicas_gc);
 496 err:
 497         kfree(c->replicas_gc.entries);
 498         c->replicas_gc.entries = NULL;
 499
 500         percpu_up_write(&c->mark_lock);
 501
 502         if (!ret)
 503                 bch2_write_super(c);
 504
 505         mutex_unlock(&c->sb_lock);
 506
 507         return ret;
 508 }
 509
 510 int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask)
 511 {
 512         struct bch_replicas_entry *e;
 513         unsigned i = 0;
 514
 515         lockdep_assert_held(&c->replicas_gc_lock);
 516
 517         mutex_lock(&c->sb_lock);
 518         BUG_ON(c->replicas_gc.entries);
 519
 520         c->replicas_gc.nr               = 0;
 521         c->replicas_gc.entry_size       = 0;
 522
 523         for_each_cpu_replicas_entry(&c->replicas, e)
 524                 if (!((1 << e->data_type) & typemask)) {
 525                         c->replicas_gc.nr++;
 526                         c->replicas_gc.entry_size =
 527                                 max_t(unsigned, c->replicas_gc.entry_size,
 528                                       replicas_entry_bytes(e));
 529                 }
 530
 531         c->replicas_gc.entries = kcalloc(c->replicas_gc.nr,
 532                                          c->replicas_gc.entry_size,
 533                                          GFP_KERNEL);
 534         if (!c->replicas_gc.entries) {
 535                 mutex_unlock(&c->sb_lock);
 536                 bch_err(c, "error allocating c->replicas_gc");
 537                 return -ENOMEM;
 538         }
 539
 540         for_each_cpu_replicas_entry(&c->replicas, e)
 541                 if (!((1 << e->data_type) & typemask))
 542                         memcpy(cpu_replicas_entry(&c->replicas_gc, i++),
 543                                e, c->replicas_gc.entry_size);
 544
 545         bch2_cpu_replicas_sort(&c->replicas_gc);
 546         mutex_unlock(&c->sb_lock);
 547
 548         return 0;
 549 }
 550
 551 /* New much simpler mechanism for clearing out unneeded replicas entries: */
 552
 553 int bch2_replicas_gc2(struct bch_fs *c)
 554 {
 555         struct bch_replicas_cpu new = { 0 };
 556         unsigned i, nr;
 557         int ret = 0;
 558
 559         bch2_journal_meta(&c->journal);
 560 retry:
 561         nr              = READ_ONCE(c->replicas.nr);
 562         new.entry_size  = READ_ONCE(c->replicas.entry_size);
 563         new.entries     = kcalloc(nr, new.entry_size, GFP_KERNEL);
 564         if (!new.entries) {
 565                 bch_err(c, "error allocating c->replicas_gc");
 566                 return -ENOMEM;
 567         }
 568
 569         mutex_lock(&c->sb_lock);
 570         percpu_down_write(&c->mark_lock);
 571
 572         if (nr                  != c->replicas.nr ||
 573             new.entry_size      != c->replicas.entry_size) {
 574                 percpu_up_write(&c->mark_lock);
 575                 mutex_unlock(&c->sb_lock);
 576                 kfree(new.entries);
 577                 goto retry;
 578         }
 579
 580         for (i = 0; i < c->replicas.nr; i++) {
 581                 struct bch_replicas_entry *e =
 582                         cpu_replicas_entry(&c->replicas, i);
 583
 584                 if (e->data_type == BCH_DATA_journal ||
 585                     c->usage_base->replicas[i] ||
 586                     percpu_u64_get(&c->usage[0]->replicas[i]) ||
 587                     percpu_u64_get(&c->usage[1]->replicas[i]) ||
 588                     percpu_u64_get(&c->usage[2]->replicas[i]) ||
 589                     percpu_u64_get(&c->usage[3]->replicas[i]))
 590                         memcpy(cpu_replicas_entry(&new, new.nr++),
 591                                e, new.entry_size);
 592         }
 593
 594         bch2_cpu_replicas_sort(&new);
 595
 596         if (bch2_cpu_replicas_to_sb_replicas(c, &new)) {
 597                 ret = -ENOSPC;
 598                 goto err;
 599         }
 600
 601         ret = replicas_table_update(c, &new);
 602 err:
 603         kfree(new.entries);
 604
 605         percpu_up_write(&c->mark_lock);
 606
 607         if (!ret)
 608                 bch2_write_super(c);
 609
 610         mutex_unlock(&c->sb_lock);
 611
 612         return ret;
 613 }
 614
 615 int bch2_replicas_set_usage(struct bch_fs *c,
 616                             struct bch_replicas_entry *r,
 617                             u64 sectors)
 618 {
 619         int ret, idx = bch2_replicas_entry_idx(c, r);
 620
 621         if (idx < 0) {
 622                 struct bch_replicas_cpu n;
 623
 624                 n = cpu_replicas_add_entry(&c->replicas, r);
 625                 if (!n.entries)
 626                         return -ENOMEM;
 627
 628                 ret = replicas_table_update(c, &n);
 629                 if (ret)
 630                         return ret;
 631
 632                 kfree(n.entries);
 633
 634                 idx = bch2_replicas_entry_idx(c, r);
 635                 BUG_ON(ret < 0);
 636         }
 637
 638         c->usage_base->replicas[idx] = sectors;
 639
 640         return 0;
 641 }
 642
 643 /* Replicas tracking - superblock: */
 644
 645 static int
 646 __bch2_sb_replicas_to_cpu_replicas(struct bch_sb_field_replicas *sb_r,
 647                                    struct bch_replicas_cpu *cpu_r)
 648 {
 649         struct bch_replicas_entry *e, *dst;
 650         unsigned nr = 0, entry_size = 0, idx = 0;
 651
 652         for_each_replicas_entry(sb_r, e) {
 653                 entry_size = max_t(unsigned, entry_size,
 654                                    replicas_entry_bytes(e));
 655                 nr++;
 656         }
 657
 658         cpu_r->entries = kcalloc(nr, entry_size, GFP_KERNEL);
 659         if (!cpu_r->entries)
 660                 return -ENOMEM;
 661
 662         cpu_r->nr               = nr;
 663         cpu_r->entry_size       = entry_size;
 664
 665         for_each_replicas_entry(sb_r, e) {
 666                 dst = cpu_replicas_entry(cpu_r, idx++);
 667                 memcpy(dst, e, replicas_entry_bytes(e));
 668                 bch2_replicas_entry_sort(dst);
 669         }
 670
 671         return 0;
 672 }
 673
 674 static int
 675 __bch2_sb_replicas_v0_to_cpu_replicas(struct bch_sb_field_replicas_v0 *sb_r,
 676                                       struct bch_replicas_cpu *cpu_r)
 677 {
 678         struct bch_replicas_entry_v0 *e;
 679         unsigned nr = 0, entry_size = 0, idx = 0;
 680
 681         for_each_replicas_entry(sb_r, e) {
 682                 entry_size = max_t(unsigned, entry_size,
 683                                    replicas_entry_bytes(e));
 684                 nr++;
 685         }
 686
 687         entry_size += sizeof(struct bch_replicas_entry) -
 688                 sizeof(struct bch_replicas_entry_v0);
 689
 690         cpu_r->entries = kcalloc(nr, entry_size, GFP_KERNEL);
 691         if (!cpu_r->entries)
 692                 return -ENOMEM;
 693
 694         cpu_r->nr               = nr;
 695         cpu_r->entry_size       = entry_size;
 696
 697         for_each_replicas_entry(sb_r, e) {
 698                 struct bch_replicas_entry *dst =
 699                         cpu_replicas_entry(cpu_r, idx++);
 700
 701                 dst->data_type  = e->data_type;
 702                 dst->nr_devs    = e->nr_devs;
 703                 dst->nr_required = 1;
 704                 memcpy(dst->devs, e->devs, e->nr_devs);
 705                 bch2_replicas_entry_sort(dst);
 706         }
 707
 708         return 0;
 709 }
 710
 711 int bch2_sb_replicas_to_cpu_replicas(struct bch_fs *c)
 712 {
 713         struct bch_sb_field_replicas *sb_v1;
 714         struct bch_sb_field_replicas_v0 *sb_v0;
 715         struct bch_replicas_cpu new_r = { 0, 0, NULL };
 716         int ret = 0;
 717
 718         if ((sb_v1 = bch2_sb_get_replicas(c->disk_sb.sb)))
 719                 ret = __bch2_sb_replicas_to_cpu_replicas(sb_v1, &new_r);
 720         else if ((sb_v0 = bch2_sb_get_replicas_v0(c->disk_sb.sb)))
 721                 ret = __bch2_sb_replicas_v0_to_cpu_replicas(sb_v0, &new_r);
 722
 723         if (ret)
 724                 return -ENOMEM;
 725
 726         bch2_cpu_replicas_sort(&new_r);
 727
 728         percpu_down_write(&c->mark_lock);
 729
 730         ret = replicas_table_update(c, &new_r);
 731         percpu_up_write(&c->mark_lock);
 732
 733         kfree(new_r.entries);
 734
 735         return 0;
 736 }
 737
 738 static int bch2_cpu_replicas_to_sb_replicas_v0(struct bch_fs *c,
 739                                                struct bch_replicas_cpu *r)
 740 {
 741         struct bch_sb_field_replicas_v0 *sb_r;
 742         struct bch_replicas_entry_v0 *dst;
 743         struct bch_replicas_entry *src;
 744         size_t bytes;
 745
 746         bytes = sizeof(struct bch_sb_field_replicas);
 747
 748         for_each_cpu_replicas_entry(r, src)
 749                 bytes += replicas_entry_bytes(src) - 1;
 750
 751         sb_r = bch2_sb_resize_replicas_v0(&c->disk_sb,
 752                         DIV_ROUND_UP(bytes, sizeof(u64)));
 753         if (!sb_r)
 754                 return -ENOSPC;
 755
 756         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas);
 757         sb_r = bch2_sb_get_replicas_v0(c->disk_sb.sb);
 758
 759         memset(&sb_r->entries, 0,
 760                vstruct_end(&sb_r->field) -
 761                (void *) &sb_r->entries);
 762
 763         dst = sb_r->entries;
 764         for_each_cpu_replicas_entry(r, src) {
 765                 dst->data_type  = src->data_type;
 766                 dst->nr_devs    = src->nr_devs;
 767                 memcpy(dst->devs, src->devs, src->nr_devs);
 768
 769                 dst = replicas_entry_next(dst);
 770
 771                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 772         }
 773
 774         return 0;
 775 }
 776
 777 static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *c,
 778                                             struct bch_replicas_cpu *r)
 779 {
 780         struct bch_sb_field_replicas *sb_r;
 781         struct bch_replicas_entry *dst, *src;
 782         bool need_v1 = false;
 783         size_t bytes;
 784
 785         bytes = sizeof(struct bch_sb_field_replicas);
 786
 787         for_each_cpu_replicas_entry(r, src) {
 788                 bytes += replicas_entry_bytes(src);
 789                 if (src->nr_required != 1)
 790                         need_v1 = true;
 791         }
 792
 793         if (!need_v1)
 794                 return bch2_cpu_replicas_to_sb_replicas_v0(c, r);
 795
 796         sb_r = bch2_sb_resize_replicas(&c->disk_sb,
 797                         DIV_ROUND_UP(bytes, sizeof(u64)));
 798         if (!sb_r)
 799                 return -ENOSPC;
 800
 801         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas_v0);
 802         sb_r = bch2_sb_get_replicas(c->disk_sb.sb);
 803
 804         memset(&sb_r->entries, 0,
 805                vstruct_end(&sb_r->field) -
 806                (void *) &sb_r->entries);
 807
 808         dst = sb_r->entries;
 809         for_each_cpu_replicas_entry(r, src) {
 810                 memcpy(dst, src, replicas_entry_bytes(src));
 811
 812                 dst = replicas_entry_next(dst);
 813
 814                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 815         }
 816
 817         return 0;
 818 }
 819
 820 static int bch2_cpu_replicas_validate(struct bch_replicas_cpu *cpu_r,
 821                                       struct bch_sb *sb,
 822                                       struct printbuf *err)
 823 {
 824         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 825         unsigned i, j;
 826
 827         sort_cmp_size(cpu_r->entries,
 828                       cpu_r->nr,
 829                       cpu_r->entry_size,
 830                       memcmp, NULL);
 831
 832         for (i = 0; i < cpu_r->nr; i++) {
 833                 struct bch_replicas_entry *e =
 834                         cpu_replicas_entry(cpu_r, i);
 835
 836                 if (e->data_type >= BCH_DATA_NR) {
 837                         pr_buf(err, "invalid data type in entry ");
 838                         bch2_replicas_entry_to_text(err, e);
 839                         return -EINVAL;
 840                 }
 841
 842                 if (!e->nr_devs) {
 843                         pr_buf(err, "no devices in entry ");
 844                         bch2_replicas_entry_to_text(err, e);
 845                         return -EINVAL;
 846                 }
 847
 848                 if (e->nr_required > 1 &&
 849                     e->nr_required >= e->nr_devs) {
 850                         pr_buf(err, "bad nr_required in entry ");
 851                         bch2_replicas_entry_to_text(err, e);
 852                         return -EINVAL;
 853                 }
 854
 855                 for (j = 0; j < e->nr_devs; j++)
 856                         if (!bch2_dev_exists(sb, mi, e->devs[j])) {
 857                                 pr_buf(err, "invalid device %u in entry ", e->devs[j]);
 858                                 bch2_replicas_entry_to_text(err, e);
 859                                 return -EINVAL;
 860                         }
 861
 862                 if (i + 1 < cpu_r->nr) {
 863                         struct bch_replicas_entry *n =
 864                                 cpu_replicas_entry(cpu_r, i + 1);
 865
 866                         BUG_ON(memcmp(e, n, cpu_r->entry_size) > 0);
 867
 868                         if (!memcmp(e, n, cpu_r->entry_size)) {
 869                                 pr_buf(err, "duplicate replicas entry ");
 870                                 bch2_replicas_entry_to_text(err, e);
 871                                 return -EINVAL;
 872                         }
 873                 }
 874         }
 875
 876         return 0;
 877 }
 878
 879 static int bch2_sb_replicas_validate(struct bch_sb *sb, struct bch_sb_field *f,
 880                                      struct printbuf *err)
 881 {
 882         struct bch_sb_field_replicas *sb_r = field_to_type(f, replicas);
 883         struct bch_replicas_cpu cpu_r;
 884         int ret;
 885
 886         if (__bch2_sb_replicas_to_cpu_replicas(sb_r, &cpu_r))
 887                 return -ENOMEM;
 888
 889         ret = bch2_cpu_replicas_validate(&cpu_r, sb, err);
 890         kfree(cpu_r.entries);
 891         return ret;
 892 }
 893
 894 static void bch2_sb_replicas_to_text(struct printbuf *out,
 895                                      struct bch_sb *sb,
 896                                      struct bch_sb_field *f)
 897 {
 898         struct bch_sb_field_replicas *r = field_to_type(f, replicas);
 899         struct bch_replicas_entry *e;
 900         bool first = true;
 901
 902         for_each_replicas_entry(r, e) {
 903                 if (!first)
 904                         pr_buf(out, " ");
 905                 first = false;
 906
 907                 bch2_replicas_entry_to_text(out, e);
 908         }
 909         pr_newline(out);
 910 }
 911
 912 const struct bch_sb_field_ops bch_sb_field_ops_replicas = {
 913         .validate       = bch2_sb_replicas_validate,
 914         .to_text        = bch2_sb_replicas_to_text,
 915 };
 916
 917 static int bch2_sb_replicas_v0_validate(struct bch_sb *sb, struct bch_sb_field *f,
 918                                         struct printbuf *err)
 919 {
 920         struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0);
 921         struct bch_replicas_cpu cpu_r;
 922         int ret;
 923
 924         if (__bch2_sb_replicas_v0_to_cpu_replicas(sb_r, &cpu_r))
 925                 return -ENOMEM;
 926
 927         ret = bch2_cpu_replicas_validate(&cpu_r, sb, err);
 928         kfree(cpu_r.entries);
 929         return ret;
 930 }
 931
 932 static void bch2_sb_replicas_v0_to_text(struct printbuf *out,
 933                                         struct bch_sb *sb,
 934                                         struct bch_sb_field *f)
 935 {
 936         struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0);
 937         struct bch_replicas_entry_v0 *e;
 938         bool first = true;
 939
 940         for_each_replicas_entry(sb_r, e) {
 941                 if (!first)
 942                         pr_buf(out, " ");
 943                 first = false;
 944
 945                 bch2_replicas_entry_v0_to_text(out, e);
 946         }
 947         pr_newline(out);
 948 }
 949
 950 const struct bch_sb_field_ops bch_sb_field_ops_replicas_v0 = {
 951         .validate       = bch2_sb_replicas_v0_validate,
 952         .to_text        = bch2_sb_replicas_v0_to_text,
 953 };
 954
 955 /* Query replicas: */
 956
 957 bool bch2_have_enough_devs(struct bch_fs *c, struct bch_devs_mask devs,
 958                            unsigned flags, bool print)
 959 {
 960         struct bch_replicas_entry *e;
 961         bool ret = true;
 962
 963         percpu_down_read(&c->mark_lock);
 964         for_each_cpu_replicas_entry(&c->replicas, e) {
 965                 unsigned i, nr_online = 0, nr_failed = 0, dflags = 0;
 966                 bool metadata = e->data_type < BCH_DATA_user;
 967
 968                 if (e->data_type == BCH_DATA_cached)
 969                         continue;
 970
 971                 for (i = 0; i < e->nr_devs; i++) {
 972                         struct bch_dev *ca = bch_dev_bkey_exists(c, e->devs[i]);
 973
 974                         nr_online += test_bit(e->devs[i], devs.d);
 975                         nr_failed += ca->mi.state == BCH_MEMBER_STATE_failed;
 976                 }
 977
 978                 if (nr_failed == e->nr_devs)
 979                         continue;
 980
 981                 if (nr_online < e->nr_required)
 982                         dflags |= metadata
 983                                 ? BCH_FORCE_IF_METADATA_LOST
 984                                 : BCH_FORCE_IF_DATA_LOST;
 985
 986                 if (nr_online < e->nr_devs)
 987                         dflags |= metadata
 988                                 ? BCH_FORCE_IF_METADATA_DEGRADED
 989                                 : BCH_FORCE_IF_DATA_DEGRADED;
 990
 991                 if (dflags & ~flags) {
 992                         if (print) {
 993                                 struct printbuf buf = PRINTBUF;
 994
 995                                 bch2_replicas_entry_to_text(&buf, e);
 996                                 bch_err(c, "insufficient devices online (%u) for replicas entry %s",
 997                                         nr_online, buf.buf);
 998                                 printbuf_exit(&buf);
 999                         }
1000                         ret = false;
1001                         break;
1002                 }
1003
1004         }
1005         percpu_up_read(&c->mark_lock);
1006
1007         return ret;
1008 }
1009
1010 unsigned bch2_sb_dev_has_data(struct bch_sb *sb, unsigned dev)
1011 {
1012         struct bch_sb_field_replicas *replicas;
1013         struct bch_sb_field_replicas_v0 *replicas_v0;
1014         unsigned i, data_has = 0;
1015
1016         replicas = bch2_sb_get_replicas(sb);
1017         replicas_v0 = bch2_sb_get_replicas_v0(sb);
1018
1019         if (replicas) {
1020                 struct bch_replicas_entry *r;
1021
1022                 for_each_replicas_entry(replicas, r)
1023                         for (i = 0; i < r->nr_devs; i++)
1024                                 if (r->devs[i] == dev)
1025                                         data_has |= 1 << r->data_type;
1026         } else if (replicas_v0) {
1027                 struct bch_replicas_entry_v0 *r;
1028
1029                 for_each_replicas_entry_v0(replicas_v0, r)
1030                         for (i = 0; i < r->nr_devs; i++)
1031                                 if (r->devs[i] == dev)
1032                                         data_has |= 1 << r->data_type;
1033         }
1034
1035
1036         return data_has;
1037 }
1038
1039 unsigned bch2_dev_has_data(struct bch_fs *c, struct bch_dev *ca)
1040 {
1041         unsigned ret;
1042
1043         mutex_lock(&c->sb_lock);
1044         ret = bch2_sb_dev_has_data(c->disk_sb.sb, ca->dev_idx);
1045         mutex_unlock(&c->sb_lock);
1046
1047         return ret;
1048 }
1049
1050 void bch2_fs_replicas_exit(struct bch_fs *c)
1051 {
1052         unsigned i;
1053
1054         kfree(c->usage_scratch);
1055         for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1056                 free_percpu(c->usage[i]);
1057         kfree(c->usage_base);
1058         kfree(c->replicas.entries);
1059         kfree(c->replicas_gc.entries);
1060
1061         mempool_exit(&c->replicas_delta_pool);
1062 }
1063
1064 int bch2_fs_replicas_init(struct bch_fs *c)
1065 {
1066         bch2_journal_entry_res_resize(&c->journal,
1067                         &c->replicas_journal_res,
1068                         reserve_journal_replicas(c, &c->replicas));
1069
1070         return mempool_init_kmalloc_pool(&c->replicas_delta_pool, 1,
1071                                          REPLICAS_DELTA_LIST_MAX) ?:
1072                 replicas_table_update(c, &c->replicas);
1073 }