fs/bcachefs/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 static void 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_to_text(struct printbuf *out,
  40                                  struct bch_replicas_entry *e)
  41 {
  42         unsigned i;
  43
  44         pr_buf(out, "%s: %u/%u [",
  45                bch2_data_types[e->data_type],
  46                e->nr_required,
  47                e->nr_devs);
  48
  49         for (i = 0; i < e->nr_devs; i++)
  50                 pr_buf(out, i ? " %u" : "%u", e->devs[i]);
  51         pr_buf(out, "]");
  52 }
  53
  54 void bch2_cpu_replicas_to_text(struct printbuf *out,
  55                               struct bch_replicas_cpu *r)
  56 {
  57         struct bch_replicas_entry *e;
  58         bool first = true;
  59
  60         for_each_cpu_replicas_entry(r, e) {
  61                 if (!first)
  62                         pr_buf(out, " ");
  63                 first = false;
  64
  65                 bch2_replicas_entry_to_text(out, e);
  66         }
  67 }
  68
  69 static void extent_to_replicas(struct bkey_s_c k,
  70                                struct bch_replicas_entry *r)
  71 {
  72         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
  73         const union bch_extent_entry *entry;
  74         struct extent_ptr_decoded p;
  75
  76         r->nr_required  = 1;
  77
  78         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
  79                 if (p.ptr.cached)
  80                         continue;
  81
  82                 if (!p.has_ec)
  83                         r->devs[r->nr_devs++] = p.ptr.dev;
  84                 else
  85                         r->nr_required = 0;
  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 void bch2_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         case KEY_TYPE_btree_ptr_v2:
 111                 e->data_type = BCH_DATA_btree;
 112                 extent_to_replicas(k, e);
 113                 break;
 114         case KEY_TYPE_extent:
 115         case KEY_TYPE_reflink_v:
 116                 e->data_type = BCH_DATA_user;
 117                 extent_to_replicas(k, e);
 118                 break;
 119         case KEY_TYPE_stripe:
 120                 e->data_type = BCH_DATA_parity;
 121                 stripe_to_replicas(k, e);
 122                 break;
 123         }
 124
 125         replicas_entry_sort(e);
 126 }
 127
 128 void bch2_devlist_to_replicas(struct bch_replicas_entry *e,
 129                               enum bch_data_type data_type,
 130                               struct bch_devs_list devs)
 131 {
 132         unsigned i;
 133
 134         BUG_ON(!data_type ||
 135                data_type == BCH_DATA_sb ||
 136                data_type >= BCH_DATA_NR);
 137
 138         e->data_type    = data_type;
 139         e->nr_devs      = 0;
 140         e->nr_required  = 1;
 141
 142         for (i = 0; i < devs.nr; i++)
 143                 e->devs[e->nr_devs++] = devs.devs[i];
 144
 145         replicas_entry_sort(e);
 146 }
 147
 148 static struct bch_replicas_cpu
 149 cpu_replicas_add_entry(struct bch_replicas_cpu *old,
 150                        struct bch_replicas_entry *new_entry)
 151 {
 152         unsigned i;
 153         struct bch_replicas_cpu new = {
 154                 .nr             = old->nr + 1,
 155                 .entry_size     = max_t(unsigned, old->entry_size,
 156                                         replicas_entry_bytes(new_entry)),
 157         };
 158
 159         BUG_ON(!new_entry->data_type);
 160         verify_replicas_entry(new_entry);
 161
 162         new.entries = kcalloc(new.nr, new.entry_size, GFP_NOIO);
 163         if (!new.entries)
 164                 return new;
 165
 166         for (i = 0; i < old->nr; i++)
 167                 memcpy(cpu_replicas_entry(&new, i),
 168                        cpu_replicas_entry(old, i),
 169                        old->entry_size);
 170
 171         memcpy(cpu_replicas_entry(&new, old->nr),
 172                new_entry,
 173                replicas_entry_bytes(new_entry));
 174
 175         bch2_cpu_replicas_sort(&new);
 176         return new;
 177 }
 178
 179 static inline int __replicas_entry_idx(struct bch_replicas_cpu *r,
 180                                        struct bch_replicas_entry *search)
 181 {
 182         int idx, entry_size = replicas_entry_bytes(search);
 183
 184         if (unlikely(entry_size > r->entry_size))
 185                 return -1;
 186
 187         verify_replicas_entry(search);
 188
 189 #define entry_cmp(_l, _r, size) memcmp(_l, _r, entry_size)
 190         idx = eytzinger0_find(r->entries, r->nr, r->entry_size,
 191                               entry_cmp, search);
 192 #undef entry_cmp
 193
 194         return idx < r->nr ? idx : -1;
 195 }
 196
 197 int bch2_replicas_entry_idx(struct bch_fs *c,
 198                             struct bch_replicas_entry *search)
 199 {
 200         replicas_entry_sort(search);
 201
 202         return __replicas_entry_idx(&c->replicas, search);
 203 }
 204
 205 static bool __replicas_has_entry(struct bch_replicas_cpu *r,
 206                                  struct bch_replicas_entry *search)
 207 {
 208         return __replicas_entry_idx(r, search) >= 0;
 209 }
 210
 211 bool bch2_replicas_marked(struct bch_fs *c,
 212                           struct bch_replicas_entry *search)
 213 {
 214         bool marked;
 215
 216         if (!search->nr_devs)
 217                 return true;
 218
 219         verify_replicas_entry(search);
 220
 221         percpu_down_read(&c->mark_lock);
 222         marked = __replicas_has_entry(&c->replicas, search) &&
 223                 (likely((!c->replicas_gc.entries)) ||
 224                  __replicas_has_entry(&c->replicas_gc, search));
 225         percpu_up_read(&c->mark_lock);
 226
 227         return marked;
 228 }
 229
 230 static void __replicas_table_update(struct bch_fs_usage *dst,
 231                                     struct bch_replicas_cpu *dst_r,
 232                                     struct bch_fs_usage *src,
 233                                     struct bch_replicas_cpu *src_r)
 234 {
 235         int src_idx, dst_idx;
 236
 237         *dst = *src;
 238
 239         for (src_idx = 0; src_idx < src_r->nr; src_idx++) {
 240                 if (!src->replicas[src_idx])
 241                         continue;
 242
 243                 dst_idx = __replicas_entry_idx(dst_r,
 244                                 cpu_replicas_entry(src_r, src_idx));
 245                 BUG_ON(dst_idx < 0);
 246
 247                 dst->replicas[dst_idx] = src->replicas[src_idx];
 248         }
 249 }
 250
 251 static void __replicas_table_update_pcpu(struct bch_fs_usage __percpu *dst_p,
 252                                     struct bch_replicas_cpu *dst_r,
 253                                     struct bch_fs_usage __percpu *src_p,
 254                                     struct bch_replicas_cpu *src_r)
 255 {
 256         unsigned src_nr = sizeof(struct bch_fs_usage) / sizeof(u64) + src_r->nr;
 257         struct bch_fs_usage *dst, *src = (void *)
 258                 bch2_acc_percpu_u64s((void *) src_p, src_nr);
 259
 260         preempt_disable();
 261         dst = this_cpu_ptr(dst_p);
 262         preempt_enable();
 263
 264         __replicas_table_update(dst, dst_r, src, src_r);
 265 }
 266
 267 /*
 268  * Resize filesystem accounting:
 269  */
 270 static int replicas_table_update(struct bch_fs *c,
 271                                  struct bch_replicas_cpu *new_r)
 272 {
 273         struct bch_fs_usage __percpu *new_usage[JOURNAL_BUF_NR];
 274         struct bch_fs_usage_online *new_scratch = NULL;
 275         struct bch_fs_usage __percpu *new_gc = NULL;
 276         struct bch_fs_usage *new_base = NULL;
 277         unsigned i, bytes = sizeof(struct bch_fs_usage) +
 278                 sizeof(u64) * new_r->nr;
 279         unsigned scratch_bytes = sizeof(struct bch_fs_usage_online) +
 280                 sizeof(u64) * new_r->nr;
 281         int ret = 0;
 282
 283         memset(new_usage, 0, sizeof(new_usage));
 284
 285         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 286                 if (!(new_usage[i] = __alloc_percpu_gfp(bytes,
 287                                         sizeof(u64), GFP_NOIO)))
 288                         goto err;
 289
 290         memset(new_usage, 0, sizeof(new_usage));
 291
 292         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 293                 if (!(new_usage[i] = __alloc_percpu_gfp(bytes,
 294                                         sizeof(u64), GFP_NOIO)))
 295                         goto err;
 296
 297         if (!(new_base = kzalloc(bytes, GFP_NOIO)) ||
 298             !(new_scratch  = kmalloc(scratch_bytes, GFP_NOIO)) ||
 299             (c->usage_gc &&
 300              !(new_gc = __alloc_percpu_gfp(bytes, sizeof(u64), GFP_NOIO))))
 301                 goto err;
 302
 303         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 304                 if (c->usage[i])
 305                         __replicas_table_update_pcpu(new_usage[i], new_r,
 306                                                      c->usage[i], &c->replicas);
 307         if (c->usage_base)
 308                 __replicas_table_update(new_base,               new_r,
 309                                         c->usage_base,          &c->replicas);
 310         if (c->usage_gc)
 311                 __replicas_table_update_pcpu(new_gc,            new_r,
 312                                              c->usage_gc,       &c->replicas);
 313
 314         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 315                 swap(c->usage[i],       new_usage[i]);
 316         swap(c->usage_base,     new_base);
 317         swap(c->usage_scratch,  new_scratch);
 318         swap(c->usage_gc,       new_gc);
 319         swap(c->replicas,       *new_r);
 320 out:
 321         free_percpu(new_gc);
 322         kfree(new_scratch);
 323         free_percpu(new_usage[1]);
 324         free_percpu(new_usage[0]);
 325         kfree(new_base);
 326         return ret;
 327 err:
 328         bch_err(c, "error updating replicas table: memory allocation failure");
 329         ret = -ENOMEM;
 330         goto out;
 331 }
 332
 333 static unsigned reserve_journal_replicas(struct bch_fs *c,
 334                                      struct bch_replicas_cpu *r)
 335 {
 336         struct bch_replicas_entry *e;
 337         unsigned journal_res_u64s = 0;
 338
 339         /* nr_inodes: */
 340         journal_res_u64s +=
 341                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 342
 343         /* key_version: */
 344         journal_res_u64s +=
 345                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 346
 347         /* persistent_reserved: */
 348         journal_res_u64s +=
 349                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64)) *
 350                 BCH_REPLICAS_MAX;
 351
 352         for_each_cpu_replicas_entry(r, e)
 353                 journal_res_u64s +=
 354                         DIV_ROUND_UP(sizeof(struct jset_entry_data_usage) +
 355                                      e->nr_devs, sizeof(u64));
 356         return journal_res_u64s;
 357 }
 358
 359 noinline
 360 static int bch2_mark_replicas_slowpath(struct bch_fs *c,
 361                                 struct bch_replicas_entry *new_entry)
 362 {
 363         struct bch_replicas_cpu new_r, new_gc;
 364         int ret = 0;
 365
 366         verify_replicas_entry(new_entry);
 367
 368         memset(&new_r, 0, sizeof(new_r));
 369         memset(&new_gc, 0, sizeof(new_gc));
 370
 371         mutex_lock(&c->sb_lock);
 372
 373         if (c->replicas_gc.entries &&
 374             !__replicas_has_entry(&c->replicas_gc, new_entry)) {
 375                 new_gc = cpu_replicas_add_entry(&c->replicas_gc, new_entry);
 376                 if (!new_gc.entries)
 377                         goto err;
 378         }
 379
 380         if (!__replicas_has_entry(&c->replicas, new_entry)) {
 381                 new_r = cpu_replicas_add_entry(&c->replicas, new_entry);
 382                 if (!new_r.entries)
 383                         goto err;
 384
 385                 ret = bch2_cpu_replicas_to_sb_replicas(c, &new_r);
 386                 if (ret)
 387                         goto err;
 388
 389                 bch2_journal_entry_res_resize(&c->journal,
 390                                 &c->replicas_journal_res,
 391                                 reserve_journal_replicas(c, &new_r));
 392         }
 393
 394         if (!new_r.entries &&
 395             !new_gc.entries)
 396                 goto out;
 397
 398         /* allocations done, now commit: */
 399
 400         if (new_r.entries)
 401                 bch2_write_super(c);
 402
 403         /* don't update in memory replicas until changes are persistent */
 404         percpu_down_write(&c->mark_lock);
 405         if (new_r.entries)
 406                 ret = replicas_table_update(c, &new_r);
 407         if (new_gc.entries)
 408                 swap(new_gc, c->replicas_gc);
 409         percpu_up_write(&c->mark_lock);
 410 out:
 411         mutex_unlock(&c->sb_lock);
 412
 413         kfree(new_r.entries);
 414         kfree(new_gc.entries);
 415
 416         return ret;
 417 err:
 418         bch_err(c, "error adding replicas entry: memory allocation failure");
 419         ret = -ENOMEM;
 420         goto out;
 421 }
 422
 423 static int __bch2_mark_replicas(struct bch_fs *c,
 424                                 struct bch_replicas_entry *r,
 425                                 bool check)
 426 {
 427         return likely(bch2_replicas_marked(c, r))       ? 0
 428                 : check                                 ? -1
 429                 : bch2_mark_replicas_slowpath(c, r);
 430 }
 431
 432 int bch2_mark_replicas(struct bch_fs *c, struct bch_replicas_entry *r)
 433 {
 434         return __bch2_mark_replicas(c, r, false);
 435 }
 436
 437 static int __bch2_mark_bkey_replicas(struct bch_fs *c, struct bkey_s_c k,
 438                                      bool check)
 439 {
 440         struct bch_replicas_padded search;
 441         struct bch_devs_list cached = bch2_bkey_cached_devs(k);
 442         unsigned i;
 443         int ret;
 444
 445         for (i = 0; i < cached.nr; i++) {
 446                 bch2_replicas_entry_cached(&search.e, cached.devs[i]);
 447
 448                 ret = __bch2_mark_replicas(c, &search.e, check);
 449                 if (ret)
 450                         return ret;
 451         }
 452
 453         bch2_bkey_to_replicas(&search.e, k);
 454
 455         ret = __bch2_mark_replicas(c, &search.e, check);
 456         if (ret)
 457                 return ret;
 458
 459         if (search.e.data_type == BCH_DATA_parity) {
 460                 search.e.data_type = BCH_DATA_cached;
 461                 ret = __bch2_mark_replicas(c, &search.e, check);
 462                 if (ret)
 463                         return ret;
 464
 465                 search.e.data_type = BCH_DATA_user;
 466                 ret = __bch2_mark_replicas(c, &search.e, check);
 467                 if (ret)
 468                         return ret;
 469         }
 470
 471         return 0;
 472 }
 473
 474 bool bch2_bkey_replicas_marked(struct bch_fs *c,
 475                                struct bkey_s_c k)
 476 {
 477         return __bch2_mark_bkey_replicas(c, k, true) == 0;
 478 }
 479
 480 int bch2_mark_bkey_replicas(struct bch_fs *c, struct bkey_s_c k)
 481 {
 482         return __bch2_mark_bkey_replicas(c, k, false);
 483 }
 484
 485 int bch2_replicas_gc_end(struct bch_fs *c, int ret)
 486 {
 487         unsigned i;
 488
 489         lockdep_assert_held(&c->replicas_gc_lock);
 490
 491         mutex_lock(&c->sb_lock);
 492         percpu_down_write(&c->mark_lock);
 493
 494         /*
 495          * this is kind of crappy; the replicas gc mechanism needs to be ripped
 496          * out
 497          */
 498
 499         for (i = 0; i < c->replicas.nr; i++) {
 500                 struct bch_replicas_entry *e =
 501                         cpu_replicas_entry(&c->replicas, i);
 502                 struct bch_replicas_cpu n;
 503
 504                 if (!__replicas_has_entry(&c->replicas_gc, e) &&
 505                     bch2_fs_usage_read_one(c, &c->usage_base->replicas[i])) {
 506                         n = cpu_replicas_add_entry(&c->replicas_gc, e);
 507                         if (!n.entries) {
 508                                 ret = -ENOSPC;
 509                                 goto err;
 510                         }
 511
 512                         swap(n, c->replicas_gc);
 513                         kfree(n.entries);
 514                 }
 515         }
 516
 517         if (bch2_cpu_replicas_to_sb_replicas(c, &c->replicas_gc)) {
 518                 ret = -ENOSPC;
 519                 goto err;
 520         }
 521
 522         ret = replicas_table_update(c, &c->replicas_gc);
 523 err:
 524         kfree(c->replicas_gc.entries);
 525         c->replicas_gc.entries = NULL;
 526
 527         percpu_up_write(&c->mark_lock);
 528
 529         if (!ret)
 530                 bch2_write_super(c);
 531
 532         mutex_unlock(&c->sb_lock);
 533
 534         return ret;
 535 }
 536
 537 int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask)
 538 {
 539         struct bch_replicas_entry *e;
 540         unsigned i = 0;
 541
 542         lockdep_assert_held(&c->replicas_gc_lock);
 543
 544         mutex_lock(&c->sb_lock);
 545         BUG_ON(c->replicas_gc.entries);
 546
 547         c->replicas_gc.nr               = 0;
 548         c->replicas_gc.entry_size       = 0;
 549
 550         for_each_cpu_replicas_entry(&c->replicas, e)
 551                 if (!((1 << e->data_type) & typemask)) {
 552                         c->replicas_gc.nr++;
 553                         c->replicas_gc.entry_size =
 554                                 max_t(unsigned, c->replicas_gc.entry_size,
 555                                       replicas_entry_bytes(e));
 556                 }
 557
 558         c->replicas_gc.entries = kcalloc(c->replicas_gc.nr,
 559                                          c->replicas_gc.entry_size,
 560                                          GFP_NOIO);
 561         if (!c->replicas_gc.entries) {
 562                 mutex_unlock(&c->sb_lock);
 563                 bch_err(c, "error allocating c->replicas_gc");
 564                 return -ENOMEM;
 565         }
 566
 567         for_each_cpu_replicas_entry(&c->replicas, e)
 568                 if (!((1 << e->data_type) & typemask))
 569                         memcpy(cpu_replicas_entry(&c->replicas_gc, i++),
 570                                e, c->replicas_gc.entry_size);
 571
 572         bch2_cpu_replicas_sort(&c->replicas_gc);
 573         mutex_unlock(&c->sb_lock);
 574
 575         return 0;
 576 }
 577
 578 int bch2_replicas_gc2(struct bch_fs *c)
 579 {
 580         struct bch_replicas_cpu new = { 0 };
 581         unsigned i, nr;
 582         int ret = 0;
 583
 584         bch2_journal_meta(&c->journal);
 585 retry:
 586         nr              = READ_ONCE(c->replicas.nr);
 587         new.entry_size  = READ_ONCE(c->replicas.entry_size);
 588         new.entries     = kcalloc(nr, new.entry_size, GFP_KERNEL);
 589         if (!new.entries) {
 590                 bch_err(c, "error allocating c->replicas_gc");
 591                 return -ENOMEM;
 592         }
 593
 594         mutex_lock(&c->sb_lock);
 595         percpu_down_write(&c->mark_lock);
 596
 597         if (nr                  != c->replicas.nr ||
 598             new.entry_size      != c->replicas.entry_size) {
 599                 percpu_up_write(&c->mark_lock);
 600                 mutex_unlock(&c->sb_lock);
 601                 kfree(new.entries);
 602                 goto retry;
 603         }
 604
 605         for (i = 0; i < c->replicas.nr; i++) {
 606                 struct bch_replicas_entry *e =
 607                         cpu_replicas_entry(&c->replicas, i);
 608
 609                 if (e->data_type == BCH_DATA_journal ||
 610                     c->usage_base->replicas[i] ||
 611                     percpu_u64_get(&c->usage[0]->replicas[i]) ||
 612                     percpu_u64_get(&c->usage[1]->replicas[i]) ||
 613                     percpu_u64_get(&c->usage[2]->replicas[i]) ||
 614                     percpu_u64_get(&c->usage[3]->replicas[i]))
 615                         memcpy(cpu_replicas_entry(&new, new.nr++),
 616                                e, new.entry_size);
 617         }
 618
 619         bch2_cpu_replicas_sort(&new);
 620
 621         if (bch2_cpu_replicas_to_sb_replicas(c, &new)) {
 622                 ret = -ENOSPC;
 623                 goto err;
 624         }
 625
 626         ret = replicas_table_update(c, &new);
 627 err:
 628         kfree(new.entries);
 629
 630         percpu_up_write(&c->mark_lock);
 631
 632         if (!ret)
 633                 bch2_write_super(c);
 634
 635         mutex_unlock(&c->sb_lock);
 636
 637         return ret;
 638 }
 639
 640 int bch2_replicas_set_usage(struct bch_fs *c,
 641                             struct bch_replicas_entry *r,
 642                             u64 sectors)
 643 {
 644         int ret, idx = bch2_replicas_entry_idx(c, r);
 645
 646         if (idx < 0) {
 647                 struct bch_replicas_cpu n;
 648
 649                 n = cpu_replicas_add_entry(&c->replicas, r);
 650                 if (!n.entries)
 651                         return -ENOMEM;
 652
 653                 ret = replicas_table_update(c, &n);
 654                 if (ret)
 655                         return ret;
 656
 657                 kfree(n.entries);
 658
 659                 idx = bch2_replicas_entry_idx(c, r);
 660                 BUG_ON(ret < 0);
 661         }
 662
 663         c->usage_base->replicas[idx] = sectors;
 664
 665         return 0;
 666 }
 667
 668 /* Replicas tracking - superblock: */
 669
 670 static int
 671 __bch2_sb_replicas_to_cpu_replicas(struct bch_sb_field_replicas *sb_r,
 672                                    struct bch_replicas_cpu *cpu_r)
 673 {
 674         struct bch_replicas_entry *e, *dst;
 675         unsigned nr = 0, entry_size = 0, idx = 0;
 676
 677         for_each_replicas_entry(sb_r, e) {
 678                 entry_size = max_t(unsigned, entry_size,
 679                                    replicas_entry_bytes(e));
 680                 nr++;
 681         }
 682
 683         cpu_r->entries = kcalloc(nr, entry_size, GFP_NOIO);
 684         if (!cpu_r->entries)
 685                 return -ENOMEM;
 686
 687         cpu_r->nr               = nr;
 688         cpu_r->entry_size       = entry_size;
 689
 690         for_each_replicas_entry(sb_r, e) {
 691                 dst = cpu_replicas_entry(cpu_r, idx++);
 692                 memcpy(dst, e, replicas_entry_bytes(e));
 693                 replicas_entry_sort(dst);
 694         }
 695
 696         return 0;
 697 }
 698
 699 static int
 700 __bch2_sb_replicas_v0_to_cpu_replicas(struct bch_sb_field_replicas_v0 *sb_r,
 701                                       struct bch_replicas_cpu *cpu_r)
 702 {
 703         struct bch_replicas_entry_v0 *e;
 704         unsigned nr = 0, entry_size = 0, idx = 0;
 705
 706         for_each_replicas_entry(sb_r, e) {
 707                 entry_size = max_t(unsigned, entry_size,
 708                                    replicas_entry_bytes(e));
 709                 nr++;
 710         }
 711
 712         entry_size += sizeof(struct bch_replicas_entry) -
 713                 sizeof(struct bch_replicas_entry_v0);
 714
 715         cpu_r->entries = kcalloc(nr, entry_size, GFP_NOIO);
 716         if (!cpu_r->entries)
 717                 return -ENOMEM;
 718
 719         cpu_r->nr               = nr;
 720         cpu_r->entry_size       = entry_size;
 721
 722         for_each_replicas_entry(sb_r, e) {
 723                 struct bch_replicas_entry *dst =
 724                         cpu_replicas_entry(cpu_r, idx++);
 725
 726                 dst->data_type  = e->data_type;
 727                 dst->nr_devs    = e->nr_devs;
 728                 dst->nr_required = 1;
 729                 memcpy(dst->devs, e->devs, e->nr_devs);
 730                 replicas_entry_sort(dst);
 731         }
 732
 733         return 0;
 734 }
 735
 736 int bch2_sb_replicas_to_cpu_replicas(struct bch_fs *c)
 737 {
 738         struct bch_sb_field_replicas *sb_v1;
 739         struct bch_sb_field_replicas_v0 *sb_v0;
 740         struct bch_replicas_cpu new_r = { 0, 0, NULL };
 741         int ret = 0;
 742
 743         if ((sb_v1 = bch2_sb_get_replicas(c->disk_sb.sb)))
 744                 ret = __bch2_sb_replicas_to_cpu_replicas(sb_v1, &new_r);
 745         else if ((sb_v0 = bch2_sb_get_replicas_v0(c->disk_sb.sb)))
 746                 ret = __bch2_sb_replicas_v0_to_cpu_replicas(sb_v0, &new_r);
 747
 748         if (ret)
 749                 return -ENOMEM;
 750
 751         bch2_cpu_replicas_sort(&new_r);
 752
 753         percpu_down_write(&c->mark_lock);
 754
 755         ret = replicas_table_update(c, &new_r);
 756         percpu_up_write(&c->mark_lock);
 757
 758         kfree(new_r.entries);
 759
 760         return 0;
 761 }
 762
 763 static int bch2_cpu_replicas_to_sb_replicas_v0(struct bch_fs *c,
 764                                                struct bch_replicas_cpu *r)
 765 {
 766         struct bch_sb_field_replicas_v0 *sb_r;
 767         struct bch_replicas_entry_v0 *dst;
 768         struct bch_replicas_entry *src;
 769         size_t bytes;
 770
 771         bytes = sizeof(struct bch_sb_field_replicas);
 772
 773         for_each_cpu_replicas_entry(r, src)
 774                 bytes += replicas_entry_bytes(src) - 1;
 775
 776         sb_r = bch2_sb_resize_replicas_v0(&c->disk_sb,
 777                         DIV_ROUND_UP(bytes, sizeof(u64)));
 778         if (!sb_r)
 779                 return -ENOSPC;
 780
 781         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas);
 782         sb_r = bch2_sb_get_replicas_v0(c->disk_sb.sb);
 783
 784         memset(&sb_r->entries, 0,
 785                vstruct_end(&sb_r->field) -
 786                (void *) &sb_r->entries);
 787
 788         dst = sb_r->entries;
 789         for_each_cpu_replicas_entry(r, src) {
 790                 dst->data_type  = src->data_type;
 791                 dst->nr_devs    = src->nr_devs;
 792                 memcpy(dst->devs, src->devs, src->nr_devs);
 793
 794                 dst = replicas_entry_next(dst);
 795
 796                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 797         }
 798
 799         return 0;
 800 }
 801
 802 static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *c,
 803                                             struct bch_replicas_cpu *r)
 804 {
 805         struct bch_sb_field_replicas *sb_r;
 806         struct bch_replicas_entry *dst, *src;
 807         bool need_v1 = false;
 808         size_t bytes;
 809
 810         bytes = sizeof(struct bch_sb_field_replicas);
 811
 812         for_each_cpu_replicas_entry(r, src) {
 813                 bytes += replicas_entry_bytes(src);
 814                 if (src->nr_required != 1)
 815                         need_v1 = true;
 816         }
 817
 818         if (!need_v1)
 819                 return bch2_cpu_replicas_to_sb_replicas_v0(c, r);
 820
 821         sb_r = bch2_sb_resize_replicas(&c->disk_sb,
 822                         DIV_ROUND_UP(bytes, sizeof(u64)));
 823         if (!sb_r)
 824                 return -ENOSPC;
 825
 826         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas_v0);
 827         sb_r = bch2_sb_get_replicas(c->disk_sb.sb);
 828
 829         memset(&sb_r->entries, 0,
 830                vstruct_end(&sb_r->field) -
 831                (void *) &sb_r->entries);
 832
 833         dst = sb_r->entries;
 834         for_each_cpu_replicas_entry(r, src) {
 835                 memcpy(dst, src, replicas_entry_bytes(src));
 836
 837                 dst = replicas_entry_next(dst);
 838
 839                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 840         }
 841
 842         return 0;
 843 }
 844
 845 static const char *check_dup_replicas_entries(struct bch_replicas_cpu *cpu_r)
 846 {
 847         unsigned i;
 848
 849         sort_cmp_size(cpu_r->entries,
 850                       cpu_r->nr,
 851                       cpu_r->entry_size,
 852                       memcmp, NULL);
 853
 854         for (i = 0; i + 1 < cpu_r->nr; i++) {
 855                 struct bch_replicas_entry *l =
 856                         cpu_replicas_entry(cpu_r, i);
 857                 struct bch_replicas_entry *r =
 858                         cpu_replicas_entry(cpu_r, i + 1);
 859
 860                 BUG_ON(memcmp(l, r, cpu_r->entry_size) > 0);
 861
 862                 if (!memcmp(l, r, cpu_r->entry_size))
 863                         return "duplicate replicas entry";
 864         }
 865
 866         return NULL;
 867 }
 868
 869 static const char *bch2_sb_validate_replicas(struct bch_sb *sb, struct bch_sb_field *f)
 870 {
 871         struct bch_sb_field_replicas *sb_r = field_to_type(f, replicas);
 872         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 873         struct bch_replicas_cpu cpu_r = { .entries = NULL };
 874         struct bch_replicas_entry *e;
 875         const char *err;
 876         unsigned i;
 877
 878         for_each_replicas_entry(sb_r, e) {
 879                 err = "invalid replicas entry: invalid data type";
 880                 if (e->data_type >= BCH_DATA_NR)
 881                         goto err;
 882
 883                 err = "invalid replicas entry: no devices";
 884                 if (!e->nr_devs)
 885                         goto err;
 886
 887                 err = "invalid replicas entry: bad nr_required";
 888                 if (e->nr_required > 1 &&
 889                     e->nr_required >= e->nr_devs)
 890                         goto err;
 891
 892                 err = "invalid replicas entry: invalid device";
 893                 for (i = 0; i < e->nr_devs; i++)
 894                         if (!bch2_dev_exists(sb, mi, e->devs[i]))
 895                                 goto err;
 896         }
 897
 898         err = "cannot allocate memory";
 899         if (__bch2_sb_replicas_to_cpu_replicas(sb_r, &cpu_r))
 900                 goto err;
 901
 902         err = check_dup_replicas_entries(&cpu_r);
 903 err:
 904         kfree(cpu_r.entries);
 905         return err;
 906 }
 907
 908 static void bch2_sb_replicas_to_text(struct printbuf *out,
 909                                      struct bch_sb *sb,
 910                                      struct bch_sb_field *f)
 911 {
 912         struct bch_sb_field_replicas *r = field_to_type(f, replicas);
 913         struct bch_replicas_entry *e;
 914         bool first = true;
 915
 916         for_each_replicas_entry(r, e) {
 917                 if (!first)
 918                         pr_buf(out, " ");
 919                 first = false;
 920
 921                 bch2_replicas_entry_to_text(out, e);
 922         }
 923 }
 924
 925 const struct bch_sb_field_ops bch_sb_field_ops_replicas = {
 926         .validate       = bch2_sb_validate_replicas,
 927         .to_text        = bch2_sb_replicas_to_text,
 928 };
 929
 930 static const char *bch2_sb_validate_replicas_v0(struct bch_sb *sb, struct bch_sb_field *f)
 931 {
 932         struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0);
 933         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 934         struct bch_replicas_cpu cpu_r = { .entries = NULL };
 935         struct bch_replicas_entry_v0 *e;
 936         const char *err;
 937         unsigned i;
 938
 939         for_each_replicas_entry_v0(sb_r, e) {
 940                 err = "invalid replicas entry: invalid data type";
 941                 if (e->data_type >= BCH_DATA_NR)
 942                         goto err;
 943
 944                 err = "invalid replicas entry: no devices";
 945                 if (!e->nr_devs)
 946                         goto err;
 947
 948                 err = "invalid replicas entry: invalid device";
 949                 for (i = 0; i < e->nr_devs; i++)
 950                         if (!bch2_dev_exists(sb, mi, e->devs[i]))
 951                                 goto err;
 952         }
 953
 954         err = "cannot allocate memory";
 955         if (__bch2_sb_replicas_v0_to_cpu_replicas(sb_r, &cpu_r))
 956                 goto err;
 957
 958         err = check_dup_replicas_entries(&cpu_r);
 959 err:
 960         kfree(cpu_r.entries);
 961         return err;
 962 }
 963
 964 const struct bch_sb_field_ops bch_sb_field_ops_replicas_v0 = {
 965         .validate       = bch2_sb_validate_replicas_v0,
 966 };
 967
 968 /* Query replicas: */
 969
 970 struct replicas_status __bch2_replicas_status(struct bch_fs *c,
 971                                               struct bch_devs_mask online_devs)
 972 {
 973         struct bch_sb_field_members *mi;
 974         struct bch_replicas_entry *e;
 975         unsigned i, nr_online, nr_offline;
 976         struct replicas_status ret;
 977
 978         memset(&ret, 0, sizeof(ret));
 979
 980         for (i = 0; i < ARRAY_SIZE(ret.replicas); i++)
 981                 ret.replicas[i].redundancy = INT_MAX;
 982
 983         mi = bch2_sb_get_members(c->disk_sb.sb);
 984
 985         percpu_down_read(&c->mark_lock);
 986
 987         for_each_cpu_replicas_entry(&c->replicas, e) {
 988                 if (e->data_type >= ARRAY_SIZE(ret.replicas))
 989                         panic("e %p data_type %u\n", e, e->data_type);
 990
 991                 nr_online = nr_offline = 0;
 992
 993                 for (i = 0; i < e->nr_devs; i++) {
 994                         BUG_ON(!bch2_dev_exists(c->disk_sb.sb, mi,
 995                                                 e->devs[i]));
 996
 997                         if (test_bit(e->devs[i], online_devs.d))
 998                                 nr_online++;
 999                         else
1000                                 nr_offline++;
1001                 }
1002
1003                 ret.replicas[e->data_type].redundancy =
1004                         min(ret.replicas[e->data_type].redundancy,
1005                             (int) nr_online - (int) e->nr_required);
1006
1007                 ret.replicas[e->data_type].nr_offline =
1008                         max(ret.replicas[e->data_type].nr_offline,
1009                             nr_offline);
1010         }
1011
1012         percpu_up_read(&c->mark_lock);
1013
1014         for (i = 0; i < ARRAY_SIZE(ret.replicas); i++)
1015                 if (ret.replicas[i].redundancy == INT_MAX)
1016                         ret.replicas[i].redundancy = 0;
1017
1018         return ret;
1019 }
1020
1021 struct replicas_status bch2_replicas_status(struct bch_fs *c)
1022 {
1023         return __bch2_replicas_status(c, bch2_online_devs(c));
1024 }
1025
1026 static bool have_enough_devs(struct replicas_status s,
1027                              enum bch_data_type type,
1028                              bool force_if_degraded,
1029                              bool force_if_lost)
1030 {
1031         return (!s.replicas[type].nr_offline || force_if_degraded) &&
1032                 (s.replicas[type].redundancy >= 0 || force_if_lost);
1033 }
1034
1035 bool bch2_have_enough_devs(struct replicas_status s, unsigned flags)
1036 {
1037         return (have_enough_devs(s, BCH_DATA_journal,
1038                                  flags & BCH_FORCE_IF_METADATA_DEGRADED,
1039                                  flags & BCH_FORCE_IF_METADATA_LOST) &&
1040                 have_enough_devs(s, BCH_DATA_btree,
1041                                  flags & BCH_FORCE_IF_METADATA_DEGRADED,
1042                                  flags & BCH_FORCE_IF_METADATA_LOST) &&
1043                 have_enough_devs(s, BCH_DATA_user,
1044                                  flags & BCH_FORCE_IF_DATA_DEGRADED,
1045                                  flags & BCH_FORCE_IF_DATA_LOST));
1046 }
1047
1048 int bch2_replicas_online(struct bch_fs *c, bool meta)
1049 {
1050         struct replicas_status s = bch2_replicas_status(c);
1051
1052         return (meta
1053                 ? min(s.replicas[BCH_DATA_journal].redundancy,
1054                       s.replicas[BCH_DATA_btree].redundancy)
1055                 : s.replicas[BCH_DATA_user].redundancy) + 1;
1056 }
1057
1058 unsigned bch2_dev_has_data(struct bch_fs *c, struct bch_dev *ca)
1059 {
1060         struct bch_replicas_entry *e;
1061         unsigned i, ret = 0;
1062
1063         percpu_down_read(&c->mark_lock);
1064
1065         for_each_cpu_replicas_entry(&c->replicas, e)
1066                 for (i = 0; i < e->nr_devs; i++)
1067                         if (e->devs[i] == ca->dev_idx)
1068                                 ret |= 1 << e->data_type;
1069
1070         percpu_up_read(&c->mark_lock);
1071
1072         return ret;
1073 }
1074
1075 int bch2_fs_replicas_init(struct bch_fs *c)
1076 {
1077         c->journal.entry_u64s_reserved +=
1078                 reserve_journal_replicas(c, &c->replicas);
1079
1080         return replicas_table_update(c, &c->replicas);
1081 }