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 inline int u8_cmp(u8 l, u8 r)
  15 {
  16         return cmp_int(l, r);
  17 }
  18
  19 static void verify_replicas_entry(struct bch_replicas_entry *e)
  20 {
  21 #ifdef CONFIG_BCACHEFS_DEBUG
  22         unsigned i;
  23
  24         BUG_ON(e->data_type >= BCH_DATA_NR);
  25         BUG_ON(!e->nr_devs);
  26         BUG_ON(e->nr_required > 1 &&
  27                e->nr_required >= e->nr_devs);
  28
  29         for (i = 0; i + 1 < e->nr_devs; i++)
  30                 BUG_ON(e->devs[i] >= e->devs[i + 1]);
  31 #endif
  32 }
  33
  34 static void replicas_entry_sort(struct bch_replicas_entry *e)
  35 {
  36         bubble_sort(e->devs, e->nr_devs, u8_cmp);
  37 }
  38
  39 static void bch2_cpu_replicas_sort(struct bch_replicas_cpu *r)
  40 {
  41         eytzinger0_sort(r->entries, r->nr, r->entry_size, memcmp, NULL);
  42 }
  43
  44 void bch2_replicas_entry_to_text(struct printbuf *out,
  45                                  struct bch_replicas_entry *e)
  46 {
  47         unsigned i;
  48
  49         pr_buf(out, "%s: %u/%u [",
  50                bch2_data_types[e->data_type],
  51                e->nr_required,
  52                e->nr_devs);
  53
  54         for (i = 0; i < e->nr_devs; i++)
  55                 pr_buf(out, i ? " %u" : "%u", e->devs[i]);
  56         pr_buf(out, "]");
  57 }
  58
  59 void bch2_cpu_replicas_to_text(struct printbuf *out,
  60                               struct bch_replicas_cpu *r)
  61 {
  62         struct bch_replicas_entry *e;
  63         bool first = true;
  64
  65         for_each_cpu_replicas_entry(r, e) {
  66                 if (!first)
  67                         pr_buf(out, " ");
  68                 first = false;
  69
  70                 bch2_replicas_entry_to_text(out, e);
  71         }
  72 }
  73
  74 static void extent_to_replicas(struct bkey_s_c k,
  75                                struct bch_replicas_entry *r)
  76 {
  77         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
  78         const union bch_extent_entry *entry;
  79         struct extent_ptr_decoded p;
  80
  81         r->nr_required  = 1;
  82
  83         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
  84                 if (p.ptr.cached)
  85                         continue;
  86
  87                 if (!p.has_ec)
  88                         r->devs[r->nr_devs++] = p.ptr.dev;
  89                 else
  90                         r->nr_required = 0;
  91         }
  92 }
  93
  94 static void stripe_to_replicas(struct bkey_s_c k,
  95                                struct bch_replicas_entry *r)
  96 {
  97         struct bkey_s_c_stripe s = bkey_s_c_to_stripe(k);
  98         const struct bch_extent_ptr *ptr;
  99
 100         r->nr_required  = s.v->nr_blocks - s.v->nr_redundant;
 101
 102         for (ptr = s.v->ptrs;
 103              ptr < s.v->ptrs + s.v->nr_blocks;
 104              ptr++)
 105                 r->devs[r->nr_devs++] = ptr->dev;
 106 }
 107
 108 void bch2_bkey_to_replicas(struct bch_replicas_entry *e,
 109                            struct bkey_s_c k)
 110 {
 111         e->nr_devs = 0;
 112
 113         switch (k.k->type) {
 114         case KEY_TYPE_btree_ptr:
 115         case KEY_TYPE_btree_ptr_v2:
 116                 e->data_type = BCH_DATA_BTREE;
 117                 extent_to_replicas(k, e);
 118                 break;
 119         case KEY_TYPE_extent:
 120         case KEY_TYPE_reflink_v:
 121                 e->data_type = BCH_DATA_USER;
 122                 extent_to_replicas(k, e);
 123                 break;
 124         case KEY_TYPE_stripe:
 125                 e->data_type = BCH_DATA_USER;
 126                 stripe_to_replicas(k, e);
 127                 break;
 128         }
 129
 130         replicas_entry_sort(e);
 131 }
 132
 133 void bch2_devlist_to_replicas(struct bch_replicas_entry *e,
 134                               enum bch_data_type data_type,
 135                               struct bch_devs_list devs)
 136 {
 137         unsigned i;
 138
 139         BUG_ON(!data_type ||
 140                data_type == BCH_DATA_SB ||
 141                data_type >= BCH_DATA_NR);
 142
 143         e->data_type    = data_type;
 144         e->nr_devs      = 0;
 145         e->nr_required  = 1;
 146
 147         for (i = 0; i < devs.nr; i++)
 148                 e->devs[e->nr_devs++] = devs.devs[i];
 149
 150         replicas_entry_sort(e);
 151 }
 152
 153 static struct bch_replicas_cpu
 154 cpu_replicas_add_entry(struct bch_replicas_cpu *old,
 155                        struct bch_replicas_entry *new_entry)
 156 {
 157         unsigned i;
 158         struct bch_replicas_cpu new = {
 159                 .nr             = old->nr + 1,
 160                 .entry_size     = max_t(unsigned, old->entry_size,
 161                                         replicas_entry_bytes(new_entry)),
 162         };
 163
 164         BUG_ON(!new_entry->data_type);
 165         verify_replicas_entry(new_entry);
 166
 167         new.entries = kcalloc(new.nr, new.entry_size, GFP_NOIO);
 168         if (!new.entries)
 169                 return new;
 170
 171         for (i = 0; i < old->nr; i++)
 172                 memcpy(cpu_replicas_entry(&new, i),
 173                        cpu_replicas_entry(old, i),
 174                        old->entry_size);
 175
 176         memcpy(cpu_replicas_entry(&new, old->nr),
 177                new_entry,
 178                replicas_entry_bytes(new_entry));
 179
 180         bch2_cpu_replicas_sort(&new);
 181         return new;
 182 }
 183
 184 static inline int __replicas_entry_idx(struct bch_replicas_cpu *r,
 185                                        struct bch_replicas_entry *search)
 186 {
 187         int idx, entry_size = replicas_entry_bytes(search);
 188
 189         if (unlikely(entry_size > r->entry_size))
 190                 return -1;
 191
 192         verify_replicas_entry(search);
 193
 194 #define entry_cmp(_l, _r, size) memcmp(_l, _r, entry_size)
 195         idx = eytzinger0_find(r->entries, r->nr, r->entry_size,
 196                               entry_cmp, search);
 197 #undef entry_cmp
 198
 199         return idx < r->nr ? idx : -1;
 200 }
 201
 202 int bch2_replicas_entry_idx(struct bch_fs *c,
 203                             struct bch_replicas_entry *search)
 204 {
 205         replicas_entry_sort(search);
 206
 207         return __replicas_entry_idx(&c->replicas, search);
 208 }
 209
 210 static bool __replicas_has_entry(struct bch_replicas_cpu *r,
 211                                  struct bch_replicas_entry *search)
 212 {
 213         return __replicas_entry_idx(r, search) >= 0;
 214 }
 215
 216 static bool bch2_replicas_marked_locked(struct bch_fs *c,
 217                           struct bch_replicas_entry *search,
 218                           bool check_gc_replicas)
 219 {
 220         if (!search->nr_devs)
 221                 return true;
 222
 223         verify_replicas_entry(search);
 224
 225         return __replicas_has_entry(&c->replicas, search) &&
 226                 (!check_gc_replicas ||
 227                  likely((!c->replicas_gc.entries)) ||
 228                  __replicas_has_entry(&c->replicas_gc, search));
 229 }
 230
 231 bool bch2_replicas_marked(struct bch_fs *c,
 232                           struct bch_replicas_entry *search,
 233                           bool check_gc_replicas)
 234 {
 235         bool marked;
 236
 237         percpu_down_read(&c->mark_lock);
 238         marked = bch2_replicas_marked_locked(c, search, check_gc_replicas);
 239         percpu_up_read(&c->mark_lock);
 240
 241         return marked;
 242 }
 243
 244 static void __replicas_table_update(struct bch_fs_usage *dst,
 245                                     struct bch_replicas_cpu *dst_r,
 246                                     struct bch_fs_usage *src,
 247                                     struct bch_replicas_cpu *src_r)
 248 {
 249         int src_idx, dst_idx;
 250
 251         *dst = *src;
 252
 253         for (src_idx = 0; src_idx < src_r->nr; src_idx++) {
 254                 if (!src->replicas[src_idx])
 255                         continue;
 256
 257                 dst_idx = __replicas_entry_idx(dst_r,
 258                                 cpu_replicas_entry(src_r, src_idx));
 259                 BUG_ON(dst_idx < 0);
 260
 261                 dst->replicas[dst_idx] = src->replicas[src_idx];
 262         }
 263 }
 264
 265 static void __replicas_table_update_pcpu(struct bch_fs_usage __percpu *dst_p,
 266                                     struct bch_replicas_cpu *dst_r,
 267                                     struct bch_fs_usage __percpu *src_p,
 268                                     struct bch_replicas_cpu *src_r)
 269 {
 270         unsigned src_nr = sizeof(struct bch_fs_usage) / sizeof(u64) + src_r->nr;
 271         struct bch_fs_usage *dst, *src = (void *)
 272                 bch2_acc_percpu_u64s((void *) src_p, src_nr);
 273
 274         preempt_disable();
 275         dst = this_cpu_ptr(dst_p);
 276         preempt_enable();
 277
 278         __replicas_table_update(dst, dst_r, src, src_r);
 279 }
 280
 281 /*
 282  * Resize filesystem accounting:
 283  */
 284 static int replicas_table_update(struct bch_fs *c,
 285                                  struct bch_replicas_cpu *new_r)
 286 {
 287         struct bch_fs_usage __percpu *new_usage[2] = { NULL, NULL };
 288         struct bch_fs_usage *new_scratch = NULL;
 289         struct bch_fs_usage __percpu *new_gc = NULL;
 290         struct bch_fs_usage *new_base = NULL;
 291         unsigned bytes = sizeof(struct bch_fs_usage) +
 292                 sizeof(u64) * new_r->nr;
 293         int ret = -ENOMEM;
 294
 295         if (!(new_base = kzalloc(bytes, GFP_NOIO)) ||
 296             !(new_usage[0] = __alloc_percpu_gfp(bytes, sizeof(u64),
 297                                                 GFP_NOIO)) ||
 298             !(new_usage[1] = __alloc_percpu_gfp(bytes, sizeof(u64),
 299                                                 GFP_NOIO)) ||
 300             !(new_scratch  = kmalloc(bytes, GFP_NOIO)) ||
 301             (c->usage_gc &&
 302              !(new_gc = __alloc_percpu_gfp(bytes, sizeof(u64), GFP_NOIO))))
 303                 goto err;
 304
 305         if (c->usage_base)
 306                 __replicas_table_update(new_base,               new_r,
 307                                         c->usage_base,          &c->replicas);
 308         if (c->usage[0])
 309                 __replicas_table_update_pcpu(new_usage[0],      new_r,
 310                                              c->usage[0],       &c->replicas);
 311         if (c->usage[1])
 312                 __replicas_table_update_pcpu(new_usage[1],      new_r,
 313                                              c->usage[1],       &c->replicas);
 314         if (c->usage_gc)
 315                 __replicas_table_update_pcpu(new_gc,            new_r,
 316                                              c->usage_gc,       &c->replicas);
 317
 318         swap(c->usage_base,     new_base);
 319         swap(c->usage[0],       new_usage[0]);
 320         swap(c->usage[1],       new_usage[1]);
 321         swap(c->usage_scratch,  new_scratch);
 322         swap(c->usage_gc,       new_gc);
 323         swap(c->replicas,       *new_r);
 324         ret = 0;
 325 err:
 326         free_percpu(new_gc);
 327         kfree(new_scratch);
 328         free_percpu(new_usage[1]);
 329         free_percpu(new_usage[0]);
 330         kfree(new_base);
 331         return ret;
 332 }
 333
 334 static unsigned reserve_journal_replicas(struct bch_fs *c,
 335                                      struct bch_replicas_cpu *r)
 336 {
 337         struct bch_replicas_entry *e;
 338         unsigned journal_res_u64s = 0;
 339
 340         /* nr_inodes: */
 341         journal_res_u64s +=
 342                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 343
 344         /* key_version: */
 345         journal_res_u64s +=
 346                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 347
 348         /* persistent_reserved: */
 349         journal_res_u64s +=
 350                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64)) *
 351                 BCH_REPLICAS_MAX;
 352
 353         for_each_cpu_replicas_entry(r, e)
 354                 journal_res_u64s +=
 355                         DIV_ROUND_UP(sizeof(struct jset_entry_data_usage) +
 356                                      e->nr_devs, sizeof(u64));
 357         return journal_res_u64s;
 358 }
 359
 360 noinline
 361 static int bch2_mark_replicas_slowpath(struct bch_fs *c,
 362                                 struct bch_replicas_entry *new_entry)
 363 {
 364         struct bch_replicas_cpu new_r, new_gc;
 365         int ret = -ENOMEM;
 366
 367         verify_replicas_entry(new_entry);
 368
 369         memset(&new_r, 0, sizeof(new_r));
 370         memset(&new_gc, 0, sizeof(new_gc));
 371
 372         mutex_lock(&c->sb_lock);
 373
 374         if (c->replicas_gc.entries &&
 375             !__replicas_has_entry(&c->replicas_gc, new_entry)) {
 376                 new_gc = cpu_replicas_add_entry(&c->replicas_gc, new_entry);
 377                 if (!new_gc.entries)
 378                         goto err;
 379         }
 380
 381         if (!__replicas_has_entry(&c->replicas, new_entry)) {
 382                 new_r = cpu_replicas_add_entry(&c->replicas, new_entry);
 383                 if (!new_r.entries)
 384                         goto err;
 385
 386                 ret = bch2_cpu_replicas_to_sb_replicas(c, &new_r);
 387                 if (ret)
 388                         goto err;
 389
 390                 bch2_journal_entry_res_resize(&c->journal,
 391                                 &c->replicas_journal_res,
 392                                 reserve_journal_replicas(c, &new_r));
 393         }
 394
 395         if (!new_r.entries &&
 396             !new_gc.entries)
 397                 goto out;
 398
 399         /* allocations done, now commit: */
 400
 401         if (new_r.entries)
 402                 bch2_write_super(c);
 403
 404         /* don't update in memory replicas until changes are persistent */
 405         percpu_down_write(&c->mark_lock);
 406         if (new_r.entries)
 407                 ret = replicas_table_update(c, &new_r);
 408         if (new_gc.entries)
 409                 swap(new_gc, c->replicas_gc);
 410         percpu_up_write(&c->mark_lock);
 411 out:
 412         ret = 0;
 413 err:
 414         mutex_unlock(&c->sb_lock);
 415
 416         kfree(new_r.entries);
 417         kfree(new_gc.entries);
 418
 419         return ret;
 420 }
 421
 422 int bch2_mark_replicas(struct bch_fs *c,
 423                        struct bch_replicas_entry *r)
 424 {
 425         return likely(bch2_replicas_marked(c, r, true))
 426                 ? 0
 427                 : bch2_mark_replicas_slowpath(c, r);
 428 }
 429
 430 bool bch2_bkey_replicas_marked_locked(struct bch_fs *c,
 431                                       struct bkey_s_c k,
 432                                       bool check_gc_replicas)
 433 {
 434         struct bch_replicas_padded search;
 435         struct bch_devs_list cached = bch2_bkey_cached_devs(k);
 436         unsigned i;
 437
 438         for (i = 0; i < cached.nr; i++) {
 439                 bch2_replicas_entry_cached(&search.e, cached.devs[i]);
 440
 441                 if (!bch2_replicas_marked_locked(c, &search.e,
 442                                                  check_gc_replicas))
 443                         return false;
 444         }
 445
 446         bch2_bkey_to_replicas(&search.e, k);
 447
 448         return bch2_replicas_marked_locked(c, &search.e, check_gc_replicas);
 449 }
 450
 451 bool bch2_bkey_replicas_marked(struct bch_fs *c,
 452                                struct bkey_s_c k,
 453                                bool check_gc_replicas)
 454 {
 455         bool marked;
 456
 457         percpu_down_read(&c->mark_lock);
 458         marked = bch2_bkey_replicas_marked_locked(c, k, check_gc_replicas);
 459         percpu_up_read(&c->mark_lock);
 460
 461         return marked;
 462 }
 463
 464 int bch2_mark_bkey_replicas(struct bch_fs *c, struct bkey_s_c k)
 465 {
 466         struct bch_replicas_padded search;
 467         struct bch_devs_list cached = bch2_bkey_cached_devs(k);
 468         unsigned i;
 469         int ret;
 470
 471         for (i = 0; i < cached.nr; i++) {
 472                 bch2_replicas_entry_cached(&search.e, cached.devs[i]);
 473
 474                 ret = bch2_mark_replicas(c, &search.e);
 475                 if (ret)
 476                         return ret;
 477         }
 478
 479         bch2_bkey_to_replicas(&search.e, k);
 480
 481         return bch2_mark_replicas(c, &search.e);
 482 }
 483
 484 int bch2_replicas_gc_end(struct bch_fs *c, int ret)
 485 {
 486         unsigned i;
 487
 488         lockdep_assert_held(&c->replicas_gc_lock);
 489
 490         mutex_lock(&c->sb_lock);
 491         percpu_down_write(&c->mark_lock);
 492
 493         /*
 494          * this is kind of crappy; the replicas gc mechanism needs to be ripped
 495          * out
 496          */
 497
 498         for (i = 0; i < c->replicas.nr; i++) {
 499                 struct bch_replicas_entry *e =
 500                         cpu_replicas_entry(&c->replicas, i);
 501                 struct bch_replicas_cpu n;
 502
 503                 if (!__replicas_has_entry(&c->replicas_gc, e) &&
 504                     (c->usage_base->replicas[i] ||
 505                      percpu_u64_get(&c->usage[0]->replicas[i]) ||
 506                      percpu_u64_get(&c->usage[1]->replicas[i]))) {
 507                         n = cpu_replicas_add_entry(&c->replicas_gc, e);
 508                         if (!n.entries) {
 509                                 ret = -ENOSPC;
 510                                 goto err;
 511                         }
 512
 513                         swap(n, c->replicas_gc);
 514                         kfree(n.entries);
 515                 }
 516         }
 517
 518         if (bch2_cpu_replicas_to_sb_replicas(c, &c->replicas_gc)) {
 519                 ret = -ENOSPC;
 520                 goto err;
 521         }
 522
 523         ret = replicas_table_update(c, &c->replicas_gc);
 524 err:
 525         kfree(c->replicas_gc.entries);
 526         c->replicas_gc.entries = NULL;
 527
 528         percpu_up_write(&c->mark_lock);
 529
 530         if (!ret)
 531                 bch2_write_super(c);
 532
 533         mutex_unlock(&c->sb_lock);
 534
 535         return ret;
 536 }
 537
 538 int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask)
 539 {
 540         struct bch_replicas_entry *e;
 541         unsigned i = 0;
 542
 543         lockdep_assert_held(&c->replicas_gc_lock);
 544
 545         mutex_lock(&c->sb_lock);
 546         BUG_ON(c->replicas_gc.entries);
 547
 548         c->replicas_gc.nr               = 0;
 549         c->replicas_gc.entry_size       = 0;
 550
 551         for_each_cpu_replicas_entry(&c->replicas, e)
 552                 if (!((1 << e->data_type) & typemask)) {
 553                         c->replicas_gc.nr++;
 554                         c->replicas_gc.entry_size =
 555                                 max_t(unsigned, c->replicas_gc.entry_size,
 556                                       replicas_entry_bytes(e));
 557                 }
 558
 559         c->replicas_gc.entries = kcalloc(c->replicas_gc.nr,
 560                                          c->replicas_gc.entry_size,
 561                                          GFP_NOIO);
 562         if (!c->replicas_gc.entries) {
 563                 mutex_unlock(&c->sb_lock);
 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                 return -ENOMEM;
 591
 592         mutex_lock(&c->sb_lock);
 593         percpu_down_write(&c->mark_lock);
 594
 595         if (nr                  != c->replicas.nr ||
 596             new.entry_size      != c->replicas.entry_size) {
 597                 percpu_up_write(&c->mark_lock);
 598                 mutex_unlock(&c->sb_lock);
 599                 kfree(new.entries);
 600                 goto retry;
 601         }
 602
 603         for (i = 0; i < c->replicas.nr; i++) {
 604                 struct bch_replicas_entry *e =
 605                         cpu_replicas_entry(&c->replicas, i);
 606
 607                 if (e->data_type == BCH_DATA_JOURNAL ||
 608                     c->usage_base->replicas[i] ||
 609                     percpu_u64_get(&c->usage[0]->replicas[i]) ||
 610                     percpu_u64_get(&c->usage[1]->replicas[i]))
 611                         memcpy(cpu_replicas_entry(&new, new.nr++),
 612                                e, new.entry_size);
 613         }
 614
 615         bch2_cpu_replicas_sort(&new);
 616
 617         if (bch2_cpu_replicas_to_sb_replicas(c, &new)) {
 618                 ret = -ENOSPC;
 619                 goto err;
 620         }
 621
 622         ret = replicas_table_update(c, &new);
 623 err:
 624         kfree(new.entries);
 625
 626         percpu_up_write(&c->mark_lock);
 627
 628         if (!ret)
 629                 bch2_write_super(c);
 630
 631         mutex_unlock(&c->sb_lock);
 632
 633         return ret;
 634 }
 635
 636 int bch2_replicas_set_usage(struct bch_fs *c,
 637                             struct bch_replicas_entry *r,
 638                             u64 sectors)
 639 {
 640         int ret, idx = bch2_replicas_entry_idx(c, r);
 641
 642         if (idx < 0) {
 643                 struct bch_replicas_cpu n;
 644
 645                 n = cpu_replicas_add_entry(&c->replicas, r);
 646                 if (!n.entries)
 647                         return -ENOMEM;
 648
 649                 ret = replicas_table_update(c, &n);
 650                 if (ret)
 651                         return ret;
 652
 653                 kfree(n.entries);
 654
 655                 idx = bch2_replicas_entry_idx(c, r);
 656                 BUG_ON(ret < 0);
 657         }
 658
 659         c->usage_base->replicas[idx] = sectors;
 660
 661         return 0;
 662 }
 663
 664 /* Replicas tracking - superblock: */
 665
 666 static int
 667 __bch2_sb_replicas_to_cpu_replicas(struct bch_sb_field_replicas *sb_r,
 668                                    struct bch_replicas_cpu *cpu_r)
 669 {
 670         struct bch_replicas_entry *e, *dst;
 671         unsigned nr = 0, entry_size = 0, idx = 0;
 672
 673         for_each_replicas_entry(sb_r, e) {
 674                 entry_size = max_t(unsigned, entry_size,
 675                                    replicas_entry_bytes(e));
 676                 nr++;
 677         }
 678
 679         cpu_r->entries = kcalloc(nr, entry_size, GFP_NOIO);
 680         if (!cpu_r->entries)
 681                 return -ENOMEM;
 682
 683         cpu_r->nr               = nr;
 684         cpu_r->entry_size       = entry_size;
 685
 686         for_each_replicas_entry(sb_r, e) {
 687                 dst = cpu_replicas_entry(cpu_r, idx++);
 688                 memcpy(dst, e, replicas_entry_bytes(e));
 689                 replicas_entry_sort(dst);
 690         }
 691
 692         return 0;
 693 }
 694
 695 static int
 696 __bch2_sb_replicas_v0_to_cpu_replicas(struct bch_sb_field_replicas_v0 *sb_r,
 697                                       struct bch_replicas_cpu *cpu_r)
 698 {
 699         struct bch_replicas_entry_v0 *e;
 700         unsigned nr = 0, entry_size = 0, idx = 0;
 701
 702         for_each_replicas_entry(sb_r, e) {
 703                 entry_size = max_t(unsigned, entry_size,
 704                                    replicas_entry_bytes(e));
 705                 nr++;
 706         }
 707
 708         entry_size += sizeof(struct bch_replicas_entry) -
 709                 sizeof(struct bch_replicas_entry_v0);
 710
 711         cpu_r->entries = kcalloc(nr, entry_size, GFP_NOIO);
 712         if (!cpu_r->entries)
 713                 return -ENOMEM;
 714
 715         cpu_r->nr               = nr;
 716         cpu_r->entry_size       = entry_size;
 717
 718         for_each_replicas_entry(sb_r, e) {
 719                 struct bch_replicas_entry *dst =
 720                         cpu_replicas_entry(cpu_r, idx++);
 721
 722                 dst->data_type  = e->data_type;
 723                 dst->nr_devs    = e->nr_devs;
 724                 dst->nr_required = 1;
 725                 memcpy(dst->devs, e->devs, e->nr_devs);
 726                 replicas_entry_sort(dst);
 727         }
 728
 729         return 0;
 730 }
 731
 732 int bch2_sb_replicas_to_cpu_replicas(struct bch_fs *c)
 733 {
 734         struct bch_sb_field_replicas *sb_v1;
 735         struct bch_sb_field_replicas_v0 *sb_v0;
 736         struct bch_replicas_cpu new_r = { 0, 0, NULL };
 737         int ret = 0;
 738
 739         if ((sb_v1 = bch2_sb_get_replicas(c->disk_sb.sb)))
 740                 ret = __bch2_sb_replicas_to_cpu_replicas(sb_v1, &new_r);
 741         else if ((sb_v0 = bch2_sb_get_replicas_v0(c->disk_sb.sb)))
 742                 ret = __bch2_sb_replicas_v0_to_cpu_replicas(sb_v0, &new_r);
 743
 744         if (ret)
 745                 return -ENOMEM;
 746
 747         bch2_cpu_replicas_sort(&new_r);
 748
 749         percpu_down_write(&c->mark_lock);
 750
 751         ret = replicas_table_update(c, &new_r);
 752         percpu_up_write(&c->mark_lock);
 753
 754         kfree(new_r.entries);
 755
 756         return 0;
 757 }
 758
 759 static int bch2_cpu_replicas_to_sb_replicas_v0(struct bch_fs *c,
 760                                                struct bch_replicas_cpu *r)
 761 {
 762         struct bch_sb_field_replicas_v0 *sb_r;
 763         struct bch_replicas_entry_v0 *dst;
 764         struct bch_replicas_entry *src;
 765         size_t bytes;
 766
 767         bytes = sizeof(struct bch_sb_field_replicas);
 768
 769         for_each_cpu_replicas_entry(r, src)
 770                 bytes += replicas_entry_bytes(src) - 1;
 771
 772         sb_r = bch2_sb_resize_replicas_v0(&c->disk_sb,
 773                         DIV_ROUND_UP(bytes, sizeof(u64)));
 774         if (!sb_r)
 775                 return -ENOSPC;
 776
 777         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas);
 778         sb_r = bch2_sb_get_replicas_v0(c->disk_sb.sb);
 779
 780         memset(&sb_r->entries, 0,
 781                vstruct_end(&sb_r->field) -
 782                (void *) &sb_r->entries);
 783
 784         dst = sb_r->entries;
 785         for_each_cpu_replicas_entry(r, src) {
 786                 dst->data_type  = src->data_type;
 787                 dst->nr_devs    = src->nr_devs;
 788                 memcpy(dst->devs, src->devs, src->nr_devs);
 789
 790                 dst = replicas_entry_next(dst);
 791
 792                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 793         }
 794
 795         return 0;
 796 }
 797
 798 static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *c,
 799                                             struct bch_replicas_cpu *r)
 800 {
 801         struct bch_sb_field_replicas *sb_r;
 802         struct bch_replicas_entry *dst, *src;
 803         bool need_v1 = false;
 804         size_t bytes;
 805
 806         bytes = sizeof(struct bch_sb_field_replicas);
 807
 808         for_each_cpu_replicas_entry(r, src) {
 809                 bytes += replicas_entry_bytes(src);
 810                 if (src->nr_required != 1)
 811                         need_v1 = true;
 812         }
 813
 814         if (!need_v1)
 815                 return bch2_cpu_replicas_to_sb_replicas_v0(c, r);
 816
 817         sb_r = bch2_sb_resize_replicas(&c->disk_sb,
 818                         DIV_ROUND_UP(bytes, sizeof(u64)));
 819         if (!sb_r)
 820                 return -ENOSPC;
 821
 822         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas_v0);
 823         sb_r = bch2_sb_get_replicas(c->disk_sb.sb);
 824
 825         memset(&sb_r->entries, 0,
 826                vstruct_end(&sb_r->field) -
 827                (void *) &sb_r->entries);
 828
 829         dst = sb_r->entries;
 830         for_each_cpu_replicas_entry(r, src) {
 831                 memcpy(dst, src, replicas_entry_bytes(src));
 832
 833                 dst = replicas_entry_next(dst);
 834
 835                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 836         }
 837
 838         return 0;
 839 }
 840
 841 static const char *check_dup_replicas_entries(struct bch_replicas_cpu *cpu_r)
 842 {
 843         unsigned i;
 844
 845         sort_cmp_size(cpu_r->entries,
 846                       cpu_r->nr,
 847                       cpu_r->entry_size,
 848                       memcmp, NULL);
 849
 850         for (i = 0; i + 1 < cpu_r->nr; i++) {
 851                 struct bch_replicas_entry *l =
 852                         cpu_replicas_entry(cpu_r, i);
 853                 struct bch_replicas_entry *r =
 854                         cpu_replicas_entry(cpu_r, i + 1);
 855
 856                 BUG_ON(memcmp(l, r, cpu_r->entry_size) > 0);
 857
 858                 if (!memcmp(l, r, cpu_r->entry_size))
 859                         return "duplicate replicas entry";
 860         }
 861
 862         return NULL;
 863 }
 864
 865 static const char *bch2_sb_validate_replicas(struct bch_sb *sb, struct bch_sb_field *f)
 866 {
 867         struct bch_sb_field_replicas *sb_r = field_to_type(f, replicas);
 868         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 869         struct bch_replicas_cpu cpu_r = { .entries = NULL };
 870         struct bch_replicas_entry *e;
 871         const char *err;
 872         unsigned i;
 873
 874         for_each_replicas_entry(sb_r, e) {
 875                 err = "invalid replicas entry: invalid data type";
 876                 if (e->data_type >= BCH_DATA_NR)
 877                         goto err;
 878
 879                 err = "invalid replicas entry: no devices";
 880                 if (!e->nr_devs)
 881                         goto err;
 882
 883                 err = "invalid replicas entry: bad nr_required";
 884                 if (e->nr_required > 1 &&
 885                     e->nr_required >= e->nr_devs)
 886                         goto err;
 887
 888                 err = "invalid replicas entry: invalid device";
 889                 for (i = 0; i < e->nr_devs; i++)
 890                         if (!bch2_dev_exists(sb, mi, e->devs[i]))
 891                                 goto err;
 892         }
 893
 894         err = "cannot allocate memory";
 895         if (__bch2_sb_replicas_to_cpu_replicas(sb_r, &cpu_r))
 896                 goto err;
 897
 898         err = check_dup_replicas_entries(&cpu_r);
 899 err:
 900         kfree(cpu_r.entries);
 901         return err;
 902 }
 903
 904 static void bch2_sb_replicas_to_text(struct printbuf *out,
 905                                      struct bch_sb *sb,
 906                                      struct bch_sb_field *f)
 907 {
 908         struct bch_sb_field_replicas *r = field_to_type(f, replicas);
 909         struct bch_replicas_entry *e;
 910         bool first = true;
 911
 912         for_each_replicas_entry(r, e) {
 913                 if (!first)
 914                         pr_buf(out, " ");
 915                 first = false;
 916
 917                 bch2_replicas_entry_to_text(out, e);
 918         }
 919 }
 920
 921 const struct bch_sb_field_ops bch_sb_field_ops_replicas = {
 922         .validate       = bch2_sb_validate_replicas,
 923         .to_text        = bch2_sb_replicas_to_text,
 924 };
 925
 926 static const char *bch2_sb_validate_replicas_v0(struct bch_sb *sb, struct bch_sb_field *f)
 927 {
 928         struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0);
 929         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 930         struct bch_replicas_cpu cpu_r = { .entries = NULL };
 931         struct bch_replicas_entry_v0 *e;
 932         const char *err;
 933         unsigned i;
 934
 935         for_each_replicas_entry_v0(sb_r, e) {
 936                 err = "invalid replicas entry: invalid data type";
 937                 if (e->data_type >= BCH_DATA_NR)
 938                         goto err;
 939
 940                 err = "invalid replicas entry: no devices";
 941                 if (!e->nr_devs)
 942                         goto err;
 943
 944                 err = "invalid replicas entry: invalid device";
 945                 for (i = 0; i < e->nr_devs; i++)
 946                         if (!bch2_dev_exists(sb, mi, e->devs[i]))
 947                                 goto err;
 948         }
 949
 950         err = "cannot allocate memory";
 951         if (__bch2_sb_replicas_v0_to_cpu_replicas(sb_r, &cpu_r))
 952                 goto err;
 953
 954         err = check_dup_replicas_entries(&cpu_r);
 955 err:
 956         kfree(cpu_r.entries);
 957         return err;
 958 }
 959
 960 const struct bch_sb_field_ops bch_sb_field_ops_replicas_v0 = {
 961         .validate       = bch2_sb_validate_replicas_v0,
 962 };
 963
 964 /* Query replicas: */
 965
 966 struct replicas_status __bch2_replicas_status(struct bch_fs *c,
 967                                               struct bch_devs_mask online_devs)
 968 {
 969         struct bch_sb_field_members *mi;
 970         struct bch_replicas_entry *e;
 971         unsigned i, nr_online, nr_offline;
 972         struct replicas_status ret;
 973
 974         memset(&ret, 0, sizeof(ret));
 975
 976         for (i = 0; i < ARRAY_SIZE(ret.replicas); i++)
 977                 ret.replicas[i].redundancy = INT_MAX;
 978
 979         mi = bch2_sb_get_members(c->disk_sb.sb);
 980
 981         percpu_down_read(&c->mark_lock);
 982
 983         for_each_cpu_replicas_entry(&c->replicas, e) {
 984                 if (e->data_type >= ARRAY_SIZE(ret.replicas))
 985                         panic("e %p data_type %u\n", e, e->data_type);
 986
 987                 nr_online = nr_offline = 0;
 988
 989                 for (i = 0; i < e->nr_devs; i++) {
 990                         BUG_ON(!bch2_dev_exists(c->disk_sb.sb, mi,
 991                                                 e->devs[i]));
 992
 993                         if (test_bit(e->devs[i], online_devs.d))
 994                                 nr_online++;
 995                         else
 996                                 nr_offline++;
 997                 }
 998
 999                 ret.replicas[e->data_type].redundancy =
1000                         min(ret.replicas[e->data_type].redundancy,
1001                             (int) nr_online - (int) e->nr_required);
1002
1003                 ret.replicas[e->data_type].nr_offline =
1004                         max(ret.replicas[e->data_type].nr_offline,
1005                             nr_offline);
1006         }
1007
1008         percpu_up_read(&c->mark_lock);
1009
1010         for (i = 0; i < ARRAY_SIZE(ret.replicas); i++)
1011                 if (ret.replicas[i].redundancy == INT_MAX)
1012                         ret.replicas[i].redundancy = 0;
1013
1014         return ret;
1015 }
1016
1017 struct replicas_status bch2_replicas_status(struct bch_fs *c)
1018 {
1019         return __bch2_replicas_status(c, bch2_online_devs(c));
1020 }
1021
1022 static bool have_enough_devs(struct replicas_status s,
1023                              enum bch_data_type type,
1024                              bool force_if_degraded,
1025                              bool force_if_lost)
1026 {
1027         return (!s.replicas[type].nr_offline || force_if_degraded) &&
1028                 (s.replicas[type].redundancy >= 0 || force_if_lost);
1029 }
1030
1031 bool bch2_have_enough_devs(struct replicas_status s, unsigned flags)
1032 {
1033         return (have_enough_devs(s, BCH_DATA_JOURNAL,
1034                                  flags & BCH_FORCE_IF_METADATA_DEGRADED,
1035                                  flags & BCH_FORCE_IF_METADATA_LOST) &&
1036                 have_enough_devs(s, BCH_DATA_BTREE,
1037                                  flags & BCH_FORCE_IF_METADATA_DEGRADED,
1038                                  flags & BCH_FORCE_IF_METADATA_LOST) &&
1039                 have_enough_devs(s, BCH_DATA_USER,
1040                                  flags & BCH_FORCE_IF_DATA_DEGRADED,
1041                                  flags & BCH_FORCE_IF_DATA_LOST));
1042 }
1043
1044 int bch2_replicas_online(struct bch_fs *c, bool meta)
1045 {
1046         struct replicas_status s = bch2_replicas_status(c);
1047
1048         return (meta
1049                 ? min(s.replicas[BCH_DATA_JOURNAL].redundancy,
1050                       s.replicas[BCH_DATA_BTREE].redundancy)
1051                 : s.replicas[BCH_DATA_USER].redundancy) + 1;
1052 }
1053
1054 unsigned bch2_dev_has_data(struct bch_fs *c, struct bch_dev *ca)
1055 {
1056         struct bch_replicas_entry *e;
1057         unsigned i, ret = 0;
1058
1059         percpu_down_read(&c->mark_lock);
1060
1061         for_each_cpu_replicas_entry(&c->replicas, e)
1062                 for (i = 0; i < e->nr_devs; i++)
1063                         if (e->devs[i] == ca->dev_idx)
1064                                 ret |= 1 << e->data_type;
1065
1066         percpu_up_read(&c->mark_lock);
1067
1068         return ret;
1069 }
1070
1071 int bch2_fs_replicas_init(struct bch_fs *c)
1072 {
1073         c->journal.entry_u64s_reserved +=
1074                 reserve_journal_replicas(c, &c->replicas);
1075
1076         return replicas_table_update(c, &c->replicas);
1077 }