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_sorted(struct bch_replicas_entry *e)
  20 {
  21 #ifdef CONFIG_BCACHES_DEBUG
  22         unsigned i;
  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.ec_nr) {
  83                         r->nr_devs = 0;
  84                         break;
  85                 }
  86
  87                 r->devs[r->nr_devs++] = p.ptr.dev;
  88         }
  89 }
  90
  91 static void stripe_to_replicas(struct bkey_s_c k,
  92                                struct bch_replicas_entry *r)
  93 {
  94         struct bkey_s_c_stripe s = bkey_s_c_to_stripe(k);
  95         const struct bch_extent_ptr *ptr;
  96
  97         r->nr_required  = s.v->nr_blocks - s.v->nr_redundant;
  98
  99         for (ptr = s.v->ptrs;
 100              ptr < s.v->ptrs + s.v->nr_blocks;
 101              ptr++)
 102                 r->devs[r->nr_devs++] = ptr->dev;
 103 }
 104
 105 void bch2_bkey_to_replicas(struct bch_replicas_entry *e,
 106                            struct bkey_s_c k)
 107 {
 108         e->nr_devs = 0;
 109
 110         switch (k.k->type) {
 111         case KEY_TYPE_btree_ptr:
 112                 e->data_type = BCH_DATA_BTREE;
 113                 extent_to_replicas(k, e);
 114                 break;
 115         case KEY_TYPE_extent:
 116         case KEY_TYPE_reflink_v:
 117                 e->data_type = BCH_DATA_USER;
 118                 extent_to_replicas(k, e);
 119                 break;
 120         case KEY_TYPE_stripe:
 121                 e->data_type = BCH_DATA_USER;
 122                 stripe_to_replicas(k, e);
 123                 break;
 124         }
 125
 126         replicas_entry_sort(e);
 127 }
 128
 129 void bch2_devlist_to_replicas(struct bch_replicas_entry *e,
 130                               enum bch_data_type data_type,
 131                               struct bch_devs_list devs)
 132 {
 133         unsigned i;
 134
 135         BUG_ON(!data_type ||
 136                data_type == BCH_DATA_SB ||
 137                data_type >= BCH_DATA_NR);
 138
 139         e->data_type    = data_type;
 140         e->nr_devs      = 0;
 141         e->nr_required  = 1;
 142
 143         for (i = 0; i < devs.nr; i++)
 144                 e->devs[e->nr_devs++] = devs.devs[i];
 145
 146         replicas_entry_sort(e);
 147 }
 148
 149 static struct bch_replicas_cpu
 150 cpu_replicas_add_entry(struct bch_replicas_cpu *old,
 151                        struct bch_replicas_entry *new_entry)
 152 {
 153         unsigned i;
 154         struct bch_replicas_cpu new = {
 155                 .nr             = old->nr + 1,
 156                 .entry_size     = max_t(unsigned, old->entry_size,
 157                                         replicas_entry_bytes(new_entry)),
 158         };
 159
 160         BUG_ON(!new_entry->data_type);
 161         verify_replicas_entry_sorted(new_entry);
 162
 163         new.entries = kcalloc(new.nr, new.entry_size, GFP_NOIO);
 164         if (!new.entries)
 165                 return new;
 166
 167         for (i = 0; i < old->nr; i++)
 168                 memcpy(cpu_replicas_entry(&new, i),
 169                        cpu_replicas_entry(old, i),
 170                        old->entry_size);
 171
 172         memcpy(cpu_replicas_entry(&new, old->nr),
 173                new_entry,
 174                replicas_entry_bytes(new_entry));
 175
 176         bch2_cpu_replicas_sort(&new);
 177         return new;
 178 }
 179
 180 static inline int __replicas_entry_idx(struct bch_replicas_cpu *r,
 181                                        struct bch_replicas_entry *search)
 182 {
 183         int idx, entry_size = replicas_entry_bytes(search);
 184
 185         if (unlikely(entry_size > r->entry_size))
 186                 return -1;
 187
 188         verify_replicas_entry_sorted(search);
 189
 190 #define entry_cmp(_l, _r, size) memcmp(_l, _r, entry_size)
 191         idx = eytzinger0_find(r->entries, r->nr, r->entry_size,
 192                               entry_cmp, search);
 193 #undef entry_cmp
 194
 195         return idx < r->nr ? idx : -1;
 196 }
 197
 198 int bch2_replicas_entry_idx(struct bch_fs *c,
 199                             struct bch_replicas_entry *search)
 200 {
 201         replicas_entry_sort(search);
 202
 203         return __replicas_entry_idx(&c->replicas, search);
 204 }
 205
 206 static bool __replicas_has_entry(struct bch_replicas_cpu *r,
 207                                  struct bch_replicas_entry *search)
 208 {
 209         return __replicas_entry_idx(r, search) >= 0;
 210 }
 211
 212 static bool bch2_replicas_marked_locked(struct bch_fs *c,
 213                           struct bch_replicas_entry *search,
 214                           bool check_gc_replicas)
 215 {
 216         if (!search->nr_devs)
 217                 return true;
 218
 219         verify_replicas_entry_sorted(search);
 220
 221         return __replicas_has_entry(&c->replicas, search) &&
 222                 (!check_gc_replicas ||
 223                  likely((!c->replicas_gc.entries)) ||
 224                  __replicas_has_entry(&c->replicas_gc, search));
 225 }
 226
 227 bool bch2_replicas_marked(struct bch_fs *c,
 228                           struct bch_replicas_entry *search,
 229                           bool check_gc_replicas)
 230 {
 231         bool marked;
 232
 233         percpu_down_read(&c->mark_lock);
 234         marked = bch2_replicas_marked_locked(c, search, check_gc_replicas);
 235         percpu_up_read(&c->mark_lock);
 236
 237         return marked;
 238 }
 239
 240 static void __replicas_table_update(struct bch_fs_usage *dst,
 241                                     struct bch_replicas_cpu *dst_r,
 242                                     struct bch_fs_usage *src,
 243                                     struct bch_replicas_cpu *src_r)
 244 {
 245         int src_idx, dst_idx;
 246
 247         *dst = *src;
 248
 249         for (src_idx = 0; src_idx < src_r->nr; src_idx++) {
 250                 if (!src->replicas[src_idx])
 251                         continue;
 252
 253                 dst_idx = __replicas_entry_idx(dst_r,
 254                                 cpu_replicas_entry(src_r, src_idx));
 255                 BUG_ON(dst_idx < 0);
 256
 257                 dst->replicas[dst_idx] = src->replicas[src_idx];
 258         }
 259 }
 260
 261 static void __replicas_table_update_pcpu(struct bch_fs_usage __percpu *dst_p,
 262                                     struct bch_replicas_cpu *dst_r,
 263                                     struct bch_fs_usage __percpu *src_p,
 264                                     struct bch_replicas_cpu *src_r)
 265 {
 266         unsigned src_nr = sizeof(struct bch_fs_usage) / sizeof(u64) + src_r->nr;
 267         struct bch_fs_usage *dst, *src = (void *)
 268                 bch2_acc_percpu_u64s((void *) src_p, src_nr);
 269
 270         preempt_disable();
 271         dst = this_cpu_ptr(dst_p);
 272         preempt_enable();
 273
 274         __replicas_table_update(dst, dst_r, src, src_r);
 275 }
 276
 277 /*
 278  * Resize filesystem accounting:
 279  */
 280 static int replicas_table_update(struct bch_fs *c,
 281                                  struct bch_replicas_cpu *new_r)
 282 {
 283         struct bch_fs_usage __percpu *new_usage[2] = { NULL, NULL };
 284         struct bch_fs_usage *new_scratch = NULL;
 285         struct bch_fs_usage __percpu *new_gc = NULL;
 286         struct bch_fs_usage *new_base = NULL;
 287         unsigned bytes = sizeof(struct bch_fs_usage) +
 288                 sizeof(u64) * new_r->nr;
 289         int ret = -ENOMEM;
 290
 291         if (!(new_base = kzalloc(bytes, GFP_NOIO)) ||
 292             !(new_usage[0] = __alloc_percpu_gfp(bytes, sizeof(u64),
 293                                                 GFP_NOIO)) ||
 294             !(new_usage[1] = __alloc_percpu_gfp(bytes, sizeof(u64),
 295                                                 GFP_NOIO)) ||
 296             !(new_scratch  = kmalloc(bytes, GFP_NOIO)) ||
 297             (c->usage_gc &&
 298              !(new_gc = __alloc_percpu_gfp(bytes, sizeof(u64), GFP_NOIO))))
 299                 goto err;
 300
 301         if (c->usage_base)
 302                 __replicas_table_update(new_base,               new_r,
 303                                         c->usage_base,          &c->replicas);
 304         if (c->usage[0])
 305                 __replicas_table_update_pcpu(new_usage[0],      new_r,
 306                                              c->usage[0],       &c->replicas);
 307         if (c->usage[1])
 308                 __replicas_table_update_pcpu(new_usage[1],      new_r,
 309                                              c->usage[1],       &c->replicas);
 310         if (c->usage_gc)
 311                 __replicas_table_update_pcpu(new_gc,            new_r,
 312                                              c->usage_gc,       &c->replicas);
 313
 314         swap(c->usage_base,     new_base);
 315         swap(c->usage[0],       new_usage[0]);
 316         swap(c->usage[1],       new_usage[1]);
 317         swap(c->usage_scratch,  new_scratch);
 318         swap(c->usage_gc,       new_gc);
 319         swap(c->replicas,       *new_r);
 320         ret = 0;
 321 err:
 322         free_percpu(new_gc);
 323         kfree(new_scratch);
 324         free_percpu(new_usage[1]);
 325         free_percpu(new_usage[0]);
 326         kfree(new_base);
 327         return ret;
 328 }
 329
 330 static unsigned reserve_journal_replicas(struct bch_fs *c,
 331                                      struct bch_replicas_cpu *r)
 332 {
 333         struct bch_replicas_entry *e;
 334         unsigned journal_res_u64s = 0;
 335
 336         /* nr_inodes: */
 337         journal_res_u64s +=
 338                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 339
 340         /* key_version: */
 341         journal_res_u64s +=
 342                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 343
 344         /* persistent_reserved: */
 345         journal_res_u64s +=
 346                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64)) *
 347                 BCH_REPLICAS_MAX;
 348
 349         for_each_cpu_replicas_entry(r, e)
 350                 journal_res_u64s +=
 351                         DIV_ROUND_UP(sizeof(struct jset_entry_data_usage) +
 352                                      e->nr_devs, sizeof(u64));
 353         return journal_res_u64s;
 354 }
 355
 356 noinline
 357 static int bch2_mark_replicas_slowpath(struct bch_fs *c,
 358                                 struct bch_replicas_entry *new_entry)
 359 {
 360         struct bch_replicas_cpu new_r, new_gc;
 361         int ret = -ENOMEM;
 362
 363         memset(&new_r, 0, sizeof(new_r));
 364         memset(&new_gc, 0, sizeof(new_gc));
 365
 366         mutex_lock(&c->sb_lock);
 367
 368         if (c->replicas_gc.entries &&
 369             !__replicas_has_entry(&c->replicas_gc, new_entry)) {
 370                 new_gc = cpu_replicas_add_entry(&c->replicas_gc, new_entry);
 371                 if (!new_gc.entries)
 372                         goto err;
 373         }
 374
 375         if (!__replicas_has_entry(&c->replicas, new_entry)) {
 376                 new_r = cpu_replicas_add_entry(&c->replicas, new_entry);
 377                 if (!new_r.entries)
 378                         goto err;
 379
 380                 ret = bch2_cpu_replicas_to_sb_replicas(c, &new_r);
 381                 if (ret)
 382                         goto err;
 383
 384                 bch2_journal_entry_res_resize(&c->journal,
 385                                 &c->replicas_journal_res,
 386                                 reserve_journal_replicas(c, &new_r));
 387         }
 388
 389         if (!new_r.entries &&
 390             !new_gc.entries)
 391                 goto out;
 392
 393         /* allocations done, now commit: */
 394
 395         if (new_r.entries)
 396                 bch2_write_super(c);
 397
 398         /* don't update in memory replicas until changes are persistent */
 399         percpu_down_write(&c->mark_lock);
 400         if (new_r.entries)
 401                 ret = replicas_table_update(c, &new_r);
 402         if (new_gc.entries)
 403                 swap(new_gc, c->replicas_gc);
 404         percpu_up_write(&c->mark_lock);
 405 out:
 406         ret = 0;
 407 err:
 408         mutex_unlock(&c->sb_lock);
 409
 410         kfree(new_r.entries);
 411         kfree(new_gc.entries);
 412
 413         return ret;
 414 }
 415
 416 int bch2_mark_replicas(struct bch_fs *c,
 417                        struct bch_replicas_entry *r)
 418 {
 419         return likely(bch2_replicas_marked(c, r, true))
 420                 ? 0
 421                 : bch2_mark_replicas_slowpath(c, r);
 422 }
 423
 424 bool bch2_bkey_replicas_marked_locked(struct bch_fs *c,
 425                                       struct bkey_s_c k,
 426                                       bool check_gc_replicas)
 427 {
 428         struct bch_replicas_padded search;
 429         struct bch_devs_list cached = bch2_bkey_cached_devs(k);
 430         unsigned i;
 431
 432         for (i = 0; i < cached.nr; i++) {
 433                 bch2_replicas_entry_cached(&search.e, cached.devs[i]);
 434
 435                 if (!bch2_replicas_marked_locked(c, &search.e,
 436                                                  check_gc_replicas))
 437                         return false;
 438         }
 439
 440         bch2_bkey_to_replicas(&search.e, k);
 441
 442         return bch2_replicas_marked_locked(c, &search.e, check_gc_replicas);
 443 }
 444
 445 bool bch2_bkey_replicas_marked(struct bch_fs *c,
 446                                struct bkey_s_c k,
 447                                bool check_gc_replicas)
 448 {
 449         bool marked;
 450
 451         percpu_down_read(&c->mark_lock);
 452         marked = bch2_bkey_replicas_marked_locked(c, k, check_gc_replicas);
 453         percpu_up_read(&c->mark_lock);
 454
 455         return marked;
 456 }
 457
 458 int bch2_mark_bkey_replicas(struct bch_fs *c, struct bkey_s_c k)
 459 {
 460         struct bch_replicas_padded search;
 461         struct bch_devs_list cached = bch2_bkey_cached_devs(k);
 462         unsigned i;
 463         int ret;
 464
 465         for (i = 0; i < cached.nr; i++) {
 466                 bch2_replicas_entry_cached(&search.e, cached.devs[i]);
 467
 468                 ret = bch2_mark_replicas(c, &search.e);
 469                 if (ret)
 470                         return ret;
 471         }
 472
 473         bch2_bkey_to_replicas(&search.e, k);
 474
 475         return bch2_mark_replicas(c, &search.e);
 476 }
 477
 478 int bch2_replicas_gc_end(struct bch_fs *c, int ret)
 479 {
 480         unsigned i;
 481
 482         lockdep_assert_held(&c->replicas_gc_lock);
 483
 484         mutex_lock(&c->sb_lock);
 485         percpu_down_write(&c->mark_lock);
 486
 487         /*
 488          * this is kind of crappy; the replicas gc mechanism needs to be ripped
 489          * out
 490          */
 491
 492         for (i = 0; i < c->replicas.nr; i++) {
 493                 struct bch_replicas_entry *e =
 494                         cpu_replicas_entry(&c->replicas, i);
 495                 struct bch_replicas_cpu n;
 496
 497                 if (!__replicas_has_entry(&c->replicas_gc, e) &&
 498                     (c->usage_base->replicas[i] ||
 499                      percpu_u64_get(&c->usage[0]->replicas[i]) ||
 500                      percpu_u64_get(&c->usage[1]->replicas[i]))) {
 501                         n = cpu_replicas_add_entry(&c->replicas_gc, e);
 502                         if (!n.entries) {
 503                                 ret = -ENOSPC;
 504                                 goto err;
 505                         }
 506
 507                         swap(n, c->replicas_gc);
 508                         kfree(n.entries);
 509                 }
 510         }
 511
 512         if (bch2_cpu_replicas_to_sb_replicas(c, &c->replicas_gc)) {
 513                 ret = -ENOSPC;
 514                 goto err;
 515         }
 516
 517         ret = replicas_table_update(c, &c->replicas_gc);
 518 err:
 519         kfree(c->replicas_gc.entries);
 520         c->replicas_gc.entries = NULL;
 521
 522         percpu_up_write(&c->mark_lock);
 523
 524         if (!ret)
 525                 bch2_write_super(c);
 526
 527         mutex_unlock(&c->sb_lock);
 528
 529         return ret;
 530 }
 531
 532 int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask)
 533 {
 534         struct bch_replicas_entry *e;
 535         unsigned i = 0;
 536
 537         lockdep_assert_held(&c->replicas_gc_lock);
 538
 539         mutex_lock(&c->sb_lock);
 540         BUG_ON(c->replicas_gc.entries);
 541
 542         c->replicas_gc.nr               = 0;
 543         c->replicas_gc.entry_size       = 0;
 544
 545         for_each_cpu_replicas_entry(&c->replicas, e)
 546                 if (!((1 << e->data_type) & typemask)) {
 547                         c->replicas_gc.nr++;
 548                         c->replicas_gc.entry_size =
 549                                 max_t(unsigned, c->replicas_gc.entry_size,
 550                                       replicas_entry_bytes(e));
 551                 }
 552
 553         c->replicas_gc.entries = kcalloc(c->replicas_gc.nr,
 554                                          c->replicas_gc.entry_size,
 555                                          GFP_NOIO);
 556         if (!c->replicas_gc.entries) {
 557                 mutex_unlock(&c->sb_lock);
 558                 return -ENOMEM;
 559         }
 560
 561         for_each_cpu_replicas_entry(&c->replicas, e)
 562                 if (!((1 << e->data_type) & typemask))
 563                         memcpy(cpu_replicas_entry(&c->replicas_gc, i++),
 564                                e, c->replicas_gc.entry_size);
 565
 566         bch2_cpu_replicas_sort(&c->replicas_gc);
 567         mutex_unlock(&c->sb_lock);
 568
 569         return 0;
 570 }
 571
 572 int bch2_replicas_gc2(struct bch_fs *c)
 573 {
 574         struct bch_replicas_cpu new = { 0 };
 575         unsigned i, nr;
 576         int ret = 0;
 577
 578         bch2_journal_meta(&c->journal);
 579 retry:
 580         nr              = READ_ONCE(c->replicas.nr);
 581         new.entry_size  = READ_ONCE(c->replicas.entry_size);
 582         new.entries     = kcalloc(nr, new.entry_size, GFP_KERNEL);
 583         if (!new.entries)
 584                 return -ENOMEM;
 585
 586         mutex_lock(&c->sb_lock);
 587         percpu_down_write(&c->mark_lock);
 588
 589         if (nr                  != c->replicas.nr ||
 590             new.entry_size      != c->replicas.entry_size) {
 591                 percpu_up_write(&c->mark_lock);
 592                 mutex_unlock(&c->sb_lock);
 593                 kfree(new.entries);
 594                 goto retry;
 595         }
 596
 597         for (i = 0; i < c->replicas.nr; i++) {
 598                 struct bch_replicas_entry *e =
 599                         cpu_replicas_entry(&c->replicas, i);
 600
 601                 if (e->data_type == BCH_DATA_JOURNAL ||
 602                     c->usage_base->replicas[i] ||
 603                     percpu_u64_get(&c->usage[0]->replicas[i]) ||
 604                     percpu_u64_get(&c->usage[1]->replicas[i]))
 605                         memcpy(cpu_replicas_entry(&new, new.nr++),
 606                                e, new.entry_size);
 607         }
 608
 609         bch2_cpu_replicas_sort(&new);
 610
 611         if (bch2_cpu_replicas_to_sb_replicas(c, &new)) {
 612                 ret = -ENOSPC;
 613                 goto err;
 614         }
 615
 616         ret = replicas_table_update(c, &new);
 617 err:
 618         kfree(new.entries);
 619
 620         percpu_up_write(&c->mark_lock);
 621
 622         if (!ret)
 623                 bch2_write_super(c);
 624
 625         mutex_unlock(&c->sb_lock);
 626
 627         return ret;
 628 }
 629
 630 int bch2_replicas_set_usage(struct bch_fs *c,
 631                             struct bch_replicas_entry *r,
 632                             u64 sectors)
 633 {
 634         int ret, idx = bch2_replicas_entry_idx(c, r);
 635
 636         if (idx < 0) {
 637                 struct bch_replicas_cpu n;
 638
 639                 n = cpu_replicas_add_entry(&c->replicas, r);
 640                 if (!n.entries)
 641                         return -ENOMEM;
 642
 643                 ret = replicas_table_update(c, &n);
 644                 if (ret)
 645                         return ret;
 646
 647                 kfree(n.entries);
 648
 649                 idx = bch2_replicas_entry_idx(c, r);
 650                 BUG_ON(ret < 0);
 651         }
 652
 653         c->usage_base->replicas[idx] = sectors;
 654
 655         return 0;
 656 }
 657
 658 /* Replicas tracking - superblock: */
 659
 660 static int
 661 __bch2_sb_replicas_to_cpu_replicas(struct bch_sb_field_replicas *sb_r,
 662                                    struct bch_replicas_cpu *cpu_r)
 663 {
 664         struct bch_replicas_entry *e, *dst;
 665         unsigned nr = 0, entry_size = 0, idx = 0;
 666
 667         for_each_replicas_entry(sb_r, e) {
 668                 entry_size = max_t(unsigned, entry_size,
 669                                    replicas_entry_bytes(e));
 670                 nr++;
 671         }
 672
 673         cpu_r->entries = kcalloc(nr, entry_size, GFP_NOIO);
 674         if (!cpu_r->entries)
 675                 return -ENOMEM;
 676
 677         cpu_r->nr               = nr;
 678         cpu_r->entry_size       = entry_size;
 679
 680         for_each_replicas_entry(sb_r, e) {
 681                 dst = cpu_replicas_entry(cpu_r, idx++);
 682                 memcpy(dst, e, replicas_entry_bytes(e));
 683                 replicas_entry_sort(dst);
 684         }
 685
 686         return 0;
 687 }
 688
 689 static int
 690 __bch2_sb_replicas_v0_to_cpu_replicas(struct bch_sb_field_replicas_v0 *sb_r,
 691                                       struct bch_replicas_cpu *cpu_r)
 692 {
 693         struct bch_replicas_entry_v0 *e;
 694         unsigned nr = 0, entry_size = 0, idx = 0;
 695
 696         for_each_replicas_entry(sb_r, e) {
 697                 entry_size = max_t(unsigned, entry_size,
 698                                    replicas_entry_bytes(e));
 699                 nr++;
 700         }
 701
 702         entry_size += sizeof(struct bch_replicas_entry) -
 703                 sizeof(struct bch_replicas_entry_v0);
 704
 705         cpu_r->entries = kcalloc(nr, entry_size, GFP_NOIO);
 706         if (!cpu_r->entries)
 707                 return -ENOMEM;
 708
 709         cpu_r->nr               = nr;
 710         cpu_r->entry_size       = entry_size;
 711
 712         for_each_replicas_entry(sb_r, e) {
 713                 struct bch_replicas_entry *dst =
 714                         cpu_replicas_entry(cpu_r, idx++);
 715
 716                 dst->data_type  = e->data_type;
 717                 dst->nr_devs    = e->nr_devs;
 718                 dst->nr_required = 1;
 719                 memcpy(dst->devs, e->devs, e->nr_devs);
 720                 replicas_entry_sort(dst);
 721         }
 722
 723         return 0;
 724 }
 725
 726 int bch2_sb_replicas_to_cpu_replicas(struct bch_fs *c)
 727 {
 728         struct bch_sb_field_replicas *sb_v1;
 729         struct bch_sb_field_replicas_v0 *sb_v0;
 730         struct bch_replicas_cpu new_r = { 0, 0, NULL };
 731         int ret = 0;
 732
 733         if ((sb_v1 = bch2_sb_get_replicas(c->disk_sb.sb)))
 734                 ret = __bch2_sb_replicas_to_cpu_replicas(sb_v1, &new_r);
 735         else if ((sb_v0 = bch2_sb_get_replicas_v0(c->disk_sb.sb)))
 736                 ret = __bch2_sb_replicas_v0_to_cpu_replicas(sb_v0, &new_r);
 737
 738         if (ret)
 739                 return -ENOMEM;
 740
 741         bch2_cpu_replicas_sort(&new_r);
 742
 743         percpu_down_write(&c->mark_lock);
 744
 745         ret = replicas_table_update(c, &new_r);
 746         percpu_up_write(&c->mark_lock);
 747
 748         kfree(new_r.entries);
 749
 750         return 0;
 751 }
 752
 753 static int bch2_cpu_replicas_to_sb_replicas_v0(struct bch_fs *c,
 754                                                struct bch_replicas_cpu *r)
 755 {
 756         struct bch_sb_field_replicas_v0 *sb_r;
 757         struct bch_replicas_entry_v0 *dst;
 758         struct bch_replicas_entry *src;
 759         size_t bytes;
 760
 761         bytes = sizeof(struct bch_sb_field_replicas);
 762
 763         for_each_cpu_replicas_entry(r, src)
 764                 bytes += replicas_entry_bytes(src) - 1;
 765
 766         sb_r = bch2_sb_resize_replicas_v0(&c->disk_sb,
 767                         DIV_ROUND_UP(bytes, sizeof(u64)));
 768         if (!sb_r)
 769                 return -ENOSPC;
 770
 771         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas);
 772         sb_r = bch2_sb_get_replicas_v0(c->disk_sb.sb);
 773
 774         memset(&sb_r->entries, 0,
 775                vstruct_end(&sb_r->field) -
 776                (void *) &sb_r->entries);
 777
 778         dst = sb_r->entries;
 779         for_each_cpu_replicas_entry(r, src) {
 780                 dst->data_type  = src->data_type;
 781                 dst->nr_devs    = src->nr_devs;
 782                 memcpy(dst->devs, src->devs, src->nr_devs);
 783
 784                 dst = replicas_entry_next(dst);
 785
 786                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 787         }
 788
 789         return 0;
 790 }
 791
 792 static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *c,
 793                                             struct bch_replicas_cpu *r)
 794 {
 795         struct bch_sb_field_replicas *sb_r;
 796         struct bch_replicas_entry *dst, *src;
 797         bool need_v1 = false;
 798         size_t bytes;
 799
 800         bytes = sizeof(struct bch_sb_field_replicas);
 801
 802         for_each_cpu_replicas_entry(r, src) {
 803                 bytes += replicas_entry_bytes(src);
 804                 if (src->nr_required != 1)
 805                         need_v1 = true;
 806         }
 807
 808         if (!need_v1)
 809                 return bch2_cpu_replicas_to_sb_replicas_v0(c, r);
 810
 811         sb_r = bch2_sb_resize_replicas(&c->disk_sb,
 812                         DIV_ROUND_UP(bytes, sizeof(u64)));
 813         if (!sb_r)
 814                 return -ENOSPC;
 815
 816         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas_v0);
 817         sb_r = bch2_sb_get_replicas(c->disk_sb.sb);
 818
 819         memset(&sb_r->entries, 0,
 820                vstruct_end(&sb_r->field) -
 821                (void *) &sb_r->entries);
 822
 823         dst = sb_r->entries;
 824         for_each_cpu_replicas_entry(r, src) {
 825                 memcpy(dst, src, replicas_entry_bytes(src));
 826
 827                 dst = replicas_entry_next(dst);
 828
 829                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 830         }
 831
 832         return 0;
 833 }
 834
 835 static const char *check_dup_replicas_entries(struct bch_replicas_cpu *cpu_r)
 836 {
 837         unsigned i;
 838
 839         sort_cmp_size(cpu_r->entries,
 840                       cpu_r->nr,
 841                       cpu_r->entry_size,
 842                       memcmp, NULL);
 843
 844         for (i = 0; i + 1 < cpu_r->nr; i++) {
 845                 struct bch_replicas_entry *l =
 846                         cpu_replicas_entry(cpu_r, i);
 847                 struct bch_replicas_entry *r =
 848                         cpu_replicas_entry(cpu_r, i + 1);
 849
 850                 BUG_ON(memcmp(l, r, cpu_r->entry_size) > 0);
 851
 852                 if (!memcmp(l, r, cpu_r->entry_size))
 853                         return "duplicate replicas entry";
 854         }
 855
 856         return NULL;
 857 }
 858
 859 static const char *bch2_sb_validate_replicas(struct bch_sb *sb, struct bch_sb_field *f)
 860 {
 861         struct bch_sb_field_replicas *sb_r = field_to_type(f, replicas);
 862         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 863         struct bch_replicas_cpu cpu_r = { .entries = NULL };
 864         struct bch_replicas_entry *e;
 865         const char *err;
 866         unsigned i;
 867
 868         for_each_replicas_entry(sb_r, e) {
 869                 err = "invalid replicas entry: invalid data type";
 870                 if (e->data_type >= BCH_DATA_NR)
 871                         goto err;
 872
 873                 err = "invalid replicas entry: no devices";
 874                 if (!e->nr_devs)
 875                         goto err;
 876
 877                 err = "invalid replicas entry: bad nr_required";
 878                 if (!e->nr_required ||
 879                     (e->nr_required > 1 &&
 880                      e->nr_required >= e->nr_devs))
 881                         goto err;
 882
 883                 err = "invalid replicas entry: invalid device";
 884                 for (i = 0; i < e->nr_devs; i++)
 885                         if (!bch2_dev_exists(sb, mi, e->devs[i]))
 886                                 goto err;
 887         }
 888
 889         err = "cannot allocate memory";
 890         if (__bch2_sb_replicas_to_cpu_replicas(sb_r, &cpu_r))
 891                 goto err;
 892
 893         err = check_dup_replicas_entries(&cpu_r);
 894 err:
 895         kfree(cpu_r.entries);
 896         return err;
 897 }
 898
 899 static void bch2_sb_replicas_to_text(struct printbuf *out,
 900                                      struct bch_sb *sb,
 901                                      struct bch_sb_field *f)
 902 {
 903         struct bch_sb_field_replicas *r = field_to_type(f, replicas);
 904         struct bch_replicas_entry *e;
 905         bool first = true;
 906
 907         for_each_replicas_entry(r, e) {
 908                 if (!first)
 909                         pr_buf(out, " ");
 910                 first = false;
 911
 912                 bch2_replicas_entry_to_text(out, e);
 913         }
 914 }
 915
 916 const struct bch_sb_field_ops bch_sb_field_ops_replicas = {
 917         .validate       = bch2_sb_validate_replicas,
 918         .to_text        = bch2_sb_replicas_to_text,
 919 };
 920
 921 static const char *bch2_sb_validate_replicas_v0(struct bch_sb *sb, struct bch_sb_field *f)
 922 {
 923         struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0);
 924         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 925         struct bch_replicas_cpu cpu_r = { .entries = NULL };
 926         struct bch_replicas_entry_v0 *e;
 927         const char *err;
 928         unsigned i;
 929
 930         for_each_replicas_entry_v0(sb_r, e) {
 931                 err = "invalid replicas entry: invalid data type";
 932                 if (e->data_type >= BCH_DATA_NR)
 933                         goto err;
 934
 935                 err = "invalid replicas entry: no devices";
 936                 if (!e->nr_devs)
 937                         goto err;
 938
 939                 err = "invalid replicas entry: invalid device";
 940                 for (i = 0; i < e->nr_devs; i++)
 941                         if (!bch2_dev_exists(sb, mi, e->devs[i]))
 942                                 goto err;
 943         }
 944
 945         err = "cannot allocate memory";
 946         if (__bch2_sb_replicas_v0_to_cpu_replicas(sb_r, &cpu_r))
 947                 goto err;
 948
 949         err = check_dup_replicas_entries(&cpu_r);
 950 err:
 951         kfree(cpu_r.entries);
 952         return err;
 953 }
 954
 955 const struct bch_sb_field_ops bch_sb_field_ops_replicas_v0 = {
 956         .validate       = bch2_sb_validate_replicas_v0,
 957 };
 958
 959 /* Query replicas: */
 960
 961 struct replicas_status __bch2_replicas_status(struct bch_fs *c,
 962                                               struct bch_devs_mask online_devs)
 963 {
 964         struct bch_sb_field_members *mi;
 965         struct bch_replicas_entry *e;
 966         unsigned i, nr_online, nr_offline;
 967         struct replicas_status ret;
 968
 969         memset(&ret, 0, sizeof(ret));
 970
 971         for (i = 0; i < ARRAY_SIZE(ret.replicas); i++)
 972                 ret.replicas[i].redundancy = INT_MAX;
 973
 974         mi = bch2_sb_get_members(c->disk_sb.sb);
 975
 976         percpu_down_read(&c->mark_lock);
 977
 978         for_each_cpu_replicas_entry(&c->replicas, e) {
 979                 if (e->data_type >= ARRAY_SIZE(ret.replicas))
 980                         panic("e %p data_type %u\n", e, e->data_type);
 981
 982                 nr_online = nr_offline = 0;
 983
 984                 for (i = 0; i < e->nr_devs; i++) {
 985                         BUG_ON(!bch2_dev_exists(c->disk_sb.sb, mi,
 986                                                 e->devs[i]));
 987
 988                         if (test_bit(e->devs[i], online_devs.d))
 989                                 nr_online++;
 990                         else
 991                                 nr_offline++;
 992                 }
 993
 994                 ret.replicas[e->data_type].redundancy =
 995                         min(ret.replicas[e->data_type].redundancy,
 996                             (int) nr_online - (int) e->nr_required);
 997
 998                 ret.replicas[e->data_type].nr_offline =
 999                         max(ret.replicas[e->data_type].nr_offline,
1000                             nr_offline);
1001         }
1002
1003         percpu_up_read(&c->mark_lock);
1004
1005         for (i = 0; i < ARRAY_SIZE(ret.replicas); i++)
1006                 if (ret.replicas[i].redundancy == INT_MAX)
1007                         ret.replicas[i].redundancy = 0;
1008
1009         return ret;
1010 }
1011
1012 struct replicas_status bch2_replicas_status(struct bch_fs *c)
1013 {
1014         return __bch2_replicas_status(c, bch2_online_devs(c));
1015 }
1016
1017 static bool have_enough_devs(struct replicas_status s,
1018                              enum bch_data_type type,
1019                              bool force_if_degraded,
1020                              bool force_if_lost)
1021 {
1022         return (!s.replicas[type].nr_offline || force_if_degraded) &&
1023                 (s.replicas[type].redundancy >= 0 || force_if_lost);
1024 }
1025
1026 bool bch2_have_enough_devs(struct replicas_status s, unsigned flags)
1027 {
1028         return (have_enough_devs(s, BCH_DATA_JOURNAL,
1029                                  flags & BCH_FORCE_IF_METADATA_DEGRADED,
1030                                  flags & BCH_FORCE_IF_METADATA_LOST) &&
1031                 have_enough_devs(s, BCH_DATA_BTREE,
1032                                  flags & BCH_FORCE_IF_METADATA_DEGRADED,
1033                                  flags & BCH_FORCE_IF_METADATA_LOST) &&
1034                 have_enough_devs(s, BCH_DATA_USER,
1035                                  flags & BCH_FORCE_IF_DATA_DEGRADED,
1036                                  flags & BCH_FORCE_IF_DATA_LOST));
1037 }
1038
1039 int bch2_replicas_online(struct bch_fs *c, bool meta)
1040 {
1041         struct replicas_status s = bch2_replicas_status(c);
1042
1043         return (meta
1044                 ? min(s.replicas[BCH_DATA_JOURNAL].redundancy,
1045                       s.replicas[BCH_DATA_BTREE].redundancy)
1046                 : s.replicas[BCH_DATA_USER].redundancy) + 1;
1047 }
1048
1049 unsigned bch2_dev_has_data(struct bch_fs *c, struct bch_dev *ca)
1050 {
1051         struct bch_replicas_entry *e;
1052         unsigned i, ret = 0;
1053
1054         percpu_down_read(&c->mark_lock);
1055
1056         for_each_cpu_replicas_entry(&c->replicas, e)
1057                 for (i = 0; i < e->nr_devs; i++)
1058                         if (e->devs[i] == ca->dev_idx)
1059                                 ret |= 1 << e->data_type;
1060
1061         percpu_up_read(&c->mark_lock);
1062
1063         return ret;
1064 }
1065
1066 int bch2_fs_replicas_init(struct bch_fs *c)
1067 {
1068         c->journal.entry_u64s_reserved +=
1069                 reserve_journal_replicas(c, &c->replicas);
1070
1071         return replicas_table_update(c, &c->replicas);
1072 }