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