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[JOURNAL_BUF_NR];
 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 i, bytes = sizeof(struct bch_fs_usage) +
 283                 sizeof(u64) * new_r->nr;
 284         int ret = 0;
 285
 286         memset(new_usage, 0, sizeof(new_usage));
 287
 288         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 289                 if (!(new_usage[i] = __alloc_percpu_gfp(bytes,
 290                                         sizeof(u64), GFP_NOIO)))
 291                         goto err;
 292
 293         if (!(new_base = kzalloc(bytes, GFP_NOIO)) ||
 294             !(new_scratch  = kmalloc(bytes, GFP_NOIO)) ||
 295             (c->usage_gc &&
 296              !(new_gc = __alloc_percpu_gfp(bytes, sizeof(u64), GFP_NOIO))))
 297                 goto err;
 298
 299         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 300                 if (c->usage[i])
 301                         __replicas_table_update_pcpu(new_usage[i], new_r,
 302                                                      c->usage[i], &c->replicas);
 303         if (c->usage_base)
 304                 __replicas_table_update(new_base,               new_r,
 305                                         c->usage_base,          &c->replicas);
 306         if (c->usage_gc)
 307                 __replicas_table_update_pcpu(new_gc,            new_r,
 308                                              c->usage_gc,       &c->replicas);
 309
 310         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 311                 swap(c->usage[i],       new_usage[i]);
 312         swap(c->usage_base,     new_base);
 313         swap(c->usage_scratch,  new_scratch);
 314         swap(c->usage_gc,       new_gc);
 315         swap(c->replicas,       *new_r);
 316 out:
 317         free_percpu(new_gc);
 318         kfree(new_scratch);
 319         free_percpu(new_usage[1]);
 320         free_percpu(new_usage[0]);
 321         kfree(new_base);
 322         return ret;
 323 err:
 324         bch_err(c, "error updating replicas table: memory allocation failure");
 325         ret = -ENOMEM;
 326         goto out;
 327 }
 328
 329 static unsigned reserve_journal_replicas(struct bch_fs *c,
 330                                      struct bch_replicas_cpu *r)
 331 {
 332         struct bch_replicas_entry *e;
 333         unsigned journal_res_u64s = 0;
 334
 335         /* nr_inodes: */
 336         journal_res_u64s +=
 337                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 338
 339         /* key_version: */
 340         journal_res_u64s +=
 341                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 342
 343         /* persistent_reserved: */
 344         journal_res_u64s +=
 345                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64)) *
 346                 BCH_REPLICAS_MAX;
 347
 348         for_each_cpu_replicas_entry(r, e)
 349                 journal_res_u64s +=
 350                         DIV_ROUND_UP(sizeof(struct jset_entry_data_usage) +
 351                                      e->nr_devs, sizeof(u64));
 352         return journal_res_u64s;
 353 }
 354
 355 noinline
 356 static int bch2_mark_replicas_slowpath(struct bch_fs *c,
 357                                 struct bch_replicas_entry *new_entry)
 358 {
 359         struct bch_replicas_cpu new_r, new_gc;
 360         int ret = 0;
 361
 362         verify_replicas_entry(new_entry);
 363
 364         memset(&new_r, 0, sizeof(new_r));
 365         memset(&new_gc, 0, sizeof(new_gc));
 366
 367         mutex_lock(&c->sb_lock);
 368
 369         if (c->replicas_gc.entries &&
 370             !__replicas_has_entry(&c->replicas_gc, new_entry)) {
 371                 new_gc = cpu_replicas_add_entry(&c->replicas_gc, new_entry);
 372                 if (!new_gc.entries)
 373                         goto err;
 374         }
 375
 376         if (!__replicas_has_entry(&c->replicas, new_entry)) {
 377                 new_r = cpu_replicas_add_entry(&c->replicas, new_entry);
 378                 if (!new_r.entries)
 379                         goto err;
 380
 381                 ret = bch2_cpu_replicas_to_sb_replicas(c, &new_r);
 382                 if (ret)
 383                         goto err;
 384
 385                 bch2_journal_entry_res_resize(&c->journal,
 386                                 &c->replicas_journal_res,
 387                                 reserve_journal_replicas(c, &new_r));
 388         }
 389
 390         if (!new_r.entries &&
 391             !new_gc.entries)
 392                 goto out;
 393
 394         /* allocations done, now commit: */
 395
 396         if (new_r.entries)
 397                 bch2_write_super(c);
 398
 399         /* don't update in memory replicas until changes are persistent */
 400         percpu_down_write(&c->mark_lock);
 401         if (new_r.entries)
 402                 ret = replicas_table_update(c, &new_r);
 403         if (new_gc.entries)
 404                 swap(new_gc, c->replicas_gc);
 405         percpu_up_write(&c->mark_lock);
 406 out:
 407         mutex_unlock(&c->sb_lock);
 408
 409         kfree(new_r.entries);
 410         kfree(new_gc.entries);
 411
 412         return ret;
 413 err:
 414         bch_err(c, "error adding replicas entry: memory allocation failure");
 415         ret = -ENOMEM;
 416         goto out;
 417 }
 418
 419 static int __bch2_mark_replicas(struct bch_fs *c,
 420                                 struct bch_replicas_entry *r,
 421                                 bool check)
 422 {
 423         return likely(bch2_replicas_marked(c, r))       ? 0
 424                 : check                                 ? -1
 425                 : bch2_mark_replicas_slowpath(c, r);
 426 }
 427
 428 int bch2_mark_replicas(struct bch_fs *c, struct bch_replicas_entry *r)
 429 {
 430         return __bch2_mark_replicas(c, r, false);
 431 }
 432
 433 static int __bch2_mark_bkey_replicas(struct bch_fs *c, struct bkey_s_c k,
 434                                      bool check)
 435 {
 436         struct bch_replicas_padded search;
 437         struct bch_devs_list cached = bch2_bkey_cached_devs(k);
 438         unsigned i;
 439         int ret;
 440
 441         for (i = 0; i < cached.nr; i++) {
 442                 bch2_replicas_entry_cached(&search.e, cached.devs[i]);
 443
 444                 ret = __bch2_mark_replicas(c, &search.e, check);
 445                 if (ret)
 446                         return ret;
 447         }
 448
 449         bch2_bkey_to_replicas(&search.e, k);
 450
 451         ret = __bch2_mark_replicas(c, &search.e, check);
 452         if (ret)
 453                 return ret;
 454
 455         if (search.e.data_type == BCH_DATA_parity) {
 456                 search.e.data_type = BCH_DATA_cached;
 457                 ret = __bch2_mark_replicas(c, &search.e, check);
 458                 if (ret)
 459                         return ret;
 460
 461                 search.e.data_type = BCH_DATA_user;
 462                 ret = __bch2_mark_replicas(c, &search.e, check);
 463                 if (ret)
 464                         return ret;
 465         }
 466
 467         return 0;
 468 }
 469
 470 bool bch2_bkey_replicas_marked(struct bch_fs *c,
 471                                struct bkey_s_c k)
 472 {
 473         return __bch2_mark_bkey_replicas(c, k, true) == 0;
 474 }
 475
 476 int bch2_mark_bkey_replicas(struct bch_fs *c, struct bkey_s_c k)
 477 {
 478         return __bch2_mark_bkey_replicas(c, k, false);
 479 }
 480
 481 int bch2_replicas_gc_end(struct bch_fs *c, int ret)
 482 {
 483         unsigned i;
 484
 485         lockdep_assert_held(&c->replicas_gc_lock);
 486
 487         mutex_lock(&c->sb_lock);
 488         percpu_down_write(&c->mark_lock);
 489
 490         /*
 491          * this is kind of crappy; the replicas gc mechanism needs to be ripped
 492          * out
 493          */
 494
 495         for (i = 0; i < c->replicas.nr; i++) {
 496                 struct bch_replicas_entry *e =
 497                         cpu_replicas_entry(&c->replicas, i);
 498                 struct bch_replicas_cpu n;
 499
 500                 if (!__replicas_has_entry(&c->replicas_gc, e) &&
 501                     bch2_fs_usage_read_one(c, &c->usage_base->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                     bch2_fs_usage_read_one(c, &c->usage_base->replicas[i]))
 607                         memcpy(cpu_replicas_entry(&new, new.nr++),
 608                                e, new.entry_size);
 609         }
 610
 611         bch2_cpu_replicas_sort(&new);
 612
 613         if (bch2_cpu_replicas_to_sb_replicas(c, &new)) {
 614                 ret = -ENOSPC;
 615                 goto err;
 616         }
 617
 618         ret = replicas_table_update(c, &new);
 619 err:
 620         kfree(new.entries);
 621
 622         percpu_up_write(&c->mark_lock);
 623
 624         if (!ret)
 625                 bch2_write_super(c);
 626
 627         mutex_unlock(&c->sb_lock);
 628
 629         return ret;
 630 }
 631
 632 int bch2_replicas_set_usage(struct bch_fs *c,
 633                             struct bch_replicas_entry *r,
 634                             u64 sectors)
 635 {
 636         int ret, idx = bch2_replicas_entry_idx(c, r);
 637
 638         if (idx < 0) {
 639                 struct bch_replicas_cpu n;
 640
 641                 n = cpu_replicas_add_entry(&c->replicas, r);
 642                 if (!n.entries)
 643                         return -ENOMEM;
 644
 645                 ret = replicas_table_update(c, &n);
 646                 if (ret)
 647                         return ret;
 648
 649                 kfree(n.entries);
 650
 651                 idx = bch2_replicas_entry_idx(c, r);
 652                 BUG_ON(ret < 0);
 653         }
 654
 655         c->usage_base->replicas[idx] = sectors;
 656
 657         return 0;
 658 }
 659
 660 /* Replicas tracking - superblock: */
 661
 662 static int
 663 __bch2_sb_replicas_to_cpu_replicas(struct bch_sb_field_replicas *sb_r,
 664                                    struct bch_replicas_cpu *cpu_r)
 665 {
 666         struct bch_replicas_entry *e, *dst;
 667         unsigned nr = 0, entry_size = 0, idx = 0;
 668
 669         for_each_replicas_entry(sb_r, e) {
 670                 entry_size = max_t(unsigned, entry_size,
 671                                    replicas_entry_bytes(e));
 672                 nr++;
 673         }
 674
 675         cpu_r->entries = kcalloc(nr, entry_size, GFP_NOIO);
 676         if (!cpu_r->entries)
 677                 return -ENOMEM;
 678
 679         cpu_r->nr               = nr;
 680         cpu_r->entry_size       = entry_size;
 681
 682         for_each_replicas_entry(sb_r, e) {
 683                 dst = cpu_replicas_entry(cpu_r, idx++);
 684                 memcpy(dst, e, replicas_entry_bytes(e));
 685                 replicas_entry_sort(dst);
 686         }
 687
 688         return 0;
 689 }
 690
 691 static int
 692 __bch2_sb_replicas_v0_to_cpu_replicas(struct bch_sb_field_replicas_v0 *sb_r,
 693                                       struct bch_replicas_cpu *cpu_r)
 694 {
 695         struct bch_replicas_entry_v0 *e;
 696         unsigned nr = 0, entry_size = 0, idx = 0;
 697
 698         for_each_replicas_entry(sb_r, e) {
 699                 entry_size = max_t(unsigned, entry_size,
 700                                    replicas_entry_bytes(e));
 701                 nr++;
 702         }
 703
 704         entry_size += sizeof(struct bch_replicas_entry) -
 705                 sizeof(struct bch_replicas_entry_v0);
 706
 707         cpu_r->entries = kcalloc(nr, entry_size, GFP_NOIO);
 708         if (!cpu_r->entries)
 709                 return -ENOMEM;
 710
 711         cpu_r->nr               = nr;
 712         cpu_r->entry_size       = entry_size;
 713
 714         for_each_replicas_entry(sb_r, e) {
 715                 struct bch_replicas_entry *dst =
 716                         cpu_replicas_entry(cpu_r, idx++);
 717
 718                 dst->data_type  = e->data_type;
 719                 dst->nr_devs    = e->nr_devs;
 720                 dst->nr_required = 1;
 721                 memcpy(dst->devs, e->devs, e->nr_devs);
 722                 replicas_entry_sort(dst);
 723         }
 724
 725         return 0;
 726 }
 727
 728 int bch2_sb_replicas_to_cpu_replicas(struct bch_fs *c)
 729 {
 730         struct bch_sb_field_replicas *sb_v1;
 731         struct bch_sb_field_replicas_v0 *sb_v0;
 732         struct bch_replicas_cpu new_r = { 0, 0, NULL };
 733         int ret = 0;
 734
 735         if ((sb_v1 = bch2_sb_get_replicas(c->disk_sb.sb)))
 736                 ret = __bch2_sb_replicas_to_cpu_replicas(sb_v1, &new_r);
 737         else if ((sb_v0 = bch2_sb_get_replicas_v0(c->disk_sb.sb)))
 738                 ret = __bch2_sb_replicas_v0_to_cpu_replicas(sb_v0, &new_r);
 739
 740         if (ret)
 741                 return -ENOMEM;
 742
 743         bch2_cpu_replicas_sort(&new_r);
 744
 745         percpu_down_write(&c->mark_lock);
 746
 747         ret = replicas_table_update(c, &new_r);
 748         percpu_up_write(&c->mark_lock);
 749
 750         kfree(new_r.entries);
 751
 752         return 0;
 753 }
 754
 755 static int bch2_cpu_replicas_to_sb_replicas_v0(struct bch_fs *c,
 756                                                struct bch_replicas_cpu *r)
 757 {
 758         struct bch_sb_field_replicas_v0 *sb_r;
 759         struct bch_replicas_entry_v0 *dst;
 760         struct bch_replicas_entry *src;
 761         size_t bytes;
 762
 763         bytes = sizeof(struct bch_sb_field_replicas);
 764
 765         for_each_cpu_replicas_entry(r, src)
 766                 bytes += replicas_entry_bytes(src) - 1;
 767
 768         sb_r = bch2_sb_resize_replicas_v0(&c->disk_sb,
 769                         DIV_ROUND_UP(bytes, sizeof(u64)));
 770         if (!sb_r)
 771                 return -ENOSPC;
 772
 773         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas);
 774         sb_r = bch2_sb_get_replicas_v0(c->disk_sb.sb);
 775
 776         memset(&sb_r->entries, 0,
 777                vstruct_end(&sb_r->field) -
 778                (void *) &sb_r->entries);
 779
 780         dst = sb_r->entries;
 781         for_each_cpu_replicas_entry(r, src) {
 782                 dst->data_type  = src->data_type;
 783                 dst->nr_devs    = src->nr_devs;
 784                 memcpy(dst->devs, src->devs, src->nr_devs);
 785
 786                 dst = replicas_entry_next(dst);
 787
 788                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 789         }
 790
 791         return 0;
 792 }
 793
 794 static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *c,
 795                                             struct bch_replicas_cpu *r)
 796 {
 797         struct bch_sb_field_replicas *sb_r;
 798         struct bch_replicas_entry *dst, *src;
 799         bool need_v1 = false;
 800         size_t bytes;
 801
 802         bytes = sizeof(struct bch_sb_field_replicas);
 803
 804         for_each_cpu_replicas_entry(r, src) {
 805                 bytes += replicas_entry_bytes(src);
 806                 if (src->nr_required != 1)
 807                         need_v1 = true;
 808         }
 809
 810         if (!need_v1)
 811                 return bch2_cpu_replicas_to_sb_replicas_v0(c, r);
 812
 813         sb_r = bch2_sb_resize_replicas(&c->disk_sb,
 814                         DIV_ROUND_UP(bytes, sizeof(u64)));
 815         if (!sb_r)
 816                 return -ENOSPC;
 817
 818         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas_v0);
 819         sb_r = bch2_sb_get_replicas(c->disk_sb.sb);
 820
 821         memset(&sb_r->entries, 0,
 822                vstruct_end(&sb_r->field) -
 823                (void *) &sb_r->entries);
 824
 825         dst = sb_r->entries;
 826         for_each_cpu_replicas_entry(r, src) {
 827                 memcpy(dst, src, replicas_entry_bytes(src));
 828
 829                 dst = replicas_entry_next(dst);
 830
 831                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 832         }
 833
 834         return 0;
 835 }
 836
 837 static const char *check_dup_replicas_entries(struct bch_replicas_cpu *cpu_r)
 838 {
 839         unsigned i;
 840
 841         sort_cmp_size(cpu_r->entries,
 842                       cpu_r->nr,
 843                       cpu_r->entry_size,
 844                       memcmp, NULL);
 845
 846         for (i = 0; i + 1 < cpu_r->nr; i++) {
 847                 struct bch_replicas_entry *l =
 848                         cpu_replicas_entry(cpu_r, i);
 849                 struct bch_replicas_entry *r =
 850                         cpu_replicas_entry(cpu_r, i + 1);
 851
 852                 BUG_ON(memcmp(l, r, cpu_r->entry_size) > 0);
 853
 854                 if (!memcmp(l, r, cpu_r->entry_size))
 855                         return "duplicate replicas entry";
 856         }
 857
 858         return NULL;
 859 }
 860
 861 static const char *bch2_sb_validate_replicas(struct bch_sb *sb, struct bch_sb_field *f)
 862 {
 863         struct bch_sb_field_replicas *sb_r = field_to_type(f, replicas);
 864         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 865         struct bch_replicas_cpu cpu_r = { .entries = NULL };
 866         struct bch_replicas_entry *e;
 867         const char *err;
 868         unsigned i;
 869
 870         for_each_replicas_entry(sb_r, e) {
 871                 err = "invalid replicas entry: invalid data type";
 872                 if (e->data_type >= BCH_DATA_NR)
 873                         goto err;
 874
 875                 err = "invalid replicas entry: no devices";
 876                 if (!e->nr_devs)
 877                         goto err;
 878
 879                 err = "invalid replicas entry: bad nr_required";
 880                 if (e->nr_required > 1 &&
 881                     e->nr_required >= e->nr_devs)
 882                         goto err;
 883
 884                 err = "invalid replicas entry: invalid device";
 885                 for (i = 0; i < e->nr_devs; i++)
 886                         if (!bch2_dev_exists(sb, mi, e->devs[i]))
 887                                 goto err;
 888         }
 889
 890         err = "cannot allocate memory";
 891         if (__bch2_sb_replicas_to_cpu_replicas(sb_r, &cpu_r))
 892                 goto err;
 893
 894         err = check_dup_replicas_entries(&cpu_r);
 895 err:
 896         kfree(cpu_r.entries);
 897         return err;
 898 }
 899
 900 static void bch2_sb_replicas_to_text(struct printbuf *out,
 901                                      struct bch_sb *sb,
 902                                      struct bch_sb_field *f)
 903 {
 904         struct bch_sb_field_replicas *r = field_to_type(f, replicas);
 905         struct bch_replicas_entry *e;
 906         bool first = true;
 907
 908         for_each_replicas_entry(r, e) {
 909                 if (!first)
 910                         pr_buf(out, " ");
 911                 first = false;
 912
 913                 bch2_replicas_entry_to_text(out, e);
 914         }
 915 }
 916
 917 const struct bch_sb_field_ops bch_sb_field_ops_replicas = {
 918         .validate       = bch2_sb_validate_replicas,
 919         .to_text        = bch2_sb_replicas_to_text,
 920 };
 921
 922 static const char *bch2_sb_validate_replicas_v0(struct bch_sb *sb, struct bch_sb_field *f)
 923 {
 924         struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0);
 925         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 926         struct bch_replicas_cpu cpu_r = { .entries = NULL };
 927         struct bch_replicas_entry_v0 *e;
 928         const char *err;
 929         unsigned i;
 930
 931         for_each_replicas_entry_v0(sb_r, e) {
 932                 err = "invalid replicas entry: invalid data type";
 933                 if (e->data_type >= BCH_DATA_NR)
 934                         goto err;
 935
 936                 err = "invalid replicas entry: no devices";
 937                 if (!e->nr_devs)
 938                         goto err;
 939
 940                 err = "invalid replicas entry: invalid device";
 941                 for (i = 0; i < e->nr_devs; i++)
 942                         if (!bch2_dev_exists(sb, mi, e->devs[i]))
 943                                 goto err;
 944         }
 945
 946         err = "cannot allocate memory";
 947         if (__bch2_sb_replicas_v0_to_cpu_replicas(sb_r, &cpu_r))
 948                 goto err;
 949
 950         err = check_dup_replicas_entries(&cpu_r);
 951 err:
 952         kfree(cpu_r.entries);
 953         return err;
 954 }
 955
 956 const struct bch_sb_field_ops bch_sb_field_ops_replicas_v0 = {
 957         .validate       = bch2_sb_validate_replicas_v0,
 958 };
 959
 960 /* Query replicas: */
 961
 962 struct replicas_status __bch2_replicas_status(struct bch_fs *c,
 963                                               struct bch_devs_mask online_devs)
 964 {
 965         struct bch_sb_field_members *mi;
 966         struct bch_replicas_entry *e;
 967         unsigned i, nr_online, nr_offline;
 968         struct replicas_status ret;
 969
 970         memset(&ret, 0, sizeof(ret));
 971
 972         for (i = 0; i < ARRAY_SIZE(ret.replicas); i++)
 973                 ret.replicas[i].redundancy = INT_MAX;
 974
 975         mi = bch2_sb_get_members(c->disk_sb.sb);
 976
 977         percpu_down_read(&c->mark_lock);
 978
 979         for_each_cpu_replicas_entry(&c->replicas, e) {
 980                 if (e->data_type >= ARRAY_SIZE(ret.replicas))
 981                         panic("e %p data_type %u\n", e, e->data_type);
 982
 983                 nr_online = nr_offline = 0;
 984
 985                 for (i = 0; i < e->nr_devs; i++) {
 986                         BUG_ON(!bch2_dev_exists(c->disk_sb.sb, mi,
 987                                                 e->devs[i]));
 988
 989                         if (test_bit(e->devs[i], online_devs.d))
 990                                 nr_online++;
 991                         else
 992                                 nr_offline++;
 993                 }
 994
 995                 ret.replicas[e->data_type].redundancy =
 996                         min(ret.replicas[e->data_type].redundancy,
 997                             (int) nr_online - (int) e->nr_required);
 998
 999                 ret.replicas[e->data_type].nr_offline =
1000                         max(ret.replicas[e->data_type].nr_offline,
1001                             nr_offline);
1002         }
1003
1004         percpu_up_read(&c->mark_lock);
1005
1006         for (i = 0; i < ARRAY_SIZE(ret.replicas); i++)
1007                 if (ret.replicas[i].redundancy == INT_MAX)
1008                         ret.replicas[i].redundancy = 0;
1009
1010         return ret;
1011 }
1012
1013 struct replicas_status bch2_replicas_status(struct bch_fs *c)
1014 {
1015         return __bch2_replicas_status(c, bch2_online_devs(c));
1016 }
1017
1018 static bool have_enough_devs(struct replicas_status s,
1019                              enum bch_data_type type,
1020                              bool force_if_degraded,
1021                              bool force_if_lost)
1022 {
1023         return (!s.replicas[type].nr_offline || force_if_degraded) &&
1024                 (s.replicas[type].redundancy >= 0 || force_if_lost);
1025 }
1026
1027 bool bch2_have_enough_devs(struct replicas_status s, unsigned flags)
1028 {
1029         return (have_enough_devs(s, BCH_DATA_journal,
1030                                  flags & BCH_FORCE_IF_METADATA_DEGRADED,
1031                                  flags & BCH_FORCE_IF_METADATA_LOST) &&
1032                 have_enough_devs(s, BCH_DATA_btree,
1033                                  flags & BCH_FORCE_IF_METADATA_DEGRADED,
1034                                  flags & BCH_FORCE_IF_METADATA_LOST) &&
1035                 have_enough_devs(s, BCH_DATA_user,
1036                                  flags & BCH_FORCE_IF_DATA_DEGRADED,
1037                                  flags & BCH_FORCE_IF_DATA_LOST));
1038 }
1039
1040 int bch2_replicas_online(struct bch_fs *c, bool meta)
1041 {
1042         struct replicas_status s = bch2_replicas_status(c);
1043
1044         return (meta
1045                 ? min(s.replicas[BCH_DATA_journal].redundancy,
1046                       s.replicas[BCH_DATA_btree].redundancy)
1047                 : s.replicas[BCH_DATA_user].redundancy) + 1;
1048 }
1049
1050 unsigned bch2_dev_has_data(struct bch_fs *c, struct bch_dev *ca)
1051 {
1052         struct bch_replicas_entry *e;
1053         unsigned i, ret = 0;
1054
1055         percpu_down_read(&c->mark_lock);
1056
1057         for_each_cpu_replicas_entry(&c->replicas, e)
1058                 for (i = 0; i < e->nr_devs; i++)
1059                         if (e->devs[i] == ca->dev_idx)
1060                                 ret |= 1 << e->data_type;
1061
1062         percpu_up_read(&c->mark_lock);
1063
1064         return ret;
1065 }
1066
1067 int bch2_fs_replicas_init(struct bch_fs *c)
1068 {
1069         c->journal.entry_u64s_reserved +=
1070                 reserve_journal_replicas(c, &c->replicas);
1071
1072         return replicas_table_update(c, &c->replicas);
1073 }