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 void verify_replicas_entry(struct bch_replicas_entry *e)
  15 {
  16 #ifdef CONFIG_BCACHEFS_DEBUG
  17         unsigned i;
  18
  19         BUG_ON(e->data_type >= BCH_DATA_NR);
  20         BUG_ON(!e->nr_devs);
  21         BUG_ON(e->nr_required > 1 &&
  22                e->nr_required >= e->nr_devs);
  23
  24         for (i = 0; i + 1 < e->nr_devs; i++)
  25                 BUG_ON(e->devs[i] >= e->devs[i + 1]);
  26 #endif
  27 }
  28
  29 static void replicas_entry_sort(struct bch_replicas_entry *e)
  30 {
  31         bubble_sort(e->devs, e->nr_devs, u8_cmp);
  32 }
  33
  34 static void bch2_cpu_replicas_sort(struct bch_replicas_cpu *r)
  35 {
  36         eytzinger0_sort(r->entries, r->nr, r->entry_size, memcmp, NULL);
  37 }
  38
  39 void bch2_replicas_entry_to_text(struct printbuf *out,
  40                                  struct bch_replicas_entry *e)
  41 {
  42         unsigned i;
  43
  44         pr_buf(out, "%s: %u/%u [",
  45                bch2_data_types[e->data_type],
  46                e->nr_required,
  47                e->nr_devs);
  48
  49         for (i = 0; i < e->nr_devs; i++)
  50                 pr_buf(out, i ? " %u" : "%u", e->devs[i]);
  51         pr_buf(out, "]");
  52 }
  53
  54 void bch2_cpu_replicas_to_text(struct printbuf *out,
  55                               struct bch_replicas_cpu *r)
  56 {
  57         struct bch_replicas_entry *e;
  58         bool first = true;
  59
  60         for_each_cpu_replicas_entry(r, e) {
  61                 if (!first)
  62                         pr_buf(out, " ");
  63                 first = false;
  64
  65                 bch2_replicas_entry_to_text(out, e);
  66         }
  67 }
  68
  69 static void extent_to_replicas(struct bkey_s_c k,
  70                                struct bch_replicas_entry *r)
  71 {
  72         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
  73         const union bch_extent_entry *entry;
  74         struct extent_ptr_decoded p;
  75
  76         r->nr_required  = 1;
  77
  78         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
  79                 if (p.ptr.cached)
  80                         continue;
  81
  82                 if (!p.has_ec)
  83                         r->devs[r->nr_devs++] = p.ptr.dev;
  84                 else
  85                         r->nr_required = 0;
  86         }
  87 }
  88
  89 static void stripe_to_replicas(struct bkey_s_c k,
  90                                struct bch_replicas_entry *r)
  91 {
  92         struct bkey_s_c_stripe s = bkey_s_c_to_stripe(k);
  93         const struct bch_extent_ptr *ptr;
  94
  95         r->nr_required  = s.v->nr_blocks - s.v->nr_redundant;
  96
  97         for (ptr = s.v->ptrs;
  98              ptr < s.v->ptrs + s.v->nr_blocks;
  99              ptr++)
 100                 r->devs[r->nr_devs++] = ptr->dev;
 101 }
 102
 103 void bch2_bkey_to_replicas(struct bch_replicas_entry *e,
 104                            struct bkey_s_c k)
 105 {
 106         e->nr_devs = 0;
 107
 108         switch (k.k->type) {
 109         case KEY_TYPE_btree_ptr:
 110         case KEY_TYPE_btree_ptr_v2:
 111                 e->data_type = BCH_DATA_btree;
 112                 extent_to_replicas(k, e);
 113                 break;
 114         case KEY_TYPE_extent:
 115         case KEY_TYPE_reflink_v:
 116                 e->data_type = BCH_DATA_user;
 117                 extent_to_replicas(k, e);
 118                 break;
 119         case KEY_TYPE_stripe:
 120                 e->data_type = BCH_DATA_parity;
 121                 stripe_to_replicas(k, e);
 122                 break;
 123         }
 124
 125         replicas_entry_sort(e);
 126 }
 127
 128 void bch2_devlist_to_replicas(struct bch_replicas_entry *e,
 129                               enum bch_data_type data_type,
 130                               struct bch_devs_list devs)
 131 {
 132         unsigned i;
 133
 134         BUG_ON(!data_type ||
 135                data_type == BCH_DATA_sb ||
 136                data_type >= BCH_DATA_NR);
 137
 138         e->data_type    = data_type;
 139         e->nr_devs      = 0;
 140         e->nr_required  = 1;
 141
 142         for (i = 0; i < devs.nr; i++)
 143                 e->devs[e->nr_devs++] = devs.devs[i];
 144
 145         replicas_entry_sort(e);
 146 }
 147
 148 static struct bch_replicas_cpu
 149 cpu_replicas_add_entry(struct bch_replicas_cpu *old,
 150                        struct bch_replicas_entry *new_entry)
 151 {
 152         unsigned i;
 153         struct bch_replicas_cpu new = {
 154                 .nr             = old->nr + 1,
 155                 .entry_size     = max_t(unsigned, old->entry_size,
 156                                         replicas_entry_bytes(new_entry)),
 157         };
 158
 159         BUG_ON(!new_entry->data_type);
 160         verify_replicas_entry(new_entry);
 161
 162         new.entries = kcalloc(new.nr, new.entry_size, GFP_NOIO);
 163         if (!new.entries)
 164                 return new;
 165
 166         for (i = 0; i < old->nr; i++)
 167                 memcpy(cpu_replicas_entry(&new, i),
 168                        cpu_replicas_entry(old, i),
 169                        old->entry_size);
 170
 171         memcpy(cpu_replicas_entry(&new, old->nr),
 172                new_entry,
 173                replicas_entry_bytes(new_entry));
 174
 175         bch2_cpu_replicas_sort(&new);
 176         return new;
 177 }
 178
 179 static inline int __replicas_entry_idx(struct bch_replicas_cpu *r,
 180                                        struct bch_replicas_entry *search)
 181 {
 182         int idx, entry_size = replicas_entry_bytes(search);
 183
 184         if (unlikely(entry_size > r->entry_size))
 185                 return -1;
 186
 187         verify_replicas_entry(search);
 188
 189 #define entry_cmp(_l, _r, size) memcmp(_l, _r, entry_size)
 190         idx = eytzinger0_find(r->entries, r->nr, r->entry_size,
 191                               entry_cmp, search);
 192 #undef entry_cmp
 193
 194         return idx < r->nr ? idx : -1;
 195 }
 196
 197 int bch2_replicas_entry_idx(struct bch_fs *c,
 198                             struct bch_replicas_entry *search)
 199 {
 200         replicas_entry_sort(search);
 201
 202         return __replicas_entry_idx(&c->replicas, search);
 203 }
 204
 205 static bool __replicas_has_entry(struct bch_replicas_cpu *r,
 206                                  struct bch_replicas_entry *search)
 207 {
 208         return __replicas_entry_idx(r, search) >= 0;
 209 }
 210
 211 bool bch2_replicas_marked(struct bch_fs *c,
 212                           struct bch_replicas_entry *search)
 213 {
 214         bool marked;
 215
 216         if (!search->nr_devs)
 217                 return true;
 218
 219         verify_replicas_entry(search);
 220
 221         percpu_down_read(&c->mark_lock);
 222         marked = __replicas_has_entry(&c->replicas, search) &&
 223                 (likely((!c->replicas_gc.entries)) ||
 224                  __replicas_has_entry(&c->replicas_gc, search));
 225         percpu_up_read(&c->mark_lock);
 226
 227         return marked;
 228 }
 229
 230 static void __replicas_table_update(struct bch_fs_usage *dst,
 231                                     struct bch_replicas_cpu *dst_r,
 232                                     struct bch_fs_usage *src,
 233                                     struct bch_replicas_cpu *src_r)
 234 {
 235         int src_idx, dst_idx;
 236
 237         *dst = *src;
 238
 239         for (src_idx = 0; src_idx < src_r->nr; src_idx++) {
 240                 if (!src->replicas[src_idx])
 241                         continue;
 242
 243                 dst_idx = __replicas_entry_idx(dst_r,
 244                                 cpu_replicas_entry(src_r, src_idx));
 245                 BUG_ON(dst_idx < 0);
 246
 247                 dst->replicas[dst_idx] = src->replicas[src_idx];
 248         }
 249 }
 250
 251 static void __replicas_table_update_pcpu(struct bch_fs_usage __percpu *dst_p,
 252                                     struct bch_replicas_cpu *dst_r,
 253                                     struct bch_fs_usage __percpu *src_p,
 254                                     struct bch_replicas_cpu *src_r)
 255 {
 256         unsigned src_nr = sizeof(struct bch_fs_usage) / sizeof(u64) + src_r->nr;
 257         struct bch_fs_usage *dst, *src = (void *)
 258                 bch2_acc_percpu_u64s((void *) src_p, src_nr);
 259
 260         preempt_disable();
 261         dst = this_cpu_ptr(dst_p);
 262         preempt_enable();
 263
 264         __replicas_table_update(dst, dst_r, src, src_r);
 265 }
 266
 267 /*
 268  * Resize filesystem accounting:
 269  */
 270 static int replicas_table_update(struct bch_fs *c,
 271                                  struct bch_replicas_cpu *new_r)
 272 {
 273         struct bch_fs_usage __percpu *new_usage[JOURNAL_BUF_NR];
 274         struct bch_fs_usage *new_scratch = NULL;
 275         struct bch_fs_usage __percpu *new_gc = NULL;
 276         struct bch_fs_usage *new_base = NULL;
 277         unsigned i, bytes = sizeof(struct bch_fs_usage) +
 278                 sizeof(u64) * new_r->nr;
 279         int ret = 0;
 280
 281         memset(new_usage, 0, sizeof(new_usage));
 282
 283         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 284                 if (!(new_usage[i] = __alloc_percpu_gfp(bytes,
 285                                         sizeof(u64), GFP_NOIO)))
 286                         goto err;
 287
 288         if (!(new_base = kzalloc(bytes, GFP_NOIO)) ||
 289             !(new_scratch  = kmalloc(bytes, GFP_NOIO)) ||
 290             (c->usage_gc &&
 291              !(new_gc = __alloc_percpu_gfp(bytes, sizeof(u64), GFP_NOIO))))
 292                 goto err;
 293
 294         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 295                 if (c->usage[i])
 296                         __replicas_table_update_pcpu(new_usage[i], new_r,
 297                                                      c->usage[i], &c->replicas);
 298         if (c->usage_base)
 299                 __replicas_table_update(new_base,               new_r,
 300                                         c->usage_base,          &c->replicas);
 301         if (c->usage_gc)
 302                 __replicas_table_update_pcpu(new_gc,            new_r,
 303                                              c->usage_gc,       &c->replicas);
 304
 305         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 306                 swap(c->usage[i],       new_usage[i]);
 307         swap(c->usage_base,     new_base);
 308         swap(c->usage_scratch,  new_scratch);
 309         swap(c->usage_gc,       new_gc);
 310         swap(c->replicas,       *new_r);
 311 out:
 312         free_percpu(new_gc);
 313         kfree(new_scratch);
 314         free_percpu(new_usage[1]);
 315         free_percpu(new_usage[0]);
 316         kfree(new_base);
 317         return ret;
 318 err:
 319         bch_err(c, "error updating replicas table: memory allocation failure");
 320         ret = -ENOMEM;
 321         goto out;
 322 }
 323
 324 static unsigned reserve_journal_replicas(struct bch_fs *c,
 325                                      struct bch_replicas_cpu *r)
 326 {
 327         struct bch_replicas_entry *e;
 328         unsigned journal_res_u64s = 0;
 329
 330         /* nr_inodes: */
 331         journal_res_u64s +=
 332                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 333
 334         /* key_version: */
 335         journal_res_u64s +=
 336                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 337
 338         /* persistent_reserved: */
 339         journal_res_u64s +=
 340                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64)) *
 341                 BCH_REPLICAS_MAX;
 342
 343         for_each_cpu_replicas_entry(r, e)
 344                 journal_res_u64s +=
 345                         DIV_ROUND_UP(sizeof(struct jset_entry_data_usage) +
 346                                      e->nr_devs, sizeof(u64));
 347         return journal_res_u64s;
 348 }
 349
 350 noinline
 351 static int bch2_mark_replicas_slowpath(struct bch_fs *c,
 352                                 struct bch_replicas_entry *new_entry)
 353 {
 354         struct bch_replicas_cpu new_r, new_gc;
 355         int ret = 0;
 356
 357         verify_replicas_entry(new_entry);
 358
 359         memset(&new_r, 0, sizeof(new_r));
 360         memset(&new_gc, 0, sizeof(new_gc));
 361
 362         mutex_lock(&c->sb_lock);
 363
 364         if (c->replicas_gc.entries &&
 365             !__replicas_has_entry(&c->replicas_gc, new_entry)) {
 366                 new_gc = cpu_replicas_add_entry(&c->replicas_gc, new_entry);
 367                 if (!new_gc.entries)
 368                         goto err;
 369         }
 370
 371         if (!__replicas_has_entry(&c->replicas, new_entry)) {
 372                 new_r = cpu_replicas_add_entry(&c->replicas, new_entry);
 373                 if (!new_r.entries)
 374                         goto err;
 375
 376                 ret = bch2_cpu_replicas_to_sb_replicas(c, &new_r);
 377                 if (ret)
 378                         goto err;
 379
 380                 bch2_journal_entry_res_resize(&c->journal,
 381                                 &c->replicas_journal_res,
 382                                 reserve_journal_replicas(c, &new_r));
 383         }
 384
 385         if (!new_r.entries &&
 386             !new_gc.entries)
 387                 goto out;
 388
 389         /* allocations done, now commit: */
 390
 391         if (new_r.entries)
 392                 bch2_write_super(c);
 393
 394         /* don't update in memory replicas until changes are persistent */
 395         percpu_down_write(&c->mark_lock);
 396         if (new_r.entries)
 397                 ret = replicas_table_update(c, &new_r);
 398         if (new_gc.entries)
 399                 swap(new_gc, c->replicas_gc);
 400         percpu_up_write(&c->mark_lock);
 401 out:
 402         mutex_unlock(&c->sb_lock);
 403
 404         kfree(new_r.entries);
 405         kfree(new_gc.entries);
 406
 407         return ret;
 408 err:
 409         bch_err(c, "error adding replicas entry: memory allocation failure");
 410         ret = -ENOMEM;
 411         goto out;
 412 }
 413
 414 static int __bch2_mark_replicas(struct bch_fs *c,
 415                                 struct bch_replicas_entry *r,
 416                                 bool check)
 417 {
 418         return likely(bch2_replicas_marked(c, r))       ? 0
 419                 : check                                 ? -1
 420                 : bch2_mark_replicas_slowpath(c, r);
 421 }
 422
 423 int bch2_mark_replicas(struct bch_fs *c, struct bch_replicas_entry *r)
 424 {
 425         return __bch2_mark_replicas(c, r, false);
 426 }
 427
 428 static int __bch2_mark_bkey_replicas(struct bch_fs *c, struct bkey_s_c k,
 429                                      bool check)
 430 {
 431         struct bch_replicas_padded search;
 432         struct bch_devs_list cached = bch2_bkey_cached_devs(k);
 433         unsigned i;
 434         int ret;
 435
 436         for (i = 0; i < cached.nr; i++) {
 437                 bch2_replicas_entry_cached(&search.e, cached.devs[i]);
 438
 439                 ret = __bch2_mark_replicas(c, &search.e, check);
 440                 if (ret)
 441                         return ret;
 442         }
 443
 444         bch2_bkey_to_replicas(&search.e, k);
 445
 446         ret = __bch2_mark_replicas(c, &search.e, check);
 447         if (ret)
 448                 return ret;
 449
 450         if (search.e.data_type == BCH_DATA_parity) {
 451                 search.e.data_type = BCH_DATA_cached;
 452                 ret = __bch2_mark_replicas(c, &search.e, check);
 453                 if (ret)
 454                         return ret;
 455
 456                 search.e.data_type = BCH_DATA_user;
 457                 ret = __bch2_mark_replicas(c, &search.e, check);
 458                 if (ret)
 459                         return ret;
 460         }
 461
 462         return 0;
 463 }
 464
 465 bool bch2_bkey_replicas_marked(struct bch_fs *c,
 466                                struct bkey_s_c k)
 467 {
 468         return __bch2_mark_bkey_replicas(c, k, true) == 0;
 469 }
 470
 471 int bch2_mark_bkey_replicas(struct bch_fs *c, struct bkey_s_c k)
 472 {
 473         return __bch2_mark_bkey_replicas(c, k, false);
 474 }
 475
 476 int bch2_replicas_gc_end(struct bch_fs *c, int ret)
 477 {
 478         unsigned i;
 479
 480         lockdep_assert_held(&c->replicas_gc_lock);
 481
 482         mutex_lock(&c->sb_lock);
 483         percpu_down_write(&c->mark_lock);
 484
 485         /*
 486          * this is kind of crappy; the replicas gc mechanism needs to be ripped
 487          * out
 488          */
 489
 490         for (i = 0; i < c->replicas.nr; i++) {
 491                 struct bch_replicas_entry *e =
 492                         cpu_replicas_entry(&c->replicas, i);
 493                 struct bch_replicas_cpu n;
 494
 495                 if (!__replicas_has_entry(&c->replicas_gc, e) &&
 496                     bch2_fs_usage_read_one(c, &c->usage_base->replicas[i])) {
 497                         n = cpu_replicas_add_entry(&c->replicas_gc, e);
 498                         if (!n.entries) {
 499                                 ret = -ENOSPC;
 500                                 goto err;
 501                         }
 502
 503                         swap(n, c->replicas_gc);
 504                         kfree(n.entries);
 505                 }
 506         }
 507
 508         if (bch2_cpu_replicas_to_sb_replicas(c, &c->replicas_gc)) {
 509                 ret = -ENOSPC;
 510                 goto err;
 511         }
 512
 513         ret = replicas_table_update(c, &c->replicas_gc);
 514 err:
 515         kfree(c->replicas_gc.entries);
 516         c->replicas_gc.entries = NULL;
 517
 518         percpu_up_write(&c->mark_lock);
 519
 520         if (!ret)
 521                 bch2_write_super(c);
 522
 523         mutex_unlock(&c->sb_lock);
 524
 525         return ret;
 526 }
 527
 528 int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask)
 529 {
 530         struct bch_replicas_entry *e;
 531         unsigned i = 0;
 532
 533         lockdep_assert_held(&c->replicas_gc_lock);
 534
 535         mutex_lock(&c->sb_lock);
 536         BUG_ON(c->replicas_gc.entries);
 537
 538         c->replicas_gc.nr               = 0;
 539         c->replicas_gc.entry_size       = 0;
 540
 541         for_each_cpu_replicas_entry(&c->replicas, e)
 542                 if (!((1 << e->data_type) & typemask)) {
 543                         c->replicas_gc.nr++;
 544                         c->replicas_gc.entry_size =
 545                                 max_t(unsigned, c->replicas_gc.entry_size,
 546                                       replicas_entry_bytes(e));
 547                 }
 548
 549         c->replicas_gc.entries = kcalloc(c->replicas_gc.nr,
 550                                          c->replicas_gc.entry_size,
 551                                          GFP_NOIO);
 552         if (!c->replicas_gc.entries) {
 553                 mutex_unlock(&c->sb_lock);
 554                 bch_err(c, "error allocating c->replicas_gc");
 555                 return -ENOMEM;
 556         }
 557
 558         for_each_cpu_replicas_entry(&c->replicas, e)
 559                 if (!((1 << e->data_type) & typemask))
 560                         memcpy(cpu_replicas_entry(&c->replicas_gc, i++),
 561                                e, c->replicas_gc.entry_size);
 562
 563         bch2_cpu_replicas_sort(&c->replicas_gc);
 564         mutex_unlock(&c->sb_lock);
 565
 566         return 0;
 567 }
 568
 569 int bch2_replicas_gc2(struct bch_fs *c)
 570 {
 571         struct bch_replicas_cpu new = { 0 };
 572         unsigned i, nr;
 573         int ret = 0;
 574
 575         bch2_journal_meta(&c->journal);
 576 retry:
 577         nr              = READ_ONCE(c->replicas.nr);
 578         new.entry_size  = READ_ONCE(c->replicas.entry_size);
 579         new.entries     = kcalloc(nr, new.entry_size, GFP_KERNEL);
 580         if (!new.entries) {
 581                 bch_err(c, "error allocating c->replicas_gc");
 582                 return -ENOMEM;
 583         }
 584
 585         mutex_lock(&c->sb_lock);
 586         percpu_down_write(&c->mark_lock);
 587
 588         if (nr                  != c->replicas.nr ||
 589             new.entry_size      != c->replicas.entry_size) {
 590                 percpu_up_write(&c->mark_lock);
 591                 mutex_unlock(&c->sb_lock);
 592                 kfree(new.entries);
 593                 goto retry;
 594         }
 595
 596         for (i = 0; i < c->replicas.nr; i++) {
 597                 struct bch_replicas_entry *e =
 598                         cpu_replicas_entry(&c->replicas, i);
 599
 600                 if (e->data_type == BCH_DATA_journal ||
 601                     c->usage_base->replicas[i] ||
 602                     percpu_u64_get(&c->usage[0]->replicas[i]) ||
 603                     percpu_u64_get(&c->usage[1]->replicas[i]) ||
 604                     percpu_u64_get(&c->usage[2]->replicas[i]) ||
 605                     percpu_u64_get(&c->usage[3]->replicas[i]))
 606                         memcpy(cpu_replicas_entry(&new, new.nr++),
 607                                e, new.entry_size);
 608         }
 609
 610         bch2_cpu_replicas_sort(&new);
 611
 612         if (bch2_cpu_replicas_to_sb_replicas(c, &new)) {
 613                 ret = -ENOSPC;
 614                 goto err;
 615         }
 616
 617         ret = replicas_table_update(c, &new);
 618 err:
 619         kfree(new.entries);
 620
 621         percpu_up_write(&c->mark_lock);
 622
 623         if (!ret)
 624                 bch2_write_super(c);
 625
 626         mutex_unlock(&c->sb_lock);
 627
 628         return ret;
 629 }
 630
 631 int bch2_replicas_set_usage(struct bch_fs *c,
 632                             struct bch_replicas_entry *r,
 633                             u64 sectors)
 634 {
 635         int ret, idx = bch2_replicas_entry_idx(c, r);
 636
 637         if (idx < 0) {
 638                 struct bch_replicas_cpu n;
 639
 640                 n = cpu_replicas_add_entry(&c->replicas, r);
 641                 if (!n.entries)
 642                         return -ENOMEM;
 643
 644                 ret = replicas_table_update(c, &n);
 645                 if (ret)
 646                         return ret;
 647
 648                 kfree(n.entries);
 649
 650                 idx = bch2_replicas_entry_idx(c, r);
 651                 BUG_ON(ret < 0);
 652         }
 653
 654         c->usage_base->replicas[idx] = sectors;
 655
 656         return 0;
 657 }
 658
 659 /* Replicas tracking - superblock: */
 660
 661 static int
 662 __bch2_sb_replicas_to_cpu_replicas(struct bch_sb_field_replicas *sb_r,
 663                                    struct bch_replicas_cpu *cpu_r)
 664 {
 665         struct bch_replicas_entry *e, *dst;
 666         unsigned nr = 0, entry_size = 0, idx = 0;
 667
 668         for_each_replicas_entry(sb_r, e) {
 669                 entry_size = max_t(unsigned, entry_size,
 670                                    replicas_entry_bytes(e));
 671                 nr++;
 672         }
 673
 674         cpu_r->entries = kcalloc(nr, entry_size, GFP_NOIO);
 675         if (!cpu_r->entries)
 676                 return -ENOMEM;
 677
 678         cpu_r->nr               = nr;
 679         cpu_r->entry_size       = entry_size;
 680
 681         for_each_replicas_entry(sb_r, e) {
 682                 dst = cpu_replicas_entry(cpu_r, idx++);
 683                 memcpy(dst, e, replicas_entry_bytes(e));
 684                 replicas_entry_sort(dst);
 685         }
 686
 687         return 0;
 688 }
 689
 690 static int
 691 __bch2_sb_replicas_v0_to_cpu_replicas(struct bch_sb_field_replicas_v0 *sb_r,
 692                                       struct bch_replicas_cpu *cpu_r)
 693 {
 694         struct bch_replicas_entry_v0 *e;
 695         unsigned nr = 0, entry_size = 0, idx = 0;
 696
 697         for_each_replicas_entry(sb_r, e) {
 698                 entry_size = max_t(unsigned, entry_size,
 699                                    replicas_entry_bytes(e));
 700                 nr++;
 701         }
 702
 703         entry_size += sizeof(struct bch_replicas_entry) -
 704                 sizeof(struct bch_replicas_entry_v0);
 705
 706         cpu_r->entries = kcalloc(nr, entry_size, GFP_NOIO);
 707         if (!cpu_r->entries)
 708                 return -ENOMEM;
 709
 710         cpu_r->nr               = nr;
 711         cpu_r->entry_size       = entry_size;
 712
 713         for_each_replicas_entry(sb_r, e) {
 714                 struct bch_replicas_entry *dst =
 715                         cpu_replicas_entry(cpu_r, idx++);
 716
 717                 dst->data_type  = e->data_type;
 718                 dst->nr_devs    = e->nr_devs;
 719                 dst->nr_required = 1;
 720                 memcpy(dst->devs, e->devs, e->nr_devs);
 721                 replicas_entry_sort(dst);
 722         }
 723
 724         return 0;
 725 }
 726
 727 int bch2_sb_replicas_to_cpu_replicas(struct bch_fs *c)
 728 {
 729         struct bch_sb_field_replicas *sb_v1;
 730         struct bch_sb_field_replicas_v0 *sb_v0;
 731         struct bch_replicas_cpu new_r = { 0, 0, NULL };
 732         int ret = 0;
 733
 734         if ((sb_v1 = bch2_sb_get_replicas(c->disk_sb.sb)))
 735                 ret = __bch2_sb_replicas_to_cpu_replicas(sb_v1, &new_r);
 736         else if ((sb_v0 = bch2_sb_get_replicas_v0(c->disk_sb.sb)))
 737                 ret = __bch2_sb_replicas_v0_to_cpu_replicas(sb_v0, &new_r);
 738
 739         if (ret)
 740                 return -ENOMEM;
 741
 742         bch2_cpu_replicas_sort(&new_r);
 743
 744         percpu_down_write(&c->mark_lock);
 745
 746         ret = replicas_table_update(c, &new_r);
 747         percpu_up_write(&c->mark_lock);
 748
 749         kfree(new_r.entries);
 750
 751         return 0;
 752 }
 753
 754 static int bch2_cpu_replicas_to_sb_replicas_v0(struct bch_fs *c,
 755                                                struct bch_replicas_cpu *r)
 756 {
 757         struct bch_sb_field_replicas_v0 *sb_r;
 758         struct bch_replicas_entry_v0 *dst;
 759         struct bch_replicas_entry *src;
 760         size_t bytes;
 761
 762         bytes = sizeof(struct bch_sb_field_replicas);
 763
 764         for_each_cpu_replicas_entry(r, src)
 765                 bytes += replicas_entry_bytes(src) - 1;
 766
 767         sb_r = bch2_sb_resize_replicas_v0(&c->disk_sb,
 768                         DIV_ROUND_UP(bytes, sizeof(u64)));
 769         if (!sb_r)
 770                 return -ENOSPC;
 771
 772         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas);
 773         sb_r = bch2_sb_get_replicas_v0(c->disk_sb.sb);
 774
 775         memset(&sb_r->entries, 0,
 776                vstruct_end(&sb_r->field) -
 777                (void *) &sb_r->entries);
 778
 779         dst = sb_r->entries;
 780         for_each_cpu_replicas_entry(r, src) {
 781                 dst->data_type  = src->data_type;
 782                 dst->nr_devs    = src->nr_devs;
 783                 memcpy(dst->devs, src->devs, src->nr_devs);
 784
 785                 dst = replicas_entry_next(dst);
 786
 787                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 788         }
 789
 790         return 0;
 791 }
 792
 793 static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *c,
 794                                             struct bch_replicas_cpu *r)
 795 {
 796         struct bch_sb_field_replicas *sb_r;
 797         struct bch_replicas_entry *dst, *src;
 798         bool need_v1 = false;
 799         size_t bytes;
 800
 801         bytes = sizeof(struct bch_sb_field_replicas);
 802
 803         for_each_cpu_replicas_entry(r, src) {
 804                 bytes += replicas_entry_bytes(src);
 805                 if (src->nr_required != 1)
 806                         need_v1 = true;
 807         }
 808
 809         if (!need_v1)
 810                 return bch2_cpu_replicas_to_sb_replicas_v0(c, r);
 811
 812         sb_r = bch2_sb_resize_replicas(&c->disk_sb,
 813                         DIV_ROUND_UP(bytes, sizeof(u64)));
 814         if (!sb_r)
 815                 return -ENOSPC;
 816
 817         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas_v0);
 818         sb_r = bch2_sb_get_replicas(c->disk_sb.sb);
 819
 820         memset(&sb_r->entries, 0,
 821                vstruct_end(&sb_r->field) -
 822                (void *) &sb_r->entries);
 823
 824         dst = sb_r->entries;
 825         for_each_cpu_replicas_entry(r, src) {
 826                 memcpy(dst, src, replicas_entry_bytes(src));
 827
 828                 dst = replicas_entry_next(dst);
 829
 830                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 831         }
 832
 833         return 0;
 834 }
 835
 836 static const char *check_dup_replicas_entries(struct bch_replicas_cpu *cpu_r)
 837 {
 838         unsigned i;
 839
 840         sort_cmp_size(cpu_r->entries,
 841                       cpu_r->nr,
 842                       cpu_r->entry_size,
 843                       memcmp, NULL);
 844
 845         for (i = 0; i + 1 < cpu_r->nr; i++) {
 846                 struct bch_replicas_entry *l =
 847                         cpu_replicas_entry(cpu_r, i);
 848                 struct bch_replicas_entry *r =
 849                         cpu_replicas_entry(cpu_r, i + 1);
 850
 851                 BUG_ON(memcmp(l, r, cpu_r->entry_size) > 0);
 852
 853                 if (!memcmp(l, r, cpu_r->entry_size))
 854                         return "duplicate replicas entry";
 855         }
 856
 857         return NULL;
 858 }
 859
 860 static const char *bch2_sb_validate_replicas(struct bch_sb *sb, struct bch_sb_field *f)
 861 {
 862         struct bch_sb_field_replicas *sb_r = field_to_type(f, replicas);
 863         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 864         struct bch_replicas_cpu cpu_r = { .entries = NULL };
 865         struct bch_replicas_entry *e;
 866         const char *err;
 867         unsigned i;
 868
 869         for_each_replicas_entry(sb_r, e) {
 870                 err = "invalid replicas entry: invalid data type";
 871                 if (e->data_type >= BCH_DATA_NR)
 872                         goto err;
 873
 874                 err = "invalid replicas entry: no devices";
 875                 if (!e->nr_devs)
 876                         goto err;
 877
 878                 err = "invalid replicas entry: bad nr_required";
 879                 if (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 }