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 void bch2_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         bch2_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         bch2_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_KERNEL);
 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         bch2_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_online *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         unsigned scratch_bytes = sizeof(struct bch_fs_usage_online) +
 280                 sizeof(u64) * new_r->nr;
 281         int ret = 0;
 282
 283         memset(new_usage, 0, sizeof(new_usage));
 284
 285         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 286                 if (!(new_usage[i] = __alloc_percpu_gfp(bytes,
 287                                         sizeof(u64), GFP_KERNEL)))
 288                         goto err;
 289
 290         if (!(new_base = kzalloc(bytes, GFP_KERNEL)) ||
 291             !(new_scratch  = kmalloc(scratch_bytes, GFP_KERNEL)) ||
 292             (c->usage_gc &&
 293              !(new_gc = __alloc_percpu_gfp(bytes, sizeof(u64), GFP_KERNEL))))
 294                 goto err;
 295
 296         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 297                 if (c->usage[i])
 298                         __replicas_table_update_pcpu(new_usage[i], new_r,
 299                                                      c->usage[i], &c->replicas);
 300         if (c->usage_base)
 301                 __replicas_table_update(new_base,               new_r,
 302                                         c->usage_base,          &c->replicas);
 303         if (c->usage_gc)
 304                 __replicas_table_update_pcpu(new_gc,            new_r,
 305                                              c->usage_gc,       &c->replicas);
 306
 307         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 308                 swap(c->usage[i],       new_usage[i]);
 309         swap(c->usage_base,     new_base);
 310         swap(c->usage_scratch,  new_scratch);
 311         swap(c->usage_gc,       new_gc);
 312         swap(c->replicas,       *new_r);
 313 out:
 314         free_percpu(new_gc);
 315         kfree(new_scratch);
 316         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 317                 free_percpu(new_usage[i]);
 318         kfree(new_base);
 319         return ret;
 320 err:
 321         bch_err(c, "error updating replicas table: memory allocation failure");
 322         ret = -ENOMEM;
 323         goto out;
 324 }
 325
 326 static unsigned reserve_journal_replicas(struct bch_fs *c,
 327                                      struct bch_replicas_cpu *r)
 328 {
 329         struct bch_replicas_entry *e;
 330         unsigned journal_res_u64s = 0;
 331
 332         /* nr_inodes: */
 333         journal_res_u64s +=
 334                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 335
 336         /* key_version: */
 337         journal_res_u64s +=
 338                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 339
 340         /* persistent_reserved: */
 341         journal_res_u64s +=
 342                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64)) *
 343                 BCH_REPLICAS_MAX;
 344
 345         for_each_cpu_replicas_entry(r, e)
 346                 journal_res_u64s +=
 347                         DIV_ROUND_UP(sizeof(struct jset_entry_data_usage) +
 348                                      e->nr_devs, sizeof(u64));
 349         return journal_res_u64s;
 350 }
 351
 352 noinline
 353 static int bch2_mark_replicas_slowpath(struct bch_fs *c,
 354                                 struct bch_replicas_entry *new_entry)
 355 {
 356         struct bch_replicas_cpu new_r, new_gc;
 357         int ret = 0;
 358
 359         verify_replicas_entry(new_entry);
 360
 361         memset(&new_r, 0, sizeof(new_r));
 362         memset(&new_gc, 0, sizeof(new_gc));
 363
 364         mutex_lock(&c->sb_lock);
 365
 366         if (c->replicas_gc.entries &&
 367             !__replicas_has_entry(&c->replicas_gc, new_entry)) {
 368                 new_gc = cpu_replicas_add_entry(&c->replicas_gc, new_entry);
 369                 if (!new_gc.entries)
 370                         goto err;
 371         }
 372
 373         if (!__replicas_has_entry(&c->replicas, new_entry)) {
 374                 new_r = cpu_replicas_add_entry(&c->replicas, new_entry);
 375                 if (!new_r.entries)
 376                         goto err;
 377
 378                 ret = bch2_cpu_replicas_to_sb_replicas(c, &new_r);
 379                 if (ret)
 380                         goto err;
 381
 382                 bch2_journal_entry_res_resize(&c->journal,
 383                                 &c->replicas_journal_res,
 384                                 reserve_journal_replicas(c, &new_r));
 385         }
 386
 387         if (!new_r.entries &&
 388             !new_gc.entries)
 389                 goto out;
 390
 391         /* allocations done, now commit: */
 392
 393         if (new_r.entries)
 394                 bch2_write_super(c);
 395
 396         /* don't update in memory replicas until changes are persistent */
 397         percpu_down_write(&c->mark_lock);
 398         if (new_r.entries)
 399                 ret = replicas_table_update(c, &new_r);
 400         if (new_gc.entries)
 401                 swap(new_gc, c->replicas_gc);
 402         percpu_up_write(&c->mark_lock);
 403 out:
 404         mutex_unlock(&c->sb_lock);
 405
 406         kfree(new_r.entries);
 407         kfree(new_gc.entries);
 408
 409         return ret;
 410 err:
 411         bch_err(c, "error adding replicas entry: memory allocation failure");
 412         ret = -ENOMEM;
 413         goto out;
 414 }
 415
 416 static int __bch2_mark_replicas(struct bch_fs *c,
 417                                 struct bch_replicas_entry *r,
 418                                 bool check)
 419 {
 420         return likely(bch2_replicas_marked(c, r))       ? 0
 421                 : check                                 ? -1
 422                 : bch2_mark_replicas_slowpath(c, r);
 423 }
 424
 425 int bch2_mark_replicas(struct bch_fs *c, struct bch_replicas_entry *r)
 426 {
 427         return __bch2_mark_replicas(c, r, false);
 428 }
 429
 430 static int __bch2_mark_bkey_replicas(struct bch_fs *c, struct bkey_s_c k,
 431                                      bool check)
 432 {
 433         struct bch_replicas_padded search;
 434         struct bch_devs_list cached = bch2_bkey_cached_devs(k);
 435         unsigned i;
 436         int ret;
 437
 438         memset(&search, 0, sizeof(search));
 439
 440         for (i = 0; i < cached.nr; i++) {
 441                 bch2_replicas_entry_cached(&search.e, cached.devs[i]);
 442
 443                 ret = __bch2_mark_replicas(c, &search.e, check);
 444                 if (ret)
 445                         return ret;
 446         }
 447
 448         bch2_bkey_to_replicas(&search.e, k);
 449
 450         ret = __bch2_mark_replicas(c, &search.e, check);
 451         if (ret)
 452                 return ret;
 453
 454         if (search.e.data_type == BCH_DATA_parity) {
 455                 search.e.data_type = BCH_DATA_cached;
 456                 ret = __bch2_mark_replicas(c, &search.e, check);
 457                 if (ret)
 458                         return ret;
 459
 460                 search.e.data_type = BCH_DATA_user;
 461                 ret = __bch2_mark_replicas(c, &search.e, check);
 462                 if (ret)
 463                         return ret;
 464         }
 465
 466         return 0;
 467 }
 468
 469 /* replicas delta list: */
 470
 471 bool bch2_replicas_delta_list_marked(struct bch_fs *c,
 472                                      struct replicas_delta_list *r)
 473 {
 474         struct replicas_delta *d = r->d;
 475         struct replicas_delta *top = (void *) r->d + r->used;
 476
 477         percpu_rwsem_assert_held(&c->mark_lock);
 478
 479         for (d = r->d; d != top; d = replicas_delta_next(d))
 480                 if (bch2_replicas_entry_idx(c, &d->r) < 0)
 481                         return false;
 482         return true;
 483 }
 484
 485 int bch2_replicas_delta_list_mark(struct bch_fs *c,
 486                                   struct replicas_delta_list *r)
 487 {
 488         struct replicas_delta *d = r->d;
 489         struct replicas_delta *top = (void *) r->d + r->used;
 490         int ret = 0;
 491
 492         for (d = r->d; !ret && d != top; d = replicas_delta_next(d))
 493                 ret = bch2_mark_replicas(c, &d->r);
 494         return ret;
 495 }
 496
 497 /* bkey replicas: */
 498
 499 bool bch2_bkey_replicas_marked(struct bch_fs *c,
 500                                struct bkey_s_c k)
 501 {
 502         return __bch2_mark_bkey_replicas(c, k, true) == 0;
 503 }
 504
 505 int bch2_mark_bkey_replicas(struct bch_fs *c, struct bkey_s_c k)
 506 {
 507         return __bch2_mark_bkey_replicas(c, k, false);
 508 }
 509
 510 /*
 511  * Old replicas_gc mechanism: only used for journal replicas entries now, should
 512  * die at some point:
 513  */
 514
 515 int bch2_replicas_gc_end(struct bch_fs *c, int ret)
 516 {
 517         unsigned i;
 518
 519         lockdep_assert_held(&c->replicas_gc_lock);
 520
 521         mutex_lock(&c->sb_lock);
 522         percpu_down_write(&c->mark_lock);
 523
 524         /*
 525          * this is kind of crappy; the replicas gc mechanism needs to be ripped
 526          * out
 527          */
 528
 529         for (i = 0; i < c->replicas.nr; i++) {
 530                 struct bch_replicas_entry *e =
 531                         cpu_replicas_entry(&c->replicas, i);
 532                 struct bch_replicas_cpu n;
 533
 534                 if (!__replicas_has_entry(&c->replicas_gc, e) &&
 535                     bch2_fs_usage_read_one(c, &c->usage_base->replicas[i])) {
 536                         n = cpu_replicas_add_entry(&c->replicas_gc, e);
 537                         if (!n.entries) {
 538                                 ret = -ENOSPC;
 539                                 goto err;
 540                         }
 541
 542                         swap(n, c->replicas_gc);
 543                         kfree(n.entries);
 544                 }
 545         }
 546
 547         if (bch2_cpu_replicas_to_sb_replicas(c, &c->replicas_gc)) {
 548                 ret = -ENOSPC;
 549                 goto err;
 550         }
 551
 552         ret = replicas_table_update(c, &c->replicas_gc);
 553 err:
 554         kfree(c->replicas_gc.entries);
 555         c->replicas_gc.entries = NULL;
 556
 557         percpu_up_write(&c->mark_lock);
 558
 559         if (!ret)
 560                 bch2_write_super(c);
 561
 562         mutex_unlock(&c->sb_lock);
 563
 564         return ret;
 565 }
 566
 567 int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask)
 568 {
 569         struct bch_replicas_entry *e;
 570         unsigned i = 0;
 571
 572         lockdep_assert_held(&c->replicas_gc_lock);
 573
 574         mutex_lock(&c->sb_lock);
 575         BUG_ON(c->replicas_gc.entries);
 576
 577         c->replicas_gc.nr               = 0;
 578         c->replicas_gc.entry_size       = 0;
 579
 580         for_each_cpu_replicas_entry(&c->replicas, e)
 581                 if (!((1 << e->data_type) & typemask)) {
 582                         c->replicas_gc.nr++;
 583                         c->replicas_gc.entry_size =
 584                                 max_t(unsigned, c->replicas_gc.entry_size,
 585                                       replicas_entry_bytes(e));
 586                 }
 587
 588         c->replicas_gc.entries = kcalloc(c->replicas_gc.nr,
 589                                          c->replicas_gc.entry_size,
 590                                          GFP_KERNEL);
 591         if (!c->replicas_gc.entries) {
 592                 mutex_unlock(&c->sb_lock);
 593                 bch_err(c, "error allocating c->replicas_gc");
 594                 return -ENOMEM;
 595         }
 596
 597         for_each_cpu_replicas_entry(&c->replicas, e)
 598                 if (!((1 << e->data_type) & typemask))
 599                         memcpy(cpu_replicas_entry(&c->replicas_gc, i++),
 600                                e, c->replicas_gc.entry_size);
 601
 602         bch2_cpu_replicas_sort(&c->replicas_gc);
 603         mutex_unlock(&c->sb_lock);
 604
 605         return 0;
 606 }
 607
 608 /* New much simpler mechanism for clearing out unneeded replicas entries: */
 609
 610 int bch2_replicas_gc2(struct bch_fs *c)
 611 {
 612         struct bch_replicas_cpu new = { 0 };
 613         unsigned i, nr;
 614         int ret = 0;
 615
 616         bch2_journal_meta(&c->journal);
 617 retry:
 618         nr              = READ_ONCE(c->replicas.nr);
 619         new.entry_size  = READ_ONCE(c->replicas.entry_size);
 620         new.entries     = kcalloc(nr, new.entry_size, GFP_KERNEL);
 621         if (!new.entries) {
 622                 bch_err(c, "error allocating c->replicas_gc");
 623                 return -ENOMEM;
 624         }
 625
 626         mutex_lock(&c->sb_lock);
 627         percpu_down_write(&c->mark_lock);
 628
 629         if (nr                  != c->replicas.nr ||
 630             new.entry_size      != c->replicas.entry_size) {
 631                 percpu_up_write(&c->mark_lock);
 632                 mutex_unlock(&c->sb_lock);
 633                 kfree(new.entries);
 634                 goto retry;
 635         }
 636
 637         for (i = 0; i < c->replicas.nr; i++) {
 638                 struct bch_replicas_entry *e =
 639                         cpu_replicas_entry(&c->replicas, i);
 640
 641                 if (e->data_type == BCH_DATA_journal ||
 642                     c->usage_base->replicas[i] ||
 643                     percpu_u64_get(&c->usage[0]->replicas[i]) ||
 644                     percpu_u64_get(&c->usage[1]->replicas[i]) ||
 645                     percpu_u64_get(&c->usage[2]->replicas[i]) ||
 646                     percpu_u64_get(&c->usage[3]->replicas[i]))
 647                         memcpy(cpu_replicas_entry(&new, new.nr++),
 648                                e, new.entry_size);
 649         }
 650
 651         bch2_cpu_replicas_sort(&new);
 652
 653         if (bch2_cpu_replicas_to_sb_replicas(c, &new)) {
 654                 ret = -ENOSPC;
 655                 goto err;
 656         }
 657
 658         ret = replicas_table_update(c, &new);
 659 err:
 660         kfree(new.entries);
 661
 662         percpu_up_write(&c->mark_lock);
 663
 664         if (!ret)
 665                 bch2_write_super(c);
 666
 667         mutex_unlock(&c->sb_lock);
 668
 669         return ret;
 670 }
 671
 672 int bch2_replicas_set_usage(struct bch_fs *c,
 673                             struct bch_replicas_entry *r,
 674                             u64 sectors)
 675 {
 676         int ret, idx = bch2_replicas_entry_idx(c, r);
 677
 678         if (idx < 0) {
 679                 struct bch_replicas_cpu n;
 680
 681                 n = cpu_replicas_add_entry(&c->replicas, r);
 682                 if (!n.entries)
 683                         return -ENOMEM;
 684
 685                 ret = replicas_table_update(c, &n);
 686                 if (ret)
 687                         return ret;
 688
 689                 kfree(n.entries);
 690
 691                 idx = bch2_replicas_entry_idx(c, r);
 692                 BUG_ON(ret < 0);
 693         }
 694
 695         c->usage_base->replicas[idx] = sectors;
 696
 697         return 0;
 698 }
 699
 700 /* Replicas tracking - superblock: */
 701
 702 static int
 703 __bch2_sb_replicas_to_cpu_replicas(struct bch_sb_field_replicas *sb_r,
 704                                    struct bch_replicas_cpu *cpu_r)
 705 {
 706         struct bch_replicas_entry *e, *dst;
 707         unsigned nr = 0, entry_size = 0, idx = 0;
 708
 709         for_each_replicas_entry(sb_r, e) {
 710                 entry_size = max_t(unsigned, entry_size,
 711                                    replicas_entry_bytes(e));
 712                 nr++;
 713         }
 714
 715         cpu_r->entries = kcalloc(nr, entry_size, GFP_KERNEL);
 716         if (!cpu_r->entries)
 717                 return -ENOMEM;
 718
 719         cpu_r->nr               = nr;
 720         cpu_r->entry_size       = entry_size;
 721
 722         for_each_replicas_entry(sb_r, e) {
 723                 dst = cpu_replicas_entry(cpu_r, idx++);
 724                 memcpy(dst, e, replicas_entry_bytes(e));
 725                 bch2_replicas_entry_sort(dst);
 726         }
 727
 728         return 0;
 729 }
 730
 731 static int
 732 __bch2_sb_replicas_v0_to_cpu_replicas(struct bch_sb_field_replicas_v0 *sb_r,
 733                                       struct bch_replicas_cpu *cpu_r)
 734 {
 735         struct bch_replicas_entry_v0 *e;
 736         unsigned nr = 0, entry_size = 0, idx = 0;
 737
 738         for_each_replicas_entry(sb_r, e) {
 739                 entry_size = max_t(unsigned, entry_size,
 740                                    replicas_entry_bytes(e));
 741                 nr++;
 742         }
 743
 744         entry_size += sizeof(struct bch_replicas_entry) -
 745                 sizeof(struct bch_replicas_entry_v0);
 746
 747         cpu_r->entries = kcalloc(nr, entry_size, GFP_KERNEL);
 748         if (!cpu_r->entries)
 749                 return -ENOMEM;
 750
 751         cpu_r->nr               = nr;
 752         cpu_r->entry_size       = entry_size;
 753
 754         for_each_replicas_entry(sb_r, e) {
 755                 struct bch_replicas_entry *dst =
 756                         cpu_replicas_entry(cpu_r, idx++);
 757
 758                 dst->data_type  = e->data_type;
 759                 dst->nr_devs    = e->nr_devs;
 760                 dst->nr_required = 1;
 761                 memcpy(dst->devs, e->devs, e->nr_devs);
 762                 bch2_replicas_entry_sort(dst);
 763         }
 764
 765         return 0;
 766 }
 767
 768 int bch2_sb_replicas_to_cpu_replicas(struct bch_fs *c)
 769 {
 770         struct bch_sb_field_replicas *sb_v1;
 771         struct bch_sb_field_replicas_v0 *sb_v0;
 772         struct bch_replicas_cpu new_r = { 0, 0, NULL };
 773         int ret = 0;
 774
 775         if ((sb_v1 = bch2_sb_get_replicas(c->disk_sb.sb)))
 776                 ret = __bch2_sb_replicas_to_cpu_replicas(sb_v1, &new_r);
 777         else if ((sb_v0 = bch2_sb_get_replicas_v0(c->disk_sb.sb)))
 778                 ret = __bch2_sb_replicas_v0_to_cpu_replicas(sb_v0, &new_r);
 779
 780         if (ret)
 781                 return -ENOMEM;
 782
 783         bch2_cpu_replicas_sort(&new_r);
 784
 785         percpu_down_write(&c->mark_lock);
 786
 787         ret = replicas_table_update(c, &new_r);
 788         percpu_up_write(&c->mark_lock);
 789
 790         kfree(new_r.entries);
 791
 792         return 0;
 793 }
 794
 795 static int bch2_cpu_replicas_to_sb_replicas_v0(struct bch_fs *c,
 796                                                struct bch_replicas_cpu *r)
 797 {
 798         struct bch_sb_field_replicas_v0 *sb_r;
 799         struct bch_replicas_entry_v0 *dst;
 800         struct bch_replicas_entry *src;
 801         size_t bytes;
 802
 803         bytes = sizeof(struct bch_sb_field_replicas);
 804
 805         for_each_cpu_replicas_entry(r, src)
 806                 bytes += replicas_entry_bytes(src) - 1;
 807
 808         sb_r = bch2_sb_resize_replicas_v0(&c->disk_sb,
 809                         DIV_ROUND_UP(bytes, sizeof(u64)));
 810         if (!sb_r)
 811                 return -ENOSPC;
 812
 813         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas);
 814         sb_r = bch2_sb_get_replicas_v0(c->disk_sb.sb);
 815
 816         memset(&sb_r->entries, 0,
 817                vstruct_end(&sb_r->field) -
 818                (void *) &sb_r->entries);
 819
 820         dst = sb_r->entries;
 821         for_each_cpu_replicas_entry(r, src) {
 822                 dst->data_type  = src->data_type;
 823                 dst->nr_devs    = src->nr_devs;
 824                 memcpy(dst->devs, src->devs, src->nr_devs);
 825
 826                 dst = replicas_entry_next(dst);
 827
 828                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 829         }
 830
 831         return 0;
 832 }
 833
 834 static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *c,
 835                                             struct bch_replicas_cpu *r)
 836 {
 837         struct bch_sb_field_replicas *sb_r;
 838         struct bch_replicas_entry *dst, *src;
 839         bool need_v1 = false;
 840         size_t bytes;
 841
 842         bytes = sizeof(struct bch_sb_field_replicas);
 843
 844         for_each_cpu_replicas_entry(r, src) {
 845                 bytes += replicas_entry_bytes(src);
 846                 if (src->nr_required != 1)
 847                         need_v1 = true;
 848         }
 849
 850         if (!need_v1)
 851                 return bch2_cpu_replicas_to_sb_replicas_v0(c, r);
 852
 853         sb_r = bch2_sb_resize_replicas(&c->disk_sb,
 854                         DIV_ROUND_UP(bytes, sizeof(u64)));
 855         if (!sb_r)
 856                 return -ENOSPC;
 857
 858         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas_v0);
 859         sb_r = bch2_sb_get_replicas(c->disk_sb.sb);
 860
 861         memset(&sb_r->entries, 0,
 862                vstruct_end(&sb_r->field) -
 863                (void *) &sb_r->entries);
 864
 865         dst = sb_r->entries;
 866         for_each_cpu_replicas_entry(r, src) {
 867                 memcpy(dst, src, replicas_entry_bytes(src));
 868
 869                 dst = replicas_entry_next(dst);
 870
 871                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 872         }
 873
 874         return 0;
 875 }
 876
 877 static const char *check_dup_replicas_entries(struct bch_replicas_cpu *cpu_r)
 878 {
 879         unsigned i;
 880
 881         sort_cmp_size(cpu_r->entries,
 882                       cpu_r->nr,
 883                       cpu_r->entry_size,
 884                       memcmp, NULL);
 885
 886         for (i = 0; i + 1 < cpu_r->nr; i++) {
 887                 struct bch_replicas_entry *l =
 888                         cpu_replicas_entry(cpu_r, i);
 889                 struct bch_replicas_entry *r =
 890                         cpu_replicas_entry(cpu_r, i + 1);
 891
 892                 BUG_ON(memcmp(l, r, cpu_r->entry_size) > 0);
 893
 894                 if (!memcmp(l, r, cpu_r->entry_size))
 895                         return "duplicate replicas entry";
 896         }
 897
 898         return NULL;
 899 }
 900
 901 static const char *bch2_sb_validate_replicas(struct bch_sb *sb, struct bch_sb_field *f)
 902 {
 903         struct bch_sb_field_replicas *sb_r = field_to_type(f, replicas);
 904         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 905         struct bch_replicas_cpu cpu_r = { .entries = NULL };
 906         struct bch_replicas_entry *e;
 907         const char *err;
 908         unsigned i;
 909
 910         for_each_replicas_entry(sb_r, e) {
 911                 err = "invalid replicas entry: invalid data type";
 912                 if (e->data_type >= BCH_DATA_NR)
 913                         goto err;
 914
 915                 err = "invalid replicas entry: no devices";
 916                 if (!e->nr_devs)
 917                         goto err;
 918
 919                 err = "invalid replicas entry: bad nr_required";
 920                 if (e->nr_required > 1 &&
 921                     e->nr_required >= e->nr_devs)
 922                         goto err;
 923
 924                 err = "invalid replicas entry: invalid device";
 925                 for (i = 0; i < e->nr_devs; i++)
 926                         if (!bch2_dev_exists(sb, mi, e->devs[i]))
 927                                 goto err;
 928         }
 929
 930         err = "cannot allocate memory";
 931         if (__bch2_sb_replicas_to_cpu_replicas(sb_r, &cpu_r))
 932                 goto err;
 933
 934         err = check_dup_replicas_entries(&cpu_r);
 935 err:
 936         kfree(cpu_r.entries);
 937         return err;
 938 }
 939
 940 static void bch2_sb_replicas_to_text(struct printbuf *out,
 941                                      struct bch_sb *sb,
 942                                      struct bch_sb_field *f)
 943 {
 944         struct bch_sb_field_replicas *r = field_to_type(f, replicas);
 945         struct bch_replicas_entry *e;
 946         bool first = true;
 947
 948         for_each_replicas_entry(r, e) {
 949                 if (!first)
 950                         pr_buf(out, " ");
 951                 first = false;
 952
 953                 bch2_replicas_entry_to_text(out, e);
 954         }
 955 }
 956
 957 const struct bch_sb_field_ops bch_sb_field_ops_replicas = {
 958         .validate       = bch2_sb_validate_replicas,
 959         .to_text        = bch2_sb_replicas_to_text,
 960 };
 961
 962 static const char *bch2_sb_validate_replicas_v0(struct bch_sb *sb, struct bch_sb_field *f)
 963 {
 964         struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0);
 965         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 966         struct bch_replicas_cpu cpu_r = { .entries = NULL };
 967         struct bch_replicas_entry_v0 *e;
 968         const char *err;
 969         unsigned i;
 970
 971         for_each_replicas_entry_v0(sb_r, e) {
 972                 err = "invalid replicas entry: invalid data type";
 973                 if (e->data_type >= BCH_DATA_NR)
 974                         goto err;
 975
 976                 err = "invalid replicas entry: no devices";
 977                 if (!e->nr_devs)
 978                         goto err;
 979
 980                 err = "invalid replicas entry: invalid device";
 981                 for (i = 0; i < e->nr_devs; i++)
 982                         if (!bch2_dev_exists(sb, mi, e->devs[i]))
 983                                 goto err;
 984         }
 985
 986         err = "cannot allocate memory";
 987         if (__bch2_sb_replicas_v0_to_cpu_replicas(sb_r, &cpu_r))
 988                 goto err;
 989
 990         err = check_dup_replicas_entries(&cpu_r);
 991 err:
 992         kfree(cpu_r.entries);
 993         return err;
 994 }
 995
 996 const struct bch_sb_field_ops bch_sb_field_ops_replicas_v0 = {
 997         .validate       = bch2_sb_validate_replicas_v0,
 998 };
 999
1000 /* Query replicas: */
1001
1002 bool bch2_have_enough_devs(struct bch_fs *c, struct bch_devs_mask devs,
1003                            unsigned flags, bool print)
1004 {
1005         struct bch_replicas_entry *e;
1006         bool ret = true;
1007
1008         percpu_down_read(&c->mark_lock);
1009         for_each_cpu_replicas_entry(&c->replicas, e) {
1010                 unsigned i, nr_online = 0, nr_failed = 0, dflags = 0;
1011                 bool metadata = e->data_type < BCH_DATA_user;
1012
1013                 for (i = 0; i < e->nr_devs; i++) {
1014                         struct bch_dev *ca = bch_dev_bkey_exists(c, e->devs[i]);
1015
1016                         nr_online += test_bit(e->devs[i], devs.d);
1017                         nr_failed += ca->mi.state == BCH_MEMBER_STATE_failed;
1018                 }
1019
1020                 if (nr_failed == e->nr_devs)
1021                         continue;
1022
1023                 if (nr_online < e->nr_required)
1024                         dflags |= metadata
1025                                 ? BCH_FORCE_IF_METADATA_LOST
1026                                 : BCH_FORCE_IF_DATA_LOST;
1027
1028                 if (nr_online < e->nr_devs)
1029                         dflags |= metadata
1030                                 ? BCH_FORCE_IF_METADATA_DEGRADED
1031                                 : BCH_FORCE_IF_DATA_DEGRADED;
1032
1033                 if (dflags & ~flags) {
1034                         if (print) {
1035                                 char buf[100];
1036
1037                                 bch2_replicas_entry_to_text(&PBUF(buf), e);
1038                                 bch_err(c, "insufficient devices online (%u) for replicas entry %s",
1039                                         nr_online, buf);
1040                         }
1041                         ret = false;
1042                         break;
1043                 }
1044
1045         }
1046         percpu_up_read(&c->mark_lock);
1047
1048         return ret;
1049 }
1050
1051 unsigned bch2_dev_has_data(struct bch_fs *c, struct bch_dev *ca)
1052 {
1053         struct bch_replicas_entry *e;
1054         unsigned i, ret = 0;
1055
1056         percpu_down_read(&c->mark_lock);
1057
1058         for_each_cpu_replicas_entry(&c->replicas, e)
1059                 for (i = 0; i < e->nr_devs; i++)
1060                         if (e->devs[i] == ca->dev_idx)
1061                                 ret |= 1 << e->data_type;
1062
1063         percpu_up_read(&c->mark_lock);
1064
1065         return ret;
1066 }
1067
1068 void bch2_fs_replicas_exit(struct bch_fs *c)
1069 {
1070         unsigned i;
1071
1072         kfree(c->usage_scratch);
1073         for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1074                 free_percpu(c->usage[i]);
1075         kfree(c->usage_base);
1076         kfree(c->replicas.entries);
1077         kfree(c->replicas_gc.entries);
1078
1079         mempool_exit(&c->replicas_delta_pool);
1080 }
1081
1082 int bch2_fs_replicas_init(struct bch_fs *c)
1083 {
1084         bch2_journal_entry_res_resize(&c->journal,
1085                         &c->replicas_journal_res,
1086                         reserve_journal_replicas(c, &c->replicas));
1087
1088         return mempool_init_kmalloc_pool(&c->replicas_delta_pool, 1,
1089                                          REPLICAS_DELTA_LIST_MAX) ?:
1090                 replicas_table_update(c, &c->replicas);
1091 }