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_USER;
 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 static bool bch2_replicas_marked_locked(struct bch_fs *c,
 217                           struct bch_replicas_entry *search,
 218                           bool check_gc_replicas)
 219 {
 220         if (!search->nr_devs)
 221                 return true;
 222
 223         verify_replicas_entry(search);
 224
 225         return __replicas_has_entry(&c->replicas, search) &&
 226                 (!check_gc_replicas ||
 227                  likely((!c->replicas_gc.entries)) ||
 228                  __replicas_has_entry(&c->replicas_gc, search));
 229 }
 230
 231 bool bch2_replicas_marked(struct bch_fs *c,
 232                           struct bch_replicas_entry *search,
 233                           bool check_gc_replicas)
 234 {
 235         bool marked;
 236
 237         percpu_down_read(&c->mark_lock);
 238         marked = bch2_replicas_marked_locked(c, search, check_gc_replicas);
 239         percpu_up_read(&c->mark_lock);
 240
 241         return marked;
 242 }
 243
 244 static void __replicas_table_update(struct bch_fs_usage *dst,
 245                                     struct bch_replicas_cpu *dst_r,
 246                                     struct bch_fs_usage *src,
 247                                     struct bch_replicas_cpu *src_r)
 248 {
 249         int src_idx, dst_idx;
 250
 251         *dst = *src;
 252
 253         for (src_idx = 0; src_idx < src_r->nr; src_idx++) {
 254                 if (!src->replicas[src_idx])
 255                         continue;
 256
 257                 dst_idx = __replicas_entry_idx(dst_r,
 258                                 cpu_replicas_entry(src_r, src_idx));
 259                 BUG_ON(dst_idx < 0);
 260
 261                 dst->replicas[dst_idx] = src->replicas[src_idx];
 262         }
 263 }
 264
 265 static void __replicas_table_update_pcpu(struct bch_fs_usage __percpu *dst_p,
 266                                     struct bch_replicas_cpu *dst_r,
 267                                     struct bch_fs_usage __percpu *src_p,
 268                                     struct bch_replicas_cpu *src_r)
 269 {
 270         unsigned src_nr = sizeof(struct bch_fs_usage) / sizeof(u64) + src_r->nr;
 271         struct bch_fs_usage *dst, *src = (void *)
 272                 bch2_acc_percpu_u64s((void *) src_p, src_nr);
 273
 274         preempt_disable();
 275         dst = this_cpu_ptr(dst_p);
 276         preempt_enable();
 277
 278         __replicas_table_update(dst, dst_r, src, src_r);
 279 }
 280
 281 /*
 282  * Resize filesystem accounting:
 283  */
 284 static int replicas_table_update(struct bch_fs *c,
 285                                  struct bch_replicas_cpu *new_r)
 286 {
 287         struct bch_fs_usage __percpu *new_usage[2] = { NULL, NULL };
 288         struct bch_fs_usage *new_scratch = NULL;
 289         struct bch_fs_usage __percpu *new_gc = NULL;
 290         struct bch_fs_usage *new_base = NULL;
 291         unsigned bytes = sizeof(struct bch_fs_usage) +
 292                 sizeof(u64) * new_r->nr;
 293         int ret = -ENOMEM;
 294
 295         if (!(new_base = kzalloc(bytes, GFP_NOIO)) ||
 296             !(new_usage[0] = __alloc_percpu_gfp(bytes, sizeof(u64),
 297                                                 GFP_NOIO)) ||
 298             !(new_usage[1] = __alloc_percpu_gfp(bytes, sizeof(u64),
 299                                                 GFP_NOIO)) ||
 300             !(new_scratch  = kmalloc(bytes, GFP_NOIO)) ||
 301             (c->usage_gc &&
 302              !(new_gc = __alloc_percpu_gfp(bytes, sizeof(u64), GFP_NOIO)))) {
 303                 bch_err(c, "error updating replicas table: memory allocation failure");
 304                 goto err;
 305         }
 306
 307         if (c->usage_base)
 308                 __replicas_table_update(new_base,               new_r,
 309                                         c->usage_base,          &c->replicas);
 310         if (c->usage[0])
 311                 __replicas_table_update_pcpu(new_usage[0],      new_r,
 312                                              c->usage[0],       &c->replicas);
 313         if (c->usage[1])
 314                 __replicas_table_update_pcpu(new_usage[1],      new_r,
 315                                              c->usage[1],       &c->replicas);
 316         if (c->usage_gc)
 317                 __replicas_table_update_pcpu(new_gc,            new_r,
 318                                              c->usage_gc,       &c->replicas);
 319
 320         swap(c->usage_base,     new_base);
 321         swap(c->usage[0],       new_usage[0]);
 322         swap(c->usage[1],       new_usage[1]);
 323         swap(c->usage_scratch,  new_scratch);
 324         swap(c->usage_gc,       new_gc);
 325         swap(c->replicas,       *new_r);
 326         ret = 0;
 327 err:
 328         free_percpu(new_gc);
 329         kfree(new_scratch);
 330         free_percpu(new_usage[1]);
 331         free_percpu(new_usage[0]);
 332         kfree(new_base);
 333         return ret;
 334 }
 335
 336 static unsigned reserve_journal_replicas(struct bch_fs *c,
 337                                      struct bch_replicas_cpu *r)
 338 {
 339         struct bch_replicas_entry *e;
 340         unsigned journal_res_u64s = 0;
 341
 342         /* nr_inodes: */
 343         journal_res_u64s +=
 344                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 345
 346         /* key_version: */
 347         journal_res_u64s +=
 348                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 349
 350         /* persistent_reserved: */
 351         journal_res_u64s +=
 352                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64)) *
 353                 BCH_REPLICAS_MAX;
 354
 355         for_each_cpu_replicas_entry(r, e)
 356                 journal_res_u64s +=
 357                         DIV_ROUND_UP(sizeof(struct jset_entry_data_usage) +
 358                                      e->nr_devs, sizeof(u64));
 359         return journal_res_u64s;
 360 }
 361
 362 noinline
 363 static int bch2_mark_replicas_slowpath(struct bch_fs *c,
 364                                 struct bch_replicas_entry *new_entry)
 365 {
 366         struct bch_replicas_cpu new_r, new_gc;
 367         int ret = 0;
 368
 369         verify_replicas_entry(new_entry);
 370
 371         memset(&new_r, 0, sizeof(new_r));
 372         memset(&new_gc, 0, sizeof(new_gc));
 373
 374         mutex_lock(&c->sb_lock);
 375
 376         if (c->replicas_gc.entries &&
 377             !__replicas_has_entry(&c->replicas_gc, new_entry)) {
 378                 new_gc = cpu_replicas_add_entry(&c->replicas_gc, new_entry);
 379                 if (!new_gc.entries)
 380                         goto err;
 381         }
 382
 383         if (!__replicas_has_entry(&c->replicas, new_entry)) {
 384                 new_r = cpu_replicas_add_entry(&c->replicas, new_entry);
 385                 if (!new_r.entries)
 386                         goto err;
 387
 388                 ret = bch2_cpu_replicas_to_sb_replicas(c, &new_r);
 389                 if (ret)
 390                         goto err;
 391
 392                 bch2_journal_entry_res_resize(&c->journal,
 393                                 &c->replicas_journal_res,
 394                                 reserve_journal_replicas(c, &new_r));
 395         }
 396
 397         if (!new_r.entries &&
 398             !new_gc.entries)
 399                 goto out;
 400
 401         /* allocations done, now commit: */
 402
 403         if (new_r.entries)
 404                 bch2_write_super(c);
 405
 406         /* don't update in memory replicas until changes are persistent */
 407         percpu_down_write(&c->mark_lock);
 408         if (new_r.entries)
 409                 ret = replicas_table_update(c, &new_r);
 410         if (new_gc.entries)
 411                 swap(new_gc, c->replicas_gc);
 412         percpu_up_write(&c->mark_lock);
 413 out:
 414         mutex_unlock(&c->sb_lock);
 415
 416         kfree(new_r.entries);
 417         kfree(new_gc.entries);
 418
 419         return ret;
 420 err:
 421         bch_err(c, "error adding replicas entry: memory allocation failure");
 422         ret = -ENOMEM;
 423         goto out;
 424 }
 425
 426 int bch2_mark_replicas(struct bch_fs *c,
 427                        struct bch_replicas_entry *r)
 428 {
 429         return likely(bch2_replicas_marked(c, r, true))
 430                 ? 0
 431                 : bch2_mark_replicas_slowpath(c, r);
 432 }
 433
 434 bool bch2_bkey_replicas_marked_locked(struct bch_fs *c,
 435                                       struct bkey_s_c k,
 436                                       bool check_gc_replicas)
 437 {
 438         struct bch_replicas_padded search;
 439         struct bch_devs_list cached = bch2_bkey_cached_devs(k);
 440         unsigned i;
 441
 442         for (i = 0; i < cached.nr; i++) {
 443                 bch2_replicas_entry_cached(&search.e, cached.devs[i]);
 444
 445                 if (!bch2_replicas_marked_locked(c, &search.e,
 446                                                  check_gc_replicas))
 447                         return false;
 448         }
 449
 450         bch2_bkey_to_replicas(&search.e, k);
 451
 452         return bch2_replicas_marked_locked(c, &search.e, check_gc_replicas);
 453 }
 454
 455 bool bch2_bkey_replicas_marked(struct bch_fs *c,
 456                                struct bkey_s_c k,
 457                                bool check_gc_replicas)
 458 {
 459         bool marked;
 460
 461         percpu_down_read(&c->mark_lock);
 462         marked = bch2_bkey_replicas_marked_locked(c, k, check_gc_replicas);
 463         percpu_up_read(&c->mark_lock);
 464
 465         return marked;
 466 }
 467
 468 int bch2_mark_bkey_replicas(struct bch_fs *c, struct bkey_s_c k)
 469 {
 470         struct bch_replicas_padded search;
 471         struct bch_devs_list cached = bch2_bkey_cached_devs(k);
 472         unsigned i;
 473         int ret;
 474
 475         for (i = 0; i < cached.nr; i++) {
 476                 bch2_replicas_entry_cached(&search.e, cached.devs[i]);
 477
 478                 ret = bch2_mark_replicas(c, &search.e);
 479                 if (ret)
 480                         return ret;
 481         }
 482
 483         bch2_bkey_to_replicas(&search.e, k);
 484
 485         return bch2_mark_replicas(c, &search.e);
 486 }
 487
 488 int bch2_replicas_gc_end(struct bch_fs *c, int ret)
 489 {
 490         unsigned i;
 491
 492         lockdep_assert_held(&c->replicas_gc_lock);
 493
 494         mutex_lock(&c->sb_lock);
 495         percpu_down_write(&c->mark_lock);
 496
 497         /*
 498          * this is kind of crappy; the replicas gc mechanism needs to be ripped
 499          * out
 500          */
 501
 502         for (i = 0; i < c->replicas.nr; i++) {
 503                 struct bch_replicas_entry *e =
 504                         cpu_replicas_entry(&c->replicas, i);
 505                 struct bch_replicas_cpu n;
 506
 507                 if (!__replicas_has_entry(&c->replicas_gc, e) &&
 508                     (c->usage_base->replicas[i] ||
 509                      percpu_u64_get(&c->usage[0]->replicas[i]) ||
 510                      percpu_u64_get(&c->usage[1]->replicas[i]))) {
 511                         n = cpu_replicas_add_entry(&c->replicas_gc, e);
 512                         if (!n.entries) {
 513                                 ret = -ENOSPC;
 514                                 goto err;
 515                         }
 516
 517                         swap(n, c->replicas_gc);
 518                         kfree(n.entries);
 519                 }
 520         }
 521
 522         if (bch2_cpu_replicas_to_sb_replicas(c, &c->replicas_gc)) {
 523                 ret = -ENOSPC;
 524                 goto err;
 525         }
 526
 527         ret = replicas_table_update(c, &c->replicas_gc);
 528 err:
 529         kfree(c->replicas_gc.entries);
 530         c->replicas_gc.entries = NULL;
 531
 532         percpu_up_write(&c->mark_lock);
 533
 534         if (!ret)
 535                 bch2_write_super(c);
 536
 537         mutex_unlock(&c->sb_lock);
 538
 539         return ret;
 540 }
 541
 542 int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask)
 543 {
 544         struct bch_replicas_entry *e;
 545         unsigned i = 0;
 546
 547         lockdep_assert_held(&c->replicas_gc_lock);
 548
 549         mutex_lock(&c->sb_lock);
 550         BUG_ON(c->replicas_gc.entries);
 551
 552         c->replicas_gc.nr               = 0;
 553         c->replicas_gc.entry_size       = 0;
 554
 555         for_each_cpu_replicas_entry(&c->replicas, e)
 556                 if (!((1 << e->data_type) & typemask)) {
 557                         c->replicas_gc.nr++;
 558                         c->replicas_gc.entry_size =
 559                                 max_t(unsigned, c->replicas_gc.entry_size,
 560                                       replicas_entry_bytes(e));
 561                 }
 562
 563         c->replicas_gc.entries = kcalloc(c->replicas_gc.nr,
 564                                          c->replicas_gc.entry_size,
 565                                          GFP_NOIO);
 566         if (!c->replicas_gc.entries) {
 567                 mutex_unlock(&c->sb_lock);
 568                 bch_err(c, "error allocating c->replicas_gc");
 569                 return -ENOMEM;
 570         }
 571
 572         for_each_cpu_replicas_entry(&c->replicas, e)
 573                 if (!((1 << e->data_type) & typemask))
 574                         memcpy(cpu_replicas_entry(&c->replicas_gc, i++),
 575                                e, c->replicas_gc.entry_size);
 576
 577         bch2_cpu_replicas_sort(&c->replicas_gc);
 578         mutex_unlock(&c->sb_lock);
 579
 580         return 0;
 581 }
 582
 583 int bch2_replicas_gc2(struct bch_fs *c)
 584 {
 585         struct bch_replicas_cpu new = { 0 };
 586         unsigned i, nr;
 587         int ret = 0;
 588
 589         bch2_journal_meta(&c->journal);
 590 retry:
 591         nr              = READ_ONCE(c->replicas.nr);
 592         new.entry_size  = READ_ONCE(c->replicas.entry_size);
 593         new.entries     = kcalloc(nr, new.entry_size, GFP_KERNEL);
 594         if (!new.entries) {
 595                 bch_err(c, "error allocating c->replicas_gc");
 596                 return -ENOMEM;
 597         }
 598
 599         mutex_lock(&c->sb_lock);
 600         percpu_down_write(&c->mark_lock);
 601
 602         if (nr                  != c->replicas.nr ||
 603             new.entry_size      != c->replicas.entry_size) {
 604                 percpu_up_write(&c->mark_lock);
 605                 mutex_unlock(&c->sb_lock);
 606                 kfree(new.entries);
 607                 goto retry;
 608         }
 609
 610         for (i = 0; i < c->replicas.nr; i++) {
 611                 struct bch_replicas_entry *e =
 612                         cpu_replicas_entry(&c->replicas, i);
 613
 614                 if (e->data_type == BCH_DATA_JOURNAL ||
 615                     c->usage_base->replicas[i] ||
 616                     percpu_u64_get(&c->usage[0]->replicas[i]) ||
 617                     percpu_u64_get(&c->usage[1]->replicas[i]))
 618                         memcpy(cpu_replicas_entry(&new, new.nr++),
 619                                e, new.entry_size);
 620         }
 621
 622         bch2_cpu_replicas_sort(&new);
 623
 624         if (bch2_cpu_replicas_to_sb_replicas(c, &new)) {
 625                 ret = -ENOSPC;
 626                 goto err;
 627         }
 628
 629         ret = replicas_table_update(c, &new);
 630 err:
 631         kfree(new.entries);
 632
 633         percpu_up_write(&c->mark_lock);
 634
 635         if (!ret)
 636                 bch2_write_super(c);
 637
 638         mutex_unlock(&c->sb_lock);
 639
 640         return ret;
 641 }
 642
 643 int bch2_replicas_set_usage(struct bch_fs *c,
 644                             struct bch_replicas_entry *r,
 645                             u64 sectors)
 646 {
 647         int ret, idx = bch2_replicas_entry_idx(c, r);
 648
 649         if (idx < 0) {
 650                 struct bch_replicas_cpu n;
 651
 652                 n = cpu_replicas_add_entry(&c->replicas, r);
 653                 if (!n.entries)
 654                         return -ENOMEM;
 655
 656                 ret = replicas_table_update(c, &n);
 657                 if (ret)
 658                         return ret;
 659
 660                 kfree(n.entries);
 661
 662                 idx = bch2_replicas_entry_idx(c, r);
 663                 BUG_ON(ret < 0);
 664         }
 665
 666         c->usage_base->replicas[idx] = sectors;
 667
 668         return 0;
 669 }
 670
 671 /* Replicas tracking - superblock: */
 672
 673 static int
 674 __bch2_sb_replicas_to_cpu_replicas(struct bch_sb_field_replicas *sb_r,
 675                                    struct bch_replicas_cpu *cpu_r)
 676 {
 677         struct bch_replicas_entry *e, *dst;
 678         unsigned nr = 0, entry_size = 0, idx = 0;
 679
 680         for_each_replicas_entry(sb_r, e) {
 681                 entry_size = max_t(unsigned, entry_size,
 682                                    replicas_entry_bytes(e));
 683                 nr++;
 684         }
 685
 686         cpu_r->entries = kcalloc(nr, entry_size, GFP_NOIO);
 687         if (!cpu_r->entries)
 688                 return -ENOMEM;
 689
 690         cpu_r->nr               = nr;
 691         cpu_r->entry_size       = entry_size;
 692
 693         for_each_replicas_entry(sb_r, e) {
 694                 dst = cpu_replicas_entry(cpu_r, idx++);
 695                 memcpy(dst, e, replicas_entry_bytes(e));
 696                 replicas_entry_sort(dst);
 697         }
 698
 699         return 0;
 700 }
 701
 702 static int
 703 __bch2_sb_replicas_v0_to_cpu_replicas(struct bch_sb_field_replicas_v0 *sb_r,
 704                                       struct bch_replicas_cpu *cpu_r)
 705 {
 706         struct bch_replicas_entry_v0 *e;
 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         entry_size += sizeof(struct bch_replicas_entry) -
 716                 sizeof(struct bch_replicas_entry_v0);
 717
 718         cpu_r->entries = kcalloc(nr, entry_size, GFP_NOIO);
 719         if (!cpu_r->entries)
 720                 return -ENOMEM;
 721
 722         cpu_r->nr               = nr;
 723         cpu_r->entry_size       = entry_size;
 724
 725         for_each_replicas_entry(sb_r, e) {
 726                 struct bch_replicas_entry *dst =
 727                         cpu_replicas_entry(cpu_r, idx++);
 728
 729                 dst->data_type  = e->data_type;
 730                 dst->nr_devs    = e->nr_devs;
 731                 dst->nr_required = 1;
 732                 memcpy(dst->devs, e->devs, e->nr_devs);
 733                 replicas_entry_sort(dst);
 734         }
 735
 736         return 0;
 737 }
 738
 739 int bch2_sb_replicas_to_cpu_replicas(struct bch_fs *c)
 740 {
 741         struct bch_sb_field_replicas *sb_v1;
 742         struct bch_sb_field_replicas_v0 *sb_v0;
 743         struct bch_replicas_cpu new_r = { 0, 0, NULL };
 744         int ret = 0;
 745
 746         if ((sb_v1 = bch2_sb_get_replicas(c->disk_sb.sb)))
 747                 ret = __bch2_sb_replicas_to_cpu_replicas(sb_v1, &new_r);
 748         else if ((sb_v0 = bch2_sb_get_replicas_v0(c->disk_sb.sb)))
 749                 ret = __bch2_sb_replicas_v0_to_cpu_replicas(sb_v0, &new_r);
 750
 751         if (ret)
 752                 return -ENOMEM;
 753
 754         bch2_cpu_replicas_sort(&new_r);
 755
 756         percpu_down_write(&c->mark_lock);
 757
 758         ret = replicas_table_update(c, &new_r);
 759         percpu_up_write(&c->mark_lock);
 760
 761         kfree(new_r.entries);
 762
 763         return 0;
 764 }
 765
 766 static int bch2_cpu_replicas_to_sb_replicas_v0(struct bch_fs *c,
 767                                                struct bch_replicas_cpu *r)
 768 {
 769         struct bch_sb_field_replicas_v0 *sb_r;
 770         struct bch_replicas_entry_v0 *dst;
 771         struct bch_replicas_entry *src;
 772         size_t bytes;
 773
 774         bytes = sizeof(struct bch_sb_field_replicas);
 775
 776         for_each_cpu_replicas_entry(r, src)
 777                 bytes += replicas_entry_bytes(src) - 1;
 778
 779         sb_r = bch2_sb_resize_replicas_v0(&c->disk_sb,
 780                         DIV_ROUND_UP(bytes, sizeof(u64)));
 781         if (!sb_r)
 782                 return -ENOSPC;
 783
 784         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas);
 785         sb_r = bch2_sb_get_replicas_v0(c->disk_sb.sb);
 786
 787         memset(&sb_r->entries, 0,
 788                vstruct_end(&sb_r->field) -
 789                (void *) &sb_r->entries);
 790
 791         dst = sb_r->entries;
 792         for_each_cpu_replicas_entry(r, src) {
 793                 dst->data_type  = src->data_type;
 794                 dst->nr_devs    = src->nr_devs;
 795                 memcpy(dst->devs, src->devs, src->nr_devs);
 796
 797                 dst = replicas_entry_next(dst);
 798
 799                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 800         }
 801
 802         return 0;
 803 }
 804
 805 static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *c,
 806                                             struct bch_replicas_cpu *r)
 807 {
 808         struct bch_sb_field_replicas *sb_r;
 809         struct bch_replicas_entry *dst, *src;
 810         bool need_v1 = false;
 811         size_t bytes;
 812
 813         bytes = sizeof(struct bch_sb_field_replicas);
 814
 815         for_each_cpu_replicas_entry(r, src) {
 816                 bytes += replicas_entry_bytes(src);
 817                 if (src->nr_required != 1)
 818                         need_v1 = true;
 819         }
 820
 821         if (!need_v1)
 822                 return bch2_cpu_replicas_to_sb_replicas_v0(c, r);
 823
 824         sb_r = bch2_sb_resize_replicas(&c->disk_sb,
 825                         DIV_ROUND_UP(bytes, sizeof(u64)));
 826         if (!sb_r)
 827                 return -ENOSPC;
 828
 829         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas_v0);
 830         sb_r = bch2_sb_get_replicas(c->disk_sb.sb);
 831
 832         memset(&sb_r->entries, 0,
 833                vstruct_end(&sb_r->field) -
 834                (void *) &sb_r->entries);
 835
 836         dst = sb_r->entries;
 837         for_each_cpu_replicas_entry(r, src) {
 838                 memcpy(dst, src, replicas_entry_bytes(src));
 839
 840                 dst = replicas_entry_next(dst);
 841
 842                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 843         }
 844
 845         return 0;
 846 }
 847
 848 static const char *check_dup_replicas_entries(struct bch_replicas_cpu *cpu_r)
 849 {
 850         unsigned i;
 851
 852         sort_cmp_size(cpu_r->entries,
 853                       cpu_r->nr,
 854                       cpu_r->entry_size,
 855                       memcmp, NULL);
 856
 857         for (i = 0; i + 1 < cpu_r->nr; i++) {
 858                 struct bch_replicas_entry *l =
 859                         cpu_replicas_entry(cpu_r, i);
 860                 struct bch_replicas_entry *r =
 861                         cpu_replicas_entry(cpu_r, i + 1);
 862
 863                 BUG_ON(memcmp(l, r, cpu_r->entry_size) > 0);
 864
 865                 if (!memcmp(l, r, cpu_r->entry_size))
 866                         return "duplicate replicas entry";
 867         }
 868
 869         return NULL;
 870 }
 871
 872 static const char *bch2_sb_validate_replicas(struct bch_sb *sb, struct bch_sb_field *f)
 873 {
 874         struct bch_sb_field_replicas *sb_r = field_to_type(f, replicas);
 875         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 876         struct bch_replicas_cpu cpu_r = { .entries = NULL };
 877         struct bch_replicas_entry *e;
 878         const char *err;
 879         unsigned i;
 880
 881         for_each_replicas_entry(sb_r, e) {
 882                 err = "invalid replicas entry: invalid data type";
 883                 if (e->data_type >= BCH_DATA_NR)
 884                         goto err;
 885
 886                 err = "invalid replicas entry: no devices";
 887                 if (!e->nr_devs)
 888                         goto err;
 889
 890                 err = "invalid replicas entry: bad nr_required";
 891                 if (e->nr_required > 1 &&
 892                     e->nr_required >= e->nr_devs)
 893                         goto err;
 894
 895                 err = "invalid replicas entry: invalid device";
 896                 for (i = 0; i < e->nr_devs; i++)
 897                         if (!bch2_dev_exists(sb, mi, e->devs[i]))
 898                                 goto err;
 899         }
 900
 901         err = "cannot allocate memory";
 902         if (__bch2_sb_replicas_to_cpu_replicas(sb_r, &cpu_r))
 903                 goto err;
 904
 905         err = check_dup_replicas_entries(&cpu_r);
 906 err:
 907         kfree(cpu_r.entries);
 908         return err;
 909 }
 910
 911 static void bch2_sb_replicas_to_text(struct printbuf *out,
 912                                      struct bch_sb *sb,
 913                                      struct bch_sb_field *f)
 914 {
 915         struct bch_sb_field_replicas *r = field_to_type(f, replicas);
 916         struct bch_replicas_entry *e;
 917         bool first = true;
 918
 919         for_each_replicas_entry(r, e) {
 920                 if (!first)
 921                         pr_buf(out, " ");
 922                 first = false;
 923
 924                 bch2_replicas_entry_to_text(out, e);
 925         }
 926 }
 927
 928 const struct bch_sb_field_ops bch_sb_field_ops_replicas = {
 929         .validate       = bch2_sb_validate_replicas,
 930         .to_text        = bch2_sb_replicas_to_text,
 931 };
 932
 933 static const char *bch2_sb_validate_replicas_v0(struct bch_sb *sb, struct bch_sb_field *f)
 934 {
 935         struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0);
 936         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 937         struct bch_replicas_cpu cpu_r = { .entries = NULL };
 938         struct bch_replicas_entry_v0 *e;
 939         const char *err;
 940         unsigned i;
 941
 942         for_each_replicas_entry_v0(sb_r, e) {
 943                 err = "invalid replicas entry: invalid data type";
 944                 if (e->data_type >= BCH_DATA_NR)
 945                         goto err;
 946
 947                 err = "invalid replicas entry: no devices";
 948                 if (!e->nr_devs)
 949                         goto err;
 950
 951                 err = "invalid replicas entry: invalid device";
 952                 for (i = 0; i < e->nr_devs; i++)
 953                         if (!bch2_dev_exists(sb, mi, e->devs[i]))
 954                                 goto err;
 955         }
 956
 957         err = "cannot allocate memory";
 958         if (__bch2_sb_replicas_v0_to_cpu_replicas(sb_r, &cpu_r))
 959                 goto err;
 960
 961         err = check_dup_replicas_entries(&cpu_r);
 962 err:
 963         kfree(cpu_r.entries);
 964         return err;
 965 }
 966
 967 const struct bch_sb_field_ops bch_sb_field_ops_replicas_v0 = {
 968         .validate       = bch2_sb_validate_replicas_v0,
 969 };
 970
 971 /* Query replicas: */
 972
 973 struct replicas_status __bch2_replicas_status(struct bch_fs *c,
 974                                               struct bch_devs_mask online_devs)
 975 {
 976         struct bch_sb_field_members *mi;
 977         struct bch_replicas_entry *e;
 978         unsigned i, nr_online, nr_offline;
 979         struct replicas_status ret;
 980
 981         memset(&ret, 0, sizeof(ret));
 982
 983         for (i = 0; i < ARRAY_SIZE(ret.replicas); i++)
 984                 ret.replicas[i].redundancy = INT_MAX;
 985
 986         mi = bch2_sb_get_members(c->disk_sb.sb);
 987
 988         percpu_down_read(&c->mark_lock);
 989
 990         for_each_cpu_replicas_entry(&c->replicas, e) {
 991                 if (e->data_type >= ARRAY_SIZE(ret.replicas))
 992                         panic("e %p data_type %u\n", e, e->data_type);
 993
 994                 nr_online = nr_offline = 0;
 995
 996                 for (i = 0; i < e->nr_devs; i++) {
 997                         BUG_ON(!bch2_dev_exists(c->disk_sb.sb, mi,
 998                                                 e->devs[i]));
 999
1000                         if (test_bit(e->devs[i], online_devs.d))
1001                                 nr_online++;
1002                         else
1003                                 nr_offline++;
1004                 }
1005
1006                 ret.replicas[e->data_type].redundancy =
1007                         min(ret.replicas[e->data_type].redundancy,
1008                             (int) nr_online - (int) e->nr_required);
1009
1010                 ret.replicas[e->data_type].nr_offline =
1011                         max(ret.replicas[e->data_type].nr_offline,
1012                             nr_offline);
1013         }
1014
1015         percpu_up_read(&c->mark_lock);
1016
1017         for (i = 0; i < ARRAY_SIZE(ret.replicas); i++)
1018                 if (ret.replicas[i].redundancy == INT_MAX)
1019                         ret.replicas[i].redundancy = 0;
1020
1021         return ret;
1022 }
1023
1024 struct replicas_status bch2_replicas_status(struct bch_fs *c)
1025 {
1026         return __bch2_replicas_status(c, bch2_online_devs(c));
1027 }
1028
1029 static bool have_enough_devs(struct replicas_status s,
1030                              enum bch_data_type type,
1031                              bool force_if_degraded,
1032                              bool force_if_lost)
1033 {
1034         return (!s.replicas[type].nr_offline || force_if_degraded) &&
1035                 (s.replicas[type].redundancy >= 0 || force_if_lost);
1036 }
1037
1038 bool bch2_have_enough_devs(struct replicas_status s, unsigned flags)
1039 {
1040         return (have_enough_devs(s, BCH_DATA_JOURNAL,
1041                                  flags & BCH_FORCE_IF_METADATA_DEGRADED,
1042                                  flags & BCH_FORCE_IF_METADATA_LOST) &&
1043                 have_enough_devs(s, BCH_DATA_BTREE,
1044                                  flags & BCH_FORCE_IF_METADATA_DEGRADED,
1045                                  flags & BCH_FORCE_IF_METADATA_LOST) &&
1046                 have_enough_devs(s, BCH_DATA_USER,
1047                                  flags & BCH_FORCE_IF_DATA_DEGRADED,
1048                                  flags & BCH_FORCE_IF_DATA_LOST));
1049 }
1050
1051 int bch2_replicas_online(struct bch_fs *c, bool meta)
1052 {
1053         struct replicas_status s = bch2_replicas_status(c);
1054
1055         return (meta
1056                 ? min(s.replicas[BCH_DATA_JOURNAL].redundancy,
1057                       s.replicas[BCH_DATA_BTREE].redundancy)
1058                 : s.replicas[BCH_DATA_USER].redundancy) + 1;
1059 }
1060
1061 unsigned bch2_dev_has_data(struct bch_fs *c, struct bch_dev *ca)
1062 {
1063         struct bch_replicas_entry *e;
1064         unsigned i, ret = 0;
1065
1066         percpu_down_read(&c->mark_lock);
1067
1068         for_each_cpu_replicas_entry(&c->replicas, e)
1069                 for (i = 0; i < e->nr_devs; i++)
1070                         if (e->devs[i] == ca->dev_idx)
1071                                 ret |= 1 << e->data_type;
1072
1073         percpu_up_read(&c->mark_lock);
1074
1075         return ret;
1076 }
1077
1078 int bch2_fs_replicas_init(struct bch_fs *c)
1079 {
1080         c->journal.entry_u64s_reserved +=
1081                 reserve_journal_replicas(c, &c->replicas);
1082
1083         return replicas_table_update(c, &c->replicas);
1084 }