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_sorted(struct bch_replicas_entry *e)
  20 {
  21 #ifdef CONFIG_BCACHES_DEBUG
  22         unsigned i;
  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.ec_nr) {
  83                         r->nr_devs = 0;
  84                         break;
  85                 }
  86
  87                 r->devs[r->nr_devs++] = p.ptr.dev;
  88         }
  89 }
  90
  91 static void stripe_to_replicas(struct bkey_s_c k,
  92                                struct bch_replicas_entry *r)
  93 {
  94         struct bkey_s_c_stripe s = bkey_s_c_to_stripe(k);
  95         const struct bch_extent_ptr *ptr;
  96
  97         r->nr_required  = s.v->nr_blocks - s.v->nr_redundant;
  98
  99         for (ptr = s.v->ptrs;
 100              ptr < s.v->ptrs + s.v->nr_blocks;
 101              ptr++)
 102                 r->devs[r->nr_devs++] = ptr->dev;
 103 }
 104
 105 void bch2_bkey_to_replicas(struct bch_replicas_entry *e,
 106                            struct bkey_s_c k)
 107 {
 108         e->nr_devs = 0;
 109
 110         switch (k.k->type) {
 111         case KEY_TYPE_btree_ptr:
 112                 e->data_type = BCH_DATA_BTREE;
 113                 extent_to_replicas(k, e);
 114                 break;
 115         case KEY_TYPE_extent:
 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_USER;
 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_sorted(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_sorted(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 static bool bch2_replicas_marked_locked(struct bch_fs *c,
 212                           struct bch_replicas_entry *search,
 213                           bool check_gc_replicas)
 214 {
 215         if (!search->nr_devs)
 216                 return true;
 217
 218         verify_replicas_entry_sorted(search);
 219
 220         return __replicas_has_entry(&c->replicas, search) &&
 221                 (!check_gc_replicas ||
 222                  likely((!c->replicas_gc.entries)) ||
 223                  __replicas_has_entry(&c->replicas_gc, search));
 224 }
 225
 226 bool bch2_replicas_marked(struct bch_fs *c,
 227                           struct bch_replicas_entry *search,
 228                           bool check_gc_replicas)
 229 {
 230         bool marked;
 231
 232         percpu_down_read(&c->mark_lock);
 233         marked = bch2_replicas_marked_locked(c, search, check_gc_replicas);
 234         percpu_up_read(&c->mark_lock);
 235
 236         return marked;
 237 }
 238
 239 static void __replicas_table_update(struct bch_fs_usage *dst,
 240                                     struct bch_replicas_cpu *dst_r,
 241                                     struct bch_fs_usage *src,
 242                                     struct bch_replicas_cpu *src_r)
 243 {
 244         int src_idx, dst_idx;
 245
 246         *dst = *src;
 247
 248         for (src_idx = 0; src_idx < src_r->nr; src_idx++) {
 249                 if (!src->replicas[src_idx])
 250                         continue;
 251
 252                 dst_idx = __replicas_entry_idx(dst_r,
 253                                 cpu_replicas_entry(src_r, src_idx));
 254                 BUG_ON(dst_idx < 0);
 255
 256                 dst->replicas[dst_idx] = src->replicas[src_idx];
 257         }
 258 }
 259
 260 static void __replicas_table_update_pcpu(struct bch_fs_usage __percpu *dst_p,
 261                                     struct bch_replicas_cpu *dst_r,
 262                                     struct bch_fs_usage __percpu *src_p,
 263                                     struct bch_replicas_cpu *src_r)
 264 {
 265         unsigned src_nr = sizeof(struct bch_fs_usage) / sizeof(u64) + src_r->nr;
 266         struct bch_fs_usage *dst, *src = (void *)
 267                 bch2_acc_percpu_u64s((void *) src_p, src_nr);
 268
 269         preempt_disable();
 270         dst = this_cpu_ptr(dst_p);
 271         preempt_enable();
 272
 273         __replicas_table_update(dst, dst_r, src, src_r);
 274 }
 275
 276 /*
 277  * Resize filesystem accounting:
 278  */
 279 static int replicas_table_update(struct bch_fs *c,
 280                                  struct bch_replicas_cpu *new_r)
 281 {
 282         struct bch_fs_usage __percpu *new_usage[2] = { NULL, NULL };
 283         struct bch_fs_usage *new_scratch = NULL;
 284         struct bch_fs_usage __percpu *new_gc = NULL;
 285         struct bch_fs_usage *new_base = NULL;
 286         unsigned bytes = sizeof(struct bch_fs_usage) +
 287                 sizeof(u64) * new_r->nr;
 288         int ret = -ENOMEM;
 289
 290         if (!(new_base = kzalloc(bytes, GFP_NOIO)) ||
 291             !(new_usage[0] = __alloc_percpu_gfp(bytes, sizeof(u64),
 292                                                 GFP_NOIO)) ||
 293             !(new_usage[1] = __alloc_percpu_gfp(bytes, sizeof(u64),
 294                                                 GFP_NOIO)) ||
 295             !(new_scratch  = kmalloc(bytes, GFP_NOIO)) ||
 296             (c->usage_gc &&
 297              !(new_gc = __alloc_percpu_gfp(bytes, sizeof(u64), GFP_NOIO))))
 298                 goto err;
 299
 300         if (c->usage_base)
 301                 __replicas_table_update(new_base,               new_r,
 302                                         c->usage_base,          &c->replicas);
 303         if (c->usage[0])
 304                 __replicas_table_update_pcpu(new_usage[0],      new_r,
 305                                              c->usage[0],       &c->replicas);
 306         if (c->usage[1])
 307                 __replicas_table_update_pcpu(new_usage[1],      new_r,
 308                                              c->usage[1],       &c->replicas);
 309         if (c->usage_gc)
 310                 __replicas_table_update_pcpu(new_gc,            new_r,
 311                                              c->usage_gc,       &c->replicas);
 312
 313         swap(c->usage_base,     new_base);
 314         swap(c->usage[0],       new_usage[0]);
 315         swap(c->usage[1],       new_usage[1]);
 316         swap(c->usage_scratch,  new_scratch);
 317         swap(c->usage_gc,       new_gc);
 318         swap(c->replicas,       *new_r);
 319         ret = 0;
 320 err:
 321         free_percpu(new_gc);
 322         kfree(new_scratch);
 323         free_percpu(new_usage[1]);
 324         free_percpu(new_usage[0]);
 325         kfree(new_base);
 326         return ret;
 327 }
 328
 329 static unsigned reserve_journal_replicas(struct bch_fs *c,
 330                                      struct bch_replicas_cpu *r)
 331 {
 332         struct bch_replicas_entry *e;
 333         unsigned journal_res_u64s = 0;
 334
 335         /* nr_inodes: */
 336         journal_res_u64s +=
 337                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 338
 339         /* key_version: */
 340         journal_res_u64s +=
 341                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 342
 343         /* persistent_reserved: */
 344         journal_res_u64s +=
 345                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64)) *
 346                 BCH_REPLICAS_MAX;
 347
 348         for_each_cpu_replicas_entry(r, e)
 349                 journal_res_u64s +=
 350                         DIV_ROUND_UP(sizeof(struct jset_entry_data_usage) +
 351                                      e->nr_devs, sizeof(u64));
 352         return journal_res_u64s;
 353 }
 354
 355 noinline
 356 static int bch2_mark_replicas_slowpath(struct bch_fs *c,
 357                                 struct bch_replicas_entry *new_entry)
 358 {
 359         struct bch_replicas_cpu new_r, new_gc;
 360         int ret = -ENOMEM;
 361
 362         memset(&new_r, 0, sizeof(new_r));
 363         memset(&new_gc, 0, sizeof(new_gc));
 364
 365         mutex_lock(&c->sb_lock);
 366
 367         if (c->replicas_gc.entries &&
 368             !__replicas_has_entry(&c->replicas_gc, new_entry)) {
 369                 new_gc = cpu_replicas_add_entry(&c->replicas_gc, new_entry);
 370                 if (!new_gc.entries)
 371                         goto err;
 372         }
 373
 374         if (!__replicas_has_entry(&c->replicas, new_entry)) {
 375                 new_r = cpu_replicas_add_entry(&c->replicas, new_entry);
 376                 if (!new_r.entries)
 377                         goto err;
 378
 379                 ret = bch2_cpu_replicas_to_sb_replicas(c, &new_r);
 380                 if (ret)
 381                         goto err;
 382
 383                 bch2_journal_entry_res_resize(&c->journal,
 384                                 &c->replicas_journal_res,
 385                                 reserve_journal_replicas(c, &new_r));
 386         }
 387
 388         if (!new_r.entries &&
 389             !new_gc.entries)
 390                 goto out;
 391
 392         /* allocations done, now commit: */
 393
 394         if (new_r.entries)
 395                 bch2_write_super(c);
 396
 397         /* don't update in memory replicas until changes are persistent */
 398         percpu_down_write(&c->mark_lock);
 399         if (new_r.entries)
 400                 ret = replicas_table_update(c, &new_r);
 401         if (new_gc.entries)
 402                 swap(new_gc, c->replicas_gc);
 403         percpu_up_write(&c->mark_lock);
 404 out:
 405         ret = 0;
 406 err:
 407         mutex_unlock(&c->sb_lock);
 408
 409         kfree(new_r.entries);
 410         kfree(new_gc.entries);
 411
 412         return ret;
 413 }
 414
 415 int bch2_mark_replicas(struct bch_fs *c,
 416                        struct bch_replicas_entry *r)
 417 {
 418         return likely(bch2_replicas_marked(c, r, true))
 419                 ? 0
 420                 : bch2_mark_replicas_slowpath(c, r);
 421 }
 422
 423 bool bch2_bkey_replicas_marked_locked(struct bch_fs *c,
 424                                       struct bkey_s_c k,
 425                                       bool check_gc_replicas)
 426 {
 427         struct bch_replicas_padded search;
 428         struct bch_devs_list cached = bch2_bkey_cached_devs(k);
 429         unsigned i;
 430
 431         for (i = 0; i < cached.nr; i++) {
 432                 bch2_replicas_entry_cached(&search.e, cached.devs[i]);
 433
 434                 if (!bch2_replicas_marked_locked(c, &search.e,
 435                                                  check_gc_replicas))
 436                         return false;
 437         }
 438
 439         bch2_bkey_to_replicas(&search.e, k);
 440
 441         return bch2_replicas_marked_locked(c, &search.e, check_gc_replicas);
 442 }
 443
 444 bool bch2_bkey_replicas_marked(struct bch_fs *c,
 445                                struct bkey_s_c k,
 446                                bool check_gc_replicas)
 447 {
 448         bool marked;
 449
 450         percpu_down_read(&c->mark_lock);
 451         marked = bch2_bkey_replicas_marked_locked(c, k, check_gc_replicas);
 452         percpu_up_read(&c->mark_lock);
 453
 454         return marked;
 455 }
 456
 457 int bch2_mark_bkey_replicas(struct bch_fs *c, struct bkey_s_c k)
 458 {
 459         struct bch_replicas_padded search;
 460         struct bch_devs_list cached = bch2_bkey_cached_devs(k);
 461         unsigned i;
 462         int ret;
 463
 464         for (i = 0; i < cached.nr; i++) {
 465                 bch2_replicas_entry_cached(&search.e, cached.devs[i]);
 466
 467                 ret = bch2_mark_replicas(c, &search.e);
 468                 if (ret)
 469                         return ret;
 470         }
 471
 472         bch2_bkey_to_replicas(&search.e, k);
 473
 474         return bch2_mark_replicas(c, &search.e);
 475 }
 476
 477 int bch2_replicas_gc_end(struct bch_fs *c, int ret)
 478 {
 479         unsigned i;
 480
 481         lockdep_assert_held(&c->replicas_gc_lock);
 482
 483         mutex_lock(&c->sb_lock);
 484         percpu_down_write(&c->mark_lock);
 485
 486         /*
 487          * this is kind of crappy; the replicas gc mechanism needs to be ripped
 488          * out
 489          */
 490
 491         for (i = 0; i < c->replicas.nr; i++) {
 492                 struct bch_replicas_entry *e =
 493                         cpu_replicas_entry(&c->replicas, i);
 494                 struct bch_replicas_cpu n;
 495
 496                 if (!__replicas_has_entry(&c->replicas_gc, e) &&
 497                     (c->usage_base->replicas[i] ||
 498                      percpu_u64_get(&c->usage[0]->replicas[i]) ||
 499                      percpu_u64_get(&c->usage[1]->replicas[i]))) {
 500                         n = cpu_replicas_add_entry(&c->replicas_gc, e);
 501                         if (!n.entries) {
 502                                 ret = -ENOSPC;
 503                                 goto err;
 504                         }
 505
 506                         swap(n, c->replicas_gc);
 507                         kfree(n.entries);
 508                 }
 509         }
 510
 511         if (bch2_cpu_replicas_to_sb_replicas(c, &c->replicas_gc)) {
 512                 ret = -ENOSPC;
 513                 goto err;
 514         }
 515
 516         ret = replicas_table_update(c, &c->replicas_gc);
 517 err:
 518         kfree(c->replicas_gc.entries);
 519         c->replicas_gc.entries = NULL;
 520
 521         percpu_up_write(&c->mark_lock);
 522
 523         if (!ret)
 524                 bch2_write_super(c);
 525
 526         mutex_unlock(&c->sb_lock);
 527
 528         return ret;
 529 }
 530
 531 int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask)
 532 {
 533         struct bch_replicas_entry *e;
 534         unsigned i = 0;
 535
 536         lockdep_assert_held(&c->replicas_gc_lock);
 537
 538         mutex_lock(&c->sb_lock);
 539         BUG_ON(c->replicas_gc.entries);
 540
 541         c->replicas_gc.nr               = 0;
 542         c->replicas_gc.entry_size       = 0;
 543
 544         for_each_cpu_replicas_entry(&c->replicas, e)
 545                 if (!((1 << e->data_type) & typemask)) {
 546                         c->replicas_gc.nr++;
 547                         c->replicas_gc.entry_size =
 548                                 max_t(unsigned, c->replicas_gc.entry_size,
 549                                       replicas_entry_bytes(e));
 550                 }
 551
 552         c->replicas_gc.entries = kcalloc(c->replicas_gc.nr,
 553                                          c->replicas_gc.entry_size,
 554                                          GFP_NOIO);
 555         if (!c->replicas_gc.entries) {
 556                 mutex_unlock(&c->sb_lock);
 557                 return -ENOMEM;
 558         }
 559
 560         for_each_cpu_replicas_entry(&c->replicas, e)
 561                 if (!((1 << e->data_type) & typemask))
 562                         memcpy(cpu_replicas_entry(&c->replicas_gc, i++),
 563                                e, c->replicas_gc.entry_size);
 564
 565         bch2_cpu_replicas_sort(&c->replicas_gc);
 566         mutex_unlock(&c->sb_lock);
 567
 568         return 0;
 569 }
 570
 571 int bch2_replicas_gc2(struct bch_fs *c)
 572 {
 573         struct bch_replicas_cpu new = { 0 };
 574         unsigned i, nr;
 575         int ret = 0;
 576
 577         bch2_journal_meta(&c->journal);
 578 retry:
 579         nr              = READ_ONCE(c->replicas.nr);
 580         new.entry_size  = READ_ONCE(c->replicas.entry_size);
 581         new.entries     = kcalloc(nr, new.entry_size, GFP_KERNEL);
 582         if (!new.entries)
 583                 return -ENOMEM;
 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                         memcpy(cpu_replicas_entry(&new, new.nr++),
 605                                e, new.entry_size);
 606         }
 607
 608         bch2_cpu_replicas_sort(&new);
 609
 610         if (bch2_cpu_replicas_to_sb_replicas(c, &new)) {
 611                 ret = -ENOSPC;
 612                 goto err;
 613         }
 614
 615         ret = replicas_table_update(c, &new);
 616 err:
 617         kfree(new.entries);
 618
 619         percpu_up_write(&c->mark_lock);
 620
 621         if (!ret)
 622                 bch2_write_super(c);
 623
 624         mutex_unlock(&c->sb_lock);
 625
 626         return ret;
 627 }
 628
 629 int bch2_replicas_set_usage(struct bch_fs *c,
 630                             struct bch_replicas_entry *r,
 631                             u64 sectors)
 632 {
 633         int ret, idx = bch2_replicas_entry_idx(c, r);
 634
 635         if (idx < 0) {
 636                 struct bch_replicas_cpu n;
 637
 638                 n = cpu_replicas_add_entry(&c->replicas, r);
 639                 if (!n.entries)
 640                         return -ENOMEM;
 641
 642                 ret = replicas_table_update(c, &n);
 643                 if (ret)
 644                         return ret;
 645
 646                 kfree(n.entries);
 647
 648                 idx = bch2_replicas_entry_idx(c, r);
 649                 BUG_ON(ret < 0);
 650         }
 651
 652         c->usage_base->replicas[idx] = sectors;
 653
 654         return 0;
 655 }
 656
 657 /* Replicas tracking - superblock: */
 658
 659 static int
 660 __bch2_sb_replicas_to_cpu_replicas(struct bch_sb_field_replicas *sb_r,
 661                                    struct bch_replicas_cpu *cpu_r)
 662 {
 663         struct bch_replicas_entry *e, *dst;
 664         unsigned nr = 0, entry_size = 0, idx = 0;
 665
 666         for_each_replicas_entry(sb_r, e) {
 667                 entry_size = max_t(unsigned, entry_size,
 668                                    replicas_entry_bytes(e));
 669                 nr++;
 670         }
 671
 672         cpu_r->entries = kcalloc(nr, entry_size, GFP_NOIO);
 673         if (!cpu_r->entries)
 674                 return -ENOMEM;
 675
 676         cpu_r->nr               = nr;
 677         cpu_r->entry_size       = entry_size;
 678
 679         for_each_replicas_entry(sb_r, e) {
 680                 dst = cpu_replicas_entry(cpu_r, idx++);
 681                 memcpy(dst, e, replicas_entry_bytes(e));
 682                 replicas_entry_sort(dst);
 683         }
 684
 685         return 0;
 686 }
 687
 688 static int
 689 __bch2_sb_replicas_v0_to_cpu_replicas(struct bch_sb_field_replicas_v0 *sb_r,
 690                                       struct bch_replicas_cpu *cpu_r)
 691 {
 692         struct bch_replicas_entry_v0 *e;
 693         unsigned nr = 0, entry_size = 0, idx = 0;
 694
 695         for_each_replicas_entry(sb_r, e) {
 696                 entry_size = max_t(unsigned, entry_size,
 697                                    replicas_entry_bytes(e));
 698                 nr++;
 699         }
 700
 701         entry_size += sizeof(struct bch_replicas_entry) -
 702                 sizeof(struct bch_replicas_entry_v0);
 703
 704         cpu_r->entries = kcalloc(nr, entry_size, GFP_NOIO);
 705         if (!cpu_r->entries)
 706                 return -ENOMEM;
 707
 708         cpu_r->nr               = nr;
 709         cpu_r->entry_size       = entry_size;
 710
 711         for_each_replicas_entry(sb_r, e) {
 712                 struct bch_replicas_entry *dst =
 713                         cpu_replicas_entry(cpu_r, idx++);
 714
 715                 dst->data_type  = e->data_type;
 716                 dst->nr_devs    = e->nr_devs;
 717                 dst->nr_required = 1;
 718                 memcpy(dst->devs, e->devs, e->nr_devs);
 719                 replicas_entry_sort(dst);
 720         }
 721
 722         return 0;
 723 }
 724
 725 int bch2_sb_replicas_to_cpu_replicas(struct bch_fs *c)
 726 {
 727         struct bch_sb_field_replicas *sb_v1;
 728         struct bch_sb_field_replicas_v0 *sb_v0;
 729         struct bch_replicas_cpu new_r = { 0, 0, NULL };
 730         int ret = 0;
 731
 732         if ((sb_v1 = bch2_sb_get_replicas(c->disk_sb.sb)))
 733                 ret = __bch2_sb_replicas_to_cpu_replicas(sb_v1, &new_r);
 734         else if ((sb_v0 = bch2_sb_get_replicas_v0(c->disk_sb.sb)))
 735                 ret = __bch2_sb_replicas_v0_to_cpu_replicas(sb_v0, &new_r);
 736
 737         if (ret)
 738                 return -ENOMEM;
 739
 740         bch2_cpu_replicas_sort(&new_r);
 741
 742         percpu_down_write(&c->mark_lock);
 743
 744         ret = replicas_table_update(c, &new_r);
 745         percpu_up_write(&c->mark_lock);
 746
 747         kfree(new_r.entries);
 748
 749         return 0;
 750 }
 751
 752 static int bch2_cpu_replicas_to_sb_replicas_v0(struct bch_fs *c,
 753                                                struct bch_replicas_cpu *r)
 754 {
 755         struct bch_sb_field_replicas_v0 *sb_r;
 756         struct bch_replicas_entry_v0 *dst;
 757         struct bch_replicas_entry *src;
 758         size_t bytes;
 759
 760         bytes = sizeof(struct bch_sb_field_replicas);
 761
 762         for_each_cpu_replicas_entry(r, src)
 763                 bytes += replicas_entry_bytes(src) - 1;
 764
 765         sb_r = bch2_sb_resize_replicas_v0(&c->disk_sb,
 766                         DIV_ROUND_UP(bytes, sizeof(u64)));
 767         if (!sb_r)
 768                 return -ENOSPC;
 769
 770         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas);
 771         sb_r = bch2_sb_get_replicas_v0(c->disk_sb.sb);
 772
 773         memset(&sb_r->entries, 0,
 774                vstruct_end(&sb_r->field) -
 775                (void *) &sb_r->entries);
 776
 777         dst = sb_r->entries;
 778         for_each_cpu_replicas_entry(r, src) {
 779                 dst->data_type  = src->data_type;
 780                 dst->nr_devs    = src->nr_devs;
 781                 memcpy(dst->devs, src->devs, src->nr_devs);
 782
 783                 dst = replicas_entry_next(dst);
 784
 785                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 786         }
 787
 788         return 0;
 789 }
 790
 791 static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *c,
 792                                             struct bch_replicas_cpu *r)
 793 {
 794         struct bch_sb_field_replicas *sb_r;
 795         struct bch_replicas_entry *dst, *src;
 796         bool need_v1 = false;
 797         size_t bytes;
 798
 799         bytes = sizeof(struct bch_sb_field_replicas);
 800
 801         for_each_cpu_replicas_entry(r, src) {
 802                 bytes += replicas_entry_bytes(src);
 803                 if (src->nr_required != 1)
 804                         need_v1 = true;
 805         }
 806
 807         if (!need_v1)
 808                 return bch2_cpu_replicas_to_sb_replicas_v0(c, r);
 809
 810         sb_r = bch2_sb_resize_replicas(&c->disk_sb,
 811                         DIV_ROUND_UP(bytes, sizeof(u64)));
 812         if (!sb_r)
 813                 return -ENOSPC;
 814
 815         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas_v0);
 816         sb_r = bch2_sb_get_replicas(c->disk_sb.sb);
 817
 818         memset(&sb_r->entries, 0,
 819                vstruct_end(&sb_r->field) -
 820                (void *) &sb_r->entries);
 821
 822         dst = sb_r->entries;
 823         for_each_cpu_replicas_entry(r, src) {
 824                 memcpy(dst, src, replicas_entry_bytes(src));
 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 const char *check_dup_replicas_entries(struct bch_replicas_cpu *cpu_r)
 835 {
 836         unsigned i;
 837
 838         sort_cmp_size(cpu_r->entries,
 839                       cpu_r->nr,
 840                       cpu_r->entry_size,
 841                       memcmp, NULL);
 842
 843         for (i = 0; i + 1 < cpu_r->nr; i++) {
 844                 struct bch_replicas_entry *l =
 845                         cpu_replicas_entry(cpu_r, i);
 846                 struct bch_replicas_entry *r =
 847                         cpu_replicas_entry(cpu_r, i + 1);
 848
 849                 BUG_ON(memcmp(l, r, cpu_r->entry_size) > 0);
 850
 851                 if (!memcmp(l, r, cpu_r->entry_size))
 852                         return "duplicate replicas entry";
 853         }
 854
 855         return NULL;
 856 }
 857
 858 static const char *bch2_sb_validate_replicas(struct bch_sb *sb, struct bch_sb_field *f)
 859 {
 860         struct bch_sb_field_replicas *sb_r = field_to_type(f, replicas);
 861         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 862         struct bch_replicas_cpu cpu_r = { .entries = NULL };
 863         struct bch_replicas_entry *e;
 864         const char *err;
 865         unsigned i;
 866
 867         for_each_replicas_entry(sb_r, e) {
 868                 err = "invalid replicas entry: invalid data type";
 869                 if (e->data_type >= BCH_DATA_NR)
 870                         goto err;
 871
 872                 err = "invalid replicas entry: no devices";
 873                 if (!e->nr_devs)
 874                         goto err;
 875
 876                 err = "invalid replicas entry: bad nr_required";
 877                 if (!e->nr_required ||
 878                     (e->nr_required > 1 &&
 879                      e->nr_required >= e->nr_devs))
 880                         goto err;
 881
 882                 err = "invalid replicas entry: invalid device";
 883                 for (i = 0; i < e->nr_devs; i++)
 884                         if (!bch2_dev_exists(sb, mi, e->devs[i]))
 885                                 goto err;
 886         }
 887
 888         err = "cannot allocate memory";
 889         if (__bch2_sb_replicas_to_cpu_replicas(sb_r, &cpu_r))
 890                 goto err;
 891
 892         err = check_dup_replicas_entries(&cpu_r);
 893 err:
 894         kfree(cpu_r.entries);
 895         return err;
 896 }
 897
 898 static void bch2_sb_replicas_to_text(struct printbuf *out,
 899                                      struct bch_sb *sb,
 900                                      struct bch_sb_field *f)
 901 {
 902         struct bch_sb_field_replicas *r = field_to_type(f, replicas);
 903         struct bch_replicas_entry *e;
 904         bool first = true;
 905
 906         for_each_replicas_entry(r, e) {
 907                 if (!first)
 908                         pr_buf(out, " ");
 909                 first = false;
 910
 911                 bch2_replicas_entry_to_text(out, e);
 912         }
 913 }
 914
 915 const struct bch_sb_field_ops bch_sb_field_ops_replicas = {
 916         .validate       = bch2_sb_validate_replicas,
 917         .to_text        = bch2_sb_replicas_to_text,
 918 };
 919
 920 static const char *bch2_sb_validate_replicas_v0(struct bch_sb *sb, struct bch_sb_field *f)
 921 {
 922         struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0);
 923         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 924         struct bch_replicas_cpu cpu_r = { .entries = NULL };
 925         struct bch_replicas_entry_v0 *e;
 926         const char *err;
 927         unsigned i;
 928
 929         for_each_replicas_entry_v0(sb_r, e) {
 930                 err = "invalid replicas entry: invalid data type";
 931                 if (e->data_type >= BCH_DATA_NR)
 932                         goto err;
 933
 934                 err = "invalid replicas entry: no devices";
 935                 if (!e->nr_devs)
 936                         goto err;
 937
 938                 err = "invalid replicas entry: invalid device";
 939                 for (i = 0; i < e->nr_devs; i++)
 940                         if (!bch2_dev_exists(sb, mi, e->devs[i]))
 941                                 goto err;
 942         }
 943
 944         err = "cannot allocate memory";
 945         if (__bch2_sb_replicas_v0_to_cpu_replicas(sb_r, &cpu_r))
 946                 goto err;
 947
 948         err = check_dup_replicas_entries(&cpu_r);
 949 err:
 950         kfree(cpu_r.entries);
 951         return err;
 952 }
 953
 954 const struct bch_sb_field_ops bch_sb_field_ops_replicas_v0 = {
 955         .validate       = bch2_sb_validate_replicas_v0,
 956 };
 957
 958 /* Query replicas: */
 959
 960 struct replicas_status __bch2_replicas_status(struct bch_fs *c,
 961                                               struct bch_devs_mask online_devs)
 962 {
 963         struct bch_sb_field_members *mi;
 964         struct bch_replicas_entry *e;
 965         unsigned i, nr_online, nr_offline;
 966         struct replicas_status ret;
 967
 968         memset(&ret, 0, sizeof(ret));
 969
 970         for (i = 0; i < ARRAY_SIZE(ret.replicas); i++)
 971                 ret.replicas[i].redundancy = INT_MAX;
 972
 973         mi = bch2_sb_get_members(c->disk_sb.sb);
 974
 975         percpu_down_read(&c->mark_lock);
 976
 977         for_each_cpu_replicas_entry(&c->replicas, e) {
 978                 if (e->data_type >= ARRAY_SIZE(ret.replicas))
 979                         panic("e %p data_type %u\n", e, e->data_type);
 980
 981                 nr_online = nr_offline = 0;
 982
 983                 for (i = 0; i < e->nr_devs; i++) {
 984                         BUG_ON(!bch2_dev_exists(c->disk_sb.sb, mi,
 985                                                 e->devs[i]));
 986
 987                         if (test_bit(e->devs[i], online_devs.d))
 988                                 nr_online++;
 989                         else
 990                                 nr_offline++;
 991                 }
 992
 993                 ret.replicas[e->data_type].redundancy =
 994                         min(ret.replicas[e->data_type].redundancy,
 995                             (int) nr_online - (int) e->nr_required);
 996
 997                 ret.replicas[e->data_type].nr_offline =
 998                         max(ret.replicas[e->data_type].nr_offline,
 999                             nr_offline);
1000         }
1001
1002         percpu_up_read(&c->mark_lock);
1003
1004         for (i = 0; i < ARRAY_SIZE(ret.replicas); i++)
1005                 if (ret.replicas[i].redundancy == INT_MAX)
1006                         ret.replicas[i].redundancy = 0;
1007
1008         return ret;
1009 }
1010
1011 struct replicas_status bch2_replicas_status(struct bch_fs *c)
1012 {
1013         return __bch2_replicas_status(c, bch2_online_devs(c));
1014 }
1015
1016 static bool have_enough_devs(struct replicas_status s,
1017                              enum bch_data_type type,
1018                              bool force_if_degraded,
1019                              bool force_if_lost)
1020 {
1021         return (!s.replicas[type].nr_offline || force_if_degraded) &&
1022                 (s.replicas[type].redundancy >= 0 || force_if_lost);
1023 }
1024
1025 bool bch2_have_enough_devs(struct replicas_status s, unsigned flags)
1026 {
1027         return (have_enough_devs(s, BCH_DATA_JOURNAL,
1028                                  flags & BCH_FORCE_IF_METADATA_DEGRADED,
1029                                  flags & BCH_FORCE_IF_METADATA_LOST) &&
1030                 have_enough_devs(s, BCH_DATA_BTREE,
1031                                  flags & BCH_FORCE_IF_METADATA_DEGRADED,
1032                                  flags & BCH_FORCE_IF_METADATA_LOST) &&
1033                 have_enough_devs(s, BCH_DATA_USER,
1034                                  flags & BCH_FORCE_IF_DATA_DEGRADED,
1035                                  flags & BCH_FORCE_IF_DATA_LOST));
1036 }
1037
1038 int bch2_replicas_online(struct bch_fs *c, bool meta)
1039 {
1040         struct replicas_status s = bch2_replicas_status(c);
1041
1042         return (meta
1043                 ? min(s.replicas[BCH_DATA_JOURNAL].redundancy,
1044                       s.replicas[BCH_DATA_BTREE].redundancy)
1045                 : s.replicas[BCH_DATA_USER].redundancy) + 1;
1046 }
1047
1048 unsigned bch2_dev_has_data(struct bch_fs *c, struct bch_dev *ca)
1049 {
1050         struct bch_replicas_entry *e;
1051         unsigned i, ret = 0;
1052
1053         percpu_down_read(&c->mark_lock);
1054
1055         for_each_cpu_replicas_entry(&c->replicas, e)
1056                 for (i = 0; i < e->nr_devs; i++)
1057                         if (e->devs[i] == ca->dev_idx)
1058                                 ret |= 1 << e->data_type;
1059
1060         percpu_up_read(&c->mark_lock);
1061
1062         return ret;
1063 }
1064
1065 int bch2_fs_replicas_init(struct bch_fs *c)
1066 {
1067         c->journal.entry_u64s_reserved +=
1068                 reserve_journal_replicas(c, &c->replicas);
1069
1070         return replicas_table_update(c, &c->replicas);
1071 }