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_v0_to_text(struct printbuf *out,
  40                                     struct bch_replicas_entry_v0 *e)
  41 {
  42         unsigned i;
  43
  44         if (e->data_type < BCH_DATA_NR)
  45                 prt_printf(out, "%s", bch2_data_types[e->data_type]);
  46         else
  47                 prt_printf(out, "(invalid data type %u)", e->data_type);
  48
  49         prt_printf(out, ": %u [", e->nr_devs);
  50         for (i = 0; i < e->nr_devs; i++)
  51                 prt_printf(out, i ? " %u" : "%u", e->devs[i]);
  52         prt_printf(out, "]");
  53 }
  54
  55 void bch2_replicas_entry_to_text(struct printbuf *out,
  56                                  struct bch_replicas_entry *e)
  57 {
  58         unsigned i;
  59
  60         if (e->data_type < BCH_DATA_NR)
  61                 prt_printf(out, "%s", bch2_data_types[e->data_type]);
  62         else
  63                 prt_printf(out, "(invalid data type %u)", e->data_type);
  64
  65         prt_printf(out, ": %u/%u [", e->nr_required, e->nr_devs);
  66         for (i = 0; i < e->nr_devs; i++)
  67                 prt_printf(out, i ? " %u" : "%u", e->devs[i]);
  68         prt_printf(out, "]");
  69 }
  70
  71 void bch2_cpu_replicas_to_text(struct printbuf *out,
  72                                struct bch_replicas_cpu *r)
  73 {
  74         struct bch_replicas_entry *e;
  75         bool first = true;
  76
  77         for_each_cpu_replicas_entry(r, e) {
  78                 if (!first)
  79                         prt_printf(out, " ");
  80                 first = false;
  81
  82                 bch2_replicas_entry_to_text(out, e);
  83         }
  84 }
  85
  86 static void extent_to_replicas(struct bkey_s_c k,
  87                                struct bch_replicas_entry *r)
  88 {
  89         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
  90         const union bch_extent_entry *entry;
  91         struct extent_ptr_decoded p;
  92
  93         r->nr_required  = 1;
  94
  95         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
  96                 if (p.ptr.cached)
  97                         continue;
  98
  99                 if (!p.has_ec)
 100                         r->devs[r->nr_devs++] = p.ptr.dev;
 101                 else
 102                         r->nr_required = 0;
 103         }
 104 }
 105
 106 static void stripe_to_replicas(struct bkey_s_c k,
 107                                struct bch_replicas_entry *r)
 108 {
 109         struct bkey_s_c_stripe s = bkey_s_c_to_stripe(k);
 110         const struct bch_extent_ptr *ptr;
 111
 112         r->nr_required  = s.v->nr_blocks - s.v->nr_redundant;
 113
 114         for (ptr = s.v->ptrs;
 115              ptr < s.v->ptrs + s.v->nr_blocks;
 116              ptr++)
 117                 r->devs[r->nr_devs++] = ptr->dev;
 118 }
 119
 120 void bch2_bkey_to_replicas(struct bch_replicas_entry *e,
 121                            struct bkey_s_c k)
 122 {
 123         e->nr_devs = 0;
 124
 125         switch (k.k->type) {
 126         case KEY_TYPE_btree_ptr:
 127         case KEY_TYPE_btree_ptr_v2:
 128                 e->data_type = BCH_DATA_btree;
 129                 extent_to_replicas(k, e);
 130                 break;
 131         case KEY_TYPE_extent:
 132         case KEY_TYPE_reflink_v:
 133                 e->data_type = BCH_DATA_user;
 134                 extent_to_replicas(k, e);
 135                 break;
 136         case KEY_TYPE_stripe:
 137                 e->data_type = BCH_DATA_parity;
 138                 stripe_to_replicas(k, e);
 139                 break;
 140         }
 141
 142         bch2_replicas_entry_sort(e);
 143 }
 144
 145 void bch2_devlist_to_replicas(struct bch_replicas_entry *e,
 146                               enum bch_data_type data_type,
 147                               struct bch_devs_list devs)
 148 {
 149         unsigned i;
 150
 151         BUG_ON(!data_type ||
 152                data_type == BCH_DATA_sb ||
 153                data_type >= BCH_DATA_NR);
 154
 155         e->data_type    = data_type;
 156         e->nr_devs      = 0;
 157         e->nr_required  = 1;
 158
 159         for (i = 0; i < devs.nr; i++)
 160                 e->devs[e->nr_devs++] = devs.devs[i];
 161
 162         bch2_replicas_entry_sort(e);
 163 }
 164
 165 static struct bch_replicas_cpu
 166 cpu_replicas_add_entry(struct bch_replicas_cpu *old,
 167                        struct bch_replicas_entry *new_entry)
 168 {
 169         unsigned i;
 170         struct bch_replicas_cpu new = {
 171                 .nr             = old->nr + 1,
 172                 .entry_size     = max_t(unsigned, old->entry_size,
 173                                         replicas_entry_bytes(new_entry)),
 174         };
 175
 176         BUG_ON(!new_entry->data_type);
 177         verify_replicas_entry(new_entry);
 178
 179         new.entries = kcalloc(new.nr, new.entry_size, GFP_KERNEL);
 180         if (!new.entries)
 181                 return new;
 182
 183         for (i = 0; i < old->nr; i++)
 184                 memcpy(cpu_replicas_entry(&new, i),
 185                        cpu_replicas_entry(old, i),
 186                        old->entry_size);
 187
 188         memcpy(cpu_replicas_entry(&new, old->nr),
 189                new_entry,
 190                replicas_entry_bytes(new_entry));
 191
 192         bch2_cpu_replicas_sort(&new);
 193         return new;
 194 }
 195
 196 static inline int __replicas_entry_idx(struct bch_replicas_cpu *r,
 197                                        struct bch_replicas_entry *search)
 198 {
 199         int idx, entry_size = replicas_entry_bytes(search);
 200
 201         if (unlikely(entry_size > r->entry_size))
 202                 return -1;
 203
 204         verify_replicas_entry(search);
 205
 206 #define entry_cmp(_l, _r, size) memcmp(_l, _r, entry_size)
 207         idx = eytzinger0_find(r->entries, r->nr, r->entry_size,
 208                               entry_cmp, search);
 209 #undef entry_cmp
 210
 211         return idx < r->nr ? idx : -1;
 212 }
 213
 214 int bch2_replicas_entry_idx(struct bch_fs *c,
 215                             struct bch_replicas_entry *search)
 216 {
 217         bch2_replicas_entry_sort(search);
 218
 219         return __replicas_entry_idx(&c->replicas, search);
 220 }
 221
 222 static bool __replicas_has_entry(struct bch_replicas_cpu *r,
 223                                  struct bch_replicas_entry *search)
 224 {
 225         return __replicas_entry_idx(r, search) >= 0;
 226 }
 227
 228 bool bch2_replicas_marked(struct bch_fs *c,
 229                           struct bch_replicas_entry *search)
 230 {
 231         bool marked;
 232
 233         if (!search->nr_devs)
 234                 return true;
 235
 236         verify_replicas_entry(search);
 237
 238         percpu_down_read(&c->mark_lock);
 239         marked = __replicas_has_entry(&c->replicas, search) &&
 240                 (likely((!c->replicas_gc.entries)) ||
 241                  __replicas_has_entry(&c->replicas_gc, search));
 242         percpu_up_read(&c->mark_lock);
 243
 244         return marked;
 245 }
 246
 247 static void __replicas_table_update(struct bch_fs_usage *dst,
 248                                     struct bch_replicas_cpu *dst_r,
 249                                     struct bch_fs_usage *src,
 250                                     struct bch_replicas_cpu *src_r)
 251 {
 252         int src_idx, dst_idx;
 253
 254         *dst = *src;
 255
 256         for (src_idx = 0; src_idx < src_r->nr; src_idx++) {
 257                 if (!src->replicas[src_idx])
 258                         continue;
 259
 260                 dst_idx = __replicas_entry_idx(dst_r,
 261                                 cpu_replicas_entry(src_r, src_idx));
 262                 BUG_ON(dst_idx < 0);
 263
 264                 dst->replicas[dst_idx] = src->replicas[src_idx];
 265         }
 266 }
 267
 268 static void __replicas_table_update_pcpu(struct bch_fs_usage __percpu *dst_p,
 269                                     struct bch_replicas_cpu *dst_r,
 270                                     struct bch_fs_usage __percpu *src_p,
 271                                     struct bch_replicas_cpu *src_r)
 272 {
 273         unsigned src_nr = sizeof(struct bch_fs_usage) / sizeof(u64) + src_r->nr;
 274         struct bch_fs_usage *dst, *src = (void *)
 275                 bch2_acc_percpu_u64s((void *) src_p, src_nr);
 276
 277         preempt_disable();
 278         dst = this_cpu_ptr(dst_p);
 279         preempt_enable();
 280
 281         __replicas_table_update(dst, dst_r, src, src_r);
 282 }
 283
 284 /*
 285  * Resize filesystem accounting:
 286  */
 287 static int replicas_table_update(struct bch_fs *c,
 288                                  struct bch_replicas_cpu *new_r)
 289 {
 290         struct bch_fs_usage __percpu *new_usage[JOURNAL_BUF_NR];
 291         struct bch_fs_usage_online *new_scratch = NULL;
 292         struct bch_fs_usage __percpu *new_gc = NULL;
 293         struct bch_fs_usage *new_base = NULL;
 294         unsigned i, bytes = sizeof(struct bch_fs_usage) +
 295                 sizeof(u64) * new_r->nr;
 296         unsigned scratch_bytes = sizeof(struct bch_fs_usage_online) +
 297                 sizeof(u64) * new_r->nr;
 298         int ret = 0;
 299
 300         memset(new_usage, 0, sizeof(new_usage));
 301
 302         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 303                 if (!(new_usage[i] = __alloc_percpu_gfp(bytes,
 304                                         sizeof(u64), GFP_KERNEL)))
 305                         goto err;
 306
 307         memset(new_usage, 0, sizeof(new_usage));
 308
 309         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 310                 if (!(new_usage[i] = __alloc_percpu_gfp(bytes,
 311                                         sizeof(u64), GFP_KERNEL)))
 312                         goto err;
 313
 314         if (!(new_base = kzalloc(bytes, GFP_KERNEL)) ||
 315             !(new_scratch  = kmalloc(scratch_bytes, GFP_KERNEL)) ||
 316             (c->usage_gc &&
 317              !(new_gc = __alloc_percpu_gfp(bytes, sizeof(u64), GFP_KERNEL))))
 318                 goto err;
 319
 320         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 321                 if (c->usage[i])
 322                         __replicas_table_update_pcpu(new_usage[i], new_r,
 323                                                      c->usage[i], &c->replicas);
 324         if (c->usage_base)
 325                 __replicas_table_update(new_base,               new_r,
 326                                         c->usage_base,          &c->replicas);
 327         if (c->usage_gc)
 328                 __replicas_table_update_pcpu(new_gc,            new_r,
 329                                              c->usage_gc,       &c->replicas);
 330
 331         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 332                 swap(c->usage[i],       new_usage[i]);
 333         swap(c->usage_base,     new_base);
 334         swap(c->usage_scratch,  new_scratch);
 335         swap(c->usage_gc,       new_gc);
 336         swap(c->replicas,       *new_r);
 337 out:
 338         free_percpu(new_gc);
 339         kfree(new_scratch);
 340         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 341                 free_percpu(new_usage[i]);
 342         kfree(new_base);
 343         return ret;
 344 err:
 345         bch_err(c, "error updating replicas table: memory allocation failure");
 346         ret = -ENOMEM;
 347         goto out;
 348 }
 349
 350 static unsigned reserve_journal_replicas(struct bch_fs *c,
 351                                      struct bch_replicas_cpu *r)
 352 {
 353         struct bch_replicas_entry *e;
 354         unsigned journal_res_u64s = 0;
 355
 356         /* nr_inodes: */
 357         journal_res_u64s +=
 358                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 359
 360         /* key_version: */
 361         journal_res_u64s +=
 362                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 363
 364         /* persistent_reserved: */
 365         journal_res_u64s +=
 366                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64)) *
 367                 BCH_REPLICAS_MAX;
 368
 369         for_each_cpu_replicas_entry(r, e)
 370                 journal_res_u64s +=
 371                         DIV_ROUND_UP(sizeof(struct jset_entry_data_usage) +
 372                                      e->nr_devs, sizeof(u64));
 373         return journal_res_u64s;
 374 }
 375
 376 noinline
 377 static int bch2_mark_replicas_slowpath(struct bch_fs *c,
 378                                 struct bch_replicas_entry *new_entry)
 379 {
 380         struct bch_replicas_cpu new_r, new_gc;
 381         int ret = 0;
 382
 383         verify_replicas_entry(new_entry);
 384
 385         memset(&new_r, 0, sizeof(new_r));
 386         memset(&new_gc, 0, sizeof(new_gc));
 387
 388         mutex_lock(&c->sb_lock);
 389
 390         if (c->replicas_gc.entries &&
 391             !__replicas_has_entry(&c->replicas_gc, new_entry)) {
 392                 new_gc = cpu_replicas_add_entry(&c->replicas_gc, new_entry);
 393                 if (!new_gc.entries)
 394                         goto err;
 395         }
 396
 397         if (!__replicas_has_entry(&c->replicas, new_entry)) {
 398                 new_r = cpu_replicas_add_entry(&c->replicas, new_entry);
 399                 if (!new_r.entries)
 400                         goto err;
 401
 402                 ret = bch2_cpu_replicas_to_sb_replicas(c, &new_r);
 403                 if (ret)
 404                         goto err;
 405
 406                 bch2_journal_entry_res_resize(&c->journal,
 407                                 &c->replicas_journal_res,
 408                                 reserve_journal_replicas(c, &new_r));
 409         }
 410
 411         if (!new_r.entries &&
 412             !new_gc.entries)
 413                 goto out;
 414
 415         /* allocations done, now commit: */
 416
 417         if (new_r.entries)
 418                 bch2_write_super(c);
 419
 420         /* don't update in memory replicas until changes are persistent */
 421         percpu_down_write(&c->mark_lock);
 422         if (new_r.entries)
 423                 ret = replicas_table_update(c, &new_r);
 424         if (new_gc.entries)
 425                 swap(new_gc, c->replicas_gc);
 426         percpu_up_write(&c->mark_lock);
 427 out:
 428         mutex_unlock(&c->sb_lock);
 429
 430         kfree(new_r.entries);
 431         kfree(new_gc.entries);
 432
 433         return ret;
 434 err:
 435         bch_err(c, "error adding replicas entry: memory allocation failure");
 436         ret = -ENOMEM;
 437         goto out;
 438 }
 439
 440 int bch2_mark_replicas(struct bch_fs *c, struct bch_replicas_entry *r)
 441 {
 442         return likely(bch2_replicas_marked(c, r))
 443                 ? 0 : bch2_mark_replicas_slowpath(c, r);
 444 }
 445
 446 /* replicas delta list: */
 447
 448 int bch2_replicas_delta_list_mark(struct bch_fs *c,
 449                                   struct replicas_delta_list *r)
 450 {
 451         struct replicas_delta *d = r->d;
 452         struct replicas_delta *top = (void *) r->d + r->used;
 453         int ret = 0;
 454
 455         for (d = r->d; !ret && d != top; d = replicas_delta_next(d))
 456                 ret = bch2_mark_replicas(c, &d->r);
 457         return ret;
 458 }
 459
 460 /*
 461  * Old replicas_gc mechanism: only used for journal replicas entries now, should
 462  * die at some point:
 463  */
 464
 465 int bch2_replicas_gc_end(struct bch_fs *c, int ret)
 466 {
 467         unsigned i;
 468
 469         lockdep_assert_held(&c->replicas_gc_lock);
 470
 471         mutex_lock(&c->sb_lock);
 472         percpu_down_write(&c->mark_lock);
 473
 474         /*
 475          * this is kind of crappy; the replicas gc mechanism needs to be ripped
 476          * out
 477          */
 478
 479         for (i = 0; i < c->replicas.nr; i++) {
 480                 struct bch_replicas_entry *e =
 481                         cpu_replicas_entry(&c->replicas, i);
 482                 struct bch_replicas_cpu n;
 483
 484                 if (!__replicas_has_entry(&c->replicas_gc, e) &&
 485                     bch2_fs_usage_read_one(c, &c->usage_base->replicas[i])) {
 486                         n = cpu_replicas_add_entry(&c->replicas_gc, e);
 487                         if (!n.entries) {
 488                                 ret = -ENOMEM;
 489                                 goto err;
 490                         }
 491
 492                         swap(n, c->replicas_gc);
 493                         kfree(n.entries);
 494                 }
 495         }
 496
 497         ret = bch2_cpu_replicas_to_sb_replicas(c, &c->replicas_gc);
 498         if (ret)
 499                 goto err;
 500
 501         ret = replicas_table_update(c, &c->replicas_gc);
 502 err:
 503         kfree(c->replicas_gc.entries);
 504         c->replicas_gc.entries = NULL;
 505
 506         percpu_up_write(&c->mark_lock);
 507
 508         if (!ret)
 509                 bch2_write_super(c);
 510
 511         mutex_unlock(&c->sb_lock);
 512
 513         return ret;
 514 }
 515
 516 int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask)
 517 {
 518         struct bch_replicas_entry *e;
 519         unsigned i = 0;
 520
 521         lockdep_assert_held(&c->replicas_gc_lock);
 522
 523         mutex_lock(&c->sb_lock);
 524         BUG_ON(c->replicas_gc.entries);
 525
 526         c->replicas_gc.nr               = 0;
 527         c->replicas_gc.entry_size       = 0;
 528
 529         for_each_cpu_replicas_entry(&c->replicas, e)
 530                 if (!((1 << e->data_type) & typemask)) {
 531                         c->replicas_gc.nr++;
 532                         c->replicas_gc.entry_size =
 533                                 max_t(unsigned, c->replicas_gc.entry_size,
 534                                       replicas_entry_bytes(e));
 535                 }
 536
 537         c->replicas_gc.entries = kcalloc(c->replicas_gc.nr,
 538                                          c->replicas_gc.entry_size,
 539                                          GFP_KERNEL);
 540         if (!c->replicas_gc.entries) {
 541                 mutex_unlock(&c->sb_lock);
 542                 bch_err(c, "error allocating c->replicas_gc");
 543                 return -ENOMEM;
 544         }
 545
 546         for_each_cpu_replicas_entry(&c->replicas, e)
 547                 if (!((1 << e->data_type) & typemask))
 548                         memcpy(cpu_replicas_entry(&c->replicas_gc, i++),
 549                                e, c->replicas_gc.entry_size);
 550
 551         bch2_cpu_replicas_sort(&c->replicas_gc);
 552         mutex_unlock(&c->sb_lock);
 553
 554         return 0;
 555 }
 556
 557 /* New much simpler mechanism for clearing out unneeded replicas entries: */
 558
 559 int bch2_replicas_gc2(struct bch_fs *c)
 560 {
 561         struct bch_replicas_cpu new = { 0 };
 562         unsigned i, nr;
 563         int ret = 0;
 564
 565         bch2_journal_meta(&c->journal);
 566 retry:
 567         nr              = READ_ONCE(c->replicas.nr);
 568         new.entry_size  = READ_ONCE(c->replicas.entry_size);
 569         new.entries     = kcalloc(nr, new.entry_size, GFP_KERNEL);
 570         if (!new.entries) {
 571                 bch_err(c, "error allocating c->replicas_gc");
 572                 return -ENOMEM;
 573         }
 574
 575         mutex_lock(&c->sb_lock);
 576         percpu_down_write(&c->mark_lock);
 577
 578         if (nr                  != c->replicas.nr ||
 579             new.entry_size      != c->replicas.entry_size) {
 580                 percpu_up_write(&c->mark_lock);
 581                 mutex_unlock(&c->sb_lock);
 582                 kfree(new.entries);
 583                 goto retry;
 584         }
 585
 586         for (i = 0; i < c->replicas.nr; i++) {
 587                 struct bch_replicas_entry *e =
 588                         cpu_replicas_entry(&c->replicas, i);
 589
 590                 if (e->data_type == BCH_DATA_journal ||
 591                     c->usage_base->replicas[i] ||
 592                     percpu_u64_get(&c->usage[0]->replicas[i]) ||
 593                     percpu_u64_get(&c->usage[1]->replicas[i]) ||
 594                     percpu_u64_get(&c->usage[2]->replicas[i]) ||
 595                     percpu_u64_get(&c->usage[3]->replicas[i]))
 596                         memcpy(cpu_replicas_entry(&new, new.nr++),
 597                                e, new.entry_size);
 598         }
 599
 600         bch2_cpu_replicas_sort(&new);
 601
 602         ret = bch2_cpu_replicas_to_sb_replicas(c, &new);
 603         if (ret)
 604                 goto err;
 605
 606         ret = replicas_table_update(c, &new);
 607 err:
 608         kfree(new.entries);
 609
 610         percpu_up_write(&c->mark_lock);
 611
 612         if (!ret)
 613                 bch2_write_super(c);
 614
 615         mutex_unlock(&c->sb_lock);
 616
 617         return ret;
 618 }
 619
 620 int bch2_replicas_set_usage(struct bch_fs *c,
 621                             struct bch_replicas_entry *r,
 622                             u64 sectors)
 623 {
 624         int ret, idx = bch2_replicas_entry_idx(c, r);
 625
 626         if (idx < 0) {
 627                 struct bch_replicas_cpu n;
 628
 629                 n = cpu_replicas_add_entry(&c->replicas, r);
 630                 if (!n.entries)
 631                         return -ENOMEM;
 632
 633                 ret = replicas_table_update(c, &n);
 634                 if (ret)
 635                         return ret;
 636
 637                 kfree(n.entries);
 638
 639                 idx = bch2_replicas_entry_idx(c, r);
 640                 BUG_ON(ret < 0);
 641         }
 642
 643         c->usage_base->replicas[idx] = sectors;
 644
 645         return 0;
 646 }
 647
 648 /* Replicas tracking - superblock: */
 649
 650 static int
 651 __bch2_sb_replicas_to_cpu_replicas(struct bch_sb_field_replicas *sb_r,
 652                                    struct bch_replicas_cpu *cpu_r)
 653 {
 654         struct bch_replicas_entry *e, *dst;
 655         unsigned nr = 0, entry_size = 0, idx = 0;
 656
 657         for_each_replicas_entry(sb_r, e) {
 658                 entry_size = max_t(unsigned, entry_size,
 659                                    replicas_entry_bytes(e));
 660                 nr++;
 661         }
 662
 663         cpu_r->entries = kcalloc(nr, entry_size, GFP_KERNEL);
 664         if (!cpu_r->entries)
 665                 return -ENOMEM;
 666
 667         cpu_r->nr               = nr;
 668         cpu_r->entry_size       = entry_size;
 669
 670         for_each_replicas_entry(sb_r, e) {
 671                 dst = cpu_replicas_entry(cpu_r, idx++);
 672                 memcpy(dst, e, replicas_entry_bytes(e));
 673                 bch2_replicas_entry_sort(dst);
 674         }
 675
 676         return 0;
 677 }
 678
 679 static int
 680 __bch2_sb_replicas_v0_to_cpu_replicas(struct bch_sb_field_replicas_v0 *sb_r,
 681                                       struct bch_replicas_cpu *cpu_r)
 682 {
 683         struct bch_replicas_entry_v0 *e;
 684         unsigned nr = 0, entry_size = 0, idx = 0;
 685
 686         for_each_replicas_entry(sb_r, e) {
 687                 entry_size = max_t(unsigned, entry_size,
 688                                    replicas_entry_bytes(e));
 689                 nr++;
 690         }
 691
 692         entry_size += sizeof(struct bch_replicas_entry) -
 693                 sizeof(struct bch_replicas_entry_v0);
 694
 695         cpu_r->entries = kcalloc(nr, entry_size, GFP_KERNEL);
 696         if (!cpu_r->entries)
 697                 return -ENOMEM;
 698
 699         cpu_r->nr               = nr;
 700         cpu_r->entry_size       = entry_size;
 701
 702         for_each_replicas_entry(sb_r, e) {
 703                 struct bch_replicas_entry *dst =
 704                         cpu_replicas_entry(cpu_r, idx++);
 705
 706                 dst->data_type  = e->data_type;
 707                 dst->nr_devs    = e->nr_devs;
 708                 dst->nr_required = 1;
 709                 memcpy(dst->devs, e->devs, e->nr_devs);
 710                 bch2_replicas_entry_sort(dst);
 711         }
 712
 713         return 0;
 714 }
 715
 716 int bch2_sb_replicas_to_cpu_replicas(struct bch_fs *c)
 717 {
 718         struct bch_sb_field_replicas *sb_v1;
 719         struct bch_sb_field_replicas_v0 *sb_v0;
 720         struct bch_replicas_cpu new_r = { 0, 0, NULL };
 721         int ret = 0;
 722
 723         if ((sb_v1 = bch2_sb_get_replicas(c->disk_sb.sb)))
 724                 ret = __bch2_sb_replicas_to_cpu_replicas(sb_v1, &new_r);
 725         else if ((sb_v0 = bch2_sb_get_replicas_v0(c->disk_sb.sb)))
 726                 ret = __bch2_sb_replicas_v0_to_cpu_replicas(sb_v0, &new_r);
 727
 728         if (ret)
 729                 return -ENOMEM;
 730
 731         bch2_cpu_replicas_sort(&new_r);
 732
 733         percpu_down_write(&c->mark_lock);
 734
 735         ret = replicas_table_update(c, &new_r);
 736         percpu_up_write(&c->mark_lock);
 737
 738         kfree(new_r.entries);
 739
 740         return 0;
 741 }
 742
 743 static int bch2_cpu_replicas_to_sb_replicas_v0(struct bch_fs *c,
 744                                                struct bch_replicas_cpu *r)
 745 {
 746         struct bch_sb_field_replicas_v0 *sb_r;
 747         struct bch_replicas_entry_v0 *dst;
 748         struct bch_replicas_entry *src;
 749         size_t bytes;
 750
 751         bytes = sizeof(struct bch_sb_field_replicas);
 752
 753         for_each_cpu_replicas_entry(r, src)
 754                 bytes += replicas_entry_bytes(src) - 1;
 755
 756         sb_r = bch2_sb_resize_replicas_v0(&c->disk_sb,
 757                         DIV_ROUND_UP(bytes, sizeof(u64)));
 758         if (!sb_r)
 759                 return -BCH_ERR_ENOSPC_sb_replicas;
 760
 761         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas);
 762         sb_r = bch2_sb_get_replicas_v0(c->disk_sb.sb);
 763
 764         memset(&sb_r->entries, 0,
 765                vstruct_end(&sb_r->field) -
 766                (void *) &sb_r->entries);
 767
 768         dst = sb_r->entries;
 769         for_each_cpu_replicas_entry(r, src) {
 770                 dst->data_type  = src->data_type;
 771                 dst->nr_devs    = src->nr_devs;
 772                 memcpy(dst->devs, src->devs, src->nr_devs);
 773
 774                 dst = replicas_entry_next(dst);
 775
 776                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 777         }
 778
 779         return 0;
 780 }
 781
 782 static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *c,
 783                                             struct bch_replicas_cpu *r)
 784 {
 785         struct bch_sb_field_replicas *sb_r;
 786         struct bch_replicas_entry *dst, *src;
 787         bool need_v1 = false;
 788         size_t bytes;
 789
 790         bytes = sizeof(struct bch_sb_field_replicas);
 791
 792         for_each_cpu_replicas_entry(r, src) {
 793                 bytes += replicas_entry_bytes(src);
 794                 if (src->nr_required != 1)
 795                         need_v1 = true;
 796         }
 797
 798         if (!need_v1)
 799                 return bch2_cpu_replicas_to_sb_replicas_v0(c, r);
 800
 801         sb_r = bch2_sb_resize_replicas(&c->disk_sb,
 802                         DIV_ROUND_UP(bytes, sizeof(u64)));
 803         if (!sb_r)
 804                 return -BCH_ERR_ENOSPC_sb_replicas;
 805
 806         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas_v0);
 807         sb_r = bch2_sb_get_replicas(c->disk_sb.sb);
 808
 809         memset(&sb_r->entries, 0,
 810                vstruct_end(&sb_r->field) -
 811                (void *) &sb_r->entries);
 812
 813         dst = sb_r->entries;
 814         for_each_cpu_replicas_entry(r, src) {
 815                 memcpy(dst, src, replicas_entry_bytes(src));
 816
 817                 dst = replicas_entry_next(dst);
 818
 819                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 820         }
 821
 822         return 0;
 823 }
 824
 825 static int bch2_cpu_replicas_validate(struct bch_replicas_cpu *cpu_r,
 826                                       struct bch_sb *sb,
 827                                       struct printbuf *err)
 828 {
 829         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 830         unsigned i, j;
 831
 832         sort_cmp_size(cpu_r->entries,
 833                       cpu_r->nr,
 834                       cpu_r->entry_size,
 835                       memcmp, NULL);
 836
 837         for (i = 0; i < cpu_r->nr; i++) {
 838                 struct bch_replicas_entry *e =
 839                         cpu_replicas_entry(cpu_r, i);
 840
 841                 if (e->data_type >= BCH_DATA_NR) {
 842                         prt_printf(err, "invalid data type in entry ");
 843                         bch2_replicas_entry_to_text(err, e);
 844                         return -BCH_ERR_invalid_sb_replicas;
 845                 }
 846
 847                 if (!e->nr_devs) {
 848                         prt_printf(err, "no devices in entry ");
 849                         bch2_replicas_entry_to_text(err, e);
 850                         return -BCH_ERR_invalid_sb_replicas;
 851                 }
 852
 853                 if (e->nr_required > 1 &&
 854                     e->nr_required >= e->nr_devs) {
 855                         prt_printf(err, "bad nr_required in entry ");
 856                         bch2_replicas_entry_to_text(err, e);
 857                         return -BCH_ERR_invalid_sb_replicas;
 858                 }
 859
 860                 for (j = 0; j < e->nr_devs; j++)
 861                         if (!bch2_dev_exists(sb, mi, e->devs[j])) {
 862                                 prt_printf(err, "invalid device %u in entry ", e->devs[j]);
 863                                 bch2_replicas_entry_to_text(err, e);
 864                                 return -BCH_ERR_invalid_sb_replicas;
 865                         }
 866
 867                 if (i + 1 < cpu_r->nr) {
 868                         struct bch_replicas_entry *n =
 869                                 cpu_replicas_entry(cpu_r, i + 1);
 870
 871                         BUG_ON(memcmp(e, n, cpu_r->entry_size) > 0);
 872
 873                         if (!memcmp(e, n, cpu_r->entry_size)) {
 874                                 prt_printf(err, "duplicate replicas entry ");
 875                                 bch2_replicas_entry_to_text(err, e);
 876                                 return -BCH_ERR_invalid_sb_replicas;
 877                         }
 878                 }
 879         }
 880
 881         return 0;
 882 }
 883
 884 static int bch2_sb_replicas_validate(struct bch_sb *sb, struct bch_sb_field *f,
 885                                      struct printbuf *err)
 886 {
 887         struct bch_sb_field_replicas *sb_r = field_to_type(f, replicas);
 888         struct bch_replicas_cpu cpu_r;
 889         int ret;
 890
 891         if (__bch2_sb_replicas_to_cpu_replicas(sb_r, &cpu_r))
 892                 return -ENOMEM;
 893
 894         ret = bch2_cpu_replicas_validate(&cpu_r, sb, err);
 895         kfree(cpu_r.entries);
 896         return ret;
 897 }
 898
 899 static void bch2_sb_replicas_to_text(struct printbuf *out,
 900                                      struct bch_sb *sb,
 901                                      struct bch_sb_field *f)
 902 {
 903         struct bch_sb_field_replicas *r = field_to_type(f, replicas);
 904         struct bch_replicas_entry *e;
 905         bool first = true;
 906
 907         for_each_replicas_entry(r, e) {
 908                 if (!first)
 909                         prt_printf(out, " ");
 910                 first = false;
 911
 912                 bch2_replicas_entry_to_text(out, e);
 913         }
 914         prt_newline(out);
 915 }
 916
 917 const struct bch_sb_field_ops bch_sb_field_ops_replicas = {
 918         .validate       = bch2_sb_replicas_validate,
 919         .to_text        = bch2_sb_replicas_to_text,
 920 };
 921
 922 static int bch2_sb_replicas_v0_validate(struct bch_sb *sb, struct bch_sb_field *f,
 923                                         struct printbuf *err)
 924 {
 925         struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0);
 926         struct bch_replicas_cpu cpu_r;
 927         int ret;
 928
 929         if (__bch2_sb_replicas_v0_to_cpu_replicas(sb_r, &cpu_r))
 930                 return -ENOMEM;
 931
 932         ret = bch2_cpu_replicas_validate(&cpu_r, sb, err);
 933         kfree(cpu_r.entries);
 934         return ret;
 935 }
 936
 937 static void bch2_sb_replicas_v0_to_text(struct printbuf *out,
 938                                         struct bch_sb *sb,
 939                                         struct bch_sb_field *f)
 940 {
 941         struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0);
 942         struct bch_replicas_entry_v0 *e;
 943         bool first = true;
 944
 945         for_each_replicas_entry(sb_r, e) {
 946                 if (!first)
 947                         prt_printf(out, " ");
 948                 first = false;
 949
 950                 bch2_replicas_entry_v0_to_text(out, e);
 951         }
 952         prt_newline(out);
 953 }
 954
 955 const struct bch_sb_field_ops bch_sb_field_ops_replicas_v0 = {
 956         .validate       = bch2_sb_replicas_v0_validate,
 957         .to_text        = bch2_sb_replicas_v0_to_text,
 958 };
 959
 960 /* Query replicas: */
 961
 962 bool bch2_have_enough_devs(struct bch_fs *c, struct bch_devs_mask devs,
 963                            unsigned flags, bool print)
 964 {
 965         struct bch_replicas_entry *e;
 966         bool ret = true;
 967
 968         percpu_down_read(&c->mark_lock);
 969         for_each_cpu_replicas_entry(&c->replicas, e) {
 970                 unsigned i, nr_online = 0, nr_failed = 0, dflags = 0;
 971                 bool metadata = e->data_type < BCH_DATA_user;
 972
 973                 if (e->data_type == BCH_DATA_cached)
 974                         continue;
 975
 976                 for (i = 0; i < e->nr_devs; i++) {
 977                         struct bch_dev *ca = bch_dev_bkey_exists(c, e->devs[i]);
 978
 979                         nr_online += test_bit(e->devs[i], devs.d);
 980                         nr_failed += ca->mi.state == BCH_MEMBER_STATE_failed;
 981                 }
 982
 983                 if (nr_failed == e->nr_devs)
 984                         continue;
 985
 986                 if (nr_online < e->nr_required)
 987                         dflags |= metadata
 988                                 ? BCH_FORCE_IF_METADATA_LOST
 989                                 : BCH_FORCE_IF_DATA_LOST;
 990
 991                 if (nr_online < e->nr_devs)
 992                         dflags |= metadata
 993                                 ? BCH_FORCE_IF_METADATA_DEGRADED
 994                                 : BCH_FORCE_IF_DATA_DEGRADED;
 995
 996                 if (dflags & ~flags) {
 997                         if (print) {
 998                                 struct printbuf buf = PRINTBUF;
 999
1000                                 bch2_replicas_entry_to_text(&buf, e);
1001                                 bch_err(c, "insufficient devices online (%u) for replicas entry %s",
1002                                         nr_online, buf.buf);
1003                                 printbuf_exit(&buf);
1004                         }
1005                         ret = false;
1006                         break;
1007                 }
1008
1009         }
1010         percpu_up_read(&c->mark_lock);
1011
1012         return ret;
1013 }
1014
1015 unsigned bch2_sb_dev_has_data(struct bch_sb *sb, unsigned dev)
1016 {
1017         struct bch_sb_field_replicas *replicas;
1018         struct bch_sb_field_replicas_v0 *replicas_v0;
1019         unsigned i, data_has = 0;
1020
1021         replicas = bch2_sb_get_replicas(sb);
1022         replicas_v0 = bch2_sb_get_replicas_v0(sb);
1023
1024         if (replicas) {
1025                 struct bch_replicas_entry *r;
1026
1027                 for_each_replicas_entry(replicas, r)
1028                         for (i = 0; i < r->nr_devs; i++)
1029                                 if (r->devs[i] == dev)
1030                                         data_has |= 1 << r->data_type;
1031         } else if (replicas_v0) {
1032                 struct bch_replicas_entry_v0 *r;
1033
1034                 for_each_replicas_entry_v0(replicas_v0, r)
1035                         for (i = 0; i < r->nr_devs; i++)
1036                                 if (r->devs[i] == dev)
1037                                         data_has |= 1 << r->data_type;
1038         }
1039
1040
1041         return data_has;
1042 }
1043
1044 unsigned bch2_dev_has_data(struct bch_fs *c, struct bch_dev *ca)
1045 {
1046         unsigned ret;
1047
1048         mutex_lock(&c->sb_lock);
1049         ret = bch2_sb_dev_has_data(c->disk_sb.sb, ca->dev_idx);
1050         mutex_unlock(&c->sb_lock);
1051
1052         return ret;
1053 }
1054
1055 void bch2_fs_replicas_exit(struct bch_fs *c)
1056 {
1057         unsigned i;
1058
1059         kfree(c->usage_scratch);
1060         for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1061                 free_percpu(c->usage[i]);
1062         kfree(c->usage_base);
1063         kfree(c->replicas.entries);
1064         kfree(c->replicas_gc.entries);
1065
1066         mempool_exit(&c->replicas_delta_pool);
1067 }
1068
1069 int bch2_fs_replicas_init(struct bch_fs *c)
1070 {
1071         bch2_journal_entry_res_resize(&c->journal,
1072                         &c->replicas_journal_res,
1073                         reserve_journal_replicas(c, &c->replicas));
1074
1075         return mempool_init_kmalloc_pool(&c->replicas_delta_pool, 1,
1076                                          REPLICAS_DELTA_LIST_MAX) ?:
1077                 replicas_table_update(c, &c->replicas);
1078 }