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 bool bch2_replicas_marked(struct bch_fs *c,
 217                           struct bch_replicas_entry *search)
 218 {
 219         bool marked;
 220
 221         if (!search->nr_devs)
 222                 return true;
 223
 224         verify_replicas_entry(search);
 225
 226         percpu_down_read(&c->mark_lock);
 227         marked = __replicas_has_entry(&c->replicas, search) &&
 228                 (likely((!c->replicas_gc.entries)) ||
 229                  __replicas_has_entry(&c->replicas_gc, search));
 230         percpu_up_read(&c->mark_lock);
 231
 232         return marked;
 233 }
 234
 235 static void __replicas_table_update(struct bch_fs_usage *dst,
 236                                     struct bch_replicas_cpu *dst_r,
 237                                     struct bch_fs_usage *src,
 238                                     struct bch_replicas_cpu *src_r)
 239 {
 240         int src_idx, dst_idx;
 241
 242         *dst = *src;
 243
 244         for (src_idx = 0; src_idx < src_r->nr; src_idx++) {
 245                 if (!src->replicas[src_idx])
 246                         continue;
 247
 248                 dst_idx = __replicas_entry_idx(dst_r,
 249                                 cpu_replicas_entry(src_r, src_idx));
 250                 BUG_ON(dst_idx < 0);
 251
 252                 dst->replicas[dst_idx] = src->replicas[src_idx];
 253         }
 254 }
 255
 256 static void __replicas_table_update_pcpu(struct bch_fs_usage __percpu *dst_p,
 257                                     struct bch_replicas_cpu *dst_r,
 258                                     struct bch_fs_usage __percpu *src_p,
 259                                     struct bch_replicas_cpu *src_r)
 260 {
 261         unsigned src_nr = sizeof(struct bch_fs_usage) / sizeof(u64) + src_r->nr;
 262         struct bch_fs_usage *dst, *src = (void *)
 263                 bch2_acc_percpu_u64s((void *) src_p, src_nr);
 264
 265         preempt_disable();
 266         dst = this_cpu_ptr(dst_p);
 267         preempt_enable();
 268
 269         __replicas_table_update(dst, dst_r, src, src_r);
 270 }
 271
 272 /*
 273  * Resize filesystem accounting:
 274  */
 275 static int replicas_table_update(struct bch_fs *c,
 276                                  struct bch_replicas_cpu *new_r)
 277 {
 278         struct bch_fs_usage __percpu *new_usage[2] = { NULL, NULL };
 279         struct bch_fs_usage *new_scratch = NULL;
 280         struct bch_fs_usage __percpu *new_gc = NULL;
 281         struct bch_fs_usage *new_base = NULL;
 282         unsigned bytes = sizeof(struct bch_fs_usage) +
 283                 sizeof(u64) * new_r->nr;
 284         int ret = -ENOMEM;
 285
 286         if (!(new_base = kzalloc(bytes, GFP_NOIO)) ||
 287             !(new_usage[0] = __alloc_percpu_gfp(bytes, sizeof(u64),
 288                                                 GFP_NOIO)) ||
 289             !(new_usage[1] = __alloc_percpu_gfp(bytes, sizeof(u64),
 290                                                 GFP_NOIO)) ||
 291             !(new_scratch  = kmalloc(bytes, GFP_NOIO)) ||
 292             (c->usage_gc &&
 293              !(new_gc = __alloc_percpu_gfp(bytes, sizeof(u64), GFP_NOIO)))) {
 294                 bch_err(c, "error updating replicas table: memory allocation failure");
 295                 goto err;
 296         }
 297
 298         if (c->usage_base)
 299                 __replicas_table_update(new_base,               new_r,
 300                                         c->usage_base,          &c->replicas);
 301         if (c->usage[0])
 302                 __replicas_table_update_pcpu(new_usage[0],      new_r,
 303                                              c->usage[0],       &c->replicas);
 304         if (c->usage[1])
 305                 __replicas_table_update_pcpu(new_usage[1],      new_r,
 306                                              c->usage[1],       &c->replicas);
 307         if (c->usage_gc)
 308                 __replicas_table_update_pcpu(new_gc,            new_r,
 309                                              c->usage_gc,       &c->replicas);
 310
 311         swap(c->usage_base,     new_base);
 312         swap(c->usage[0],       new_usage[0]);
 313         swap(c->usage[1],       new_usage[1]);
 314         swap(c->usage_scratch,  new_scratch);
 315         swap(c->usage_gc,       new_gc);
 316         swap(c->replicas,       *new_r);
 317         ret = 0;
 318 err:
 319         free_percpu(new_gc);
 320         kfree(new_scratch);
 321         free_percpu(new_usage[1]);
 322         free_percpu(new_usage[0]);
 323         kfree(new_base);
 324         return ret;
 325 }
 326
 327 static unsigned reserve_journal_replicas(struct bch_fs *c,
 328                                      struct bch_replicas_cpu *r)
 329 {
 330         struct bch_replicas_entry *e;
 331         unsigned journal_res_u64s = 0;
 332
 333         /* nr_inodes: */
 334         journal_res_u64s +=
 335                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 336
 337         /* key_version: */
 338         journal_res_u64s +=
 339                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 340
 341         /* persistent_reserved: */
 342         journal_res_u64s +=
 343                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64)) *
 344                 BCH_REPLICAS_MAX;
 345
 346         for_each_cpu_replicas_entry(r, e)
 347                 journal_res_u64s +=
 348                         DIV_ROUND_UP(sizeof(struct jset_entry_data_usage) +
 349                                      e->nr_devs, sizeof(u64));
 350         return journal_res_u64s;
 351 }
 352
 353 noinline
 354 static int bch2_mark_replicas_slowpath(struct bch_fs *c,
 355                                 struct bch_replicas_entry *new_entry)
 356 {
 357         struct bch_replicas_cpu new_r, new_gc;
 358         int ret = 0;
 359
 360         verify_replicas_entry(new_entry);
 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         mutex_unlock(&c->sb_lock);
 406
 407         kfree(new_r.entries);
 408         kfree(new_gc.entries);
 409
 410         return ret;
 411 err:
 412         bch_err(c, "error adding replicas entry: memory allocation failure");
 413         ret = -ENOMEM;
 414         goto out;
 415 }
 416
 417 static int __bch2_mark_replicas(struct bch_fs *c,
 418                                 struct bch_replicas_entry *r,
 419                                 bool check)
 420 {
 421         return likely(bch2_replicas_marked(c, r))       ? 0
 422                 : check                                 ? -1
 423                 : bch2_mark_replicas_slowpath(c, r);
 424 }
 425
 426 int bch2_mark_replicas(struct bch_fs *c, struct bch_replicas_entry *r)
 427 {
 428         return __bch2_mark_replicas(c, r, false);
 429 }
 430
 431 static int __bch2_mark_bkey_replicas(struct bch_fs *c, struct bkey_s_c k,
 432                                      bool check)
 433 {
 434         struct bch_replicas_padded search;
 435         struct bch_devs_list cached = bch2_bkey_cached_devs(k);
 436         unsigned i;
 437         int ret;
 438
 439         for (i = 0; i < cached.nr; i++) {
 440                 bch2_replicas_entry_cached(&search.e, cached.devs[i]);
 441
 442                 ret = __bch2_mark_replicas(c, &search.e, check);
 443                 if (ret)
 444                         return ret;
 445         }
 446
 447         bch2_bkey_to_replicas(&search.e, k);
 448
 449         return __bch2_mark_replicas(c, &search.e, check);
 450 }
 451
 452 bool bch2_bkey_replicas_marked(struct bch_fs *c,
 453                                struct bkey_s_c k)
 454 {
 455         return __bch2_mark_bkey_replicas(c, k, true) == 0;
 456 }
 457
 458 int bch2_mark_bkey_replicas(struct bch_fs *c, struct bkey_s_c k)
 459 {
 460         return __bch2_mark_bkey_replicas(c, k, false);
 461 }
 462
 463 int bch2_replicas_gc_end(struct bch_fs *c, int ret)
 464 {
 465         unsigned i;
 466
 467         lockdep_assert_held(&c->replicas_gc_lock);
 468
 469         mutex_lock(&c->sb_lock);
 470         percpu_down_write(&c->mark_lock);
 471
 472         /*
 473          * this is kind of crappy; the replicas gc mechanism needs to be ripped
 474          * out
 475          */
 476
 477         for (i = 0; i < c->replicas.nr; i++) {
 478                 struct bch_replicas_entry *e =
 479                         cpu_replicas_entry(&c->replicas, i);
 480                 struct bch_replicas_cpu n;
 481
 482                 if (!__replicas_has_entry(&c->replicas_gc, e) &&
 483                     (c->usage_base->replicas[i] ||
 484                      percpu_u64_get(&c->usage[0]->replicas[i]) ||
 485                      percpu_u64_get(&c->usage[1]->replicas[i]))) {
 486                         n = cpu_replicas_add_entry(&c->replicas_gc, e);
 487                         if (!n.entries) {
 488                                 ret = -ENOSPC;
 489                                 goto err;
 490                         }
 491
 492                         swap(n, c->replicas_gc);
 493                         kfree(n.entries);
 494                 }
 495         }
 496
 497         if (bch2_cpu_replicas_to_sb_replicas(c, &c->replicas_gc)) {
 498                 ret = -ENOSPC;
 499                 goto err;
 500         }
 501
 502         ret = replicas_table_update(c, &c->replicas_gc);
 503 err:
 504         kfree(c->replicas_gc.entries);
 505         c->replicas_gc.entries = NULL;
 506
 507         percpu_up_write(&c->mark_lock);
 508
 509         if (!ret)
 510                 bch2_write_super(c);
 511
 512         mutex_unlock(&c->sb_lock);
 513
 514         return ret;
 515 }
 516
 517 int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask)
 518 {
 519         struct bch_replicas_entry *e;
 520         unsigned i = 0;
 521
 522         lockdep_assert_held(&c->replicas_gc_lock);
 523
 524         mutex_lock(&c->sb_lock);
 525         BUG_ON(c->replicas_gc.entries);
 526
 527         c->replicas_gc.nr               = 0;
 528         c->replicas_gc.entry_size       = 0;
 529
 530         for_each_cpu_replicas_entry(&c->replicas, e)
 531                 if (!((1 << e->data_type) & typemask)) {
 532                         c->replicas_gc.nr++;
 533                         c->replicas_gc.entry_size =
 534                                 max_t(unsigned, c->replicas_gc.entry_size,
 535                                       replicas_entry_bytes(e));
 536                 }
 537
 538         c->replicas_gc.entries = kcalloc(c->replicas_gc.nr,
 539                                          c->replicas_gc.entry_size,
 540                                          GFP_NOIO);
 541         if (!c->replicas_gc.entries) {
 542                 mutex_unlock(&c->sb_lock);
 543                 bch_err(c, "error allocating c->replicas_gc");
 544                 return -ENOMEM;
 545         }
 546
 547         for_each_cpu_replicas_entry(&c->replicas, e)
 548                 if (!((1 << e->data_type) & typemask))
 549                         memcpy(cpu_replicas_entry(&c->replicas_gc, i++),
 550                                e, c->replicas_gc.entry_size);
 551
 552         bch2_cpu_replicas_sort(&c->replicas_gc);
 553         mutex_unlock(&c->sb_lock);
 554
 555         return 0;
 556 }
 557
 558 int bch2_replicas_gc2(struct bch_fs *c)
 559 {
 560         struct bch_replicas_cpu new = { 0 };
 561         unsigned i, nr;
 562         int ret = 0;
 563
 564         bch2_journal_meta(&c->journal);
 565 retry:
 566         nr              = READ_ONCE(c->replicas.nr);
 567         new.entry_size  = READ_ONCE(c->replicas.entry_size);
 568         new.entries     = kcalloc(nr, new.entry_size, GFP_KERNEL);
 569         if (!new.entries) {
 570                 bch_err(c, "error allocating c->replicas_gc");
 571                 return -ENOMEM;
 572         }
 573
 574         mutex_lock(&c->sb_lock);
 575         percpu_down_write(&c->mark_lock);
 576
 577         if (nr                  != c->replicas.nr ||
 578             new.entry_size      != c->replicas.entry_size) {
 579                 percpu_up_write(&c->mark_lock);
 580                 mutex_unlock(&c->sb_lock);
 581                 kfree(new.entries);
 582                 goto retry;
 583         }
 584
 585         for (i = 0; i < c->replicas.nr; i++) {
 586                 struct bch_replicas_entry *e =
 587                         cpu_replicas_entry(&c->replicas, i);
 588
 589                 if (e->data_type == BCH_DATA_journal ||
 590                     c->usage_base->replicas[i] ||
 591                     percpu_u64_get(&c->usage[0]->replicas[i]) ||
 592                     percpu_u64_get(&c->usage[1]->replicas[i]))
 593                         memcpy(cpu_replicas_entry(&new, new.nr++),
 594                                e, new.entry_size);
 595         }
 596
 597         bch2_cpu_replicas_sort(&new);
 598
 599         if (bch2_cpu_replicas_to_sb_replicas(c, &new)) {
 600                 ret = -ENOSPC;
 601                 goto err;
 602         }
 603
 604         ret = replicas_table_update(c, &new);
 605 err:
 606         kfree(new.entries);
 607
 608         percpu_up_write(&c->mark_lock);
 609
 610         if (!ret)
 611                 bch2_write_super(c);
 612
 613         mutex_unlock(&c->sb_lock);
 614
 615         return ret;
 616 }
 617
 618 int bch2_replicas_set_usage(struct bch_fs *c,
 619                             struct bch_replicas_entry *r,
 620                             u64 sectors)
 621 {
 622         int ret, idx = bch2_replicas_entry_idx(c, r);
 623
 624         if (idx < 0) {
 625                 struct bch_replicas_cpu n;
 626
 627                 n = cpu_replicas_add_entry(&c->replicas, r);
 628                 if (!n.entries)
 629                         return -ENOMEM;
 630
 631                 ret = replicas_table_update(c, &n);
 632                 if (ret)
 633                         return ret;
 634
 635                 kfree(n.entries);
 636
 637                 idx = bch2_replicas_entry_idx(c, r);
 638                 BUG_ON(ret < 0);
 639         }
 640
 641         c->usage_base->replicas[idx] = sectors;
 642
 643         return 0;
 644 }
 645
 646 /* Replicas tracking - superblock: */
 647
 648 static int
 649 __bch2_sb_replicas_to_cpu_replicas(struct bch_sb_field_replicas *sb_r,
 650                                    struct bch_replicas_cpu *cpu_r)
 651 {
 652         struct bch_replicas_entry *e, *dst;
 653         unsigned nr = 0, entry_size = 0, idx = 0;
 654
 655         for_each_replicas_entry(sb_r, e) {
 656                 entry_size = max_t(unsigned, entry_size,
 657                                    replicas_entry_bytes(e));
 658                 nr++;
 659         }
 660
 661         cpu_r->entries = kcalloc(nr, entry_size, GFP_NOIO);
 662         if (!cpu_r->entries)
 663                 return -ENOMEM;
 664
 665         cpu_r->nr               = nr;
 666         cpu_r->entry_size       = entry_size;
 667
 668         for_each_replicas_entry(sb_r, e) {
 669                 dst = cpu_replicas_entry(cpu_r, idx++);
 670                 memcpy(dst, e, replicas_entry_bytes(e));
 671                 replicas_entry_sort(dst);
 672         }
 673
 674         return 0;
 675 }
 676
 677 static int
 678 __bch2_sb_replicas_v0_to_cpu_replicas(struct bch_sb_field_replicas_v0 *sb_r,
 679                                       struct bch_replicas_cpu *cpu_r)
 680 {
 681         struct bch_replicas_entry_v0 *e;
 682         unsigned nr = 0, entry_size = 0, idx = 0;
 683
 684         for_each_replicas_entry(sb_r, e) {
 685                 entry_size = max_t(unsigned, entry_size,
 686                                    replicas_entry_bytes(e));
 687                 nr++;
 688         }
 689
 690         entry_size += sizeof(struct bch_replicas_entry) -
 691                 sizeof(struct bch_replicas_entry_v0);
 692
 693         cpu_r->entries = kcalloc(nr, entry_size, GFP_NOIO);
 694         if (!cpu_r->entries)
 695                 return -ENOMEM;
 696
 697         cpu_r->nr               = nr;
 698         cpu_r->entry_size       = entry_size;
 699
 700         for_each_replicas_entry(sb_r, e) {
 701                 struct bch_replicas_entry *dst =
 702                         cpu_replicas_entry(cpu_r, idx++);
 703
 704                 dst->data_type  = e->data_type;
 705                 dst->nr_devs    = e->nr_devs;
 706                 dst->nr_required = 1;
 707                 memcpy(dst->devs, e->devs, e->nr_devs);
 708                 replicas_entry_sort(dst);
 709         }
 710
 711         return 0;
 712 }
 713
 714 int bch2_sb_replicas_to_cpu_replicas(struct bch_fs *c)
 715 {
 716         struct bch_sb_field_replicas *sb_v1;
 717         struct bch_sb_field_replicas_v0 *sb_v0;
 718         struct bch_replicas_cpu new_r = { 0, 0, NULL };
 719         int ret = 0;
 720
 721         if ((sb_v1 = bch2_sb_get_replicas(c->disk_sb.sb)))
 722                 ret = __bch2_sb_replicas_to_cpu_replicas(sb_v1, &new_r);
 723         else if ((sb_v0 = bch2_sb_get_replicas_v0(c->disk_sb.sb)))
 724                 ret = __bch2_sb_replicas_v0_to_cpu_replicas(sb_v0, &new_r);
 725
 726         if (ret)
 727                 return -ENOMEM;
 728
 729         bch2_cpu_replicas_sort(&new_r);
 730
 731         percpu_down_write(&c->mark_lock);
 732
 733         ret = replicas_table_update(c, &new_r);
 734         percpu_up_write(&c->mark_lock);
 735
 736         kfree(new_r.entries);
 737
 738         return 0;
 739 }
 740
 741 static int bch2_cpu_replicas_to_sb_replicas_v0(struct bch_fs *c,
 742                                                struct bch_replicas_cpu *r)
 743 {
 744         struct bch_sb_field_replicas_v0 *sb_r;
 745         struct bch_replicas_entry_v0 *dst;
 746         struct bch_replicas_entry *src;
 747         size_t bytes;
 748
 749         bytes = sizeof(struct bch_sb_field_replicas);
 750
 751         for_each_cpu_replicas_entry(r, src)
 752                 bytes += replicas_entry_bytes(src) - 1;
 753
 754         sb_r = bch2_sb_resize_replicas_v0(&c->disk_sb,
 755                         DIV_ROUND_UP(bytes, sizeof(u64)));
 756         if (!sb_r)
 757                 return -ENOSPC;
 758
 759         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas);
 760         sb_r = bch2_sb_get_replicas_v0(c->disk_sb.sb);
 761
 762         memset(&sb_r->entries, 0,
 763                vstruct_end(&sb_r->field) -
 764                (void *) &sb_r->entries);
 765
 766         dst = sb_r->entries;
 767         for_each_cpu_replicas_entry(r, src) {
 768                 dst->data_type  = src->data_type;
 769                 dst->nr_devs    = src->nr_devs;
 770                 memcpy(dst->devs, src->devs, src->nr_devs);
 771
 772                 dst = replicas_entry_next(dst);
 773
 774                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 775         }
 776
 777         return 0;
 778 }
 779
 780 static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *c,
 781                                             struct bch_replicas_cpu *r)
 782 {
 783         struct bch_sb_field_replicas *sb_r;
 784         struct bch_replicas_entry *dst, *src;
 785         bool need_v1 = false;
 786         size_t bytes;
 787
 788         bytes = sizeof(struct bch_sb_field_replicas);
 789
 790         for_each_cpu_replicas_entry(r, src) {
 791                 bytes += replicas_entry_bytes(src);
 792                 if (src->nr_required != 1)
 793                         need_v1 = true;
 794         }
 795
 796         if (!need_v1)
 797                 return bch2_cpu_replicas_to_sb_replicas_v0(c, r);
 798
 799         sb_r = bch2_sb_resize_replicas(&c->disk_sb,
 800                         DIV_ROUND_UP(bytes, sizeof(u64)));
 801         if (!sb_r)
 802                 return -ENOSPC;
 803
 804         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas_v0);
 805         sb_r = bch2_sb_get_replicas(c->disk_sb.sb);
 806
 807         memset(&sb_r->entries, 0,
 808                vstruct_end(&sb_r->field) -
 809                (void *) &sb_r->entries);
 810
 811         dst = sb_r->entries;
 812         for_each_cpu_replicas_entry(r, src) {
 813                 memcpy(dst, src, replicas_entry_bytes(src));
 814
 815                 dst = replicas_entry_next(dst);
 816
 817                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 818         }
 819
 820         return 0;
 821 }
 822
 823 static const char *check_dup_replicas_entries(struct bch_replicas_cpu *cpu_r)
 824 {
 825         unsigned i;
 826
 827         sort_cmp_size(cpu_r->entries,
 828                       cpu_r->nr,
 829                       cpu_r->entry_size,
 830                       memcmp, NULL);
 831
 832         for (i = 0; i + 1 < cpu_r->nr; i++) {
 833                 struct bch_replicas_entry *l =
 834                         cpu_replicas_entry(cpu_r, i);
 835                 struct bch_replicas_entry *r =
 836                         cpu_replicas_entry(cpu_r, i + 1);
 837
 838                 BUG_ON(memcmp(l, r, cpu_r->entry_size) > 0);
 839
 840                 if (!memcmp(l, r, cpu_r->entry_size))
 841                         return "duplicate replicas entry";
 842         }
 843
 844         return NULL;
 845 }
 846
 847 static const char *bch2_sb_validate_replicas(struct bch_sb *sb, struct bch_sb_field *f)
 848 {
 849         struct bch_sb_field_replicas *sb_r = field_to_type(f, replicas);
 850         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 851         struct bch_replicas_cpu cpu_r = { .entries = NULL };
 852         struct bch_replicas_entry *e;
 853         const char *err;
 854         unsigned i;
 855
 856         for_each_replicas_entry(sb_r, e) {
 857                 err = "invalid replicas entry: invalid data type";
 858                 if (e->data_type >= BCH_DATA_NR)
 859                         goto err;
 860
 861                 err = "invalid replicas entry: no devices";
 862                 if (!e->nr_devs)
 863                         goto err;
 864
 865                 err = "invalid replicas entry: bad nr_required";
 866                 if (e->nr_required > 1 &&
 867                     e->nr_required >= e->nr_devs)
 868                         goto err;
 869
 870                 err = "invalid replicas entry: invalid device";
 871                 for (i = 0; i < e->nr_devs; i++)
 872                         if (!bch2_dev_exists(sb, mi, e->devs[i]))
 873                                 goto err;
 874         }
 875
 876         err = "cannot allocate memory";
 877         if (__bch2_sb_replicas_to_cpu_replicas(sb_r, &cpu_r))
 878                 goto err;
 879
 880         err = check_dup_replicas_entries(&cpu_r);
 881 err:
 882         kfree(cpu_r.entries);
 883         return err;
 884 }
 885
 886 static void bch2_sb_replicas_to_text(struct printbuf *out,
 887                                      struct bch_sb *sb,
 888                                      struct bch_sb_field *f)
 889 {
 890         struct bch_sb_field_replicas *r = field_to_type(f, replicas);
 891         struct bch_replicas_entry *e;
 892         bool first = true;
 893
 894         for_each_replicas_entry(r, e) {
 895                 if (!first)
 896                         pr_buf(out, " ");
 897                 first = false;
 898
 899                 bch2_replicas_entry_to_text(out, e);
 900         }
 901 }
 902
 903 const struct bch_sb_field_ops bch_sb_field_ops_replicas = {
 904         .validate       = bch2_sb_validate_replicas,
 905         .to_text        = bch2_sb_replicas_to_text,
 906 };
 907
 908 static const char *bch2_sb_validate_replicas_v0(struct bch_sb *sb, struct bch_sb_field *f)
 909 {
 910         struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0);
 911         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 912         struct bch_replicas_cpu cpu_r = { .entries = NULL };
 913         struct bch_replicas_entry_v0 *e;
 914         const char *err;
 915         unsigned i;
 916
 917         for_each_replicas_entry_v0(sb_r, e) {
 918                 err = "invalid replicas entry: invalid data type";
 919                 if (e->data_type >= BCH_DATA_NR)
 920                         goto err;
 921
 922                 err = "invalid replicas entry: no devices";
 923                 if (!e->nr_devs)
 924                         goto err;
 925
 926                 err = "invalid replicas entry: invalid device";
 927                 for (i = 0; i < e->nr_devs; i++)
 928                         if (!bch2_dev_exists(sb, mi, e->devs[i]))
 929                                 goto err;
 930         }
 931
 932         err = "cannot allocate memory";
 933         if (__bch2_sb_replicas_v0_to_cpu_replicas(sb_r, &cpu_r))
 934                 goto err;
 935
 936         err = check_dup_replicas_entries(&cpu_r);
 937 err:
 938         kfree(cpu_r.entries);
 939         return err;
 940 }
 941
 942 const struct bch_sb_field_ops bch_sb_field_ops_replicas_v0 = {
 943         .validate       = bch2_sb_validate_replicas_v0,
 944 };
 945
 946 /* Query replicas: */
 947
 948 struct replicas_status __bch2_replicas_status(struct bch_fs *c,
 949                                               struct bch_devs_mask online_devs)
 950 {
 951         struct bch_sb_field_members *mi;
 952         struct bch_replicas_entry *e;
 953         unsigned i, nr_online, nr_offline;
 954         struct replicas_status ret;
 955
 956         memset(&ret, 0, sizeof(ret));
 957
 958         for (i = 0; i < ARRAY_SIZE(ret.replicas); i++)
 959                 ret.replicas[i].redundancy = INT_MAX;
 960
 961         mi = bch2_sb_get_members(c->disk_sb.sb);
 962
 963         percpu_down_read(&c->mark_lock);
 964
 965         for_each_cpu_replicas_entry(&c->replicas, e) {
 966                 if (e->data_type >= ARRAY_SIZE(ret.replicas))
 967                         panic("e %p data_type %u\n", e, e->data_type);
 968
 969                 nr_online = nr_offline = 0;
 970
 971                 for (i = 0; i < e->nr_devs; i++) {
 972                         BUG_ON(!bch2_dev_exists(c->disk_sb.sb, mi,
 973                                                 e->devs[i]));
 974
 975                         if (test_bit(e->devs[i], online_devs.d))
 976                                 nr_online++;
 977                         else
 978                                 nr_offline++;
 979                 }
 980
 981                 ret.replicas[e->data_type].redundancy =
 982                         min(ret.replicas[e->data_type].redundancy,
 983                             (int) nr_online - (int) e->nr_required);
 984
 985                 ret.replicas[e->data_type].nr_offline =
 986                         max(ret.replicas[e->data_type].nr_offline,
 987                             nr_offline);
 988         }
 989
 990         percpu_up_read(&c->mark_lock);
 991
 992         for (i = 0; i < ARRAY_SIZE(ret.replicas); i++)
 993                 if (ret.replicas[i].redundancy == INT_MAX)
 994                         ret.replicas[i].redundancy = 0;
 995
 996         return ret;
 997 }
 998
 999 struct replicas_status bch2_replicas_status(struct bch_fs *c)
1000 {
1001         return __bch2_replicas_status(c, bch2_online_devs(c));
1002 }
1003
1004 static bool have_enough_devs(struct replicas_status s,
1005                              enum bch_data_type type,
1006                              bool force_if_degraded,
1007                              bool force_if_lost)
1008 {
1009         return (!s.replicas[type].nr_offline || force_if_degraded) &&
1010                 (s.replicas[type].redundancy >= 0 || force_if_lost);
1011 }
1012
1013 bool bch2_have_enough_devs(struct replicas_status s, unsigned flags)
1014 {
1015         return (have_enough_devs(s, BCH_DATA_journal,
1016                                  flags & BCH_FORCE_IF_METADATA_DEGRADED,
1017                                  flags & BCH_FORCE_IF_METADATA_LOST) &&
1018                 have_enough_devs(s, BCH_DATA_btree,
1019                                  flags & BCH_FORCE_IF_METADATA_DEGRADED,
1020                                  flags & BCH_FORCE_IF_METADATA_LOST) &&
1021                 have_enough_devs(s, BCH_DATA_user,
1022                                  flags & BCH_FORCE_IF_DATA_DEGRADED,
1023                                  flags & BCH_FORCE_IF_DATA_LOST));
1024 }
1025
1026 int bch2_replicas_online(struct bch_fs *c, bool meta)
1027 {
1028         struct replicas_status s = bch2_replicas_status(c);
1029
1030         return (meta
1031                 ? min(s.replicas[BCH_DATA_journal].redundancy,
1032                       s.replicas[BCH_DATA_btree].redundancy)
1033                 : s.replicas[BCH_DATA_user].redundancy) + 1;
1034 }
1035
1036 unsigned bch2_dev_has_data(struct bch_fs *c, struct bch_dev *ca)
1037 {
1038         struct bch_replicas_entry *e;
1039         unsigned i, ret = 0;
1040
1041         percpu_down_read(&c->mark_lock);
1042
1043         for_each_cpu_replicas_entry(&c->replicas, e)
1044                 for (i = 0; i < e->nr_devs; i++)
1045                         if (e->devs[i] == ca->dev_idx)
1046                                 ret |= 1 << e->data_type;
1047
1048         percpu_up_read(&c->mark_lock);
1049
1050         return ret;
1051 }
1052
1053 int bch2_fs_replicas_init(struct bch_fs *c)
1054 {
1055         c->journal.entry_u64s_reserved +=
1056                 reserve_journal_replicas(c, &c->replicas);
1057
1058         return replicas_table_update(c, &c->replicas);
1059 }