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 #include <linux/sort.h>
  10
  11 static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *,
  12                                             struct bch_replicas_cpu *);
  13
  14 /* Some (buggy!) compilers don't allow memcmp to be passed as a pointer */
  15 static int bch2_memcmp(const void *l, const void *r,  const void *priv)
  16 {
  17         size_t size = (size_t) priv;
  18         return memcmp(l, r, size);
  19 }
  20
  21 /* Replicas tracking - in memory: */
  22
  23 static void verify_replicas_entry(struct bch_replicas_entry_v1 *e)
  24 {
  25 #ifdef CONFIG_BCACHEFS_DEBUG
  26         unsigned i;
  27
  28         BUG_ON(e->data_type >= BCH_DATA_NR);
  29         BUG_ON(!e->nr_devs);
  30         BUG_ON(e->nr_required > 1 &&
  31                e->nr_required >= e->nr_devs);
  32
  33         for (i = 0; i + 1 < e->nr_devs; i++)
  34                 BUG_ON(e->devs[i] >= e->devs[i + 1]);
  35 #endif
  36 }
  37
  38 void bch2_replicas_entry_sort(struct bch_replicas_entry_v1 *e)
  39 {
  40         bubble_sort(e->devs, e->nr_devs, u8_cmp);
  41 }
  42
  43 static void bch2_cpu_replicas_sort(struct bch_replicas_cpu *r)
  44 {
  45         eytzinger0_sort_r(r->entries, r->nr, r->entry_size,
  46                           bch2_memcmp, NULL, (void *)(size_t)r->entry_size);
  47 }
  48
  49 static void bch2_replicas_entry_v0_to_text(struct printbuf *out,
  50                                            struct bch_replicas_entry_v0 *e)
  51 {
  52         bch2_prt_data_type(out, e->data_type);
  53
  54         prt_printf(out, ": %u [", e->nr_devs);
  55         for (unsigned i = 0; i < e->nr_devs; i++)
  56                 prt_printf(out, i ? " %u" : "%u", e->devs[i]);
  57         prt_printf(out, "]");
  58 }
  59
  60 void bch2_replicas_entry_to_text(struct printbuf *out,
  61                                  struct bch_replicas_entry_v1 *e)
  62 {
  63         bch2_prt_data_type(out, e->data_type);
  64
  65         prt_printf(out, ": %u/%u [", e->nr_required, e->nr_devs);
  66         for (unsigned i = 0; i < e->nr_devs; i++)
  67                 prt_printf(out, i ? " %u" : "%u", e->devs[i]);
  68         prt_printf(out, "]");
  69 }
  70
  71 int bch2_replicas_entry_validate(struct bch_replicas_entry_v1 *r,
  72                                  struct bch_sb *sb,
  73                                  struct printbuf *err)
  74 {
  75         if (!r->nr_devs) {
  76                 prt_printf(err, "no devices in entry ");
  77                 goto bad;
  78         }
  79
  80         if (r->nr_required > 1 &&
  81             r->nr_required >= r->nr_devs) {
  82                 prt_printf(err, "bad nr_required in entry ");
  83                 goto bad;
  84         }
  85
  86         for (unsigned i = 0; i < r->nr_devs; i++)
  87                 if (!bch2_dev_exists(sb, r->devs[i])) {
  88                         prt_printf(err, "invalid device %u in entry ", r->devs[i]);
  89                         goto bad;
  90                 }
  91
  92         return 0;
  93 bad:
  94         bch2_replicas_entry_to_text(err, r);
  95         return -BCH_ERR_invalid_replicas_entry;
  96 }
  97
  98 void bch2_cpu_replicas_to_text(struct printbuf *out,
  99                                struct bch_replicas_cpu *r)
 100 {
 101         struct bch_replicas_entry_v1 *e;
 102         bool first = true;
 103
 104         for_each_cpu_replicas_entry(r, e) {
 105                 if (!first)
 106                         prt_printf(out, " ");
 107                 first = false;
 108
 109                 bch2_replicas_entry_to_text(out, e);
 110         }
 111 }
 112
 113 static void extent_to_replicas(struct bkey_s_c k,
 114                                struct bch_replicas_entry_v1 *r)
 115 {
 116         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 117         const union bch_extent_entry *entry;
 118         struct extent_ptr_decoded p;
 119
 120         r->nr_required  = 1;
 121
 122         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 123                 if (p.ptr.cached)
 124                         continue;
 125
 126                 if (!p.has_ec)
 127                         r->devs[r->nr_devs++] = p.ptr.dev;
 128                 else
 129                         r->nr_required = 0;
 130         }
 131 }
 132
 133 static void stripe_to_replicas(struct bkey_s_c k,
 134                                struct bch_replicas_entry_v1 *r)
 135 {
 136         struct bkey_s_c_stripe s = bkey_s_c_to_stripe(k);
 137         const struct bch_extent_ptr *ptr;
 138
 139         r->nr_required  = s.v->nr_blocks - s.v->nr_redundant;
 140
 141         for (ptr = s.v->ptrs;
 142              ptr < s.v->ptrs + s.v->nr_blocks;
 143              ptr++)
 144                 r->devs[r->nr_devs++] = ptr->dev;
 145 }
 146
 147 void bch2_bkey_to_replicas(struct bch_replicas_entry_v1 *e,
 148                            struct bkey_s_c k)
 149 {
 150         e->nr_devs = 0;
 151
 152         switch (k.k->type) {
 153         case KEY_TYPE_btree_ptr:
 154         case KEY_TYPE_btree_ptr_v2:
 155                 e->data_type = BCH_DATA_btree;
 156                 extent_to_replicas(k, e);
 157                 break;
 158         case KEY_TYPE_extent:
 159         case KEY_TYPE_reflink_v:
 160                 e->data_type = BCH_DATA_user;
 161                 extent_to_replicas(k, e);
 162                 break;
 163         case KEY_TYPE_stripe:
 164                 e->data_type = BCH_DATA_parity;
 165                 stripe_to_replicas(k, e);
 166                 break;
 167         }
 168
 169         bch2_replicas_entry_sort(e);
 170 }
 171
 172 void bch2_devlist_to_replicas(struct bch_replicas_entry_v1 *e,
 173                               enum bch_data_type data_type,
 174                               struct bch_devs_list devs)
 175 {
 176         BUG_ON(!data_type ||
 177                data_type == BCH_DATA_sb ||
 178                data_type >= BCH_DATA_NR);
 179
 180         e->data_type    = data_type;
 181         e->nr_devs      = 0;
 182         e->nr_required  = 1;
 183
 184         darray_for_each(devs, i)
 185                 e->devs[e->nr_devs++] = *i;
 186
 187         bch2_replicas_entry_sort(e);
 188 }
 189
 190 static struct bch_replicas_cpu
 191 cpu_replicas_add_entry(struct bch_fs *c,
 192                        struct bch_replicas_cpu *old,
 193                        struct bch_replicas_entry_v1 *new_entry)
 194 {
 195         unsigned i;
 196         struct bch_replicas_cpu new = {
 197                 .nr             = old->nr + 1,
 198                 .entry_size     = max_t(unsigned, old->entry_size,
 199                                         replicas_entry_bytes(new_entry)),
 200         };
 201
 202         for (i = 0; i < new_entry->nr_devs; i++)
 203                 BUG_ON(!bch2_dev_exists2(c, new_entry->devs[i]));
 204
 205         BUG_ON(!new_entry->data_type);
 206         verify_replicas_entry(new_entry);
 207
 208         new.entries = kcalloc(new.nr, new.entry_size, GFP_KERNEL);
 209         if (!new.entries)
 210                 return new;
 211
 212         for (i = 0; i < old->nr; i++)
 213                 memcpy(cpu_replicas_entry(&new, i),
 214                        cpu_replicas_entry(old, i),
 215                        old->entry_size);
 216
 217         memcpy(cpu_replicas_entry(&new, old->nr),
 218                new_entry,
 219                replicas_entry_bytes(new_entry));
 220
 221         bch2_cpu_replicas_sort(&new);
 222         return new;
 223 }
 224
 225 static inline int __replicas_entry_idx(struct bch_replicas_cpu *r,
 226                                        struct bch_replicas_entry_v1 *search)
 227 {
 228         int idx, entry_size = replicas_entry_bytes(search);
 229
 230         if (unlikely(entry_size > r->entry_size))
 231                 return -1;
 232
 233         verify_replicas_entry(search);
 234
 235 #define entry_cmp(_l, _r)       memcmp(_l, _r, entry_size)
 236         idx = eytzinger0_find(r->entries, r->nr, r->entry_size,
 237                               entry_cmp, search);
 238 #undef entry_cmp
 239
 240         return idx < r->nr ? idx : -1;
 241 }
 242
 243 int bch2_replicas_entry_idx(struct bch_fs *c,
 244                             struct bch_replicas_entry_v1 *search)
 245 {
 246         bch2_replicas_entry_sort(search);
 247
 248         return __replicas_entry_idx(&c->replicas, search);
 249 }
 250
 251 static bool __replicas_has_entry(struct bch_replicas_cpu *r,
 252                                  struct bch_replicas_entry_v1 *search)
 253 {
 254         return __replicas_entry_idx(r, search) >= 0;
 255 }
 256
 257 bool bch2_replicas_marked(struct bch_fs *c,
 258                           struct bch_replicas_entry_v1 *search)
 259 {
 260         bool marked;
 261
 262         if (!search->nr_devs)
 263                 return true;
 264
 265         verify_replicas_entry(search);
 266
 267         percpu_down_read(&c->mark_lock);
 268         marked = __replicas_has_entry(&c->replicas, search) &&
 269                 (likely((!c->replicas_gc.entries)) ||
 270                  __replicas_has_entry(&c->replicas_gc, search));
 271         percpu_up_read(&c->mark_lock);
 272
 273         return marked;
 274 }
 275
 276 static void __replicas_table_update(struct bch_fs_usage *dst,
 277                                     struct bch_replicas_cpu *dst_r,
 278                                     struct bch_fs_usage *src,
 279                                     struct bch_replicas_cpu *src_r)
 280 {
 281         int src_idx, dst_idx;
 282
 283         *dst = *src;
 284
 285         for (src_idx = 0; src_idx < src_r->nr; src_idx++) {
 286                 if (!src->replicas[src_idx])
 287                         continue;
 288
 289                 dst_idx = __replicas_entry_idx(dst_r,
 290                                 cpu_replicas_entry(src_r, src_idx));
 291                 BUG_ON(dst_idx < 0);
 292
 293                 dst->replicas[dst_idx] = src->replicas[src_idx];
 294         }
 295 }
 296
 297 static void __replicas_table_update_pcpu(struct bch_fs_usage __percpu *dst_p,
 298                                     struct bch_replicas_cpu *dst_r,
 299                                     struct bch_fs_usage __percpu *src_p,
 300                                     struct bch_replicas_cpu *src_r)
 301 {
 302         unsigned src_nr = sizeof(struct bch_fs_usage) / sizeof(u64) + src_r->nr;
 303         struct bch_fs_usage *dst, *src = (void *)
 304                 bch2_acc_percpu_u64s((u64 __percpu *) src_p, src_nr);
 305
 306         preempt_disable();
 307         dst = this_cpu_ptr(dst_p);
 308         preempt_enable();
 309
 310         __replicas_table_update(dst, dst_r, src, src_r);
 311 }
 312
 313 /*
 314  * Resize filesystem accounting:
 315  */
 316 static int replicas_table_update(struct bch_fs *c,
 317                                  struct bch_replicas_cpu *new_r)
 318 {
 319         struct bch_fs_usage __percpu *new_usage[JOURNAL_BUF_NR];
 320         struct bch_fs_usage_online *new_scratch = NULL;
 321         struct bch_fs_usage __percpu *new_gc = NULL;
 322         struct bch_fs_usage *new_base = NULL;
 323         unsigned i, bytes = sizeof(struct bch_fs_usage) +
 324                 sizeof(u64) * new_r->nr;
 325         unsigned scratch_bytes = sizeof(struct bch_fs_usage_online) +
 326                 sizeof(u64) * new_r->nr;
 327         int ret = 0;
 328
 329         memset(new_usage, 0, sizeof(new_usage));
 330
 331         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 332                 if (!(new_usage[i] = __alloc_percpu_gfp(bytes,
 333                                         sizeof(u64), GFP_KERNEL)))
 334                         goto err;
 335
 336         if (!(new_base = kzalloc(bytes, GFP_KERNEL)) ||
 337             !(new_scratch  = kmalloc(scratch_bytes, GFP_KERNEL)) ||
 338             (c->usage_gc &&
 339              !(new_gc = __alloc_percpu_gfp(bytes, sizeof(u64), GFP_KERNEL))))
 340                 goto err;
 341
 342         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 343                 if (c->usage[i])
 344                         __replicas_table_update_pcpu(new_usage[i], new_r,
 345                                                      c->usage[i], &c->replicas);
 346         if (c->usage_base)
 347                 __replicas_table_update(new_base,               new_r,
 348                                         c->usage_base,          &c->replicas);
 349         if (c->usage_gc)
 350                 __replicas_table_update_pcpu(new_gc,            new_r,
 351                                              c->usage_gc,       &c->replicas);
 352
 353         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 354                 swap(c->usage[i],       new_usage[i]);
 355         swap(c->usage_base,     new_base);
 356         swap(c->usage_scratch,  new_scratch);
 357         swap(c->usage_gc,       new_gc);
 358         swap(c->replicas,       *new_r);
 359 out:
 360         free_percpu(new_gc);
 361         kfree(new_scratch);
 362         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 363                 free_percpu(new_usage[i]);
 364         kfree(new_base);
 365         return ret;
 366 err:
 367         bch_err(c, "error updating replicas table: memory allocation failure");
 368         ret = -BCH_ERR_ENOMEM_replicas_table;
 369         goto out;
 370 }
 371
 372 static unsigned reserve_journal_replicas(struct bch_fs *c,
 373                                      struct bch_replicas_cpu *r)
 374 {
 375         struct bch_replicas_entry_v1 *e;
 376         unsigned journal_res_u64s = 0;
 377
 378         /* nr_inodes: */
 379         journal_res_u64s +=
 380                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 381
 382         /* key_version: */
 383         journal_res_u64s +=
 384                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 385
 386         /* persistent_reserved: */
 387         journal_res_u64s +=
 388                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64)) *
 389                 BCH_REPLICAS_MAX;
 390
 391         for_each_cpu_replicas_entry(r, e)
 392                 journal_res_u64s +=
 393                         DIV_ROUND_UP(sizeof(struct jset_entry_data_usage) +
 394                                      e->nr_devs, sizeof(u64));
 395         return journal_res_u64s;
 396 }
 397
 398 noinline
 399 static int bch2_mark_replicas_slowpath(struct bch_fs *c,
 400                                 struct bch_replicas_entry_v1 *new_entry)
 401 {
 402         struct bch_replicas_cpu new_r, new_gc;
 403         int ret = 0;
 404
 405         verify_replicas_entry(new_entry);
 406
 407         memset(&new_r, 0, sizeof(new_r));
 408         memset(&new_gc, 0, sizeof(new_gc));
 409
 410         mutex_lock(&c->sb_lock);
 411
 412         if (c->replicas_gc.entries &&
 413             !__replicas_has_entry(&c->replicas_gc, new_entry)) {
 414                 new_gc = cpu_replicas_add_entry(c, &c->replicas_gc, new_entry);
 415                 if (!new_gc.entries) {
 416                         ret = -BCH_ERR_ENOMEM_cpu_replicas;
 417                         goto err;
 418                 }
 419         }
 420
 421         if (!__replicas_has_entry(&c->replicas, new_entry)) {
 422                 new_r = cpu_replicas_add_entry(c, &c->replicas, new_entry);
 423                 if (!new_r.entries) {
 424                         ret = -BCH_ERR_ENOMEM_cpu_replicas;
 425                         goto err;
 426                 }
 427
 428                 ret = bch2_cpu_replicas_to_sb_replicas(c, &new_r);
 429                 if (ret)
 430                         goto err;
 431
 432                 bch2_journal_entry_res_resize(&c->journal,
 433                                 &c->replicas_journal_res,
 434                                 reserve_journal_replicas(c, &new_r));
 435         }
 436
 437         if (!new_r.entries &&
 438             !new_gc.entries)
 439                 goto out;
 440
 441         /* allocations done, now commit: */
 442
 443         if (new_r.entries)
 444                 bch2_write_super(c);
 445
 446         /* don't update in memory replicas until changes are persistent */
 447         percpu_down_write(&c->mark_lock);
 448         if (new_r.entries)
 449                 ret = replicas_table_update(c, &new_r);
 450         if (new_gc.entries)
 451                 swap(new_gc, c->replicas_gc);
 452         percpu_up_write(&c->mark_lock);
 453 out:
 454         mutex_unlock(&c->sb_lock);
 455
 456         kfree(new_r.entries);
 457         kfree(new_gc.entries);
 458
 459         return ret;
 460 err:
 461         bch_err_msg(c, ret, "adding replicas entry");
 462         goto out;
 463 }
 464
 465 int bch2_mark_replicas(struct bch_fs *c, struct bch_replicas_entry_v1 *r)
 466 {
 467         return likely(bch2_replicas_marked(c, r))
 468                 ? 0 : bch2_mark_replicas_slowpath(c, r);
 469 }
 470
 471 /* replicas delta list: */
 472
 473 int bch2_replicas_delta_list_mark(struct bch_fs *c,
 474                                   struct replicas_delta_list *r)
 475 {
 476         struct replicas_delta *d = r->d;
 477         struct replicas_delta *top = (void *) r->d + r->used;
 478         int ret = 0;
 479
 480         for (d = r->d; !ret && d != top; d = replicas_delta_next(d))
 481                 ret = bch2_mark_replicas(c, &d->r);
 482         return ret;
 483 }
 484
 485 /*
 486  * Old replicas_gc mechanism: only used for journal replicas entries now, should
 487  * die at some point:
 488  */
 489
 490 int bch2_replicas_gc_end(struct bch_fs *c, int ret)
 491 {
 492         lockdep_assert_held(&c->replicas_gc_lock);
 493
 494         mutex_lock(&c->sb_lock);
 495         percpu_down_write(&c->mark_lock);
 496
 497         ret =   ret ?:
 498                 bch2_cpu_replicas_to_sb_replicas(c, &c->replicas_gc) ?:
 499                 replicas_table_update(c, &c->replicas_gc);
 500
 501         kfree(c->replicas_gc.entries);
 502         c->replicas_gc.entries = NULL;
 503
 504         percpu_up_write(&c->mark_lock);
 505
 506         if (!ret)
 507                 bch2_write_super(c);
 508
 509         mutex_unlock(&c->sb_lock);
 510
 511         return ret;
 512 }
 513
 514 int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask)
 515 {
 516         struct bch_replicas_entry_v1 *e;
 517         unsigned i = 0;
 518
 519         lockdep_assert_held(&c->replicas_gc_lock);
 520
 521         mutex_lock(&c->sb_lock);
 522         BUG_ON(c->replicas_gc.entries);
 523
 524         c->replicas_gc.nr               = 0;
 525         c->replicas_gc.entry_size       = 0;
 526
 527         for_each_cpu_replicas_entry(&c->replicas, e)
 528                 if (!((1 << e->data_type) & typemask)) {
 529                         c->replicas_gc.nr++;
 530                         c->replicas_gc.entry_size =
 531                                 max_t(unsigned, c->replicas_gc.entry_size,
 532                                       replicas_entry_bytes(e));
 533                 }
 534
 535         c->replicas_gc.entries = kcalloc(c->replicas_gc.nr,
 536                                          c->replicas_gc.entry_size,
 537                                          GFP_KERNEL);
 538         if (!c->replicas_gc.entries) {
 539                 mutex_unlock(&c->sb_lock);
 540                 bch_err(c, "error allocating c->replicas_gc");
 541                 return -BCH_ERR_ENOMEM_replicas_gc;
 542         }
 543
 544         for_each_cpu_replicas_entry(&c->replicas, e)
 545                 if (!((1 << e->data_type) & typemask))
 546                         memcpy(cpu_replicas_entry(&c->replicas_gc, i++),
 547                                e, c->replicas_gc.entry_size);
 548
 549         bch2_cpu_replicas_sort(&c->replicas_gc);
 550         mutex_unlock(&c->sb_lock);
 551
 552         return 0;
 553 }
 554
 555 /*
 556  * New much simpler mechanism for clearing out unneeded replicas entries - drop
 557  * replicas entries that have 0 sectors used.
 558  *
 559  * However, we don't track sector counts for journal usage, so this doesn't drop
 560  * any BCH_DATA_journal entries; the old bch2_replicas_gc_(start|end) mechanism
 561  * is retained for that.
 562  */
 563 int bch2_replicas_gc2(struct bch_fs *c)
 564 {
 565         struct bch_replicas_cpu new = { 0 };
 566         unsigned i, nr;
 567         int ret = 0;
 568
 569         bch2_journal_meta(&c->journal);
 570 retry:
 571         nr              = READ_ONCE(c->replicas.nr);
 572         new.entry_size  = READ_ONCE(c->replicas.entry_size);
 573         new.entries     = kcalloc(nr, new.entry_size, GFP_KERNEL);
 574         if (!new.entries) {
 575                 bch_err(c, "error allocating c->replicas_gc");
 576                 return -BCH_ERR_ENOMEM_replicas_gc;
 577         }
 578
 579         mutex_lock(&c->sb_lock);
 580         percpu_down_write(&c->mark_lock);
 581
 582         if (nr                  != c->replicas.nr ||
 583             new.entry_size      != c->replicas.entry_size) {
 584                 percpu_up_write(&c->mark_lock);
 585                 mutex_unlock(&c->sb_lock);
 586                 kfree(new.entries);
 587                 goto retry;
 588         }
 589
 590         for (i = 0; i < c->replicas.nr; i++) {
 591                 struct bch_replicas_entry_v1 *e =
 592                         cpu_replicas_entry(&c->replicas, i);
 593
 594                 if (e->data_type == BCH_DATA_journal ||
 595                     c->usage_base->replicas[i] ||
 596                     percpu_u64_get(&c->usage[0]->replicas[i]) ||
 597                     percpu_u64_get(&c->usage[1]->replicas[i]) ||
 598                     percpu_u64_get(&c->usage[2]->replicas[i]) ||
 599                     percpu_u64_get(&c->usage[3]->replicas[i]))
 600                         memcpy(cpu_replicas_entry(&new, new.nr++),
 601                                e, new.entry_size);
 602         }
 603
 604         bch2_cpu_replicas_sort(&new);
 605
 606         ret =   bch2_cpu_replicas_to_sb_replicas(c, &new) ?:
 607                 replicas_table_update(c, &new);
 608
 609         kfree(new.entries);
 610
 611         percpu_up_write(&c->mark_lock);
 612
 613         if (!ret)
 614                 bch2_write_super(c);
 615
 616         mutex_unlock(&c->sb_lock);
 617
 618         return ret;
 619 }
 620
 621 int bch2_replicas_set_usage(struct bch_fs *c,
 622                             struct bch_replicas_entry_v1 *r,
 623                             u64 sectors)
 624 {
 625         int ret, idx = bch2_replicas_entry_idx(c, r);
 626
 627         if (idx < 0) {
 628                 struct bch_replicas_cpu n;
 629
 630                 n = cpu_replicas_add_entry(c, &c->replicas, r);
 631                 if (!n.entries)
 632                         return -BCH_ERR_ENOMEM_cpu_replicas;
 633
 634                 ret = replicas_table_update(c, &n);
 635                 if (ret)
 636                         return ret;
 637
 638                 kfree(n.entries);
 639
 640                 idx = bch2_replicas_entry_idx(c, r);
 641                 BUG_ON(ret < 0);
 642         }
 643
 644         c->usage_base->replicas[idx] = sectors;
 645
 646         return 0;
 647 }
 648
 649 /* Replicas tracking - superblock: */
 650
 651 static int
 652 __bch2_sb_replicas_to_cpu_replicas(struct bch_sb_field_replicas *sb_r,
 653                                    struct bch_replicas_cpu *cpu_r)
 654 {
 655         struct bch_replicas_entry_v1 *e, *dst;
 656         unsigned nr = 0, entry_size = 0, idx = 0;
 657
 658         for_each_replicas_entry(sb_r, e) {
 659                 entry_size = max_t(unsigned, entry_size,
 660                                    replicas_entry_bytes(e));
 661                 nr++;
 662         }
 663
 664         cpu_r->entries = kcalloc(nr, entry_size, GFP_KERNEL);
 665         if (!cpu_r->entries)
 666                 return -BCH_ERR_ENOMEM_cpu_replicas;
 667
 668         cpu_r->nr               = nr;
 669         cpu_r->entry_size       = entry_size;
 670
 671         for_each_replicas_entry(sb_r, e) {
 672                 dst = cpu_replicas_entry(cpu_r, idx++);
 673                 memcpy(dst, e, replicas_entry_bytes(e));
 674                 bch2_replicas_entry_sort(dst);
 675         }
 676
 677         return 0;
 678 }
 679
 680 static int
 681 __bch2_sb_replicas_v0_to_cpu_replicas(struct bch_sb_field_replicas_v0 *sb_r,
 682                                       struct bch_replicas_cpu *cpu_r)
 683 {
 684         struct bch_replicas_entry_v0 *e;
 685         unsigned nr = 0, entry_size = 0, idx = 0;
 686
 687         for_each_replicas_entry(sb_r, e) {
 688                 entry_size = max_t(unsigned, entry_size,
 689                                    replicas_entry_bytes(e));
 690                 nr++;
 691         }
 692
 693         entry_size += sizeof(struct bch_replicas_entry_v1) -
 694                 sizeof(struct bch_replicas_entry_v0);
 695
 696         cpu_r->entries = kcalloc(nr, entry_size, GFP_KERNEL);
 697         if (!cpu_r->entries)
 698                 return -BCH_ERR_ENOMEM_cpu_replicas;
 699
 700         cpu_r->nr               = nr;
 701         cpu_r->entry_size       = entry_size;
 702
 703         for_each_replicas_entry(sb_r, e) {
 704                 struct bch_replicas_entry_v1 *dst =
 705                         cpu_replicas_entry(cpu_r, idx++);
 706
 707                 dst->data_type  = e->data_type;
 708                 dst->nr_devs    = e->nr_devs;
 709                 dst->nr_required = 1;
 710                 memcpy(dst->devs, e->devs, e->nr_devs);
 711                 bch2_replicas_entry_sort(dst);
 712         }
 713
 714         return 0;
 715 }
 716
 717 int bch2_sb_replicas_to_cpu_replicas(struct bch_fs *c)
 718 {
 719         struct bch_sb_field_replicas *sb_v1;
 720         struct bch_sb_field_replicas_v0 *sb_v0;
 721         struct bch_replicas_cpu new_r = { 0, 0, NULL };
 722         int ret = 0;
 723
 724         if ((sb_v1 = bch2_sb_field_get(c->disk_sb.sb, replicas)))
 725                 ret = __bch2_sb_replicas_to_cpu_replicas(sb_v1, &new_r);
 726         else if ((sb_v0 = bch2_sb_field_get(c->disk_sb.sb, replicas_v0)))
 727                 ret = __bch2_sb_replicas_v0_to_cpu_replicas(sb_v0, &new_r);
 728         if (ret)
 729                 return ret;
 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_v1 *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_field_resize(&c->disk_sb, replicas_v0,
 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_field_get(c->disk_sb.sb, replicas_v0);
 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_v1 *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_field_resize(&c->disk_sb, replicas,
 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_field_get(c->disk_sb.sb, replicas);
 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         unsigned i;
 830
 831         sort_r(cpu_r->entries,
 832                cpu_r->nr,
 833                cpu_r->entry_size,
 834                bch2_memcmp, NULL,
 835                (void *)(size_t)cpu_r->entry_size);
 836
 837         for (i = 0; i < cpu_r->nr; i++) {
 838                 struct bch_replicas_entry_v1 *e =
 839                         cpu_replicas_entry(cpu_r, i);
 840
 841                 int ret = bch2_replicas_entry_validate(e, sb, err);
 842                 if (ret)
 843                         return ret;
 844
 845                 if (i + 1 < cpu_r->nr) {
 846                         struct bch_replicas_entry_v1 *n =
 847                                 cpu_replicas_entry(cpu_r, i + 1);
 848
 849                         BUG_ON(memcmp(e, n, cpu_r->entry_size) > 0);
 850
 851                         if (!memcmp(e, n, cpu_r->entry_size)) {
 852                                 prt_printf(err, "duplicate replicas entry ");
 853                                 bch2_replicas_entry_to_text(err, e);
 854                                 return -BCH_ERR_invalid_sb_replicas;
 855                         }
 856                 }
 857         }
 858
 859         return 0;
 860 }
 861
 862 static int bch2_sb_replicas_validate(struct bch_sb *sb, struct bch_sb_field *f,
 863                                      struct printbuf *err)
 864 {
 865         struct bch_sb_field_replicas *sb_r = field_to_type(f, replicas);
 866         struct bch_replicas_cpu cpu_r;
 867         int ret;
 868
 869         ret = __bch2_sb_replicas_to_cpu_replicas(sb_r, &cpu_r);
 870         if (ret)
 871                 return ret;
 872
 873         ret = bch2_cpu_replicas_validate(&cpu_r, sb, err);
 874         kfree(cpu_r.entries);
 875         return ret;
 876 }
 877
 878 static void bch2_sb_replicas_to_text(struct printbuf *out,
 879                                      struct bch_sb *sb,
 880                                      struct bch_sb_field *f)
 881 {
 882         struct bch_sb_field_replicas *r = field_to_type(f, replicas);
 883         struct bch_replicas_entry_v1 *e;
 884         bool first = true;
 885
 886         for_each_replicas_entry(r, e) {
 887                 if (!first)
 888                         prt_printf(out, " ");
 889                 first = false;
 890
 891                 bch2_replicas_entry_to_text(out, e);
 892         }
 893         prt_newline(out);
 894 }
 895
 896 const struct bch_sb_field_ops bch_sb_field_ops_replicas = {
 897         .validate       = bch2_sb_replicas_validate,
 898         .to_text        = bch2_sb_replicas_to_text,
 899 };
 900
 901 static int bch2_sb_replicas_v0_validate(struct bch_sb *sb, struct bch_sb_field *f,
 902                                         struct printbuf *err)
 903 {
 904         struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0);
 905         struct bch_replicas_cpu cpu_r;
 906         int ret;
 907
 908         ret = __bch2_sb_replicas_v0_to_cpu_replicas(sb_r, &cpu_r);
 909         if (ret)
 910                 return ret;
 911
 912         ret = bch2_cpu_replicas_validate(&cpu_r, sb, err);
 913         kfree(cpu_r.entries);
 914         return ret;
 915 }
 916
 917 static void bch2_sb_replicas_v0_to_text(struct printbuf *out,
 918                                         struct bch_sb *sb,
 919                                         struct bch_sb_field *f)
 920 {
 921         struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0);
 922         struct bch_replicas_entry_v0 *e;
 923         bool first = true;
 924
 925         for_each_replicas_entry(sb_r, e) {
 926                 if (!first)
 927                         prt_printf(out, " ");
 928                 first = false;
 929
 930                 bch2_replicas_entry_v0_to_text(out, e);
 931         }
 932         prt_newline(out);
 933 }
 934
 935 const struct bch_sb_field_ops bch_sb_field_ops_replicas_v0 = {
 936         .validate       = bch2_sb_replicas_v0_validate,
 937         .to_text        = bch2_sb_replicas_v0_to_text,
 938 };
 939
 940 /* Query replicas: */
 941
 942 bool bch2_have_enough_devs(struct bch_fs *c, struct bch_devs_mask devs,
 943                            unsigned flags, bool print)
 944 {
 945         struct bch_replicas_entry_v1 *e;
 946         bool ret = true;
 947
 948         percpu_down_read(&c->mark_lock);
 949         for_each_cpu_replicas_entry(&c->replicas, e) {
 950                 unsigned i, nr_online = 0, nr_failed = 0, dflags = 0;
 951                 bool metadata = e->data_type < BCH_DATA_user;
 952
 953                 if (e->data_type == BCH_DATA_cached)
 954                         continue;
 955
 956                 for (i = 0; i < e->nr_devs; i++) {
 957                         struct bch_dev *ca = bch_dev_bkey_exists(c, e->devs[i]);
 958
 959                         nr_online += test_bit(e->devs[i], devs.d);
 960                         nr_failed += ca->mi.state == BCH_MEMBER_STATE_failed;
 961                 }
 962
 963                 if (nr_failed == e->nr_devs)
 964                         continue;
 965
 966                 if (nr_online < e->nr_required)
 967                         dflags |= metadata
 968                                 ? BCH_FORCE_IF_METADATA_LOST
 969                                 : BCH_FORCE_IF_DATA_LOST;
 970
 971                 if (nr_online < e->nr_devs)
 972                         dflags |= metadata
 973                                 ? BCH_FORCE_IF_METADATA_DEGRADED
 974                                 : BCH_FORCE_IF_DATA_DEGRADED;
 975
 976                 if (dflags & ~flags) {
 977                         if (print) {
 978                                 struct printbuf buf = PRINTBUF;
 979
 980                                 bch2_replicas_entry_to_text(&buf, e);
 981                                 bch_err(c, "insufficient devices online (%u) for replicas entry %s",
 982                                         nr_online, buf.buf);
 983                                 printbuf_exit(&buf);
 984                         }
 985                         ret = false;
 986                         break;
 987                 }
 988
 989         }
 990         percpu_up_read(&c->mark_lock);
 991
 992         return ret;
 993 }
 994
 995 unsigned bch2_sb_dev_has_data(struct bch_sb *sb, unsigned dev)
 996 {
 997         struct bch_sb_field_replicas *replicas;
 998         struct bch_sb_field_replicas_v0 *replicas_v0;
 999         unsigned i, data_has = 0;
1000
1001         replicas = bch2_sb_field_get(sb, replicas);
1002         replicas_v0 = bch2_sb_field_get(sb, replicas_v0);
1003
1004         if (replicas) {
1005                 struct bch_replicas_entry_v1 *r;
1006
1007                 for_each_replicas_entry(replicas, r)
1008                         for (i = 0; i < r->nr_devs; i++)
1009                                 if (r->devs[i] == dev)
1010                                         data_has |= 1 << r->data_type;
1011         } else if (replicas_v0) {
1012                 struct bch_replicas_entry_v0 *r;
1013
1014                 for_each_replicas_entry_v0(replicas_v0, r)
1015                         for (i = 0; i < r->nr_devs; i++)
1016                                 if (r->devs[i] == dev)
1017                                         data_has |= 1 << r->data_type;
1018         }
1019
1020
1021         return data_has;
1022 }
1023
1024 unsigned bch2_dev_has_data(struct bch_fs *c, struct bch_dev *ca)
1025 {
1026         unsigned ret;
1027
1028         mutex_lock(&c->sb_lock);
1029         ret = bch2_sb_dev_has_data(c->disk_sb.sb, ca->dev_idx);
1030         mutex_unlock(&c->sb_lock);
1031
1032         return ret;
1033 }
1034
1035 void bch2_fs_replicas_exit(struct bch_fs *c)
1036 {
1037         unsigned i;
1038
1039         kfree(c->usage_scratch);
1040         for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1041                 free_percpu(c->usage[i]);
1042         kfree(c->usage_base);
1043         kfree(c->replicas.entries);
1044         kfree(c->replicas_gc.entries);
1045
1046         mempool_exit(&c->replicas_delta_pool);
1047 }
1048
1049 int bch2_fs_replicas_init(struct bch_fs *c)
1050 {
1051         bch2_journal_entry_res_resize(&c->journal,
1052                         &c->replicas_journal_res,
1053                         reserve_journal_replicas(c, &c->replicas));
1054
1055         return mempool_init_kmalloc_pool(&c->replicas_delta_pool, 1,
1056                                          REPLICAS_DELTA_LIST_MAX) ?:
1057                 replicas_table_update(c, &c->replicas);
1058 }