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_to_text(struct printbuf *out,
  40                                  struct bch_replicas_entry *e)
  41 {
  42         unsigned i;
  43
  44         pr_buf(out, "%s: %u/%u [",
  45                bch2_data_types[e->data_type],
  46                e->nr_required,
  47                e->nr_devs);
  48
  49         for (i = 0; i < e->nr_devs; i++)
  50                 pr_buf(out, i ? " %u" : "%u", e->devs[i]);
  51         pr_buf(out, "]");
  52 }
  53
  54 void bch2_cpu_replicas_to_text(struct printbuf *out,
  55                               struct bch_replicas_cpu *r)
  56 {
  57         struct bch_replicas_entry *e;
  58         bool first = true;
  59
  60         for_each_cpu_replicas_entry(r, e) {
  61                 if (!first)
  62                         pr_buf(out, " ");
  63                 first = false;
  64
  65                 bch2_replicas_entry_to_text(out, e);
  66         }
  67 }
  68
  69 static void extent_to_replicas(struct bkey_s_c k,
  70                                struct bch_replicas_entry *r)
  71 {
  72         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
  73         const union bch_extent_entry *entry;
  74         struct extent_ptr_decoded p;
  75
  76         r->nr_required  = 1;
  77
  78         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
  79                 if (p.ptr.cached)
  80                         continue;
  81
  82                 if (!p.has_ec)
  83                         r->devs[r->nr_devs++] = p.ptr.dev;
  84                 else
  85                         r->nr_required = 0;
  86         }
  87 }
  88
  89 static void stripe_to_replicas(struct bkey_s_c k,
  90                                struct bch_replicas_entry *r)
  91 {
  92         struct bkey_s_c_stripe s = bkey_s_c_to_stripe(k);
  93         const struct bch_extent_ptr *ptr;
  94
  95         r->nr_required  = s.v->nr_blocks - s.v->nr_redundant;
  96
  97         for (ptr = s.v->ptrs;
  98              ptr < s.v->ptrs + s.v->nr_blocks;
  99              ptr++)
 100                 r->devs[r->nr_devs++] = ptr->dev;
 101 }
 102
 103 void bch2_bkey_to_replicas(struct bch_replicas_entry *e,
 104                            struct bkey_s_c k)
 105 {
 106         e->nr_devs = 0;
 107
 108         switch (k.k->type) {
 109         case KEY_TYPE_btree_ptr:
 110         case KEY_TYPE_btree_ptr_v2:
 111                 e->data_type = BCH_DATA_btree;
 112                 extent_to_replicas(k, e);
 113                 break;
 114         case KEY_TYPE_extent:
 115         case KEY_TYPE_reflink_v:
 116                 e->data_type = BCH_DATA_user;
 117                 extent_to_replicas(k, e);
 118                 break;
 119         case KEY_TYPE_stripe:
 120                 e->data_type = BCH_DATA_parity;
 121                 stripe_to_replicas(k, e);
 122                 break;
 123         }
 124
 125         bch2_replicas_entry_sort(e);
 126 }
 127
 128 void bch2_devlist_to_replicas(struct bch_replicas_entry *e,
 129                               enum bch_data_type data_type,
 130                               struct bch_devs_list devs)
 131 {
 132         unsigned i;
 133
 134         BUG_ON(!data_type ||
 135                data_type == BCH_DATA_sb ||
 136                data_type >= BCH_DATA_NR);
 137
 138         e->data_type    = data_type;
 139         e->nr_devs      = 0;
 140         e->nr_required  = 1;
 141
 142         for (i = 0; i < devs.nr; i++)
 143                 e->devs[e->nr_devs++] = devs.devs[i];
 144
 145         bch2_replicas_entry_sort(e);
 146 }
 147
 148 static struct bch_replicas_cpu
 149 cpu_replicas_add_entry(struct bch_replicas_cpu *old,
 150                        struct bch_replicas_entry *new_entry)
 151 {
 152         unsigned i;
 153         struct bch_replicas_cpu new = {
 154                 .nr             = old->nr + 1,
 155                 .entry_size     = max_t(unsigned, old->entry_size,
 156                                         replicas_entry_bytes(new_entry)),
 157         };
 158
 159         BUG_ON(!new_entry->data_type);
 160         verify_replicas_entry(new_entry);
 161
 162         new.entries = kcalloc(new.nr, new.entry_size, GFP_KERNEL);
 163         if (!new.entries)
 164                 return new;
 165
 166         for (i = 0; i < old->nr; i++)
 167                 memcpy(cpu_replicas_entry(&new, i),
 168                        cpu_replicas_entry(old, i),
 169                        old->entry_size);
 170
 171         memcpy(cpu_replicas_entry(&new, old->nr),
 172                new_entry,
 173                replicas_entry_bytes(new_entry));
 174
 175         bch2_cpu_replicas_sort(&new);
 176         return new;
 177 }
 178
 179 static inline int __replicas_entry_idx(struct bch_replicas_cpu *r,
 180                                        struct bch_replicas_entry *search)
 181 {
 182         int idx, entry_size = replicas_entry_bytes(search);
 183
 184         if (unlikely(entry_size > r->entry_size))
 185                 return -1;
 186
 187         verify_replicas_entry(search);
 188
 189 #define entry_cmp(_l, _r, size) memcmp(_l, _r, entry_size)
 190         idx = eytzinger0_find(r->entries, r->nr, r->entry_size,
 191                               entry_cmp, search);
 192 #undef entry_cmp
 193
 194         return idx < r->nr ? idx : -1;
 195 }
 196
 197 int bch2_replicas_entry_idx(struct bch_fs *c,
 198                             struct bch_replicas_entry *search)
 199 {
 200         bch2_replicas_entry_sort(search);
 201
 202         return __replicas_entry_idx(&c->replicas, search);
 203 }
 204
 205 static bool __replicas_has_entry(struct bch_replicas_cpu *r,
 206                                  struct bch_replicas_entry *search)
 207 {
 208         return __replicas_entry_idx(r, search) >= 0;
 209 }
 210
 211 bool bch2_replicas_marked(struct bch_fs *c,
 212                           struct bch_replicas_entry *search)
 213 {
 214         bool marked;
 215
 216         if (!search->nr_devs)
 217                 return true;
 218
 219         verify_replicas_entry(search);
 220
 221         percpu_down_read(&c->mark_lock);
 222         marked = __replicas_has_entry(&c->replicas, search) &&
 223                 (likely((!c->replicas_gc.entries)) ||
 224                  __replicas_has_entry(&c->replicas_gc, search));
 225         percpu_up_read(&c->mark_lock);
 226
 227         return marked;
 228 }
 229
 230 static void __replicas_table_update(struct bch_fs_usage *dst,
 231                                     struct bch_replicas_cpu *dst_r,
 232                                     struct bch_fs_usage *src,
 233                                     struct bch_replicas_cpu *src_r)
 234 {
 235         int src_idx, dst_idx;
 236
 237         *dst = *src;
 238
 239         for (src_idx = 0; src_idx < src_r->nr; src_idx++) {
 240                 if (!src->replicas[src_idx])
 241                         continue;
 242
 243                 dst_idx = __replicas_entry_idx(dst_r,
 244                                 cpu_replicas_entry(src_r, src_idx));
 245                 BUG_ON(dst_idx < 0);
 246
 247                 dst->replicas[dst_idx] = src->replicas[src_idx];
 248         }
 249 }
 250
 251 static void __replicas_table_update_pcpu(struct bch_fs_usage __percpu *dst_p,
 252                                     struct bch_replicas_cpu *dst_r,
 253                                     struct bch_fs_usage __percpu *src_p,
 254                                     struct bch_replicas_cpu *src_r)
 255 {
 256         unsigned src_nr = sizeof(struct bch_fs_usage) / sizeof(u64) + src_r->nr;
 257         struct bch_fs_usage *dst, *src = (void *)
 258                 bch2_acc_percpu_u64s((void *) src_p, src_nr);
 259
 260         preempt_disable();
 261         dst = this_cpu_ptr(dst_p);
 262         preempt_enable();
 263
 264         __replicas_table_update(dst, dst_r, src, src_r);
 265 }
 266
 267 /*
 268  * Resize filesystem accounting:
 269  */
 270 static int replicas_table_update(struct bch_fs *c,
 271                                  struct bch_replicas_cpu *new_r)
 272 {
 273         struct bch_fs_usage __percpu *new_usage[JOURNAL_BUF_NR];
 274         struct bch_fs_usage_online *new_scratch = NULL;
 275         struct bch_fs_usage __percpu *new_gc = NULL;
 276         struct bch_fs_usage *new_base = NULL;
 277         unsigned i, bytes = sizeof(struct bch_fs_usage) +
 278                 sizeof(u64) * new_r->nr;
 279         unsigned scratch_bytes = sizeof(struct bch_fs_usage_online) +
 280                 sizeof(u64) * new_r->nr;
 281         int ret = 0;
 282
 283         memset(new_usage, 0, sizeof(new_usage));
 284
 285         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 286                 if (!(new_usage[i] = __alloc_percpu_gfp(bytes,
 287                                         sizeof(u64), GFP_KERNEL)))
 288                         goto err;
 289
 290         if (!(new_base = kzalloc(bytes, GFP_KERNEL)) ||
 291             !(new_scratch  = kmalloc(scratch_bytes, GFP_KERNEL)) ||
 292             (c->usage_gc &&
 293              !(new_gc = __alloc_percpu_gfp(bytes, sizeof(u64), GFP_KERNEL))))
 294                 goto err;
 295
 296         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 297                 if (c->usage[i])
 298                         __replicas_table_update_pcpu(new_usage[i], new_r,
 299                                                      c->usage[i], &c->replicas);
 300         if (c->usage_base)
 301                 __replicas_table_update(new_base,               new_r,
 302                                         c->usage_base,          &c->replicas);
 303         if (c->usage_gc)
 304                 __replicas_table_update_pcpu(new_gc,            new_r,
 305                                              c->usage_gc,       &c->replicas);
 306
 307         for (i = 0; i < ARRAY_SIZE(new_usage); i++)
 308                 swap(c->usage[i],       new_usage[i]);
 309         swap(c->usage_base,     new_base);
 310         swap(c->usage_scratch,  new_scratch);
 311         swap(c->usage_gc,       new_gc);
 312         swap(c->replicas,       *new_r);
 313 out:
 314         free_percpu(new_gc);
 315         kfree(new_scratch);
 316         free_percpu(new_usage[1]);
 317         free_percpu(new_usage[0]);
 318         kfree(new_base);
 319         return ret;
 320 err:
 321         bch_err(c, "error updating replicas table: memory allocation failure");
 322         ret = -ENOMEM;
 323         goto out;
 324 }
 325
 326 static unsigned reserve_journal_replicas(struct bch_fs *c,
 327                                      struct bch_replicas_cpu *r)
 328 {
 329         struct bch_replicas_entry *e;
 330         unsigned journal_res_u64s = 0;
 331
 332         /* nr_inodes: */
 333         journal_res_u64s +=
 334                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 335
 336         /* key_version: */
 337         journal_res_u64s +=
 338                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
 339
 340         /* persistent_reserved: */
 341         journal_res_u64s +=
 342                 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64)) *
 343                 BCH_REPLICAS_MAX;
 344
 345         for_each_cpu_replicas_entry(r, e)
 346                 journal_res_u64s +=
 347                         DIV_ROUND_UP(sizeof(struct jset_entry_data_usage) +
 348                                      e->nr_devs, sizeof(u64));
 349         return journal_res_u64s;
 350 }
 351
 352 noinline
 353 static int bch2_mark_replicas_slowpath(struct bch_fs *c,
 354                                 struct bch_replicas_entry *new_entry)
 355 {
 356         struct bch_replicas_cpu new_r, new_gc;
 357         int ret = 0;
 358
 359         verify_replicas_entry(new_entry);
 360
 361         memset(&new_r, 0, sizeof(new_r));
 362         memset(&new_gc, 0, sizeof(new_gc));
 363
 364         mutex_lock(&c->sb_lock);
 365
 366         if (c->replicas_gc.entries &&
 367             !__replicas_has_entry(&c->replicas_gc, new_entry)) {
 368                 new_gc = cpu_replicas_add_entry(&c->replicas_gc, new_entry);
 369                 if (!new_gc.entries)
 370                         goto err;
 371         }
 372
 373         if (!__replicas_has_entry(&c->replicas, new_entry)) {
 374                 new_r = cpu_replicas_add_entry(&c->replicas, new_entry);
 375                 if (!new_r.entries)
 376                         goto err;
 377
 378                 ret = bch2_cpu_replicas_to_sb_replicas(c, &new_r);
 379                 if (ret)
 380                         goto err;
 381
 382                 bch2_journal_entry_res_resize(&c->journal,
 383                                 &c->replicas_journal_res,
 384                                 reserve_journal_replicas(c, &new_r));
 385         }
 386
 387         if (!new_r.entries &&
 388             !new_gc.entries)
 389                 goto out;
 390
 391         /* allocations done, now commit: */
 392
 393         if (new_r.entries)
 394                 bch2_write_super(c);
 395
 396         /* don't update in memory replicas until changes are persistent */
 397         percpu_down_write(&c->mark_lock);
 398         if (new_r.entries)
 399                 ret = replicas_table_update(c, &new_r);
 400         if (new_gc.entries)
 401                 swap(new_gc, c->replicas_gc);
 402         percpu_up_write(&c->mark_lock);
 403 out:
 404         mutex_unlock(&c->sb_lock);
 405
 406         kfree(new_r.entries);
 407         kfree(new_gc.entries);
 408
 409         return ret;
 410 err:
 411         bch_err(c, "error adding replicas entry: memory allocation failure");
 412         ret = -ENOMEM;
 413         goto out;
 414 }
 415
 416 static int __bch2_mark_replicas(struct bch_fs *c,
 417                                 struct bch_replicas_entry *r,
 418                                 bool check)
 419 {
 420         return likely(bch2_replicas_marked(c, r))       ? 0
 421                 : check                                 ? -1
 422                 : bch2_mark_replicas_slowpath(c, r);
 423 }
 424
 425 int bch2_mark_replicas(struct bch_fs *c, struct bch_replicas_entry *r)
 426 {
 427         return __bch2_mark_replicas(c, r, false);
 428 }
 429
 430 static int __bch2_mark_bkey_replicas(struct bch_fs *c, struct bkey_s_c k,
 431                                      bool check)
 432 {
 433         struct bch_replicas_padded search;
 434         struct bch_devs_list cached = bch2_bkey_cached_devs(k);
 435         unsigned i;
 436         int ret;
 437
 438         for (i = 0; i < cached.nr; i++) {
 439                 bch2_replicas_entry_cached(&search.e, cached.devs[i]);
 440
 441                 ret = __bch2_mark_replicas(c, &search.e, check);
 442                 if (ret)
 443                         return ret;
 444         }
 445
 446         bch2_bkey_to_replicas(&search.e, k);
 447
 448         ret = __bch2_mark_replicas(c, &search.e, check);
 449         if (ret)
 450                 return ret;
 451
 452         if (search.e.data_type == BCH_DATA_parity) {
 453                 search.e.data_type = BCH_DATA_cached;
 454                 ret = __bch2_mark_replicas(c, &search.e, check);
 455                 if (ret)
 456                         return ret;
 457
 458                 search.e.data_type = BCH_DATA_user;
 459                 ret = __bch2_mark_replicas(c, &search.e, check);
 460                 if (ret)
 461                         return ret;
 462         }
 463
 464         return 0;
 465 }
 466
 467 /* replicas delta list: */
 468
 469 bool bch2_replicas_delta_list_marked(struct bch_fs *c,
 470                                      struct replicas_delta_list *r)
 471 {
 472         struct replicas_delta *d = r->d;
 473         struct replicas_delta *top = (void *) r->d + r->used;
 474
 475         percpu_rwsem_assert_held(&c->mark_lock);
 476
 477         for (d = r->d; d != top; d = replicas_delta_next(d))
 478                 if (bch2_replicas_entry_idx(c, &d->r) < 0)
 479                         return false;
 480         return true;
 481 }
 482
 483 int bch2_replicas_delta_list_mark(struct bch_fs *c,
 484                                   struct replicas_delta_list *r)
 485 {
 486         struct replicas_delta *d = r->d;
 487         struct replicas_delta *top = (void *) r->d + r->used;
 488         int ret = 0;
 489
 490         for (d = r->d; !ret && d != top; d = replicas_delta_next(d))
 491                 ret = bch2_mark_replicas(c, &d->r);
 492         return ret;
 493 }
 494
 495 /* bkey replicas: */
 496
 497 bool bch2_bkey_replicas_marked(struct bch_fs *c,
 498                                struct bkey_s_c k)
 499 {
 500         return __bch2_mark_bkey_replicas(c, k, true) == 0;
 501 }
 502
 503 int bch2_mark_bkey_replicas(struct bch_fs *c, struct bkey_s_c k)
 504 {
 505         return __bch2_mark_bkey_replicas(c, k, false);
 506 }
 507
 508 /*
 509  * Old replicas_gc mechanism: only used for journal replicas entries now, should
 510  * die at some point:
 511  */
 512
 513 int bch2_replicas_gc_end(struct bch_fs *c, int ret)
 514 {
 515         unsigned i;
 516
 517         lockdep_assert_held(&c->replicas_gc_lock);
 518
 519         mutex_lock(&c->sb_lock);
 520         percpu_down_write(&c->mark_lock);
 521
 522         /*
 523          * this is kind of crappy; the replicas gc mechanism needs to be ripped
 524          * out
 525          */
 526
 527         for (i = 0; i < c->replicas.nr; i++) {
 528                 struct bch_replicas_entry *e =
 529                         cpu_replicas_entry(&c->replicas, i);
 530                 struct bch_replicas_cpu n;
 531
 532                 if (!__replicas_has_entry(&c->replicas_gc, e) &&
 533                     bch2_fs_usage_read_one(c, &c->usage_base->replicas[i])) {
 534                         n = cpu_replicas_add_entry(&c->replicas_gc, e);
 535                         if (!n.entries) {
 536                                 ret = -ENOSPC;
 537                                 goto err;
 538                         }
 539
 540                         swap(n, c->replicas_gc);
 541                         kfree(n.entries);
 542                 }
 543         }
 544
 545         if (bch2_cpu_replicas_to_sb_replicas(c, &c->replicas_gc)) {
 546                 ret = -ENOSPC;
 547                 goto err;
 548         }
 549
 550         ret = replicas_table_update(c, &c->replicas_gc);
 551 err:
 552         kfree(c->replicas_gc.entries);
 553         c->replicas_gc.entries = NULL;
 554
 555         percpu_up_write(&c->mark_lock);
 556
 557         if (!ret)
 558                 bch2_write_super(c);
 559
 560         mutex_unlock(&c->sb_lock);
 561
 562         return ret;
 563 }
 564
 565 int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask)
 566 {
 567         struct bch_replicas_entry *e;
 568         unsigned i = 0;
 569
 570         lockdep_assert_held(&c->replicas_gc_lock);
 571
 572         mutex_lock(&c->sb_lock);
 573         BUG_ON(c->replicas_gc.entries);
 574
 575         c->replicas_gc.nr               = 0;
 576         c->replicas_gc.entry_size       = 0;
 577
 578         for_each_cpu_replicas_entry(&c->replicas, e)
 579                 if (!((1 << e->data_type) & typemask)) {
 580                         c->replicas_gc.nr++;
 581                         c->replicas_gc.entry_size =
 582                                 max_t(unsigned, c->replicas_gc.entry_size,
 583                                       replicas_entry_bytes(e));
 584                 }
 585
 586         c->replicas_gc.entries = kcalloc(c->replicas_gc.nr,
 587                                          c->replicas_gc.entry_size,
 588                                          GFP_KERNEL);
 589         if (!c->replicas_gc.entries) {
 590                 mutex_unlock(&c->sb_lock);
 591                 bch_err(c, "error allocating c->replicas_gc");
 592                 return -ENOMEM;
 593         }
 594
 595         for_each_cpu_replicas_entry(&c->replicas, e)
 596                 if (!((1 << e->data_type) & typemask))
 597                         memcpy(cpu_replicas_entry(&c->replicas_gc, i++),
 598                                e, c->replicas_gc.entry_size);
 599
 600         bch2_cpu_replicas_sort(&c->replicas_gc);
 601         mutex_unlock(&c->sb_lock);
 602
 603         return 0;
 604 }
 605
 606 /* New much simpler mechanism for clearing out unneeded replicas entries: */
 607
 608 int bch2_replicas_gc2(struct bch_fs *c)
 609 {
 610         struct bch_replicas_cpu new = { 0 };
 611         unsigned i, nr;
 612         int ret = 0;
 613
 614         bch2_journal_meta(&c->journal);
 615 retry:
 616         nr              = READ_ONCE(c->replicas.nr);
 617         new.entry_size  = READ_ONCE(c->replicas.entry_size);
 618         new.entries     = kcalloc(nr, new.entry_size, GFP_KERNEL);
 619         if (!new.entries) {
 620                 bch_err(c, "error allocating c->replicas_gc");
 621                 return -ENOMEM;
 622         }
 623
 624         mutex_lock(&c->sb_lock);
 625         percpu_down_write(&c->mark_lock);
 626
 627         if (nr                  != c->replicas.nr ||
 628             new.entry_size      != c->replicas.entry_size) {
 629                 percpu_up_write(&c->mark_lock);
 630                 mutex_unlock(&c->sb_lock);
 631                 kfree(new.entries);
 632                 goto retry;
 633         }
 634
 635         for (i = 0; i < c->replicas.nr; i++) {
 636                 struct bch_replicas_entry *e =
 637                         cpu_replicas_entry(&c->replicas, i);
 638
 639                 if (e->data_type == BCH_DATA_journal ||
 640                     c->usage_base->replicas[i] ||
 641                     percpu_u64_get(&c->usage[0]->replicas[i]) ||
 642                     percpu_u64_get(&c->usage[1]->replicas[i]) ||
 643                     percpu_u64_get(&c->usage[2]->replicas[i]) ||
 644                     percpu_u64_get(&c->usage[3]->replicas[i]))
 645                         memcpy(cpu_replicas_entry(&new, new.nr++),
 646                                e, new.entry_size);
 647         }
 648
 649         bch2_cpu_replicas_sort(&new);
 650
 651         if (bch2_cpu_replicas_to_sb_replicas(c, &new)) {
 652                 ret = -ENOSPC;
 653                 goto err;
 654         }
 655
 656         ret = replicas_table_update(c, &new);
 657 err:
 658         kfree(new.entries);
 659
 660         percpu_up_write(&c->mark_lock);
 661
 662         if (!ret)
 663                 bch2_write_super(c);
 664
 665         mutex_unlock(&c->sb_lock);
 666
 667         return ret;
 668 }
 669
 670 int bch2_replicas_set_usage(struct bch_fs *c,
 671                             struct bch_replicas_entry *r,
 672                             u64 sectors)
 673 {
 674         int ret, idx = bch2_replicas_entry_idx(c, r);
 675
 676         if (idx < 0) {
 677                 struct bch_replicas_cpu n;
 678
 679                 n = cpu_replicas_add_entry(&c->replicas, r);
 680                 if (!n.entries)
 681                         return -ENOMEM;
 682
 683                 ret = replicas_table_update(c, &n);
 684                 if (ret)
 685                         return ret;
 686
 687                 kfree(n.entries);
 688
 689                 idx = bch2_replicas_entry_idx(c, r);
 690                 BUG_ON(ret < 0);
 691         }
 692
 693         c->usage_base->replicas[idx] = sectors;
 694
 695         return 0;
 696 }
 697
 698 /* Replicas tracking - superblock: */
 699
 700 static int
 701 __bch2_sb_replicas_to_cpu_replicas(struct bch_sb_field_replicas *sb_r,
 702                                    struct bch_replicas_cpu *cpu_r)
 703 {
 704         struct bch_replicas_entry *e, *dst;
 705         unsigned nr = 0, entry_size = 0, idx = 0;
 706
 707         for_each_replicas_entry(sb_r, e) {
 708                 entry_size = max_t(unsigned, entry_size,
 709                                    replicas_entry_bytes(e));
 710                 nr++;
 711         }
 712
 713         cpu_r->entries = kcalloc(nr, entry_size, GFP_KERNEL);
 714         if (!cpu_r->entries)
 715                 return -ENOMEM;
 716
 717         cpu_r->nr               = nr;
 718         cpu_r->entry_size       = entry_size;
 719
 720         for_each_replicas_entry(sb_r, e) {
 721                 dst = cpu_replicas_entry(cpu_r, idx++);
 722                 memcpy(dst, e, replicas_entry_bytes(e));
 723                 bch2_replicas_entry_sort(dst);
 724         }
 725
 726         return 0;
 727 }
 728
 729 static int
 730 __bch2_sb_replicas_v0_to_cpu_replicas(struct bch_sb_field_replicas_v0 *sb_r,
 731                                       struct bch_replicas_cpu *cpu_r)
 732 {
 733         struct bch_replicas_entry_v0 *e;
 734         unsigned nr = 0, entry_size = 0, idx = 0;
 735
 736         for_each_replicas_entry(sb_r, e) {
 737                 entry_size = max_t(unsigned, entry_size,
 738                                    replicas_entry_bytes(e));
 739                 nr++;
 740         }
 741
 742         entry_size += sizeof(struct bch_replicas_entry) -
 743                 sizeof(struct bch_replicas_entry_v0);
 744
 745         cpu_r->entries = kcalloc(nr, entry_size, GFP_KERNEL);
 746         if (!cpu_r->entries)
 747                 return -ENOMEM;
 748
 749         cpu_r->nr               = nr;
 750         cpu_r->entry_size       = entry_size;
 751
 752         for_each_replicas_entry(sb_r, e) {
 753                 struct bch_replicas_entry *dst =
 754                         cpu_replicas_entry(cpu_r, idx++);
 755
 756                 dst->data_type  = e->data_type;
 757                 dst->nr_devs    = e->nr_devs;
 758                 dst->nr_required = 1;
 759                 memcpy(dst->devs, e->devs, e->nr_devs);
 760                 bch2_replicas_entry_sort(dst);
 761         }
 762
 763         return 0;
 764 }
 765
 766 int bch2_sb_replicas_to_cpu_replicas(struct bch_fs *c)
 767 {
 768         struct bch_sb_field_replicas *sb_v1;
 769         struct bch_sb_field_replicas_v0 *sb_v0;
 770         struct bch_replicas_cpu new_r = { 0, 0, NULL };
 771         int ret = 0;
 772
 773         if ((sb_v1 = bch2_sb_get_replicas(c->disk_sb.sb)))
 774                 ret = __bch2_sb_replicas_to_cpu_replicas(sb_v1, &new_r);
 775         else if ((sb_v0 = bch2_sb_get_replicas_v0(c->disk_sb.sb)))
 776                 ret = __bch2_sb_replicas_v0_to_cpu_replicas(sb_v0, &new_r);
 777
 778         if (ret)
 779                 return -ENOMEM;
 780
 781         bch2_cpu_replicas_sort(&new_r);
 782
 783         percpu_down_write(&c->mark_lock);
 784
 785         ret = replicas_table_update(c, &new_r);
 786         percpu_up_write(&c->mark_lock);
 787
 788         kfree(new_r.entries);
 789
 790         return 0;
 791 }
 792
 793 static int bch2_cpu_replicas_to_sb_replicas_v0(struct bch_fs *c,
 794                                                struct bch_replicas_cpu *r)
 795 {
 796         struct bch_sb_field_replicas_v0 *sb_r;
 797         struct bch_replicas_entry_v0 *dst;
 798         struct bch_replicas_entry *src;
 799         size_t bytes;
 800
 801         bytes = sizeof(struct bch_sb_field_replicas);
 802
 803         for_each_cpu_replicas_entry(r, src)
 804                 bytes += replicas_entry_bytes(src) - 1;
 805
 806         sb_r = bch2_sb_resize_replicas_v0(&c->disk_sb,
 807                         DIV_ROUND_UP(bytes, sizeof(u64)));
 808         if (!sb_r)
 809                 return -ENOSPC;
 810
 811         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas);
 812         sb_r = bch2_sb_get_replicas_v0(c->disk_sb.sb);
 813
 814         memset(&sb_r->entries, 0,
 815                vstruct_end(&sb_r->field) -
 816                (void *) &sb_r->entries);
 817
 818         dst = sb_r->entries;
 819         for_each_cpu_replicas_entry(r, src) {
 820                 dst->data_type  = src->data_type;
 821                 dst->nr_devs    = src->nr_devs;
 822                 memcpy(dst->devs, src->devs, src->nr_devs);
 823
 824                 dst = replicas_entry_next(dst);
 825
 826                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 827         }
 828
 829         return 0;
 830 }
 831
 832 static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *c,
 833                                             struct bch_replicas_cpu *r)
 834 {
 835         struct bch_sb_field_replicas *sb_r;
 836         struct bch_replicas_entry *dst, *src;
 837         bool need_v1 = false;
 838         size_t bytes;
 839
 840         bytes = sizeof(struct bch_sb_field_replicas);
 841
 842         for_each_cpu_replicas_entry(r, src) {
 843                 bytes += replicas_entry_bytes(src);
 844                 if (src->nr_required != 1)
 845                         need_v1 = true;
 846         }
 847
 848         if (!need_v1)
 849                 return bch2_cpu_replicas_to_sb_replicas_v0(c, r);
 850
 851         sb_r = bch2_sb_resize_replicas(&c->disk_sb,
 852                         DIV_ROUND_UP(bytes, sizeof(u64)));
 853         if (!sb_r)
 854                 return -ENOSPC;
 855
 856         bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas_v0);
 857         sb_r = bch2_sb_get_replicas(c->disk_sb.sb);
 858
 859         memset(&sb_r->entries, 0,
 860                vstruct_end(&sb_r->field) -
 861                (void *) &sb_r->entries);
 862
 863         dst = sb_r->entries;
 864         for_each_cpu_replicas_entry(r, src) {
 865                 memcpy(dst, src, replicas_entry_bytes(src));
 866
 867                 dst = replicas_entry_next(dst);
 868
 869                 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
 870         }
 871
 872         return 0;
 873 }
 874
 875 static const char *check_dup_replicas_entries(struct bch_replicas_cpu *cpu_r)
 876 {
 877         unsigned i;
 878
 879         sort_cmp_size(cpu_r->entries,
 880                       cpu_r->nr,
 881                       cpu_r->entry_size,
 882                       memcmp, NULL);
 883
 884         for (i = 0; i + 1 < cpu_r->nr; i++) {
 885                 struct bch_replicas_entry *l =
 886                         cpu_replicas_entry(cpu_r, i);
 887                 struct bch_replicas_entry *r =
 888                         cpu_replicas_entry(cpu_r, i + 1);
 889
 890                 BUG_ON(memcmp(l, r, cpu_r->entry_size) > 0);
 891
 892                 if (!memcmp(l, r, cpu_r->entry_size))
 893                         return "duplicate replicas entry";
 894         }
 895
 896         return NULL;
 897 }
 898
 899 static const char *bch2_sb_validate_replicas(struct bch_sb *sb, struct bch_sb_field *f)
 900 {
 901         struct bch_sb_field_replicas *sb_r = field_to_type(f, replicas);
 902         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 903         struct bch_replicas_cpu cpu_r = { .entries = NULL };
 904         struct bch_replicas_entry *e;
 905         const char *err;
 906         unsigned i;
 907
 908         for_each_replicas_entry(sb_r, e) {
 909                 err = "invalid replicas entry: invalid data type";
 910                 if (e->data_type >= BCH_DATA_NR)
 911                         goto err;
 912
 913                 err = "invalid replicas entry: no devices";
 914                 if (!e->nr_devs)
 915                         goto err;
 916
 917                 err = "invalid replicas entry: bad nr_required";
 918                 if (e->nr_required > 1 &&
 919                     e->nr_required >= e->nr_devs)
 920                         goto err;
 921
 922                 err = "invalid replicas entry: invalid device";
 923                 for (i = 0; i < e->nr_devs; i++)
 924                         if (!bch2_dev_exists(sb, mi, e->devs[i]))
 925                                 goto err;
 926         }
 927
 928         err = "cannot allocate memory";
 929         if (__bch2_sb_replicas_to_cpu_replicas(sb_r, &cpu_r))
 930                 goto err;
 931
 932         err = check_dup_replicas_entries(&cpu_r);
 933 err:
 934         kfree(cpu_r.entries);
 935         return err;
 936 }
 937
 938 static void bch2_sb_replicas_to_text(struct printbuf *out,
 939                                      struct bch_sb *sb,
 940                                      struct bch_sb_field *f)
 941 {
 942         struct bch_sb_field_replicas *r = field_to_type(f, replicas);
 943         struct bch_replicas_entry *e;
 944         bool first = true;
 945
 946         for_each_replicas_entry(r, e) {
 947                 if (!first)
 948                         pr_buf(out, " ");
 949                 first = false;
 950
 951                 bch2_replicas_entry_to_text(out, e);
 952         }
 953 }
 954
 955 const struct bch_sb_field_ops bch_sb_field_ops_replicas = {
 956         .validate       = bch2_sb_validate_replicas,
 957         .to_text        = bch2_sb_replicas_to_text,
 958 };
 959
 960 static const char *bch2_sb_validate_replicas_v0(struct bch_sb *sb, struct bch_sb_field *f)
 961 {
 962         struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0);
 963         struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
 964         struct bch_replicas_cpu cpu_r = { .entries = NULL };
 965         struct bch_replicas_entry_v0 *e;
 966         const char *err;
 967         unsigned i;
 968
 969         for_each_replicas_entry_v0(sb_r, e) {
 970                 err = "invalid replicas entry: invalid data type";
 971                 if (e->data_type >= BCH_DATA_NR)
 972                         goto err;
 973
 974                 err = "invalid replicas entry: no devices";
 975                 if (!e->nr_devs)
 976                         goto err;
 977
 978                 err = "invalid replicas entry: invalid device";
 979                 for (i = 0; i < e->nr_devs; i++)
 980                         if (!bch2_dev_exists(sb, mi, e->devs[i]))
 981                                 goto err;
 982         }
 983
 984         err = "cannot allocate memory";
 985         if (__bch2_sb_replicas_v0_to_cpu_replicas(sb_r, &cpu_r))
 986                 goto err;
 987
 988         err = check_dup_replicas_entries(&cpu_r);
 989 err:
 990         kfree(cpu_r.entries);
 991         return err;
 992 }
 993
 994 const struct bch_sb_field_ops bch_sb_field_ops_replicas_v0 = {
 995         .validate       = bch2_sb_validate_replicas_v0,
 996 };
 997
 998 /* Query replicas: */
 999
1000 bool bch2_have_enough_devs(struct bch_fs *c, struct bch_devs_mask devs,
1001                            unsigned flags, bool print)
1002 {
1003         struct bch_replicas_entry *e;
1004         bool ret = true;
1005
1006         percpu_down_read(&c->mark_lock);
1007         for_each_cpu_replicas_entry(&c->replicas, e) {
1008                 unsigned i, nr_online = 0, nr_failed = 0, dflags = 0;
1009                 bool metadata = e->data_type < BCH_DATA_user;
1010
1011                 for (i = 0; i < e->nr_devs; i++) {
1012                         struct bch_dev *ca = bch_dev_bkey_exists(c, e->devs[i]);
1013
1014                         nr_online += test_bit(e->devs[i], devs.d);
1015                         nr_failed += ca->mi.state == BCH_MEMBER_STATE_failed;
1016                 }
1017
1018                 if (nr_failed == e->nr_devs)
1019                         continue;
1020
1021                 if (nr_online < e->nr_required)
1022                         dflags |= metadata
1023                                 ? BCH_FORCE_IF_METADATA_LOST
1024                                 : BCH_FORCE_IF_DATA_LOST;
1025
1026                 if (nr_online < e->nr_devs)
1027                         dflags |= metadata
1028                                 ? BCH_FORCE_IF_METADATA_DEGRADED
1029                                 : BCH_FORCE_IF_DATA_DEGRADED;
1030
1031                 if (dflags & ~flags) {
1032                         if (print) {
1033                                 char buf[100];
1034
1035                                 bch2_replicas_entry_to_text(&PBUF(buf), e);
1036                                 bch_err(c, "insufficient devices online (%u) for replicas entry %s",
1037                                         nr_online, buf);
1038                         }
1039                         ret = false;
1040                         break;
1041                 }
1042
1043         }
1044         percpu_up_read(&c->mark_lock);
1045
1046         return ret;
1047 }
1048
1049 unsigned bch2_dev_has_data(struct bch_fs *c, struct bch_dev *ca)
1050 {
1051         struct bch_replicas_entry *e;
1052         unsigned i, ret = 0;
1053
1054         percpu_down_read(&c->mark_lock);
1055
1056         for_each_cpu_replicas_entry(&c->replicas, e)
1057                 for (i = 0; i < e->nr_devs; i++)
1058                         if (e->devs[i] == ca->dev_idx)
1059                                 ret |= 1 << e->data_type;
1060
1061         percpu_up_read(&c->mark_lock);
1062
1063         return ret;
1064 }
1065
1066 int bch2_fs_replicas_init(struct bch_fs *c)
1067 {
1068         bch2_journal_entry_res_resize(&c->journal,
1069                         &c->replicas_journal_res,
1070                         reserve_journal_replicas(c, &c->replicas));
1071
1072         return replicas_table_update(c, &c->replicas);
1073 }