drivers/md/bcache/extents.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2010 Kent Overstreet <kent.overstreet@gmail.com>
   4  *
   5  * Uses a block device as cache for other block devices; optimized for SSDs.
   6  * All allocation is done in buckets, which should match the erase block size
   7  * of the device.
   8  *
   9  * Buckets containing cached data are kept on a heap sorted by priority;
  10  * bucket priority is increased on cache hit, and periodically all the buckets
  11  * on the heap have their priority scaled down. This currently is just used as
  12  * an LRU but in the future should allow for more intelligent heuristics.
  13  *
  14  * Buckets have an 8 bit counter; freeing is accomplished by incrementing the
  15  * counter. Garbage collection is used to remove stale pointers.
  16  *
  17  * Indexing is done via a btree; nodes are not necessarily fully sorted, rather
  18  * as keys are inserted we only sort the pages that have not yet been written.
  19  * When garbage collection is run, we resort the entire node.
  20  *
  21  * All configuration is done via sysfs; see Documentation/admin-guide/bcache.rst.
  22  */
  23
  24 #include "bcache.h"
  25 #include "btree.h"
  26 #include "debug.h"
  27 #include "extents.h"
  28 #include "writeback.h"
  29
  30 static void sort_key_next(struct btree_iter *iter,
  31                           struct btree_iter_set *i)
  32 {
  33         i->k = bkey_next(i->k);
  34
  35         if (i->k == i->end)
  36                 *i = iter->data[--iter->used];
  37 }
  38
  39 static bool bch_key_sort_cmp(struct btree_iter_set l,
  40                              struct btree_iter_set r)
  41 {
  42         int64_t c = bkey_cmp(l.k, r.k);
  43
  44         return c ? c > 0 : l.k < r.k;
  45 }
  46
  47 static bool __ptr_invalid(struct cache_set *c, const struct bkey *k)
  48 {
  49         unsigned int i;
  50
  51         for (i = 0; i < KEY_PTRS(k); i++)
  52                 if (ptr_available(c, k, i)) {
  53                         struct cache *ca = PTR_CACHE(c, k, i);
  54                         size_t bucket = PTR_BUCKET_NR(c, k, i);
  55                         size_t r = bucket_remainder(c, PTR_OFFSET(k, i));
  56
  57                         if (KEY_SIZE(k) + r > c->sb.bucket_size ||
  58                             bucket <  ca->sb.first_bucket ||
  59                             bucket >= ca->sb.nbuckets)
  60                                 return true;
  61                 }
  62
  63         return false;
  64 }
  65
  66 /* Common among btree and extent ptrs */
  67
  68 static const char *bch_ptr_status(struct cache_set *c, const struct bkey *k)
  69 {
  70         unsigned int i;
  71
  72         for (i = 0; i < KEY_PTRS(k); i++)
  73                 if (ptr_available(c, k, i)) {
  74                         struct cache *ca = PTR_CACHE(c, k, i);
  75                         size_t bucket = PTR_BUCKET_NR(c, k, i);
  76                         size_t r = bucket_remainder(c, PTR_OFFSET(k, i));
  77
  78                         if (KEY_SIZE(k) + r > c->sb.bucket_size)
  79                                 return "bad, length too big";
  80                         if (bucket <  ca->sb.first_bucket)
  81                                 return "bad, short offset";
  82                         if (bucket >= ca->sb.nbuckets)
  83                                 return "bad, offset past end of device";
  84                         if (ptr_stale(c, k, i))
  85                                 return "stale";
  86                 }
  87
  88         if (!bkey_cmp(k, &ZERO_KEY))
  89                 return "bad, null key";
  90         if (!KEY_PTRS(k))
  91                 return "bad, no pointers";
  92         if (!KEY_SIZE(k))
  93                 return "zeroed key";
  94         return "";
  95 }
  96
  97 void bch_extent_to_text(char *buf, size_t size, const struct bkey *k)
  98 {
  99         unsigned int i = 0;
 100         char *out = buf, *end = buf + size;
 101
 102 #define p(...)  (out += scnprintf(out, end - out, __VA_ARGS__))
 103
 104         p("%llu:%llu len %llu -> [", KEY_INODE(k), KEY_START(k), KEY_SIZE(k));
 105
 106         for (i = 0; i < KEY_PTRS(k); i++) {
 107                 if (i)
 108                         p(", ");
 109
 110                 if (PTR_DEV(k, i) == PTR_CHECK_DEV)
 111                         p("check dev");
 112                 else
 113                         p("%llu:%llu gen %llu", PTR_DEV(k, i),
 114                           PTR_OFFSET(k, i), PTR_GEN(k, i));
 115         }
 116
 117         p("]");
 118
 119         if (KEY_DIRTY(k))
 120                 p(" dirty");
 121         if (KEY_CSUM(k))
 122                 p(" cs%llu %llx", KEY_CSUM(k), k->ptr[1]);
 123 #undef p
 124 }
 125
 126 static void bch_bkey_dump(struct btree_keys *keys, const struct bkey *k)
 127 {
 128         struct btree *b = container_of(keys, struct btree, keys);
 129         unsigned int j;
 130         char buf[80];
 131
 132         bch_extent_to_text(buf, sizeof(buf), k);
 133         pr_err(" %s", buf);
 134
 135         for (j = 0; j < KEY_PTRS(k); j++) {
 136                 size_t n = PTR_BUCKET_NR(b->c, k, j);
 137
 138                 pr_err(" bucket %zu", n);
 139                 if (n >= b->c->sb.first_bucket && n < b->c->sb.nbuckets)
 140                         pr_err(" prio %i",
 141                                PTR_BUCKET(b->c, k, j)->prio);
 142         }
 143
 144         pr_err(" %s\n", bch_ptr_status(b->c, k));
 145 }
 146
 147 /* Btree ptrs */
 148
 149 bool __bch_btree_ptr_invalid(struct cache_set *c, const struct bkey *k)
 150 {
 151         char buf[80];
 152
 153         if (!KEY_PTRS(k) || !KEY_SIZE(k) || KEY_DIRTY(k))
 154                 goto bad;
 155
 156         if (__ptr_invalid(c, k))
 157                 goto bad;
 158
 159         return false;
 160 bad:
 161         bch_extent_to_text(buf, sizeof(buf), k);
 162         cache_bug(c, "spotted btree ptr %s: %s", buf, bch_ptr_status(c, k));
 163         return true;
 164 }
 165
 166 static bool bch_btree_ptr_invalid(struct btree_keys *bk, const struct bkey *k)
 167 {
 168         struct btree *b = container_of(bk, struct btree, keys);
 169
 170         return __bch_btree_ptr_invalid(b->c, k);
 171 }
 172
 173 static bool btree_ptr_bad_expensive(struct btree *b, const struct bkey *k)
 174 {
 175         unsigned int i;
 176         char buf[80];
 177         struct bucket *g;
 178
 179         if (mutex_trylock(&b->c->bucket_lock)) {
 180                 for (i = 0; i < KEY_PTRS(k); i++)
 181                         if (ptr_available(b->c, k, i)) {
 182                                 g = PTR_BUCKET(b->c, k, i);
 183
 184                                 if (KEY_DIRTY(k) ||
 185                                     g->prio != BTREE_PRIO ||
 186                                     (b->c->gc_mark_valid &&
 187                                      GC_MARK(g) != GC_MARK_METADATA))
 188                                         goto err;
 189                         }
 190
 191                 mutex_unlock(&b->c->bucket_lock);
 192         }
 193
 194         return false;
 195 err:
 196         mutex_unlock(&b->c->bucket_lock);
 197         bch_extent_to_text(buf, sizeof(buf), k);
 198         btree_bug(b,
 199 "inconsistent btree pointer %s: bucket %zi pin %i prio %i gen %i last_gc %i mark %llu",
 200                   buf, PTR_BUCKET_NR(b->c, k, i), atomic_read(&g->pin),
 201                   g->prio, g->gen, g->last_gc, GC_MARK(g));
 202         return true;
 203 }
 204
 205 static bool bch_btree_ptr_bad(struct btree_keys *bk, const struct bkey *k)
 206 {
 207         struct btree *b = container_of(bk, struct btree, keys);
 208         unsigned int i;
 209
 210         if (!bkey_cmp(k, &ZERO_KEY) ||
 211             !KEY_PTRS(k) ||
 212             bch_ptr_invalid(bk, k))
 213                 return true;
 214
 215         for (i = 0; i < KEY_PTRS(k); i++)
 216                 if (!ptr_available(b->c, k, i) ||
 217                     ptr_stale(b->c, k, i))
 218                         return true;
 219
 220         if (expensive_debug_checks(b->c) &&
 221             btree_ptr_bad_expensive(b, k))
 222                 return true;
 223
 224         return false;
 225 }
 226
 227 static bool bch_btree_ptr_insert_fixup(struct btree_keys *bk,
 228                                        struct bkey *insert,
 229                                        struct btree_iter *iter,
 230                                        struct bkey *replace_key)
 231 {
 232         struct btree *b = container_of(bk, struct btree, keys);
 233
 234         if (!KEY_OFFSET(insert))
 235                 btree_current_write(b)->prio_blocked++;
 236
 237         return false;
 238 }
 239
 240 const struct btree_keys_ops bch_btree_keys_ops = {
 241         .sort_cmp       = bch_key_sort_cmp,
 242         .insert_fixup   = bch_btree_ptr_insert_fixup,
 243         .key_invalid    = bch_btree_ptr_invalid,
 244         .key_bad        = bch_btree_ptr_bad,
 245         .key_to_text    = bch_extent_to_text,
 246         .key_dump       = bch_bkey_dump,
 247 };
 248
 249 /* Extents */
 250
 251 /*
 252  * Returns true if l > r - unless l == r, in which case returns true if l is
 253  * older than r.
 254  *
 255  * Necessary for btree_sort_fixup() - if there are multiple keys that compare
 256  * equal in different sets, we have to process them newest to oldest.
 257  */
 258 static bool bch_extent_sort_cmp(struct btree_iter_set l,
 259                                 struct btree_iter_set r)
 260 {
 261         int64_t c = bkey_cmp(&START_KEY(l.k), &START_KEY(r.k));
 262
 263         return c ? c > 0 : l.k < r.k;
 264 }
 265
 266 static struct bkey *bch_extent_sort_fixup(struct btree_iter *iter,
 267                                           struct bkey *tmp)
 268 {
 269         while (iter->used > 1) {
 270                 struct btree_iter_set *top = iter->data, *i = top + 1;
 271
 272                 if (iter->used > 2 &&
 273                     bch_extent_sort_cmp(i[0], i[1]))
 274                         i++;
 275
 276                 if (bkey_cmp(top->k, &START_KEY(i->k)) <= 0)
 277                         break;
 278
 279                 if (!KEY_SIZE(i->k)) {
 280                         sort_key_next(iter, i);
 281                         heap_sift(iter, i - top, bch_extent_sort_cmp);
 282                         continue;
 283                 }
 284
 285                 if (top->k > i->k) {
 286                         if (bkey_cmp(top->k, i->k) >= 0)
 287                                 sort_key_next(iter, i);
 288                         else
 289                                 bch_cut_front(top->k, i->k);
 290
 291                         heap_sift(iter, i - top, bch_extent_sort_cmp);
 292                 } else {
 293                         /* can't happen because of comparison func */
 294                         BUG_ON(!bkey_cmp(&START_KEY(top->k), &START_KEY(i->k)));
 295
 296                         if (bkey_cmp(i->k, top->k) < 0) {
 297                                 bkey_copy(tmp, top->k);
 298
 299                                 bch_cut_back(&START_KEY(i->k), tmp);
 300                                 bch_cut_front(i->k, top->k);
 301                                 heap_sift(iter, 0, bch_extent_sort_cmp);
 302
 303                                 return tmp;
 304                         } else {
 305                                 bch_cut_back(&START_KEY(i->k), top->k);
 306                         }
 307                 }
 308         }
 309
 310         return NULL;
 311 }
 312
 313 static void bch_subtract_dirty(struct bkey *k,
 314                            struct cache_set *c,
 315                            uint64_t offset,
 316                            int sectors)
 317 {
 318         if (KEY_DIRTY(k))
 319                 bcache_dev_sectors_dirty_add(c, KEY_INODE(k),
 320                                              offset, -sectors);
 321 }
 322
 323 static bool bch_extent_insert_fixup(struct btree_keys *b,
 324                                     struct bkey *insert,
 325                                     struct btree_iter *iter,
 326                                     struct bkey *replace_key)
 327 {
 328         struct cache_set *c = container_of(b, struct btree, keys)->c;
 329
 330         uint64_t old_offset;
 331         unsigned int old_size, sectors_found = 0;
 332
 333         BUG_ON(!KEY_OFFSET(insert));
 334         BUG_ON(!KEY_SIZE(insert));
 335
 336         while (1) {
 337                 struct bkey *k = bch_btree_iter_next(iter);
 338
 339                 if (!k)
 340                         break;
 341
 342                 if (bkey_cmp(&START_KEY(k), insert) >= 0) {
 343                         if (KEY_SIZE(k))
 344                                 break;
 345                         else
 346                                 continue;
 347                 }
 348
 349                 if (bkey_cmp(k, &START_KEY(insert)) <= 0)
 350                         continue;
 351
 352                 old_offset = KEY_START(k);
 353                 old_size = KEY_SIZE(k);
 354
 355                 /*
 356                  * We might overlap with 0 size extents; we can't skip these
 357                  * because if they're in the set we're inserting to we have to
 358                  * adjust them so they don't overlap with the key we're
 359                  * inserting. But we don't want to check them for replace
 360                  * operations.
 361                  */
 362
 363                 if (replace_key && KEY_SIZE(k)) {
 364                         /*
 365                          * k might have been split since we inserted/found the
 366                          * key we're replacing
 367                          */
 368                         unsigned int i;
 369                         uint64_t offset = KEY_START(k) -
 370                                 KEY_START(replace_key);
 371
 372                         /* But it must be a subset of the replace key */
 373                         if (KEY_START(k) < KEY_START(replace_key) ||
 374                             KEY_OFFSET(k) > KEY_OFFSET(replace_key))
 375                                 goto check_failed;
 376
 377                         /* We didn't find a key that we were supposed to */
 378                         if (KEY_START(k) > KEY_START(insert) + sectors_found)
 379                                 goto check_failed;
 380
 381                         if (!bch_bkey_equal_header(k, replace_key))
 382                                 goto check_failed;
 383
 384                         /* skip past gen */
 385                         offset <<= 8;
 386
 387                         BUG_ON(!KEY_PTRS(replace_key));
 388
 389                         for (i = 0; i < KEY_PTRS(replace_key); i++)
 390                                 if (k->ptr[i] != replace_key->ptr[i] + offset)
 391                                         goto check_failed;
 392
 393                         sectors_found = KEY_OFFSET(k) - KEY_START(insert);
 394                 }
 395
 396                 if (bkey_cmp(insert, k) < 0 &&
 397                     bkey_cmp(&START_KEY(insert), &START_KEY(k)) > 0) {
 398                         /*
 399                          * We overlapped in the middle of an existing key: that
 400                          * means we have to split the old key. But we have to do
 401                          * slightly different things depending on whether the
 402                          * old key has been written out yet.
 403                          */
 404
 405                         struct bkey *top;
 406
 407                         bch_subtract_dirty(k, c, KEY_START(insert),
 408                                        KEY_SIZE(insert));
 409
 410                         if (bkey_written(b, k)) {
 411                                 /*
 412                                  * We insert a new key to cover the top of the
 413                                  * old key, and the old key is modified in place
 414                                  * to represent the bottom split.
 415                                  *
 416                                  * It's completely arbitrary whether the new key
 417                                  * is the top or the bottom, but it has to match
 418                                  * up with what btree_sort_fixup() does - it
 419                                  * doesn't check for this kind of overlap, it
 420                                  * depends on us inserting a new key for the top
 421                                  * here.
 422                                  */
 423                                 top = bch_bset_search(b, bset_tree_last(b),
 424                                                       insert);
 425                                 bch_bset_insert(b, top, k);
 426                         } else {
 427                                 BKEY_PADDED(key) temp;
 428                                 bkey_copy(&temp.key, k);
 429                                 bch_bset_insert(b, k, &temp.key);
 430                                 top = bkey_next(k);
 431                         }
 432
 433                         bch_cut_front(insert, top);
 434                         bch_cut_back(&START_KEY(insert), k);
 435                         bch_bset_fix_invalidated_key(b, k);
 436                         goto out;
 437                 }
 438
 439                 if (bkey_cmp(insert, k) < 0) {
 440                         bch_cut_front(insert, k);
 441                 } else {
 442                         if (bkey_cmp(&START_KEY(insert), &START_KEY(k)) > 0)
 443                                 old_offset = KEY_START(insert);
 444
 445                         if (bkey_written(b, k) &&
 446                             bkey_cmp(&START_KEY(insert), &START_KEY(k)) <= 0) {
 447                                 /*
 448                                  * Completely overwrote, so we don't have to
 449                                  * invalidate the binary search tree
 450                                  */
 451                                 bch_cut_front(k, k);
 452                         } else {
 453                                 __bch_cut_back(&START_KEY(insert), k);
 454                                 bch_bset_fix_invalidated_key(b, k);
 455                         }
 456                 }
 457
 458                 bch_subtract_dirty(k, c, old_offset, old_size - KEY_SIZE(k));
 459         }
 460
 461 check_failed:
 462         if (replace_key) {
 463                 if (!sectors_found) {
 464                         return true;
 465                 } else if (sectors_found < KEY_SIZE(insert)) {
 466                         SET_KEY_OFFSET(insert, KEY_OFFSET(insert) -
 467                                        (KEY_SIZE(insert) - sectors_found));
 468                         SET_KEY_SIZE(insert, sectors_found);
 469                 }
 470         }
 471 out:
 472         if (KEY_DIRTY(insert))
 473                 bcache_dev_sectors_dirty_add(c, KEY_INODE(insert),
 474                                              KEY_START(insert),
 475                                              KEY_SIZE(insert));
 476
 477         return false;
 478 }
 479
 480 bool __bch_extent_invalid(struct cache_set *c, const struct bkey *k)
 481 {
 482         char buf[80];
 483
 484         if (!KEY_SIZE(k))
 485                 return true;
 486
 487         if (KEY_SIZE(k) > KEY_OFFSET(k))
 488                 goto bad;
 489
 490         if (__ptr_invalid(c, k))
 491                 goto bad;
 492
 493         return false;
 494 bad:
 495         bch_extent_to_text(buf, sizeof(buf), k);
 496         cache_bug(c, "spotted extent %s: %s", buf, bch_ptr_status(c, k));
 497         return true;
 498 }
 499
 500 static bool bch_extent_invalid(struct btree_keys *bk, const struct bkey *k)
 501 {
 502         struct btree *b = container_of(bk, struct btree, keys);
 503
 504         return __bch_extent_invalid(b->c, k);
 505 }
 506
 507 static bool bch_extent_bad_expensive(struct btree *b, const struct bkey *k,
 508                                      unsigned int ptr)
 509 {
 510         struct bucket *g = PTR_BUCKET(b->c, k, ptr);
 511         char buf[80];
 512
 513         if (mutex_trylock(&b->c->bucket_lock)) {
 514                 if (b->c->gc_mark_valid &&
 515                     (!GC_MARK(g) ||
 516                      GC_MARK(g) == GC_MARK_METADATA ||
 517                      (GC_MARK(g) != GC_MARK_DIRTY && KEY_DIRTY(k))))
 518                         goto err;
 519
 520                 if (g->prio == BTREE_PRIO)
 521                         goto err;
 522
 523                 mutex_unlock(&b->c->bucket_lock);
 524         }
 525
 526         return false;
 527 err:
 528         mutex_unlock(&b->c->bucket_lock);
 529         bch_extent_to_text(buf, sizeof(buf), k);
 530         btree_bug(b,
 531 "inconsistent extent pointer %s:\nbucket %zu pin %i prio %i gen %i last_gc %i mark %llu",
 532                   buf, PTR_BUCKET_NR(b->c, k, ptr), atomic_read(&g->pin),
 533                   g->prio, g->gen, g->last_gc, GC_MARK(g));
 534         return true;
 535 }
 536
 537 static bool bch_extent_bad(struct btree_keys *bk, const struct bkey *k)
 538 {
 539         struct btree *b = container_of(bk, struct btree, keys);
 540         unsigned int i, stale;
 541         char buf[80];
 542
 543         if (!KEY_PTRS(k) ||
 544             bch_extent_invalid(bk, k))
 545                 return true;
 546
 547         for (i = 0; i < KEY_PTRS(k); i++)
 548                 if (!ptr_available(b->c, k, i))
 549                         return true;
 550
 551         for (i = 0; i < KEY_PTRS(k); i++) {
 552                 stale = ptr_stale(b->c, k, i);
 553
 554                 if (stale && KEY_DIRTY(k)) {
 555                         bch_extent_to_text(buf, sizeof(buf), k);
 556                         pr_info("stale dirty pointer, stale %u, key: %s",
 557                                 stale, buf);
 558                 }
 559
 560                 btree_bug_on(stale > BUCKET_GC_GEN_MAX, b,
 561                              "key too stale: %i, need_gc %u",
 562                              stale, b->c->need_gc);
 563
 564                 if (stale)
 565                         return true;
 566
 567                 if (expensive_debug_checks(b->c) &&
 568                     bch_extent_bad_expensive(b, k, i))
 569                         return true;
 570         }
 571
 572         return false;
 573 }
 574
 575 static uint64_t merge_chksums(struct bkey *l, struct bkey *r)
 576 {
 577         return (l->ptr[KEY_PTRS(l)] + r->ptr[KEY_PTRS(r)]) &
 578                 ~((uint64_t)1 << 63);
 579 }
 580
 581 static bool bch_extent_merge(struct btree_keys *bk,
 582                              struct bkey *l,
 583                              struct bkey *r)
 584 {
 585         struct btree *b = container_of(bk, struct btree, keys);
 586         unsigned int i;
 587
 588         if (key_merging_disabled(b->c))
 589                 return false;
 590
 591         for (i = 0; i < KEY_PTRS(l); i++)
 592                 if (l->ptr[i] + MAKE_PTR(0, KEY_SIZE(l), 0) != r->ptr[i] ||
 593                     PTR_BUCKET_NR(b->c, l, i) != PTR_BUCKET_NR(b->c, r, i))
 594                         return false;
 595
 596         /* Keys with no pointers aren't restricted to one bucket and could
 597          * overflow KEY_SIZE
 598          */
 599         if (KEY_SIZE(l) + KEY_SIZE(r) > USHRT_MAX) {
 600                 SET_KEY_OFFSET(l, KEY_OFFSET(l) + USHRT_MAX - KEY_SIZE(l));
 601                 SET_KEY_SIZE(l, USHRT_MAX);
 602
 603                 bch_cut_front(l, r);
 604                 return false;
 605         }
 606
 607         if (KEY_CSUM(l)) {
 608                 if (KEY_CSUM(r))
 609                         l->ptr[KEY_PTRS(l)] = merge_chksums(l, r);
 610                 else
 611                         SET_KEY_CSUM(l, 0);
 612         }
 613
 614         SET_KEY_OFFSET(l, KEY_OFFSET(l) + KEY_SIZE(r));
 615         SET_KEY_SIZE(l, KEY_SIZE(l) + KEY_SIZE(r));
 616
 617         return true;
 618 }
 619
 620 const struct btree_keys_ops bch_extent_keys_ops = {
 621         .sort_cmp       = bch_extent_sort_cmp,
 622         .sort_fixup     = bch_extent_sort_fixup,
 623         .insert_fixup   = bch_extent_insert_fixup,
 624         .key_invalid    = bch_extent_invalid,
 625         .key_bad        = bch_extent_bad,
 626         .key_merge      = bch_extent_merge,
 627         .key_to_text    = bch_extent_to_text,
 628         .key_dump       = bch_bkey_dump,
 629         .is_extents     = true,
 630 };