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/bcache.txt.
  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 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 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 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 j;
 130         char buf[80];
 131
 132         bch_extent_to_text(buf, sizeof(buf), k);
 133         printk(" %s", buf);
 134
 135         for (j = 0; j < KEY_PTRS(k); j++) {
 136                 size_t n = PTR_BUCKET_NR(b->c, k, j);
 137                 printk(" bucket %zu", n);
 138
 139                 if (n >= b->c->sb.first_bucket && n < b->c->sb.nbuckets)
 140                         printk(" prio %i",
 141                                PTR_BUCKET(b->c, k, j)->prio);
 142         }
 143
 144         printk(" %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         return __bch_btree_ptr_invalid(b->c, k);
 170 }
 171
 172 static bool btree_ptr_bad_expensive(struct btree *b, const struct bkey *k)
 173 {
 174         unsigned i;
 175         char buf[80];
 176         struct bucket *g;
 177
 178         if (mutex_trylock(&b->c->bucket_lock)) {
 179                 for (i = 0; i < KEY_PTRS(k); i++)
 180                         if (ptr_available(b->c, k, i)) {
 181                                 g = PTR_BUCKET(b->c, k, i);
 182
 183                                 if (KEY_DIRTY(k) ||
 184                                     g->prio != BTREE_PRIO ||
 185                                     (b->c->gc_mark_valid &&
 186                                      GC_MARK(g) != GC_MARK_METADATA))
 187                                         goto err;
 188                         }
 189
 190                 mutex_unlock(&b->c->bucket_lock);
 191         }
 192
 193         return false;
 194 err:
 195         mutex_unlock(&b->c->bucket_lock);
 196         bch_extent_to_text(buf, sizeof(buf), k);
 197         btree_bug(b,
 198 "inconsistent btree pointer %s: bucket %zi pin %i prio %i gen %i last_gc %i mark %llu",
 199                   buf, PTR_BUCKET_NR(b->c, k, i), atomic_read(&g->pin),
 200                   g->prio, g->gen, g->last_gc, GC_MARK(g));
 201         return true;
 202 }
 203
 204 static bool bch_btree_ptr_bad(struct btree_keys *bk, const struct bkey *k)
 205 {
 206         struct btree *b = container_of(bk, struct btree, keys);
 207         unsigned i;
 208
 209         if (!bkey_cmp(k, &ZERO_KEY) ||
 210             !KEY_PTRS(k) ||
 211             bch_ptr_invalid(bk, k))
 212                 return true;
 213
 214         for (i = 0; i < KEY_PTRS(k); i++)
 215                 if (!ptr_available(b->c, k, i) ||
 216                     ptr_stale(b->c, k, i))
 217                         return true;
 218
 219         if (expensive_debug_checks(b->c) &&
 220             btree_ptr_bad_expensive(b, k))
 221                 return true;
 222
 223         return false;
 224 }
 225
 226 static bool bch_btree_ptr_insert_fixup(struct btree_keys *bk,
 227                                        struct bkey *insert,
 228                                        struct btree_iter *iter,
 229                                        struct bkey *replace_key)
 230 {
 231         struct btree *b = container_of(bk, struct btree, keys);
 232
 233         if (!KEY_OFFSET(insert))
 234                 btree_current_write(b)->prio_blocked++;
 235
 236         return false;
 237 }
 238
 239 const struct btree_keys_ops bch_btree_keys_ops = {
 240         .sort_cmp       = bch_key_sort_cmp,
 241         .insert_fixup   = bch_btree_ptr_insert_fixup,
 242         .key_invalid    = bch_btree_ptr_invalid,
 243         .key_bad        = bch_btree_ptr_bad,
 244         .key_to_text    = bch_extent_to_text,
 245         .key_dump       = bch_bkey_dump,
 246 };
 247
 248 /* Extents */
 249
 250 /*
 251  * Returns true if l > r - unless l == r, in which case returns true if l is
 252  * older than r.
 253  *
 254  * Necessary for btree_sort_fixup() - if there are multiple keys that compare
 255  * equal in different sets, we have to process them newest to oldest.
 256  */
 257 static bool bch_extent_sort_cmp(struct btree_iter_set l,
 258                                 struct btree_iter_set r)
 259 {
 260         int64_t c = bkey_cmp(&START_KEY(l.k), &START_KEY(r.k));
 261
 262         return c ? c > 0 : l.k < r.k;
 263 }
 264
 265 static struct bkey *bch_extent_sort_fixup(struct btree_iter *iter,
 266                                           struct bkey *tmp)
 267 {
 268         while (iter->used > 1) {
 269                 struct btree_iter_set *top = iter->data, *i = top + 1;
 270
 271                 if (iter->used > 2 &&
 272                     bch_extent_sort_cmp(i[0], i[1]))
 273                         i++;
 274
 275                 if (bkey_cmp(top->k, &START_KEY(i->k)) <= 0)
 276                         break;
 277
 278                 if (!KEY_SIZE(i->k)) {
 279                         sort_key_next(iter, i);
 280                         heap_sift(iter, i - top, bch_extent_sort_cmp);
 281                         continue;
 282                 }
 283
 284                 if (top->k > i->k) {
 285                         if (bkey_cmp(top->k, i->k) >= 0)
 286                                 sort_key_next(iter, i);
 287                         else
 288                                 bch_cut_front(top->k, i->k);
 289
 290                         heap_sift(iter, i - top, bch_extent_sort_cmp);
 291                 } else {
 292                         /* can't happen because of comparison func */
 293                         BUG_ON(!bkey_cmp(&START_KEY(top->k), &START_KEY(i->k)));
 294
 295                         if (bkey_cmp(i->k, top->k) < 0) {
 296                                 bkey_copy(tmp, top->k);
 297
 298                                 bch_cut_back(&START_KEY(i->k), tmp);
 299                                 bch_cut_front(i->k, top->k);
 300                                 heap_sift(iter, 0, bch_extent_sort_cmp);
 301
 302                                 return tmp;
 303                         } else {
 304                                 bch_cut_back(&START_KEY(i->k), top->k);
 305                         }
 306                 }
 307         }
 308
 309         return NULL;
 310 }
 311
 312 static void bch_subtract_dirty(struct bkey *k,
 313                            struct cache_set *c,
 314                            uint64_t offset,
 315                            int sectors)
 316 {
 317         if (KEY_DIRTY(k))
 318                 bcache_dev_sectors_dirty_add(c, KEY_INODE(k),
 319                                              offset, -sectors);
 320 }
 321
 322 static bool bch_extent_insert_fixup(struct btree_keys *b,
 323                                     struct bkey *insert,
 324                                     struct btree_iter *iter,
 325                                     struct bkey *replace_key)
 326 {
 327         struct cache_set *c = container_of(b, struct btree, keys)->c;
 328
 329         uint64_t old_offset;
 330         unsigned old_size, sectors_found = 0;
 331
 332         BUG_ON(!KEY_OFFSET(insert));
 333         BUG_ON(!KEY_SIZE(insert));
 334
 335         while (1) {
 336                 struct bkey *k = bch_btree_iter_next(iter);
 337                 if (!k)
 338                         break;
 339
 340                 if (bkey_cmp(&START_KEY(k), insert) >= 0) {
 341                         if (KEY_SIZE(k))
 342                                 break;
 343                         else
 344                                 continue;
 345                 }
 346
 347                 if (bkey_cmp(k, &START_KEY(insert)) <= 0)
 348                         continue;
 349
 350                 old_offset = KEY_START(k);
 351                 old_size = KEY_SIZE(k);
 352
 353                 /*
 354                  * We might overlap with 0 size extents; we can't skip these
 355                  * because if they're in the set we're inserting to we have to
 356                  * adjust them so they don't overlap with the key we're
 357                  * inserting. But we don't want to check them for replace
 358                  * operations.
 359                  */
 360
 361                 if (replace_key && KEY_SIZE(k)) {
 362                         /*
 363                          * k might have been split since we inserted/found the
 364                          * key we're replacing
 365                          */
 366                         unsigned i;
 367                         uint64_t offset = KEY_START(k) -
 368                                 KEY_START(replace_key);
 369
 370                         /* But it must be a subset of the replace key */
 371                         if (KEY_START(k) < KEY_START(replace_key) ||
 372                             KEY_OFFSET(k) > KEY_OFFSET(replace_key))
 373                                 goto check_failed;
 374
 375                         /* We didn't find a key that we were supposed to */
 376                         if (KEY_START(k) > KEY_START(insert) + sectors_found)
 377                                 goto check_failed;
 378
 379                         if (!bch_bkey_equal_header(k, replace_key))
 380                                 goto check_failed;
 381
 382                         /* skip past gen */
 383                         offset <<= 8;
 384
 385                         BUG_ON(!KEY_PTRS(replace_key));
 386
 387                         for (i = 0; i < KEY_PTRS(replace_key); i++)
 388                                 if (k->ptr[i] != replace_key->ptr[i] + offset)
 389                                         goto check_failed;
 390
 391                         sectors_found = KEY_OFFSET(k) - KEY_START(insert);
 392                 }
 393
 394                 if (bkey_cmp(insert, k) < 0 &&
 395                     bkey_cmp(&START_KEY(insert), &START_KEY(k)) > 0) {
 396                         /*
 397                          * We overlapped in the middle of an existing key: that
 398                          * means we have to split the old key. But we have to do
 399                          * slightly different things depending on whether the
 400                          * old key has been written out yet.
 401                          */
 402
 403                         struct bkey *top;
 404
 405                         bch_subtract_dirty(k, c, KEY_START(insert),
 406                                        KEY_SIZE(insert));
 407
 408                         if (bkey_written(b, k)) {
 409                                 /*
 410                                  * We insert a new key to cover the top of the
 411                                  * old key, and the old key is modified in place
 412                                  * to represent the bottom split.
 413                                  *
 414                                  * It's completely arbitrary whether the new key
 415                                  * is the top or the bottom, but it has to match
 416                                  * up with what btree_sort_fixup() does - it
 417                                  * doesn't check for this kind of overlap, it
 418                                  * depends on us inserting a new key for the top
 419                                  * here.
 420                                  */
 421                                 top = bch_bset_search(b, bset_tree_last(b),
 422                                                       insert);
 423                                 bch_bset_insert(b, top, k);
 424                         } else {
 425                                 BKEY_PADDED(key) temp;
 426                                 bkey_copy(&temp.key, k);
 427                                 bch_bset_insert(b, k, &temp.key);
 428                                 top = bkey_next(k);
 429                         }
 430
 431                         bch_cut_front(insert, top);
 432                         bch_cut_back(&START_KEY(insert), k);
 433                         bch_bset_fix_invalidated_key(b, k);
 434                         goto out;
 435                 }
 436
 437                 if (bkey_cmp(insert, k) < 0) {
 438                         bch_cut_front(insert, k);
 439                 } else {
 440                         if (bkey_cmp(&START_KEY(insert), &START_KEY(k)) > 0)
 441                                 old_offset = KEY_START(insert);
 442
 443                         if (bkey_written(b, k) &&
 444                             bkey_cmp(&START_KEY(insert), &START_KEY(k)) <= 0) {
 445                                 /*
 446                                  * Completely overwrote, so we don't have to
 447                                  * invalidate the binary search tree
 448                                  */
 449                                 bch_cut_front(k, k);
 450                         } else {
 451                                 __bch_cut_back(&START_KEY(insert), k);
 452                                 bch_bset_fix_invalidated_key(b, k);
 453                         }
 454                 }
 455
 456                 bch_subtract_dirty(k, c, old_offset, old_size - KEY_SIZE(k));
 457         }
 458
 459 check_failed:
 460         if (replace_key) {
 461                 if (!sectors_found) {
 462                         return true;
 463                 } else if (sectors_found < KEY_SIZE(insert)) {
 464                         SET_KEY_OFFSET(insert, KEY_OFFSET(insert) -
 465                                        (KEY_SIZE(insert) - sectors_found));
 466                         SET_KEY_SIZE(insert, sectors_found);
 467                 }
 468         }
 469 out:
 470         if (KEY_DIRTY(insert))
 471                 bcache_dev_sectors_dirty_add(c, KEY_INODE(insert),
 472                                              KEY_START(insert),
 473                                              KEY_SIZE(insert));
 474
 475         return false;
 476 }
 477
 478 bool __bch_extent_invalid(struct cache_set *c, const struct bkey *k)
 479 {
 480         char buf[80];
 481
 482         if (!KEY_SIZE(k))
 483                 return true;
 484
 485         if (KEY_SIZE(k) > KEY_OFFSET(k))
 486                 goto bad;
 487
 488         if (__ptr_invalid(c, k))
 489                 goto bad;
 490
 491         return false;
 492 bad:
 493         bch_extent_to_text(buf, sizeof(buf), k);
 494         cache_bug(c, "spotted extent %s: %s", buf, bch_ptr_status(c, k));
 495         return true;
 496 }
 497
 498 static bool bch_extent_invalid(struct btree_keys *bk, const struct bkey *k)
 499 {
 500         struct btree *b = container_of(bk, struct btree, keys);
 501         return __bch_extent_invalid(b->c, k);
 502 }
 503
 504 static bool bch_extent_bad_expensive(struct btree *b, const struct bkey *k,
 505                                      unsigned ptr)
 506 {
 507         struct bucket *g = PTR_BUCKET(b->c, k, ptr);
 508         char buf[80];
 509
 510         if (mutex_trylock(&b->c->bucket_lock)) {
 511                 if (b->c->gc_mark_valid &&
 512                     (!GC_MARK(g) ||
 513                      GC_MARK(g) == GC_MARK_METADATA ||
 514                      (GC_MARK(g) != GC_MARK_DIRTY && KEY_DIRTY(k))))
 515                         goto err;
 516
 517                 if (g->prio == BTREE_PRIO)
 518                         goto err;
 519
 520                 mutex_unlock(&b->c->bucket_lock);
 521         }
 522
 523         return false;
 524 err:
 525         mutex_unlock(&b->c->bucket_lock);
 526         bch_extent_to_text(buf, sizeof(buf), k);
 527         btree_bug(b,
 528 "inconsistent extent pointer %s:\nbucket %zu pin %i prio %i gen %i last_gc %i mark %llu",
 529                   buf, PTR_BUCKET_NR(b->c, k, ptr), atomic_read(&g->pin),
 530                   g->prio, g->gen, g->last_gc, GC_MARK(g));
 531         return true;
 532 }
 533
 534 static bool bch_extent_bad(struct btree_keys *bk, const struct bkey *k)
 535 {
 536         struct btree *b = container_of(bk, struct btree, keys);
 537         struct bucket *g;
 538         unsigned i, stale;
 539
 540         if (!KEY_PTRS(k) ||
 541             bch_extent_invalid(bk, k))
 542                 return true;
 543
 544         for (i = 0; i < KEY_PTRS(k); i++)
 545                 if (!ptr_available(b->c, k, i))
 546                         return true;
 547
 548         if (!expensive_debug_checks(b->c) && KEY_DIRTY(k))
 549                 return false;
 550
 551         for (i = 0; i < KEY_PTRS(k); i++) {
 552                 g = PTR_BUCKET(b->c, k, i);
 553                 stale = ptr_stale(b->c, k, i);
 554
 555                 btree_bug_on(stale > 96, b,
 556                              "key too stale: %i, need_gc %u",
 557                              stale, b->c->need_gc);
 558
 559                 btree_bug_on(stale && KEY_DIRTY(k) && KEY_SIZE(k),
 560                              b, "stale dirty pointer");
 561
 562                 if (stale)
 563                         return true;
 564
 565                 if (expensive_debug_checks(b->c) &&
 566                     bch_extent_bad_expensive(b, k, i))
 567                         return true;
 568         }
 569
 570         return false;
 571 }
 572
 573 static uint64_t merge_chksums(struct bkey *l, struct bkey *r)
 574 {
 575         return (l->ptr[KEY_PTRS(l)] + r->ptr[KEY_PTRS(r)]) &
 576                 ~((uint64_t)1 << 63);
 577 }
 578
 579 static bool bch_extent_merge(struct btree_keys *bk, struct bkey *l, struct bkey *r)
 580 {
 581         struct btree *b = container_of(bk, struct btree, keys);
 582         unsigned i;
 583
 584         if (key_merging_disabled(b->c))
 585                 return false;
 586
 587         for (i = 0; i < KEY_PTRS(l); i++)
 588                 if (l->ptr[i] + MAKE_PTR(0, KEY_SIZE(l), 0) != r->ptr[i] ||
 589                     PTR_BUCKET_NR(b->c, l, i) != PTR_BUCKET_NR(b->c, r, i))
 590                         return false;
 591
 592         /* Keys with no pointers aren't restricted to one bucket and could
 593          * overflow KEY_SIZE
 594          */
 595         if (KEY_SIZE(l) + KEY_SIZE(r) > USHRT_MAX) {
 596                 SET_KEY_OFFSET(l, KEY_OFFSET(l) + USHRT_MAX - KEY_SIZE(l));
 597                 SET_KEY_SIZE(l, USHRT_MAX);
 598
 599                 bch_cut_front(l, r);
 600                 return false;
 601         }
 602
 603         if (KEY_CSUM(l)) {
 604                 if (KEY_CSUM(r))
 605                         l->ptr[KEY_PTRS(l)] = merge_chksums(l, r);
 606                 else
 607                         SET_KEY_CSUM(l, 0);
 608         }
 609
 610         SET_KEY_OFFSET(l, KEY_OFFSET(l) + KEY_SIZE(r));
 611         SET_KEY_SIZE(l, KEY_SIZE(l) + KEY_SIZE(r));
 612
 613         return true;
 614 }
 615
 616 const struct btree_keys_ops bch_extent_keys_ops = {
 617         .sort_cmp       = bch_extent_sort_cmp,
 618         .sort_fixup     = bch_extent_sort_fixup,
 619         .insert_fixup   = bch_extent_insert_fixup,
 620         .key_invalid    = bch_extent_invalid,
 621         .key_bad        = bch_extent_bad,
 622         .key_merge      = bch_extent_merge,
 623         .key_to_text    = bch_extent_to_text,
 624         .key_dump       = bch_bkey_dump,
 625         .is_extents     = true,
 626 };