drivers/md/bcache/request.c

   1 /*
   2  * Main bcache entry point - handle a read or a write request and decide what to
   3  * do with it; the make_request functions are called by the block layer.
   4  *
   5  * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
   6  * Copyright 2012 Google, Inc.
   7  */
   8
   9 #include "bcache.h"
  10 #include "btree.h"
  11 #include "debug.h"
  12 #include "request.h"
  13 #include "writeback.h"
  14
  15 #include <linux/module.h>
  16 #include <linux/hash.h>
  17 #include <linux/random.h>
  18 #include <linux/backing-dev.h>
  19
  20 #include <trace/events/bcache.h>
  21
  22 #define CUTOFF_CACHE_ADD        95
  23 #define CUTOFF_CACHE_READA      90
  24
  25 struct kmem_cache *bch_search_cache;
  26
  27 static void bch_data_insert_start(struct closure *);
  28
  29 static unsigned cache_mode(struct cached_dev *dc, struct bio *bio)
  30 {
  31         return BDEV_CACHE_MODE(&dc->sb);
  32 }
  33
  34 static bool verify(struct cached_dev *dc, struct bio *bio)
  35 {
  36         return dc->verify;
  37 }
  38
  39 static void bio_csum(struct bio *bio, struct bkey *k)
  40 {
  41         struct bio_vec bv;
  42         struct bvec_iter iter;
  43         uint64_t csum = 0;
  44
  45         bio_for_each_segment(bv, bio, iter) {
  46                 void *d = kmap(bv.bv_page) + bv.bv_offset;
  47                 csum = bch_crc64_update(csum, d, bv.bv_len);
  48                 kunmap(bv.bv_page);
  49         }
  50
  51         k->ptr[KEY_PTRS(k)] = csum & (~0ULL >> 1);
  52 }
  53
  54 /* Insert data into cache */
  55
  56 static void bch_data_insert_keys(struct closure *cl)
  57 {
  58         struct data_insert_op *op = container_of(cl, struct data_insert_op, cl);
  59         atomic_t *journal_ref = NULL;
  60         struct bkey *replace_key = op->replace ? &op->replace_key : NULL;
  61         int ret;
  62
  63         /*
  64          * If we're looping, might already be waiting on
  65          * another journal write - can't wait on more than one journal write at
  66          * a time
  67          *
  68          * XXX: this looks wrong
  69          */
  70 #if 0
  71         while (atomic_read(&s->cl.remaining) & CLOSURE_WAITING)
  72                 closure_sync(&s->cl);
  73 #endif
  74
  75         if (!op->replace)
  76                 journal_ref = bch_journal(op->c, &op->insert_keys,
  77                                           op->flush_journal ? cl : NULL);
  78
  79         ret = bch_btree_insert(op->c, &op->insert_keys,
  80                                journal_ref, replace_key);
  81         if (ret == -ESRCH) {
  82                 op->replace_collision = true;
  83         } else if (ret) {
  84                 op->error               = -ENOMEM;
  85                 op->insert_data_done    = true;
  86         }
  87
  88         if (journal_ref)
  89                 atomic_dec_bug(journal_ref);
  90
  91         if (!op->insert_data_done)
  92                 continue_at(cl, bch_data_insert_start, op->wq);
  93
  94         bch_keylist_free(&op->insert_keys);
  95         closure_return(cl);
  96 }
  97
  98 static int bch_keylist_realloc(struct keylist *l, unsigned u64s,
  99                                struct cache_set *c)
 100 {
 101         size_t oldsize = bch_keylist_nkeys(l);
 102         size_t newsize = oldsize + u64s;
 103
 104         /*
 105          * The journalling code doesn't handle the case where the keys to insert
 106          * is bigger than an empty write: If we just return -ENOMEM here,
 107          * bio_insert() and bio_invalidate() will insert the keys created so far
 108          * and finish the rest when the keylist is empty.
 109          */
 110         if (newsize * sizeof(uint64_t) > block_bytes(c) - sizeof(struct jset))
 111                 return -ENOMEM;
 112
 113         return __bch_keylist_realloc(l, u64s);
 114 }
 115
 116 static void bch_data_invalidate(struct closure *cl)
 117 {
 118         struct data_insert_op *op = container_of(cl, struct data_insert_op, cl);
 119         struct bio *bio = op->bio;
 120
 121         pr_debug("invalidating %i sectors from %llu",
 122                  bio_sectors(bio), (uint64_t) bio->bi_iter.bi_sector);
 123
 124         while (bio_sectors(bio)) {
 125                 unsigned sectors = min(bio_sectors(bio),
 126                                        1U << (KEY_SIZE_BITS - 1));
 127
 128                 if (bch_keylist_realloc(&op->insert_keys, 2, op->c))
 129                         goto out;
 130
 131                 bio->bi_iter.bi_sector  += sectors;
 132                 bio->bi_iter.bi_size    -= sectors << 9;
 133
 134                 bch_keylist_add(&op->insert_keys,
 135                                 &KEY(op->inode, bio->bi_iter.bi_sector, sectors));
 136         }
 137
 138         op->insert_data_done = true;
 139         bio_put(bio);
 140 out:
 141         continue_at(cl, bch_data_insert_keys, op->wq);
 142 }
 143
 144 static void bch_data_insert_error(struct closure *cl)
 145 {
 146         struct data_insert_op *op = container_of(cl, struct data_insert_op, cl);
 147
 148         /*
 149          * Our data write just errored, which means we've got a bunch of keys to
 150          * insert that point to data that wasn't succesfully written.
 151          *
 152          * We don't have to insert those keys but we still have to invalidate
 153          * that region of the cache - so, if we just strip off all the pointers
 154          * from the keys we'll accomplish just that.
 155          */
 156
 157         struct bkey *src = op->insert_keys.keys, *dst = op->insert_keys.keys;
 158
 159         while (src != op->insert_keys.top) {
 160                 struct bkey *n = bkey_next(src);
 161
 162                 SET_KEY_PTRS(src, 0);
 163                 memmove(dst, src, bkey_bytes(src));
 164
 165                 dst = bkey_next(dst);
 166                 src = n;
 167         }
 168
 169         op->insert_keys.top = dst;
 170
 171         bch_data_insert_keys(cl);
 172 }
 173
 174 static void bch_data_insert_endio(struct bio *bio, int error)
 175 {
 176         struct closure *cl = bio->bi_private;
 177         struct data_insert_op *op = container_of(cl, struct data_insert_op, cl);
 178
 179         if (error) {
 180                 /* TODO: We could try to recover from this. */
 181                 if (op->writeback)
 182                         op->error = error;
 183                 else if (!op->replace)
 184                         set_closure_fn(cl, bch_data_insert_error, op->wq);
 185                 else
 186                         set_closure_fn(cl, NULL, NULL);
 187         }
 188
 189         bch_bbio_endio(op->c, bio, error, "writing data to cache");
 190 }
 191
 192 static void bch_data_insert_start(struct closure *cl)
 193 {
 194         struct data_insert_op *op = container_of(cl, struct data_insert_op, cl);
 195         struct bio *bio = op->bio, *n;
 196
 197         if (atomic_sub_return(bio_sectors(bio), &op->c->sectors_to_gc) < 0) {
 198                 set_gc_sectors(op->c);
 199                 wake_up_gc(op->c);
 200         }
 201
 202         if (op->bypass)
 203                 return bch_data_invalidate(cl);
 204
 205         /*
 206          * Journal writes are marked REQ_FLUSH; if the original write was a
 207          * flush, it'll wait on the journal write.
 208          */
 209         bio->bi_rw &= ~(REQ_FLUSH|REQ_FUA);
 210
 211         do {
 212                 unsigned i;
 213                 struct bkey *k;
 214                 struct bio_set *split = op->c->bio_split;
 215
 216                 /* 1 for the device pointer and 1 for the chksum */
 217                 if (bch_keylist_realloc(&op->insert_keys,
 218                                         3 + (op->csum ? 1 : 0),
 219                                         op->c))
 220                         continue_at(cl, bch_data_insert_keys, op->wq);
 221
 222                 k = op->insert_keys.top;
 223                 bkey_init(k);
 224                 SET_KEY_INODE(k, op->inode);
 225                 SET_KEY_OFFSET(k, bio->bi_iter.bi_sector);
 226
 227                 if (!bch_alloc_sectors(op->c, k, bio_sectors(bio),
 228                                        op->write_point, op->write_prio,
 229                                        op->writeback))
 230                         goto err;
 231
 232                 n = bio_next_split(bio, KEY_SIZE(k), GFP_NOIO, split);
 233
 234                 n->bi_end_io    = bch_data_insert_endio;
 235                 n->bi_private   = cl;
 236
 237                 if (op->writeback) {
 238                         SET_KEY_DIRTY(k, true);
 239
 240                         for (i = 0; i < KEY_PTRS(k); i++)
 241                                 SET_GC_MARK(PTR_BUCKET(op->c, k, i),
 242                                             GC_MARK_DIRTY);
 243                 }
 244
 245                 SET_KEY_CSUM(k, op->csum);
 246                 if (KEY_CSUM(k))
 247                         bio_csum(n, k);
 248
 249                 trace_bcache_cache_insert(k);
 250                 bch_keylist_push(&op->insert_keys);
 251
 252                 n->bi_rw |= REQ_WRITE;
 253                 bch_submit_bbio(n, op->c, k, 0);
 254         } while (n != bio);
 255
 256         op->insert_data_done = true;
 257         continue_at(cl, bch_data_insert_keys, op->wq);
 258 err:
 259         /* bch_alloc_sectors() blocks if s->writeback = true */
 260         BUG_ON(op->writeback);
 261
 262         /*
 263          * But if it's not a writeback write we'd rather just bail out if
 264          * there aren't any buckets ready to write to - it might take awhile and
 265          * we might be starving btree writes for gc or something.
 266          */
 267
 268         if (!op->replace) {
 269                 /*
 270                  * Writethrough write: We can't complete the write until we've
 271                  * updated the index. But we don't want to delay the write while
 272                  * we wait for buckets to be freed up, so just invalidate the
 273                  * rest of the write.
 274                  */
 275                 op->bypass = true;
 276                 return bch_data_invalidate(cl);
 277         } else {
 278                 /*
 279                  * From a cache miss, we can just insert the keys for the data
 280                  * we have written or bail out if we didn't do anything.
 281                  */
 282                 op->insert_data_done = true;
 283                 bio_put(bio);
 284
 285                 if (!bch_keylist_empty(&op->insert_keys))
 286                         continue_at(cl, bch_data_insert_keys, op->wq);
 287                 else
 288                         closure_return(cl);
 289         }
 290 }
 291
 292 /**
 293  * bch_data_insert - stick some data in the cache
 294  *
 295  * This is the starting point for any data to end up in a cache device; it could
 296  * be from a normal write, or a writeback write, or a write to a flash only
 297  * volume - it's also used by the moving garbage collector to compact data in
 298  * mostly empty buckets.
 299  *
 300  * It first writes the data to the cache, creating a list of keys to be inserted
 301  * (if the data had to be fragmented there will be multiple keys); after the
 302  * data is written it calls bch_journal, and after the keys have been added to
 303  * the next journal write they're inserted into the btree.
 304  *
 305  * It inserts the data in s->cache_bio; bi_sector is used for the key offset,
 306  * and op->inode is used for the key inode.
 307  *
 308  * If s->bypass is true, instead of inserting the data it invalidates the
 309  * region of the cache represented by s->cache_bio and op->inode.
 310  */
 311 void bch_data_insert(struct closure *cl)
 312 {
 313         struct data_insert_op *op = container_of(cl, struct data_insert_op, cl);
 314
 315         trace_bcache_write(op->c, op->inode, op->bio,
 316                            op->writeback, op->bypass);
 317
 318         bch_keylist_init(&op->insert_keys);
 319         bio_get(op->bio);
 320         bch_data_insert_start(cl);
 321 }
 322
 323 /* Congested? */
 324
 325 unsigned bch_get_congested(struct cache_set *c)
 326 {
 327         int i;
 328         long rand;
 329
 330         if (!c->congested_read_threshold_us &&
 331             !c->congested_write_threshold_us)
 332                 return 0;
 333
 334         i = (local_clock_us() - c->congested_last_us) / 1024;
 335         if (i < 0)
 336                 return 0;
 337
 338         i += atomic_read(&c->congested);
 339         if (i >= 0)
 340                 return 0;
 341
 342         i += CONGESTED_MAX;
 343
 344         if (i > 0)
 345                 i = fract_exp_two(i, 6);
 346
 347         rand = get_random_int();
 348         i -= bitmap_weight(&rand, BITS_PER_LONG);
 349
 350         return i > 0 ? i : 1;
 351 }
 352
 353 static void add_sequential(struct task_struct *t)
 354 {
 355         ewma_add(t->sequential_io_avg,
 356                  t->sequential_io, 8, 0);
 357
 358         t->sequential_io = 0;
 359 }
 360
 361 static struct hlist_head *iohash(struct cached_dev *dc, uint64_t k)
 362 {
 363         return &dc->io_hash[hash_64(k, RECENT_IO_BITS)];
 364 }
 365
 366 static bool check_should_bypass(struct cached_dev *dc, struct bio *bio)
 367 {
 368         struct cache_set *c = dc->disk.c;
 369         unsigned mode = cache_mode(dc, bio);
 370         unsigned sectors, congested = bch_get_congested(c);
 371         struct task_struct *task = current;
 372         struct io *i;
 373
 374         if (test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags) ||
 375             c->gc_stats.in_use > CUTOFF_CACHE_ADD ||
 376             (bio->bi_rw & REQ_DISCARD))
 377                 goto skip;
 378
 379         if (mode == CACHE_MODE_NONE ||
 380             (mode == CACHE_MODE_WRITEAROUND &&
 381              (bio->bi_rw & REQ_WRITE)))
 382                 goto skip;
 383
 384         if (bio->bi_iter.bi_sector & (c->sb.block_size - 1) ||
 385             bio_sectors(bio) & (c->sb.block_size - 1)) {
 386                 pr_debug("skipping unaligned io");
 387                 goto skip;
 388         }
 389
 390         if (bypass_torture_test(dc)) {
 391                 if ((get_random_int() & 3) == 3)
 392                         goto skip;
 393                 else
 394                         goto rescale;
 395         }
 396
 397         if (!congested && !dc->sequential_cutoff)
 398                 goto rescale;
 399
 400         if (!congested &&
 401             mode == CACHE_MODE_WRITEBACK &&
 402             (bio->bi_rw & REQ_WRITE) &&
 403             (bio->bi_rw & REQ_SYNC))
 404                 goto rescale;
 405
 406         spin_lock(&dc->io_lock);
 407
 408         hlist_for_each_entry(i, iohash(dc, bio->bi_iter.bi_sector), hash)
 409                 if (i->last == bio->bi_iter.bi_sector &&
 410                     time_before(jiffies, i->jiffies))
 411                         goto found;
 412
 413         i = list_first_entry(&dc->io_lru, struct io, lru);
 414
 415         add_sequential(task);
 416         i->sequential = 0;
 417 found:
 418         if (i->sequential + bio->bi_iter.bi_size > i->sequential)
 419                 i->sequential   += bio->bi_iter.bi_size;
 420
 421         i->last                  = bio_end_sector(bio);
 422         i->jiffies               = jiffies + msecs_to_jiffies(5000);
 423         task->sequential_io      = i->sequential;
 424
 425         hlist_del(&i->hash);
 426         hlist_add_head(&i->hash, iohash(dc, i->last));
 427         list_move_tail(&i->lru, &dc->io_lru);
 428
 429         spin_unlock(&dc->io_lock);
 430
 431         sectors = max(task->sequential_io,
 432                       task->sequential_io_avg) >> 9;
 433
 434         if (dc->sequential_cutoff &&
 435             sectors >= dc->sequential_cutoff >> 9) {
 436                 trace_bcache_bypass_sequential(bio);
 437                 goto skip;
 438         }
 439
 440         if (congested && sectors >= congested) {
 441                 trace_bcache_bypass_congested(bio);
 442                 goto skip;
 443         }
 444
 445 rescale:
 446         bch_rescale_priorities(c, bio_sectors(bio));
 447         return false;
 448 skip:
 449         bch_mark_sectors_bypassed(c, dc, bio_sectors(bio));
 450         return true;
 451 }
 452
 453 /* Cache lookup */
 454
 455 struct search {
 456         /* Stack frame for bio_complete */
 457         struct closure          cl;
 458
 459         struct bbio             bio;
 460         struct bio              *orig_bio;
 461         struct bio              *cache_miss;
 462         struct bcache_device    *d;
 463
 464         unsigned                insert_bio_sectors;
 465         unsigned                recoverable:1;
 466         unsigned                write:1;
 467         unsigned                read_dirty_data:1;
 468
 469         unsigned long           start_time;
 470
 471         struct btree_op         op;
 472         struct data_insert_op   iop;
 473 };
 474
 475 static void bch_cache_read_endio(struct bio *bio, int error)
 476 {
 477         struct bbio *b = container_of(bio, struct bbio, bio);
 478         struct closure *cl = bio->bi_private;
 479         struct search *s = container_of(cl, struct search, cl);
 480
 481         /*
 482          * If the bucket was reused while our bio was in flight, we might have
 483          * read the wrong data. Set s->error but not error so it doesn't get
 484          * counted against the cache device, but we'll still reread the data
 485          * from the backing device.
 486          */
 487
 488         if (error)
 489                 s->iop.error = error;
 490         else if (!KEY_DIRTY(&b->key) &&
 491                  ptr_stale(s->iop.c, &b->key, 0)) {
 492                 atomic_long_inc(&s->iop.c->cache_read_races);
 493                 s->iop.error = -EINTR;
 494         }
 495
 496         bch_bbio_endio(s->iop.c, bio, error, "reading from cache");
 497 }
 498
 499 /*
 500  * Read from a single key, handling the initial cache miss if the key starts in
 501  * the middle of the bio
 502  */
 503 static int cache_lookup_fn(struct btree_op *op, struct btree *b, struct bkey *k)
 504 {
 505         struct search *s = container_of(op, struct search, op);
 506         struct bio *n, *bio = &s->bio.bio;
 507         struct bkey *bio_key;
 508         unsigned ptr;
 509
 510         if (bkey_cmp(k, &KEY(s->iop.inode, bio->bi_iter.bi_sector, 0)) <= 0)
 511                 return MAP_CONTINUE;
 512
 513         if (KEY_INODE(k) != s->iop.inode ||
 514             KEY_START(k) > bio->bi_iter.bi_sector) {
 515                 unsigned bio_sectors = bio_sectors(bio);
 516                 unsigned sectors = KEY_INODE(k) == s->iop.inode
 517                         ? min_t(uint64_t, INT_MAX,
 518                                 KEY_START(k) - bio->bi_iter.bi_sector)
 519                         : INT_MAX;
 520
 521                 int ret = s->d->cache_miss(b, s, bio, sectors);
 522                 if (ret != MAP_CONTINUE)
 523                         return ret;
 524
 525                 /* if this was a complete miss we shouldn't get here */
 526                 BUG_ON(bio_sectors <= sectors);
 527         }
 528
 529         if (!KEY_SIZE(k))
 530                 return MAP_CONTINUE;
 531
 532         /* XXX: figure out best pointer - for multiple cache devices */
 533         ptr = 0;
 534
 535         PTR_BUCKET(b->c, k, ptr)->prio = INITIAL_PRIO;
 536
 537         if (KEY_DIRTY(k))
 538                 s->read_dirty_data = true;
 539
 540         n = bio_next_split(bio, min_t(uint64_t, INT_MAX,
 541                                       KEY_OFFSET(k) - bio->bi_iter.bi_sector),
 542                            GFP_NOIO, s->d->bio_split);
 543
 544         bio_key = &container_of(n, struct bbio, bio)->key;
 545         bch_bkey_copy_single_ptr(bio_key, k, ptr);
 546
 547         bch_cut_front(&KEY(s->iop.inode, n->bi_iter.bi_sector, 0), bio_key);
 548         bch_cut_back(&KEY(s->iop.inode, bio_end_sector(n), 0), bio_key);
 549
 550         n->bi_end_io    = bch_cache_read_endio;
 551         n->bi_private   = &s->cl;
 552
 553         /*
 554          * The bucket we're reading from might be reused while our bio
 555          * is in flight, and we could then end up reading the wrong
 556          * data.
 557          *
 558          * We guard against this by checking (in cache_read_endio()) if
 559          * the pointer is stale again; if so, we treat it as an error
 560          * and reread from the backing device (but we don't pass that
 561          * error up anywhere).
 562          */
 563
 564         __bch_submit_bbio(n, b->c);
 565         return n == bio ? MAP_DONE : MAP_CONTINUE;
 566 }
 567
 568 static void cache_lookup(struct closure *cl)
 569 {
 570         struct search *s = container_of(cl, struct search, iop.cl);
 571         struct bio *bio = &s->bio.bio;
 572         int ret;
 573
 574         bch_btree_op_init(&s->op, -1);
 575
 576         ret = bch_btree_map_keys(&s->op, s->iop.c,
 577                                  &KEY(s->iop.inode, bio->bi_iter.bi_sector, 0),
 578                                  cache_lookup_fn, MAP_END_KEY);
 579         if (ret == -EAGAIN)
 580                 continue_at(cl, cache_lookup, bcache_wq);
 581
 582         closure_return(cl);
 583 }
 584
 585 /* Common code for the make_request functions */
 586
 587 static void request_endio(struct bio *bio, int error)
 588 {
 589         struct closure *cl = bio->bi_private;
 590
 591         if (error) {
 592                 struct search *s = container_of(cl, struct search, cl);
 593                 s->iop.error = error;
 594                 /* Only cache read errors are recoverable */
 595                 s->recoverable = false;
 596         }
 597
 598         bio_put(bio);
 599         closure_put(cl);
 600 }
 601
 602 static void bio_complete(struct search *s)
 603 {
 604         if (s->orig_bio) {
 605                 generic_end_io_acct(bio_data_dir(s->orig_bio),
 606                                     &s->d->disk->part0, s->start_time);
 607
 608                 trace_bcache_request_end(s->d, s->orig_bio);
 609                 bio_endio(s->orig_bio, s->iop.error);
 610                 s->orig_bio = NULL;
 611         }
 612 }
 613
 614 static void do_bio_hook(struct search *s, struct bio *orig_bio)
 615 {
 616         struct bio *bio = &s->bio.bio;
 617
 618         bio_init(bio);
 619         __bio_clone_fast(bio, orig_bio);
 620         bio->bi_end_io          = request_endio;
 621         bio->bi_private         = &s->cl;
 622
 623         bio_cnt_set(bio, 3);
 624 }
 625
 626 static void search_free(struct closure *cl)
 627 {
 628         struct search *s = container_of(cl, struct search, cl);
 629         bio_complete(s);
 630
 631         if (s->iop.bio)
 632                 bio_put(s->iop.bio);
 633
 634         closure_debug_destroy(cl);
 635         mempool_free(s, s->d->c->search);
 636 }
 637
 638 static inline struct search *search_alloc(struct bio *bio,
 639                                           struct bcache_device *d)
 640 {
 641         struct search *s;
 642
 643         s = mempool_alloc(d->c->search, GFP_NOIO);
 644
 645         closure_init(&s->cl, NULL);
 646         do_bio_hook(s, bio);
 647
 648         s->orig_bio             = bio;
 649         s->cache_miss           = NULL;
 650         s->d                    = d;
 651         s->recoverable          = 1;
 652         s->write                = (bio->bi_rw & REQ_WRITE) != 0;
 653         s->read_dirty_data      = 0;
 654         s->start_time           = jiffies;
 655
 656         s->iop.c                = d->c;
 657         s->iop.bio              = NULL;
 658         s->iop.inode            = d->id;
 659         s->iop.write_point      = hash_long((unsigned long) current, 16);
 660         s->iop.write_prio       = 0;
 661         s->iop.error            = 0;
 662         s->iop.flags            = 0;
 663         s->iop.flush_journal    = (bio->bi_rw & (REQ_FLUSH|REQ_FUA)) != 0;
 664         s->iop.wq               = bcache_wq;
 665
 666         return s;
 667 }
 668
 669 /* Cached devices */
 670
 671 static void cached_dev_bio_complete(struct closure *cl)
 672 {
 673         struct search *s = container_of(cl, struct search, cl);
 674         struct cached_dev *dc = container_of(s->d, struct cached_dev, disk);
 675
 676         search_free(cl);
 677         cached_dev_put(dc);
 678 }
 679
 680 /* Process reads */
 681
 682 static void cached_dev_cache_miss_done(struct closure *cl)
 683 {
 684         struct search *s = container_of(cl, struct search, cl);
 685
 686         if (s->iop.replace_collision)
 687                 bch_mark_cache_miss_collision(s->iop.c, s->d);
 688
 689         if (s->iop.bio) {
 690                 int i;
 691                 struct bio_vec *bv;
 692
 693                 bio_for_each_segment_all(bv, s->iop.bio, i)
 694                         __free_page(bv->bv_page);
 695         }
 696
 697         cached_dev_bio_complete(cl);
 698 }
 699
 700 static void cached_dev_read_error(struct closure *cl)
 701 {
 702         struct search *s = container_of(cl, struct search, cl);
 703         struct bio *bio = &s->bio.bio;
 704
 705         if (s->recoverable) {
 706                 /* Retry from the backing device: */
 707                 trace_bcache_read_retry(s->orig_bio);
 708
 709                 s->iop.error = 0;
 710                 do_bio_hook(s, s->orig_bio);
 711
 712                 /* XXX: invalidate cache */
 713
 714                 closure_bio_submit(bio, cl, s->d);
 715         }
 716
 717         continue_at(cl, cached_dev_cache_miss_done, NULL);
 718 }
 719
 720 static void cached_dev_read_done(struct closure *cl)
 721 {
 722         struct search *s = container_of(cl, struct search, cl);
 723         struct cached_dev *dc = container_of(s->d, struct cached_dev, disk);
 724
 725         /*
 726          * We had a cache miss; cache_bio now contains data ready to be inserted
 727          * into the cache.
 728          *
 729          * First, we copy the data we just read from cache_bio's bounce buffers
 730          * to the buffers the original bio pointed to:
 731          */
 732
 733         if (s->iop.bio) {
 734                 bio_reset(s->iop.bio);
 735                 s->iop.bio->bi_iter.bi_sector = s->cache_miss->bi_iter.bi_sector;
 736                 s->iop.bio->bi_bdev = s->cache_miss->bi_bdev;
 737                 s->iop.bio->bi_iter.bi_size = s->insert_bio_sectors << 9;
 738                 bch_bio_map(s->iop.bio, NULL);
 739
 740                 bio_copy_data(s->cache_miss, s->iop.bio);
 741
 742                 bio_put(s->cache_miss);
 743                 s->cache_miss = NULL;
 744         }
 745
 746         if (verify(dc, &s->bio.bio) && s->recoverable && !s->read_dirty_data)
 747                 bch_data_verify(dc, s->orig_bio);
 748
 749         bio_complete(s);
 750
 751         if (s->iop.bio &&
 752             !test_bit(CACHE_SET_STOPPING, &s->iop.c->flags)) {
 753                 BUG_ON(!s->iop.replace);
 754                 closure_call(&s->iop.cl, bch_data_insert, NULL, cl);
 755         }
 756
 757         continue_at(cl, cached_dev_cache_miss_done, NULL);
 758 }
 759
 760 static void cached_dev_read_done_bh(struct closure *cl)
 761 {
 762         struct search *s = container_of(cl, struct search, cl);
 763         struct cached_dev *dc = container_of(s->d, struct cached_dev, disk);
 764
 765         bch_mark_cache_accounting(s->iop.c, s->d,
 766                                   !s->cache_miss, s->iop.bypass);
 767         trace_bcache_read(s->orig_bio, !s->cache_miss, s->iop.bypass);
 768
 769         if (s->iop.error)
 770                 continue_at_nobarrier(cl, cached_dev_read_error, bcache_wq);
 771         else if (s->iop.bio || verify(dc, &s->bio.bio))
 772                 continue_at_nobarrier(cl, cached_dev_read_done, bcache_wq);
 773         else
 774                 continue_at_nobarrier(cl, cached_dev_bio_complete, NULL);
 775 }
 776
 777 static int cached_dev_cache_miss(struct btree *b, struct search *s,
 778                                  struct bio *bio, unsigned sectors)
 779 {
 780         int ret = MAP_CONTINUE;
 781         unsigned reada = 0;
 782         struct cached_dev *dc = container_of(s->d, struct cached_dev, disk);
 783         struct bio *miss, *cache_bio;
 784
 785         if (s->cache_miss || s->iop.bypass) {
 786                 miss = bio_next_split(bio, sectors, GFP_NOIO, s->d->bio_split);
 787                 ret = miss == bio ? MAP_DONE : MAP_CONTINUE;
 788                 goto out_submit;
 789         }
 790
 791         if (!(bio->bi_rw & REQ_RAHEAD) &&
 792             !(bio->bi_rw & REQ_META) &&
 793             s->iop.c->gc_stats.in_use < CUTOFF_CACHE_READA)
 794                 reada = min_t(sector_t, dc->readahead >> 9,
 795                               bdev_sectors(bio->bi_bdev) - bio_end_sector(bio));
 796
 797         s->insert_bio_sectors = min(sectors, bio_sectors(bio) + reada);
 798
 799         s->iop.replace_key = KEY(s->iop.inode,
 800                                  bio->bi_iter.bi_sector + s->insert_bio_sectors,
 801                                  s->insert_bio_sectors);
 802
 803         ret = bch_btree_insert_check_key(b, &s->op, &s->iop.replace_key);
 804         if (ret)
 805                 return ret;
 806
 807         s->iop.replace = true;
 808
 809         miss = bio_next_split(bio, sectors, GFP_NOIO, s->d->bio_split);
 810
 811         /* btree_search_recurse()'s btree iterator is no good anymore */
 812         ret = miss == bio ? MAP_DONE : -EINTR;
 813
 814         cache_bio = bio_alloc_bioset(GFP_NOWAIT,
 815                         DIV_ROUND_UP(s->insert_bio_sectors, PAGE_SECTORS),
 816                         dc->disk.bio_split);
 817         if (!cache_bio)
 818                 goto out_submit;
 819
 820         cache_bio->bi_iter.bi_sector    = miss->bi_iter.bi_sector;
 821         cache_bio->bi_bdev              = miss->bi_bdev;
 822         cache_bio->bi_iter.bi_size      = s->insert_bio_sectors << 9;
 823
 824         cache_bio->bi_end_io    = request_endio;
 825         cache_bio->bi_private   = &s->cl;
 826
 827         bch_bio_map(cache_bio, NULL);
 828         if (bio_alloc_pages(cache_bio, __GFP_NOWARN|GFP_NOIO))
 829                 goto out_put;
 830
 831         if (reada)
 832                 bch_mark_cache_readahead(s->iop.c, s->d);
 833
 834         s->cache_miss   = miss;
 835         s->iop.bio      = cache_bio;
 836         bio_get(cache_bio);
 837         closure_bio_submit(cache_bio, &s->cl, s->d);
 838
 839         return ret;
 840 out_put:
 841         bio_put(cache_bio);
 842 out_submit:
 843         miss->bi_end_io         = request_endio;
 844         miss->bi_private        = &s->cl;
 845         closure_bio_submit(miss, &s->cl, s->d);
 846         return ret;
 847 }
 848
 849 static void cached_dev_read(struct cached_dev *dc, struct search *s)
 850 {
 851         struct closure *cl = &s->cl;
 852
 853         closure_call(&s->iop.cl, cache_lookup, NULL, cl);
 854         continue_at(cl, cached_dev_read_done_bh, NULL);
 855 }
 856
 857 /* Process writes */
 858
 859 static void cached_dev_write_complete(struct closure *cl)
 860 {
 861         struct search *s = container_of(cl, struct search, cl);
 862         struct cached_dev *dc = container_of(s->d, struct cached_dev, disk);
 863
 864         up_read_non_owner(&dc->writeback_lock);
 865         cached_dev_bio_complete(cl);
 866 }
 867
 868 static void cached_dev_write(struct cached_dev *dc, struct search *s)
 869 {
 870         struct closure *cl = &s->cl;
 871         struct bio *bio = &s->bio.bio;
 872         struct bkey start = KEY(dc->disk.id, bio->bi_iter.bi_sector, 0);
 873         struct bkey end = KEY(dc->disk.id, bio_end_sector(bio), 0);
 874
 875         bch_keybuf_check_overlapping(&s->iop.c->moving_gc_keys, &start, &end);
 876
 877         down_read_non_owner(&dc->writeback_lock);
 878         if (bch_keybuf_check_overlapping(&dc->writeback_keys, &start, &end)) {
 879                 /*
 880                  * We overlap with some dirty data undergoing background
 881                  * writeback, force this write to writeback
 882                  */
 883                 s->iop.bypass = false;
 884                 s->iop.writeback = true;
 885         }
 886
 887         /*
 888          * Discards aren't _required_ to do anything, so skipping if
 889          * check_overlapping returned true is ok
 890          *
 891          * But check_overlapping drops dirty keys for which io hasn't started,
 892          * so we still want to call it.
 893          */
 894         if (bio->bi_rw & REQ_DISCARD)
 895                 s->iop.bypass = true;
 896
 897         if (should_writeback(dc, s->orig_bio,
 898                              cache_mode(dc, bio),
 899                              s->iop.bypass)) {
 900                 s->iop.bypass = false;
 901                 s->iop.writeback = true;
 902         }
 903
 904         if (s->iop.bypass) {
 905                 s->iop.bio = s->orig_bio;
 906                 bio_get(s->iop.bio);
 907
 908                 if (!(bio->bi_rw & REQ_DISCARD) ||
 909                     blk_queue_discard(bdev_get_queue(dc->bdev)))
 910                         closure_bio_submit(bio, cl, s->d);
 911         } else if (s->iop.writeback) {
 912                 bch_writeback_add(dc);
 913                 s->iop.bio = bio;
 914
 915                 if (bio->bi_rw & REQ_FLUSH) {
 916                         /* Also need to send a flush to the backing device */
 917                         struct bio *flush = bio_alloc_bioset(GFP_NOIO, 0,
 918                                                              dc->disk.bio_split);
 919
 920                         flush->bi_rw    = WRITE_FLUSH;
 921                         flush->bi_bdev  = bio->bi_bdev;
 922                         flush->bi_end_io = request_endio;
 923                         flush->bi_private = cl;
 924
 925                         closure_bio_submit(flush, cl, s->d);
 926                 }
 927         } else {
 928                 s->iop.bio = bio_clone_fast(bio, GFP_NOIO, dc->disk.bio_split);
 929
 930                 closure_bio_submit(bio, cl, s->d);
 931         }
 932
 933         closure_call(&s->iop.cl, bch_data_insert, NULL, cl);
 934         continue_at(cl, cached_dev_write_complete, NULL);
 935 }
 936
 937 static void cached_dev_nodata(struct closure *cl)
 938 {
 939         struct search *s = container_of(cl, struct search, cl);
 940         struct bio *bio = &s->bio.bio;
 941
 942         if (s->iop.flush_journal)
 943                 bch_journal_meta(s->iop.c, cl);
 944
 945         /* If it's a flush, we send the flush to the backing device too */
 946         closure_bio_submit(bio, cl, s->d);
 947
 948         continue_at(cl, cached_dev_bio_complete, NULL);
 949 }
 950
 951 /* Cached devices - read & write stuff */
 952
 953 static void cached_dev_make_request(struct request_queue *q, struct bio *bio)
 954 {
 955         struct search *s;
 956         struct bcache_device *d = bio->bi_bdev->bd_disk->private_data;
 957         struct cached_dev *dc = container_of(d, struct cached_dev, disk);
 958         int rw = bio_data_dir(bio);
 959
 960         generic_start_io_acct(rw, bio_sectors(bio), &d->disk->part0);
 961
 962         bio->bi_bdev = dc->bdev;
 963         bio->bi_iter.bi_sector += dc->sb.data_offset;
 964
 965         if (cached_dev_get(dc)) {
 966                 s = search_alloc(bio, d);
 967                 trace_bcache_request_start(s->d, bio);
 968
 969                 if (!bio->bi_iter.bi_size) {
 970                         /*
 971                          * can't call bch_journal_meta from under
 972                          * generic_make_request
 973                          */
 974                         continue_at_nobarrier(&s->cl,
 975                                               cached_dev_nodata,
 976                                               bcache_wq);
 977                 } else {
 978                         s->iop.bypass = check_should_bypass(dc, bio);
 979
 980                         if (rw)
 981                                 cached_dev_write(dc, s);
 982                         else
 983                                 cached_dev_read(dc, s);
 984                 }
 985         } else {
 986                 if ((bio->bi_rw & REQ_DISCARD) &&
 987                     !blk_queue_discard(bdev_get_queue(dc->bdev)))
 988                         bio_endio(bio, 0);
 989                 else
 990                         bch_generic_make_request(bio, &d->bio_split_hook);
 991         }
 992 }
 993
 994 static int cached_dev_ioctl(struct bcache_device *d, fmode_t mode,
 995                             unsigned int cmd, unsigned long arg)
 996 {
 997         struct cached_dev *dc = container_of(d, struct cached_dev, disk);
 998         return __blkdev_driver_ioctl(dc->bdev, mode, cmd, arg);
 999 }
1000
1001 static int cached_dev_congested(void *data, int bits)
1002 {
1003         struct bcache_device *d = data;
1004         struct cached_dev *dc = container_of(d, struct cached_dev, disk);
1005         struct request_queue *q = bdev_get_queue(dc->bdev);
1006         int ret = 0;
1007
1008         if (bdi_congested(&q->backing_dev_info, bits))
1009                 return 1;
1010
1011         if (cached_dev_get(dc)) {
1012                 unsigned i;
1013                 struct cache *ca;
1014
1015                 for_each_cache(ca, d->c, i) {
1016                         q = bdev_get_queue(ca->bdev);
1017                         ret |= bdi_congested(&q->backing_dev_info, bits);
1018                 }
1019
1020                 cached_dev_put(dc);
1021         }
1022
1023         return ret;
1024 }
1025
1026 void bch_cached_dev_request_init(struct cached_dev *dc)
1027 {
1028         struct gendisk *g = dc->disk.disk;
1029
1030         g->queue->make_request_fn               = cached_dev_make_request;
1031         g->queue->backing_dev_info.congested_fn = cached_dev_congested;
1032         dc->disk.cache_miss                     = cached_dev_cache_miss;
1033         dc->disk.ioctl                          = cached_dev_ioctl;
1034 }
1035
1036 /* Flash backed devices */
1037
1038 static int flash_dev_cache_miss(struct btree *b, struct search *s,
1039                                 struct bio *bio, unsigned sectors)
1040 {
1041         unsigned bytes = min(sectors, bio_sectors(bio)) << 9;
1042
1043         swap(bio->bi_iter.bi_size, bytes);
1044         zero_fill_bio(bio);
1045         swap(bio->bi_iter.bi_size, bytes);
1046
1047         bio_advance(bio, bytes);
1048
1049         if (!bio->bi_iter.bi_size)
1050                 return MAP_DONE;
1051
1052         return MAP_CONTINUE;
1053 }
1054
1055 static void flash_dev_nodata(struct closure *cl)
1056 {
1057         struct search *s = container_of(cl, struct search, cl);
1058
1059         if (s->iop.flush_journal)
1060                 bch_journal_meta(s->iop.c, cl);
1061
1062         continue_at(cl, search_free, NULL);
1063 }
1064
1065 static void flash_dev_make_request(struct request_queue *q, struct bio *bio)
1066 {
1067         struct search *s;
1068         struct closure *cl;
1069         struct bcache_device *d = bio->bi_bdev->bd_disk->private_data;
1070         int rw = bio_data_dir(bio);
1071
1072         generic_start_io_acct(rw, bio_sectors(bio), &d->disk->part0);
1073
1074         s = search_alloc(bio, d);
1075         cl = &s->cl;
1076         bio = &s->bio.bio;
1077
1078         trace_bcache_request_start(s->d, bio);
1079
1080         if (!bio->bi_iter.bi_size) {
1081                 /*
1082                  * can't call bch_journal_meta from under
1083                  * generic_make_request
1084                  */
1085                 continue_at_nobarrier(&s->cl,
1086                                       flash_dev_nodata,
1087                                       bcache_wq);
1088         } else if (rw) {
1089                 bch_keybuf_check_overlapping(&s->iop.c->moving_gc_keys,
1090                                         &KEY(d->id, bio->bi_iter.bi_sector, 0),
1091                                         &KEY(d->id, bio_end_sector(bio), 0));
1092
1093                 s->iop.bypass           = (bio->bi_rw & REQ_DISCARD) != 0;
1094                 s->iop.writeback        = true;
1095                 s->iop.bio              = bio;
1096
1097                 closure_call(&s->iop.cl, bch_data_insert, NULL, cl);
1098         } else {
1099                 closure_call(&s->iop.cl, cache_lookup, NULL, cl);
1100         }
1101
1102         continue_at(cl, search_free, NULL);
1103 }
1104
1105 static int flash_dev_ioctl(struct bcache_device *d, fmode_t mode,
1106                            unsigned int cmd, unsigned long arg)
1107 {
1108         return -ENOTTY;
1109 }
1110
1111 static int flash_dev_congested(void *data, int bits)
1112 {
1113         struct bcache_device *d = data;
1114         struct request_queue *q;
1115         struct cache *ca;
1116         unsigned i;
1117         int ret = 0;
1118
1119         for_each_cache(ca, d->c, i) {
1120                 q = bdev_get_queue(ca->bdev);
1121                 ret |= bdi_congested(&q->backing_dev_info, bits);
1122         }
1123
1124         return ret;
1125 }
1126
1127 void bch_flash_dev_request_init(struct bcache_device *d)
1128 {
1129         struct gendisk *g = d->disk;
1130
1131         g->queue->make_request_fn               = flash_dev_make_request;
1132         g->queue->backing_dev_info.congested_fn = flash_dev_congested;
1133         d->cache_miss                           = flash_dev_cache_miss;
1134         d->ioctl                                = flash_dev_ioctl;
1135 }
1136
1137 void bch_request_exit(void)
1138 {
1139         if (bch_search_cache)
1140                 kmem_cache_destroy(bch_search_cache);
1141 }
1142
1143 int __init bch_request_init(void)
1144 {
1145         bch_search_cache = KMEM_CACHE(search, 0);
1146         if (!bch_search_cache)
1147                 return -ENOMEM;
1148
1149         return 0;
1150 }